Context Navigation

source: rtems/cpukit/libnetworking/netinet/tcp_input.c @ b6ea96ef

4.104.114.84.95

Last change on this file since b6ea96ef was 3fb76805, checked in by Joel Sherrill <joel.sherrill@…>, on 10/04/02 at 14:47:11

2002-10-04 Jay Monkman <jtm@…>

netinet/in_cksum.c, netinet/ip_icmp.h, netinet/ip_input.c, netinet/tcp_input.c, netinet/tcp_subr.c, netinet/tcp_var.h, sys/queue.h: Address alignment requirements for the ARM.

Property mode set to 100644

File size: 60.4 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 acknowledgment:
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	* Call with ti==0 after become established to
160	* force pre-ESTABLISHED data up to user socket.
161	*/
162	if (ti == 0)
163	goto present;
164
165	/*
166	* Find a segment which begins after this one does.
167	*/
168	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
169	q = (struct tcpiphdr *)q->ti_next)
170	if (SEQ_GT(q->ti_seq, ti->ti_seq))
171	break;
172
173	/*
174	* If there is a preceding segment, it may provide some of
175	* our data already. If so, drop the data from the incoming
176	* segment. If it provides all of our data, drop us.
177	*/
178	if ((struct tcpiphdr )q->ti_prev != (struct tcpiphdr )tp) {
179	register int i;
180	q = (struct tcpiphdr *)q->ti_prev;
181	/* conversion to int (in i) handles seq wraparound */
182	i = q->ti_seq + q->ti_len - ti->ti_seq;
183	if (i > 0) {
184	if (i >= ti->ti_len) {
185	tcpstat.tcps_rcvduppack++;
186	tcpstat.tcps_rcvdupbyte += ti->ti_len;
187	m_freem(m);
188	/*
189	* Try to present any queued data
190	* at the left window edge to the user.
191	* This is needed after the 3-WHS
192	* completes.
193	*/
194	goto present; /* ??? */
195	}
196	m_adj(m, i);
197	ti->ti_len -= i;
198	ti->ti_seq += i;
199	}
200	q = (struct tcpiphdr *)(q->ti_next);
201	}
202	tcpstat.tcps_rcvoopack++;
203	tcpstat.tcps_rcvoobyte += ti->ti_len;
204	#if (defined(__GNUC__) && defined(__arm__))
205	STR32_UNALGN(ti,m);
206	#else
207	REASS_MBUF(ti) = m; /* XXX */
208	#endif
209	/*
210	* While we overlap succeeding segments trim them or,
211	* if they are completely covered, dequeue them.
212	*/
213	while (q != (struct tcpiphdr *)tp) {
214	register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
215	if (i <= 0)
216	break;
217	if (i < q->ti_len) {
218	q->ti_seq += i;
219	q->ti_len -= i;
220	#if (defined(__GNUC__) && defined(__arm__))
221	LD32_UNALGN(q,m);
222	m_adj(m, i);
223	#else
224	m_adj(REASS_MBUF(q), i);
225	#endif
226	break;
227	}
228	q = (struct tcpiphdr *)q->ti_next;
229	#if (defined(__GNUC__) && defined(__arm__))
230	LD32_UNALGN((struct tcpiphdr *)q->ti_prev,m);
231	#else
232	m = REASS_MBUF((struct tcpiphdr *)q->ti_prev);
233	#endif
234	remque(q->ti_prev);
235	m_freem(m);
236	}
237
238	/*
239	* Stick new segment in its place.
240	*/
241	insque(ti, q->ti_prev);
242
243	present:
244	/*
245	* Present data to user, advancing rcv_nxt through
246	* completed sequence space.
247	*/
248	if (!TCPS_HAVEESTABLISHED(tp->t_state))
249	return (0);
250	ti = tp->seg_next;
251	if (ti == (struct tcpiphdr *)tp \|\| ti->ti_seq != tp->rcv_nxt)
252	return (0);
253	do {
254	tp->rcv_nxt += ti->ti_len;
255	flags = ti->ti_flags & TH_FIN;
256	remque(ti);
257	#if (defined(__GNUC__) && defined(__arm__))
258	LD32_UNALGN(ti,m);
259	#else
260	m = REASS_MBUF(ti);
261	#endif
262	ti = (struct tcpiphdr *)ti->ti_next;
263	if (so->so_state & SS_CANTRCVMORE)
264	m_freem(m);
265	else
266	sbappend(&so->so_rcv, m);
267	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
268	sorwakeup(so);
269	return (flags);
270	}
271
272	/*
273	* TCP input routine, follows pages 65-76 of the
274	* protocol specification dated September, 1981 very closely.
275	*/
276	void
277	tcp_input(m, iphlen)
278	register struct mbuf *m;
279	int iphlen;
280	{
281	register struct tcpiphdr *ti;
282	register struct inpcb *inp;
283	u_char *optp = NULL;
284	int optlen = 0;
285	int len, tlen, off;
286	register struct tcpcb *tp = 0;
287	register int tiflags;
288	struct socket *so = 0;
289	int todrop, acked, ourfinisacked, needoutput = 0;
290	struct in_addr laddr;
291	int dropsocket = 0;
292	int iss = 0;
293	u_long tiwin;
294	struct tcpopt to; /* options in this segment */
295	struct rmxp_tao taop; / pointer to our TAO cache entry */
296	struct rmxp_tao tao_noncached; /* in case there's no cached entry */
297	#ifdef TCPDEBUG
298	short ostate = 0;
299	#endif
300
301	bzero((char *)&to, sizeof(to));
302
303	tcpstat.tcps_rcvtotal++;
304	/*
305	* Get IP and TCP header together in first mbuf.
306	* Note: IP leaves IP header in first mbuf.
307	*/
308	ti = mtod(m, struct tcpiphdr *);
309	if (iphlen > sizeof (struct ip))
310	ip_stripoptions(m, (struct mbuf *)0);
311	if (m->m_len < sizeof (struct tcpiphdr)) {
312	if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
313	tcpstat.tcps_rcvshort++;
314	return;
315	}
316	ti = mtod(m, struct tcpiphdr *);
317	}
318
319	/*
320	* Checksum extended TCP header and data.
321	*/
322	tlen = ((struct ip *)ti)->ip_len;
323	len = sizeof (struct ip) + tlen;
324	ti->ti_next = ti->ti_prev = 0;
325	ti->ti_x1 = 0;
326	ti->ti_len = (u_short)tlen;
327	HTONS(ti->ti_len);
328	ti->ti_sum = in_cksum(m, len);
329	if (ti->ti_sum) {
330	tcpstat.tcps_rcvbadsum++;
331	goto drop;
332	}
333	#endif /* TUBA_INCLUDE */
334
335	/*
336	* Check that TCP offset makes sense,
337	* pull out TCP options and adjust length. XXX
338	*/
339	off = ti->ti_off << 2;
340	if (off < sizeof (struct tcphdr) \|\| off > tlen) {
341	tcpstat.tcps_rcvbadoff++;
342	goto drop;
343	}
344	tlen -= off;
345	ti->ti_len = tlen;
346	if (off > sizeof (struct tcphdr)) {
347	if (m->m_len < sizeof(struct ip) + off) {
348	if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
349	tcpstat.tcps_rcvshort++;
350	return;
351	}
352	ti = mtod(m, struct tcpiphdr *);
353	}
354	optlen = off - sizeof (struct tcphdr);
355	optp = mtod(m, u_char *) + sizeof (struct tcpiphdr);
356	}
357	tiflags = ti->ti_flags;
358
359	/*
360	* Convert TCP protocol specific fields to host format.
361	*/
362	NTOHL(ti->ti_seq);
363	NTOHL(ti->ti_ack);
364	NTOHS(ti->ti_win);
365	NTOHS(ti->ti_urp);
366
367	/*
368	* Drop TCP, IP headers and TCP options.
369	*/
370	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
371	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
372
373	/*
374	* Locate pcb for segment.
375	*/
376	findpcb:
377	inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport,
378	ti->ti_dst, ti->ti_dport, 1);
379
380	/*
381	* If the state is CLOSED (i.e., TCB does not exist) then
382	* all data in the incoming segment is discarded.
383	* If the TCB exists but is in CLOSED state, it is embryonic,
384	* but should either do a listen or a connect soon.
385	*/
386	if (inp == NULL) {
387	if (log_in_vain && tiflags & TH_SYN) {
388	char buf[4*sizeof "123"];
389
390	strcpy(buf, inet_ntoa(ti->ti_dst));
391	log(LOG_INFO, "Connection attempt to TCP %s:%d"
392	" from %s:%d\n",
393	buf, ntohs(ti->ti_dport),
394	inet_ntoa(ti->ti_src), ntohs(ti->ti_sport));
395	}
396	goto dropwithreset;
397	}
398	tp = intotcpcb(inp);
399	if (tp == 0)
400	goto dropwithreset;
401	if (tp->t_state == TCPS_CLOSED)
402	goto drop;
403
404	/* Unscale the window into a 32-bit value. */
405	if ((tiflags & TH_SYN) == 0)
406	tiwin = ti->ti_win << tp->snd_scale;
407	else
408	tiwin = ti->ti_win;
409
410	so = inp->inp_socket;
411	if (so->so_options & (SO_DEBUG\|SO_ACCEPTCONN)) {
412	#ifdef TCPDEBUG
413	if (so->so_options & SO_DEBUG) {
414	ostate = tp->t_state;
415	tcp_saveti = *ti;
416	}
417	#endif
418	if (so->so_options & SO_ACCEPTCONN) {
419	register struct tcpcb *tp0 = tp;
420	struct socket *so2;
421	if ((tiflags & (TH_RST\|TH_ACK\|TH_SYN)) != TH_SYN) {
422	/*
423	* Note: dropwithreset makes sure we don't
424	* send a RST in response to a RST.
425	*/
426	if (tiflags & TH_ACK) {
427	tcpstat.tcps_badsyn++;
428	goto dropwithreset;
429	}
430	goto drop;
431	}
432	so2 = sonewconn(so, 0);
433	if (so2 == 0) {
434	tcpstat.tcps_listendrop++;
435	so2 = sodropablereq(so);
436	if (so2) {
437	tcp_drop(sototcpcb(so2), ETIMEDOUT);
438	so2 = sonewconn(so, 0);
439	}
440	if (!so2)
441	goto drop;
442	}
443	so = so2;
444	/*
445	* This is ugly, but ....
446	*
447	* Mark socket as temporary until we're
448	* committed to keeping it. The code at
449	* ``drop'' and ``dropwithreset'' check the
450	* flag dropsocket to see if the temporary
451	* socket created here should be discarded.
452	* We mark the socket as discardable until
453	* we're committed to it below in TCPS_LISTEN.
454	*/
455	dropsocket++;
456	inp = (struct inpcb *)so->so_pcb;
457	inp->inp_laddr = ti->ti_dst;
458	inp->inp_lport = ti->ti_dport;
459	in_pcbrehash(inp);
460	#if BSD>=43
461	inp->inp_options = ip_srcroute();
462	#endif
463	tp = intotcpcb(inp);
464	tp->t_state = TCPS_LISTEN;
465	tp->t_flags \|= tp0->t_flags & (TF_NOPUSH\|TF_NOOPT);
466
467	/* Compute proper scaling value from buffer space */
468	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
469	TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
470	tp->request_r_scale++;
471	}
472	}
473
474	/*
475	* Segment received on connection.
476	* Reset idle time and keep-alive timer.
477	*/
478	tp->t_idle = 0;
479	if (TCPS_HAVEESTABLISHED(tp->t_state))
480	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
481
482	/*
483	* Process options if not in LISTEN state,
484	* else do it below (after getting remote address).
485	*/
486	if (tp->t_state != TCPS_LISTEN)
487	tcp_dooptions(tp, optp, optlen, ti, &to);
488
489	/*
490	* Header prediction: check for the two common cases
491	* of a uni-directional data xfer. If the packet has
492	* no control flags, is in-sequence, the window didn't
493	* change and we're not retransmitting, it's a
494	* candidate. If the length is zero and the ack moved
495	* forward, we're the sender side of the xfer. Just
496	* free the data acked & wake any higher level process
497	* that was blocked waiting for space. If the length
498	* is non-zero and the ack didn't move, we're the
499	* receiver side. If we're getting packets in-order
500	* (the reassembly queue is empty), add the data to
501	* the socket buffer and note that we need a delayed ack.
502	* Make sure that the hidden state-flags are also off.
503	* Since we check for TCPS_ESTABLISHED above, it can only
504	* be TH_NEEDSYN.
505	*/
506	if (tp->t_state == TCPS_ESTABLISHED &&
507	(tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK &&
508	((tp->t_flags & (TF_NEEDSYN\|TF_NEEDFIN)) == 0) &&
509	((to.to_flag & TOF_TS) == 0 \|\|
510	TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
511	/*
512	* Using the CC option is compulsory if once started:
513	* the segment is OK if no T/TCP was negotiated or
514	* if the segment has a CC option equal to CCrecv
515	*/
516	((tp->t_flags & (TF_REQ_CC\|TF_RCVD_CC)) != (TF_REQ_CC\|TF_RCVD_CC) \|\|
517	((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
518	ti->ti_seq == tp->rcv_nxt &&
519	tiwin && tiwin == tp->snd_wnd &&
520	tp->snd_nxt == tp->snd_max) {
521
522	/*
523	* If last ACK falls within this segment's sequence numbers,
524	* record the timestamp.
525	* NOTE that the test is modified according to the latest
526	* proposal of the tcplw@cray.com list (Braden 1993/04/26).
527	*/
528	if ((to.to_flag & TOF_TS) != 0 &&
529	SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
530	tp->ts_recent_age = tcp_now;
531	tp->ts_recent = to.to_tsval;
532	}
533
534	if (ti->ti_len == 0) {
535	if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
536	SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
537	tp->snd_cwnd >= tp->snd_wnd &&
538	tp->t_dupacks < tcprexmtthresh) {
539	/*
540	* this is a pure ack for outstanding data.
541	*/
542	++tcpstat.tcps_predack;
543	if ((to.to_flag & TOF_TS) != 0)
544	tcp_xmit_timer(tp,
545	tcp_now - to.to_tsecr + 1);
546	else if (tp->t_rtt &&
547	SEQ_GT(ti->ti_ack, tp->t_rtseq))
548	tcp_xmit_timer(tp, tp->t_rtt);
549	acked = ti->ti_ack - tp->snd_una;
550	tcpstat.tcps_rcvackpack++;
551	tcpstat.tcps_rcvackbyte += acked;
552	sbdrop(&so->so_snd, acked);
553	tp->snd_una = ti->ti_ack;
554	m_freem(m);
555
556	/*
557	* If all outstanding data are acked, stop
558	* retransmit timer, otherwise restart timer
559	* using current (possibly backed-off) value.
560	* If process is waiting for space,
561	* wakeup/selwakeup/signal. If data
562	* are ready to send, let tcp_output
563	* decide between more output or persist.
564	*/
565	if (tp->snd_una == tp->snd_max)
566	tp->t_timer[TCPT_REXMT] = 0;
567	else if (tp->t_timer[TCPT_PERSIST] == 0)
568	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
569
570	if (so->so_snd.sb_flags & SB_NOTIFY)
571	sowwakeup(so);
572	if (so->so_snd.sb_cc)
573	(void) tcp_output(tp);
574	return;
575	}
576	} else if (ti->ti_ack == tp->snd_una &&
577	tp->seg_next == (struct tcpiphdr *)tp &&
578	ti->ti_len <= sbspace(&so->so_rcv)) {
579	/*
580	* this is a pure, in-sequence data packet
581	* with nothing on the reassembly queue and
582	* we have enough buffer space to take it.
583	*/
584	++tcpstat.tcps_preddat;
585	tp->rcv_nxt += ti->ti_len;
586	tcpstat.tcps_rcvpack++;
587	tcpstat.tcps_rcvbyte += ti->ti_len;
588	/*
589	* Add data to socket buffer.
590	*/
591	sbappend(&so->so_rcv, m);
592	sorwakeup(so);
593	#ifdef TCP_ACK_HACK
594	/*
595	* If this is a short packet, then ACK now - with Nagel
596	* congestion avoidance sender won't send more until
597	* he gets an ACK.
598	*/
599	if (tiflags & TH_PUSH) {
600	tp->t_flags \|= TF_ACKNOW;
601	tcp_output(tp);
602	} else {
603	tp->t_flags \|= TF_DELACK;
604	}
605	#else
606	tp->t_flags \|= TF_DELACK;
607	#endif
608	return;
609	}
610	}
611
612	/*
613	* Calculate amount of space in receive window,
614	* and then do TCP input processing.
615	* Receive window is amount of space in rcv queue,
616	* but not less than advertised window.
617	*/
618	{ int win;
619
620	win = sbspace(&so->so_rcv);
621	if (win < 0)
622	win = 0;
623	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
624	}
625
626	switch (tp->t_state) {
627
628	/*
629	* If the state is LISTEN then ignore segment if it contains an RST.
630	* If the segment contains an ACK then it is bad and send a RST.
631	* If it does not contain a SYN then it is not interesting; drop it.
632	* If it is from this socket, drop it, it must be forged.
633	* Don't bother responding if the destination was a broadcast.
634	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
635	* tp->iss, and send a segment:
636	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
637	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
638	* Fill in remote peer address fields if not previously specified.
639	* Enter SYN_RECEIVED state, and process any other fields of this
640	* segment in this state.
641	*/
642	case TCPS_LISTEN: {
643	struct mbuf *am;
644	register struct sockaddr_in *sin;
645
646	if (tiflags & TH_RST)
647	goto drop;
648	if (tiflags & TH_ACK)
649	goto dropwithreset;
650	if ((tiflags & TH_SYN) == 0)
651	goto drop;
652	if ((ti->ti_dport == ti->ti_sport) &&
653	(ti->ti_dst.s_addr == ti->ti_src.s_addr))
654	goto drop;
655	/*
656	* RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
657	* in_broadcast() should never return true on a received
658	* packet with M_BCAST not set.
659	*/
660	if (m->m_flags & (M_BCAST\|M_MCAST) \|\|
661	IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
662	goto drop;
663	am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */
664	if (am == NULL)
665	goto drop;
666	am->m_len = sizeof (struct sockaddr_in);
667	sin = mtod(am, struct sockaddr_in *);
668	sin->sin_family = AF_INET;
669	sin->sin_len = sizeof(*sin);
670	sin->sin_addr = ti->ti_src;
671	sin->sin_port = ti->ti_sport;
672	bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
673	laddr = inp->inp_laddr;
674	if (inp->inp_laddr.s_addr == INADDR_ANY)
675	inp->inp_laddr = ti->ti_dst;
676	if (in_pcbconnect(inp, am)) {
677	inp->inp_laddr = laddr;
678	(void) m_free(am);
679	goto drop;
680	}
681	(void) m_free(am);
682	tp->t_template = tcp_template(tp);
683	if (tp->t_template == 0) {
684	tp = tcp_drop(tp, ENOBUFS);
685	dropsocket = 0; /* socket is already gone */
686	goto drop;
687	}
688	if ((taop = tcp_gettaocache(inp)) == NULL) {
689	taop = &tao_noncached;
690	bzero(taop, sizeof(*taop));
691	}
692	tcp_dooptions(tp, optp, optlen, ti, &to);
693	if (iss)
694	tp->iss = iss;
695	else
696	tp->iss = tcp_iss;
697	tcp_iss += TCP_ISSINCR/4;
698	tp->irs = ti->ti_seq;
699	tcp_sendseqinit(tp);
700	tcp_rcvseqinit(tp);
701	/*
702	* Initialization of the tcpcb for transaction;
703	* set SND.WND = SEG.WND,
704	* initialize CCsend and CCrecv.
705	*/
706	tp->snd_wnd = tiwin; /* initial send-window */
707	tp->cc_send = CC_INC(tcp_ccgen);
708	tp->cc_recv = to.to_cc;
709	/*
710	* Perform TAO test on incoming CC (SEG.CC) option, if any.
711	* - compare SEG.CC against cached CC from the same host,
712	* if any.
713	* - if SEG.CC > chached value, SYN must be new and is accepted
714	* immediately: save new CC in the cache, mark the socket
715	* connected, enter ESTABLISHED state, turn on flag to
716	* send a SYN in the next segment.
717	* A virtual advertised window is set in rcv_adv to
718	* initialize SWS prevention. Then enter normal segment
719	* processing: drop SYN, process data and FIN.
720	* - otherwise do a normal 3-way handshake.
721	*/
722	if ((to.to_flag & TOF_CC) != 0) {
723	if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
724	taop->tao_cc = to.to_cc;
725	tp->t_state = TCPS_ESTABLISHED;
726
727	/*
728	* If there is a FIN, or if there is data and the
729	* connection is local, then delay SYN,ACK(SYN) in
730	* the hope of piggy-backing it on a response
731	* segment. Otherwise must send ACK now in case
732	* the other side is slow starting.
733	*/
734	if ((tiflags & TH_FIN) \|\| (ti->ti_len != 0 &&
735	in_localaddr(inp->inp_faddr)))
736	tp->t_flags \|= (TF_DELACK \| TF_NEEDSYN);
737	else
738	tp->t_flags \|= (TF_ACKNOW \| TF_NEEDSYN);
739
740	/*
741	* Limit the `virtual advertised window' to TCP_MAXWIN
742	* here. Even if we requested window scaling, it will
743	* become effective only later when our SYN is acked.
744	*/
745	tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
746	tcpstat.tcps_connects++;
747	soisconnected(so);
748	tp->t_timer[TCPT_KEEP] = tcp_keepinit;
749	dropsocket = 0; /* committed to socket */
750	tcpstat.tcps_accepts++;
751	goto trimthenstep6;
752	}
753	/* else do standard 3-way handshake */
754	} else {
755	/*
756	* No CC option, but maybe CC.NEW:
757	* invalidate cached value.
758	*/
759	taop->tao_cc = 0;
760	}
761	/*
762	* TAO test failed or there was no CC option,
763	* do a standard 3-way handshake.
764	*/
765	tp->t_flags \|= TF_ACKNOW;
766	tp->t_state = TCPS_SYN_RECEIVED;
767	tp->t_timer[TCPT_KEEP] = tcp_keepinit;
768	dropsocket = 0; /* committed to socket */
769	tcpstat.tcps_accepts++;
770	goto trimthenstep6;
771	}
772
773	/*
774	* If the state is SYN_RECEIVED:
775	* if seg contains SYN/ACK, send a RST.
776	* if seg contains an ACK, but not for our SYN/ACK, send a RST.
777	*/
778	case TCPS_SYN_RECEIVED:
779	if (tiflags & TH_ACK) {
780	if (tiflags & TH_SYN) {
781	tcpstat.tcps_badsyn++;
782	goto dropwithreset;
783	}
784	if (SEQ_LEQ(ti->ti_ack, tp->snd_una) \|\|
785	SEQ_GT(ti->ti_ack, tp->snd_max))
786	goto dropwithreset;
787	}
788	break;
789
790	/*
791	* If the state is SYN_SENT:
792	* if seg contains an ACK, but not for our SYN, drop the input.
793	* if seg contains a RST, then drop the connection.
794	* if seg does not contain SYN, then drop it.
795	* Otherwise this is an acceptable SYN segment
796	* initialize tp->rcv_nxt and tp->irs
797	* if seg contains ack then advance tp->snd_una
798	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
799	* arrange for segment to be acked (eventually)
800	* continue processing rest of data/controls, beginning with URG
801	*/
802	case TCPS_SYN_SENT:
803	if ((taop = tcp_gettaocache(inp)) == NULL) {
804	taop = &tao_noncached;
805	bzero(taop, sizeof(*taop));
806	}
807
808	if ((tiflags & TH_ACK) &&
809	(SEQ_LEQ(ti->ti_ack, tp->iss) \|\|
810	SEQ_GT(ti->ti_ack, tp->snd_max))) {
811	/*
812	* If we have a cached CCsent for the remote host,
813	* hence we haven't just crashed and restarted,
814	* do not send a RST. This may be a retransmission
815	* from the other side after our earlier ACK was lost.
816	* Our new SYN, when it arrives, will serve as the
817	* needed ACK.
818	*/
819	if (taop->tao_ccsent != 0)
820	goto drop;
821	else
822	goto dropwithreset;
823	}
824	if (tiflags & TH_RST) {
825	if (tiflags & TH_ACK)
826	tp = tcp_drop(tp, ECONNREFUSED);
827	goto drop;
828	}
829	if ((tiflags & TH_SYN) == 0)
830	goto drop;
831	tp->snd_wnd = ti->ti_win; /* initial send window */
832	tp->cc_recv = to.to_cc; /* foreign CC */
833
834	tp->irs = ti->ti_seq;
835	tcp_rcvseqinit(tp);
836	if (tiflags & TH_ACK) {
837	/*
838	* Our SYN was acked. If segment contains CC.ECHO
839	* option, check it to make sure this segment really
840	* matches our SYN. If not, just drop it as old
841	* duplicate, but send an RST if we're still playing
842	* by the old rules. If no CC.ECHO option, make sure
843	* we don't get fooled into using T/TCP.
844	*/
845	if (to.to_flag & TOF_CCECHO) {
846	if (tp->cc_send != to.to_ccecho) {
847	if (taop->tao_ccsent != 0)
848	goto drop;
849	else
850	goto dropwithreset;
851	}
852	} else
853	tp->t_flags &= ~TF_RCVD_CC;
854	tcpstat.tcps_connects++;
855	soisconnected(so);
856	/* Do window scaling on this connection? */
857	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
858	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
859	tp->snd_scale = tp->requested_s_scale;
860	tp->rcv_scale = tp->request_r_scale;
861	}
862	/* Segment is acceptable, update cache if undefined. */
863	if (taop->tao_ccsent == 0)
864	taop->tao_ccsent = to.to_ccecho;
865
866	tp->rcv_adv += tp->rcv_wnd;
867	tp->snd_una++; /* SYN is acked */
868	/*
869	* If there's data, delay ACK; if there's also a FIN
870	* ACKNOW will be turned on later.
871	*/
872	if (ti->ti_len != 0)
873	tp->t_flags \|= TF_DELACK;
874	else
875	tp->t_flags \|= TF_ACKNOW;
876	/*
877	* Received <SYN,ACK> in SYN_SENT[*] state.
878	* Transitions:
879	* SYN_SENT --> ESTABLISHED
880	* SYN_SENT* --> FIN_WAIT_1
881	*/
882	if (tp->t_flags & TF_NEEDFIN) {
883	tp->t_state = TCPS_FIN_WAIT_1;
884	tp->t_flags &= ~TF_NEEDFIN;
885	tiflags &= ~TH_SYN;
886	} else {
887	tp->t_state = TCPS_ESTABLISHED;
888	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
889	}
890	} else {
891	/*
892	* Received initial SYN in SYN-SENT[*] state => simul-
893	* taneous open. If segment contains CC option and there is
894	* a cached CC, apply TAO test; if it succeeds, connection is
895	* half-synchronized. Otherwise, do 3-way handshake:
896	* SYN-SENT -> SYN-RECEIVED
897	* SYN-SENT* -> SYN-RECEIVED*
898	* If there was no CC option, clear cached CC value.
899	*/
900	tp->t_flags \|= TF_ACKNOW;
901	tp->t_timer[TCPT_REXMT] = 0;
902	if (to.to_flag & TOF_CC) {
903	if (taop->tao_cc != 0 &&
904	CC_GT(to.to_cc, taop->tao_cc)) {
905	/*
906	* update cache and make transition:
907	* SYN-SENT -> ESTABLISHED*
908	* SYN-SENT* -> FIN-WAIT-1*
909	*/
910	taop->tao_cc = to.to_cc;
911	if (tp->t_flags & TF_NEEDFIN) {
912	tp->t_state = TCPS_FIN_WAIT_1;
913	tp->t_flags &= ~TF_NEEDFIN;
914	} else {
915	tp->t_state = TCPS_ESTABLISHED;
916	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
917	}
918	tp->t_flags \|= TF_NEEDSYN;
919	} else
920	tp->t_state = TCPS_SYN_RECEIVED;
921	} else {
922	/* CC.NEW or no option => invalidate cache */
923	taop->tao_cc = 0;
924	tp->t_state = TCPS_SYN_RECEIVED;
925	}
926	}
927
928	trimthenstep6:
929	/*
930	* Advance ti->ti_seq to correspond to first data byte.
931	* If data, trim to stay within window,
932	* dropping FIN if necessary.
933	*/
934	ti->ti_seq++;
935	if (ti->ti_len > tp->rcv_wnd) {
936	todrop = ti->ti_len - tp->rcv_wnd;
937	m_adj(m, -todrop);
938	ti->ti_len = tp->rcv_wnd;
939	tiflags &= ~TH_FIN;
940	tcpstat.tcps_rcvpackafterwin++;
941	tcpstat.tcps_rcvbyteafterwin += todrop;
942	}
943	tp->snd_wl1 = ti->ti_seq - 1;
944	tp->rcv_up = ti->ti_seq;
945	/*
946	* Client side of transaction: already sent SYN and data.
947	* If the remote host used T/TCP to validate the SYN,
948	* our data will be ACK'd; if so, enter normal data segment
949	* processing in the middle of step 5, ack processing.
950	* Otherwise, goto step 6.
951	*/
952	if (tiflags & TH_ACK)
953	goto process_ACK;
954	goto step6;
955	/*
956	* If the state is LAST_ACK or CLOSING or TIME_WAIT:
957	* if segment contains a SYN and CC [not CC.NEW] option:
958	* if state == TIME_WAIT and connection duration > MSL,
959	* drop packet and send RST;
960	*
961	* if SEG.CC > CCrecv then is new SYN, and can implicitly
962	* ack the FIN (and data) in retransmission queue.
963	* Complete close and delete TCPCB. Then reprocess
964	* segment, hoping to find new TCPCB in LISTEN state;
965	*
966	* else must be old SYN; drop it.
967	* else do normal processing.
968	*/
969	case TCPS_LAST_ACK:
970	case TCPS_CLOSING:
971	case TCPS_TIME_WAIT:
972	if ((tiflags & TH_SYN) &&
973	(to.to_flag & TOF_CC) && tp->cc_recv != 0) {
974	if (tp->t_state == TCPS_TIME_WAIT &&
975	tp->t_duration > TCPTV_MSL)
976	goto dropwithreset;
977	if (CC_GT(to.to_cc, tp->cc_recv)) {
978	tp = tcp_close(tp);
979	goto findpcb;
980	}
981	else
982	goto drop;
983	}
984	break; /* continue normal processing */
985	}
986
987	/*
988	* States other than LISTEN or SYN_SENT.
989	* First check timestamp, if present.
990	* Then check the connection count, if present.
991	* Then check that at least some bytes of segment are within
992	* receive window. If segment begins before rcv_nxt,
993	* drop leading data (and SYN); if nothing left, just ack.
994	*
995	* RFC 1323 PAWS: If we have a timestamp reply on this segment
996	* and it's less than ts_recent, drop it.
997	*/
998	if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 &&
999	tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1000
1001	/* Check to see if ts_recent is over 24 days old. */
1002	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1003	/*
1004	* Invalidate ts_recent. If this segment updates
1005	* ts_recent, the age will be reset later and ts_recent
1006	* will get a valid value. If it does not, setting
1007	* ts_recent to zero will at least satisfy the
1008	* requirement that zero be placed in the timestamp
1009	* echo reply when ts_recent isn't valid. The
1010	* age isn't reset until we get a valid ts_recent
1011	* because we don't want out-of-order segments to be
1012	* dropped when ts_recent is old.
1013	*/
1014	tp->ts_recent = 0;
1015	} else {
1016	tcpstat.tcps_rcvduppack++;
1017	tcpstat.tcps_rcvdupbyte += ti->ti_len;
1018	tcpstat.tcps_pawsdrop++;
1019	goto dropafterack;
1020	}
1021	}
1022
1023	/*
1024	* T/TCP mechanism
1025	* If T/TCP was negotiated and the segment doesn't have CC,
1026	* or if it's CC is wrong then drop the segment.
1027	* RST segments do not have to comply with this.
1028	*/
1029	if ((tp->t_flags & (TF_REQ_CC\|TF_RCVD_CC)) == (TF_REQ_CC\|TF_RCVD_CC) &&
1030	((to.to_flag & TOF_CC) == 0 \|\| tp->cc_recv != to.to_cc) &&
1031	(tiflags & TH_RST) == 0)
1032	goto dropafterack;
1033
1034	todrop = tp->rcv_nxt - ti->ti_seq;
1035	if (todrop > 0) {
1036	if (tiflags & TH_SYN) {
1037	tiflags &= ~TH_SYN;
1038	ti->ti_seq++;
1039	if (ti->ti_urp > 1)
1040	ti->ti_urp--;
1041	else
1042	tiflags &= ~TH_URG;
1043	todrop--;
1044	}
1045	/*
1046	* Following if statement from Stevens, vol. 2, p. 960.
1047	*/
1048	if (todrop > ti->ti_len
1049	\|\| (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
1050	/*
1051	* Any valid FIN must be to the left of the window.
1052	* At this point the FIN must be a duplicate or out
1053	* of sequence; drop it.
1054	*/
1055	tiflags &= ~TH_FIN;
1056
1057	/*
1058	* Send an ACK to resynchronize and drop any data.
1059	* But keep on processing for RST or ACK.
1060	*/
1061	tp->t_flags \|= TF_ACKNOW;
1062	todrop = ti->ti_len;
1063	tcpstat.tcps_rcvduppack++;
1064	tcpstat.tcps_rcvdupbyte += todrop;
1065	} else {
1066	tcpstat.tcps_rcvpartduppack++;
1067	tcpstat.tcps_rcvpartdupbyte += todrop;
1068	}
1069	m_adj(m, todrop);
1070	ti->ti_seq += todrop;
1071	ti->ti_len -= todrop;
1072	if (ti->ti_urp > todrop)
1073	ti->ti_urp -= todrop;
1074	else {
1075	tiflags &= ~TH_URG;
1076	ti->ti_urp = 0;
1077	}
1078	}
1079
1080	/*
1081	* If new data are received on a connection after the
1082	* user processes are gone, then RST the other end.
1083	*/
1084	if ((so->so_state & SS_NOFDREF) &&
1085	tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
1086	tp = tcp_close(tp);
1087	tcpstat.tcps_rcvafterclose++;
1088	goto dropwithreset;
1089	}
1090
1091	/*
1092	* If segment ends after window, drop trailing data
1093	* (and PUSH and FIN); if nothing left, just ACK.
1094	*/
1095	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1096	if (todrop > 0) {
1097	tcpstat.tcps_rcvpackafterwin++;
1098	if (todrop >= ti->ti_len) {
1099	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1100	/*
1101	* If a new connection request is received
1102	* while in TIME_WAIT, drop the old connection
1103	* and start over if the sequence numbers
1104	* are above the previous ones.
1105	*/
1106	if (tiflags & TH_SYN &&
1107	tp->t_state == TCPS_TIME_WAIT &&
1108	SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
1109	iss = tp->rcv_nxt + TCP_ISSINCR;
1110	tp = tcp_close(tp);
1111	goto findpcb;
1112	}
1113	/*
1114	* If window is closed can only take segments at
1115	* window edge, and have to drop data and PUSH from
1116	* incoming segments. Continue processing, but
1117	* remember to ack. Otherwise, drop segment
1118	* and ack.
1119	*/
1120	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
1121	tp->t_flags \|= TF_ACKNOW;
1122	tcpstat.tcps_rcvwinprobe++;
1123	} else
1124	goto dropafterack;
1125	} else
1126	tcpstat.tcps_rcvbyteafterwin += todrop;
1127	m_adj(m, -todrop);
1128	ti->ti_len -= todrop;
1129	tiflags &= ~(TH_PUSH\|TH_FIN);
1130	}
1131
1132	/*
1133	* If last ACK falls within this segment's sequence numbers,
1134	* record its timestamp.
1135	* NOTE that the test is modified according to the latest
1136	* proposal of the tcplw@cray.com list (Braden 1993/04/26).
1137	*/
1138	if ((to.to_flag & TOF_TS) != 0 &&
1139	SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
1140	tp->ts_recent_age = tcp_now;
1141	tp->ts_recent = to.to_tsval;
1142	}
1143
1144	/*
1145	* If the RST bit is set examine the state:
1146	* SYN_RECEIVED STATE:
1147	* If passive open, return to LISTEN state.
1148	* If active open, inform user that connection was refused.
1149	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1150	* Inform user that connection was reset, and close tcb.
1151	* CLOSING, LAST_ACK, TIME_WAIT STATES
1152	* Close the tcb.
1153	*/
1154	if (tiflags&TH_RST) switch (tp->t_state) {
1155
1156	case TCPS_SYN_RECEIVED:
1157	so->so_error = ECONNREFUSED;
1158	goto close;
1159
1160	case TCPS_ESTABLISHED:
1161	case TCPS_FIN_WAIT_1:
1162	case TCPS_FIN_WAIT_2:
1163	case TCPS_CLOSE_WAIT:
1164	so->so_error = ECONNRESET;
1165	close:
1166	tp->t_state = TCPS_CLOSED;
1167	tcpstat.tcps_drops++;
1168	tp = tcp_close(tp);
1169	goto drop;
1170
1171	case TCPS_CLOSING:
1172	case TCPS_LAST_ACK:
1173	case TCPS_TIME_WAIT:
1174	tp = tcp_close(tp);
1175	goto drop;
1176	}
1177
1178	/*
1179	* If a SYN is in the window, then this is an
1180	* error and we send an RST and drop the connection.
1181	*/
1182	if (tiflags & TH_SYN) {
1183	tp = tcp_drop(tp, ECONNRESET);
1184	goto dropwithreset;
1185	}
1186
1187	/*
1188	* If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1189	* flag is on (half-synchronized state), then queue data for
1190	* later processing; else drop segment and return.
1191	*/
1192	if ((tiflags & TH_ACK) == 0) {
1193	if (tp->t_state == TCPS_SYN_RECEIVED \|\|
1194	(tp->t_flags & TF_NEEDSYN))
1195	goto step6;
1196	else
1197	goto drop;
1198	}
1199
1200	/*
1201	* Ack processing.
1202	*/
1203	switch (tp->t_state) {
1204
1205	/*
1206	* In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1207	* ESTABLISHED state and continue processing.
1208	* The ACK was checked above.
1209	*/
1210	case TCPS_SYN_RECEIVED:
1211
1212	tcpstat.tcps_connects++;
1213	soisconnected(so);
1214	/* Do window scaling? */
1215	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
1216	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
1217	tp->snd_scale = tp->requested_s_scale;
1218	tp->rcv_scale = tp->request_r_scale;
1219	}
1220	/*
1221	* Upon successful completion of 3-way handshake,
1222	* update cache.CC if it was undefined, pass any queued
1223	* data to the user, and advance state appropriately.
1224	*/
1225	if ((taop = tcp_gettaocache(inp)) != NULL &&
1226	taop->tao_cc == 0)
1227	taop->tao_cc = tp->cc_recv;
1228
1229	/*
1230	* Make transitions:
1231	* SYN-RECEIVED -> ESTABLISHED
1232	* SYN-RECEIVED* -> FIN-WAIT-1
1233	*/
1234	if (tp->t_flags & TF_NEEDFIN) {
1235	tp->t_state = TCPS_FIN_WAIT_1;
1236	tp->t_flags &= ~TF_NEEDFIN;
1237	} else {
1238	tp->t_state = TCPS_ESTABLISHED;
1239	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1240	}
1241	/*
1242	* If segment contains data or ACK, will call tcp_reass()
1243	* later; if not, do so now to pass queued data to user.
1244	*/
1245	if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0)
1246	(void) tcp_reass(tp, (struct tcpiphdr *)0,
1247	(struct mbuf *)0);
1248	tp->snd_wl1 = ti->ti_seq - 1;
1249	/* fall into ... */
1250
1251	/*
1252	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1253	* ACKs. If the ack is in the range
1254	* tp->snd_una < ti->ti_ack <= tp->snd_max
1255	* then advance tp->snd_una to ti->ti_ack and drop
1256	* data from the retransmission queue. If this ACK reflects
1257	* more up to date window information we update our window information.
1258	*/
1259	case TCPS_ESTABLISHED:
1260	case TCPS_FIN_WAIT_1:
1261	case TCPS_FIN_WAIT_2:
1262	case TCPS_CLOSE_WAIT:
1263	case TCPS_CLOSING:
1264	case TCPS_LAST_ACK:
1265	case TCPS_TIME_WAIT:
1266
1267	if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1268	if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1269	tcpstat.tcps_rcvdupack++;
1270	/*
1271	* If we have outstanding data (other than
1272	* a window probe), this is a completely
1273	* duplicate ack (ie, window info didn't
1274	* change), the ack is the biggest we've
1275	* seen and we've seen exactly our rexmt
1276	* threshhold of them, assume a packet
1277	* has been dropped and retransmit it.
1278	* Kludge snd_nxt & the congestion
1279	* window so we send only this one
1280	* packet.
1281	*
1282	* We know we're losing at the current
1283	* window size so do congestion avoidance
1284	* (set ssthresh to half the current window
1285	* and pull our congestion window back to
1286	* the new ssthresh).
1287	*
1288	* Dup acks mean that packets have left the
1289	* network (they're now cached at the receiver)
1290	* so bump cwnd by the amount in the receiver
1291	* to keep a constant cwnd packets in the
1292	* network.
1293	*/
1294	if (tp->t_timer[TCPT_REXMT] == 0 \|\|
1295	ti->ti_ack != tp->snd_una)
1296	tp->t_dupacks = 0;
1297	else if (++tp->t_dupacks == tcprexmtthresh) {
1298	tcp_seq onxt = tp->snd_nxt;
1299	u_int win =
1300	min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1301	tp->t_maxseg;
1302
1303	if (win < 2)
1304	win = 2;
1305	tp->snd_ssthresh = win * tp->t_maxseg;
1306	tp->t_timer[TCPT_REXMT] = 0;
1307	tp->t_rtt = 0;
1308	tp->snd_nxt = ti->ti_ack;
1309	tp->snd_cwnd = tp->t_maxseg;
1310	(void) tcp_output(tp);
1311	tp->snd_cwnd = tp->snd_ssthresh +
1312	tp->t_maxseg * tp->t_dupacks;
1313	if (SEQ_GT(onxt, tp->snd_nxt))
1314	tp->snd_nxt = onxt;
1315	goto drop;
1316	} else if (tp->t_dupacks > tcprexmtthresh) {
1317	tp->snd_cwnd += tp->t_maxseg;
1318	(void) tcp_output(tp);
1319	goto drop;
1320	}
1321	} else
1322	tp->t_dupacks = 0;
1323	break;
1324	}
1325	/*
1326	* If the congestion window was inflated to account
1327	* for the other side's cached packets, retract it.
1328	*/
1329	if (tp->t_dupacks >= tcprexmtthresh &&
1330	tp->snd_cwnd > tp->snd_ssthresh)
1331	tp->snd_cwnd = tp->snd_ssthresh;
1332	tp->t_dupacks = 0;
1333	if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1334	tcpstat.tcps_rcvacktoomuch++;
1335	goto dropafterack;
1336	}
1337	/*
1338	* If we reach this point, ACK is not a duplicate,
1339	* i.e., it ACKs something we sent.
1340	*/
1341	if (tp->t_flags & TF_NEEDSYN) {
1342	/*
1343	* T/TCP: Connection was half-synchronized, and our
1344	* SYN has been ACK'd (so connection is now fully
1345	* synchronized). Go to non-starred state,
1346	* increment snd_una for ACK of SYN, and check if
1347	* we can do window scaling.
1348	*/
1349	tp->t_flags &= ~TF_NEEDSYN;
1350	tp->snd_una++;
1351	/* Do window scaling? */
1352	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
1353	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
1354	tp->snd_scale = tp->requested_s_scale;
1355	tp->rcv_scale = tp->request_r_scale;
1356	}
1357	}
1358
1359	process_ACK:
1360	acked = ti->ti_ack - tp->snd_una;
1361	tcpstat.tcps_rcvackpack++;
1362	tcpstat.tcps_rcvackbyte += acked;
1363
1364	/*
1365	* If we have a timestamp reply, update smoothed
1366	* round trip time. If no timestamp is present but
1367	* transmit timer is running and timed sequence
1368	* number was acked, update smoothed round trip time.
1369	* Since we now have an rtt measurement, cancel the
1370	* timer backoff (cf., Phil Karn's retransmit alg.).
1371	* Recompute the initial retransmit timer.
1372	*/
1373	if (to.to_flag & TOF_TS)
1374	tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1);
1375	else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1376	tcp_xmit_timer(tp,tp->t_rtt);
1377
1378	/*
1379	* If all outstanding data is acked, stop retransmit
1380	* timer and remember to restart (more output or persist).
1381	* If there is more data to be acked, restart retransmit
1382	* timer, using current (possibly backed-off) value.
1383	*/
1384	if (ti->ti_ack == tp->snd_max) {
1385	tp->t_timer[TCPT_REXMT] = 0;
1386	needoutput = 1;
1387	} else if (tp->t_timer[TCPT_PERSIST] == 0)
1388	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1389
1390	/*
1391	* If no data (only SYN) was ACK'd,
1392	* skip rest of ACK processing.
1393	*/
1394	if (acked == 0)
1395	goto step6;
1396
1397	/*
1398	* When new data is acked, open the congestion window.
1399	* If the window gives us less than ssthresh packets
1400	* in flight, open exponentially (maxseg per packet).
1401	* Otherwise open linearly: maxseg per window
1402	* (maxseg^2 / cwnd per packet).
1403	*/
1404	{
1405	register u_int cw = tp->snd_cwnd;
1406	register u_int incr = tp->t_maxseg;
1407
1408	if (cw > tp->snd_ssthresh)
1409	incr = incr * incr / cw;
1410	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1411	}
1412	if (acked > so->so_snd.sb_cc) {
1413	tp->snd_wnd -= so->so_snd.sb_cc;
1414	sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1415	ourfinisacked = 1;
1416	} else {
1417	sbdrop(&so->so_snd, acked);
1418	tp->snd_wnd -= acked;
1419	ourfinisacked = 0;
1420	}
1421	if (so->so_snd.sb_flags & SB_NOTIFY)
1422	sowwakeup(so);
1423	tp->snd_una = ti->ti_ack;
1424	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1425	tp->snd_nxt = tp->snd_una;
1426
1427	switch (tp->t_state) {
1428
1429	/*
1430	* In FIN_WAIT_1 STATE in addition to the processing
1431	* for the ESTABLISHED state if our FIN is now acknowledged
1432	* then enter FIN_WAIT_2.
1433	*/
1434	case TCPS_FIN_WAIT_1:
1435	if (ourfinisacked) {
1436	/*
1437	* If we can't receive any more
1438	* data, then closing user can proceed.
1439	* Starting the timer is contrary to the
1440	* specification, but if we don't get a FIN
1441	* we'll hang forever.
1442	*/
1443	if (so->so_state & SS_CANTRCVMORE) {
1444	soisdisconnected(so);
1445	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1446	}
1447	tp->t_state = TCPS_FIN_WAIT_2;
1448	}
1449	break;
1450
1451	/*
1452	* In CLOSING STATE in addition to the processing for
1453	* the ESTABLISHED state if the ACK acknowledges our FIN
1454	* then enter the TIME-WAIT state, otherwise ignore
1455	* the segment.
1456	*/
1457	case TCPS_CLOSING:
1458	if (ourfinisacked) {
1459	tp->t_state = TCPS_TIME_WAIT;
1460	tcp_canceltimers(tp);
1461	/* Shorten TIME_WAIT [RFC-1644, p.28] */
1462	if (tp->cc_recv != 0 &&
1463	tp->t_duration < TCPTV_MSL)
1464	tp->t_timer[TCPT_2MSL] =
1465	tp->t_rxtcur * TCPTV_TWTRUNC;
1466	else
1467	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1468	soisdisconnected(so);
1469	}
1470	break;
1471
1472	/*
1473	* In LAST_ACK, we may still be waiting for data to drain
1474	* and/or to be acked, as well as for the ack of our FIN.
1475	* If our FIN is now acknowledged, delete the TCB,
1476	* enter the closed state and return.
1477	*/
1478	case TCPS_LAST_ACK:
1479	if (ourfinisacked) {
1480	tp = tcp_close(tp);
1481	goto drop;
1482	}
1483	break;
1484
1485	/*
1486	* In TIME_WAIT state the only thing that should arrive
1487	* is a retransmission of the remote FIN. Acknowledge
1488	* it and restart the finack timer.
1489	*/
1490	case TCPS_TIME_WAIT:
1491	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1492	goto dropafterack;
1493	}
1494	}
1495
1496	step6:
1497	/*
1498	* Update window information.
1499	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1500	*/
1501	if ((tiflags & TH_ACK) &&
1502	(SEQ_LT(tp->snd_wl1, ti->ti_seq) \|\|
1503	(tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) \|\|
1504	(tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1505	/* keep track of pure window updates */
1506	if (ti->ti_len == 0 &&
1507	tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1508	tcpstat.tcps_rcvwinupd++;
1509	tp->snd_wnd = tiwin;
1510	tp->snd_wl1 = ti->ti_seq;
1511	tp->snd_wl2 = ti->ti_ack;
1512	if (tp->snd_wnd > tp->max_sndwnd)
1513	tp->max_sndwnd = tp->snd_wnd;
1514	needoutput = 1;
1515	}
1516
1517	/*
1518	* Process segments with URG.
1519	*/
1520	if ((tiflags & TH_URG) && ti->ti_urp &&
1521	TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1522	/*
1523	* This is a kludge, but if we receive and accept
1524	* random urgent pointers, we'll crash in
1525	* soreceive. It's hard to imagine someone
1526	* actually wanting to send this much urgent data.
1527	*/
1528	if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1529	ti->ti_urp = 0; /* XXX */
1530	tiflags &= ~TH_URG; /* XXX */
1531	goto dodata; /* XXX */
1532	}
1533	/*
1534	* If this segment advances the known urgent pointer,
1535	* then mark the data stream. This should not happen
1536	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1537	* a FIN has been received from the remote side.
1538	* In these states we ignore the URG.
1539	*
1540	* According to RFC961 (Assigned Protocols),
1541	* the urgent pointer points to the last octet
1542	* of urgent data. We continue, however,
1543	* to consider it to indicate the first octet
1544	* of data past the urgent section as the original
1545	* spec states (in one of two places).
1546	*/
1547	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1548	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1549	so->so_oobmark = so->so_rcv.sb_cc +
1550	(tp->rcv_up - tp->rcv_nxt) - 1;
1551	if (so->so_oobmark == 0)
1552	so->so_state \|= SS_RCVATMARK;
1553	sohasoutofband(so);
1554	tp->t_oobflags &= ~(TCPOOB_HAVEDATA \| TCPOOB_HADDATA);
1555	}
1556	/*
1557	* Remove out of band data so doesn't get presented to user.
1558	* This can happen independent of advancing the URG pointer,
1559	* but if two URG's are pending at once, some out-of-band
1560	* data may creep in... ick.
1561	*/
1562	if (ti->ti_urp <= (u_long)ti->ti_len
1563	#ifdef SO_OOBINLINE
1564	&& (so->so_options & SO_OOBINLINE) == 0
1565	#endif
1566	)
1567	tcp_pulloutofband(so, ti, m);
1568	} else
1569	/*
1570	* If no out of band data is expected,
1571	* pull receive urgent pointer along
1572	* with the receive window.
1573	*/
1574	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1575	tp->rcv_up = tp->rcv_nxt;
1576	dodata: /* XXX */
1577
1578	/*
1579	* Process the segment text, merging it into the TCP sequencing queue,
1580	* and arranging for acknowledgment of receipt if necessary.
1581	* This process logically involves adjusting tp->rcv_wnd as data
1582	* is presented to the user (this happens in tcp_usrreq.c,
1583	* case PRU_RCVD). If a FIN has already been received on this
1584	* connection then we just ignore the text.
1585	*/
1586	if ((ti->ti_len \|\| (tiflags&TH_FIN)) &&
1587	TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1588	TCP_REASS(tp, ti, m, so, tiflags);
1589	/*
1590	* Note the amount of data that peer has sent into
1591	* our window, in order to estimate the sender's
1592	* buffer size.
1593	*/
1594	len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1595	} else {
1596	m_freem(m);
1597	tiflags &= ~TH_FIN;
1598	}
1599
1600	/*
1601	* If FIN is received ACK the FIN and let the user know
1602	* that the connection is closing.
1603	*/
1604	if (tiflags & TH_FIN) {
1605	if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1606	socantrcvmore(so);
1607	/*
1608	* If connection is half-synchronized
1609	* (ie NEEDSYN flag on) then delay ACK,
1610	* so it may be piggybacked when SYN is sent.
1611	* Otherwise, since we received a FIN then no
1612	* more input can be expected, send ACK now.
1613	*/
1614	if (tp->t_flags & TF_NEEDSYN)
1615	tp->t_flags \|= TF_DELACK;
1616	else
1617	tp->t_flags \|= TF_ACKNOW;
1618	tp->rcv_nxt++;
1619	}
1620	switch (tp->t_state) {
1621
1622	/*
1623	* In SYN_RECEIVED and ESTABLISHED STATES
1624	* enter the CLOSE_WAIT state.
1625	*/
1626	case TCPS_SYN_RECEIVED:
1627	case TCPS_ESTABLISHED:
1628	tp->t_state = TCPS_CLOSE_WAIT;
1629	break;
1630
1631	/*
1632	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1633	* enter the CLOSING state.
1634	*/
1635	case TCPS_FIN_WAIT_1:
1636	tp->t_state = TCPS_CLOSING;
1637	break;
1638
1639	/*
1640	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1641	* starting the time-wait timer, turning off the other
1642	* standard timers.
1643	*/
1644	case TCPS_FIN_WAIT_2:
1645	tp->t_state = TCPS_TIME_WAIT;
1646	tcp_canceltimers(tp);
1647	/* Shorten TIME_WAIT [RFC-1644, p.28] */
1648	if (tp->cc_recv != 0 &&
1649	tp->t_duration < TCPTV_MSL) {
1650	tp->t_timer[TCPT_2MSL] =
1651	tp->t_rxtcur * TCPTV_TWTRUNC;
1652	/* For transaction client, force ACK now. */
1653	tp->t_flags \|= TF_ACKNOW;
1654	}
1655	else
1656	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1657	soisdisconnected(so);
1658	break;
1659
1660	/*
1661	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1662	*/
1663	case TCPS_TIME_WAIT:
1664	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1665	break;
1666	}
1667	}
1668	#ifdef TCPDEBUG
1669	if (so->so_options & SO_DEBUG)
1670	tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1671	#endif
1672
1673	/*
1674	* Return any desired output.
1675	*/
1676	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1677	(void) tcp_output(tp);
1678	return;
1679
1680	dropafterack:
1681	/*
1682	* Generate an ACK dropping incoming segment if it occupies
1683	* sequence space, where the ACK reflects our state.
1684	*/
1685	if (tiflags & TH_RST)
1686	goto drop;
1687	#ifdef TCPDEBUG
1688	if (so->so_options & SO_DEBUG)
1689	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1690	#endif
1691	m_freem(m);
1692	tp->t_flags \|= TF_ACKNOW;
1693	(void) tcp_output(tp);
1694	return;
1695
1696	dropwithreset:
1697	/*
1698	* Generate a RST, dropping incoming segment.
1699	* Make ACK acceptable to originator of segment.
1700	* Don't bother to respond if destination was broadcast/multicast.
1701	*/
1702	if ((tiflags & TH_RST) \|\| m->m_flags & (M_BCAST\|M_MCAST) \|\|
1703	IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1704	goto drop;
1705	#ifdef TCPDEBUG
1706	if (tp == 0 \|\| (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1707	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1708	#endif
1709	if (tiflags & TH_ACK)
1710	tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1711	else {
1712	if (tiflags & TH_SYN)
1713	ti->ti_len++;
1714	tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1715	TH_RST\|TH_ACK);
1716	}
1717	/* destroy temporarily created socket */
1718	if (dropsocket)
1719	(void) soabort(so);
1720	return;
1721
1722	drop:
1723	/*
1724	* Drop space held by incoming segment and return.
1725	*/
1726	#ifdef TCPDEBUG
1727	if (tp == 0 \|\| (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1728	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1729	#endif
1730	m_freem(m);
1731	/* destroy temporarily created socket */
1732	if (dropsocket)
1733	(void) soabort(so);
1734	return;
1735	#ifndef TUBA_INCLUDE
1736	}
1737
1738	static void
1739	tcp_dooptions(tp, cp, cnt, ti, to)
1740	struct tcpcb *tp;
1741	u_char *cp;
1742	int cnt;
1743	struct tcpiphdr *ti;
1744	struct tcpopt *to;
1745	{
1746	u_short mss = 0;
1747	int opt, optlen;
1748
1749	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1750	opt = cp[0];
1751	if (opt == TCPOPT_EOL)
1752	break;
1753	if (opt == TCPOPT_NOP)
1754	optlen = 1;
1755	else {
1756	optlen = cp[1];
1757	if (optlen <= 0)
1758	break;
1759	}
1760	switch (opt) {
1761
1762	default:
1763	continue;
1764
1765	case TCPOPT_MAXSEG:
1766	if (optlen != TCPOLEN_MAXSEG)
1767	continue;
1768	if (!(ti->ti_flags & TH_SYN))
1769	continue;
1770	bcopy((char ) cp + 2, (char ) &mss, sizeof(mss));
1771	NTOHS(mss);
1772	break;
1773
1774	case TCPOPT_WINDOW:
1775	if (optlen != TCPOLEN_WINDOW)
1776	continue;
1777	if (!(ti->ti_flags & TH_SYN))
1778	continue;
1779	tp->t_flags \|= TF_RCVD_SCALE;
1780	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1781	break;
1782
1783	case TCPOPT_TIMESTAMP:
1784	if (optlen != TCPOLEN_TIMESTAMP)
1785	continue;
1786	to->to_flag \|= TOF_TS;
1787	bcopy((char *)cp + 2,
1788	(char *)&to->to_tsval, sizeof(to->to_tsval));
1789	NTOHL(to->to_tsval);
1790	bcopy((char *)cp + 6,
1791	(char *)&to->to_tsecr, sizeof(to->to_tsecr));
1792	NTOHL(to->to_tsecr);
1793
1794	/*
1795	* A timestamp received in a SYN makes
1796	* it ok to send timestamp requests and replies.
1797	*/
1798	if (ti->ti_flags & TH_SYN) {
1799	tp->t_flags \|= TF_RCVD_TSTMP;
1800	tp->ts_recent = to->to_tsval;
1801	tp->ts_recent_age = tcp_now;
1802	}
1803	break;
1804	case TCPOPT_CC:
1805	if (optlen != TCPOLEN_CC)
1806	continue;
1807	to->to_flag \|= TOF_CC;
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	if (ti->ti_flags & TH_SYN)
1816	tp->t_flags \|= TF_RCVD_CC;
1817	break;
1818	case TCPOPT_CCNEW:
1819	if (optlen != TCPOLEN_CC)
1820	continue;
1821	if (!(ti->ti_flags & TH_SYN))
1822	continue;
1823	to->to_flag \|= TOF_CCNEW;
1824	bcopy((char *)cp + 2,
1825	(char *)&to->to_cc, sizeof(to->to_cc));
1826	NTOHL(to->to_cc);
1827	/*
1828	* A CC or CC.new option received in a SYN makes
1829	* it ok to send CC in subsequent segments.
1830	*/
1831	tp->t_flags \|= TF_RCVD_CC;
1832	break;
1833	case TCPOPT_CCECHO:
1834	if (optlen != TCPOLEN_CC)
1835	continue;
1836	if (!(ti->ti_flags & TH_SYN))
1837	continue;
1838	to->to_flag \|= TOF_CCECHO;
1839	bcopy((char *)cp + 2,
1840	(char *)&to->to_ccecho, sizeof(to->to_ccecho));
1841	NTOHL(to->to_ccecho);
1842	break;
1843	}
1844	}
1845	if (ti->ti_flags & TH_SYN)
1846	tcp_mss(tp, mss); /* sets t_maxseg */
1847	}
1848
1849	/*
1850	* Pull out of band byte out of a segment so
1851	* it doesn't appear in the user's data queue.
1852	* It is still reflected in the segment length for
1853	* sequencing purposes.
1854	*/
1855	static void
1856	tcp_pulloutofband(so, ti, m)
1857	struct socket *so;
1858	struct tcpiphdr *ti;
1859	register struct mbuf *m;
1860	{
1861	int cnt = ti->ti_urp - 1;
1862
1863	while (cnt >= 0) {
1864	if (m->m_len > cnt) {
1865	char *cp = mtod(m, caddr_t) + cnt;
1866	struct tcpcb *tp = sototcpcb(so);
1867
1868	tp->t_iobc = *cp;
1869	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1870	bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1871	m->m_len--;
1872	return;
1873	}
1874	cnt -= m->m_len;
1875	m = m->m_next;
1876	if (m == 0)
1877	break;
1878	}
1879	panic("tcp_pulloutofband");
1880	}
1881
1882	/*
1883	* Collect new round-trip time estimate
1884	* and update averages and current timeout.
1885	*/
1886	static void
1887	tcp_xmit_timer(tp, rtt)
1888	register struct tcpcb *tp;
1889	short rtt;
1890	{
1891	register int delta;
1892
1893	tcpstat.tcps_rttupdated++;
1894	tp->t_rttupdated++;
1895	if (tp->t_srtt != 0) {
1896	/*
1897	* srtt is stored as fixed point with 5 bits after the
1898	* binary point (i.e., scaled by 8). The following magic
1899	* is equivalent to the smoothing algorithm in rfc793 with
1900	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1901	* point). Adjust rtt to origin 0.
1902	*/
1903	delta = ((rtt - 1) << TCP_DELTA_SHIFT)
1904	- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
1905
1906	if ((tp->t_srtt += delta) <= 0)
1907	tp->t_srtt = 1;
1908
1909	/*
1910	* We accumulate a smoothed rtt variance (actually, a
1911	* smoothed mean difference), then set the retransmit
1912	* timer to smoothed rtt + 4 times the smoothed variance.
1913	* rttvar is stored as fixed point with 4 bits after the
1914	* binary point (scaled by 16). The following is
1915	* equivalent to rfc793 smoothing with an alpha of .75
1916	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1917	* rfc793's wired-in beta.
1918	*/
1919	if (delta < 0)
1920	delta = -delta;
1921	delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
1922	if ((tp->t_rttvar += delta) <= 0)
1923	tp->t_rttvar = 1;
1924	} else {
1925	/*
1926	* No rtt measurement yet - use the unsmoothed rtt.
1927	* Set the variance to half the rtt (so our first
1928	* retransmit happens at 3*rtt).
1929	*/
1930	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1931	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1932	}
1933	tp->t_rtt = 0;
1934	tp->t_rxtshift = 0;
1935
1936	/*
1937	* the retransmit should happen at rtt + 4 * rttvar.
1938	* Because of the way we do the smoothing, srtt and rttvar
1939	* will each average +1/2 tick of bias. When we compute
1940	* the retransmit timer, we want 1/2 tick of rounding and
1941	* 1 extra tick because of +-1/2 tick uncertainty in the
1942	* firing of the timer. The bias will give us exactly the
1943	* 1.5 tick we need. But, because the bias is
1944	* statistical, we have to test that we don't drop below
1945	* the minimum feasible timer (which is 2 ticks).
1946	*/
1947	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1948	max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
1949
1950	/*
1951	* We received an ack for a packet that wasn't retransmitted;
1952	* it is probably safe to discard any error indications we've
1953	* received recently. This isn't quite right, but close enough
1954	* for now (a route might have failed after we sent a segment,
1955	* and the return path might not be symmetrical).
1956	*/
1957	tp->t_softerror = 0;
1958	}
1959
1960	/*
1961	* Determine a reasonable value for maxseg size.
1962	* If the route is known, check route for mtu.
1963	* If none, use an mss that can be handled on the outgoing
1964	* interface without forcing IP to fragment; if bigger than
1965	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1966	* to utilize large mbufs. If no route is found, route has no mtu,
1967	* or the destination isn't local, use a default, hopefully conservative
1968	* size (usually 512 or the default IP max size, but no more than the mtu
1969	* of the interface), as we can't discover anything about intervening
1970	* gateways or networks. We also initialize the congestion/slow start
1971	* window to be a single segment if the destination isn't local.
1972	* While looking at the routing entry, we also initialize other path-dependent
1973	* parameters from pre-set or cached values in the routing entry.
1974	*
1975	* Also take into account the space needed for options that we
1976	* send regularly. Make maxseg shorter by that amount to assure
1977	* that we can send maxseg amount of data even when the options
1978	* are present. Store the upper limit of the length of options plus
1979	* data in maxopd.
1980	*
1981	* NOTE that this routine is only called when we process an incoming
1982	* segment, for outgoing segments only tcp_mssopt is called.
1983	*
1984	* In case of T/TCP, we call this routine during implicit connection
1985	* setup as well (offer = -1), to initialize maxseg from the cached
1986	* MSS of our peer.
1987	*/
1988	void
1989	tcp_mss(tp, offer)
1990	struct tcpcb *tp;
1991	int offer;
1992	{
1993	register struct rtentry *rt;
1994	struct ifnet *ifp;
1995	register int rtt, mss;
1996	u_long bufsize;
1997	struct inpcb *inp;
1998	struct socket *so;
1999	struct rmxp_tao *taop;
2000	int origoffer = offer;
2001
2002	inp = tp->t_inpcb;
2003	if ((rt = tcp_rtlookup(inp)) == NULL) {
2004	tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2005	return;
2006	}
2007	ifp = rt->rt_ifp;
2008	so = inp->inp_socket;
2009
2010	taop = rmx_taop(rt->rt_rmx);
2011	/*
2012	* Offer == -1 means that we didn't receive SYN yet,
2013	* use cached value in that case;
2014	*/
2015	if (offer == -1)
2016	offer = taop->tao_mssopt;
2017	/*
2018	* Offer == 0 means that there was no MSS on the SYN segment,
2019	* in this case we use tcp_mssdflt.
2020	*/
2021	if (offer == 0)
2022	offer = tcp_mssdflt;
2023	else
2024	/*
2025	* Sanity check: make sure that maxopd will be large
2026	* enough to allow some data on segments even is the
2027	* all the option space is used (40bytes). Otherwise
2028	* funny things may happen in tcp_output.
2029	*/
2030	offer = max(offer, 64);
2031	taop->tao_mssopt = offer;
2032
2033	/*
2034	* While we're here, check if there's an initial rtt
2035	* or rttvar. Convert from the route-table units
2036	* to scaled multiples of the slow timeout timer.
2037	*/
2038	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2039	/*
2040	* XXX the lock bit for RTT indicates that the value
2041	* is also a minimum value; this is subject to time.
2042	*/
2043	if (rt->rt_rmx.rmx_locks & RTV_RTT)
2044	tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
2045	tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
2046	tcpstat.tcps_usedrtt++;
2047	if (rt->rt_rmx.rmx_rttvar) {
2048	tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2049	(RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
2050	tcpstat.tcps_usedrttvar++;
2051	} else {
2052	/* default variation is +- 1 rtt */
2053	tp->t_rttvar =
2054	tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2055	}
2056	TCPT_RANGESET(tp->t_rxtcur,
2057	((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2058	tp->t_rttmin, TCPTV_REXMTMAX);
2059	}
2060	/*
2061	* if there's an mtu associated with the route, use it
2062	*/
2063	if (rt->rt_rmx.rmx_mtu)
2064	mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
2065	else
2066	{
2067	mss = ifp->if_mtu - sizeof(struct tcpiphdr);
2068	if (!in_localaddr(inp->inp_faddr))
2069	mss = min(mss, tcp_mssdflt);
2070	}
2071	mss = min(mss, offer);
2072	/*
2073	* maxopd stores the maximum length of data AND options
2074	* in a segment; maxseg is the amount of data in a normal
2075	* segment. We need to store this value (maxopd) apart
2076	* from maxseg, because now every segment carries options
2077	* and thus we normally have somewhat less data in segments.
2078	*/
2079	tp->t_maxopd = mss;
2080
2081	/*
2082	* In case of T/TCP, origoffer==-1 indicates, that no segments
2083	* were received yet. In this case we just guess, otherwise
2084	* we do the same as before T/TCP.
2085	*/
2086	if ((tp->t_flags & (TF_REQ_TSTMP\|TF_NOOPT)) == TF_REQ_TSTMP &&
2087	(origoffer == -1 \|\|
2088	(tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2089	mss -= TCPOLEN_TSTAMP_APPA;
2090	if ((tp->t_flags & (TF_REQ_CC\|TF_NOOPT)) == TF_REQ_CC &&
2091	(origoffer == -1 \|\|
2092	(tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2093	mss -= TCPOLEN_CC_APPA;
2094
2095	#if (MCLBYTES & (MCLBYTES - 1)) == 0
2096	if (mss > MCLBYTES)
2097	mss &= ~(MCLBYTES-1);
2098	#else
2099	if (mss > MCLBYTES)
2100	mss = mss / MCLBYTES * MCLBYTES;
2101	#endif
2102	/*
2103	* If there's a pipesize, change the socket buffer
2104	* to that size. Make the socket buffers an integral
2105	* number of mss units; if the mss is larger than
2106	* the socket buffer, decrease the mss.
2107	*/
2108	#ifdef RTV_SPIPE
2109	if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2110	#endif
2111	bufsize = so->so_snd.sb_hiwat;
2112	if (bufsize < mss)
2113	mss = bufsize;
2114	else {
2115	bufsize = roundup(bufsize, mss);
2116	if (bufsize > sb_max)
2117	bufsize = sb_max;
2118	(void)sbreserve(&so->so_snd, bufsize);
2119	}
2120	tp->t_maxseg = mss;
2121
2122	#ifdef RTV_RPIPE
2123	if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2124	#endif
2125	bufsize = so->so_rcv.sb_hiwat;
2126	if (bufsize > mss) {
2127	bufsize = roundup(bufsize, mss);
2128	if (bufsize > sb_max)
2129	bufsize = sb_max;
2130	(void)sbreserve(&so->so_rcv, bufsize);
2131	}
2132	/*
2133	* Don't force slow-start on local network.
2134	*/
2135	if (!in_localaddr(inp->inp_faddr))
2136	tp->snd_cwnd = mss;
2137
2138	if (rt->rt_rmx.rmx_ssthresh) {
2139	/*
2140	* There's some sort of gateway or interface
2141	* buffer limit on the path. Use this to set
2142	* the slow start threshhold, but set the
2143	* threshold to no less than 2*mss.
2144	*/
2145	tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2146	tcpstat.tcps_usedssthresh++;
2147	}
2148	}
2149
2150	/*
2151	* Determine the MSS option to send on an outgoing SYN.
2152	*/
2153	int
2154	tcp_mssopt(tp)
2155	struct tcpcb *tp;
2156	{
2157	struct rtentry *rt;
2158
2159	rt = tcp_rtlookup(tp->t_inpcb);
2160	if (rt == NULL)
2161	return tcp_mssdflt;
2162
2163	return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr);
2164	}
2165	#endif /* TUBA_INCLUDE */

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

Download in other formats: