Remove the advertising clause from the slirp license
[qemu] / slirp / tcp_input.c
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *      @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
30  * tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
31  */
32
33 /*
34  * Changes and additions relating to SLiRP
35  * Copyright (c) 1995 Danny Gasparovski.
36  *
37  * Please read the file COPYRIGHT for the
38  * terms and conditions of the copyright.
39  */
40
41 #include <slirp.h>
42 #include "ip_icmp.h"
43
44 struct socket tcb;
45
46 #define TCPREXMTTHRESH 3
47 struct  socket *tcp_last_so = &tcb;
48
49 tcp_seq tcp_iss;                /* tcp initial send seq # */
50
51 #define TCP_PAWS_IDLE   (24 * 24 * 60 * 60 * PR_SLOWHZ)
52
53 /* for modulo comparisons of timestamps */
54 #define TSTMP_LT(a,b)   ((int)((a)-(b)) < 0)
55 #define TSTMP_GEQ(a,b)  ((int)((a)-(b)) >= 0)
56
57 /*
58  * Insert segment ti into reassembly queue of tcp with
59  * control block tp.  Return TH_FIN if reassembly now includes
60  * a segment with FIN.  The macro form does the common case inline
61  * (segment is the next to be received on an established connection,
62  * and the queue is empty), avoiding linkage into and removal
63  * from the queue and repetition of various conversions.
64  * Set DELACK for segments received in order, but ack immediately
65  * when segments are out of order (so fast retransmit can work).
66  */
67 #ifdef TCP_ACK_HACK
68 #define TCP_REASS(tp, ti, m, so, flags) {\
69        if ((ti)->ti_seq == (tp)->rcv_nxt && \
70            tcpfrag_list_empty(tp) && \
71            (tp)->t_state == TCPS_ESTABLISHED) {\
72                if (ti->ti_flags & TH_PUSH) \
73                        tp->t_flags |= TF_ACKNOW; \
74                else \
75                        tp->t_flags |= TF_DELACK; \
76                (tp)->rcv_nxt += (ti)->ti_len; \
77                flags = (ti)->ti_flags & TH_FIN; \
78                STAT(tcpstat.tcps_rcvpack++);         \
79                STAT(tcpstat.tcps_rcvbyte += (ti)->ti_len);   \
80                if (so->so_emu) { \
81                        if (tcp_emu((so),(m))) sbappend((so), (m)); \
82                } else \
83                        sbappend((so), (m)); \
84 /*               sorwakeup(so); */ \
85         } else {\
86                (flags) = tcp_reass((tp), (ti), (m)); \
87                tp->t_flags |= TF_ACKNOW; \
88        } \
89 }
90 #else
91 #define TCP_REASS(tp, ti, m, so, flags) { \
92         if ((ti)->ti_seq == (tp)->rcv_nxt && \
93         tcpfrag_list_empty(tp) && \
94             (tp)->t_state == TCPS_ESTABLISHED) { \
95                 tp->t_flags |= TF_DELACK; \
96                 (tp)->rcv_nxt += (ti)->ti_len; \
97                 flags = (ti)->ti_flags & TH_FIN; \
98                 STAT(tcpstat.tcps_rcvpack++);        \
99                 STAT(tcpstat.tcps_rcvbyte += (ti)->ti_len);  \
100                 if (so->so_emu) { \
101                         if (tcp_emu((so),(m))) sbappend(so, (m)); \
102                 } else \
103                         sbappend((so), (m)); \
104 /*              sorwakeup(so); */ \
105         } else { \
106                 (flags) = tcp_reass((tp), (ti), (m)); \
107                 tp->t_flags |= TF_ACKNOW; \
108         } \
109 }
110 #endif
111 static void tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt,
112                           struct tcpiphdr *ti);
113 static void tcp_xmit_timer(register struct tcpcb *tp, int rtt);
114
115 static int
116 tcp_reass(register struct tcpcb *tp, register struct tcpiphdr *ti,
117           struct mbuf *m)
118 {
119         register struct tcpiphdr *q;
120         struct socket *so = tp->t_socket;
121         int flags;
122
123         /*
124          * Call with ti==0 after become established to
125          * force pre-ESTABLISHED data up to user socket.
126          */
127         if (ti == 0)
128                 goto present;
129
130         /*
131          * Find a segment which begins after this one does.
132          */
133         for (q = tcpfrag_list_first(tp); !tcpfrag_list_end(q, tp);
134             q = tcpiphdr_next(q))
135                 if (SEQ_GT(q->ti_seq, ti->ti_seq))
136                         break;
137
138         /*
139          * If there is a preceding segment, it may provide some of
140          * our data already.  If so, drop the data from the incoming
141          * segment.  If it provides all of our data, drop us.
142          */
143         if (!tcpfrag_list_end(tcpiphdr_prev(q), tp)) {
144                 register int i;
145                 q = tcpiphdr_prev(q);
146                 /* conversion to int (in i) handles seq wraparound */
147                 i = q->ti_seq + q->ti_len - ti->ti_seq;
148                 if (i > 0) {
149                         if (i >= ti->ti_len) {
150                                 STAT(tcpstat.tcps_rcvduppack++);
151                                 STAT(tcpstat.tcps_rcvdupbyte += ti->ti_len);
152                                 m_freem(m);
153                                 /*
154                                  * Try to present any queued data
155                                  * at the left window edge to the user.
156                                  * This is needed after the 3-WHS
157                                  * completes.
158                                  */
159                                 goto present;   /* ??? */
160                         }
161                         m_adj(m, i);
162                         ti->ti_len -= i;
163                         ti->ti_seq += i;
164                 }
165                 q = tcpiphdr_next(q);
166         }
167         STAT(tcpstat.tcps_rcvoopack++);
168         STAT(tcpstat.tcps_rcvoobyte += ti->ti_len);
169         ti->ti_mbuf = m;
170
171         /*
172          * While we overlap succeeding segments trim them or,
173          * if they are completely covered, dequeue them.
174          */
175         while (!tcpfrag_list_end(q, tp)) {
176                 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
177                 if (i <= 0)
178                         break;
179                 if (i < q->ti_len) {
180                         q->ti_seq += i;
181                         q->ti_len -= i;
182                         m_adj(q->ti_mbuf, i);
183                         break;
184                 }
185                 q = tcpiphdr_next(q);
186                 m = tcpiphdr_prev(q)->ti_mbuf;
187                 remque(tcpiphdr2qlink(tcpiphdr_prev(q)));
188                 m_freem(m);
189         }
190
191         /*
192          * Stick new segment in its place.
193          */
194         insque(tcpiphdr2qlink(ti), tcpiphdr2qlink(tcpiphdr_prev(q)));
195
196 present:
197         /*
198          * Present data to user, advancing rcv_nxt through
199          * completed sequence space.
200          */
201         if (!TCPS_HAVEESTABLISHED(tp->t_state))
202                 return (0);
203         ti = tcpfrag_list_first(tp);
204         if (tcpfrag_list_end(ti, tp) || ti->ti_seq != tp->rcv_nxt)
205                 return (0);
206         if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
207                 return (0);
208         do {
209                 tp->rcv_nxt += ti->ti_len;
210                 flags = ti->ti_flags & TH_FIN;
211                 remque(tcpiphdr2qlink(ti));
212                 m = ti->ti_mbuf;
213                 ti = tcpiphdr_next(ti);
214 /*              if (so->so_state & SS_FCANTRCVMORE) */
215                 if (so->so_state & SS_FCANTSENDMORE)
216                         m_freem(m);
217                 else {
218                         if (so->so_emu) {
219                                 if (tcp_emu(so,m)) sbappend(so, m);
220                         } else
221                                 sbappend(so, m);
222                 }
223         } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
224 /*      sorwakeup(so); */
225         return (flags);
226 }
227
228 /*
229  * TCP input routine, follows pages 65-76 of the
230  * protocol specification dated September, 1981 very closely.
231  */
232 void
233 tcp_input(m, iphlen, inso)
234         register struct mbuf *m;
235         int iphlen;
236         struct socket *inso;
237 {
238         struct ip save_ip, *ip;
239         register struct tcpiphdr *ti;
240         caddr_t optp = NULL;
241         int optlen = 0;
242         int len, tlen, off;
243         register struct tcpcb *tp = 0;
244         register int tiflags;
245         struct socket *so = 0;
246         int todrop, acked, ourfinisacked, needoutput = 0;
247 /*      int dropsocket = 0; */
248         int iss = 0;
249         u_long tiwin;
250         int ret;
251 /*      int ts_present = 0; */
252     struct ex_list *ex_ptr;
253
254         DEBUG_CALL("tcp_input");
255         DEBUG_ARGS((dfd," m = %8lx  iphlen = %2d  inso = %lx\n",
256                     (long )m, iphlen, (long )inso ));
257
258         /*
259          * If called with m == 0, then we're continuing the connect
260          */
261         if (m == NULL) {
262                 so = inso;
263
264                 /* Re-set a few variables */
265                 tp = sototcpcb(so);
266                 m = so->so_m;
267                 so->so_m = 0;
268                 ti = so->so_ti;
269                 tiwin = ti->ti_win;
270                 tiflags = ti->ti_flags;
271
272                 goto cont_conn;
273         }
274
275
276         STAT(tcpstat.tcps_rcvtotal++);
277         /*
278          * Get IP and TCP header together in first mbuf.
279          * Note: IP leaves IP header in first mbuf.
280          */
281         ti = mtod(m, struct tcpiphdr *);
282         if (iphlen > sizeof(struct ip )) {
283           ip_stripoptions(m, (struct mbuf *)0);
284           iphlen=sizeof(struct ip );
285         }
286         /* XXX Check if too short */
287
288
289         /*
290          * Save a copy of the IP header in case we want restore it
291          * for sending an ICMP error message in response.
292          */
293         ip=mtod(m, struct ip *);
294         save_ip = *ip;
295         save_ip.ip_len+= iphlen;
296
297         /*
298          * Checksum extended TCP header and data.
299          */
300         tlen = ((struct ip *)ti)->ip_len;
301         tcpiphdr2qlink(ti)->next = tcpiphdr2qlink(ti)->prev = 0;
302     memset(&ti->ti_i.ih_mbuf, 0 , sizeof(struct mbuf_ptr));
303         ti->ti_x1 = 0;
304         ti->ti_len = htons((u_int16_t)tlen);
305         len = sizeof(struct ip ) + tlen;
306         /* keep checksum for ICMP reply
307          * ti->ti_sum = cksum(m, len);
308          * if (ti->ti_sum) { */
309         if(cksum(m, len)) {
310           STAT(tcpstat.tcps_rcvbadsum++);
311           goto drop;
312         }
313
314         /*
315          * Check that TCP offset makes sense,
316          * pull out TCP options and adjust length.              XXX
317          */
318         off = ti->ti_off << 2;
319         if (off < sizeof (struct tcphdr) || off > tlen) {
320           STAT(tcpstat.tcps_rcvbadoff++);
321           goto drop;
322         }
323         tlen -= off;
324         ti->ti_len = tlen;
325         if (off > sizeof (struct tcphdr)) {
326           optlen = off - sizeof (struct tcphdr);
327           optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
328
329                 /*
330                  * Do quick retrieval of timestamp options ("options
331                  * prediction?").  If timestamp is the only option and it's
332                  * formatted as recommended in RFC 1323 appendix A, we
333                  * quickly get the values now and not bother calling
334                  * tcp_dooptions(), etc.
335                  */
336 /*              if ((optlen == TCPOLEN_TSTAMP_APPA ||
337  *                   (optlen > TCPOLEN_TSTAMP_APPA &&
338  *                      optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
339  *                   *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
340  *                   (ti->ti_flags & TH_SYN) == 0) {
341  *                      ts_present = 1;
342  *                      ts_val = ntohl(*(u_int32_t *)(optp + 4));
343  *                      ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
344  *                      optp = NULL;   / * we've parsed the options * /
345  *              }
346  */
347         }
348         tiflags = ti->ti_flags;
349
350         /*
351          * Convert TCP protocol specific fields to host format.
352          */
353         NTOHL(ti->ti_seq);
354         NTOHL(ti->ti_ack);
355         NTOHS(ti->ti_win);
356         NTOHS(ti->ti_urp);
357
358         /*
359          * Drop TCP, IP headers and TCP options.
360          */
361         m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
362         m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
363
364     if (slirp_restrict) {
365         for (ex_ptr = exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next)
366             if (ex_ptr->ex_fport == ti->ti_dport &&
367                     (ntohl(ti->ti_dst.s_addr) & 0xff) == ex_ptr->ex_addr)
368                 break;
369
370         if (!ex_ptr)
371             goto drop;
372     }
373         /*
374          * Locate pcb for segment.
375          */
376 findso:
377         so = tcp_last_so;
378         if (so->so_fport != ti->ti_dport ||
379             so->so_lport != ti->ti_sport ||
380             so->so_laddr.s_addr != ti->ti_src.s_addr ||
381             so->so_faddr.s_addr != ti->ti_dst.s_addr) {
382                 so = solookup(&tcb, ti->ti_src, ti->ti_sport,
383                                ti->ti_dst, ti->ti_dport);
384                 if (so)
385                         tcp_last_so = so;
386                 STAT(tcpstat.tcps_socachemiss++);
387         }
388
389         /*
390          * If the state is CLOSED (i.e., TCB does not exist) then
391          * all data in the incoming segment is discarded.
392          * If the TCB exists but is in CLOSED state, it is embryonic,
393          * but should either do a listen or a connect soon.
394          *
395          * state == CLOSED means we've done socreate() but haven't
396          * attached it to a protocol yet...
397          *
398          * XXX If a TCB does not exist, and the TH_SYN flag is
399          * the only flag set, then create a session, mark it
400          * as if it was LISTENING, and continue...
401          */
402         if (so == 0) {
403           if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
404             goto dropwithreset;
405
406           if ((so = socreate()) == NULL)
407             goto dropwithreset;
408           if (tcp_attach(so) < 0) {
409             free(so); /* Not sofree (if it failed, it's not insqued) */
410             goto dropwithreset;
411           }
412
413           sbreserve(&so->so_snd, TCP_SNDSPACE);
414           sbreserve(&so->so_rcv, TCP_RCVSPACE);
415
416           /*            tcp_last_so = so; */  /* XXX ? */
417           /*            tp = sototcpcb(so);    */
418
419           so->so_laddr = ti->ti_src;
420           so->so_lport = ti->ti_sport;
421           so->so_faddr = ti->ti_dst;
422           so->so_fport = ti->ti_dport;
423
424           if ((so->so_iptos = tcp_tos(so)) == 0)
425             so->so_iptos = ((struct ip *)ti)->ip_tos;
426
427           tp = sototcpcb(so);
428           tp->t_state = TCPS_LISTEN;
429         }
430
431         /*
432          * If this is a still-connecting socket, this probably
433          * a retransmit of the SYN.  Whether it's a retransmit SYN
434          * or something else, we nuke it.
435          */
436         if (so->so_state & SS_ISFCONNECTING)
437                 goto drop;
438
439         tp = sototcpcb(so);
440
441         /* XXX Should never fail */
442         if (tp == 0)
443                 goto dropwithreset;
444         if (tp->t_state == TCPS_CLOSED)
445                 goto drop;
446
447         /* Unscale the window into a 32-bit value. */
448 /*      if ((tiflags & TH_SYN) == 0)
449  *              tiwin = ti->ti_win << tp->snd_scale;
450  *      else
451  */
452                 tiwin = ti->ti_win;
453
454         /*
455          * Segment received on connection.
456          * Reset idle time and keep-alive timer.
457          */
458         tp->t_idle = 0;
459         if (SO_OPTIONS)
460            tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
461         else
462            tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
463
464         /*
465          * Process options if not in LISTEN state,
466          * else do it below (after getting remote address).
467          */
468         if (optp && tp->t_state != TCPS_LISTEN)
469                 tcp_dooptions(tp, (u_char *)optp, optlen, ti);
470 /* , */
471 /*                      &ts_present, &ts_val, &ts_ecr); */
472
473         /*
474          * Header prediction: check for the two common cases
475          * of a uni-directional data xfer.  If the packet has
476          * no control flags, is in-sequence, the window didn't
477          * change and we're not retransmitting, it's a
478          * candidate.  If the length is zero and the ack moved
479          * forward, we're the sender side of the xfer.  Just
480          * free the data acked & wake any higher level process
481          * that was blocked waiting for space.  If the length
482          * is non-zero and the ack didn't move, we're the
483          * receiver side.  If we're getting packets in-order
484          * (the reassembly queue is empty), add the data to
485          * the socket buffer and note that we need a delayed ack.
486          *
487          * XXX Some of these tests are not needed
488          * eg: the tiwin == tp->snd_wnd prevents many more
489          * predictions.. with no *real* advantage..
490          */
491         if (tp->t_state == TCPS_ESTABLISHED &&
492             (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
493 /*          (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && */
494             ti->ti_seq == tp->rcv_nxt &&
495             tiwin && tiwin == tp->snd_wnd &&
496             tp->snd_nxt == tp->snd_max) {
497                 /*
498                  * If last ACK falls within this segment's sequence numbers,
499                  *  record the timestamp.
500                  */
501 /*              if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
502  *                 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
503  *                      tp->ts_recent_age = tcp_now;
504  *                      tp->ts_recent = ts_val;
505  *              }
506  */
507                 if (ti->ti_len == 0) {
508                         if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
509                             SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
510                             tp->snd_cwnd >= tp->snd_wnd) {
511                                 /*
512                                  * this is a pure ack for outstanding data.
513                                  */
514                                 STAT(tcpstat.tcps_predack++);
515 /*                              if (ts_present)
516  *                                      tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
517  *                              else
518  */                                  if (tp->t_rtt &&
519                                             SEQ_GT(ti->ti_ack, tp->t_rtseq))
520                                         tcp_xmit_timer(tp, tp->t_rtt);
521                                 acked = ti->ti_ack - tp->snd_una;
522                                 STAT(tcpstat.tcps_rcvackpack++);
523                                 STAT(tcpstat.tcps_rcvackbyte += acked);
524                                 sbdrop(&so->so_snd, acked);
525                                 tp->snd_una = ti->ti_ack;
526                                 m_freem(m);
527
528                                 /*
529                                  * If all outstanding data are acked, stop
530                                  * retransmit timer, otherwise restart timer
531                                  * using current (possibly backed-off) value.
532                                  * If process is waiting for space,
533                                  * wakeup/selwakeup/signal.  If data
534                                  * are ready to send, let tcp_output
535                                  * decide between more output or persist.
536                                  */
537                                 if (tp->snd_una == tp->snd_max)
538                                         tp->t_timer[TCPT_REXMT] = 0;
539                                 else if (tp->t_timer[TCPT_PERSIST] == 0)
540                                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
541
542                                 /*
543                                  * There's room in so_snd, sowwakup will read()
544                                  * from the socket if we can
545                                  */
546 /*                              if (so->so_snd.sb_flags & SB_NOTIFY)
547  *                                      sowwakeup(so);
548  */
549                                 /*
550                                  * This is called because sowwakeup might have
551                                  * put data into so_snd.  Since we don't so sowwakeup,
552                                  * we don't need this.. XXX???
553                                  */
554                                 if (so->so_snd.sb_cc)
555                                         (void) tcp_output(tp);
556
557                                 return;
558                         }
559                 } else if (ti->ti_ack == tp->snd_una &&
560                     tcpfrag_list_empty(tp) &&
561                     ti->ti_len <= sbspace(&so->so_rcv)) {
562                         /*
563                          * this is a pure, in-sequence data packet
564                          * with nothing on the reassembly queue and
565                          * we have enough buffer space to take it.
566                          */
567                         STAT(tcpstat.tcps_preddat++);
568                         tp->rcv_nxt += ti->ti_len;
569                         STAT(tcpstat.tcps_rcvpack++);
570                         STAT(tcpstat.tcps_rcvbyte += ti->ti_len);
571                         /*
572                          * Add data to socket buffer.
573                          */
574                         if (so->so_emu) {
575                                 if (tcp_emu(so,m)) sbappend(so, m);
576                         } else
577                                 sbappend(so, m);
578
579                         /*
580                          * XXX This is called when data arrives.  Later, check
581                          * if we can actually write() to the socket
582                          * XXX Need to check? It's be NON_BLOCKING
583                          */
584 /*                      sorwakeup(so); */
585
586                         /*
587                          * If this is a short packet, then ACK now - with Nagel
588                          *      congestion avoidance sender won't send more until
589                          *      he gets an ACK.
590                          *
591                          * It is better to not delay acks at all to maximize
592                          * TCP throughput.  See RFC 2581.
593                          */
594                         tp->t_flags |= TF_ACKNOW;
595                         tcp_output(tp);
596                         return;
597                 }
598         } /* header prediction */
599         /*
600          * Calculate amount of space in receive window,
601          * and then do TCP input processing.
602          * Receive window is amount of space in rcv queue,
603          * but not less than advertised window.
604          */
605         { int win;
606           win = sbspace(&so->so_rcv);
607           if (win < 0)
608             win = 0;
609           tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
610         }
611
612         switch (tp->t_state) {
613
614         /*
615          * If the state is LISTEN then ignore segment if it contains an RST.
616          * If the segment contains an ACK then it is bad and send a RST.
617          * If it does not contain a SYN then it is not interesting; drop it.
618          * Don't bother responding if the destination was a broadcast.
619          * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
620          * tp->iss, and send a segment:
621          *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
622          * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
623          * Fill in remote peer address fields if not previously specified.
624          * Enter SYN_RECEIVED state, and process any other fields of this
625          * segment in this state.
626          */
627         case TCPS_LISTEN: {
628
629           if (tiflags & TH_RST)
630             goto drop;
631           if (tiflags & TH_ACK)
632             goto dropwithreset;
633           if ((tiflags & TH_SYN) == 0)
634             goto drop;
635
636           /*
637            * This has way too many gotos...
638            * But a bit of spaghetti code never hurt anybody :)
639            */
640
641           /*
642            * If this is destined for the control address, then flag to
643            * tcp_ctl once connected, otherwise connect
644            */
645           if ((so->so_faddr.s_addr&htonl(0xffffff00)) == special_addr.s_addr) {
646             int lastbyte=ntohl(so->so_faddr.s_addr) & 0xff;
647             if (lastbyte!=CTL_ALIAS && lastbyte!=CTL_DNS) {
648 #if 0
649               if(lastbyte==CTL_CMD || lastbyte==CTL_EXEC) {
650                 /* Command or exec adress */
651                 so->so_state |= SS_CTL;
652               } else
653 #endif
654               {
655                 /* May be an add exec */
656                 for(ex_ptr = exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next) {
657                   if(ex_ptr->ex_fport == so->so_fport &&
658                      lastbyte == ex_ptr->ex_addr) {
659                     so->so_state |= SS_CTL;
660                     break;
661                   }
662                 }
663               }
664               if(so->so_state & SS_CTL) goto cont_input;
665             }
666             /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
667           }
668
669           if (so->so_emu & EMU_NOCONNECT) {
670             so->so_emu &= ~EMU_NOCONNECT;
671             goto cont_input;
672           }
673
674           if((tcp_fconnect(so) == -1) && (errno != EINPROGRESS) && (errno != EWOULDBLOCK)) {
675             u_char code=ICMP_UNREACH_NET;
676             DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
677                         errno,strerror(errno)));
678             if(errno == ECONNREFUSED) {
679               /* ACK the SYN, send RST to refuse the connection */
680               tcp_respond(tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
681                           TH_RST|TH_ACK);
682             } else {
683               if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
684               HTONL(ti->ti_seq);             /* restore tcp header */
685               HTONL(ti->ti_ack);
686               HTONS(ti->ti_win);
687               HTONS(ti->ti_urp);
688               m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
689               m->m_len  += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
690               *ip=save_ip;
691               icmp_error(m, ICMP_UNREACH,code, 0,strerror(errno));
692             }
693             tp = tcp_close(tp);
694             m_free(m);
695           } else {
696             /*
697              * Haven't connected yet, save the current mbuf
698              * and ti, and return
699              * XXX Some OS's don't tell us whether the connect()
700              * succeeded or not.  So we must time it out.
701              */
702             so->so_m = m;
703             so->so_ti = ti;
704             tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
705             tp->t_state = TCPS_SYN_RECEIVED;
706           }
707           return;
708
709         cont_conn:
710           /* m==NULL
711            * Check if the connect succeeded
712            */
713           if (so->so_state & SS_NOFDREF) {
714             tp = tcp_close(tp);
715             goto dropwithreset;
716           }
717         cont_input:
718           tcp_template(tp);
719
720           if (optp)
721             tcp_dooptions(tp, (u_char *)optp, optlen, ti);
722           /* , */
723           /*                            &ts_present, &ts_val, &ts_ecr); */
724
725           if (iss)
726             tp->iss = iss;
727           else
728             tp->iss = tcp_iss;
729           tcp_iss += TCP_ISSINCR/2;
730           tp->irs = ti->ti_seq;
731           tcp_sendseqinit(tp);
732           tcp_rcvseqinit(tp);
733           tp->t_flags |= TF_ACKNOW;
734           tp->t_state = TCPS_SYN_RECEIVED;
735           tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
736           STAT(tcpstat.tcps_accepts++);
737           goto trimthenstep6;
738         } /* case TCPS_LISTEN */
739
740         /*
741          * If the state is SYN_SENT:
742          *      if seg contains an ACK, but not for our SYN, drop the input.
743          *      if seg contains a RST, then drop the connection.
744          *      if seg does not contain SYN, then drop it.
745          * Otherwise this is an acceptable SYN segment
746          *      initialize tp->rcv_nxt and tp->irs
747          *      if seg contains ack then advance tp->snd_una
748          *      if SYN has been acked change to ESTABLISHED else SYN_RCVD state
749          *      arrange for segment to be acked (eventually)
750          *      continue processing rest of data/controls, beginning with URG
751          */
752         case TCPS_SYN_SENT:
753                 if ((tiflags & TH_ACK) &&
754                     (SEQ_LEQ(ti->ti_ack, tp->iss) ||
755                      SEQ_GT(ti->ti_ack, tp->snd_max)))
756                         goto dropwithreset;
757
758                 if (tiflags & TH_RST) {
759                         if (tiflags & TH_ACK)
760                                 tp = tcp_drop(tp,0); /* XXX Check t_softerror! */
761                         goto drop;
762                 }
763
764                 if ((tiflags & TH_SYN) == 0)
765                         goto drop;
766                 if (tiflags & TH_ACK) {
767                         tp->snd_una = ti->ti_ack;
768                         if (SEQ_LT(tp->snd_nxt, tp->snd_una))
769                                 tp->snd_nxt = tp->snd_una;
770                 }
771
772                 tp->t_timer[TCPT_REXMT] = 0;
773                 tp->irs = ti->ti_seq;
774                 tcp_rcvseqinit(tp);
775                 tp->t_flags |= TF_ACKNOW;
776                 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
777                         STAT(tcpstat.tcps_connects++);
778                         soisfconnected(so);
779                         tp->t_state = TCPS_ESTABLISHED;
780
781                         /* Do window scaling on this connection? */
782 /*                      if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
783  *                              (TF_RCVD_SCALE|TF_REQ_SCALE)) {
784  *                              tp->snd_scale = tp->requested_s_scale;
785  *                              tp->rcv_scale = tp->request_r_scale;
786  *                      }
787  */
788                         (void) tcp_reass(tp, (struct tcpiphdr *)0,
789                                 (struct mbuf *)0);
790                         /*
791                          * if we didn't have to retransmit the SYN,
792                          * use its rtt as our initial srtt & rtt var.
793                          */
794                         if (tp->t_rtt)
795                                 tcp_xmit_timer(tp, tp->t_rtt);
796                 } else
797                         tp->t_state = TCPS_SYN_RECEIVED;
798
799 trimthenstep6:
800                 /*
801                  * Advance ti->ti_seq to correspond to first data byte.
802                  * If data, trim to stay within window,
803                  * dropping FIN if necessary.
804                  */
805                 ti->ti_seq++;
806                 if (ti->ti_len > tp->rcv_wnd) {
807                         todrop = ti->ti_len - tp->rcv_wnd;
808                         m_adj(m, -todrop);
809                         ti->ti_len = tp->rcv_wnd;
810                         tiflags &= ~TH_FIN;
811                         STAT(tcpstat.tcps_rcvpackafterwin++);
812                         STAT(tcpstat.tcps_rcvbyteafterwin += todrop);
813                 }
814                 tp->snd_wl1 = ti->ti_seq - 1;
815                 tp->rcv_up = ti->ti_seq;
816                 goto step6;
817         } /* switch tp->t_state */
818         /*
819          * States other than LISTEN or SYN_SENT.
820          * First check timestamp, if present.
821          * Then check that at least some bytes of segment are within
822          * receive window.  If segment begins before rcv_nxt,
823          * drop leading data (and SYN); if nothing left, just ack.
824          *
825          * RFC 1323 PAWS: If we have a timestamp reply on this segment
826          * and it's less than ts_recent, drop it.
827          */
828 /*      if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
829  *          TSTMP_LT(ts_val, tp->ts_recent)) {
830  *
831  */             /* Check to see if ts_recent is over 24 days old.  */
832 /*              if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
833  */                     /*
834  *                       * Invalidate ts_recent.  If this segment updates
835  *                       * ts_recent, the age will be reset later and ts_recent
836  *                       * will get a valid value.  If it does not, setting
837  *                       * ts_recent to zero will at least satisfy the
838  *                       * requirement that zero be placed in the timestamp
839  *                       * echo reply when ts_recent isn't valid.  The
840  *                       * age isn't reset until we get a valid ts_recent
841  *                       * because we don't want out-of-order segments to be
842  *                       * dropped when ts_recent is old.
843  *                       */
844 /*                      tp->ts_recent = 0;
845  *              } else {
846  *                      tcpstat.tcps_rcvduppack++;
847  *                      tcpstat.tcps_rcvdupbyte += ti->ti_len;
848  *                      tcpstat.tcps_pawsdrop++;
849  *                      goto dropafterack;
850  *              }
851  *      }
852  */
853
854         todrop = tp->rcv_nxt - ti->ti_seq;
855         if (todrop > 0) {
856                 if (tiflags & TH_SYN) {
857                         tiflags &= ~TH_SYN;
858                         ti->ti_seq++;
859                         if (ti->ti_urp > 1)
860                                 ti->ti_urp--;
861                         else
862                                 tiflags &= ~TH_URG;
863                         todrop--;
864                 }
865                 /*
866                  * Following if statement from Stevens, vol. 2, p. 960.
867                  */
868                 if (todrop > ti->ti_len
869                     || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
870                         /*
871                          * Any valid FIN must be to the left of the window.
872                          * At this point the FIN must be a duplicate or out
873                          * of sequence; drop it.
874                          */
875                         tiflags &= ~TH_FIN;
876
877                         /*
878                          * Send an ACK to resynchronize and drop any data.
879                          * But keep on processing for RST or ACK.
880                          */
881                         tp->t_flags |= TF_ACKNOW;
882                         todrop = ti->ti_len;
883                         STAT(tcpstat.tcps_rcvduppack++);
884                         STAT(tcpstat.tcps_rcvdupbyte += todrop);
885                 } else {
886                         STAT(tcpstat.tcps_rcvpartduppack++);
887                         STAT(tcpstat.tcps_rcvpartdupbyte += todrop);
888                 }
889                 m_adj(m, todrop);
890                 ti->ti_seq += todrop;
891                 ti->ti_len -= todrop;
892                 if (ti->ti_urp > todrop)
893                         ti->ti_urp -= todrop;
894                 else {
895                         tiflags &= ~TH_URG;
896                         ti->ti_urp = 0;
897                 }
898         }
899         /*
900          * If new data are received on a connection after the
901          * user processes are gone, then RST the other end.
902          */
903         if ((so->so_state & SS_NOFDREF) &&
904             tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
905                 tp = tcp_close(tp);
906                 STAT(tcpstat.tcps_rcvafterclose++);
907                 goto dropwithreset;
908         }
909
910         /*
911          * If segment ends after window, drop trailing data
912          * (and PUSH and FIN); if nothing left, just ACK.
913          */
914         todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
915         if (todrop > 0) {
916                 STAT(tcpstat.tcps_rcvpackafterwin++);
917                 if (todrop >= ti->ti_len) {
918                         STAT(tcpstat.tcps_rcvbyteafterwin += ti->ti_len);
919                         /*
920                          * If a new connection request is received
921                          * while in TIME_WAIT, drop the old connection
922                          * and start over if the sequence numbers
923                          * are above the previous ones.
924                          */
925                         if (tiflags & TH_SYN &&
926                             tp->t_state == TCPS_TIME_WAIT &&
927                             SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
928                                 iss = tp->rcv_nxt + TCP_ISSINCR;
929                                 tp = tcp_close(tp);
930                                 goto findso;
931                         }
932                         /*
933                          * If window is closed can only take segments at
934                          * window edge, and have to drop data and PUSH from
935                          * incoming segments.  Continue processing, but
936                          * remember to ack.  Otherwise, drop segment
937                          * and ack.
938                          */
939                         if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
940                                 tp->t_flags |= TF_ACKNOW;
941                                 STAT(tcpstat.tcps_rcvwinprobe++);
942                         } else
943                                 goto dropafterack;
944                 } else
945                         STAT(tcpstat.tcps_rcvbyteafterwin += todrop);
946                 m_adj(m, -todrop);
947                 ti->ti_len -= todrop;
948                 tiflags &= ~(TH_PUSH|TH_FIN);
949         }
950
951         /*
952          * If last ACK falls within this segment's sequence numbers,
953          * record its timestamp.
954          */
955 /*      if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
956  *          SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
957  *                 ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
958  *              tp->ts_recent_age = tcp_now;
959  *              tp->ts_recent = ts_val;
960  *      }
961  */
962
963         /*
964          * If the RST bit is set examine the state:
965          *    SYN_RECEIVED STATE:
966          *      If passive open, return to LISTEN state.
967          *      If active open, inform user that connection was refused.
968          *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
969          *      Inform user that connection was reset, and close tcb.
970          *    CLOSING, LAST_ACK, TIME_WAIT STATES
971          *      Close the tcb.
972          */
973         if (tiflags&TH_RST) switch (tp->t_state) {
974
975         case TCPS_SYN_RECEIVED:
976 /*              so->so_error = ECONNREFUSED; */
977                 goto close;
978
979         case TCPS_ESTABLISHED:
980         case TCPS_FIN_WAIT_1:
981         case TCPS_FIN_WAIT_2:
982         case TCPS_CLOSE_WAIT:
983 /*              so->so_error = ECONNRESET; */
984         close:
985                 tp->t_state = TCPS_CLOSED;
986                 STAT(tcpstat.tcps_drops++);
987                 tp = tcp_close(tp);
988                 goto drop;
989
990         case TCPS_CLOSING:
991         case TCPS_LAST_ACK:
992         case TCPS_TIME_WAIT:
993                 tp = tcp_close(tp);
994                 goto drop;
995         }
996
997         /*
998          * If a SYN is in the window, then this is an
999          * error and we send an RST and drop the connection.
1000          */
1001         if (tiflags & TH_SYN) {
1002                 tp = tcp_drop(tp,0);
1003                 goto dropwithreset;
1004         }
1005
1006         /*
1007          * If the ACK bit is off we drop the segment and return.
1008          */
1009         if ((tiflags & TH_ACK) == 0) goto drop;
1010
1011         /*
1012          * Ack processing.
1013          */
1014         switch (tp->t_state) {
1015         /*
1016          * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1017          * ESTABLISHED state and continue processing, otherwise
1018          * send an RST.  una<=ack<=max
1019          */
1020         case TCPS_SYN_RECEIVED:
1021
1022                 if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
1023                     SEQ_GT(ti->ti_ack, tp->snd_max))
1024                         goto dropwithreset;
1025                 STAT(tcpstat.tcps_connects++);
1026                 tp->t_state = TCPS_ESTABLISHED;
1027                 /*
1028                  * The sent SYN is ack'ed with our sequence number +1
1029                  * The first data byte already in the buffer will get
1030                  * lost if no correction is made.  This is only needed for
1031                  * SS_CTL since the buffer is empty otherwise.
1032                  * tp->snd_una++; or:
1033                  */
1034                 tp->snd_una=ti->ti_ack;
1035                 if (so->so_state & SS_CTL) {
1036                   /* So tcp_ctl reports the right state */
1037                   ret = tcp_ctl(so);
1038                   if (ret == 1) {
1039                     soisfconnected(so);
1040                     so->so_state &= ~SS_CTL;   /* success XXX */
1041                   } else if (ret == 2) {
1042                     so->so_state = SS_NOFDREF; /* CTL_CMD */
1043                   } else {
1044                     needoutput = 1;
1045                     tp->t_state = TCPS_FIN_WAIT_1;
1046                   }
1047                 } else {
1048                   soisfconnected(so);
1049                 }
1050
1051                 /* Do window scaling? */
1052 /*              if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1053  *                      (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1054  *                      tp->snd_scale = tp->requested_s_scale;
1055  *                      tp->rcv_scale = tp->request_r_scale;
1056  *              }
1057  */
1058                 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
1059                 tp->snd_wl1 = ti->ti_seq - 1;
1060                 /* Avoid ack processing; snd_una==ti_ack  =>  dup ack */
1061                 goto synrx_to_est;
1062                 /* fall into ... */
1063
1064         /*
1065          * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1066          * ACKs.  If the ack is in the range
1067          *      tp->snd_una < ti->ti_ack <= tp->snd_max
1068          * then advance tp->snd_una to ti->ti_ack and drop
1069          * data from the retransmission queue.  If this ACK reflects
1070          * more up to date window information we update our window information.
1071          */
1072         case TCPS_ESTABLISHED:
1073         case TCPS_FIN_WAIT_1:
1074         case TCPS_FIN_WAIT_2:
1075         case TCPS_CLOSE_WAIT:
1076         case TCPS_CLOSING:
1077         case TCPS_LAST_ACK:
1078         case TCPS_TIME_WAIT:
1079
1080                 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1081                         if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1082                           STAT(tcpstat.tcps_rcvdupack++);
1083                           DEBUG_MISC((dfd," dup ack  m = %lx  so = %lx \n",
1084                                       (long )m, (long )so));
1085                                 /*
1086                                  * If we have outstanding data (other than
1087                                  * a window probe), this is a completely
1088                                  * duplicate ack (ie, window info didn't
1089                                  * change), the ack is the biggest we've
1090                                  * seen and we've seen exactly our rexmt
1091                                  * threshold of them, assume a packet
1092                                  * has been dropped and retransmit it.
1093                                  * Kludge snd_nxt & the congestion
1094                                  * window so we send only this one
1095                                  * packet.
1096                                  *
1097                                  * We know we're losing at the current
1098                                  * window size so do congestion avoidance
1099                                  * (set ssthresh to half the current window
1100                                  * and pull our congestion window back to
1101                                  * the new ssthresh).
1102                                  *
1103                                  * Dup acks mean that packets have left the
1104                                  * network (they're now cached at the receiver)
1105                                  * so bump cwnd by the amount in the receiver
1106                                  * to keep a constant cwnd packets in the
1107                                  * network.
1108                                  */
1109                                 if (tp->t_timer[TCPT_REXMT] == 0 ||
1110                                     ti->ti_ack != tp->snd_una)
1111                                         tp->t_dupacks = 0;
1112                                 else if (++tp->t_dupacks == TCPREXMTTHRESH) {
1113                                         tcp_seq onxt = tp->snd_nxt;
1114                                         u_int win =
1115                                             min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1116                                                 tp->t_maxseg;
1117
1118                                         if (win < 2)
1119                                                 win = 2;
1120                                         tp->snd_ssthresh = win * tp->t_maxseg;
1121                                         tp->t_timer[TCPT_REXMT] = 0;
1122                                         tp->t_rtt = 0;
1123                                         tp->snd_nxt = ti->ti_ack;
1124                                         tp->snd_cwnd = tp->t_maxseg;
1125                                         (void) tcp_output(tp);
1126                                         tp->snd_cwnd = tp->snd_ssthresh +
1127                                                tp->t_maxseg * tp->t_dupacks;
1128                                         if (SEQ_GT(onxt, tp->snd_nxt))
1129                                                 tp->snd_nxt = onxt;
1130                                         goto drop;
1131                                 } else if (tp->t_dupacks > TCPREXMTTHRESH) {
1132                                         tp->snd_cwnd += tp->t_maxseg;
1133                                         (void) tcp_output(tp);
1134                                         goto drop;
1135                                 }
1136                         } else
1137                                 tp->t_dupacks = 0;
1138                         break;
1139                 }
1140         synrx_to_est:
1141                 /*
1142                  * If the congestion window was inflated to account
1143                  * for the other side's cached packets, retract it.
1144                  */
1145                 if (tp->t_dupacks > TCPREXMTTHRESH &&
1146                     tp->snd_cwnd > tp->snd_ssthresh)
1147                         tp->snd_cwnd = tp->snd_ssthresh;
1148                 tp->t_dupacks = 0;
1149                 if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1150                         STAT(tcpstat.tcps_rcvacktoomuch++);
1151                         goto dropafterack;
1152                 }
1153                 acked = ti->ti_ack - tp->snd_una;
1154                 STAT(tcpstat.tcps_rcvackpack++);
1155                 STAT(tcpstat.tcps_rcvackbyte += acked);
1156
1157                 /*
1158                  * If we have a timestamp reply, update smoothed
1159                  * round trip time.  If no timestamp is present but
1160                  * transmit timer is running and timed sequence
1161                  * number was acked, update smoothed round trip time.
1162                  * Since we now have an rtt measurement, cancel the
1163                  * timer backoff (cf., Phil Karn's retransmit alg.).
1164                  * Recompute the initial retransmit timer.
1165                  */
1166 /*              if (ts_present)
1167  *                      tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1168  *              else
1169  */
1170                      if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1171                         tcp_xmit_timer(tp,tp->t_rtt);
1172
1173                 /*
1174                  * If all outstanding data is acked, stop retransmit
1175                  * timer and remember to restart (more output or persist).
1176                  * If there is more data to be acked, restart retransmit
1177                  * timer, using current (possibly backed-off) value.
1178                  */
1179                 if (ti->ti_ack == tp->snd_max) {
1180                         tp->t_timer[TCPT_REXMT] = 0;
1181                         needoutput = 1;
1182                 } else if (tp->t_timer[TCPT_PERSIST] == 0)
1183                         tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1184                 /*
1185                  * When new data is acked, open the congestion window.
1186                  * If the window gives us less than ssthresh packets
1187                  * in flight, open exponentially (maxseg per packet).
1188                  * Otherwise open linearly: maxseg per window
1189                  * (maxseg^2 / cwnd per packet).
1190                  */
1191                 {
1192                   register u_int cw = tp->snd_cwnd;
1193                   register u_int incr = tp->t_maxseg;
1194
1195                   if (cw > tp->snd_ssthresh)
1196                     incr = incr * incr / cw;
1197                   tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1198                 }
1199                 if (acked > so->so_snd.sb_cc) {
1200                         tp->snd_wnd -= so->so_snd.sb_cc;
1201                         sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1202                         ourfinisacked = 1;
1203                 } else {
1204                         sbdrop(&so->so_snd, acked);
1205                         tp->snd_wnd -= acked;
1206                         ourfinisacked = 0;
1207                 }
1208                 /*
1209                  * XXX sowwakup is called when data is acked and there's room for
1210                  * for more data... it should read() the socket
1211                  */
1212 /*              if (so->so_snd.sb_flags & SB_NOTIFY)
1213  *                      sowwakeup(so);
1214  */
1215                 tp->snd_una = ti->ti_ack;
1216                 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1217                         tp->snd_nxt = tp->snd_una;
1218
1219                 switch (tp->t_state) {
1220
1221                 /*
1222                  * In FIN_WAIT_1 STATE in addition to the processing
1223                  * for the ESTABLISHED state if our FIN is now acknowledged
1224                  * then enter FIN_WAIT_2.
1225                  */
1226                 case TCPS_FIN_WAIT_1:
1227                         if (ourfinisacked) {
1228                                 /*
1229                                  * If we can't receive any more
1230                                  * data, then closing user can proceed.
1231                                  * Starting the timer is contrary to the
1232                                  * specification, but if we don't get a FIN
1233                                  * we'll hang forever.
1234                                  */
1235                                 if (so->so_state & SS_FCANTRCVMORE) {
1236                                         soisfdisconnected(so);
1237                                         tp->t_timer[TCPT_2MSL] = TCP_MAXIDLE;
1238                                 }
1239                                 tp->t_state = TCPS_FIN_WAIT_2;
1240                         }
1241                         break;
1242
1243                 /*
1244                  * In CLOSING STATE in addition to the processing for
1245                  * the ESTABLISHED state if the ACK acknowledges our FIN
1246                  * then enter the TIME-WAIT state, otherwise ignore
1247                  * the segment.
1248                  */
1249                 case TCPS_CLOSING:
1250                         if (ourfinisacked) {
1251                                 tp->t_state = TCPS_TIME_WAIT;
1252                                 tcp_canceltimers(tp);
1253                                 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1254                                 soisfdisconnected(so);
1255                         }
1256                         break;
1257
1258                 /*
1259                  * In LAST_ACK, we may still be waiting for data to drain
1260                  * and/or to be acked, as well as for the ack of our FIN.
1261                  * If our FIN is now acknowledged, delete the TCB,
1262                  * enter the closed state and return.
1263                  */
1264                 case TCPS_LAST_ACK:
1265                         if (ourfinisacked) {
1266                                 tp = tcp_close(tp);
1267                                 goto drop;
1268                         }
1269                         break;
1270
1271                 /*
1272                  * In TIME_WAIT state the only thing that should arrive
1273                  * is a retransmission of the remote FIN.  Acknowledge
1274                  * it and restart the finack timer.
1275                  */
1276                 case TCPS_TIME_WAIT:
1277                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1278                         goto dropafterack;
1279                 }
1280         } /* switch(tp->t_state) */
1281
1282 step6:
1283         /*
1284          * Update window information.
1285          * Don't look at window if no ACK: TAC's send garbage on first SYN.
1286          */
1287         if ((tiflags & TH_ACK) &&
1288             (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1289             (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1290             (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1291                 /* keep track of pure window updates */
1292                 if (ti->ti_len == 0 &&
1293                     tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1294                         STAT(tcpstat.tcps_rcvwinupd++);
1295                 tp->snd_wnd = tiwin;
1296                 tp->snd_wl1 = ti->ti_seq;
1297                 tp->snd_wl2 = ti->ti_ack;
1298                 if (tp->snd_wnd > tp->max_sndwnd)
1299                         tp->max_sndwnd = tp->snd_wnd;
1300                 needoutput = 1;
1301         }
1302
1303         /*
1304          * Process segments with URG.
1305          */
1306         if ((tiflags & TH_URG) && ti->ti_urp &&
1307             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1308                 /*
1309                  * This is a kludge, but if we receive and accept
1310                  * random urgent pointers, we'll crash in
1311                  * soreceive.  It's hard to imagine someone
1312                  * actually wanting to send this much urgent data.
1313                  */
1314                 if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
1315                         ti->ti_urp = 0;
1316                         tiflags &= ~TH_URG;
1317                         goto dodata;
1318                 }
1319                 /*
1320                  * If this segment advances the known urgent pointer,
1321                  * then mark the data stream.  This should not happen
1322                  * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1323                  * a FIN has been received from the remote side.
1324                  * In these states we ignore the URG.
1325                  *
1326                  * According to RFC961 (Assigned Protocols),
1327                  * the urgent pointer points to the last octet
1328                  * of urgent data.  We continue, however,
1329                  * to consider it to indicate the first octet
1330                  * of data past the urgent section as the original
1331                  * spec states (in one of two places).
1332                  */
1333                 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1334                         tp->rcv_up = ti->ti_seq + ti->ti_urp;
1335                         so->so_urgc =  so->so_rcv.sb_cc +
1336                                 (tp->rcv_up - tp->rcv_nxt); /* -1; */
1337                         tp->rcv_up = ti->ti_seq + ti->ti_urp;
1338
1339                 }
1340         } else
1341                 /*
1342                  * If no out of band data is expected,
1343                  * pull receive urgent pointer along
1344                  * with the receive window.
1345                  */
1346                 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1347                         tp->rcv_up = tp->rcv_nxt;
1348 dodata:
1349
1350         /*
1351          * Process the segment text, merging it into the TCP sequencing queue,
1352          * and arranging for acknowledgment of receipt if necessary.
1353          * This process logically involves adjusting tp->rcv_wnd as data
1354          * is presented to the user (this happens in tcp_usrreq.c,
1355          * case PRU_RCVD).  If a FIN has already been received on this
1356          * connection then we just ignore the text.
1357          */
1358         if ((ti->ti_len || (tiflags&TH_FIN)) &&
1359             TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1360                 TCP_REASS(tp, ti, m, so, tiflags);
1361                 /*
1362                  * Note the amount of data that peer has sent into
1363                  * our window, in order to estimate the sender's
1364                  * buffer size.
1365                  */
1366                 len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1367         } else {
1368                 m_free(m);
1369                 tiflags &= ~TH_FIN;
1370         }
1371
1372         /*
1373          * If FIN is received ACK the FIN and let the user know
1374          * that the connection is closing.
1375          */
1376         if (tiflags & TH_FIN) {
1377                 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1378                         /*
1379                          * If we receive a FIN we can't send more data,
1380                          * set it SS_FDRAIN
1381                          * Shutdown the socket if there is no rx data in the
1382                          * buffer.
1383                          * soread() is called on completion of shutdown() and
1384                          * will got to TCPS_LAST_ACK, and use tcp_output()
1385                          * to send the FIN.
1386                          */
1387 /*                      sofcantrcvmore(so); */
1388                         sofwdrain(so);
1389
1390                         tp->t_flags |= TF_ACKNOW;
1391                         tp->rcv_nxt++;
1392                 }
1393                 switch (tp->t_state) {
1394
1395                 /*
1396                  * In SYN_RECEIVED and ESTABLISHED STATES
1397                  * enter the CLOSE_WAIT state.
1398                  */
1399                 case TCPS_SYN_RECEIVED:
1400                 case TCPS_ESTABLISHED:
1401                   if(so->so_emu == EMU_CTL)        /* no shutdown on socket */
1402                     tp->t_state = TCPS_LAST_ACK;
1403                   else
1404                     tp->t_state = TCPS_CLOSE_WAIT;
1405                   break;
1406
1407                 /*
1408                  * If still in FIN_WAIT_1 STATE FIN has not been acked so
1409                  * enter the CLOSING state.
1410                  */
1411                 case TCPS_FIN_WAIT_1:
1412                         tp->t_state = TCPS_CLOSING;
1413                         break;
1414
1415                 /*
1416                  * In FIN_WAIT_2 state enter the TIME_WAIT state,
1417                  * starting the time-wait timer, turning off the other
1418                  * standard timers.
1419                  */
1420                 case TCPS_FIN_WAIT_2:
1421                         tp->t_state = TCPS_TIME_WAIT;
1422                         tcp_canceltimers(tp);
1423                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1424                         soisfdisconnected(so);
1425                         break;
1426
1427                 /*
1428                  * In TIME_WAIT state restart the 2 MSL time_wait timer.
1429                  */
1430                 case TCPS_TIME_WAIT:
1431                         tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1432                         break;
1433                 }
1434         }
1435
1436         /*
1437          * If this is a small packet, then ACK now - with Nagel
1438          *      congestion avoidance sender won't send more until
1439          *      he gets an ACK.
1440          *
1441          * See above.
1442          */
1443 /*      if (ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg) {
1444  */
1445 /*      if ((ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg &&
1446  *              (so->so_iptos & IPTOS_LOWDELAY) == 0) ||
1447  *             ((so->so_iptos & IPTOS_LOWDELAY) &&
1448  *             ((struct tcpiphdr_2 *)ti)->first_char == (char)27)) {
1449  */
1450         if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
1451             ((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
1452                 tp->t_flags |= TF_ACKNOW;
1453         }
1454
1455         /*
1456          * Return any desired output.
1457          */
1458         if (needoutput || (tp->t_flags & TF_ACKNOW)) {
1459                 (void) tcp_output(tp);
1460         }
1461         return;
1462
1463 dropafterack:
1464         /*
1465          * Generate an ACK dropping incoming segment if it occupies
1466          * sequence space, where the ACK reflects our state.
1467          */
1468         if (tiflags & TH_RST)
1469                 goto drop;
1470         m_freem(m);
1471         tp->t_flags |= TF_ACKNOW;
1472         (void) tcp_output(tp);
1473         return;
1474
1475 dropwithreset:
1476         /* reuses m if m!=NULL, m_free() unnecessary */
1477         if (tiflags & TH_ACK)
1478                 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1479         else {
1480                 if (tiflags & TH_SYN) ti->ti_len++;
1481                 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1482                     TH_RST|TH_ACK);
1483         }
1484
1485         return;
1486
1487 drop:
1488         /*
1489          * Drop space held by incoming segment and return.
1490          */
1491         m_free(m);
1492
1493         return;
1494 }
1495
1496  /* , ts_present, ts_val, ts_ecr) */
1497 /*      int *ts_present;
1498  *      u_int32_t *ts_val, *ts_ecr;
1499  */
1500 static void
1501 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcpiphdr *ti)
1502 {
1503         u_int16_t mss;
1504         int opt, optlen;
1505
1506         DEBUG_CALL("tcp_dooptions");
1507         DEBUG_ARGS((dfd," tp = %lx  cnt=%i \n", (long )tp, cnt));
1508
1509         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1510                 opt = cp[0];
1511                 if (opt == TCPOPT_EOL)
1512                         break;
1513                 if (opt == TCPOPT_NOP)
1514                         optlen = 1;
1515                 else {
1516                         optlen = cp[1];
1517                         if (optlen <= 0)
1518                                 break;
1519                 }
1520                 switch (opt) {
1521
1522                 default:
1523                         continue;
1524
1525                 case TCPOPT_MAXSEG:
1526                         if (optlen != TCPOLEN_MAXSEG)
1527                                 continue;
1528                         if (!(ti->ti_flags & TH_SYN))
1529                                 continue;
1530                         memcpy((char *) &mss, (char *) cp + 2, sizeof(mss));
1531                         NTOHS(mss);
1532                         (void) tcp_mss(tp, mss);        /* sets t_maxseg */
1533                         break;
1534
1535 /*              case TCPOPT_WINDOW:
1536  *                      if (optlen != TCPOLEN_WINDOW)
1537  *                              continue;
1538  *                      if (!(ti->ti_flags & TH_SYN))
1539  *                              continue;
1540  *                      tp->t_flags |= TF_RCVD_SCALE;
1541  *                      tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1542  *                      break;
1543  */
1544 /*              case TCPOPT_TIMESTAMP:
1545  *                      if (optlen != TCPOLEN_TIMESTAMP)
1546  *                              continue;
1547  *                      *ts_present = 1;
1548  *                      memcpy((char *) ts_val, (char *)cp + 2, sizeof(*ts_val));
1549  *                      NTOHL(*ts_val);
1550  *                      memcpy((char *) ts_ecr, (char *)cp + 6, sizeof(*ts_ecr));
1551  *                      NTOHL(*ts_ecr);
1552  *
1553  */                     /*
1554  *                       * A timestamp received in a SYN makes
1555  *                       * it ok to send timestamp requests and replies.
1556  *                       */
1557 /*                      if (ti->ti_flags & TH_SYN) {
1558  *                              tp->t_flags |= TF_RCVD_TSTMP;
1559  *                              tp->ts_recent = *ts_val;
1560  *                              tp->ts_recent_age = tcp_now;
1561  *                      }
1562  */                     break;
1563                 }
1564         }
1565 }
1566
1567
1568 /*
1569  * Pull out of band byte out of a segment so
1570  * it doesn't appear in the user's data queue.
1571  * It is still reflected in the segment length for
1572  * sequencing purposes.
1573  */
1574
1575 #ifdef notdef
1576
1577 void
1578 tcp_pulloutofband(so, ti, m)
1579         struct socket *so;
1580         struct tcpiphdr *ti;
1581         register struct mbuf *m;
1582 {
1583         int cnt = ti->ti_urp - 1;
1584
1585         while (cnt >= 0) {
1586                 if (m->m_len > cnt) {
1587                         char *cp = mtod(m, caddr_t) + cnt;
1588                         struct tcpcb *tp = sototcpcb(so);
1589
1590                         tp->t_iobc = *cp;
1591                         tp->t_oobflags |= TCPOOB_HAVEDATA;
1592                         memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1593                         m->m_len--;
1594                         return;
1595                 }
1596                 cnt -= m->m_len;
1597                 m = m->m_next; /* XXX WRONG! Fix it! */
1598                 if (m == 0)
1599                         break;
1600         }
1601         panic("tcp_pulloutofband");
1602 }
1603
1604 #endif /* notdef */
1605
1606 /*
1607  * Collect new round-trip time estimate
1608  * and update averages and current timeout.
1609  */
1610
1611 static void
1612 tcp_xmit_timer(register struct tcpcb *tp, int rtt)
1613 {
1614         register short delta;
1615
1616         DEBUG_CALL("tcp_xmit_timer");
1617         DEBUG_ARG("tp = %lx", (long)tp);
1618         DEBUG_ARG("rtt = %d", rtt);
1619
1620         STAT(tcpstat.tcps_rttupdated++);
1621         if (tp->t_srtt != 0) {
1622                 /*
1623                  * srtt is stored as fixed point with 3 bits after the
1624                  * binary point (i.e., scaled by 8).  The following magic
1625                  * is equivalent to the smoothing algorithm in rfc793 with
1626                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1627                  * point).  Adjust rtt to origin 0.
1628                  */
1629                 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1630                 if ((tp->t_srtt += delta) <= 0)
1631                         tp->t_srtt = 1;
1632                 /*
1633                  * We accumulate a smoothed rtt variance (actually, a
1634                  * smoothed mean difference), then set the retransmit
1635                  * timer to smoothed rtt + 4 times the smoothed variance.
1636                  * rttvar is stored as fixed point with 2 bits after the
1637                  * binary point (scaled by 4).  The following is
1638                  * equivalent to rfc793 smoothing with an alpha of .75
1639                  * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1640                  * rfc793's wired-in beta.
1641                  */
1642                 if (delta < 0)
1643                         delta = -delta;
1644                 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1645                 if ((tp->t_rttvar += delta) <= 0)
1646                         tp->t_rttvar = 1;
1647         } else {
1648                 /*
1649                  * No rtt measurement yet - use the unsmoothed rtt.
1650                  * Set the variance to half the rtt (so our first
1651                  * retransmit happens at 3*rtt).
1652                  */
1653                 tp->t_srtt = rtt << TCP_RTT_SHIFT;
1654                 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1655         }
1656         tp->t_rtt = 0;
1657         tp->t_rxtshift = 0;
1658
1659         /*
1660          * the retransmit should happen at rtt + 4 * rttvar.
1661          * Because of the way we do the smoothing, srtt and rttvar
1662          * will each average +1/2 tick of bias.  When we compute
1663          * the retransmit timer, we want 1/2 tick of rounding and
1664          * 1 extra tick because of +-1/2 tick uncertainty in the
1665          * firing of the timer.  The bias will give us exactly the
1666          * 1.5 tick we need.  But, because the bias is
1667          * statistical, we have to test that we don't drop below
1668          * the minimum feasible timer (which is 2 ticks).
1669          */
1670         TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1671             (short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1672
1673         /*
1674          * We received an ack for a packet that wasn't retransmitted;
1675          * it is probably safe to discard any error indications we've
1676          * received recently.  This isn't quite right, but close enough
1677          * for now (a route might have failed after we sent a segment,
1678          * and the return path might not be symmetrical).
1679          */
1680         tp->t_softerror = 0;
1681 }
1682
1683 /*
1684  * Determine a reasonable value for maxseg size.
1685  * If the route is known, check route for mtu.
1686  * If none, use an mss that can be handled on the outgoing
1687  * interface without forcing IP to fragment; if bigger than
1688  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1689  * to utilize large mbufs.  If no route is found, route has no mtu,
1690  * or the destination isn't local, use a default, hopefully conservative
1691  * size (usually 512 or the default IP max size, but no more than the mtu
1692  * of the interface), as we can't discover anything about intervening
1693  * gateways or networks.  We also initialize the congestion/slow start
1694  * window to be a single segment if the destination isn't local.
1695  * While looking at the routing entry, we also initialize other path-dependent
1696  * parameters from pre-set or cached values in the routing entry.
1697  */
1698
1699 int
1700 tcp_mss(tp, offer)
1701         register struct tcpcb *tp;
1702         u_int offer;
1703 {
1704         struct socket *so = tp->t_socket;
1705         int mss;
1706
1707         DEBUG_CALL("tcp_mss");
1708         DEBUG_ARG("tp = %lx", (long)tp);
1709         DEBUG_ARG("offer = %d", offer);
1710
1711         mss = min(IF_MTU, IF_MRU) - sizeof(struct tcpiphdr);
1712         if (offer)
1713                 mss = min(mss, offer);
1714         mss = max(mss, 32);
1715         if (mss < tp->t_maxseg || offer != 0)
1716            tp->t_maxseg = mss;
1717
1718         tp->snd_cwnd = mss;
1719
1720         sbreserve(&so->so_snd, TCP_SNDSPACE + ((TCP_SNDSPACE % mss) ?
1721                                                (mss - (TCP_SNDSPACE % mss)) :
1722                                                0));
1723         sbreserve(&so->so_rcv, TCP_RCVSPACE + ((TCP_RCVSPACE % mss) ?
1724                                                (mss - (TCP_RCVSPACE % mss)) :
1725                                                0));
1726
1727         DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1728
1729         return mss;
1730 }