Context Navigation

tcp_input.c @ 0e16fa45

Last change on this file since 0e16fa45 was cb68253, checked in by Sebastian Huber <sebastian.huber@…>, on 09/07/18 at 04:19:02

network: Use kernel/user space header files

Add and use <machine/rtems-bsd-kernel-space.h> and
<machine/rtems-bsd-user-space.h> similar to the libbsd to avoid command
line defines and defines scattered throught the code base.

Simplify cpukit/libnetworking/Makefile.am.

Update #3375.

Property mode set to 100644

File size: 60.2 KB

Line
1	#include <machine/rtems-bsd-kernel-space.h>
2
3	/*
4	* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5	* The Regents of the University of California. All rights reserved.
6	*
7	* Redistribution and use in source and binary forms, with or without
8	* modification, are permitted provided that the following conditions
9	* are met:
10	* 1. Redistributions of source code must retain the above copyright
11	* notice, this list of conditions and the following disclaimer.
12	* 2. Redistributions in binary form must reproduce the above copyright
13	* notice, this list of conditions and the following disclaimer in the
14	* documentation and/or other materials provided with the distribution.
15	* 4. Neither the name of the University nor the names of its contributors
16	* may be used to endorse or promote products derived from this software
17	* without specific prior written permission.
18	*
19	* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22	* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29	* SUCH DAMAGE.
30	*
31	* @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
32	*/
33
34	#ifdef HAVE_CONFIG_H
35	#include "config.h"
36	#endif
37
38	#include "opt_tcpdebug.h"
39
40	#ifndef TUBA_INCLUDE
41	#include <sys/param.h>
42	#include <sys/queue.h>
43	#include <sys/systm.h>
44	#include <sys/kernel.h>
45	#include <sys/sysctl.h>
46	#include <sys/malloc.h>
47	#include <sys/mbuf.h>
48	#include <sys/protosw.h>
49	#include <sys/socket.h>
50	#include <sys/socketvar.h>
51	#include <errno.h>
52	#include <sys/syslog.h>
53
54	#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
55
56	#include <net/if.h>
57	#include <net/if_var.h>
58	#include <net/route.h>
59
60	#include <netinet/in.h>
61	#include <rtems/rtems_netinet_in.h>
62	#include <netinet/in_systm.h>
63	#include <netinet/ip.h>
64	#include <netinet/in_pcb.h>
65	#include <netinet/ip_var.h>
66	#include <netinet/tcp.h>
67	#include <netinet/tcp_fsm.h>
68	#include <netinet/tcp_seq.h>
69	#include <netinet/tcp_timer.h>
70	#include <netinet/tcp_var.h>
71	#include <netinet/tcpip.h>
72	#ifdef TCPDEBUG
73	#include <netinet/tcp_debug.h>
74	static struct tcpiphdr tcp_saveti;
75	#endif
76
77	static int tcprexmtthresh = 3;
78	tcp_seq tcp_iss;
79	tcp_cc tcp_ccgen;
80
81	struct tcpstat tcpstat;
82	SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats,
83	CTLFLAG_RD, &tcpstat , tcpstat, "");
84
85	static int log_in_vain = 0;
86	SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
87	&log_in_vain, 0, "");
88
89	u_long tcp_now;
90	struct inpcbhead tcb;
91	struct inpcbinfo tcbinfo;
92
93	static void tcp_dooptions(struct tcpcb *,
94	u_char , int, struct tcpiphdr , struct tcpopt *);
95	static void tcp_pulloutofband(struct socket *,
96	struct tcpiphdr , struct mbuf );
97	static int tcp_reass(struct tcpcb , struct tcpiphdr , struct mbuf *);
98	static void tcp_xmit_timer(struct tcpcb *, int);
99
100	#endif /* TUBA_INCLUDE */
101
102	/*
103	* Insert segment ti into reassembly queue of tcp with
104	* control block tp. Return TH_FIN if reassembly now includes
105	* a segment with FIN. The macro form does the common case inline
106	* (segment is the next to be received on an established connection,
107	* and the queue is empty), avoiding linkage into and removal
108	* from the queue and repetition of various conversions.
109	* Set DELACK for segments received in order, but ack immediately
110	* when segments are out of order (so fast retransmit can work).
111	*/
112	#ifdef TCP_ACK_HACK
113	#define TCP_REASS(tp, ti, m, so, flags) { \
114	if ((ti)->ti_seq == (tp)->rcv_nxt && \
115	(tp)->seg_next == (struct tcpiphdr *)(tp) && \
116	(tp)->t_state == TCPS_ESTABLISHED) { \
117	if (ti->ti_flags & TH_PUSH) \
118	tp->t_flags \|= TF_ACKNOW; \
119	else \
120	tp->t_flags \|= TF_DELACK; \
121	(tp)->rcv_nxt += (ti)->ti_len; \
122	flags = (ti)->ti_flags & TH_FIN; \
123	tcpstat.tcps_rcvpack++;\
124	tcpstat.tcps_rcvbyte += (ti)->ti_len;\
125	sbappend(&(so)->so_rcv, (m)); \
126	sorwakeup(so); \
127	} else { \
128	(flags) = tcp_reass((tp), (ti), (m)); \
129	tp->t_flags \|= TF_ACKNOW; \
130	} \
131	}
132	#else
133	#define TCP_REASS(tp, ti, m, so, flags) { \
134	if ((ti)->ti_seq == (tp)->rcv_nxt && \
135	(tp)->seg_next == (struct tcpiphdr *)(tp) && \
136	(tp)->t_state == TCPS_ESTABLISHED) { \
137	tp->t_flags \|= TF_DELACK; \
138	(tp)->rcv_nxt += (ti)->ti_len; \
139	flags = (ti)->ti_flags & TH_FIN; \
140	tcpstat.tcps_rcvpack++;\
141	tcpstat.tcps_rcvbyte += (ti)->ti_len;\
142	sbappend(&(so)->so_rcv, (m)); \
143	sorwakeup(so); \
144	} else { \
145	(flags) = tcp_reass((tp), (ti), (m)); \
146	tp->t_flags \|= TF_ACKNOW; \
147	} \
148	}
149	#endif
150	#ifndef TUBA_INCLUDE
151
152	static int
153	tcp_reass(struct tcpcb tp, struct tcpiphdr ti, 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__) \|\| defined(__mips__)))
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__) \|\| defined(__mips__)))
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__) \|\| defined(__mips__)))
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__) \|\| defined(__mips__)))
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(struct mbuf *m, int iphlen)
278	{
279	register struct tcpiphdr *ti;
280	register struct inpcb *inp;
281	u_char *optp = NULL;
282	int optlen = 0;
283	int len, tlen, off;
284	register struct tcpcb *tp = 0;
285	register int tiflags;
286	struct socket *so = 0;
287	int todrop, acked, ourfinisacked, needoutput = 0;
288	struct in_addr laddr;
289	int dropsocket = 0;
290	int iss = 0;
291	u_long tiwin;
292	struct tcpopt to; /* options in this segment */
293	struct rmxp_tao taop; / pointer to our TAO cache entry */
294	struct rmxp_tao tao_noncached; /* in case there's no cached entry */
295	#ifdef TCPDEBUG
296	short ostate = 0;
297	#endif
298
299	bzero((char *)&to, sizeof(to));
300
301	tcpstat.tcps_rcvtotal++;
302	/*
303	* Get IP and TCP header together in first mbuf.
304	* Note: IP leaves IP header in first mbuf.
305	*/
306	ti = mtod(m, struct tcpiphdr *);
307	if (iphlen > sizeof (struct ip))
308	ip_stripoptions(m, (struct mbuf *)0);
309	if (m->m_len < sizeof (struct tcpiphdr)) {
310	if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
311	tcpstat.tcps_rcvshort++;
312	return;
313	}
314	ti = mtod(m, struct tcpiphdr *);
315	}
316
317	/*
318	* Checksum extended TCP header and data.
319	*/
320	tlen = ((struct ip *)ti)->ip_len;
321	len = sizeof (struct ip) + tlen;
322	ti->ti_next = ti->ti_prev = 0;
323	ti->ti_x1 = 0;
324	ti->ti_len = (u_short)tlen;
325	HTONS(ti->ti_len);
326	ti->ti_sum = in_cksum(m, len);
327	if (ti->ti_sum) {
328	tcpstat.tcps_rcvbadsum++;
329	goto drop;
330	}
331	#endif /* TUBA_INCLUDE */
332
333	/*
334	* Check that TCP offset makes sense,
335	* pull out TCP options and adjust length. XXX
336	*/
337	off = ti->ti_off << 2;
338	if (off < sizeof (struct tcphdr) \|\| off > tlen) {
339	tcpstat.tcps_rcvbadoff++;
340	goto drop;
341	}
342	tlen -= off;
343	ti->ti_len = tlen;
344	if (off > sizeof (struct tcphdr)) {
345	if (m->m_len < sizeof(struct ip) + off) {
346	if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
347	tcpstat.tcps_rcvshort++;
348	return;
349	}
350	ti = mtod(m, struct tcpiphdr *);
351	}
352	optlen = off - sizeof (struct tcphdr);
353	optp = mtod(m, u_char *) + sizeof (struct tcpiphdr);
354	}
355	tiflags = ti->ti_flags;
356
357	/*
358	* Convert TCP protocol specific fields to host format.
359	*/
360	NTOHL(ti->ti_seq);
361	NTOHL(ti->ti_ack);
362	NTOHS(ti->ti_win);
363	NTOHS(ti->ti_urp);
364
365	/*
366	* Drop TCP, IP headers and TCP options.
367	*/
368	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
369	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
370
371	/*
372	* Locate pcb for segment.
373	*/
374	findpcb:
375	inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport,
376	ti->ti_dst, ti->ti_dport, 1);
377
378	/*
379	* If the state is CLOSED (i.e., TCB does not exist) then
380	* all data in the incoming segment is discarded.
381	* If the TCB exists but is in CLOSED state, it is embryonic,
382	* but should either do a listen or a connect soon.
383	*/
384	if (inp == NULL) {
385	if (log_in_vain && tiflags & TH_SYN) {
386	char buf0[INET_ADDRSTRLEN];
387	char buf1[INET_ADDRSTRLEN];
388
389	log(LOG_INFO, "Connection attempt to TCP %s:%d"
390	" from %s:%d\n",
391	inet_ntoa_r(ti->ti_dst, buf0), ntohs(ti->ti_dport),
392	inet_ntoa_r(ti->ti_src, buf1), ntohs(ti->ti_sport));
393	}
394	goto dropwithreset;
395	}
396	tp = intotcpcb(inp);
397	if (tp == 0)
398	goto dropwithreset;
399	if (tp->t_state == TCPS_CLOSED)
400	goto drop;
401
402	/* Unscale the window into a 32-bit value. */
403	if ((tiflags & TH_SYN) == 0)
404	tiwin = ti->ti_win << tp->snd_scale;
405	else
406	tiwin = ti->ti_win;
407
408	so = inp->inp_socket;
409	if (so->so_options & (SO_DEBUG\|SO_ACCEPTCONN)) {
410	#ifdef TCPDEBUG
411	if (so->so_options & SO_DEBUG) {
412	ostate = tp->t_state;
413	tcp_saveti = *ti;
414	}
415	#endif
416	if (so->so_options & SO_ACCEPTCONN) {
417	register struct tcpcb *tp0 = tp;
418	struct socket *so2;
419	if ((tiflags & (TH_RST\|TH_ACK\|TH_SYN)) != TH_SYN) {
420	/*
421	* Note: dropwithreset makes sure we don't
422	* send a RST in response to a RST.
423	*/
424	if (tiflags & TH_ACK) {
425	tcpstat.tcps_badsyn++;
426	goto dropwithreset;
427	}
428	goto drop;
429	}
430	so2 = sonewconn(so, 0);
431	if (so2 == 0) {
432	tcpstat.tcps_listendrop++;
433	so2 = sodropablereq(so);
434	if (so2) {
435	tcp_drop(sototcpcb(so2), ETIMEDOUT);
436	so2 = sonewconn(so, 0);
437	}
438	if (!so2)
439	goto drop;
440	}
441	so = so2;
442	/*
443	* This is ugly, but ....
444	*
445	* Mark socket as temporary until we're
446	* committed to keeping it. The code at
447	* ``drop'' and ``dropwithreset'' check the
448	* flag dropsocket to see if the temporary
449	* socket created here should be discarded.
450	* We mark the socket as discardable until
451	* we're committed to it below in TCPS_LISTEN.
452	*/
453	dropsocket++;
454	inp = (struct inpcb *)so->so_pcb;
455	inp->inp_laddr = ti->ti_dst;
456	inp->inp_lport = ti->ti_dport;
457	in_pcbrehash(inp);
458	#if BSD>=43
459	inp->inp_options = ip_srcroute();
460	#endif
461	tp = intotcpcb(inp);
462	tp->t_state = TCPS_LISTEN;
463	tp->t_flags \|= tp0->t_flags & (TF_NOPUSH\|TF_NOOPT);
464
465	/* Compute proper scaling value from buffer space */
466	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
467	TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
468	tp->request_r_scale++;
469	}
470	}
471
472	/*
473	* Segment received on connection.
474	* Reset idle time and keep-alive timer.
475	*/
476	tp->t_idle = 0;
477	if (TCPS_HAVEESTABLISHED(tp->t_state))
478	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
479
480	/*
481	* Process options if not in LISTEN state,
482	* else do it below (after getting remote address).
483	*/
484	if (tp->t_state != TCPS_LISTEN)
485	tcp_dooptions(tp, optp, optlen, ti, &to);
486
487	/*
488	* Header prediction: check for the two common cases
489	* of a uni-directional data xfer. If the packet has
490	* no control flags, is in-sequence, the window didn't
491	* change and we're not retransmitting, it's a
492	* candidate. If the length is zero and the ack moved
493	* forward, we're the sender side of the xfer. Just
494	* free the data acked & wake any higher level process
495	* that was blocked waiting for space. If the length
496	* is non-zero and the ack didn't move, we're the
497	* receiver side. If we're getting packets in-order
498	* (the reassembly queue is empty), add the data to
499	* the socket buffer and note that we need a delayed ack.
500	* Make sure that the hidden state-flags are also off.
501	* Since we check for TCPS_ESTABLISHED above, it can only
502	* be TH_NEEDSYN.
503	*/
504	if (tp->t_state == TCPS_ESTABLISHED &&
505	(tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK &&
506	((tp->t_flags & (TF_NEEDSYN\|TF_NEEDFIN)) == 0) &&
507	((to.to_flags & TOF_TS) == 0 \|\|
508	TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
509	/*
510	* Using the CC option is compulsory if once started:
511	* the segment is OK if no T/TCP was negotiated or
512	* if the segment has a CC option equal to CCrecv
513	*/
514	((tp->t_flags & (TF_REQ_CC\|TF_RCVD_CC)) != (TF_REQ_CC\|TF_RCVD_CC) \|\|
515	((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
516	ti->ti_seq == tp->rcv_nxt &&
517	tiwin && tiwin == tp->snd_wnd &&
518	tp->snd_nxt == tp->snd_max) {
519
520	/*
521	* If last ACK falls within this segment's sequence numbers,
522	* record the timestamp.
523	* NOTE that the test is modified according to the latest
524	* proposal of the tcplw@cray.com list (Braden 1993/04/26).
525	*/
526	if ((to.to_flags & TOF_TS) != 0 &&
527	SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
528	tp->ts_recent_age = tcp_now;
529	tp->ts_recent = to.to_tsval;
530	}
531
532	if (ti->ti_len == 0) {
533	if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
534	SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
535	tp->snd_cwnd >= tp->snd_wnd &&
536	tp->t_dupacks < tcprexmtthresh) {
537	/*
538	* this is a pure ack for outstanding data.
539	*/
540	++tcpstat.tcps_predack;
541	if ((to.to_flags & TOF_TS) != 0)
542	tcp_xmit_timer(tp,
543	tcp_now - to.to_tsecr + 1);
544	else if (tp->t_rtt &&
545	SEQ_GT(ti->ti_ack, tp->t_rtseq))
546	tcp_xmit_timer(tp, tp->t_rtt);
547	acked = ti->ti_ack - tp->snd_una;
548	tcpstat.tcps_rcvackpack++;
549	tcpstat.tcps_rcvackbyte += acked;
550	sbdrop(&so->so_snd, acked);
551	tp->snd_una = ti->ti_ack;
552	m_freem(m);
553
554	/*
555	* If all outstanding data are acked, stop
556	* retransmit timer, otherwise restart timer
557	* using current (possibly backed-off) value.
558	* If process is waiting for space,
559	* wakeup/selwakeup/signal. If data
560	* are ready to send, let tcp_output
561	* decide between more output or persist.
562	*/
563	if (tp->snd_una == tp->snd_max)
564	tp->t_timer[TCPT_REXMT] = 0;
565	else if (tp->t_timer[TCPT_PERSIST] == 0)
566	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
567
568	if (so->so_snd.sb_flags & SB_NOTIFY)
569	sowwakeup(so);
570	if (so->so_snd.sb_cc)
571	(void) tcp_output(tp);
572	return;
573	}
574	} else if (ti->ti_ack == tp->snd_una &&
575	tp->seg_next == (struct tcpiphdr *)tp &&
576	ti->ti_len <= sbspace(&so->so_rcv)) {
577	/*
578	* this is a pure, in-sequence data packet
579	* with nothing on the reassembly queue and
580	* we have enough buffer space to take it.
581	*/
582	++tcpstat.tcps_preddat;
583	tp->rcv_nxt += ti->ti_len;
584	tcpstat.tcps_rcvpack++;
585	tcpstat.tcps_rcvbyte += ti->ti_len;
586	/*
587	* Add data to socket buffer.
588	*/
589	sbappend(&so->so_rcv, m);
590	sorwakeup(so);
591	#ifdef TCP_ACK_HACK
592	/*
593	* If this is a short packet, then ACK now - with Nagel
594	* congestion avoidance sender won't send more until
595	* he gets an ACK.
596	*/
597	if (tiflags & TH_PUSH) {
598	tp->t_flags \|= TF_ACKNOW;
599	tcp_output(tp);
600	} else {
601	tp->t_flags \|= TF_DELACK;
602	}
603	#else
604	tp->t_flags \|= TF_DELACK;
605	#endif
606	return;
607	}
608	}
609
610	/*
611	* Calculate amount of space in receive window,
612	* and then do TCP input processing.
613	* Receive window is amount of space in rcv queue,
614	* but not less than advertised window.
615	*/
616	{ int win;
617
618	win = sbspace(&so->so_rcv);
619	if (win < 0)
620	win = 0;
621	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
622	}
623
624	switch (tp->t_state) {
625
626	/*
627	* If the state is LISTEN then ignore segment if it contains an RST.
628	* If the segment contains an ACK then it is bad and send a RST.
629	* If it does not contain a SYN then it is not interesting; drop it.
630	* If it is from this socket, drop it, it must be forged.
631	* Don't bother responding if the destination was a broadcast.
632	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
633	* tp->iss, and send a segment:
634	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
635	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
636	* Fill in remote peer address fields if not previously specified.
637	* Enter SYN_RECEIVED state, and process any other fields of this
638	* segment in this state.
639	*/
640	case TCPS_LISTEN: {
641	struct mbuf *am;
642	register struct sockaddr_in *sin;
643
644	if (tiflags & TH_RST)
645	goto drop;
646	if (tiflags & TH_ACK)
647	goto dropwithreset;
648	if ((tiflags & TH_SYN) == 0)
649	goto drop;
650	if ((ti->ti_dport == ti->ti_sport) &&
651	(ti->ti_dst.s_addr == ti->ti_src.s_addr))
652	goto drop;
653	/*
654	* RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
655	* in_broadcast() should never return true on a received
656	* packet with M_BCAST not set.
657	*/
658	if (m->m_flags & (M_BCAST\|M_MCAST) \|\|
659	IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
660	goto drop;
661	am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */
662	if (am == NULL)
663	goto drop;
664	am->m_len = sizeof (struct sockaddr_in);
665	sin = mtod(am, struct sockaddr_in *);
666	sin->sin_family = AF_INET;
667	sin->sin_len = sizeof(*sin);
668	sin->sin_addr = ti->ti_src;
669	sin->sin_port = ti->ti_sport;
670	bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
671	laddr = inp->inp_laddr;
672	if (inp->inp_laddr.s_addr == INADDR_ANY)
673	inp->inp_laddr = ti->ti_dst;
674	if (in_pcbconnect(inp, am)) {
675	inp->inp_laddr = laddr;
676	(void) m_free(am);
677	goto drop;
678	}
679	(void) m_free(am);
680	tp->t_template = tcp_template(tp);
681	if (tp->t_template == 0) {
682	tp = tcp_drop(tp, ENOBUFS);
683	dropsocket = 0; /* socket is already gone */
684	goto drop;
685	}
686	if ((taop = tcp_gettaocache(inp)) == NULL) {
687	taop = &tao_noncached;
688	bzero(taop, sizeof(*taop));
689	}
690	tcp_dooptions(tp, optp, optlen, ti, &to);
691	if (iss)
692	tp->iss = iss;
693	else
694	tp->iss = tcp_iss;
695	tcp_iss += TCP_ISSINCR/4;
696	tp->irs = ti->ti_seq;
697	tcp_sendseqinit(tp);
698	tcp_rcvseqinit(tp);
699	/*
700	* Initialization of the tcpcb for transaction;
701	* set SND.WND = SEG.WND,
702	* initialize CCsend and CCrecv.
703	*/
704	tp->snd_wnd = tiwin; /* initial send-window */
705	tp->cc_send = CC_INC(tcp_ccgen);
706	tp->cc_recv = to.to_cc;
707	/*
708	* Perform TAO test on incoming CC (SEG.CC) option, if any.
709	* - compare SEG.CC against cached CC from the same host,
710	* if any.
711	* - if SEG.CC > chached value, SYN must be new and is accepted
712	* immediately: save new CC in the cache, mark the socket
713	* connected, enter ESTABLISHED state, turn on flag to
714	* send a SYN in the next segment.
715	* A virtual advertised window is set in rcv_adv to
716	* initialize SWS prevention. Then enter normal segment
717	* processing: drop SYN, process data and FIN.
718	* - otherwise do a normal 3-way handshake.
719	*/
720	if ((to.to_flags & TOF_CC) != 0) {
721	if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
722	taop->tao_cc = to.to_cc;
723	tp->t_state = TCPS_ESTABLISHED;
724
725	/*
726	* If there is a FIN, or if there is data and the
727	* connection is local, then delay SYN,ACK(SYN) in
728	* the hope of piggy-backing it on a response
729	* segment. Otherwise must send ACK now in case
730	* the other side is slow starting.
731	*/
732	if ((tiflags & TH_FIN) \|\| (ti->ti_len != 0 &&
733	in_localaddr(inp->inp_faddr)))
734	tp->t_flags \|= (TF_DELACK \| TF_NEEDSYN);
735	else
736	tp->t_flags \|= (TF_ACKNOW \| TF_NEEDSYN);
737
738	/*
739	* Limit the `virtual advertised window' to TCP_MAXWIN
740	* here. Even if we requested window scaling, it will
741	* become effective only later when our SYN is acked.
742	*/
743	tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
744	tcpstat.tcps_connects++;
745	soisconnected(so);
746	tp->t_timer[TCPT_KEEP] = tcp_keepinit;
747	dropsocket = 0; /* committed to socket */
748	tcpstat.tcps_accepts++;
749	goto trimthenstep6;
750	}
751	/* else do standard 3-way handshake */
752	} else {
753	/*
754	* No CC option, but maybe CC.NEW:
755	* invalidate cached value.
756	*/
757	taop->tao_cc = 0;
758	}
759	/*
760	* TAO test failed or there was no CC option,
761	* do a standard 3-way handshake.
762	*/
763	tp->t_flags \|= TF_ACKNOW;
764	tp->t_state = TCPS_SYN_RECEIVED;
765	tp->t_timer[TCPT_KEEP] = tcp_keepinit;
766	dropsocket = 0; /* committed to socket */
767	tcpstat.tcps_accepts++;
768	goto trimthenstep6;
769	}
770
771	/*
772	* If the state is SYN_RECEIVED:
773	* if seg contains SYN/ACK, send a RST.
774	* if seg contains an ACK, but not for our SYN/ACK, send a RST.
775	*/
776	case TCPS_SYN_RECEIVED:
777	if (tiflags & TH_ACK) {
778	if (tiflags & TH_SYN) {
779	tcpstat.tcps_badsyn++;
780	goto dropwithreset;
781	}
782	if (SEQ_LEQ(ti->ti_ack, tp->snd_una) \|\|
783	SEQ_GT(ti->ti_ack, tp->snd_max))
784	goto dropwithreset;
785	}
786	break;
787
788	/*
789	* If the state is SYN_SENT:
790	* if seg contains an ACK, but not for our SYN, drop the input.
791	* if seg contains a RST, then drop the connection.
792	* if seg does not contain SYN, then drop it.
793	* Otherwise this is an acceptable SYN segment
794	* initialize tp->rcv_nxt and tp->irs
795	* if seg contains ack then advance tp->snd_una
796	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
797	* arrange for segment to be acked (eventually)
798	* continue processing rest of data/controls, beginning with URG
799	*/
800	case TCPS_SYN_SENT:
801	if ((taop = tcp_gettaocache(inp)) == NULL) {
802	taop = &tao_noncached;
803	bzero(taop, sizeof(*taop));
804	}
805
806	if ((tiflags & TH_ACK) &&
807	(SEQ_LEQ(ti->ti_ack, tp->iss) \|\|
808	SEQ_GT(ti->ti_ack, tp->snd_max))) {
809	/*
810	* If we have a cached CCsent for the remote host,
811	* hence we haven't just crashed and restarted,
812	* do not send a RST. This may be a retransmission
813	* from the other side after our earlier ACK was lost.
814	* Our new SYN, when it arrives, will serve as the
815	* needed ACK.
816	*/
817	if (taop->tao_ccsent != 0)
818	goto drop;
819	else
820	goto dropwithreset;
821	}
822	if (tiflags & TH_RST) {
823	if (tiflags & TH_ACK)
824	tp = tcp_drop(tp, ECONNREFUSED);
825	goto drop;
826	}
827	if ((tiflags & TH_SYN) == 0)
828	goto drop;
829	tp->snd_wnd = ti->ti_win; /* initial send window */
830	tp->cc_recv = to.to_cc; /* foreign CC */
831
832	tp->irs = ti->ti_seq;
833	tcp_rcvseqinit(tp);
834	if (tiflags & TH_ACK) {
835	/*
836	* Our SYN was acked. If segment contains CC.ECHO
837	* option, check it to make sure this segment really
838	* matches our SYN. If not, just drop it as old
839	* duplicate, but send an RST if we're still playing
840	* by the old rules. If no CC.ECHO option, make sure
841	* we don't get fooled into using T/TCP.
842	*/
843	if (to.to_flags & TOF_CCECHO) {
844	if (tp->cc_send != to.to_ccecho) {
845	if (taop->tao_ccsent != 0)
846	goto drop;
847	else
848	goto dropwithreset;
849	}
850	} else
851	tp->t_flags &= ~TF_RCVD_CC;
852	tcpstat.tcps_connects++;
853	soisconnected(so);
854	/* Do window scaling on this connection? */
855	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
856	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
857	tp->snd_scale = tp->requested_s_scale;
858	tp->rcv_scale = tp->request_r_scale;
859	}
860	/* Segment is acceptable, update cache if undefined. */
861	if (taop->tao_ccsent == 0)
862	taop->tao_ccsent = to.to_ccecho;
863
864	tp->rcv_adv += tp->rcv_wnd;
865	tp->snd_una++; /* SYN is acked */
866	/*
867	* If there's data, delay ACK; if there's also a FIN
868	* ACKNOW will be turned on later.
869	*/
870	if (ti->ti_len != 0)
871	tp->t_flags \|= TF_DELACK;
872	else
873	tp->t_flags \|= TF_ACKNOW;
874	/*
875	* Received <SYN,ACK> in SYN_SENT[*] state.
876	* Transitions:
877	* SYN_SENT --> ESTABLISHED
878	* SYN_SENT* --> FIN_WAIT_1
879	*/
880	if (tp->t_flags & TF_NEEDFIN) {
881	tp->t_state = TCPS_FIN_WAIT_1;
882	tp->t_flags &= ~TF_NEEDFIN;
883	tiflags &= ~TH_SYN;
884	} else {
885	tp->t_state = TCPS_ESTABLISHED;
886	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
887	}
888	} else {
889	/*
890	* Received initial SYN in SYN-SENT[*] state => simul-
891	* taneous open. If segment contains CC option and there is
892	* a cached CC, apply TAO test; if it succeeds, connection is
893	* half-synchronized. Otherwise, do 3-way handshake:
894	* SYN-SENT -> SYN-RECEIVED
895	* SYN-SENT* -> SYN-RECEIVED*
896	* If there was no CC option, clear cached CC value.
897	*/
898	tp->t_flags \|= TF_ACKNOW;
899	tp->t_timer[TCPT_REXMT] = 0;
900	if (to.to_flags & TOF_CC) {
901	if (taop->tao_cc != 0 &&
902	CC_GT(to.to_cc, taop->tao_cc)) {
903	/*
904	* update cache and make transition:
905	* SYN-SENT -> ESTABLISHED*
906	* SYN-SENT* -> FIN-WAIT-1*
907	*/
908	taop->tao_cc = to.to_cc;
909	if (tp->t_flags & TF_NEEDFIN) {
910	tp->t_state = TCPS_FIN_WAIT_1;
911	tp->t_flags &= ~TF_NEEDFIN;
912	} else {
913	tp->t_state = TCPS_ESTABLISHED;
914	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
915	}
916	tp->t_flags \|= TF_NEEDSYN;
917	} else
918	tp->t_state = TCPS_SYN_RECEIVED;
919	} else {
920	/* CC.NEW or no option => invalidate cache */
921	taop->tao_cc = 0;
922	tp->t_state = TCPS_SYN_RECEIVED;
923	}
924	}
925
926	trimthenstep6:
927	/*
928	* Advance ti->ti_seq to correspond to first data byte.
929	* If data, trim to stay within window,
930	* dropping FIN if necessary.
931	*/
932	ti->ti_seq++;
933	if (ti->ti_len > tp->rcv_wnd) {
934	todrop = ti->ti_len - tp->rcv_wnd;
935	m_adj(m, -todrop);
936	ti->ti_len = tp->rcv_wnd;
937	tiflags &= ~TH_FIN;
938	tcpstat.tcps_rcvpackafterwin++;
939	tcpstat.tcps_rcvbyteafterwin += todrop;
940	}
941	tp->snd_wl1 = ti->ti_seq - 1;
942	tp->rcv_up = ti->ti_seq;
943	/*
944	* Client side of transaction: already sent SYN and data.
945	* If the remote host used T/TCP to validate the SYN,
946	* our data will be ACK'd; if so, enter normal data segment
947	* processing in the middle of step 5, ack processing.
948	* Otherwise, goto step 6.
949	*/
950	if (tiflags & TH_ACK)
951	goto process_ACK;
952	goto step6;
953	/*
954	* If the state is LAST_ACK or CLOSING or TIME_WAIT:
955	* if segment contains a SYN and CC [not CC.NEW] option:
956	* if state == TIME_WAIT and connection duration > MSL,
957	* drop packet and send RST;
958	*
959	* if SEG.CC > CCrecv then is new SYN, and can implicitly
960	* ack the FIN (and data) in retransmission queue.
961	* Complete close and delete TCPCB. Then reprocess
962	* segment, hoping to find new TCPCB in LISTEN state;
963	*
964	* else must be old SYN; drop it.
965	* else do normal processing.
966	*/
967	case TCPS_LAST_ACK:
968	case TCPS_CLOSING:
969	case TCPS_TIME_WAIT:
970	if ((tiflags & TH_SYN) &&
971	(to.to_flags & TOF_CC) && tp->cc_recv != 0) {
972	if (tp->t_state == TCPS_TIME_WAIT &&
973	tp->t_duration > TCPTV_MSL)
974	goto dropwithreset;
975	if (CC_GT(to.to_cc, tp->cc_recv)) {
976	tp = tcp_close(tp);
977	goto findpcb;
978	}
979	else
980	goto drop;
981	}
982	break; /* continue normal processing */
983	}
984
985	/*
986	* States other than LISTEN or SYN_SENT.
987	* First check timestamp, if present.
988	* Then check the connection count, if present.
989	* Then check that at least some bytes of segment are within
990	* receive window. If segment begins before rcv_nxt,
991	* drop leading data (and SYN); if nothing left, just ack.
992	*
993	* RFC 1323 PAWS: If we have a timestamp reply on this segment
994	* and it's less than ts_recent, drop it.
995	*/
996	if ((to.to_flags & TOF_TS) != 0 && (tiflags & TH_RST) == 0 &&
997	tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) {
998
999	/* Check to see if ts_recent is over 24 days old. */
1000	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1001	/*
1002	* Invalidate ts_recent. If this segment updates
1003	* ts_recent, the age will be reset later and ts_recent
1004	* will get a valid value. If it does not, setting
1005	* ts_recent to zero will at least satisfy the
1006	* requirement that zero be placed in the timestamp
1007	* echo reply when ts_recent isn't valid. The
1008	* age isn't reset until we get a valid ts_recent
1009	* because we don't want out-of-order segments to be
1010	* dropped when ts_recent is old.
1011	*/
1012	tp->ts_recent = 0;
1013	} else {
1014	tcpstat.tcps_rcvduppack++;
1015	tcpstat.tcps_rcvdupbyte += ti->ti_len;
1016	tcpstat.tcps_pawsdrop++;
1017	goto dropafterack;
1018	}
1019	}
1020
1021	/*
1022	* T/TCP mechanism
1023	* If T/TCP was negotiated and the segment doesn't have CC,
1024	* or if it's CC is wrong then drop the segment.
1025	* RST segments do not have to comply with this.
1026	*/
1027	if ((tp->t_flags & (TF_REQ_CC\|TF_RCVD_CC)) == (TF_REQ_CC\|TF_RCVD_CC) &&
1028	((to.to_flags & TOF_CC) == 0 \|\| tp->cc_recv != to.to_cc) &&
1029	(tiflags & TH_RST) == 0)
1030	goto dropafterack;
1031
1032	todrop = tp->rcv_nxt - ti->ti_seq;
1033	if (todrop > 0) {
1034	if (tiflags & TH_SYN) {
1035	tiflags &= ~TH_SYN;
1036	ti->ti_seq++;
1037	if (ti->ti_urp > 1)
1038	ti->ti_urp--;
1039	else
1040	tiflags &= ~TH_URG;
1041	todrop--;
1042	}
1043	/*
1044	* Following if statement from Stevens, vol. 2, p. 960.
1045	*/
1046	if (todrop > ti->ti_len
1047	\|\| (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
1048	/*
1049	* Any valid FIN must be to the left of the window.
1050	* At this point the FIN must be a duplicate or out
1051	* of sequence; drop it.
1052	*/
1053	tiflags &= ~TH_FIN;
1054
1055	/*
1056	* Send an ACK to resynchronize and drop any data.
1057	* But keep on processing for RST or ACK.
1058	*/
1059	tp->t_flags \|= TF_ACKNOW;
1060	todrop = ti->ti_len;
1061	tcpstat.tcps_rcvduppack++;
1062	tcpstat.tcps_rcvdupbyte += todrop;
1063	} else {
1064	tcpstat.tcps_rcvpartduppack++;
1065	tcpstat.tcps_rcvpartdupbyte += todrop;
1066	}
1067	m_adj(m, todrop);
1068	ti->ti_seq += todrop;
1069	ti->ti_len -= todrop;
1070	if (ti->ti_urp > todrop)
1071	ti->ti_urp -= todrop;
1072	else {
1073	tiflags &= ~TH_URG;
1074	ti->ti_urp = 0;
1075	}
1076	}
1077
1078	/*
1079	* If new data are received on a connection after the
1080	* user processes are gone, then RST the other end.
1081	*/
1082	if ((so->so_state & SS_NOFDREF) &&
1083	tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
1084	tp = tcp_close(tp);
1085	tcpstat.tcps_rcvafterclose++;
1086	goto dropwithreset;
1087	}
1088
1089	/*
1090	* If segment ends after window, drop trailing data
1091	* (and PUSH and FIN); if nothing left, just ACK.
1092	*/
1093	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1094	if (todrop > 0) {
1095	tcpstat.tcps_rcvpackafterwin++;
1096	if (todrop >= ti->ti_len) {
1097	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1098	/*
1099	* If a new connection request is received
1100	* while in TIME_WAIT, drop the old connection
1101	* and start over if the sequence numbers
1102	* are above the previous ones.
1103	*/
1104	if (tiflags & TH_SYN &&
1105	tp->t_state == TCPS_TIME_WAIT &&
1106	SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
1107	iss = tp->rcv_nxt + TCP_ISSINCR;
1108	tp = tcp_close(tp);
1109	goto findpcb;
1110	}
1111	/*
1112	* If window is closed can only take segments at
1113	* window edge, and have to drop data and PUSH from
1114	* incoming segments. Continue processing, but
1115	* remember to ack. Otherwise, drop segment
1116	* and ack.
1117	*/
1118	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
1119	tp->t_flags \|= TF_ACKNOW;
1120	tcpstat.tcps_rcvwinprobe++;
1121	} else
1122	goto dropafterack;
1123	} else
1124	tcpstat.tcps_rcvbyteafterwin += todrop;
1125	m_adj(m, -todrop);
1126	ti->ti_len -= todrop;
1127	tiflags &= ~(TH_PUSH\|TH_FIN);
1128	}
1129
1130	/*
1131	* If last ACK falls within this segment's sequence numbers,
1132	* record its timestamp.
1133	* NOTE that the test is modified according to the latest
1134	* proposal of the tcplw@cray.com list (Braden 1993/04/26).
1135	*/
1136	if ((to.to_flags & TOF_TS) != 0 &&
1137	SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
1138	tp->ts_recent_age = tcp_now;
1139	tp->ts_recent = to.to_tsval;
1140	}
1141
1142	/*
1143	* If the RST bit is set examine the state:
1144	* SYN_RECEIVED STATE:
1145	* If passive open, return to LISTEN state.
1146	* If active open, inform user that connection was refused.
1147	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1148	* Inform user that connection was reset, and close tcb.
1149	* CLOSING, LAST_ACK, TIME_WAIT STATES
1150	* Close the tcb.
1151	*/
1152	if (tiflags&TH_RST) switch (tp->t_state) {
1153
1154	case TCPS_SYN_RECEIVED:
1155	so->so_error = ECONNREFUSED;
1156	goto close;
1157
1158	case TCPS_ESTABLISHED:
1159	case TCPS_FIN_WAIT_1:
1160	case TCPS_FIN_WAIT_2:
1161	case TCPS_CLOSE_WAIT:
1162	so->so_error = ECONNRESET;
1163	close:
1164	tp->t_state = TCPS_CLOSED;
1165	tcpstat.tcps_drops++;
1166	tp = tcp_close(tp);
1167	goto drop;
1168
1169	case TCPS_CLOSING:
1170	case TCPS_LAST_ACK:
1171	case TCPS_TIME_WAIT:
1172	tp = tcp_close(tp);
1173	goto drop;
1174	}
1175
1176	/*
1177	* If a SYN is in the window, then this is an
1178	* error and we send an RST and drop the connection.
1179	*/
1180	if (tiflags & TH_SYN) {
1181	tp = tcp_drop(tp, ECONNRESET);
1182	goto dropwithreset;
1183	}
1184
1185	/*
1186	* If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1187	* flag is on (half-synchronized state), then queue data for
1188	* later processing; else drop segment and return.
1189	*/
1190	if ((tiflags & TH_ACK) == 0) {
1191	if (tp->t_state == TCPS_SYN_RECEIVED \|\|
1192	(tp->t_flags & TF_NEEDSYN))
1193	goto step6;
1194	else
1195	goto drop;
1196	}
1197
1198	/*
1199	* Ack processing.
1200	*/
1201	switch (tp->t_state) {
1202
1203	/*
1204	* In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1205	* ESTABLISHED state and continue processing.
1206	* The ACK was checked above.
1207	*/
1208	case TCPS_SYN_RECEIVED:
1209
1210	tcpstat.tcps_connects++;
1211	soisconnected(so);
1212	/* Do window scaling? */
1213	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
1214	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
1215	tp->snd_scale = tp->requested_s_scale;
1216	tp->rcv_scale = tp->request_r_scale;
1217	}
1218	/*
1219	* Upon successful completion of 3-way handshake,
1220	* update cache.CC if it was undefined, pass any queued
1221	* data to the user, and advance state appropriately.
1222	*/
1223	if ((taop = tcp_gettaocache(inp)) != NULL &&
1224	taop->tao_cc == 0)
1225	taop->tao_cc = tp->cc_recv;
1226
1227	/*
1228	* Make transitions:
1229	* SYN-RECEIVED -> ESTABLISHED
1230	* SYN-RECEIVED* -> FIN-WAIT-1
1231	*/
1232	if (tp->t_flags & TF_NEEDFIN) {
1233	tp->t_state = TCPS_FIN_WAIT_1;
1234	tp->t_flags &= ~TF_NEEDFIN;
1235	} else {
1236	tp->t_state = TCPS_ESTABLISHED;
1237	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1238	}
1239	/*
1240	* If segment contains data or ACK, will call tcp_reass()
1241	* later; if not, do so now to pass queued data to user.
1242	*/
1243	if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0)
1244	(void) tcp_reass(tp, (struct tcpiphdr *)0,
1245	(struct mbuf *)0);
1246	tp->snd_wl1 = ti->ti_seq - 1;
1247	/* fall into ... */
1248
1249	/*
1250	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1251	* ACKs. If the ack is in the range
1252	* tp->snd_una < ti->ti_ack <= tp->snd_max
1253	* then advance tp->snd_una to ti->ti_ack and drop
1254	* data from the retransmission queue. If this ACK reflects
1255	* more up to date window information we update our window information.
1256	*/
1257	case TCPS_ESTABLISHED:
1258	case TCPS_FIN_WAIT_1:
1259	case TCPS_FIN_WAIT_2:
1260	case TCPS_CLOSE_WAIT:
1261	case TCPS_CLOSING:
1262	case TCPS_LAST_ACK:
1263	case TCPS_TIME_WAIT:
1264
1265	if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1266	if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1267	tcpstat.tcps_rcvdupack++;
1268	/*
1269	* If we have outstanding data (other than
1270	* a window probe), this is a completely
1271	* duplicate ack (ie, window info didn't
1272	* change), the ack is the biggest we've
1273	* seen and we've seen exactly our rexmt
1274	* threshhold of them, assume a packet
1275	* has been dropped and retransmit it.
1276	* Kludge snd_nxt & the congestion
1277	* window so we send only this one
1278	* packet.
1279	*
1280	* We know we're losing at the current
1281	* window size so do congestion avoidance
1282	* (set ssthresh to half the current window
1283	* and pull our congestion window back to
1284	* the new ssthresh).
1285	*
1286	* Dup acks mean that packets have left the
1287	* network (they're now cached at the receiver)
1288	* so bump cwnd by the amount in the receiver
1289	* to keep a constant cwnd packets in the
1290	* network.
1291	*/
1292	if (tp->t_timer[TCPT_REXMT] == 0 \|\|
1293	ti->ti_ack != tp->snd_una)
1294	tp->t_dupacks = 0;
1295	else if (++tp->t_dupacks == tcprexmtthresh) {
1296	tcp_seq onxt = tp->snd_nxt;
1297	u_int win =
1298	min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1299	tp->t_maxseg;
1300
1301	if (win < 2)
1302	win = 2;
1303	tp->snd_ssthresh = win * tp->t_maxseg;
1304	tp->t_timer[TCPT_REXMT] = 0;
1305	tp->t_rtt = 0;
1306	tp->snd_nxt = ti->ti_ack;
1307	tp->snd_cwnd = tp->t_maxseg;
1308	(void) tcp_output(tp);
1309	tp->snd_cwnd = tp->snd_ssthresh +
1310	tp->t_maxseg * tp->t_dupacks;
1311	if (SEQ_GT(onxt, tp->snd_nxt))
1312	tp->snd_nxt = onxt;
1313	goto drop;
1314	} else if (tp->t_dupacks > tcprexmtthresh) {
1315	tp->snd_cwnd += tp->t_maxseg;
1316	(void) tcp_output(tp);
1317	goto drop;
1318	}
1319	} else
1320	tp->t_dupacks = 0;
1321	break;
1322	}
1323	/*
1324	* If the congestion window was inflated to account
1325	* for the other side's cached packets, retract it.
1326	*/
1327	if (tp->t_dupacks >= tcprexmtthresh &&
1328	tp->snd_cwnd > tp->snd_ssthresh)
1329	tp->snd_cwnd = tp->snd_ssthresh;
1330	tp->t_dupacks = 0;
1331	if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1332	tcpstat.tcps_rcvacktoomuch++;
1333	goto dropafterack;
1334	}
1335	/*
1336	* If we reach this point, ACK is not a duplicate,
1337	* i.e., it ACKs something we sent.
1338	*/
1339	if (tp->t_flags & TF_NEEDSYN) {
1340	/*
1341	* T/TCP: Connection was half-synchronized, and our
1342	* SYN has been ACK'd (so connection is now fully
1343	* synchronized). Go to non-starred state,
1344	* increment snd_una for ACK of SYN, and check if
1345	* we can do window scaling.
1346	*/
1347	tp->t_flags &= ~TF_NEEDSYN;
1348	tp->snd_una++;
1349	/* Do window scaling? */
1350	if ((tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE)) ==
1351	(TF_RCVD_SCALE\|TF_REQ_SCALE)) {
1352	tp->snd_scale = tp->requested_s_scale;
1353	tp->rcv_scale = tp->request_r_scale;
1354	}
1355	}
1356
1357	process_ACK:
1358	acked = ti->ti_ack - tp->snd_una;
1359	tcpstat.tcps_rcvackpack++;
1360	tcpstat.tcps_rcvackbyte += acked;
1361
1362	/*
1363	* If we have a timestamp reply, update smoothed
1364	* round trip time. If no timestamp is present but
1365	* transmit timer is running and timed sequence
1366	* number was acked, update smoothed round trip time.
1367	* Since we now have an rtt measurement, cancel the
1368	* timer backoff (cf., Phil Karn's retransmit alg.).
1369	* Recompute the initial retransmit timer.
1370	*/
1371	if (to.to_flags & TOF_TS)
1372	tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1);
1373	else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1374	tcp_xmit_timer(tp,tp->t_rtt);
1375
1376	/*
1377	* If all outstanding data is acked, stop retransmit
1378	* timer and remember to restart (more output or persist).
1379	* If there is more data to be acked, restart retransmit
1380	* timer, using current (possibly backed-off) value.
1381	*/
1382	if (ti->ti_ack == tp->snd_max) {
1383	tp->t_timer[TCPT_REXMT] = 0;
1384	needoutput = 1;
1385	} else if (tp->t_timer[TCPT_PERSIST] == 0)
1386	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1387
1388	/*
1389	* If no data (only SYN) was ACK'd,
1390	* skip rest of ACK processing.
1391	*/
1392	if (acked == 0)
1393	goto step6;
1394
1395	/*
1396	* When new data is acked, open the congestion window.
1397	* If the window gives us less than ssthresh packets
1398	* in flight, open exponentially (maxseg per packet).
1399	* Otherwise open linearly: maxseg per window
1400	* (maxseg^2 / cwnd per packet).
1401	*/
1402	{
1403	register u_int cw = tp->snd_cwnd;
1404	register u_int incr = tp->t_maxseg;
1405
1406	if (cw > tp->snd_ssthresh)
1407	incr = incr * incr / cw;
1408	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1409	}
1410	if (acked > so->so_snd.sb_cc) {
1411	tp->snd_wnd -= so->so_snd.sb_cc;
1412	sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1413	ourfinisacked = 1;
1414	} else {
1415	sbdrop(&so->so_snd, acked);
1416	tp->snd_wnd -= acked;
1417	ourfinisacked = 0;
1418	}
1419	if (so->so_snd.sb_flags & SB_NOTIFY)
1420	sowwakeup(so);
1421	tp->snd_una = ti->ti_ack;
1422	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1423	tp->snd_nxt = tp->snd_una;
1424
1425	switch (tp->t_state) {
1426
1427	/*
1428	* In FIN_WAIT_1 STATE in addition to the processing
1429	* for the ESTABLISHED state if our FIN is now acknowledged
1430	* then enter FIN_WAIT_2.
1431	*/
1432	case TCPS_FIN_WAIT_1:
1433	if (ourfinisacked) {
1434	/*
1435	* If we can't receive any more
1436	* data, then closing user can proceed.
1437	* Starting the timer is contrary to the
1438	* specification, but if we don't get a FIN
1439	* we'll hang forever.
1440	*/
1441	if (so->so_state & SS_CANTRCVMORE) {
1442	soisdisconnected(so);
1443	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1444	}
1445	tp->t_state = TCPS_FIN_WAIT_2;
1446	}
1447	break;
1448
1449	/*
1450	* In CLOSING STATE in addition to the processing for
1451	* the ESTABLISHED state if the ACK acknowledges our FIN
1452	* then enter the TIME-WAIT state, otherwise ignore
1453	* the segment.
1454	*/
1455	case TCPS_CLOSING:
1456	if (ourfinisacked) {
1457	tp->t_state = TCPS_TIME_WAIT;
1458	tcp_canceltimers(tp);
1459	/* Shorten TIME_WAIT [RFC-1644, p.28] */
1460	if (tp->cc_recv != 0 &&
1461	tp->t_duration < TCPTV_MSL)
1462	tp->t_timer[TCPT_2MSL] =
1463	tp->t_rxtcur * TCPTV_TWTRUNC;
1464	else
1465	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1466	soisdisconnected(so);
1467	}
1468	break;
1469
1470	/*
1471	* In LAST_ACK, we may still be waiting for data to drain
1472	* and/or to be acked, as well as for the ack of our FIN.
1473	* If our FIN is now acknowledged, delete the TCB,
1474	* enter the closed state and return.
1475	*/
1476	case TCPS_LAST_ACK:
1477	if (ourfinisacked) {
1478	tp = tcp_close(tp);
1479	goto drop;
1480	}
1481	break;
1482
1483	/*
1484	* In TIME_WAIT state the only thing that should arrive
1485	* is a retransmission of the remote FIN. Acknowledge
1486	* it and restart the finack timer.
1487	*/
1488	case TCPS_TIME_WAIT:
1489	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1490	goto dropafterack;
1491	}
1492	}
1493
1494	step6:
1495	/*
1496	* Update window information.
1497	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1498	*/
1499	if ((tiflags & TH_ACK) &&
1500	(SEQ_LT(tp->snd_wl1, ti->ti_seq) \|\|
1501	(tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) \|\|
1502	(tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1503	/* keep track of pure window updates */
1504	if (ti->ti_len == 0 &&
1505	tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1506	tcpstat.tcps_rcvwinupd++;
1507	tp->snd_wnd = tiwin;
1508	tp->snd_wl1 = ti->ti_seq;
1509	tp->snd_wl2 = ti->ti_ack;
1510	if (tp->snd_wnd > tp->max_sndwnd)
1511	tp->max_sndwnd = tp->snd_wnd;
1512	needoutput = 1;
1513	}
1514
1515	/*
1516	* Process segments with URG.
1517	*/
1518	if ((tiflags & TH_URG) && ti->ti_urp &&
1519	TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1520	/*
1521	* This is a kludge, but if we receive and accept
1522	* random urgent pointers, we'll crash in
1523	* soreceive. It's hard to imagine someone
1524	* actually wanting to send this much urgent data.
1525	*/
1526	if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1527	ti->ti_urp = 0; /* XXX */
1528	tiflags &= ~TH_URG; /* XXX */
1529	goto dodata; /* XXX */
1530	}
1531	/*
1532	* If this segment advances the known urgent pointer,
1533	* then mark the data stream. This should not happen
1534	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1535	* a FIN has been received from the remote side.
1536	* In these states we ignore the URG.
1537	*
1538	* According to RFC961 (Assigned Protocols),
1539	* the urgent pointer points to the last octet
1540	* of urgent data. We continue, however,
1541	* to consider it to indicate the first octet
1542	* of data past the urgent section as the original
1543	* spec states (in one of two places).
1544	*/
1545	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1546	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1547	so->so_oobmark = so->so_rcv.sb_cc +
1548	(tp->rcv_up - tp->rcv_nxt) - 1;
1549	if (so->so_oobmark == 0)
1550	so->so_state \|= SS_RCVATMARK;
1551	sohasoutofband(so);
1552	tp->t_oobflags &= ~(TCPOOB_HAVEDATA \| TCPOOB_HADDATA);
1553	}
1554	/*
1555	* Remove out of band data so doesn't get presented to user.
1556	* This can happen independent of advancing the URG pointer,
1557	* but if two URG's are pending at once, some out-of-band
1558	* data may creep in... ick.
1559	*/
1560	if (ti->ti_urp <= (u_long)ti->ti_len
1561	#ifdef SO_OOBINLINE
1562	&& (so->so_options & SO_OOBINLINE) == 0
1563	#endif
1564	)
1565	tcp_pulloutofband(so, ti, m);
1566	} else
1567	/*
1568	* If no out of band data is expected,
1569	* pull receive urgent pointer along
1570	* with the receive window.
1571	*/
1572	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1573	tp->rcv_up = tp->rcv_nxt;
1574	dodata: /* XXX */
1575
1576	/*
1577	* Process the segment text, merging it into the TCP sequencing queue,
1578	* and arranging for acknowledgment of receipt if necessary.
1579	* This process logically involves adjusting tp->rcv_wnd as data
1580	* is presented to the user (this happens in tcp_usrreq.c,
1581	* case PRU_RCVD). If a FIN has already been received on this
1582	* connection then we just ignore the text.
1583	*/
1584	if ((ti->ti_len \|\| (tiflags&TH_FIN)) &&
1585	TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1586	TCP_REASS(tp, ti, m, so, tiflags);
1587	/*
1588	* Note the amount of data that peer has sent into
1589	* our window, in order to estimate the sender's
1590	* buffer size.
1591	*/
1592	len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1593	} else {
1594	m_freem(m);
1595	tiflags &= ~TH_FIN;
1596	}
1597
1598	/*
1599	* If FIN is received ACK the FIN and let the user know
1600	* that the connection is closing.
1601	*/
1602	if (tiflags & TH_FIN) {
1603	if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1604	socantrcvmore(so);
1605	/*
1606	* If connection is half-synchronized
1607	* (ie NEEDSYN flag on) then delay ACK,
1608	* so it may be piggybacked when SYN is sent.
1609	* Otherwise, since we received a FIN then no
1610	* more input can be expected, send ACK now.
1611	*/
1612	if (tp->t_flags & TF_NEEDSYN)
1613	tp->t_flags \|= TF_DELACK;
1614	else
1615	tp->t_flags \|= TF_ACKNOW;
1616	tp->rcv_nxt++;
1617	}
1618	switch (tp->t_state) {
1619
1620	/*
1621	* In SYN_RECEIVED and ESTABLISHED STATES
1622	* enter the CLOSE_WAIT state.
1623	*/
1624	case TCPS_SYN_RECEIVED:
1625	case TCPS_ESTABLISHED:
1626	tp->t_state = TCPS_CLOSE_WAIT;
1627	break;
1628
1629	/*
1630	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1631	* enter the CLOSING state.
1632	*/
1633	case TCPS_FIN_WAIT_1:
1634	tp->t_state = TCPS_CLOSING;
1635	break;
1636
1637	/*
1638	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1639	* starting the time-wait timer, turning off the other
1640	* standard timers.
1641	*/
1642	case TCPS_FIN_WAIT_2:
1643	tp->t_state = TCPS_TIME_WAIT;
1644	tcp_canceltimers(tp);
1645	/* Shorten TIME_WAIT [RFC-1644, p.28] */
1646	if (tp->cc_recv != 0 &&
1647	tp->t_duration < TCPTV_MSL) {
1648	tp->t_timer[TCPT_2MSL] =
1649	tp->t_rxtcur * TCPTV_TWTRUNC;
1650	/* For transaction client, force ACK now. */
1651	tp->t_flags \|= TF_ACKNOW;
1652	}
1653	else
1654	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1655	soisdisconnected(so);
1656	break;
1657
1658	/*
1659	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1660	*/
1661	case TCPS_TIME_WAIT:
1662	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1663	break;
1664	}
1665	}
1666	#ifdef TCPDEBUG
1667	if (so->so_options & SO_DEBUG)
1668	tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1669	#endif
1670
1671	/*
1672	* Return any desired output.
1673	*/
1674	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1675	(void) tcp_output(tp);
1676	return;
1677
1678	dropafterack:
1679	/*
1680	* Generate an ACK dropping incoming segment if it occupies
1681	* sequence space, where the ACK reflects our state.
1682	*/
1683	if (tiflags & TH_RST)
1684	goto drop;
1685	#ifdef TCPDEBUG
1686	if (so->so_options & SO_DEBUG)
1687	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1688	#endif
1689	m_freem(m);
1690	tp->t_flags \|= TF_ACKNOW;
1691	(void) tcp_output(tp);
1692	return;
1693
1694	dropwithreset:
1695	/*
1696	* Generate a RST, dropping incoming segment.
1697	* Make ACK acceptable to originator of segment.
1698	* Don't bother to respond if destination was broadcast/multicast.
1699	*/
1700	if ((tiflags & TH_RST) \|\| m->m_flags & (M_BCAST\|M_MCAST) \|\|
1701	IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1702	goto drop;
1703	#ifdef TCPDEBUG
1704	if (tp == 0 \|\| (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1705	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1706	#endif
1707	if (tiflags & TH_ACK)
1708	tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1709	else {
1710	if (tiflags & TH_SYN)
1711	ti->ti_len++;
1712	tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1713	TH_RST\|TH_ACK);
1714	}
1715	/* destroy temporarily created socket */
1716	if (dropsocket)
1717	(void) soabort(so);
1718	return;
1719
1720	drop:
1721	/*
1722	* Drop space held by incoming segment and return.
1723	*/
1724	#ifdef TCPDEBUG
1725	if (tp == 0 \|\| (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1726	tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1727	#endif
1728	m_freem(m);
1729	/* destroy temporarily created socket */
1730	if (dropsocket)
1731	(void) soabort(so);
1732	return;
1733	#ifndef TUBA_INCLUDE
1734	}
1735
1736	static void
1737	tcp_dooptions(struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti,
1738	struct tcpopt *to)
1739	{
1740	u_short mss = 0;
1741	int opt, optlen;
1742
1743	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1744	opt = cp[0];
1745	if (opt == TCPOPT_EOL)
1746	break;
1747	if (opt == TCPOPT_NOP)
1748	optlen = 1;
1749	else {
1750	optlen = cp[1];
1751	if (optlen <= 0)
1752	break;
1753	}
1754	switch (opt) {
1755
1756	default:
1757	continue;
1758
1759	case TCPOPT_MAXSEG:
1760	if (optlen != TCPOLEN_MAXSEG)
1761	continue;
1762	if (!(ti->ti_flags & TH_SYN))
1763	continue;
1764	bcopy((char ) cp + 2, (char ) &mss, sizeof(mss));
1765	NTOHS(mss);
1766	break;
1767
1768	case TCPOPT_WINDOW:
1769	if (optlen != TCPOLEN_WINDOW)
1770	continue;
1771	if (!(ti->ti_flags & TH_SYN))
1772	continue;
1773	tp->t_flags \|= TF_RCVD_SCALE;
1774	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1775	break;
1776
1777	case TCPOPT_TIMESTAMP:
1778	if (optlen != TCPOLEN_TIMESTAMP)
1779	continue;
1780	to->to_flags \|= TOF_TS;
1781	bcopy((char *)cp + 2,
1782	(char *)&to->to_tsval, sizeof(to->to_tsval));
1783	NTOHL(to->to_tsval);
1784	bcopy((char *)cp + 6,
1785	(char *)&to->to_tsecr, sizeof(to->to_tsecr));
1786	NTOHL(to->to_tsecr);
1787
1788	/*
1789	* A timestamp received in a SYN makes
1790	* it ok to send timestamp requests and replies.
1791	*/
1792	if (ti->ti_flags & TH_SYN) {
1793	tp->t_flags \|= TF_RCVD_TSTMP;
1794	tp->ts_recent = to->to_tsval;
1795	tp->ts_recent_age = tcp_now;
1796	}
1797	break;
1798	case TCPOPT_CC:
1799	if (optlen != TCPOLEN_CC)
1800	continue;
1801	to->to_flags \|= TOF_CC;
1802	bcopy((char *)cp + 2,
1803	(char *)&to->to_cc, sizeof(to->to_cc));
1804	NTOHL(to->to_cc);
1805	/*
1806	* A CC or CC.new option received in a SYN makes
1807	* it ok to send CC in subsequent segments.
1808	*/
1809	if (ti->ti_flags & TH_SYN)
1810	tp->t_flags \|= TF_RCVD_CC;
1811	break;
1812	case TCPOPT_CCNEW:
1813	if (optlen != TCPOLEN_CC)
1814	continue;
1815	if (!(ti->ti_flags & TH_SYN))
1816	continue;
1817	to->to_flags \|= TOF_CCNEW;
1818	bcopy((char *)cp + 2,
1819	(char *)&to->to_cc, sizeof(to->to_cc));
1820	NTOHL(to->to_cc);
1821	/*
1822	* A CC or CC.new option received in a SYN makes
1823	* it ok to send CC in subsequent segments.
1824	*/
1825	tp->t_flags \|= TF_RCVD_CC;
1826	break;
1827	case TCPOPT_CCECHO:
1828	if (optlen != TCPOLEN_CC)
1829	continue;
1830	if (!(ti->ti_flags & TH_SYN))
1831	continue;
1832	to->to_flags \|= TOF_CCECHO;
1833	bcopy((char *)cp + 2,
1834	(char *)&to->to_ccecho, sizeof(to->to_ccecho));
1835	NTOHL(to->to_ccecho);
1836	break;
1837	}
1838	}
1839	if (ti->ti_flags & TH_SYN)
1840	tcp_mss(tp, mss); /* sets t_maxseg */
1841	}
1842
1843	/*
1844	* Pull out of band byte out of a segment so
1845	* it doesn't appear in the user's data queue.
1846	* It is still reflected in the segment length for
1847	* sequencing purposes.
1848	*/
1849	static void
1850	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1851	{
1852	int cnt = ti->ti_urp - 1;
1853
1854	while (cnt >= 0) {
1855	if (m->m_len > cnt) {
1856	char *cp = mtod(m, caddr_t) + cnt;
1857	struct tcpcb *tp = sototcpcb(so);
1858
1859	tp->t_iobc = *cp;
1860	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1861	bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1862	m->m_len--;
1863	return;
1864	}
1865	cnt -= m->m_len;
1866	m = m->m_next;
1867	if (m == 0)
1868	break;
1869	}
1870	panic("tcp_pulloutofband");
1871	}
1872
1873	/*
1874	* Collect new round-trip time estimate
1875	* and update averages and current timeout.
1876	*/
1877	static void
1878	tcp_xmit_timer(struct tcpcb *tp, int rtt)
1879	{
1880	int delta;
1881
1882	tcpstat.tcps_rttupdated++;
1883	tp->t_rttupdated++;
1884	if (tp->t_srtt != 0) {
1885	/*
1886	* srtt is stored as fixed point with 5 bits after the
1887	* binary point (i.e., scaled by 8). The following magic
1888	* is equivalent to the smoothing algorithm in rfc793 with
1889	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1890	* point). Adjust rtt to origin 0.
1891	*/
1892	delta = ((rtt - 1) << TCP_DELTA_SHIFT)
1893	- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
1894
1895	if ((tp->t_srtt += delta) <= 0)
1896	tp->t_srtt = 1;
1897
1898	/*
1899	* We accumulate a smoothed rtt variance (actually, a
1900	* smoothed mean difference), then set the retransmit
1901	* timer to smoothed rtt + 4 times the smoothed variance.
1902	* rttvar is stored as fixed point with 4 bits after the
1903	* binary point (scaled by 16). The following is
1904	* equivalent to rfc793 smoothing with an alpha of .75
1905	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1906	* rfc793's wired-in beta.
1907	*/
1908	if (delta < 0)
1909	delta = -delta;
1910	delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
1911	if ((tp->t_rttvar += delta) <= 0)
1912	tp->t_rttvar = 1;
1913	} else {
1914	/*
1915	* No rtt measurement yet - use the unsmoothed rtt.
1916	* Set the variance to half the rtt (so our first
1917	* retransmit happens at 3*rtt).
1918	*/
1919	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1920	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1921	}
1922	tp->t_rtt = 0;
1923	tp->t_rxtshift = 0;
1924
1925	/*
1926	* the retransmit should happen at rtt + 4 * rttvar.
1927	* Because of the way we do the smoothing, srtt and rttvar
1928	* will each average +1/2 tick of bias. When we compute
1929	* the retransmit timer, we want 1/2 tick of rounding and
1930	* 1 extra tick because of +-1/2 tick uncertainty in the
1931	* firing of the timer. The bias will give us exactly the
1932	* 1.5 tick we need. But, because the bias is
1933	* statistical, we have to test that we don't drop below
1934	* the minimum feasible timer (which is 2 ticks).
1935	*/
1936	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1937	max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
1938
1939	/*
1940	* We received an ack for a packet that wasn't retransmitted;
1941	* it is probably safe to discard any error indications we've
1942	* received recently. This isn't quite right, but close enough
1943	* for now (a route might have failed after we sent a segment,
1944	* and the return path might not be symmetrical).
1945	*/
1946	tp->t_softerror = 0;
1947	}
1948
1949	/*
1950	* Determine a reasonable value for maxseg size.
1951	* If the route is known, check route for mtu.
1952	* If none, use an mss that can be handled on the outgoing
1953	* interface without forcing IP to fragment; if bigger than
1954	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1955	* to utilize large mbufs. If no route is found, route has no mtu,
1956	* or the destination isn't local, use a default, hopefully conservative
1957	* size (usually 512 or the default IP max size, but no more than the mtu
1958	* of the interface), as we can't discover anything about intervening
1959	* gateways or networks. We also initialize the congestion/slow start
1960	* window to be a single segment if the destination isn't local.
1961	* While looking at the routing entry, we also initialize other path-dependent
1962	* parameters from pre-set or cached values in the routing entry.
1963	*
1964	* Also take into account the space needed for options that we
1965	* send regularly. Make maxseg shorter by that amount to assure
1966	* that we can send maxseg amount of data even when the options
1967	* are present. Store the upper limit of the length of options plus
1968	* data in maxopd.
1969	*
1970	* NOTE that this routine is only called when we process an incoming
1971	* segment, for outgoing segments only tcp_mssopt is called.
1972	*
1973	* In case of T/TCP, we call this routine during implicit connection
1974	* setup as well (offer = -1), to initialize maxseg from the cached
1975	* MSS of our peer.
1976	*/
1977	void
1978	tcp_mss(struct tcpcb *tp, int offer)
1979	{
1980	register struct rtentry *rt;
1981	struct ifnet *ifp;
1982	register int rtt, mss;
1983	u_long bufsize;
1984	struct inpcb *inp;
1985	struct socket *so;
1986	struct rmxp_tao *taop;
1987	int origoffer = offer;
1988
1989	inp = tp->t_inpcb;
1990	if ((rt = tcp_rtlookup(inp)) == NULL) {
1991	tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
1992	return;
1993	}
1994	ifp = rt->rt_ifp;
1995	so = inp->inp_socket;
1996
1997	taop = rmx_taop(rt->rt_rmx);
1998	/*
1999	* Offer == -1 means that we didn't receive SYN yet,
2000	* use cached value in that case;
2001	*/
2002	if (offer == -1)
2003	offer = taop->tao_mssopt;
2004	/*
2005	* Offer == 0 means that there was no MSS on the SYN segment,
2006	* in this case we use tcp_mssdflt.
2007	*/
2008	if (offer == 0)
2009	offer = tcp_mssdflt;
2010	else
2011	/*
2012	* Sanity check: make sure that maxopd will be large
2013	* enough to allow some data on segments even is the
2014	* all the option space is used (40bytes). Otherwise
2015	* funny things may happen in tcp_output.
2016	*/
2017	offer = max(offer, 64);
2018	taop->tao_mssopt = offer;
2019
2020	/*
2021	* While we're here, check if there's an initial rtt
2022	* or rttvar. Convert from the route-table units
2023	* to scaled multiples of the slow timeout timer.
2024	*/
2025	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2026	/*
2027	* XXX the lock bit for RTT indicates that the value
2028	* is also a minimum value; this is subject to time.
2029	*/
2030	if (rt->rt_rmx.rmx_locks & RTV_RTT)
2031	tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
2032	tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
2033	tcpstat.tcps_usedrtt++;
2034	if (rt->rt_rmx.rmx_rttvar) {
2035	tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2036	(RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
2037	tcpstat.tcps_usedrttvar++;
2038	} else {
2039	/* default variation is +- 1 rtt */
2040	tp->t_rttvar =
2041	tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2042	}
2043	TCPT_RANGESET(tp->t_rxtcur,
2044	((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2045	tp->t_rttmin, TCPTV_REXMTMAX);
2046	}
2047	/*
2048	* if there's an mtu associated with the route, use it
2049	*/
2050	if (rt->rt_rmx.rmx_mtu)
2051	mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
2052	else
2053	{
2054	mss = ifp->if_mtu - sizeof(struct tcpiphdr);
2055	if (!in_localaddr(inp->inp_faddr))
2056	mss = min(mss, tcp_mssdflt);
2057	}
2058	mss = min(mss, offer);
2059	/*
2060	* maxopd stores the maximum length of data AND options
2061	* in a segment; maxseg is the amount of data in a normal
2062	* segment. We need to store this value (maxopd) apart
2063	* from maxseg, because now every segment carries options
2064	* and thus we normally have somewhat less data in segments.
2065	*/
2066	tp->t_maxopd = mss;
2067
2068	/*
2069	* In case of T/TCP, origoffer==-1 indicates, that no segments
2070	* were received yet. In this case we just guess, otherwise
2071	* we do the same as before T/TCP.
2072	*/
2073	if ((tp->t_flags & (TF_REQ_TSTMP\|TF_NOOPT)) == TF_REQ_TSTMP &&
2074	(origoffer == -1 \|\|
2075	(tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2076	mss -= TCPOLEN_TSTAMP_APPA;
2077	if ((tp->t_flags & (TF_REQ_CC\|TF_NOOPT)) == TF_REQ_CC &&
2078	(origoffer == -1 \|\|
2079	(tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2080	mss -= TCPOLEN_CC_APPA;
2081
2082	#if (MCLBYTES & (MCLBYTES - 1)) == 0
2083	if (mss > MCLBYTES)
2084	mss &= ~(MCLBYTES-1);
2085	#else
2086	if (mss > MCLBYTES)
2087	mss = mss / MCLBYTES * MCLBYTES;
2088	#endif
2089	/*
2090	* If there's a pipesize, change the socket buffer
2091	* to that size. Make the socket buffers an integral
2092	* number of mss units; if the mss is larger than
2093	* the socket buffer, decrease the mss.
2094	*/
2095	#ifdef RTV_SPIPE
2096	if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2097	#endif
2098	bufsize = so->so_snd.sb_hiwat;
2099	if (bufsize < mss)
2100	mss = bufsize;
2101	else {
2102	bufsize = roundup(bufsize, mss);
2103	if (bufsize > sb_max)
2104	bufsize = sb_max;
2105	(void)sbreserve(&so->so_snd, bufsize);
2106	}
2107	tp->t_maxseg = mss;
2108
2109	#ifdef RTV_RPIPE
2110	if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2111	#endif
2112	bufsize = so->so_rcv.sb_hiwat;
2113	if (bufsize > mss) {
2114	bufsize = roundup(bufsize, mss);
2115	if (bufsize > sb_max)
2116	bufsize = sb_max;
2117	(void)sbreserve(&so->so_rcv, bufsize);
2118	}
2119	/*
2120	* Don't force slow-start on local network.
2121	*/
2122	if (!in_localaddr(inp->inp_faddr))
2123	tp->snd_cwnd = mss;
2124
2125	if (rt->rt_rmx.rmx_ssthresh) {
2126	/*
2127	* There's some sort of gateway or interface
2128	* buffer limit on the path. Use this to set
2129	* the slow start threshhold, but set the
2130	* threshold to no less than 2*mss.
2131	*/
2132	tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2133	tcpstat.tcps_usedssthresh++;
2134	}
2135	}
2136
2137	/*
2138	* Determine the MSS option to send on an outgoing SYN.
2139	*/
2140	int
2141	tcp_mssopt(struct tcpcb *tp)
2142	{
2143	struct rtentry *rt;
2144
2145	rt = tcp_rtlookup(tp->t_inpcb);
2146	if (rt == NULL)
2147	return tcp_mssdflt;
2148
2149	return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr);
2150	}
2151	#endif /* TUBA_INCLUDE */

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

Context Navigation

source: rtems/cpukit/libnetworking/netinet/tcp_input.c @ 0e16fa45

Download in other formats: