2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
34 * tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
38 * Changes and additions relating to SLiRP
39 * Copyright (c) 1995 Danny Gasparovski.
41 * Please read the file COPYRIGHT for the
42 * terms and conditions of the copyright.
50 #define TCPREXMTTHRESH 3
51 struct socket
*tcp_last_so
= &tcb
;
53 tcp_seq tcp_iss
; /* tcp initial send seq # */
55 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
57 /* for modulo comparisons of timestamps */
58 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0)
59 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0)
62 * Insert segment ti into reassembly queue of tcp with
63 * control block tp. Return TH_FIN if reassembly now includes
64 * a segment with FIN. The macro form does the common case inline
65 * (segment is the next to be received on an established connection,
66 * and the queue is empty), avoiding linkage into and removal
67 * from the queue and repetition of various conversions.
68 * Set DELACK for segments received in order, but ack immediately
69 * when segments are out of order (so fast retransmit can work).
72 #define TCP_REASS(tp, ti, m, so, flags) {\
73 if ((ti)->ti_seq == (tp)->rcv_nxt && \
74 (tp)->seg_next == (tcpiphdrp_32)(tp) && \
75 (tp)->t_state == TCPS_ESTABLISHED) {\
76 if (ti->ti_flags & TH_PUSH) \
77 tp->t_flags |= TF_ACKNOW; \
79 tp->t_flags |= TF_DELACK; \
80 (tp)->rcv_nxt += (ti)->ti_len; \
81 flags = (ti)->ti_flags & TH_FIN; \
82 STAT(tcpstat.tcps_rcvpack++); \
83 STAT(tcpstat.tcps_rcvbyte += (ti)->ti_len); \
85 if (tcp_emu((so),(m))) sbappend((so), (m)); \
87 sbappend((so), (m)); \
88 /* sorwakeup(so); */ \
90 (flags) = tcp_reass((tp), (ti), (m)); \
91 tp->t_flags |= TF_ACKNOW; \
95 #define TCP_REASS(tp, ti, m, so, flags) { \
96 if ((ti)->ti_seq == (tp)->rcv_nxt && \
97 (tp)->seg_next == (tcpiphdrp_32)(tp) && \
98 (tp)->t_state == TCPS_ESTABLISHED) { \
99 tp->t_flags |= TF_DELACK; \
100 (tp)->rcv_nxt += (ti)->ti_len; \
101 flags = (ti)->ti_flags & TH_FIN; \
102 STAT(tcpstat.tcps_rcvpack++); \
103 STAT(tcpstat.tcps_rcvbyte += (ti)->ti_len); \
105 if (tcp_emu((so),(m))) sbappend(so, (m)); \
107 sbappend((so), (m)); \
108 /* sorwakeup(so); */ \
110 (flags) = tcp_reass((tp), (ti), (m)); \
111 tp->t_flags |= TF_ACKNOW; \
115 static void tcp_dooptions(struct tcpcb
*tp
, u_char
*cp
, int cnt
,
116 struct tcpiphdr
*ti
);
117 static void tcp_xmit_timer(register struct tcpcb
*tp
, int rtt
);
120 tcp_reass(register struct tcpcb
*tp
, register struct tcpiphdr
*ti
,
123 register struct tcpiphdr
*q
;
124 struct socket
*so
= tp
->t_socket
;
128 * Call with ti==0 after become established to
129 * force pre-ESTABLISHED data up to user socket.
135 * Find a segment which begins after this one does.
137 for (q
= (struct tcpiphdr
*)tp
->seg_next
; q
!= (struct tcpiphdr
*)tp
;
138 q
= (struct tcpiphdr
*)q
->ti_next
)
139 if (SEQ_GT(q
->ti_seq
, ti
->ti_seq
))
143 * If there is a preceding segment, it may provide some of
144 * our data already. If so, drop the data from the incoming
145 * segment. If it provides all of our data, drop us.
147 if ((struct tcpiphdr
*)q
->ti_prev
!= (struct tcpiphdr
*)tp
) {
149 q
= (struct tcpiphdr
*)q
->ti_prev
;
150 /* conversion to int (in i) handles seq wraparound */
151 i
= q
->ti_seq
+ q
->ti_len
- ti
->ti_seq
;
153 if (i
>= ti
->ti_len
) {
154 STAT(tcpstat
.tcps_rcvduppack
++);
155 STAT(tcpstat
.tcps_rcvdupbyte
+= ti
->ti_len
);
158 * Try to present any queued data
159 * at the left window edge to the user.
160 * This is needed after the 3-WHS
163 goto present
; /* ??? */
169 q
= (struct tcpiphdr
*)(q
->ti_next
);
171 STAT(tcpstat
.tcps_rcvoopack
++);
172 STAT(tcpstat
.tcps_rcvoobyte
+= ti
->ti_len
);
173 REASS_MBUF(ti
) = (mbufp_32
) m
; /* XXX */
176 * While we overlap succeeding segments trim them or,
177 * if they are completely covered, dequeue them.
179 while (q
!= (struct tcpiphdr
*)tp
) {
180 register int i
= (ti
->ti_seq
+ ti
->ti_len
) - q
->ti_seq
;
186 m_adj((struct mbuf
*) REASS_MBUF(q
), i
);
189 q
= (struct tcpiphdr
*)q
->ti_next
;
190 m
= (struct mbuf
*) REASS_MBUF((struct tcpiphdr
*)q
->ti_prev
);
191 remque_32((void *)(q
->ti_prev
));
196 * Stick new segment in its place.
198 insque_32(ti
, (void *)(q
->ti_prev
));
202 * Present data to user, advancing rcv_nxt through
203 * completed sequence space.
205 if (!TCPS_HAVEESTABLISHED(tp
->t_state
))
207 ti
= (struct tcpiphdr
*) tp
->seg_next
;
208 if (ti
== (struct tcpiphdr
*)tp
|| ti
->ti_seq
!= tp
->rcv_nxt
)
210 if (tp
->t_state
== TCPS_SYN_RECEIVED
&& ti
->ti_len
)
213 tp
->rcv_nxt
+= ti
->ti_len
;
214 flags
= ti
->ti_flags
& TH_FIN
;
216 m
= (struct mbuf
*) REASS_MBUF(ti
); /* XXX */
217 ti
= (struct tcpiphdr
*)ti
->ti_next
;
218 /* if (so->so_state & SS_FCANTRCVMORE) */
219 if (so
->so_state
& SS_FCANTSENDMORE
)
223 if (tcp_emu(so
,m
)) sbappend(so
, m
);
227 } while (ti
!= (struct tcpiphdr
*)tp
&& ti
->ti_seq
== tp
->rcv_nxt
);
233 * TCP input routine, follows pages 65-76 of the
234 * protocol specification dated September, 1981 very closely.
237 tcp_input(m
, iphlen
, inso
)
238 register struct mbuf
*m
;
242 struct ip save_ip
, *ip
;
243 register struct tcpiphdr
*ti
;
247 register struct tcpcb
*tp
= 0;
248 register int tiflags
;
249 struct socket
*so
= 0;
250 int todrop
, acked
, ourfinisacked
, needoutput
= 0;
251 /* int dropsocket = 0; */
255 /* int ts_present = 0; */
256 struct ex_list
*ex_ptr
;
258 DEBUG_CALL("tcp_input");
259 DEBUG_ARGS((dfd
," m = %8lx iphlen = %2d inso = %lx\n",
260 (long )m
, iphlen
, (long )inso
));
263 * If called with m == 0, then we're continuing the connect
268 /* Re-set a few variables */
274 tiflags
= ti
->ti_flags
;
280 STAT(tcpstat
.tcps_rcvtotal
++);
282 * Get IP and TCP header together in first mbuf.
283 * Note: IP leaves IP header in first mbuf.
285 ti
= mtod(m
, struct tcpiphdr
*);
286 if (iphlen
> sizeof(struct ip
)) {
287 ip_stripoptions(m
, (struct mbuf
*)0);
288 iphlen
=sizeof(struct ip
);
290 /* XXX Check if too short */
294 * Save a copy of the IP header in case we want restore it
295 * for sending an ICMP error message in response.
297 ip
=mtod(m
, struct ip
*);
299 save_ip
.ip_len
+= iphlen
;
302 * Checksum extended TCP header and data.
304 tlen
= ((struct ip
*)ti
)->ip_len
;
305 ti
->ti_next
= ti
->ti_prev
= 0;
307 ti
->ti_len
= htons((u_int16_t
)tlen
);
308 len
= sizeof(struct ip
) + tlen
;
309 /* keep checksum for ICMP reply
310 * ti->ti_sum = cksum(m, len);
311 * if (ti->ti_sum) { */
313 STAT(tcpstat
.tcps_rcvbadsum
++);
318 * Check that TCP offset makes sense,
319 * pull out TCP options and adjust length. XXX
321 off
= ti
->ti_off
<< 2;
322 if (off
< sizeof (struct tcphdr
) || off
> tlen
) {
323 STAT(tcpstat
.tcps_rcvbadoff
++);
328 if (off
> sizeof (struct tcphdr
)) {
329 optlen
= off
- sizeof (struct tcphdr
);
330 optp
= mtod(m
, caddr_t
) + sizeof (struct tcpiphdr
);
333 * Do quick retrieval of timestamp options ("options
334 * prediction?"). If timestamp is the only option and it's
335 * formatted as recommended in RFC 1323 appendix A, we
336 * quickly get the values now and not bother calling
337 * tcp_dooptions(), etc.
339 /* if ((optlen == TCPOLEN_TSTAMP_APPA ||
340 * (optlen > TCPOLEN_TSTAMP_APPA &&
341 * optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
342 * *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
343 * (ti->ti_flags & TH_SYN) == 0) {
345 * ts_val = ntohl(*(u_int32_t *)(optp + 4));
346 * ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
347 * optp = NULL; / * we've parsed the options * /
351 tiflags
= ti
->ti_flags
;
354 * Convert TCP protocol specific fields to host format.
362 * Drop TCP, IP headers and TCP options.
364 m
->m_data
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
365 m
->m_len
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
367 if (slirp_restrict
) {
368 for (ex_ptr
= exec_list
; ex_ptr
; ex_ptr
= ex_ptr
->ex_next
)
369 if (ex_ptr
->ex_fport
== ti
->ti_dport
&&
370 (ntohl(ti
->ti_dst
.s_addr
) & 0xff) == ex_ptr
->ex_addr
)
377 * Locate pcb for segment.
381 if (so
->so_fport
!= ti
->ti_dport
||
382 so
->so_lport
!= ti
->ti_sport
||
383 so
->so_laddr
.s_addr
!= ti
->ti_src
.s_addr
||
384 so
->so_faddr
.s_addr
!= ti
->ti_dst
.s_addr
) {
385 so
= solookup(&tcb
, ti
->ti_src
, ti
->ti_sport
,
386 ti
->ti_dst
, ti
->ti_dport
);
389 STAT(tcpstat
.tcps_socachemiss
++);
393 * If the state is CLOSED (i.e., TCB does not exist) then
394 * all data in the incoming segment is discarded.
395 * If the TCB exists but is in CLOSED state, it is embryonic,
396 * but should either do a listen or a connect soon.
398 * state == CLOSED means we've done socreate() but haven't
399 * attached it to a protocol yet...
401 * XXX If a TCB does not exist, and the TH_SYN flag is
402 * the only flag set, then create a session, mark it
403 * as if it was LISTENING, and continue...
406 if ((tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) != TH_SYN
)
409 if ((so
= socreate()) == NULL
)
411 if (tcp_attach(so
) < 0) {
412 free(so
); /* Not sofree (if it failed, it's not insqued) */
416 sbreserve(&so
->so_snd
, TCP_SNDSPACE
);
417 sbreserve(&so
->so_rcv
, TCP_RCVSPACE
);
419 /* tcp_last_so = so; */ /* XXX ? */
420 /* tp = sototcpcb(so); */
422 so
->so_laddr
= ti
->ti_src
;
423 so
->so_lport
= ti
->ti_sport
;
424 so
->so_faddr
= ti
->ti_dst
;
425 so
->so_fport
= ti
->ti_dport
;
427 if ((so
->so_iptos
= tcp_tos(so
)) == 0)
428 so
->so_iptos
= ((struct ip
*)ti
)->ip_tos
;
431 tp
->t_state
= TCPS_LISTEN
;
435 * If this is a still-connecting socket, this probably
436 * a retransmit of the SYN. Whether it's a retransmit SYN
437 * or something else, we nuke it.
439 if (so
->so_state
& SS_ISFCONNECTING
)
444 /* XXX Should never fail */
447 if (tp
->t_state
== TCPS_CLOSED
)
450 /* Unscale the window into a 32-bit value. */
451 /* if ((tiflags & TH_SYN) == 0)
452 * tiwin = ti->ti_win << tp->snd_scale;
458 * Segment received on connection.
459 * Reset idle time and keep-alive timer.
463 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEPINTVL
;
465 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_IDLE
;
468 * Process options if not in LISTEN state,
469 * else do it below (after getting remote address).
471 if (optp
&& tp
->t_state
!= TCPS_LISTEN
)
472 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
474 /* &ts_present, &ts_val, &ts_ecr); */
477 * Header prediction: check for the two common cases
478 * of a uni-directional data xfer. If the packet has
479 * no control flags, is in-sequence, the window didn't
480 * change and we're not retransmitting, it's a
481 * candidate. If the length is zero and the ack moved
482 * forward, we're the sender side of the xfer. Just
483 * free the data acked & wake any higher level process
484 * that was blocked waiting for space. If the length
485 * is non-zero and the ack didn't move, we're the
486 * receiver side. If we're getting packets in-order
487 * (the reassembly queue is empty), add the data to
488 * the socket buffer and note that we need a delayed ack.
490 * XXX Some of these tests are not needed
491 * eg: the tiwin == tp->snd_wnd prevents many more
492 * predictions.. with no *real* advantage..
494 if (tp
->t_state
== TCPS_ESTABLISHED
&&
495 (tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) == TH_ACK
&&
496 /* (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && */
497 ti
->ti_seq
== tp
->rcv_nxt
&&
498 tiwin
&& tiwin
== tp
->snd_wnd
&&
499 tp
->snd_nxt
== tp
->snd_max
) {
501 * If last ACK falls within this segment's sequence numbers,
502 * record the timestamp.
504 /* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
505 * SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
506 * tp->ts_recent_age = tcp_now;
507 * tp->ts_recent = ts_val;
510 if (ti
->ti_len
== 0) {
511 if (SEQ_GT(ti
->ti_ack
, tp
->snd_una
) &&
512 SEQ_LEQ(ti
->ti_ack
, tp
->snd_max
) &&
513 tp
->snd_cwnd
>= tp
->snd_wnd
) {
515 * this is a pure ack for outstanding data.
517 STAT(tcpstat
.tcps_predack
++);
519 * tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
522 SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
523 tcp_xmit_timer(tp
, tp
->t_rtt
);
524 acked
= ti
->ti_ack
- tp
->snd_una
;
525 STAT(tcpstat
.tcps_rcvackpack
++);
526 STAT(tcpstat
.tcps_rcvackbyte
+= acked
);
527 sbdrop(&so
->so_snd
, acked
);
528 tp
->snd_una
= ti
->ti_ack
;
532 * If all outstanding data are acked, stop
533 * retransmit timer, otherwise restart timer
534 * using current (possibly backed-off) value.
535 * If process is waiting for space,
536 * wakeup/selwakeup/signal. If data
537 * are ready to send, let tcp_output
538 * decide between more output or persist.
540 if (tp
->snd_una
== tp
->snd_max
)
541 tp
->t_timer
[TCPT_REXMT
] = 0;
542 else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
543 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
546 * There's room in so_snd, sowwakup will read()
547 * from the socket if we can
549 /* if (so->so_snd.sb_flags & SB_NOTIFY)
553 * This is called because sowwakeup might have
554 * put data into so_snd. Since we don't so sowwakeup,
555 * we don't need this.. XXX???
557 if (so
->so_snd
.sb_cc
)
558 (void) tcp_output(tp
);
562 } else if (ti
->ti_ack
== tp
->snd_una
&&
563 tp
->seg_next
== (tcpiphdrp_32
)tp
&&
564 ti
->ti_len
<= sbspace(&so
->so_rcv
)) {
566 * this is a pure, in-sequence data packet
567 * with nothing on the reassembly queue and
568 * we have enough buffer space to take it.
570 STAT(tcpstat
.tcps_preddat
++);
571 tp
->rcv_nxt
+= ti
->ti_len
;
572 STAT(tcpstat
.tcps_rcvpack
++);
573 STAT(tcpstat
.tcps_rcvbyte
+= ti
->ti_len
);
575 * Add data to socket buffer.
578 if (tcp_emu(so
,m
)) sbappend(so
, m
);
583 * XXX This is called when data arrives. Later, check
584 * if we can actually write() to the socket
585 * XXX Need to check? It's be NON_BLOCKING
590 * If this is a short packet, then ACK now - with Nagel
591 * congestion avoidance sender won't send more until
594 * It is better to not delay acks at all to maximize
595 * TCP throughput. See RFC 2581.
597 tp
->t_flags
|= TF_ACKNOW
;
601 } /* header prediction */
603 * Calculate amount of space in receive window,
604 * and then do TCP input processing.
605 * Receive window is amount of space in rcv queue,
606 * but not less than advertised window.
609 win
= sbspace(&so
->so_rcv
);
612 tp
->rcv_wnd
= max(win
, (int)(tp
->rcv_adv
- tp
->rcv_nxt
));
615 switch (tp
->t_state
) {
618 * If the state is LISTEN then ignore segment if it contains an RST.
619 * If the segment contains an ACK then it is bad and send a RST.
620 * If it does not contain a SYN then it is not interesting; drop it.
621 * Don't bother responding if the destination was a broadcast.
622 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
623 * tp->iss, and send a segment:
624 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
625 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
626 * Fill in remote peer address fields if not previously specified.
627 * Enter SYN_RECEIVED state, and process any other fields of this
628 * segment in this state.
632 if (tiflags
& TH_RST
)
634 if (tiflags
& TH_ACK
)
636 if ((tiflags
& TH_SYN
) == 0)
640 * This has way too many gotos...
641 * But a bit of spaghetti code never hurt anybody :)
645 * If this is destined for the control address, then flag to
646 * tcp_ctl once connected, otherwise connect
648 if ((so
->so_faddr
.s_addr
&htonl(0xffffff00)) == special_addr
.s_addr
) {
649 int lastbyte
=ntohl(so
->so_faddr
.s_addr
) & 0xff;
650 if (lastbyte
!=CTL_ALIAS
&& lastbyte
!=CTL_DNS
) {
652 if(lastbyte
==CTL_CMD
|| lastbyte
==CTL_EXEC
) {
653 /* Command or exec adress */
654 so
->so_state
|= SS_CTL
;
658 /* May be an add exec */
659 for(ex_ptr
= exec_list
; ex_ptr
; ex_ptr
= ex_ptr
->ex_next
) {
660 if(ex_ptr
->ex_fport
== so
->so_fport
&&
661 lastbyte
== ex_ptr
->ex_addr
) {
662 so
->so_state
|= SS_CTL
;
667 if(so
->so_state
& SS_CTL
) goto cont_input
;
669 /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
672 if (so
->so_emu
& EMU_NOCONNECT
) {
673 so
->so_emu
&= ~EMU_NOCONNECT
;
677 if((tcp_fconnect(so
) == -1) && (errno
!= EINPROGRESS
) && (errno
!= EWOULDBLOCK
)) {
678 u_char code
=ICMP_UNREACH_NET
;
679 DEBUG_MISC((dfd
," tcp fconnect errno = %d-%s\n",
680 errno
,strerror(errno
)));
681 if(errno
== ECONNREFUSED
) {
682 /* ACK the SYN, send RST to refuse the connection */
683 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+1, (tcp_seq
)0,
686 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
687 HTONL(ti
->ti_seq
); /* restore tcp header */
691 m
->m_data
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
692 m
->m_len
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
694 icmp_error(m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
700 * Haven't connected yet, save the current mbuf
702 * XXX Some OS's don't tell us whether the connect()
703 * succeeded or not. So we must time it out.
707 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
708 tp
->t_state
= TCPS_SYN_RECEIVED
;
714 * Check if the connect succeeded
716 if (so
->so_state
& SS_NOFDREF
) {
724 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
726 /* &ts_present, &ts_val, &ts_ecr); */
732 tcp_iss
+= TCP_ISSINCR
/2;
733 tp
->irs
= ti
->ti_seq
;
736 tp
->t_flags
|= TF_ACKNOW
;
737 tp
->t_state
= TCPS_SYN_RECEIVED
;
738 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
739 STAT(tcpstat
.tcps_accepts
++);
741 } /* case TCPS_LISTEN */
744 * If the state is SYN_SENT:
745 * if seg contains an ACK, but not for our SYN, drop the input.
746 * if seg contains a RST, then drop the connection.
747 * if seg does not contain SYN, then drop it.
748 * Otherwise this is an acceptable SYN segment
749 * initialize tp->rcv_nxt and tp->irs
750 * if seg contains ack then advance tp->snd_una
751 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
752 * arrange for segment to be acked (eventually)
753 * continue processing rest of data/controls, beginning with URG
756 if ((tiflags
& TH_ACK
) &&
757 (SEQ_LEQ(ti
->ti_ack
, tp
->iss
) ||
758 SEQ_GT(ti
->ti_ack
, tp
->snd_max
)))
761 if (tiflags
& TH_RST
) {
762 if (tiflags
& TH_ACK
)
763 tp
= tcp_drop(tp
,0); /* XXX Check t_softerror! */
767 if ((tiflags
& TH_SYN
) == 0)
769 if (tiflags
& TH_ACK
) {
770 tp
->snd_una
= ti
->ti_ack
;
771 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
772 tp
->snd_nxt
= tp
->snd_una
;
775 tp
->t_timer
[TCPT_REXMT
] = 0;
776 tp
->irs
= ti
->ti_seq
;
778 tp
->t_flags
|= TF_ACKNOW
;
779 if (tiflags
& TH_ACK
&& SEQ_GT(tp
->snd_una
, tp
->iss
)) {
780 STAT(tcpstat
.tcps_connects
++);
782 tp
->t_state
= TCPS_ESTABLISHED
;
784 /* Do window scaling on this connection? */
785 /* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
786 * (TF_RCVD_SCALE|TF_REQ_SCALE)) {
787 * tp->snd_scale = tp->requested_s_scale;
788 * tp->rcv_scale = tp->request_r_scale;
791 (void) tcp_reass(tp
, (struct tcpiphdr
*)0,
794 * if we didn't have to retransmit the SYN,
795 * use its rtt as our initial srtt & rtt var.
798 tcp_xmit_timer(tp
, tp
->t_rtt
);
800 tp
->t_state
= TCPS_SYN_RECEIVED
;
804 * Advance ti->ti_seq to correspond to first data byte.
805 * If data, trim to stay within window,
806 * dropping FIN if necessary.
809 if (ti
->ti_len
> tp
->rcv_wnd
) {
810 todrop
= ti
->ti_len
- tp
->rcv_wnd
;
812 ti
->ti_len
= tp
->rcv_wnd
;
814 STAT(tcpstat
.tcps_rcvpackafterwin
++);
815 STAT(tcpstat
.tcps_rcvbyteafterwin
+= todrop
);
817 tp
->snd_wl1
= ti
->ti_seq
- 1;
818 tp
->rcv_up
= ti
->ti_seq
;
820 } /* switch tp->t_state */
822 * States other than LISTEN or SYN_SENT.
823 * First check timestamp, if present.
824 * Then check that at least some bytes of segment are within
825 * receive window. If segment begins before rcv_nxt,
826 * drop leading data (and SYN); if nothing left, just ack.
828 * RFC 1323 PAWS: If we have a timestamp reply on this segment
829 * and it's less than ts_recent, drop it.
831 /* if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
832 * TSTMP_LT(ts_val, tp->ts_recent)) {
834 */ /* Check to see if ts_recent is over 24 days old. */
835 /* if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
837 * * Invalidate ts_recent. If this segment updates
838 * * ts_recent, the age will be reset later and ts_recent
839 * * will get a valid value. If it does not, setting
840 * * ts_recent to zero will at least satisfy the
841 * * requirement that zero be placed in the timestamp
842 * * echo reply when ts_recent isn't valid. The
843 * * age isn't reset until we get a valid ts_recent
844 * * because we don't want out-of-order segments to be
845 * * dropped when ts_recent is old.
847 /* tp->ts_recent = 0;
849 * tcpstat.tcps_rcvduppack++;
850 * tcpstat.tcps_rcvdupbyte += ti->ti_len;
851 * tcpstat.tcps_pawsdrop++;
857 todrop
= tp
->rcv_nxt
- ti
->ti_seq
;
859 if (tiflags
& TH_SYN
) {
869 * Following if statement from Stevens, vol. 2, p. 960.
871 if (todrop
> ti
->ti_len
872 || (todrop
== ti
->ti_len
&& (tiflags
& TH_FIN
) == 0)) {
874 * Any valid FIN must be to the left of the window.
875 * At this point the FIN must be a duplicate or out
876 * of sequence; drop it.
881 * Send an ACK to resynchronize and drop any data.
882 * But keep on processing for RST or ACK.
884 tp
->t_flags
|= TF_ACKNOW
;
886 STAT(tcpstat
.tcps_rcvduppack
++);
887 STAT(tcpstat
.tcps_rcvdupbyte
+= todrop
);
889 STAT(tcpstat
.tcps_rcvpartduppack
++);
890 STAT(tcpstat
.tcps_rcvpartdupbyte
+= todrop
);
893 ti
->ti_seq
+= todrop
;
894 ti
->ti_len
-= todrop
;
895 if (ti
->ti_urp
> todrop
)
896 ti
->ti_urp
-= todrop
;
903 * If new data are received on a connection after the
904 * user processes are gone, then RST the other end.
906 if ((so
->so_state
& SS_NOFDREF
) &&
907 tp
->t_state
> TCPS_CLOSE_WAIT
&& ti
->ti_len
) {
909 STAT(tcpstat
.tcps_rcvafterclose
++);
914 * If segment ends after window, drop trailing data
915 * (and PUSH and FIN); if nothing left, just ACK.
917 todrop
= (ti
->ti_seq
+ti
->ti_len
) - (tp
->rcv_nxt
+tp
->rcv_wnd
);
919 STAT(tcpstat
.tcps_rcvpackafterwin
++);
920 if (todrop
>= ti
->ti_len
) {
921 STAT(tcpstat
.tcps_rcvbyteafterwin
+= ti
->ti_len
);
923 * If a new connection request is received
924 * while in TIME_WAIT, drop the old connection
925 * and start over if the sequence numbers
926 * are above the previous ones.
928 if (tiflags
& TH_SYN
&&
929 tp
->t_state
== TCPS_TIME_WAIT
&&
930 SEQ_GT(ti
->ti_seq
, tp
->rcv_nxt
)) {
931 iss
= tp
->rcv_nxt
+ TCP_ISSINCR
;
936 * If window is closed can only take segments at
937 * window edge, and have to drop data and PUSH from
938 * incoming segments. Continue processing, but
939 * remember to ack. Otherwise, drop segment
942 if (tp
->rcv_wnd
== 0 && ti
->ti_seq
== tp
->rcv_nxt
) {
943 tp
->t_flags
|= TF_ACKNOW
;
944 STAT(tcpstat
.tcps_rcvwinprobe
++);
948 STAT(tcpstat
.tcps_rcvbyteafterwin
+= todrop
);
950 ti
->ti_len
-= todrop
;
951 tiflags
&= ~(TH_PUSH
|TH_FIN
);
955 * If last ACK falls within this segment's sequence numbers,
956 * record its timestamp.
958 /* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
959 * SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
960 * ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
961 * tp->ts_recent_age = tcp_now;
962 * tp->ts_recent = ts_val;
967 * If the RST bit is set examine the state:
968 * SYN_RECEIVED STATE:
969 * If passive open, return to LISTEN state.
970 * If active open, inform user that connection was refused.
971 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
972 * Inform user that connection was reset, and close tcb.
973 * CLOSING, LAST_ACK, TIME_WAIT STATES
976 if (tiflags
&TH_RST
) switch (tp
->t_state
) {
978 case TCPS_SYN_RECEIVED
:
979 /* so->so_error = ECONNREFUSED; */
982 case TCPS_ESTABLISHED
:
983 case TCPS_FIN_WAIT_1
:
984 case TCPS_FIN_WAIT_2
:
985 case TCPS_CLOSE_WAIT
:
986 /* so->so_error = ECONNRESET; */
988 tp
->t_state
= TCPS_CLOSED
;
989 STAT(tcpstat
.tcps_drops
++);
1001 * If a SYN is in the window, then this is an
1002 * error and we send an RST and drop the connection.
1004 if (tiflags
& TH_SYN
) {
1005 tp
= tcp_drop(tp
,0);
1010 * If the ACK bit is off we drop the segment and return.
1012 if ((tiflags
& TH_ACK
) == 0) goto drop
;
1017 switch (tp
->t_state
) {
1019 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1020 * ESTABLISHED state and continue processing, otherwise
1021 * send an RST. una<=ack<=max
1023 case TCPS_SYN_RECEIVED
:
1025 if (SEQ_GT(tp
->snd_una
, ti
->ti_ack
) ||
1026 SEQ_GT(ti
->ti_ack
, tp
->snd_max
))
1028 STAT(tcpstat
.tcps_connects
++);
1029 tp
->t_state
= TCPS_ESTABLISHED
;
1031 * The sent SYN is ack'ed with our sequence number +1
1032 * The first data byte already in the buffer will get
1033 * lost if no correction is made. This is only needed for
1034 * SS_CTL since the buffer is empty otherwise.
1035 * tp->snd_una++; or:
1037 tp
->snd_una
=ti
->ti_ack
;
1038 if (so
->so_state
& SS_CTL
) {
1039 /* So tcp_ctl reports the right state */
1043 so
->so_state
&= ~SS_CTL
; /* success XXX */
1044 } else if (ret
== 2) {
1045 so
->so_state
= SS_NOFDREF
; /* CTL_CMD */
1048 tp
->t_state
= TCPS_FIN_WAIT_1
;
1054 /* Do window scaling? */
1055 /* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1056 * (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1057 * tp->snd_scale = tp->requested_s_scale;
1058 * tp->rcv_scale = tp->request_r_scale;
1061 (void) tcp_reass(tp
, (struct tcpiphdr
*)0, (struct mbuf
*)0);
1062 tp
->snd_wl1
= ti
->ti_seq
- 1;
1063 /* Avoid ack processing; snd_una==ti_ack => dup ack */
1068 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1069 * ACKs. If the ack is in the range
1070 * tp->snd_una < ti->ti_ack <= tp->snd_max
1071 * then advance tp->snd_una to ti->ti_ack and drop
1072 * data from the retransmission queue. If this ACK reflects
1073 * more up to date window information we update our window information.
1075 case TCPS_ESTABLISHED
:
1076 case TCPS_FIN_WAIT_1
:
1077 case TCPS_FIN_WAIT_2
:
1078 case TCPS_CLOSE_WAIT
:
1081 case TCPS_TIME_WAIT
:
1083 if (SEQ_LEQ(ti
->ti_ack
, tp
->snd_una
)) {
1084 if (ti
->ti_len
== 0 && tiwin
== tp
->snd_wnd
) {
1085 STAT(tcpstat
.tcps_rcvdupack
++);
1086 DEBUG_MISC((dfd
," dup ack m = %lx so = %lx \n",
1087 (long )m
, (long )so
));
1089 * If we have outstanding data (other than
1090 * a window probe), this is a completely
1091 * duplicate ack (ie, window info didn't
1092 * change), the ack is the biggest we've
1093 * seen and we've seen exactly our rexmt
1094 * threshold of them, assume a packet
1095 * has been dropped and retransmit it.
1096 * Kludge snd_nxt & the congestion
1097 * window so we send only this one
1100 * We know we're losing at the current
1101 * window size so do congestion avoidance
1102 * (set ssthresh to half the current window
1103 * and pull our congestion window back to
1104 * the new ssthresh).
1106 * Dup acks mean that packets have left the
1107 * network (they're now cached at the receiver)
1108 * so bump cwnd by the amount in the receiver
1109 * to keep a constant cwnd packets in the
1112 if (tp
->t_timer
[TCPT_REXMT
] == 0 ||
1113 ti
->ti_ack
!= tp
->snd_una
)
1115 else if (++tp
->t_dupacks
== TCPREXMTTHRESH
) {
1116 tcp_seq onxt
= tp
->snd_nxt
;
1118 min(tp
->snd_wnd
, tp
->snd_cwnd
) / 2 /
1123 tp
->snd_ssthresh
= win
* tp
->t_maxseg
;
1124 tp
->t_timer
[TCPT_REXMT
] = 0;
1126 tp
->snd_nxt
= ti
->ti_ack
;
1127 tp
->snd_cwnd
= tp
->t_maxseg
;
1128 (void) tcp_output(tp
);
1129 tp
->snd_cwnd
= tp
->snd_ssthresh
+
1130 tp
->t_maxseg
* tp
->t_dupacks
;
1131 if (SEQ_GT(onxt
, tp
->snd_nxt
))
1134 } else if (tp
->t_dupacks
> TCPREXMTTHRESH
) {
1135 tp
->snd_cwnd
+= tp
->t_maxseg
;
1136 (void) tcp_output(tp
);
1145 * If the congestion window was inflated to account
1146 * for the other side's cached packets, retract it.
1148 if (tp
->t_dupacks
> TCPREXMTTHRESH
&&
1149 tp
->snd_cwnd
> tp
->snd_ssthresh
)
1150 tp
->snd_cwnd
= tp
->snd_ssthresh
;
1152 if (SEQ_GT(ti
->ti_ack
, tp
->snd_max
)) {
1153 STAT(tcpstat
.tcps_rcvacktoomuch
++);
1156 acked
= ti
->ti_ack
- tp
->snd_una
;
1157 STAT(tcpstat
.tcps_rcvackpack
++);
1158 STAT(tcpstat
.tcps_rcvackbyte
+= acked
);
1161 * If we have a timestamp reply, update smoothed
1162 * round trip time. If no timestamp is present but
1163 * transmit timer is running and timed sequence
1164 * number was acked, update smoothed round trip time.
1165 * Since we now have an rtt measurement, cancel the
1166 * timer backoff (cf., Phil Karn's retransmit alg.).
1167 * Recompute the initial retransmit timer.
1170 * tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1173 if (tp
->t_rtt
&& SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
1174 tcp_xmit_timer(tp
,tp
->t_rtt
);
1177 * If all outstanding data is acked, stop retransmit
1178 * timer and remember to restart (more output or persist).
1179 * If there is more data to be acked, restart retransmit
1180 * timer, using current (possibly backed-off) value.
1182 if (ti
->ti_ack
== tp
->snd_max
) {
1183 tp
->t_timer
[TCPT_REXMT
] = 0;
1185 } else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
1186 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
1188 * When new data is acked, open the congestion window.
1189 * If the window gives us less than ssthresh packets
1190 * in flight, open exponentially (maxseg per packet).
1191 * Otherwise open linearly: maxseg per window
1192 * (maxseg^2 / cwnd per packet).
1195 register u_int cw
= tp
->snd_cwnd
;
1196 register u_int incr
= tp
->t_maxseg
;
1198 if (cw
> tp
->snd_ssthresh
)
1199 incr
= incr
* incr
/ cw
;
1200 tp
->snd_cwnd
= min(cw
+ incr
, TCP_MAXWIN
<<tp
->snd_scale
);
1202 if (acked
> so
->so_snd
.sb_cc
) {
1203 tp
->snd_wnd
-= so
->so_snd
.sb_cc
;
1204 sbdrop(&so
->so_snd
, (int )so
->so_snd
.sb_cc
);
1207 sbdrop(&so
->so_snd
, acked
);
1208 tp
->snd_wnd
-= acked
;
1212 * XXX sowwakup is called when data is acked and there's room for
1213 * for more data... it should read() the socket
1215 /* if (so->so_snd.sb_flags & SB_NOTIFY)
1218 tp
->snd_una
= ti
->ti_ack
;
1219 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
1220 tp
->snd_nxt
= tp
->snd_una
;
1222 switch (tp
->t_state
) {
1225 * In FIN_WAIT_1 STATE in addition to the processing
1226 * for the ESTABLISHED state if our FIN is now acknowledged
1227 * then enter FIN_WAIT_2.
1229 case TCPS_FIN_WAIT_1
:
1230 if (ourfinisacked
) {
1232 * If we can't receive any more
1233 * data, then closing user can proceed.
1234 * Starting the timer is contrary to the
1235 * specification, but if we don't get a FIN
1236 * we'll hang forever.
1238 if (so
->so_state
& SS_FCANTRCVMORE
) {
1239 soisfdisconnected(so
);
1240 tp
->t_timer
[TCPT_2MSL
] = TCP_MAXIDLE
;
1242 tp
->t_state
= TCPS_FIN_WAIT_2
;
1247 * In CLOSING STATE in addition to the processing for
1248 * the ESTABLISHED state if the ACK acknowledges our FIN
1249 * then enter the TIME-WAIT state, otherwise ignore
1253 if (ourfinisacked
) {
1254 tp
->t_state
= TCPS_TIME_WAIT
;
1255 tcp_canceltimers(tp
);
1256 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1257 soisfdisconnected(so
);
1262 * In LAST_ACK, we may still be waiting for data to drain
1263 * and/or to be acked, as well as for the ack of our FIN.
1264 * If our FIN is now acknowledged, delete the TCB,
1265 * enter the closed state and return.
1268 if (ourfinisacked
) {
1275 * In TIME_WAIT state the only thing that should arrive
1276 * is a retransmission of the remote FIN. Acknowledge
1277 * it and restart the finack timer.
1279 case TCPS_TIME_WAIT
:
1280 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1283 } /* switch(tp->t_state) */
1287 * Update window information.
1288 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1290 if ((tiflags
& TH_ACK
) &&
1291 (SEQ_LT(tp
->snd_wl1
, ti
->ti_seq
) ||
1292 (tp
->snd_wl1
== ti
->ti_seq
&& (SEQ_LT(tp
->snd_wl2
, ti
->ti_ack
) ||
1293 (tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
))))) {
1294 /* keep track of pure window updates */
1295 if (ti
->ti_len
== 0 &&
1296 tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
)
1297 STAT(tcpstat
.tcps_rcvwinupd
++);
1298 tp
->snd_wnd
= tiwin
;
1299 tp
->snd_wl1
= ti
->ti_seq
;
1300 tp
->snd_wl2
= ti
->ti_ack
;
1301 if (tp
->snd_wnd
> tp
->max_sndwnd
)
1302 tp
->max_sndwnd
= tp
->snd_wnd
;
1307 * Process segments with URG.
1309 if ((tiflags
& TH_URG
) && ti
->ti_urp
&&
1310 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1312 * This is a kludge, but if we receive and accept
1313 * random urgent pointers, we'll crash in
1314 * soreceive. It's hard to imagine someone
1315 * actually wanting to send this much urgent data.
1317 if (ti
->ti_urp
+ so
->so_rcv
.sb_cc
> so
->so_rcv
.sb_datalen
) {
1323 * If this segment advances the known urgent pointer,
1324 * then mark the data stream. This should not happen
1325 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1326 * a FIN has been received from the remote side.
1327 * In these states we ignore the URG.
1329 * According to RFC961 (Assigned Protocols),
1330 * the urgent pointer points to the last octet
1331 * of urgent data. We continue, however,
1332 * to consider it to indicate the first octet
1333 * of data past the urgent section as the original
1334 * spec states (in one of two places).
1336 if (SEQ_GT(ti
->ti_seq
+ti
->ti_urp
, tp
->rcv_up
)) {
1337 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1338 so
->so_urgc
= so
->so_rcv
.sb_cc
+
1339 (tp
->rcv_up
- tp
->rcv_nxt
); /* -1; */
1340 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1345 * If no out of band data is expected,
1346 * pull receive urgent pointer along
1347 * with the receive window.
1349 if (SEQ_GT(tp
->rcv_nxt
, tp
->rcv_up
))
1350 tp
->rcv_up
= tp
->rcv_nxt
;
1354 * Process the segment text, merging it into the TCP sequencing queue,
1355 * and arranging for acknowledgment of receipt if necessary.
1356 * This process logically involves adjusting tp->rcv_wnd as data
1357 * is presented to the user (this happens in tcp_usrreq.c,
1358 * case PRU_RCVD). If a FIN has already been received on this
1359 * connection then we just ignore the text.
1361 if ((ti
->ti_len
|| (tiflags
&TH_FIN
)) &&
1362 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1363 TCP_REASS(tp
, ti
, m
, so
, tiflags
);
1365 * Note the amount of data that peer has sent into
1366 * our window, in order to estimate the sender's
1369 len
= so
->so_rcv
.sb_datalen
- (tp
->rcv_adv
- tp
->rcv_nxt
);
1376 * If FIN is received ACK the FIN and let the user know
1377 * that the connection is closing.
1379 if (tiflags
& TH_FIN
) {
1380 if (TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1382 * If we receive a FIN we can't send more data,
1384 * Shutdown the socket if there is no rx data in the
1386 * soread() is called on completion of shutdown() and
1387 * will got to TCPS_LAST_ACK, and use tcp_output()
1390 /* sofcantrcvmore(so); */
1393 tp
->t_flags
|= TF_ACKNOW
;
1396 switch (tp
->t_state
) {
1399 * In SYN_RECEIVED and ESTABLISHED STATES
1400 * enter the CLOSE_WAIT state.
1402 case TCPS_SYN_RECEIVED
:
1403 case TCPS_ESTABLISHED
:
1404 if(so
->so_emu
== EMU_CTL
) /* no shutdown on socket */
1405 tp
->t_state
= TCPS_LAST_ACK
;
1407 tp
->t_state
= TCPS_CLOSE_WAIT
;
1411 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1412 * enter the CLOSING state.
1414 case TCPS_FIN_WAIT_1
:
1415 tp
->t_state
= TCPS_CLOSING
;
1419 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1420 * starting the time-wait timer, turning off the other
1423 case TCPS_FIN_WAIT_2
:
1424 tp
->t_state
= TCPS_TIME_WAIT
;
1425 tcp_canceltimers(tp
);
1426 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1427 soisfdisconnected(so
);
1431 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1433 case TCPS_TIME_WAIT
:
1434 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1440 * If this is a small packet, then ACK now - with Nagel
1441 * congestion avoidance sender won't send more until
1446 /* if (ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg) {
1448 /* if ((ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg &&
1449 * (so->so_iptos & IPTOS_LOWDELAY) == 0) ||
1450 * ((so->so_iptos & IPTOS_LOWDELAY) &&
1451 * ((struct tcpiphdr_2 *)ti)->first_char == (char)27)) {
1453 if (ti
->ti_len
&& (unsigned)ti
->ti_len
<= 5 &&
1454 ((struct tcpiphdr_2
*)ti
)->first_char
== (char)27) {
1455 tp
->t_flags
|= TF_ACKNOW
;
1459 * Return any desired output.
1461 if (needoutput
|| (tp
->t_flags
& TF_ACKNOW
)) {
1462 (void) tcp_output(tp
);
1468 * Generate an ACK dropping incoming segment if it occupies
1469 * sequence space, where the ACK reflects our state.
1471 if (tiflags
& TH_RST
)
1474 tp
->t_flags
|= TF_ACKNOW
;
1475 (void) tcp_output(tp
);
1479 /* reuses m if m!=NULL, m_free() unnecessary */
1480 if (tiflags
& TH_ACK
)
1481 tcp_respond(tp
, ti
, m
, (tcp_seq
)0, ti
->ti_ack
, TH_RST
);
1483 if (tiflags
& TH_SYN
) ti
->ti_len
++;
1484 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+ti
->ti_len
, (tcp_seq
)0,
1492 * Drop space held by incoming segment and return.
1499 /* , ts_present, ts_val, ts_ecr) */
1501 * u_int32_t *ts_val, *ts_ecr;
1504 tcp_dooptions(struct tcpcb
*tp
, u_char
*cp
, int cnt
, struct tcpiphdr
*ti
)
1509 DEBUG_CALL("tcp_dooptions");
1510 DEBUG_ARGS((dfd
," tp = %lx cnt=%i \n", (long )tp
, cnt
));
1512 for (; cnt
> 0; cnt
-= optlen
, cp
+= optlen
) {
1514 if (opt
== TCPOPT_EOL
)
1516 if (opt
== TCPOPT_NOP
)
1529 if (optlen
!= TCPOLEN_MAXSEG
)
1531 if (!(ti
->ti_flags
& TH_SYN
))
1533 memcpy((char *) &mss
, (char *) cp
+ 2, sizeof(mss
));
1535 (void) tcp_mss(tp
, mss
); /* sets t_maxseg */
1538 /* case TCPOPT_WINDOW:
1539 * if (optlen != TCPOLEN_WINDOW)
1541 * if (!(ti->ti_flags & TH_SYN))
1543 * tp->t_flags |= TF_RCVD_SCALE;
1544 * tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1547 /* case TCPOPT_TIMESTAMP:
1548 * if (optlen != TCPOLEN_TIMESTAMP)
1551 * memcpy((char *) ts_val, (char *)cp + 2, sizeof(*ts_val));
1553 * memcpy((char *) ts_ecr, (char *)cp + 6, sizeof(*ts_ecr));
1557 * * A timestamp received in a SYN makes
1558 * * it ok to send timestamp requests and replies.
1560 /* if (ti->ti_flags & TH_SYN) {
1561 * tp->t_flags |= TF_RCVD_TSTMP;
1562 * tp->ts_recent = *ts_val;
1563 * tp->ts_recent_age = tcp_now;
1572 * Pull out of band byte out of a segment so
1573 * it doesn't appear in the user's data queue.
1574 * It is still reflected in the segment length for
1575 * sequencing purposes.
1581 tcp_pulloutofband(so
, ti
, m
)
1583 struct tcpiphdr
*ti
;
1584 register struct mbuf
*m
;
1586 int cnt
= ti
->ti_urp
- 1;
1589 if (m
->m_len
> cnt
) {
1590 char *cp
= mtod(m
, caddr_t
) + cnt
;
1591 struct tcpcb
*tp
= sototcpcb(so
);
1594 tp
->t_oobflags
|= TCPOOB_HAVEDATA
;
1595 memcpy(sp
, cp
+1, (unsigned)(m
->m_len
- cnt
- 1));
1600 m
= m
->m_next
; /* XXX WRONG! Fix it! */
1604 panic("tcp_pulloutofband");
1610 * Collect new round-trip time estimate
1611 * and update averages and current timeout.
1615 tcp_xmit_timer(register struct tcpcb
*tp
, int rtt
)
1617 register short delta
;
1619 DEBUG_CALL("tcp_xmit_timer");
1620 DEBUG_ARG("tp = %lx", (long)tp
);
1621 DEBUG_ARG("rtt = %d", rtt
);
1623 STAT(tcpstat
.tcps_rttupdated
++);
1624 if (tp
->t_srtt
!= 0) {
1626 * srtt is stored as fixed point with 3 bits after the
1627 * binary point (i.e., scaled by 8). The following magic
1628 * is equivalent to the smoothing algorithm in rfc793 with
1629 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1630 * point). Adjust rtt to origin 0.
1632 delta
= rtt
- 1 - (tp
->t_srtt
>> TCP_RTT_SHIFT
);
1633 if ((tp
->t_srtt
+= delta
) <= 0)
1636 * We accumulate a smoothed rtt variance (actually, a
1637 * smoothed mean difference), then set the retransmit
1638 * timer to smoothed rtt + 4 times the smoothed variance.
1639 * rttvar is stored as fixed point with 2 bits after the
1640 * binary point (scaled by 4). The following is
1641 * equivalent to rfc793 smoothing with an alpha of .75
1642 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
1643 * rfc793's wired-in beta.
1647 delta
-= (tp
->t_rttvar
>> TCP_RTTVAR_SHIFT
);
1648 if ((tp
->t_rttvar
+= delta
) <= 0)
1652 * No rtt measurement yet - use the unsmoothed rtt.
1653 * Set the variance to half the rtt (so our first
1654 * retransmit happens at 3*rtt).
1656 tp
->t_srtt
= rtt
<< TCP_RTT_SHIFT
;
1657 tp
->t_rttvar
= rtt
<< (TCP_RTTVAR_SHIFT
- 1);
1663 * the retransmit should happen at rtt + 4 * rttvar.
1664 * Because of the way we do the smoothing, srtt and rttvar
1665 * will each average +1/2 tick of bias. When we compute
1666 * the retransmit timer, we want 1/2 tick of rounding and
1667 * 1 extra tick because of +-1/2 tick uncertainty in the
1668 * firing of the timer. The bias will give us exactly the
1669 * 1.5 tick we need. But, because the bias is
1670 * statistical, we have to test that we don't drop below
1671 * the minimum feasible timer (which is 2 ticks).
1673 TCPT_RANGESET(tp
->t_rxtcur
, TCP_REXMTVAL(tp
),
1674 (short)tp
->t_rttmin
, TCPTV_REXMTMAX
); /* XXX */
1677 * We received an ack for a packet that wasn't retransmitted;
1678 * it is probably safe to discard any error indications we've
1679 * received recently. This isn't quite right, but close enough
1680 * for now (a route might have failed after we sent a segment,
1681 * and the return path might not be symmetrical).
1683 tp
->t_softerror
= 0;
1687 * Determine a reasonable value for maxseg size.
1688 * If the route is known, check route for mtu.
1689 * If none, use an mss that can be handled on the outgoing
1690 * interface without forcing IP to fragment; if bigger than
1691 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1692 * to utilize large mbufs. If no route is found, route has no mtu,
1693 * or the destination isn't local, use a default, hopefully conservative
1694 * size (usually 512 or the default IP max size, but no more than the mtu
1695 * of the interface), as we can't discover anything about intervening
1696 * gateways or networks. We also initialize the congestion/slow start
1697 * window to be a single segment if the destination isn't local.
1698 * While looking at the routing entry, we also initialize other path-dependent
1699 * parameters from pre-set or cached values in the routing entry.
1704 register struct tcpcb
*tp
;
1707 struct socket
*so
= tp
->t_socket
;
1710 DEBUG_CALL("tcp_mss");
1711 DEBUG_ARG("tp = %lx", (long)tp
);
1712 DEBUG_ARG("offer = %d", offer
);
1714 mss
= min(IF_MTU
, IF_MRU
) - sizeof(struct tcpiphdr
);
1716 mss
= min(mss
, offer
);
1718 if (mss
< tp
->t_maxseg
|| offer
!= 0)
1723 sbreserve(&so
->so_snd
, TCP_SNDSPACE
+ ((TCP_SNDSPACE
% mss
) ?
1724 (mss
- (TCP_SNDSPACE
% mss
)) :
1726 sbreserve(&so
->so_rcv
, TCP_RCVSPACE
+ ((TCP_RCVSPACE
% mss
) ?
1727 (mss
- (TCP_RCVSPACE
% mss
)) :
1730 DEBUG_MISC((dfd
, " returning mss = %d\n", mss
));