tcp_input.c@ 35923

Last change on this file since 35923 was 35923, checked in by vboxsync, 14 years ago
NAT: 'icmp_error` frees source mbuf.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 62.2 KB

Line
1	/* $Id: tcp_input.c 35923 2011-02-10 03:48:27Z vboxsync $ */
2	/** @file
3	* NAT - TCP input.
4	*/
5
6	/*
7	* Copyright (C) 2006-2010 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*
19	* This code is based on:
20	*
21	* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
22	* The Regents of the University of California. All rights reserved.
23	*
24	* Redistribution and use in source and binary forms, with or without
25	* modification, are permitted provided that the following conditions
26	* are met:
27	* 1. Redistributions of source code must retain the above copyright
28	* notice, this list of conditions and the following disclaimer.
29	* 2. Redistributions in binary form must reproduce the above copyright
30	* notice, this list of conditions and the following disclaimer in the
31	* documentation and/or other materials provided with the distribution.
32	* 3. All advertising materials mentioning features or use of this software
33	* must display the following acknowledgement:
34	* This product includes software developed by the University of
35	* California, Berkeley and its contributors.
36	* 4. Neither the name of the University nor the names of its contributors
37	* may be used to endorse or promote products derived from this software
38	* without specific prior written permission.
39	*
40	* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
41	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
42	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
43	* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
44	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
45	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
46	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
47	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
48	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
49	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
50	* SUCH DAMAGE.
51	*
52	* @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
53	* tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
54	*/
55
56	/*
57	* Changes and additions relating to SLiRP
58	* Copyright (c) 1995 Danny Gasparovski.
59	*
60	* Please read the file COPYRIGHT for the
61	* terms and conditions of the copyright.
62	*/
63
64	#include <slirp.h>
65	#include "ip_icmp.h"
66
67
68	#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
69
70	/* for modulo comparisons of timestamps */
71	#define TSTMP_LT(a, b) ((int)((a)-(b)) < 0)
72	#define TSTMP_GEQ(a, b) ((int)((a)-(b)) >= 0)
73
74	#ifndef TCP_ACK_HACK
75	#define DELAY_ACK(tp, ti) \
76	if (ti->ti_flags & TH_PUSH) \
77	tp->t_flags \|= TF_ACKNOW; \
78	else \
79	tp->t_flags \|= TF_DELACK;
80	#else /* !TCP_ACK_HACK */
81	#define DELAY_ACK(tp, ign) \
82	tp->t_flags \|= TF_DELACK;
83	#endif /* TCP_ACK_HACK */
84
85
86	/*
87	* deps: netinet/tcp_reass.c
88	* tcp_reass_maxqlen = 48 (deafault)
89	* tcp_reass_maxseg = nmbclusters/16 (nmbclusters = 1024 + maxusers * 64 from kern/kern_mbuf.c let's say 256)
90	*/
91	int
92	tcp_reass(PNATState pData, struct tcpcb tp, struct tcphdr th, int tlenp, struct mbuf m)
93	{
94	struct tseg_qent *q;
95	struct tseg_qent *p = NULL;
96	struct tseg_qent *nq;
97	struct tseg_qent *te = NULL;
98	struct socket *so = tp->t_socket;
99	int flags;
100	STAM_PROFILE_START(&pData->StatTCP_reassamble, tcp_reassamble);
101
102	/*
103	* XXX: tcp_reass() is rather inefficient with its data structures
104	* and should be rewritten (see NetBSD for optimizations). While
105	* doing that it should move to its own file tcp_reass.c.
106	*/
107
108	/*
109	* Call with th==NULL after become established to
110	* force pre-ESTABLISHED data up to user socket.
111	*/
112	if (th == NULL)
113	goto present;
114
115	/*
116	* Limit the number of segments in the reassembly queue to prevent
117	* holding on to too many segments (and thus running out of mbufs).
118	* Make sure to let the missing segment through which caused this
119	* queue. Always keep one global queue entry spare to be able to
120	* process the missing segment.
121	*/
122	if ( th->th_seq != tp->rcv_nxt
123	&& ( tcp_reass_qsize + 1 >= tcp_reass_maxseg
124	\|\| tp->t_segqlen >= tcp_reass_maxqlen))
125	{
126	tcp_reass_overflows++;
127	tcpstat.tcps_rcvmemdrop++;
128	m_freem(pData, m);
129	*tlenp = 0;
130	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
131	return (0);
132	}
133
134	/*
135	* Allocate a new queue entry. If we can't, or hit the zone limit
136	* just drop the pkt.
137	*/
138	te = RTMemAlloc(sizeof(struct tseg_qent));
139	if (te == NULL)
140	{
141	tcpstat.tcps_rcvmemdrop++;
142	m_freem(pData, m);
143	*tlenp = 0;
144	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
145	return (0);
146	}
147	tp->t_segqlen++;
148	tcp_reass_qsize++;
149
150	/*
151	* Find a segment which begins after this one does.
152	*/
153	LIST_FOREACH(q, &tp->t_segq, tqe_q)
154	{
155	if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
156	break;
157	p = q;
158	}
159
160	/*
161	* If there is a preceding segment, it may provide some of
162	* our data already. If so, drop the data from the incoming
163	* segment. If it provides all of our data, drop us.
164	*/
165	if (p != NULL)
166	{
167	int i;
168	/* conversion to int (in i) handles seq wraparound */
169	i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
170	if (i > 0)
171	{
172	if (i >= *tlenp)
173	{
174	tcpstat.tcps_rcvduppack++;
175	tcpstat.tcps_rcvdupbyte += *tlenp;
176	m_freem(pData, m);
177	RTMemFree(te);
178	tp->t_segqlen--;
179	tcp_reass_qsize--;
180	/*
181	* Try to present any queued data
182	* at the left window edge to the user.
183	* This is needed after the 3-WHS
184	* completes.
185	*/
186	goto present; /* ??? */
187	}
188	m_adj(m, i);
189	*tlenp -= i;
190	th->th_seq += i;
191	}
192	}
193	tcpstat.tcps_rcvoopack++;
194	tcpstat.tcps_rcvoobyte += *tlenp;
195
196	/*
197	* While we overlap succeeding segments trim them or,
198	* if they are completely covered, dequeue them.
199	*/
200	while (q)
201	{
202	int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
203	if (i <= 0)
204	break;
205	if (i < q->tqe_len)
206	{
207	q->tqe_th->th_seq += i;
208	q->tqe_len -= i;
209	m_adj(q->tqe_m, i);
210	break;
211	}
212
213	nq = LIST_NEXT(q, tqe_q);
214	LIST_REMOVE(q, tqe_q);
215	m_freem(pData, q->tqe_m);
216	RTMemFree(q);
217	tp->t_segqlen--;
218	tcp_reass_qsize--;
219	q = nq;
220	}
221
222	/* Insert the new segment queue entry into place. */
223	te->tqe_m = m;
224	te->tqe_th = th;
225	te->tqe_len = *tlenp;
226
227	if (p == NULL)
228	{
229	LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
230	}
231	else
232	{
233	LIST_INSERT_AFTER(p, te, tqe_q);
234	}
235
236	present:
237	/*
238	* Present data to user, advancing rcv_nxt through
239	* completed sequence space.
240	*/
241	if (!TCPS_HAVEESTABLISHED(tp->t_state))
242	{
243	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
244	return (0);
245	}
246	q = LIST_FIRST(&tp->t_segq);
247	if (!q \|\| q->tqe_th->th_seq != tp->rcv_nxt)
248	{
249	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
250	return (0);
251	}
252	do
253	{
254	tp->rcv_nxt += q->tqe_len;
255	flags = q->tqe_th->th_flags & TH_FIN;
256	nq = LIST_NEXT(q, tqe_q);
257	LIST_REMOVE(q, tqe_q);
258	/* XXX: This place should be checked for the same code in
259	* original BSD code for Slirp and current BSD used SS_FCANTRCVMORE
260	*/
261	if (so->so_state & SS_FCANTSENDMORE)
262	m_freem(pData, q->tqe_m);
263	else
264	sbappend(pData, so, q->tqe_m);
265	RTMemFree(q);
266	tp->t_segqlen--;
267	tcp_reass_qsize--;
268	q = nq;
269	}
270	while (q && q->tqe_th->th_seq == tp->rcv_nxt);
271
272	STAM_PROFILE_STOP(&pData->StatTCP_reassamble, tcp_reassamble);
273	return flags;
274	}
275
276	/*
277	* TCP input routine, follows pages 65-76 of the
278	* protocol specification dated September, 1981 very closely.
279	*/
280	void
281	tcp_input(PNATState pData, register struct mbuf m, int iphlen, struct socket inso)
282	{
283	struct ip save_ip, *ip;
284	register struct tcpiphdr *ti;
285	caddr_t optp = NULL;
286	int optlen = 0;
287	int len, tlen, off;
288	register struct tcpcb *tp = 0;
289	register int tiflags;
290	struct socket *so = 0;
291	int todrop, acked, ourfinisacked, needoutput = 0;
292	/* int dropsocket = 0; */
293	int iss = 0;
294	u_long tiwin;
295	/* int ts_present = 0; */
296	STAM_PROFILE_START(&pData->StatTCP_input, counter_input);
297
298	LogFlow(("tcp_input: m = %8lx, iphlen = %2d, inso = %lx\n",
299	(long)m, iphlen, (long)inso));
300
301	if (inso != NULL)
302	{
303	QSOCKET_LOCK(tcb);
304	SOCKET_LOCK(inso);
305	QSOCKET_UNLOCK(tcb);
306	}
307	/*
308	* If called with m == 0, then we're continuing the connect
309	*/
310	if (m == NULL)
311	{
312	so = inso;
313	Log4(("NAT: tcp_input: %R[natsock]\n", so));
314	/* Re-set a few variables */
315	tp = sototcpcb(so);
316	m = so->so_m;
317
318	so->so_m = 0;
319	ti = so->so_ti;
320
321	/** @todo (vvl) clarify why it might happens */
322	if (ti == NULL)
323	{
324	LogRel(("NAT: ti is null. can't do any reseting connection actions\n"));
325	/* mbuf should be cleared in sofree called from tcp_close */
326	tcp_close(pData, tp);
327	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
328	return;
329	}
330
331	tiwin = ti->ti_win;
332	tiflags = ti->ti_flags;
333
334	goto cont_conn;
335	}
336
337	tcpstat.tcps_rcvtotal++;
338	/*
339	* Get IP and TCP header together in first mbuf.
340	* Note: IP leaves IP header in first mbuf.
341	*/
342	ti = mtod(m, struct tcpiphdr *);
343	if (iphlen > sizeof(struct ip))
344	{
345	ip_stripoptions(m, (struct mbuf *)0);
346	iphlen = sizeof(struct ip);
347	}
348	/* XXX Check if too short */
349
350
351	/*
352	* Save a copy of the IP header in case we want restore it
353	* for sending an ICMP error message in response.
354	*/
355	ip = mtod(m, struct ip *);
356	/*
357	* (vvl) ip_input substracts IP header length from ip->ip_len value.
358	* here we do the test the same as input method of UDP protocol.
359	*/
360	Assert((ip->ip_len + iphlen == m_length(m, NULL)));
361	save_ip = *ip;
362	save_ip.ip_len+= iphlen;
363
364	/*
365	* Checksum extended TCP header and data.
366	*/
367	tlen = ((struct ip *)ti)->ip_len;
368	memset(ti->ti_x1, 0, 9);
369	ti->ti_len = RT_H2N_U16((u_int16_t)tlen);
370	len = sizeof(struct ip) + tlen;
371	/* keep checksum for ICMP reply
372	* ti->ti_sum = cksum(m, len);
373	* if (ti->ti_sum) { */
374	if (cksum(m, len))
375	{
376	tcpstat.tcps_rcvbadsum++;
377	goto drop;
378	}
379
380	/*
381	* Check that TCP offset makes sense,
382	* pull out TCP options and adjust length. XXX
383	*/
384	off = ti->ti_off << 2;
385	if ( off < sizeof (struct tcphdr)
386	\|\| off > tlen)
387	{
388	tcpstat.tcps_rcvbadoff++;
389	goto drop;
390	}
391	tlen -= off;
392	ti->ti_len = tlen;
393	if (off > sizeof (struct tcphdr))
394	{
395	optlen = off - sizeof (struct tcphdr);
396	optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
397
398	/*
399	* Do quick retrieval of timestamp options ("options
400	* prediction?"). If timestamp is the only option and it's
401	* formatted as recommended in RFC 1323 appendix A, we
402	* quickly get the values now and not bother calling
403	* tcp_dooptions(), etc.
404	*/
405	#if 0
406	if (( optlen == TCPOLEN_TSTAMP_APPA
407	\|\| ( optlen > TCPOLEN_TSTAMP_APPA
408	&& optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
409	(u_int32_t )optp == RT_H2N_U32_C(TCPOPT_TSTAMP_HDR) &&
410	(ti->ti_flags & TH_SYN) == 0)
411	{
412	ts_present = 1;
413	ts_val = RT_N2H_U32((u_int32_t )(optp + 4));
414	ts_ecr = RT_N2H_U32((u_int32_t )(optp + 8));
415	optp = NULL; / * we have parsed the options * /
416	}
417	#endif
418	}
419	tiflags = ti->ti_flags;
420
421	/*
422	* Convert TCP protocol specific fields to host format.
423	*/
424	NTOHL(ti->ti_seq);
425	NTOHL(ti->ti_ack);
426	NTOHS(ti->ti_win);
427	NTOHS(ti->ti_urp);
428
429	/*
430	* Drop TCP, IP headers and TCP options.
431	*/
432	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
433	m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
434
435	/*
436	* Locate pcb for segment.
437	*/
438	findso:
439	if (so != NULL && so != &tcb)
440	SOCKET_UNLOCK(so);
441	QSOCKET_LOCK(tcb);
442	so = tcp_last_so;
443	if ( so->so_fport != ti->ti_dport
444	\|\| so->so_lport != ti->ti_sport
445	\|\| so->so_laddr.s_addr != ti->ti_src.s_addr
446	\|\| so->so_faddr.s_addr != ti->ti_dst.s_addr)
447	{
448	#ifdef VBOX_WITH_SLIRP_MT
449	struct socket *sonxt;
450	#endif
451	QSOCKET_UNLOCK(tcb);
452	/* @todo fix SOLOOKUP macrodefinition to be usable here */
453	#ifndef VBOX_WITH_SLIRP_MT
454	so = solookup(&tcb, ti->ti_src, ti->ti_sport,
455	ti->ti_dst, ti->ti_dport);
456	#else
457	so = NULL;
458	QSOCKET_FOREACH(so, sonxt, tcp)
459	/* { */
460	if ( so->so_lport == ti->ti_sport
461	&& so->so_laddr.s_addr == ti->ti_src.s_addr
462	&& so->so_faddr.s_addr == ti->ti_dst.s_addr
463	&& so->so_fport == ti->ti_dport
464	&& so->so_deleted != 1)
465	{
466	break; /* so is locked here */
467	}
468	LOOP_LABEL(tcp, so, sonxt);
469	}
470	if (so == &tcb) {
471	so = NULL;
472	}
473	#endif
474	if (so)
475	{
476	tcp_last_so = so;
477	}
478	++tcpstat.tcps_socachemiss;
479	}
480	else
481	{
482	SOCKET_LOCK(so);
483	QSOCKET_UNLOCK(tcb);
484	}
485
486	/*
487	* If the state is CLOSED (i.e., TCB does not exist) then
488	* all data in the incoming segment is discarded.
489	* If the TCB exists but is in CLOSED state, it is embryonic,
490	* but should either do a listen or a connect soon.
491	*
492	* state == CLOSED means we've done socreate() but haven't
493	* attached it to a protocol yet...
494	*
495	* XXX If a TCB does not exist, and the TH_SYN flag is
496	* the only flag set, then create a session, mark it
497	* as if it was LISTENING, and continue...
498	*/
499	if (so == 0)
500	{
501	if ((tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) != TH_SYN)
502	goto dropwithreset;
503
504	if ((so = socreate()) == NULL)
505	goto dropwithreset;
506	if (tcp_attach(pData, so) < 0)
507	{
508	RTMemFree(so); /* Not sofree (if it failed, it's not insqued) */
509	goto dropwithreset;
510	}
511	SOCKET_LOCK(so);
512	#ifndef VBOX_WITH_SLIRP_BSD_SBUF
513	sbreserve(pData, &so->so_snd, tcp_sndspace);
514	sbreserve(pData, &so->so_rcv, tcp_rcvspace);
515	#else
516	sbuf_new(&so->so_snd, NULL, tcp_sndspace, SBUF_AUTOEXTEND);
517	sbuf_new(&so->so_rcv, NULL, tcp_rcvspace, SBUF_AUTOEXTEND);
518	#endif
519
520	/* tcp_last_so = so; / / XXX ? */
521	/* tp = sototcpcb(so); */
522
523	so->so_laddr = ti->ti_src;
524	so->so_lport = ti->ti_sport;
525	so->so_faddr = ti->ti_dst;
526	so->so_fport = ti->ti_dport;
527
528	so->so_iptos = ((struct ip *)ti)->ip_tos;
529
530	tp = sototcpcb(so);
531	tp->t_state = TCPS_LISTEN;
532	}
533
534	/*
535	* If this is a still-connecting socket, this probably
536	* a retransmit of the SYN. Whether it's a retransmit SYN
537	* or something else, we nuke it.
538	*/
539	if (so->so_state & SS_ISFCONNECTING)
540	{
541	goto drop;
542	}
543
544	tp = sototcpcb(so);
545
546	/* XXX Should never fail */
547	if (tp == 0)
548	goto dropwithreset;
549	if (tp->t_state == TCPS_CLOSED)
550	{
551	goto drop;
552	}
553
554	/* Unscale the window into a 32-bit value. */
555	/* if ((tiflags & TH_SYN) == 0)
556	* tiwin = ti->ti_win << tp->snd_scale;
557	* else
558	*/
559	tiwin = ti->ti_win;
560
561	/*
562	* Segment received on connection.
563	* Reset idle time and keep-alive timer.
564	*/
565	tp->t_idle = 0;
566	if (so_options)
567	tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
568	else
569	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
570
571	/*
572	* Process options if not in LISTEN state,
573	* else do it below (after getting remote address).
574	*/
575	if (optp && tp->t_state != TCPS_LISTEN)
576	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
577	/* , */
578	/* &ts_present, &ts_val, &ts_ecr); */
579
580	/*
581	* Header prediction: check for the two common cases
582	* of a uni-directional data xfer. If the packet has
583	* no control flags, is in-sequence, the window didn't
584	* change and we're not retransmitting, it's a
585	* candidate. If the length is zero and the ack moved
586	* forward, we're the sender side of the xfer. Just
587	* free the data acked & wake any higher level process
588	* that was blocked waiting for space. If the length
589	* is non-zero and the ack didn't move, we're the
590	* receiver side. If we're getting packets in-order
591	* (the reassembly queue is empty), add the data to
592	* the socket buffer and note that we need a delayed ack.
593	*
594	* XXX Some of these tests are not needed
595	* eg: the tiwin == tp->snd_wnd prevents many more
596	* predictions.. with no real advantage..
597	*/
598	if ( tp->t_state == TCPS_ESTABLISHED
599	&& (tiflags & (TH_SYN\|TH_FIN\|TH_RST\|TH_URG\|TH_ACK)) == TH_ACK
600	/* && (!ts_present \|\| TSTMP_GEQ(ts_val, tp->ts_recent)) */
601	&& ti->ti_seq == tp->rcv_nxt
602	&& tiwin && tiwin == tp->snd_wnd
603	&& tp->snd_nxt == tp->snd_max)
604	{
605	/*
606	* If last ACK falls within this segment's sequence numbers,
607	* record the timestamp.
608	*/
609	#if 0
610	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
611	SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len))
612	{
613	tp->ts_recent_age = tcp_now;
614	tp->ts_recent = ts_val;
615	}
616	#endif
617
618	if (ti->ti_len == 0)
619	{
620	if ( SEQ_GT(ti->ti_ack, tp->snd_una)
621	&& SEQ_LEQ(ti->ti_ack, tp->snd_max)
622	&& tp->snd_cwnd >= tp->snd_wnd)
623	{
624	/*
625	* this is a pure ack for outstanding data.
626	*/
627	++tcpstat.tcps_predack;
628	#if 0
629	if (ts_present)
630	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
631	else
632	#endif
633	if ( tp->t_rtt
634	&& SEQ_GT(ti->ti_ack, tp->t_rtseq))
635	tcp_xmit_timer(pData, tp, tp->t_rtt);
636	acked = ti->ti_ack - tp->snd_una;
637	tcpstat.tcps_rcvackpack++;
638	tcpstat.tcps_rcvackbyte += acked;
639	#ifndef VBOX_WITH_SLIRP_BSD_SBUF
640	sbdrop(&so->so_snd, acked);
641	#else
642	if (sbuf_len(&so->so_snd) < acked)
643	/* drop all what sbuf have */
644	sbuf_setpos(&so->so_snd, 0);
645	else
646	sbuf_setpos(&so->so_snd, sbuf_len(&so->so_snd) - acked);
647	#endif
648	tp->snd_una = ti->ti_ack;
649	m_freem(pData, m);
650
651	/*
652	* If all outstanding data are acked, stop
653	* retransmit timer, otherwise restart timer
654	* using current (possibly backed-off) value.
655	* If process is waiting for space,
656	* wakeup/selwakeup/signal. If data
657	* are ready to send, let tcp_output
658	* decide between more output or persist.
659	*/
660	if (tp->snd_una == tp->snd_max)
661	tp->t_timer[TCPT_REXMT] = 0;
662	else if (tp->t_timer[TCPT_PERSIST] == 0)
663	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
664
665	/*
666	* There's room in so_snd, sowwakup will read()
667	* from the socket if we can
668	*/
669	#if 0
670	if (so->so_snd.sb_flags & SB_NOTIFY)
671	sowwakeup(so);
672	#endif
673	/*
674	* This is called because sowwakeup might have
675	* put data into so_snd. Since we don't so sowwakeup,
676	* we don't need this.. XXX???
677	*/
678	if (SBUF_LEN(&so->so_snd))
679	(void) tcp_output(pData, tp);
680
681	SOCKET_UNLOCK(so);
682	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
683	return;
684	}
685	}
686	else if ( ti->ti_ack == tp->snd_una
687	&& LIST_FIRST(&tp->t_segq)
688	&& ti->ti_len <= sbspace(&so->so_rcv))
689	{
690	/*
691	* this is a pure, in-sequence data packet
692	* with nothing on the reassembly queue and
693	* we have enough buffer space to take it.
694	*/
695	++tcpstat.tcps_preddat;
696	tp->rcv_nxt += ti->ti_len;
697	tcpstat.tcps_rcvpack++;
698	tcpstat.tcps_rcvbyte += ti->ti_len;
699	/*
700	* Add data to socket buffer.
701	*/
702	sbappend(pData, so, m);
703
704	/*
705	* XXX This is called when data arrives. Later, check
706	* if we can actually write() to the socket
707	* XXX Need to check? It's be NON_BLOCKING
708	*/
709	/* sorwakeup(so); */
710
711	/*
712	* If this is a short packet, then ACK now - with Nagel
713	* congestion avoidance sender won't send more until
714	* he gets an ACK.
715	*
716	* It is better to not delay acks at all to maximize
717	* TCP throughput. See RFC 2581.
718	*/
719	tp->t_flags \|= TF_ACKNOW;
720	tcp_output(pData, tp);
721	SOCKET_UNLOCK(so);
722	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
723	return;
724	}
725	} /* header prediction */
726	/*
727	* Calculate amount of space in receive window,
728	* and then do TCP input processing.
729	* Receive window is amount of space in rcv queue,
730	* but not less than advertised window.
731	*/
732	{
733	int win;
734	win = sbspace(&so->so_rcv);
735	if (win < 0)
736	win = 0;
737	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
738	}
739
740	switch (tp->t_state)
741	{
742	/*
743	* If the state is LISTEN then ignore segment if it contains an RST.
744	* If the segment contains an ACK then it is bad and send a RST.
745	* If it does not contain a SYN then it is not interesting; drop it.
746	* Don't bother responding if the destination was a broadcast.
747	* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
748	* tp->iss, and send a segment:
749	* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
750	* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
751	* Fill in remote peer address fields if not previously specified.
752	* Enter SYN_RECEIVED state, and process any other fields of this
753	* segment in this state.
754	*/
755	case TCPS_LISTEN:
756	{
757	if (tiflags & TH_RST)
758	goto drop;
759	if (tiflags & TH_ACK)
760	goto dropwithreset;
761	if ((tiflags & TH_SYN) == 0)
762	goto drop;
763
764	/*
765	* This has way too many gotos...
766	* But a bit of spaghetti code never hurt anybody :)
767	*/
768	if ( (tcp_fconnect(pData, so) == -1)
769	&& errno != EINPROGRESS
770	&& errno != EWOULDBLOCK)
771	{
772	u_char code = ICMP_UNREACH_NET;
773	Log2((" tcp fconnect errno = %d (%s)\n", errno, strerror(errno)));
774	if (errno == ECONNREFUSED)
775	{
776	/* ACK the SYN, send RST to refuse the connection */
777	tcp_respond(pData, tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
778	TH_RST\|TH_ACK);
779	}
780	else
781	{
782	if (errno == EHOSTUNREACH)
783	code = ICMP_UNREACH_HOST;
784	HTONL(ti->ti_seq); /* restore tcp header */
785	HTONL(ti->ti_ack);
786	HTONS(ti->ti_win);
787	HTONS(ti->ti_urp);
788	m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
789	m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
790	*ip = save_ip;
791	icmp_error(pData, m, ICMP_UNREACH, code, 0, strerror(errno));
792	tp->t_socket->so_m = NULL;
793	}
794	tp = tcp_close(pData, tp);
795	}
796	else
797	{
798	/*
799	* Haven't connected yet, save the current mbuf
800	* and ti, and return
801	* XXX Some OS's don't tell us whether the connect()
802	* succeeded or not. So we must time it out.
803	*/
804	so->so_m = m;
805	so->so_ti = ti;
806	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
807	tp->t_state = TCPS_SYN_RECEIVED;
808	}
809	SOCKET_UNLOCK(so);
810	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
811	return;
812
813	cont_conn:
814	/* m==NULL
815	* Check if the connect succeeded
816	*/
817	if (so->so_state & SS_NOFDREF)
818	{
819	tp = tcp_close(pData, tp);
820	goto dropwithreset;
821	}
822	cont_input:
823	tcp_template(tp);
824
825	if (optp)
826	tcp_dooptions(pData, tp, (u_char *)optp, optlen, ti);
827
828	if (iss)
829	tp->iss = iss;
830	else
831	tp->iss = tcp_iss;
832	tcp_iss += TCP_ISSINCR/2;
833	tp->irs = ti->ti_seq;
834	tcp_sendseqinit(tp);
835	tcp_rcvseqinit(tp);
836	tp->t_flags \|= TF_ACKNOW;
837	tp->t_state = TCPS_SYN_RECEIVED;
838	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
839	tcpstat.tcps_accepts++;
840	goto trimthenstep6;
841	} /* case TCPS_LISTEN */
842
843	/*
844	* If the state is SYN_SENT:
845	* if seg contains an ACK, but not for our SYN, drop the input.
846	* if seg contains a RST, then drop the connection.
847	* if seg does not contain SYN, then drop it.
848	* Otherwise this is an acceptable SYN segment
849	* initialize tp->rcv_nxt and tp->irs
850	* if seg contains ack then advance tp->snd_una
851	* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
852	* arrange for segment to be acked (eventually)
853	* continue processing rest of data/controls, beginning with URG
854	*/
855	case TCPS_SYN_SENT:
856	if ( (tiflags & TH_ACK)
857	&& ( SEQ_LEQ(ti->ti_ack, tp->iss)
858	\|\| SEQ_GT(ti->ti_ack, tp->snd_max)))
859	goto dropwithreset;
860
861	if (tiflags & TH_RST)
862	{
863	if (tiflags & TH_ACK)
864	tp = tcp_drop(pData, tp, 0); /* XXX Check t_softerror! */
865	goto drop;
866	}
867
868	if ((tiflags & TH_SYN) == 0)
869	{
870	goto drop;
871	}
872	if (tiflags & TH_ACK)
873	{
874	tp->snd_una = ti->ti_ack;
875	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
876	tp->snd_nxt = tp->snd_una;
877	}
878
879	tp->t_timer[TCPT_REXMT] = 0;
880	tp->irs = ti->ti_seq;
881	tcp_rcvseqinit(tp);
882	tp->t_flags \|= TF_ACKNOW;
883	if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss))
884	{
885	tcpstat.tcps_connects++;
886	soisfconnected(so);
887	tp->t_state = TCPS_ESTABLISHED;
888
889	/* Do window scaling on this connection? */
890	#if 0
891	if (( tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
892	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
893	{
894	tp->snd_scale = tp->requested_s_scale;
895	tp->rcv_scale = tp->request_r_scale;
896	}
897	#endif
898	(void) tcp_reass(pData, tp, (struct tcphdr )0, NULL, (struct mbuf )0);
899	/*
900	* if we didn't have to retransmit the SYN,
901	* use its rtt as our initial srtt & rtt var.
902	*/
903	if (tp->t_rtt)
904	tcp_xmit_timer(pData, tp, tp->t_rtt);
905	}
906	else
907	tp->t_state = TCPS_SYN_RECEIVED;
908
909	trimthenstep6:
910	/*
911	* Advance ti->ti_seq to correspond to first data byte.
912	* If data, trim to stay within window,
913	* dropping FIN if necessary.
914	*/
915	ti->ti_seq++;
916	if (ti->ti_len > tp->rcv_wnd)
917	{
918	todrop = ti->ti_len - tp->rcv_wnd;
919	m_adj(m, -todrop);
920	ti->ti_len = tp->rcv_wnd;
921	tiflags &= ~TH_FIN;
922	tcpstat.tcps_rcvpackafterwin++;
923	tcpstat.tcps_rcvbyteafterwin += todrop;
924	}
925	tp->snd_wl1 = ti->ti_seq - 1;
926	tp->rcv_up = ti->ti_seq;
927	Log2(("hit6\n"));
928	goto step6;
929	} /* switch tp->t_state */
930	/*
931	* States other than LISTEN or SYN_SENT.
932	* First check timestamp, if present.
933	* Then check that at least some bytes of segment are within
934	* receive window. If segment begins before rcv_nxt,
935	* drop leading data (and SYN); if nothing left, just ack.
936	*
937	* RFC 1323 PAWS: If we have a timestamp reply on this segment
938	* and it's less than ts_recent, drop it.
939	*/
940	#if 0
941	if ( ts_present
942	&& (tiflags & TH_RST) == 0
943	&& tp->ts_recent
944	&& TSTMP_LT(ts_val, tp->ts_recent))
945	{
946	/* Check to see if ts_recent is over 24 days old. */
947	if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE)
948	{
949	/*
950	* Invalidate ts_recent. If this segment updates
951	* ts_recent, the age will be reset later and ts_recent
952	* will get a valid value. If it does not, setting
953	* ts_recent to zero will at least satisfy the
954	* requirement that zero be placed in the timestamp
955	* echo reply when ts_recent isn't valid. The
956	* age isn't reset until we get a valid ts_recent
957	* because we don't want out-of-order segments to be
958	* dropped when ts_recent is old.
959	*/
960	tp->ts_recent = 0;
961	}
962	else
963	{
964	tcpstat.tcps_rcvduppack++;
965	tcpstat.tcps_rcvdupbyte += ti->ti_len;
966	tcpstat.tcps_pawsdrop++;
967	goto dropafterack;
968	}
969	}
970	#endif
971
972	todrop = tp->rcv_nxt - ti->ti_seq;
973	if (todrop > 0)
974	{
975	if (tiflags & TH_SYN)
976	{
977	tiflags &= ~TH_SYN;
978	ti->ti_seq++;
979	if (ti->ti_urp > 1)
980	ti->ti_urp--;
981	else
982	tiflags &= ~TH_URG;
983	todrop--;
984	}
985	/*
986	* Following if statement from Stevens, vol. 2, p. 960.
987	*/
988	if ( todrop > ti->ti_len
989	\|\| ( todrop == ti->ti_len
990	&& (tiflags & TH_FIN) == 0))
991	{
992	/*
993	* Any valid FIN must be to the left of the window.
994	* At this point the FIN must be a duplicate or out
995	* of sequence; drop it.
996	*/
997	tiflags &= ~TH_FIN;
998
999	/*
1000	* Send an ACK to resynchronize and drop any data.
1001	* But keep on processing for RST or ACK.
1002	*/
1003	tp->t_flags \|= TF_ACKNOW;
1004	todrop = ti->ti_len;
1005	tcpstat.tcps_rcvduppack++;
1006	tcpstat.tcps_rcvdupbyte += todrop;
1007	}
1008	else
1009	{
1010	tcpstat.tcps_rcvpartduppack++;
1011	tcpstat.tcps_rcvpartdupbyte += todrop;
1012	}
1013	m_adj(m, todrop);
1014	ti->ti_seq += todrop;
1015	ti->ti_len -= todrop;
1016	if (ti->ti_urp > todrop)
1017	ti->ti_urp -= todrop;
1018	else
1019	{
1020	tiflags &= ~TH_URG;
1021	ti->ti_urp = 0;
1022	}
1023	}
1024	/*
1025	* If new data are received on a connection after the
1026	* user processes are gone, then RST the other end.
1027	*/
1028	if ( (so->so_state & SS_NOFDREF)
1029	&& tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len)
1030	{
1031	tp = tcp_close(pData, tp);
1032	tcpstat.tcps_rcvafterclose++;
1033	goto dropwithreset;
1034	}
1035
1036	/*
1037	* If segment ends after window, drop trailing data
1038	* (and PUSH and FIN); if nothing left, just ACK.
1039	*/
1040	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1041	if (todrop > 0)
1042	{
1043	tcpstat.tcps_rcvpackafterwin++;
1044	if (todrop >= ti->ti_len)
1045	{
1046	tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1047	/*
1048	* If a new connection request is received
1049	* while in TIME_WAIT, drop the old connection
1050	* and start over if the sequence numbers
1051	* are above the previous ones.
1052	*/
1053	if ( tiflags & TH_SYN
1054	&& tp->t_state == TCPS_TIME_WAIT
1055	&& SEQ_GT(ti->ti_seq, tp->rcv_nxt))
1056	{
1057	iss = tp->rcv_nxt + TCP_ISSINCR;
1058	tp = tcp_close(pData, tp);
1059	SOCKET_UNLOCK(tp->t_socket);
1060	goto findso;
1061	}
1062	/*
1063	* If window is closed can only take segments at
1064	* window edge, and have to drop data and PUSH from
1065	* incoming segments. Continue processing, but
1066	* remember to ack. Otherwise, drop segment
1067	* and ack.
1068	*/
1069	if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt)
1070	{
1071	tp->t_flags \|= TF_ACKNOW;
1072	tcpstat.tcps_rcvwinprobe++;
1073	}
1074	else
1075	goto dropafterack;
1076	}
1077	else
1078	tcpstat.tcps_rcvbyteafterwin += todrop;
1079	m_adj(m, -todrop);
1080	ti->ti_len -= todrop;
1081	tiflags &= ~(TH_PUSH\|TH_FIN);
1082	}
1083
1084	/*
1085	* If last ACK falls within this segment's sequence numbers,
1086	* record its timestamp.
1087	*/
1088	#if 0
1089	if ( ts_present
1090	&& SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)
1091	&& SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + ((tiflags & (TH_SYN\|TH_FIN)) != 0)))
1092	{
1093	tp->ts_recent_age = tcp_now;
1094	tp->ts_recent = ts_val;
1095	}
1096	#endif
1097
1098	/*
1099	* If the RST bit is set examine the state:
1100	* SYN_RECEIVED STATE:
1101	* If passive open, return to LISTEN state.
1102	* If active open, inform user that connection was refused.
1103	* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1104	* Inform user that connection was reset, and close tcb.
1105	* CLOSING, LAST_ACK, TIME_WAIT STATES
1106	* Close the tcb.
1107	*/
1108	if (tiflags&TH_RST)
1109	switch (tp->t_state)
1110	{
1111	case TCPS_SYN_RECEIVED:
1112	/* so->so_error = ECONNREFUSED; */
1113	goto close;
1114
1115	case TCPS_ESTABLISHED:
1116	case TCPS_FIN_WAIT_1:
1117	case TCPS_FIN_WAIT_2:
1118	case TCPS_CLOSE_WAIT:
1119	/* so->so_error = ECONNRESET; */
1120	close:
1121	tp->t_state = TCPS_CLOSED;
1122	tcpstat.tcps_drops++;
1123	tp = tcp_close(pData, tp);
1124	goto drop;
1125
1126	case TCPS_CLOSING:
1127	case TCPS_LAST_ACK:
1128	case TCPS_TIME_WAIT:
1129	tp = tcp_close(pData, tp);
1130	goto drop;
1131	}
1132
1133	/*
1134	* If a SYN is in the window, then this is an
1135	* error and we send an RST and drop the connection.
1136	*/
1137	if (tiflags & TH_SYN)
1138	{
1139	tp = tcp_drop(pData, tp, 0);
1140	goto dropwithreset;
1141	}
1142
1143	/*
1144	* If the ACK bit is off we drop the segment and return.
1145	*/
1146	if ((tiflags & TH_ACK) == 0)
1147	{
1148	goto drop;
1149	}
1150
1151	/*
1152	* Ack processing.
1153	*/
1154	switch (tp->t_state)
1155	{
1156	/*
1157	* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1158	* ESTABLISHED state and continue processing, otherwise
1159	* send an RST. una<=ack<=max
1160	*/
1161	case TCPS_SYN_RECEIVED:
1162	if ( SEQ_GT(tp->snd_una, ti->ti_ack)
1163	\|\| SEQ_GT(ti->ti_ack, tp->snd_max))
1164	goto dropwithreset;
1165	tcpstat.tcps_connects++;
1166	tp->t_state = TCPS_ESTABLISHED;
1167	/*
1168	* The sent SYN is ack'ed with our sequence number +1
1169	* The first data byte already in the buffer will get
1170	* lost if no correction is made. This is only needed for
1171	* SS_CTL since the buffer is empty otherwise.
1172	* tp->snd_una++; or:
1173	*/
1174	tp->snd_una = ti->ti_ack;
1175	soisfconnected(so);
1176
1177	/* Do window scaling? */
1178	#if 0
1179	if ( (tp->t_flags & (TF_RCVD_SCALE\|TF_REQ_SCALE))
1180	== (TF_RCVD_SCALE\|TF_REQ_SCALE))
1181	{
1182	tp->snd_scale = tp->requested_s_scale;
1183	tp->rcv_scale = tp->request_r_scale;
1184	}
1185	#endif
1186	(void) tcp_reass(pData, tp, (struct tcphdr )0, (int )0, (struct mbuf *)0);
1187	tp->snd_wl1 = ti->ti_seq - 1;
1188	/* Avoid ack processing; snd_una==ti_ack => dup ack */
1189	goto synrx_to_est;
1190	/* fall into ... */
1191
1192	/*
1193	* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1194	* ACKs. If the ack is in the range
1195	* tp->snd_una < ti->ti_ack <= tp->snd_max
1196	* then advance tp->snd_una to ti->ti_ack and drop
1197	* data from the retransmission queue. If this ACK reflects
1198	* more up to date window information we update our window information.
1199	*/
1200	case TCPS_ESTABLISHED:
1201	case TCPS_FIN_WAIT_1:
1202	case TCPS_FIN_WAIT_2:
1203	case TCPS_CLOSE_WAIT:
1204	case TCPS_CLOSING:
1205	case TCPS_LAST_ACK:
1206	case TCPS_TIME_WAIT:
1207	if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
1208	{
1209	if (ti->ti_len == 0 && tiwin == tp->snd_wnd)
1210	{
1211	tcpstat.tcps_rcvdupack++;
1212	Log2((" dup ack m = %lx, so = %lx\n", (long)m, (long)so));
1213	/*
1214	* If we have outstanding data (other than
1215	* a window probe), this is a completely
1216	* duplicate ack (ie, window info didn't
1217	* change), the ack is the biggest we've
1218	* seen and we've seen exactly our rexmt
1219	* threshold of them, assume a packet
1220	* has been dropped and retransmit it.
1221	* Kludge snd_nxt & the congestion
1222	* window so we send only this one
1223	* packet.
1224	*
1225	* We know we're losing at the current
1226	* window size so do congestion avoidance
1227	* (set ssthresh to half the current window
1228	* and pull our congestion window back to
1229	* the new ssthresh).
1230	*
1231	* Dup acks mean that packets have left the
1232	* network (they're now cached at the receiver)
1233	* so bump cwnd by the amount in the receiver
1234	* to keep a constant cwnd packets in the
1235	* network.
1236	*/
1237	if ( tp->t_timer[TCPT_REXMT] == 0
1238	\|\| ti->ti_ack != tp->snd_una)
1239	tp->t_dupacks = 0;
1240	else if (++tp->t_dupacks == tcprexmtthresh)
1241	{
1242	tcp_seq onxt = tp->snd_nxt;
1243	u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
1244	if (win < 2)
1245	win = 2;
1246	tp->snd_ssthresh = win * tp->t_maxseg;
1247	tp->t_timer[TCPT_REXMT] = 0;
1248	tp->t_rtt = 0;
1249	tp->snd_nxt = ti->ti_ack;
1250	tp->snd_cwnd = tp->t_maxseg;
1251	(void) tcp_output(pData, tp);
1252	tp->snd_cwnd = tp->snd_ssthresh +
1253	tp->t_maxseg * tp->t_dupacks;
1254	if (SEQ_GT(onxt, tp->snd_nxt))
1255	tp->snd_nxt = onxt;
1256	goto drop;
1257	}
1258	else if (tp->t_dupacks > tcprexmtthresh)
1259	{
1260	tp->snd_cwnd += tp->t_maxseg;
1261	(void) tcp_output(pData, tp);
1262	goto drop;
1263	}
1264	}
1265	else
1266	tp->t_dupacks = 0;
1267	break;
1268	}
1269	synrx_to_est:
1270	/*
1271	* If the congestion window was inflated to account
1272	* for the other side's cached packets, retract it.
1273	*/
1274	if ( tp->t_dupacks > tcprexmtthresh
1275	&& tp->snd_cwnd > tp->snd_ssthresh)
1276	tp->snd_cwnd = tp->snd_ssthresh;
1277	tp->t_dupacks = 0;
1278	if (SEQ_GT(ti->ti_ack, tp->snd_max))
1279	{
1280	tcpstat.tcps_rcvacktoomuch++;
1281	goto dropafterack;
1282	}
1283	acked = ti->ti_ack - tp->snd_una;
1284	tcpstat.tcps_rcvackpack++;
1285	tcpstat.tcps_rcvackbyte += acked;
1286
1287	/*
1288	* If we have a timestamp reply, update smoothed
1289	* round trip time. If no timestamp is present but
1290	* transmit timer is running and timed sequence
1291	* number was acked, update smoothed round trip time.
1292	* Since we now have an rtt measurement, cancel the
1293	* timer backoff (cf., Phil Karn's retransmit alg.).
1294	* Recompute the initial retransmit timer.
1295	*/
1296	#if 0
1297	if (ts_present)
1298	tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1299	else
1300	#endif
1301	if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1302	tcp_xmit_timer(pData, tp, tp->t_rtt);
1303
1304	/*
1305	* If all outstanding data is acked, stop retransmit
1306	* timer and remember to restart (more output or persist).
1307	* If there is more data to be acked, restart retransmit
1308	* timer, using current (possibly backed-off) value.
1309	*/
1310	if (ti->ti_ack == tp->snd_max)
1311	{
1312	tp->t_timer[TCPT_REXMT] = 0;
1313	needoutput = 1;
1314	}
1315	else if (tp->t_timer[TCPT_PERSIST] == 0)
1316	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1317	/*
1318	* When new data is acked, open the congestion window.
1319	* If the window gives us less than ssthresh packets
1320	* in flight, open exponentially (maxseg per packet).
1321	* Otherwise open linearly: maxseg per window
1322	* (maxseg^2 / cwnd per packet).
1323	*/
1324	{
1325	register u_int cw = tp->snd_cwnd;
1326	register u_int incr = tp->t_maxseg;
1327
1328	if (cw > tp->snd_ssthresh)
1329	incr = incr * incr / cw;
1330	tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1331	}
1332	if (acked > SBUF_LEN(&so->so_snd))
1333	{
1334	tp->snd_wnd -= SBUF_LEN(&so->so_snd);
1335	#ifndef VBOX_WITH_SLIRP_BSD_SBUF
1336	sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1337	#else
1338	sbuf_clear(&so->so_snd);
1339	#endif
1340	ourfinisacked = 1;
1341	}
1342	else
1343	{
1344	#ifndef VBOX_WITH_SLIRP_BSD_SBUF
1345	sbdrop(&so->so_snd, acked);
1346	#else
1347	sbuf_setpos(&so->so_snd, sbuf_len(&so->so_snd) - acked);
1348	#endif
1349	tp->snd_wnd -= acked;
1350	ourfinisacked = 0;
1351	}
1352	/*
1353	* XXX sowwakup is called when data is acked and there's room for
1354	* for more data... it should read() the socket
1355	*/
1356	#if 0
1357	if (so->so_snd.sb_flags & SB_NOTIFY)
1358	sowwakeup(so);
1359	#endif
1360	tp->snd_una = ti->ti_ack;
1361	if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1362	tp->snd_nxt = tp->snd_una;
1363
1364	switch (tp->t_state)
1365	{
1366	/*
1367	* In FIN_WAIT_1 STATE in addition to the processing
1368	* for the ESTABLISHED state if our FIN is now acknowledged
1369	* then enter FIN_WAIT_2.
1370	*/
1371	case TCPS_FIN_WAIT_1:
1372	if (ourfinisacked)
1373	{
1374	/*
1375	* If we can't receive any more
1376	* data, then closing user can proceed.
1377	* Starting the timer is contrary to the
1378	* specification, but if we don't get a FIN
1379	* we'll hang forever.
1380	*/
1381	if (so->so_state & SS_FCANTRCVMORE)
1382	{
1383	soisfdisconnected(so);
1384	tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1385	}
1386	tp->t_state = TCPS_FIN_WAIT_2;
1387	}
1388	break;
1389
1390	/*
1391	* In CLOSING STATE in addition to the processing for
1392	* the ESTABLISHED state if the ACK acknowledges our FIN
1393	* then enter the TIME-WAIT state, otherwise ignore
1394	* the segment.
1395	*/
1396	case TCPS_CLOSING:
1397	if (ourfinisacked)
1398	{
1399	tp->t_state = TCPS_TIME_WAIT;
1400	tcp_canceltimers(tp);
1401	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1402	soisfdisconnected(so);
1403	}
1404	break;
1405
1406	/*
1407	* In LAST_ACK, we may still be waiting for data to drain
1408	* and/or to be acked, as well as for the ack of our FIN.
1409	* If our FIN is now acknowledged, delete the TCB,
1410	* enter the closed state and return.
1411	*/
1412	case TCPS_LAST_ACK:
1413	if (ourfinisacked)
1414	{
1415	tp = tcp_close(pData, tp);
1416	goto drop;
1417	}
1418	break;
1419
1420	/*
1421	* In TIME_WAIT state the only thing that should arrive
1422	* is a retransmission of the remote FIN. Acknowledge
1423	* it and restart the finack timer.
1424	*/
1425	case TCPS_TIME_WAIT:
1426	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1427	goto dropafterack;
1428	}
1429	} /* switch(tp->t_state) */
1430
1431	step6:
1432	/*
1433	* Update window information.
1434	* Don't look at window if no ACK: TAC's send garbage on first SYN.
1435	*/
1436	if ( (tiflags & TH_ACK)
1437	&& ( SEQ_LT(tp->snd_wl1, ti->ti_seq)
1438	\|\| ( tp->snd_wl1 == ti->ti_seq
1439	&& ( SEQ_LT(tp->snd_wl2, ti->ti_ack)
1440	\|\| ( tp->snd_wl2 == ti->ti_ack
1441	&& tiwin > tp->snd_wnd)))))
1442	{
1443	/* keep track of pure window updates */
1444	if ( ti->ti_len == 0
1445	&& tp->snd_wl2 == ti->ti_ack
1446	&& tiwin > tp->snd_wnd)
1447	tcpstat.tcps_rcvwinupd++;
1448	tp->snd_wnd = tiwin;
1449	tp->snd_wl1 = ti->ti_seq;
1450	tp->snd_wl2 = ti->ti_ack;
1451	if (tp->snd_wnd > tp->max_sndwnd)
1452	tp->max_sndwnd = tp->snd_wnd;
1453	needoutput = 1;
1454	}
1455
1456	/*
1457	* Process segments with URG.
1458	*/
1459	if ((tiflags & TH_URG) && ti->ti_urp &&
1460	TCPS_HAVERCVDFIN(tp->t_state) == 0)
1461	{
1462	/* BSD's sbufs are auto extent so we shouldn't worry here */
1463	#ifndef VBOX_WITH_SLIRP_BSD_SBUF
1464	/*
1465	* This is a kludge, but if we receive and accept
1466	* random urgent pointers, we'll crash in
1467	* soreceive. It's hard to imagine someone
1468	* actually wanting to send this much urgent data.
1469	*/
1470	if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen)
1471	{
1472	ti->ti_urp = 0;
1473	tiflags &= ~TH_URG;
1474	goto dodata;
1475	}
1476	#endif
1477	/*
1478	* If this segment advances the known urgent pointer,
1479	* then mark the data stream. This should not happen
1480	* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1481	* a FIN has been received from the remote side.
1482	* In these states we ignore the URG.
1483	*
1484	* According to RFC961 (Assigned Protocols),
1485	* the urgent pointer points to the last octet
1486	* of urgent data. We continue, however,
1487	* to consider it to indicate the first octet
1488	* of data past the urgent section as the original
1489	* spec states (in one of two places).
1490	*/
1491	if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up))
1492	{
1493	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1494	so->so_urgc = SBUF_LEN(&so->so_rcv) +
1495	(tp->rcv_up - tp->rcv_nxt); /* -1; */
1496	tp->rcv_up = ti->ti_seq + ti->ti_urp;
1497	}
1498	}
1499	else
1500	/*
1501	* If no out of band data is expected,
1502	* pull receive urgent pointer along
1503	* with the receive window.
1504	*/
1505	if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1506	tp->rcv_up = tp->rcv_nxt;
1507	dodata:
1508
1509	/*
1510	* If this is a small packet, then ACK now - with Nagel
1511	* congestion avoidance sender won't send more until
1512	* he gets an ACK.
1513	*
1514	* See above.
1515	*/
1516	if ( ti->ti_len
1517	&& (unsigned)ti->ti_len <= 5
1518	&& ((struct tcpiphdr_2 *)ti)->first_char == (char)27)
1519	{
1520	tp->t_flags \|= TF_ACKNOW;
1521	}
1522
1523	/*
1524	* Process the segment text, merging it into the TCP sequencing queue,
1525	* and arranging for acknowledgment of receipt if necessary.
1526	* This process logically involves adjusting tp->rcv_wnd as data
1527	* is presented to the user (this happens in tcp_usrreq.c,
1528	* case PRU_RCVD). If a FIN has already been received on this
1529	* connection then we just ignore the text.
1530	*/
1531	if ( (ti->ti_len \|\| (tiflags&TH_FIN))
1532	&& TCPS_HAVERCVDFIN(tp->t_state) == 0)
1533	{
1534	if ( ti->ti_seq == tp->rcv_nxt
1535	&& LIST_EMPTY(&tp->t_segq)
1536	&& tp->t_state == TCPS_ESTABLISHED)
1537	{
1538	DELAY_ACK(tp, ti); /* little bit different from BSD declaration see netinet/tcp_input.c */
1539	tp->rcv_nxt += tlen;
1540	tiflags = ti->ti_t.th_flags & TH_FIN;
1541	tcpstat.tcps_rcvpack++;
1542	tcpstat.tcps_rcvbyte += tlen;
1543	if (so->so_state & SS_FCANTRCVMORE)
1544	m_freem(pData, m);
1545	else
1546	sbappend(pData, so, m);
1547	}
1548	else
1549	{
1550	tiflags = tcp_reass(pData, tp, &ti->ti_t, &tlen, m);
1551	tiflags \|= TF_ACKNOW;
1552	}
1553	/*
1554	* Note the amount of data that peer has sent into
1555	* our window, in order to estimate the sender's
1556	* buffer size.
1557	*/
1558	len = SBUF_SIZE(&so->so_rcv) - (tp->rcv_adv - tp->rcv_nxt);
1559	}
1560	else
1561	{
1562	m_freem(pData, m);
1563	tiflags &= ~TH_FIN;
1564	}
1565
1566	/*
1567	* If FIN is received ACK the FIN and let the user know
1568	* that the connection is closing.
1569	*/
1570	if (tiflags & TH_FIN)
1571	{
1572	if (TCPS_HAVERCVDFIN(tp->t_state) == 0)
1573	{
1574	/*
1575	* If we receive a FIN we can't send more data,
1576	* set it SS_FDRAIN
1577	* Shutdown the socket if there is no rx data in the
1578	* buffer.
1579	* soread() is called on completion of shutdown() and
1580	* will got to TCPS_LAST_ACK, and use tcp_output()
1581	* to send the FIN.
1582	*/
1583	/* sofcantrcvmore(so); */
1584	sofwdrain(so);
1585
1586	tp->t_flags \|= TF_ACKNOW;
1587	tp->rcv_nxt++;
1588	}
1589	switch (tp->t_state)
1590	{
1591	/*
1592	* In SYN_RECEIVED and ESTABLISHED STATES
1593	* enter the CLOSE_WAIT state.
1594	*/
1595	case TCPS_SYN_RECEIVED:
1596	case TCPS_ESTABLISHED:
1597	tp->t_state = TCPS_CLOSE_WAIT;
1598	break;
1599
1600	/*
1601	* If still in FIN_WAIT_1 STATE FIN has not been acked so
1602	* enter the CLOSING state.
1603	*/
1604	case TCPS_FIN_WAIT_1:
1605	tp->t_state = TCPS_CLOSING;
1606	break;
1607
1608	/*
1609	* In FIN_WAIT_2 state enter the TIME_WAIT state,
1610	* starting the time-wait timer, turning off the other
1611	* standard timers.
1612	*/
1613	case TCPS_FIN_WAIT_2:
1614	tp->t_state = TCPS_TIME_WAIT;
1615	tcp_canceltimers(tp);
1616	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1617	soisfdisconnected(so);
1618	break;
1619
1620	/*
1621	* In TIME_WAIT state restart the 2 MSL time_wait timer.
1622	*/
1623	case TCPS_TIME_WAIT:
1624	tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1625	break;
1626	}
1627	}
1628
1629	/*
1630	* Return any desired output.
1631	*/
1632	if (needoutput \|\| (tp->t_flags & TF_ACKNOW))
1633	tcp_output(pData, tp);
1634
1635	SOCKET_UNLOCK(so);
1636	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1637	return;
1638
1639	dropafterack:
1640	Log2(("drop after ack\n"));
1641	/*
1642	* Generate an ACK dropping incoming segment if it occupies
1643	* sequence space, where the ACK reflects our state.
1644	*/
1645	if (tiflags & TH_RST)
1646	goto drop;
1647	m_freem(pData, m);
1648	tp->t_flags \|= TF_ACKNOW;
1649	(void) tcp_output(pData, tp);
1650	SOCKET_UNLOCK(so);
1651	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1652	return;
1653
1654	dropwithreset:
1655	/* reuses m if m!=NULL, m_free() unnecessary */
1656	if (tiflags & TH_ACK)
1657	tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1658	else
1659	{
1660	if (tiflags & TH_SYN)
1661	ti->ti_len++;
1662	tcp_respond(pData, tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1663	TH_RST\|TH_ACK);
1664	}
1665
1666	if (so != &tcb)
1667	SOCKET_UNLOCK(so);
1668	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1669	return;
1670
1671	drop:
1672	/*
1673	* Drop space held by incoming segment and return.
1674	*/
1675	m_freem(pData, m);
1676
1677	#ifdef VBOX_WITH_SLIRP_MT
1678	if (RTCritSectIsOwned(&so->so_mutex))
1679	{
1680	SOCKET_UNLOCK(so);
1681	}
1682	#endif
1683
1684	STAM_PROFILE_STOP(&pData->StatTCP_input, counter_input);
1685	return;
1686	}
1687
1688	void
1689	tcp_dooptions(PNATState pData, struct tcpcb tp, u_char cp, int cnt, struct tcpiphdr *ti)
1690	{
1691	u_int16_t mss;
1692	int opt, optlen;
1693
1694	LogFlow(("tcp_dooptions: tp = %lx, cnt=%i\n", (long)tp, cnt));
1695
1696	for (; cnt > 0; cnt -= optlen, cp += optlen)
1697	{
1698	opt = cp[0];
1699	if (opt == TCPOPT_EOL)
1700	break;
1701	if (opt == TCPOPT_NOP)
1702	optlen = 1;
1703	else
1704	{
1705	optlen = cp[1];
1706	if (optlen <= 0)
1707	break;
1708	}
1709	switch (opt)
1710	{
1711	default:
1712	continue;
1713
1714	case TCPOPT_MAXSEG:
1715	if (optlen != TCPOLEN_MAXSEG)
1716	continue;
1717	if (!(ti->ti_flags & TH_SYN))
1718	continue;
1719	memcpy((char ) &mss, (char ) cp + 2, sizeof(mss));
1720	NTOHS(mss);
1721	(void) tcp_mss(pData, tp, mss); /* sets t_maxseg */
1722	break;
1723
1724	#if 0
1725	case TCPOPT_WINDOW:
1726	if (optlen != TCPOLEN_WINDOW)
1727	continue;
1728	if (!(ti->ti_flags & TH_SYN))
1729	continue;
1730	tp->t_flags \|= TF_RCVD_SCALE;
1731	tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1732	break;
1733
1734	case TCPOPT_TIMESTAMP:
1735	if (optlen != TCPOLEN_TIMESTAMP)
1736	continue;
1737	*ts_present = 1;
1738	memcpy((char ) ts_val, (char )cp + 2, sizeof(*ts_val));
1739	NTOHL(*ts_val);
1740	memcpy((char ) ts_ecr, (char )cp + 6, sizeof(*ts_ecr));
1741	NTOHL(*ts_ecr);
1742
1743	/*
1744	* A timestamp received in a SYN makes
1745	* it ok to send timestamp requests and replies.
1746	*/
1747	if (ti->ti_flags & TH_SYN)
1748	{
1749	tp->t_flags \|= TF_RCVD_TSTMP;
1750	tp->ts_recent = *ts_val;
1751	tp->ts_recent_age = tcp_now;
1752	}
1753	break;
1754	#endif
1755	}
1756	}
1757	}
1758
1759
1760	/*
1761	* Pull out of band byte out of a segment so
1762	* it doesn't appear in the user's data queue.
1763	* It is still reflected in the segment length for
1764	* sequencing purposes.
1765	*/
1766
1767	#if 0
1768	void
1769	tcp_pulloutofband(struct socket so, struct tcpiphdr ti, struct mbuf *m)
1770	{
1771	int cnt = ti->ti_urp - 1;
1772
1773	while (cnt >= 0)
1774	{
1775	if (m->m_len > cnt)
1776	{
1777	char *cp = mtod(m, caddr_t) + cnt;
1778	struct tcpcb *tp = sototcpcb(so);
1779
1780	tp->t_iobc = *cp;
1781	tp->t_oobflags \|= TCPOOB_HAVEDATA;
1782	memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1783	m->m_len--;
1784	return;
1785	}
1786	cnt -= m->m_len;
1787	m = m->m_next; /* XXX WRONG! Fix it! */
1788	if (m == 0)
1789	break;
1790	}
1791	panic("tcp_pulloutofband");
1792	}
1793	#endif
1794
1795	/*
1796	* Collect new round-trip time estimate
1797	* and update averages and current timeout.
1798	*/
1799
1800	void
1801	tcp_xmit_timer(PNATState pData, register struct tcpcb *tp, int rtt)
1802	{
1803	register short delta;
1804
1805	LogFlow(("tcp_xmit_timer: tp = %lx rtt = %d\n", (long)tp, rtt));
1806
1807	tcpstat.tcps_rttupdated++;
1808	if (tp->t_srtt != 0)
1809	{
1810	/*
1811	* srtt is stored as fixed point with 3 bits after the
1812	* binary point (i.e., scaled by 8). The following magic
1813	* is equivalent to the smoothing algorithm in rfc793 with
1814	* an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1815	* point). Adjust rtt to origin 0.
1816	*/
1817	delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1818	if ((tp->t_srtt += delta) <= 0)
1819	tp->t_srtt = 1;
1820	/*
1821	* We accumulate a smoothed rtt variance (actually, a
1822	* smoothed mean difference), then set the retransmit
1823	* timer to smoothed rtt + 4 times the smoothed variance.
1824	* rttvar is stored as fixed point with 2 bits after the
1825	* binary point (scaled by 4). The following is
1826	* equivalent to rfc793 smoothing with an alpha of .75
1827	* (rttvar = rttvar*3/4 + \|delta\| / 4). This replaces
1828	* rfc793's wired-in beta.
1829	*/
1830	if (delta < 0)
1831	delta = -delta;
1832	delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1833	if ((tp->t_rttvar += delta) <= 0)
1834	tp->t_rttvar = 1;
1835	}
1836	else
1837	{
1838	/*
1839	* No rtt measurement yet - use the unsmoothed rtt.
1840	* Set the variance to half the rtt (so our first
1841	* retransmit happens at 3*rtt).
1842	*/
1843	tp->t_srtt = rtt << TCP_RTT_SHIFT;
1844	tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1845	}
1846	tp->t_rtt = 0;
1847	tp->t_rxtshift = 0;
1848
1849	/*
1850	* the retransmit should happen at rtt + 4 * rttvar.
1851	* Because of the way we do the smoothing, srtt and rttvar
1852	* will each average +1/2 tick of bias. When we compute
1853	* the retransmit timer, we want 1/2 tick of rounding and
1854	* 1 extra tick because of +-1/2 tick uncertainty in the
1855	* firing of the timer. The bias will give us exactly the
1856	* 1.5 tick we need. But, because the bias is
1857	* statistical, we have to test that we don't drop below
1858	* the minimum feasible timer (which is 2 ticks).
1859	*/
1860	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1861	(short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1862
1863	/*
1864	* We received an ack for a packet that wasn't retransmitted;
1865	* it is probably safe to discard any error indications we've
1866	* received recently. This isn't quite right, but close enough
1867	* for now (a route might have failed after we sent a segment,
1868	* and the return path might not be symmetrical).
1869	*/
1870	tp->t_softerror = 0;
1871	}
1872
1873	/*
1874	* Determine a reasonable value for maxseg size.
1875	* If the route is known, check route for mtu.
1876	* If none, use an mss that can be handled on the outgoing
1877	* interface without forcing IP to fragment; if bigger than
1878	* an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1879	* to utilize large mbufs. If no route is found, route has no mtu,
1880	* or the destination isn't local, use a default, hopefully conservative
1881	* size (usually 512 or the default IP max size, but no more than the mtu
1882	* of the interface), as we can't discover anything about intervening
1883	* gateways or networks. We also initialize the congestion/slow start
1884	* window to be a single segment if the destination isn't local.
1885	* While looking at the routing entry, we also initialize other path-dependent
1886	* parameters from pre-set or cached values in the routing entry.
1887	*/
1888
1889	int
1890	tcp_mss(PNATState pData, register struct tcpcb *tp, u_int offer)
1891	{
1892	struct socket *so = tp->t_socket;
1893	int mss;
1894
1895	LogFlow(("tcp_mss: tp = %lx, offet = %d\n", (long)tp, offer));
1896
1897	mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1898	if (offer)
1899	mss = min(mss, offer);
1900	mss = max(mss, 32);
1901	if (mss < tp->t_maxseg \|\| offer != 0)
1902	tp->t_maxseg = mss;
1903
1904	tp->snd_cwnd = mss;
1905
1906	#ifndef VBOX_WITH_SLIRP_BSD_SBUF
1907	sbreserve(pData, &so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1908	sbreserve(pData, &so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1909	#else
1910	sbuf_new(&so->so_snd, NULL, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0), SBUF_AUTOEXTEND);
1911	sbuf_new(&so->so_rcv, NULL, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0), SBUF_AUTOEXTEND);
1912	#endif
1913
1914	Log2((" returning mss = %d\n", mss));
1915
1916	return mss;
1917	}

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/tcp_input.c@ 35923

Download in other formats: