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.
54 #include <vdecommon.h>
58 #define min(x,y) ((x) < (y) ? (x) : (y))
59 #define max(x,y) ((x) > (y) ? (x) : (y))
61 int tcprexmtthresh
= 3;
62 struct socket
*tcp_last_so
= &tcb
;
64 tcp_seq tcp_iss
; /* tcp initial send seq # */
66 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
68 /* for modulo comparisons of timestamps */
69 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0)
70 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0)
73 * Insert segment ti into reassembly queue of tcp with
74 * control block tp. Return TH_FIN if reassembly now includes
75 * a segment with FIN. The macro form does the common case inline
76 * (segment is the next to be received on an established connection,
77 * and the queue is empty), avoiding linkage into and removal
78 * from the queue and repetition of various conversions.
79 * Set DELACK for segments received in order, but ack immediately
80 * when segments are out of order (so fast retransmit can work).
83 #define TCP_REASS(tp, ti, m, so, flags) {\
84 if ((ti)->ti_seq == (tp)->rcv_nxt && \
85 (tp)->seg_next == (tcpiphdrp_32)(tp) && \
86 (tp)->t_state == TCPS_ESTABLISHED) {\
87 if (ti->ti_flags & TH_PUSH) \
88 tp->t_flags |= TF_ACKNOW; \
90 tp->t_flags |= TF_DELACK; \
91 (tp)->rcv_nxt += (ti)->ti_len; \
92 flags = (ti)->ti_flags & TH_FIN; \
93 tcpstat.tcps_rcvpack++;\
94 tcpstat.tcps_rcvbyte += (ti)->ti_len;\
96 if (tcp_emu((so),(m))) sbappend((so), (m)); \
98 sbappend((so), (m)); \
99 /* sorwakeup(so); */ \
101 (flags) = tcp_reass((tp), (ti), (m)); \
102 tp->t_flags |= TF_ACKNOW; \
106 #define TCP_REASS(tp, ti, m, so, flags) { \
107 if ((ti)->ti_seq == (tp)->rcv_nxt && \
108 (tp)->seg_next == (tcpiphdrp_32)(tp) && \
109 (tp)->t_state == TCPS_ESTABLISHED) { \
110 tp->t_flags |= TF_DELACK; \
111 (tp)->rcv_nxt += (ti)->ti_len; \
112 flags = (ti)->ti_flags & TH_FIN; \
113 tcpstat.tcps_rcvpack++;\
114 tcpstat.tcps_rcvbyte += (ti)->ti_len;\
116 if (tcp_emu((so),(m))) sbappend(so, (m)); \
118 sbappend((so), (m)); \
119 /* sorwakeup(so); */ \
121 (flags) = tcp_reass((tp), (ti), (m)); \
122 tp->t_flags |= TF_ACKNOW; \
129 register struct tcpcb
*tp
;
130 register struct tcpiphdr
*ti
;
133 register struct tcpiphdr
*q
;
134 struct socket
*so
= tp
->t_socket
;
138 * Call with ti==0 after become established to
139 * force pre-ESTABLISHED data up to user socket.
145 * Find a segment which begins after this one does.
147 for (q
= (struct tcpiphdr
*)tp
->seg_next
; q
!= (struct tcpiphdr
*)tp
;
148 q
= (struct tcpiphdr
*)q
->ti_next
)
149 if (SEQ_GT(q
->ti_seq
, ti
->ti_seq
))
153 * If there is a preceding segment, it may provide some of
154 * our data already. If so, drop the data from the incoming
155 * segment. If it provides all of our data, drop us.
157 if ((struct tcpiphdr
*)q
->ti_prev
!= (struct tcpiphdr
*)tp
) {
159 q
= (struct tcpiphdr
*)q
->ti_prev
;
160 /* conversion to int (in i) handles seq wraparound */
161 i
= q
->ti_seq
+ q
->ti_len
- ti
->ti_seq
;
163 if (i
>= ti
->ti_len
) {
164 tcpstat
.tcps_rcvduppack
++;
165 tcpstat
.tcps_rcvdupbyte
+= ti
->ti_len
;
168 * Try to present any queued data
169 * at the left window edge to the user.
170 * This is needed after the 3-WHS
173 goto present
; /* ??? */
179 q
= (struct tcpiphdr
*)(q
->ti_next
);
181 tcpstat
.tcps_rcvoopack
++;
182 tcpstat
.tcps_rcvoobyte
+= ti
->ti_len
;
183 REASS_MBUF(ti
) = (mbufp_32
) m
; /* XXX */
186 * While we overlap succeeding segments trim them or,
187 * if they are completely covered, dequeue them.
189 while (q
!= (struct tcpiphdr
*)tp
) {
190 register int i
= (ti
->ti_seq
+ ti
->ti_len
) - q
->ti_seq
;
196 m_adj((struct mbuf
*) REASS_MBUF(q
), i
);
199 q
= (struct tcpiphdr
*)q
->ti_next
;
200 m
= (struct mbuf
*) REASS_MBUF((struct tcpiphdr
*)q
->ti_prev
);
201 remque_32((void *)(q
->ti_prev
));
206 * Stick new segment in its place.
208 insque_32(ti
, (void *)(q
->ti_prev
));
212 * Present data to user, advancing rcv_nxt through
213 * completed sequence space.
215 if (!TCPS_HAVEESTABLISHED(tp
->t_state
))
217 ti
= (struct tcpiphdr
*) tp
->seg_next
;
218 if (ti
== (struct tcpiphdr
*)tp
|| ti
->ti_seq
!= tp
->rcv_nxt
)
220 if (tp
->t_state
== TCPS_SYN_RECEIVED
&& ti
->ti_len
)
223 tp
->rcv_nxt
+= ti
->ti_len
;
224 flags
= ti
->ti_flags
& TH_FIN
;
226 m
= (struct mbuf
*) REASS_MBUF(ti
); /* XXX */
227 ti
= (struct tcpiphdr
*)ti
->ti_next
;
228 /* if (so->so_state & SS_FCANTRCVMORE) */
229 if (so
->so_state
& SS_FCANTSENDMORE
)
233 if (tcp_emu(so
,m
)) sbappend(so
, m
);
237 } while (ti
!= (struct tcpiphdr
*)tp
&& ti
->ti_seq
== tp
->rcv_nxt
);
243 * TCP input routine, follows pages 65-76 of the
244 * protocol specification dated September, 1981 very closely.
247 tcp_input(m
, iphlen
, inso
)
248 register struct mbuf
*m
;
252 struct ip save_ip
, *ip
;
253 register struct tcpiphdr
*ti
;
257 register struct tcpcb
*tp
= 0;
258 register int tiflags
;
259 struct socket
*so
= 0;
260 int todrop
, acked
, ourfinisacked
, needoutput
= 0;
261 /* int dropsocket = 0; */
265 /* int ts_present = 0; */
267 DEBUG_CALL("tcp_input");
268 DEBUG_ARGS((dfd
," m = %8lx iphlen = %2d inso = %lx\n",
269 (long )m
, iphlen
, (long )inso
));
272 * If called with m == 0, then we're continuing the connect
277 /* Re-set a few variables */
283 tiflags
= ti
->ti_flags
;
289 tcpstat
.tcps_rcvtotal
++;
291 * Get IP and TCP header together in first mbuf.
292 * Note: IP leaves IP header in first mbuf.
294 ti
= mtod(m
, struct tcpiphdr
*);
295 if (iphlen
> sizeof(struct ip
)) {
296 ip_stripoptions(m
, (struct mbuf
*)0);
297 iphlen
=sizeof(struct ip
);
299 /* XXX Check if too short */
303 * Save a copy of the IP header in case we want restore it
304 * for sending an ICMP error message in response.
306 ip
=mtod(m
, struct ip
*);
310 * Checksum extended TCP header and data.
312 //tlen = ((struct ip *)ti)->ip_len;
313 // use save_ip instead of ti to be avoid gcc aliasing optimization problems
315 ti
->ti_next
= ti
->ti_prev
= 0;
317 ti
->ti_len
= htons((u_int16_t
)tlen
);
318 len
= sizeof(struct ip
) + tlen
;
319 /* keep checksum for ICMP reply
320 * ti->ti_sum = cksum(m, len);
321 * if (ti->ti_sum) { */
323 tcpstat
.tcps_rcvbadsum
++;
327 save_ip
.ip_len
+= iphlen
;
330 * Check that TCP offset makes sense,
331 * pull out TCP options and adjust length. XXX
333 off
= ti
->ti_off
<< 2;
334 if (off
< sizeof (struct tcphdr
) || off
> tlen
) {
335 tcpstat
.tcps_rcvbadoff
++;
340 if (off
> sizeof (struct tcphdr
)) {
341 optlen
= off
- sizeof (struct tcphdr
);
342 optp
= mtod(m
, caddr_t
) + sizeof (struct tcpiphdr
);
345 * Do quick retrieval of timestamp options ("options
346 * prediction?"). If timestamp is the only option and it's
347 * formatted as recommended in RFC 1323 appendix A, we
348 * quickly get the values now and not bother calling
349 * tcp_dooptions(), etc.
351 /* if ((optlen == TCPOLEN_TSTAMP_APPA ||
352 * (optlen > TCPOLEN_TSTAMP_APPA &&
353 * optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
354 * *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
355 * (ti->ti_flags & TH_SYN) == 0) {
357 * ts_val = ntohl(*(u_int32_t *)(optp + 4));
358 * ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
359 * optp = NULL; / * we've parsed the options * /
363 tiflags
= ti
->ti_flags
;
366 * Convert TCP protocol specific fields to host format.
374 * Drop TCP, IP headers and TCP options.
376 m
->m_data
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
377 m
->m_len
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
380 * Locate pcb for segment.
384 if (so
->so_fport
!= ti
->ti_dport
||
385 so
->so_lport
!= ti
->ti_sport
||
386 so
->so_laddr
.s_addr
!= ti
->ti_src
.s_addr
||
387 so
->so_faddr
.s_addr
!= ti
->ti_dst
.s_addr
) {
388 so
= solookup(&tcb
, ti
->ti_src
, ti
->ti_sport
,
389 ti
->ti_dst
, ti
->ti_dport
);
392 ++tcpstat
.tcps_socachemiss
;
396 * If the state is CLOSED (i.e., TCB does not exist) then
397 * all data in the incoming segment is discarded.
398 * If the TCB exists but is in CLOSED state, it is embryonic,
399 * but should either do a listen or a connect soon.
401 * state == CLOSED means we've done socreate() but haven't
402 * attached it to a protocol yet...
404 * XXX If a TCB does not exist, and the TH_SYN flag is
405 * the only flag set, then create a session, mark it
406 * as if it was LISTENING, and continue...
409 if ((tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) != TH_SYN
)
412 if ((so
= socreate()) == NULL
)
414 if (tcp_attach(so
) < 0) {
415 free(so
); /* Not sofree (if it failed, it's not insqued) */
419 sbreserve(&so
->so_snd
, tcp_sndspace
);
420 sbreserve(&so
->so_rcv
, tcp_rcvspace
);
422 /* tcp_last_so = so; */ /* XXX ? */
423 /* tp = sototcpcb(so); */
425 so
->so_laddr
= ti
->ti_src
;
426 so
->so_lport
= ti
->ti_sport
;
427 so
->so_faddr
= ti
->ti_dst
;
428 so
->so_fport
= ti
->ti_dport
;
430 if ((so
->so_iptos
= tcp_tos(so
)) == 0)
431 so
->so_iptos
= ((struct ip
*)ti
)->ip_tos
;
434 tp
->t_state
= TCPS_LISTEN
;
438 * If this is a still-connecting socket, this probably
439 * a retransmit of the SYN. Whether it's a retransmit SYN
440 * or something else, we nuke it.
442 if (so
->so_state
& SS_ISFCONNECTING
)
446 /* XXX Should never fail */
449 if (tp
->t_state
== TCPS_CLOSED
)
452 /* Unscale the window into a 32-bit value. */
453 /* if ((tiflags & TH_SYN) == 0)
454 * tiwin = ti->ti_win << tp->snd_scale;
460 * Segment received on connection.
461 * Reset idle time and keep-alive timer.
465 tp
->t_timer
[TCPT_KEEP
] = tcp_keepintvl
;
467 tp
->t_timer
[TCPT_KEEP
] = tcp_keepidle
;
470 * Process options if not in LISTEN state,
471 * else do it below (after getting remote address).
473 if (optp
&& tp
->t_state
!= TCPS_LISTEN
)
474 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
476 /* &ts_present, &ts_val, &ts_ecr); */
479 * Header prediction: check for the two common cases
480 * of a uni-directional data xfer. If the packet has
481 * no control flags, is in-sequence, the window didn't
482 * change and we're not retransmitting, it's a
483 * candidate. If the length is zero and the ack moved
484 * forward, we're the sender side of the xfer. Just
485 * free the data acked & wake any higher level process
486 * that was blocked waiting for space. If the length
487 * is non-zero and the ack didn't move, we're the
488 * receiver side. If we're getting packets in-order
489 * (the reassembly queue is empty), add the data to
490 * the socket buffer and note that we need a delayed ack.
492 * XXX Some of these tests are not needed
493 * eg: the tiwin == tp->snd_wnd prevents many more
494 * predictions.. with no *real* advantage..
496 if (tp
->t_state
== TCPS_ESTABLISHED
&&
497 (tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) == TH_ACK
&&
498 /* (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && */
499 ti
->ti_seq
== tp
->rcv_nxt
&&
500 tiwin
&& tiwin
== tp
->snd_wnd
&&
501 tp
->snd_nxt
== tp
->snd_max
) {
503 * If last ACK falls within this segment's sequence numbers,
504 * record the timestamp.
506 /* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
507 * SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
508 * tp->ts_recent_age = tcp_now;
509 * tp->ts_recent = ts_val;
512 if (ti
->ti_len
== 0) {
513 if (SEQ_GT(ti
->ti_ack
, tp
->snd_una
) &&
514 SEQ_LEQ(ti
->ti_ack
, tp
->snd_max
) &&
515 tp
->snd_cwnd
>= tp
->snd_wnd
) {
517 * this is a pure ack for outstanding data.
519 ++tcpstat
.tcps_predack
;
521 * tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
524 SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
525 tcp_xmit_timer(tp
, tp
->t_rtt
);
526 acked
= ti
->ti_ack
- tp
->snd_una
;
527 tcpstat
.tcps_rcvackpack
++;
528 tcpstat
.tcps_rcvackbyte
+= acked
;
529 sbdrop(&so
->so_snd
, acked
);
530 tp
->snd_una
= ti
->ti_ack
;
534 * If all outstanding data are acked, stop
535 * retransmit timer, otherwise restart timer
536 * using current (possibly backed-off) value.
537 * If process is waiting for space,
538 * wakeup/selwakeup/signal. If data
539 * are ready to send, let tcp_output
540 * decide between more output or persist.
542 if (tp
->snd_una
== tp
->snd_max
)
543 tp
->t_timer
[TCPT_REXMT
] = 0;
544 else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
545 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
548 * There's room in so_snd, sowwakup will read()
549 * from the socket if we can
551 /* if (so->so_snd.sb_flags & SB_NOTIFY)
555 * This is called because sowwakeup might have
556 * put data into so_snd. Since we don't so sowwakeup,
557 * we don't need this.. XXX???
559 if (so
->so_snd
.sb_cc
)
560 (void) tcp_output(tp
);
564 } else if (ti
->ti_ack
== tp
->snd_una
&&
565 tp
->seg_next
== (tcpiphdrp_32
)tp
&&
566 ti
->ti_len
<= sbspace(&so
->so_rcv
)) {
568 * this is a pure, in-sequence data packet
569 * with nothing on the reassembly queue and
570 * we have enough buffer space to take it.
572 ++tcpstat
.tcps_preddat
;
573 tp
->rcv_nxt
+= ti
->ti_len
;
574 tcpstat
.tcps_rcvpack
++;
575 tcpstat
.tcps_rcvbyte
+= ti
->ti_len
;
577 * Add data to socket buffer.
580 if (tcp_emu(so
,m
)) sbappend(so
, m
);
585 * XXX This is called when data arrives. Later, check
586 * if we can actually write() to the socket
587 * XXX Need to check? It's be NON_BLOCKING
592 * If this is a short packet, then ACK now - with Nagel
593 * congestion avoidance sender won't send more until
596 * Here are 3 interpretations of what should happen.
597 * The best (for me) is to delay-ack everything except
598 * if it's a one-byte packet containing an ESC
599 * (this means it's an arrow key (or similar) sent using
600 * Nagel, hence there will be no echo)
601 * The first of these is the original, the second is the
602 * middle ground between the other 2
604 /* if (((unsigned)ti->ti_len < tp->t_maxseg)) {
606 /* if (((unsigned)ti->ti_len < tp->t_maxseg &&
607 * (so->so_iptos & IPTOS_LOWDELAY) == 0) ||
608 * ((so->so_iptos & IPTOS_LOWDELAY) &&
609 * ((struct tcpiphdr_2 *)ti)->first_char == (char)27)) {
611 if ((unsigned)ti
->ti_len
== 1 &&
612 ((struct tcpiphdr_2
*)ti
)->first_char
== (char)27) {
613 tp
->t_flags
|= TF_ACKNOW
;
616 tp
->t_flags
|= TF_DELACK
;
620 } /* header prediction */
622 * Calculate amount of space in receive window,
623 * and then do TCP input processing.
624 * Receive window is amount of space in rcv queue,
625 * but not less than advertised window.
628 win
= sbspace(&so
->so_rcv
);
631 tp
->rcv_wnd
= max(win
, (int)(tp
->rcv_adv
- tp
->rcv_nxt
));
634 switch (tp
->t_state
) {
637 * If the state is LISTEN then ignore segment if it contains an RST.
638 * If the segment contains an ACK then it is bad and send a RST.
639 * If it does not contain a SYN then it is not interesting; drop it.
640 * Don't bother responding if the destination was a broadcast.
641 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
642 * tp->iss, and send a segment:
643 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
644 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
645 * Fill in remote peer address fields if not previously specified.
646 * Enter SYN_RECEIVED state, and process any other fields of this
647 * segment in this state.
651 if (tiflags
& TH_RST
)
653 if (tiflags
& TH_ACK
)
655 if ((tiflags
& TH_SYN
) == 0)
659 * This has way too many gotos...
660 * But a bit of spaghetti code never hurt anybody :)
664 * If this is destined for the control address, then flag to
665 * tcp_ctl once connected, otherwise connect
667 if ((so
->so_faddr
.s_addr
&htonl(0xffffff00)) == special_addr
.s_addr
) {
668 int lastbyte
=ntohl(so
->so_faddr
.s_addr
) & 0xff;
669 if (lastbyte
!=CTL_ALIAS
&& lastbyte
!=CTL_DNS
) {
671 if(lastbyte
==CTL_CMD
|| lastbyte
==CTL_EXEC
) {
672 /* Command or exec adress */
673 so
->so_state
|= SS_CTL
;
675 /* May be an add exec */
676 struct ex_list
*ex_ptr
;
678 for(ex_ptr
= exec_list
; ex_ptr
; ex_ptr
= ex_ptr
->ex_next
) {
679 if(ex_ptr
->ex_fport
== so
->so_fport
&&
680 lastbyte
== ex_ptr
->ex_addr
) {
681 so
->so_state
|= SS_CTL
;
686 if(so
->so_state
& SS_CTL
) goto cont_input
;
689 /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
692 if (so
->so_emu
& EMU_NOCONNECT
) {
693 so
->so_emu
&= ~EMU_NOCONNECT
;
697 if(tcp_fconnect(so
) == -1 && errno
!= EINPROGRESS
) {
698 u_char code
=ICMP_UNREACH_NET
;
699 DEBUG_MISC((dfd
," tcp fconnect errno = %d-%s\n",
700 errno
,strerror(errno
)));
701 if(errno
== ECONNREFUSED
) {
702 /* ACK the SYN, send RST to refuse the connection */
703 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+1, (tcp_seq
)0,
706 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
707 HTONL(ti
->ti_seq
); /* restore tcp header */
711 m
->m_data
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
712 m
->m_len
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
714 icmp_error(m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
720 * Haven't connected yet, save the current mbuf
722 * XXX Some OS's don't tell us whether the connect()
723 * succeeded or not. So we must time it out.
727 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
728 tp
->t_state
= TCPS_SYN_RECEIVED
;
734 * Check if the connect succeeded
736 if (so
->so_state
& SS_NOFDREF
) {
744 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
746 /* &ts_present, &ts_val, &ts_ecr); */
752 tcp_iss
+= TCP_ISSINCR
/2;
753 tp
->irs
= ti
->ti_seq
;
756 tp
->t_flags
|= TF_ACKNOW
;
757 tp
->t_state
= TCPS_SYN_RECEIVED
;
758 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
759 tcpstat
.tcps_accepts
++;
761 } /* case TCPS_LISTEN */
764 * If the state is SYN_SENT:
765 * if seg contains an ACK, but not for our SYN, drop the input.
766 * if seg contains a RST, then drop the connection.
767 * if seg does not contain SYN, then drop it.
768 * Otherwise this is an acceptable SYN segment
769 * initialize tp->rcv_nxt and tp->irs
770 * if seg contains ack then advance tp->snd_una
771 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
772 * arrange for segment to be acked (eventually)
773 * continue processing rest of data/controls, beginning with URG
776 if ((tiflags
& TH_ACK
) &&
777 (SEQ_LEQ(ti
->ti_ack
, tp
->iss
) ||
778 SEQ_GT(ti
->ti_ack
, tp
->snd_max
)))
781 if (tiflags
& TH_RST
) {
782 if (tiflags
& TH_ACK
)
783 tp
= tcp_drop(tp
,0); /* XXX Check t_softerror! */
787 if ((tiflags
& TH_SYN
) == 0)
789 if (tiflags
& TH_ACK
) {
790 tp
->snd_una
= ti
->ti_ack
;
791 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
792 tp
->snd_nxt
= tp
->snd_una
;
795 tp
->t_timer
[TCPT_REXMT
] = 0;
796 tp
->irs
= ti
->ti_seq
;
798 tp
->t_flags
|= TF_ACKNOW
;
799 if (tiflags
& TH_ACK
&& SEQ_GT(tp
->snd_una
, tp
->iss
)) {
800 tcpstat
.tcps_connects
++;
802 tp
->t_state
= TCPS_ESTABLISHED
;
804 /* Do window scaling on this connection? */
805 /* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
806 * (TF_RCVD_SCALE|TF_REQ_SCALE)) {
807 * tp->snd_scale = tp->requested_s_scale;
808 * tp->rcv_scale = tp->request_r_scale;
811 (void) tcp_reass(tp
, (struct tcpiphdr
*)0,
814 * if we didn't have to retransmit the SYN,
815 * use its rtt as our initial srtt & rtt var.
818 tcp_xmit_timer(tp
, tp
->t_rtt
);
820 tp
->t_state
= TCPS_SYN_RECEIVED
;
824 * Advance ti->ti_seq to correspond to first data byte.
825 * If data, trim to stay within window,
826 * dropping FIN if necessary.
829 if (ti
->ti_len
> tp
->rcv_wnd
) {
830 todrop
= ti
->ti_len
- tp
->rcv_wnd
;
832 ti
->ti_len
= tp
->rcv_wnd
;
834 tcpstat
.tcps_rcvpackafterwin
++;
835 tcpstat
.tcps_rcvbyteafterwin
+= todrop
;
837 tp
->snd_wl1
= ti
->ti_seq
- 1;
838 tp
->rcv_up
= ti
->ti_seq
;
840 } /* switch tp->t_state */
842 * States other than LISTEN or SYN_SENT.
843 * First check timestamp, if present.
844 * Then check that at least some bytes of segment are within
845 * receive window. If segment begins before rcv_nxt,
846 * drop leading data (and SYN); if nothing left, just ack.
848 * RFC 1323 PAWS: If we have a timestamp reply on this segment
849 * and it's less than ts_recent, drop it.
851 /* if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
852 * TSTMP_LT(ts_val, tp->ts_recent)) {
854 */ /* Check to see if ts_recent is over 24 days old. */
855 /* if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
857 * * Invalidate ts_recent. If this segment updates
858 * * ts_recent, the age will be reset later and ts_recent
859 * * will get a valid value. If it does not, setting
860 * * ts_recent to zero will at least satisfy the
861 * * requirement that zero be placed in the timestamp
862 * * echo reply when ts_recent isn't valid. The
863 * * age isn't reset until we get a valid ts_recent
864 * * because we don't want out-of-order segments to be
865 * * dropped when ts_recent is old.
867 /* tp->ts_recent = 0;
869 * tcpstat.tcps_rcvduppack++;
870 * tcpstat.tcps_rcvdupbyte += ti->ti_len;
871 * tcpstat.tcps_pawsdrop++;
877 todrop
= tp
->rcv_nxt
- ti
->ti_seq
;
879 if (tiflags
& TH_SYN
) {
889 * Following if statement from Stevens, vol. 2, p. 960.
891 if (todrop
> ti
->ti_len
892 || (todrop
== ti
->ti_len
&& (tiflags
& TH_FIN
) == 0)) {
894 * Any valid FIN must be to the left of the window.
895 * At this point the FIN must be a duplicate or out
896 * of sequence; drop it.
901 * Send an ACK to resynchronize and drop any data.
902 * But keep on processing for RST or ACK.
904 tp
->t_flags
|= TF_ACKNOW
;
906 tcpstat
.tcps_rcvduppack
++;
907 tcpstat
.tcps_rcvdupbyte
+= todrop
;
909 tcpstat
.tcps_rcvpartduppack
++;
910 tcpstat
.tcps_rcvpartdupbyte
+= todrop
;
913 ti
->ti_seq
+= todrop
;
914 ti
->ti_len
-= todrop
;
915 if (ti
->ti_urp
> todrop
)
916 ti
->ti_urp
-= todrop
;
923 * If new data are received on a connection after the
924 * user processes are gone, then RST the other end.
926 if ((so
->so_state
& SS_NOFDREF
) &&
927 tp
->t_state
> TCPS_CLOSE_WAIT
&& ti
->ti_len
) {
929 tcpstat
.tcps_rcvafterclose
++;
934 * If segment ends after window, drop trailing data
935 * (and PUSH and FIN); if nothing left, just ACK.
937 todrop
= (ti
->ti_seq
+ti
->ti_len
) - (tp
->rcv_nxt
+tp
->rcv_wnd
);
939 tcpstat
.tcps_rcvpackafterwin
++;
940 if (todrop
>= ti
->ti_len
) {
941 tcpstat
.tcps_rcvbyteafterwin
+= ti
->ti_len
;
943 * If a new connection request is received
944 * while in TIME_WAIT, drop the old connection
945 * and start over if the sequence numbers
946 * are above the previous ones.
948 if (tiflags
& TH_SYN
&&
949 tp
->t_state
== TCPS_TIME_WAIT
&&
950 SEQ_GT(ti
->ti_seq
, tp
->rcv_nxt
)) {
951 iss
= tp
->rcv_nxt
+ TCP_ISSINCR
;
956 * If window is closed can only take segments at
957 * window edge, and have to drop data and PUSH from
958 * incoming segments. Continue processing, but
959 * remember to ack. Otherwise, drop segment
962 if (tp
->rcv_wnd
== 0 && ti
->ti_seq
== tp
->rcv_nxt
) {
963 tp
->t_flags
|= TF_ACKNOW
;
964 tcpstat
.tcps_rcvwinprobe
++;
968 tcpstat
.tcps_rcvbyteafterwin
+= todrop
;
970 ti
->ti_len
-= todrop
;
971 tiflags
&= ~(TH_PUSH
|TH_FIN
);
975 * If last ACK falls within this segment's sequence numbers,
976 * record its timestamp.
978 /* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
979 * SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
980 * ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
981 * tp->ts_recent_age = tcp_now;
982 * tp->ts_recent = ts_val;
987 * If the RST bit is set examine the state:
988 * SYN_RECEIVED STATE:
989 * If passive open, return to LISTEN state.
990 * If active open, inform user that connection was refused.
991 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
992 * Inform user that connection was reset, and close tcb.
993 * CLOSING, LAST_ACK, TIME_WAIT STATES
996 if (tiflags
&TH_RST
) switch (tp
->t_state
) {
998 case TCPS_SYN_RECEIVED
:
999 /* so->so_error = ECONNREFUSED; */
1002 case TCPS_ESTABLISHED
:
1003 case TCPS_FIN_WAIT_1
:
1004 case TCPS_FIN_WAIT_2
:
1005 case TCPS_CLOSE_WAIT
:
1006 /* so->so_error = ECONNRESET; */
1008 tp
->t_state
= TCPS_CLOSED
;
1009 tcpstat
.tcps_drops
++;
1015 case TCPS_TIME_WAIT
:
1021 * If a SYN is in the window, then this is an
1022 * error and we send an RST and drop the connection.
1024 if (tiflags
& TH_SYN
) {
1025 tp
= tcp_drop(tp
,0);
1030 * If the ACK bit is off we drop the segment and return.
1032 if ((tiflags
& TH_ACK
) == 0) goto drop
;
1037 switch (tp
->t_state
) {
1039 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1040 * ESTABLISHED state and continue processing, otherwise
1041 * send an RST. una<=ack<=max
1043 case TCPS_SYN_RECEIVED
:
1045 if (SEQ_GT(tp
->snd_una
, ti
->ti_ack
) ||
1046 SEQ_GT(ti
->ti_ack
, tp
->snd_max
))
1048 tcpstat
.tcps_connects
++;
1049 tp
->t_state
= TCPS_ESTABLISHED
;
1051 * The sent SYN is ack'ed with our sequence number +1
1052 * The first data byte already in the buffer will get
1053 * lost if no correction is made. This is only needed for
1054 * SS_CTL since the buffer is empty otherwise.
1055 * tp->snd_una++; or:
1057 tp
->snd_una
=ti
->ti_ack
;
1058 if (so
->so_state
& SS_CTL
) {
1059 /* So tcp_ctl reports the right state */
1063 so
->so_state
&= ~SS_CTL
; /* success XXX */
1064 } else if (ret
== 2) {
1065 so
->so_state
= SS_NOFDREF
; /* CTL_CMD */
1068 tp
->t_state
= TCPS_FIN_WAIT_1
;
1074 /* Do window scaling? */
1075 /* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1076 * (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1077 * tp->snd_scale = tp->requested_s_scale;
1078 * tp->rcv_scale = tp->request_r_scale;
1081 (void) tcp_reass(tp
, (struct tcpiphdr
*)0, (struct mbuf
*)0);
1082 tp
->snd_wl1
= ti
->ti_seq
- 1;
1083 /* Avoid ack processing; snd_una==ti_ack => dup ack */
1088 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1089 * ACKs. If the ack is in the range
1090 * tp->snd_una < ti->ti_ack <= tp->snd_max
1091 * then advance tp->snd_una to ti->ti_ack and drop
1092 * data from the retransmission queue. If this ACK reflects
1093 * more up to date window information we update our window information.
1095 case TCPS_ESTABLISHED
:
1096 case TCPS_FIN_WAIT_1
:
1097 case TCPS_FIN_WAIT_2
:
1098 case TCPS_CLOSE_WAIT
:
1101 case TCPS_TIME_WAIT
:
1103 if (SEQ_LEQ(ti
->ti_ack
, tp
->snd_una
)) {
1104 if (ti
->ti_len
== 0 && tiwin
== tp
->snd_wnd
) {
1105 tcpstat
.tcps_rcvdupack
++;
1106 DEBUG_MISC((dfd
," dup ack m = %lx so = %lx \n",
1107 (long )m
, (long )so
));
1109 * If we have outstanding data (other than
1110 * a window probe), this is a completely
1111 * duplicate ack (ie, window info didn't
1112 * change), the ack is the biggest we've
1113 * seen and we've seen exactly our rexmt
1114 * threshold of them, assume a packet
1115 * has been dropped and retransmit it.
1116 * Kludge snd_nxt & the congestion
1117 * window so we send only this one
1120 * We know we're losing at the current
1121 * window size so do congestion avoidance
1122 * (set ssthresh to half the current window
1123 * and pull our congestion window back to
1124 * the new ssthresh).
1126 * Dup acks mean that packets have left the
1127 * network (they're now cached at the receiver)
1128 * so bump cwnd by the amount in the receiver
1129 * to keep a constant cwnd packets in the
1132 if (tp
->t_timer
[TCPT_REXMT
] == 0 ||
1133 ti
->ti_ack
!= tp
->snd_una
)
1135 else if (++tp
->t_dupacks
== tcprexmtthresh
) {
1136 tcp_seq onxt
= tp
->snd_nxt
;
1138 min(tp
->snd_wnd
, tp
->snd_cwnd
) / 2 /
1143 tp
->snd_ssthresh
= win
* tp
->t_maxseg
;
1144 tp
->t_timer
[TCPT_REXMT
] = 0;
1146 tp
->snd_nxt
= ti
->ti_ack
;
1147 tp
->snd_cwnd
= tp
->t_maxseg
;
1148 (void) tcp_output(tp
);
1149 tp
->snd_cwnd
= tp
->snd_ssthresh
+
1150 tp
->t_maxseg
* tp
->t_dupacks
;
1151 if (SEQ_GT(onxt
, tp
->snd_nxt
))
1154 } else if (tp
->t_dupacks
> tcprexmtthresh
) {
1155 tp
->snd_cwnd
+= tp
->t_maxseg
;
1156 (void) tcp_output(tp
);
1165 * If the congestion window was inflated to account
1166 * for the other side's cached packets, retract it.
1168 if (tp
->t_dupacks
> tcprexmtthresh
&&
1169 tp
->snd_cwnd
> tp
->snd_ssthresh
)
1170 tp
->snd_cwnd
= tp
->snd_ssthresh
;
1172 if (SEQ_GT(ti
->ti_ack
, tp
->snd_max
)) {
1173 tcpstat
.tcps_rcvacktoomuch
++;
1176 acked
= ti
->ti_ack
- tp
->snd_una
;
1177 tcpstat
.tcps_rcvackpack
++;
1178 tcpstat
.tcps_rcvackbyte
+= acked
;
1181 * If we have a timestamp reply, update smoothed
1182 * round trip time. If no timestamp is present but
1183 * transmit timer is running and timed sequence
1184 * number was acked, update smoothed round trip time.
1185 * Since we now have an rtt measurement, cancel the
1186 * timer backoff (cf., Phil Karn's retransmit alg.).
1187 * Recompute the initial retransmit timer.
1190 * tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1193 if (tp
->t_rtt
&& SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
1194 tcp_xmit_timer(tp
,tp
->t_rtt
);
1197 * If all outstanding data is acked, stop retransmit
1198 * timer and remember to restart (more output or persist).
1199 * If there is more data to be acked, restart retransmit
1200 * timer, using current (possibly backed-off) value.
1202 if (ti
->ti_ack
== tp
->snd_max
) {
1203 tp
->t_timer
[TCPT_REXMT
] = 0;
1205 } else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
1206 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
1208 * When new data is acked, open the congestion window.
1209 * If the window gives us less than ssthresh packets
1210 * in flight, open exponentially (maxseg per packet).
1211 * Otherwise open linearly: maxseg per window
1212 * (maxseg^2 / cwnd per packet).
1215 register u_int cw
= tp
->snd_cwnd
;
1216 register u_int incr
= tp
->t_maxseg
;
1218 if (cw
> tp
->snd_ssthresh
)
1219 incr
= incr
* incr
/ cw
;
1220 tp
->snd_cwnd
= min(cw
+ incr
, TCP_MAXWIN
<<tp
->snd_scale
);
1222 if (acked
> so
->so_snd
.sb_cc
) {
1223 tp
->snd_wnd
-= so
->so_snd
.sb_cc
;
1224 sbdrop(&so
->so_snd
, (int )so
->so_snd
.sb_cc
);
1227 sbdrop(&so
->so_snd
, acked
);
1228 tp
->snd_wnd
-= acked
;
1232 * XXX sowwakup is called when data is acked and there's room for
1233 * for more data... it should read() the socket
1235 /* if (so->so_snd.sb_flags & SB_NOTIFY)
1238 tp
->snd_una
= ti
->ti_ack
;
1239 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
1240 tp
->snd_nxt
= tp
->snd_una
;
1242 switch (tp
->t_state
) {
1245 * In FIN_WAIT_1 STATE in addition to the processing
1246 * for the ESTABLISHED state if our FIN is now acknowledged
1247 * then enter FIN_WAIT_2.
1249 case TCPS_FIN_WAIT_1
:
1250 if (ourfinisacked
) {
1252 * If we can't receive any more
1253 * data, then closing user can proceed.
1254 * Starting the timer is contrary to the
1255 * specification, but if we don't get a FIN
1256 * we'll hang forever.
1258 if (so
->so_state
& SS_FCANTRCVMORE
) {
1259 soisfdisconnected(so
);
1260 tp
->t_timer
[TCPT_2MSL
] = tcp_maxidle
;
1262 tp
->t_state
= TCPS_FIN_WAIT_2
;
1267 * In CLOSING STATE in addition to the processing for
1268 * the ESTABLISHED state if the ACK acknowledges our FIN
1269 * then enter the TIME-WAIT state, otherwise ignore
1273 if (ourfinisacked
) {
1274 tp
->t_state
= TCPS_TIME_WAIT
;
1275 tcp_canceltimers(tp
);
1276 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1277 soisfdisconnected(so
);
1282 * In LAST_ACK, we may still be waiting for data to drain
1283 * and/or to be acked, as well as for the ack of our FIN.
1284 * If our FIN is now acknowledged, delete the TCB,
1285 * enter the closed state and return.
1288 if (ourfinisacked
) {
1295 * In TIME_WAIT state the only thing that should arrive
1296 * is a retransmission of the remote FIN. Acknowledge
1297 * it and restart the finack timer.
1299 case TCPS_TIME_WAIT
:
1300 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1303 } /* switch(tp->t_state) */
1307 * Update window information.
1308 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1310 if ((tiflags
& TH_ACK
) &&
1311 (SEQ_LT(tp
->snd_wl1
, ti
->ti_seq
) ||
1312 (tp
->snd_wl1
== ti
->ti_seq
&& (SEQ_LT(tp
->snd_wl2
, ti
->ti_ack
) ||
1313 (tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
))))) {
1314 /* keep track of pure window updates */
1315 if (ti
->ti_len
== 0 &&
1316 tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
)
1317 tcpstat
.tcps_rcvwinupd
++;
1318 tp
->snd_wnd
= tiwin
;
1319 tp
->snd_wl1
= ti
->ti_seq
;
1320 tp
->snd_wl2
= ti
->ti_ack
;
1321 if (tp
->snd_wnd
> tp
->max_sndwnd
)
1322 tp
->max_sndwnd
= tp
->snd_wnd
;
1327 * Process segments with URG.
1329 if ((tiflags
& TH_URG
) && ti
->ti_urp
&&
1330 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1332 * This is a kludge, but if we receive and accept
1333 * random urgent pointers, we'll crash in
1334 * soreceive. It's hard to imagine someone
1335 * actually wanting to send this much urgent data.
1337 if (ti
->ti_urp
+ so
->so_rcv
.sb_cc
> so
->so_rcv
.sb_datalen
) {
1343 * If this segment advances the known urgent pointer,
1344 * then mark the data stream. This should not happen
1345 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1346 * a FIN has been received from the remote side.
1347 * In these states we ignore the URG.
1349 * According to RFC961 (Assigned Protocols),
1350 * the urgent pointer points to the last octet
1351 * of urgent data. We continue, however,
1352 * to consider it to indicate the first octet
1353 * of data past the urgent section as the original
1354 * spec states (in one of two places).
1356 if (SEQ_GT(ti
->ti_seq
+ti
->ti_urp
, tp
->rcv_up
)) {
1357 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1358 so
->so_urgc
= so
->so_rcv
.sb_cc
+
1359 (tp
->rcv_up
- tp
->rcv_nxt
); /* -1; */
1360 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1365 * If no out of band data is expected,
1366 * pull receive urgent pointer along
1367 * with the receive window.
1369 if (SEQ_GT(tp
->rcv_nxt
, tp
->rcv_up
))
1370 tp
->rcv_up
= tp
->rcv_nxt
;
1374 * Process the segment text, merging it into the TCP sequencing queue,
1375 * and arranging for acknowledgment of receipt if necessary.
1376 * This process logically involves adjusting tp->rcv_wnd as data
1377 * is presented to the user (this happens in tcp_usrreq.c,
1378 * case PRU_RCVD). If a FIN has already been received on this
1379 * connection then we just ignore the text.
1381 if ((ti
->ti_len
|| (tiflags
&TH_FIN
)) &&
1382 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1383 TCP_REASS(tp
, ti
, m
, so
, tiflags
);
1385 * Note the amount of data that peer has sent into
1386 * our window, in order to estimate the sender's
1389 len
= so
->so_rcv
.sb_datalen
- (tp
->rcv_adv
- tp
->rcv_nxt
);
1396 * If FIN is received ACK the FIN and let the user know
1397 * that the connection is closing.
1399 if (tiflags
& TH_FIN
) {
1400 if (TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1402 * If we receive a FIN we can't send more data,
1404 * Shutdown the socket if there is no rx data in the
1406 * soread() is called on completion of shutdown() and
1407 * will got to TCPS_LAST_ACK, and use tcp_output()
1410 /* sofcantrcvmore(so); */
1413 tp
->t_flags
|= TF_ACKNOW
;
1416 switch (tp
->t_state
) {
1419 * In SYN_RECEIVED and ESTABLISHED STATES
1420 * enter the CLOSE_WAIT state.
1422 case TCPS_SYN_RECEIVED
:
1423 case TCPS_ESTABLISHED
:
1424 if(so
->so_emu
== EMU_CTL
) /* no shutdown on socket */
1425 tp
->t_state
= TCPS_LAST_ACK
;
1427 tp
->t_state
= TCPS_CLOSE_WAIT
;
1431 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1432 * enter the CLOSING state.
1434 case TCPS_FIN_WAIT_1
:
1435 tp
->t_state
= TCPS_CLOSING
;
1439 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1440 * starting the time-wait timer, turning off the other
1443 case TCPS_FIN_WAIT_2
:
1444 tp
->t_state
= TCPS_TIME_WAIT
;
1445 tcp_canceltimers(tp
);
1446 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1447 soisfdisconnected(so
);
1451 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1453 case TCPS_TIME_WAIT
:
1454 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1460 * If this is a small packet, then ACK now - with Nagel
1461 * congestion avoidance sender won't send more until
1466 /* if (ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg) {
1468 /* if ((ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg &&
1469 * (so->so_iptos & IPTOS_LOWDELAY) == 0) ||
1470 * ((so->so_iptos & IPTOS_LOWDELAY) &&
1471 * ((struct tcpiphdr_2 *)ti)->first_char == (char)27)) {
1473 if (ti
->ti_len
&& (unsigned)ti
->ti_len
<= 5 &&
1474 ((struct tcpiphdr_2
*)ti
)->first_char
== (char)27) {
1475 tp
->t_flags
|= TF_ACKNOW
;
1479 * Return any desired output.
1481 if (needoutput
|| (tp
->t_flags
& TF_ACKNOW
)) {
1482 (void) tcp_output(tp
);
1488 * Generate an ACK dropping incoming segment if it occupies
1489 * sequence space, where the ACK reflects our state.
1491 if (tiflags
& TH_RST
)
1494 tp
->t_flags
|= TF_ACKNOW
;
1495 (void) tcp_output(tp
);
1499 /* reuses m if m!=NULL, m_free() unnecessary */
1500 if (tiflags
& TH_ACK
)
1501 tcp_respond(tp
, ti
, m
, (tcp_seq
)0, ti
->ti_ack
, TH_RST
);
1503 if (tiflags
& TH_SYN
) ti
->ti_len
++;
1504 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+ti
->ti_len
, (tcp_seq
)0,
1512 * Drop space held by incoming segment and return.
1519 /* , ts_present, ts_val, ts_ecr) */
1521 * u_int32_t *ts_val, *ts_ecr;
1524 tcp_dooptions(tp
, cp
, cnt
, ti
)
1528 struct tcpiphdr
*ti
;
1533 DEBUG_CALL("tcp_dooptions");
1534 DEBUG_ARGS((dfd
," tp = %lx cnt=%i \n", (long )tp
, cnt
));
1536 for (; cnt
> 0; cnt
-= optlen
, cp
+= optlen
) {
1538 if (opt
== TCPOPT_EOL
)
1540 if (opt
== TCPOPT_NOP
)
1553 if (optlen
!= TCPOLEN_MAXSEG
)
1555 if (!(ti
->ti_flags
& TH_SYN
))
1557 memcpy((char *) &mss
, (char *) cp
+ 2, sizeof(mss
));
1559 (void) tcp_mss(tp
, mss
); /* sets t_maxseg */
1562 /* case TCPOPT_WINDOW:
1563 * if (optlen != TCPOLEN_WINDOW)
1565 * if (!(ti->ti_flags & TH_SYN))
1567 * tp->t_flags |= TF_RCVD_SCALE;
1568 * tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1571 /* case TCPOPT_TIMESTAMP:
1572 * if (optlen != TCPOLEN_TIMESTAMP)
1575 * memcpy((char *) ts_val, (char *)cp + 2, sizeof(*ts_val));
1577 * memcpy((char *) ts_ecr, (char *)cp + 6, sizeof(*ts_ecr));
1581 * * A timestamp received in a SYN makes
1582 * * it ok to send timestamp requests and replies.
1584 /* if (ti->ti_flags & TH_SYN) {
1585 * tp->t_flags |= TF_RCVD_TSTMP;
1586 * tp->ts_recent = *ts_val;
1587 * tp->ts_recent_age = tcp_now;
1596 * Pull out of band byte out of a segment so
1597 * it doesn't appear in the user's data queue.
1598 * It is still reflected in the segment length for
1599 * sequencing purposes.
1605 tcp_pulloutofband(so
, ti
, m
)
1607 struct tcpiphdr
*ti
;
1608 register struct mbuf
*m
;
1610 int cnt
= ti
->ti_urp
- 1;
1613 if (m
->m_len
> cnt
) {
1614 char *cp
= mtod(m
, caddr_t
) + cnt
;
1615 struct tcpcb
*tp
= sototcpcb(so
);
1618 tp
->t_oobflags
|= TCPOOB_HAVEDATA
;
1619 memcpy(sp
, cp
+1, (unsigned)(m
->m_len
- cnt
- 1));
1624 m
= m
->m_next
; /* XXX WRONG! Fix it! */
1628 panic("tcp_pulloutofband");
1634 * Collect new round-trip time estimate
1635 * and update averages and current timeout.
1639 tcp_xmit_timer(tp
, rtt
)
1640 register struct tcpcb
*tp
;
1643 register short delta
;
1645 DEBUG_CALL("tcp_xmit_timer");
1646 DEBUG_ARG("tp = %lx", (long)tp
);
1647 DEBUG_ARG("rtt = %d", rtt
);
1649 tcpstat
.tcps_rttupdated
++;
1650 if (tp
->t_srtt
!= 0) {
1652 * srtt is stored as fixed point with 3 bits after the
1653 * binary point (i.e., scaled by 8). The following magic
1654 * is equivalent to the smoothing algorithm in rfc793 with
1655 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1656 * point). Adjust rtt to origin 0.
1658 delta
= rtt
- 1 - (tp
->t_srtt
>> TCP_RTT_SHIFT
);
1659 if ((tp
->t_srtt
+= delta
) <= 0)
1662 * We accumulate a smoothed rtt variance (actually, a
1663 * smoothed mean difference), then set the retransmit
1664 * timer to smoothed rtt + 4 times the smoothed variance.
1665 * rttvar is stored as fixed point with 2 bits after the
1666 * binary point (scaled by 4). The following is
1667 * equivalent to rfc793 smoothing with an alpha of .75
1668 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
1669 * rfc793's wired-in beta.
1673 delta
-= (tp
->t_rttvar
>> TCP_RTTVAR_SHIFT
);
1674 if ((tp
->t_rttvar
+= delta
) <= 0)
1678 * No rtt measurement yet - use the unsmoothed rtt.
1679 * Set the variance to half the rtt (so our first
1680 * retransmit happens at 3*rtt).
1682 tp
->t_srtt
= rtt
<< TCP_RTT_SHIFT
;
1683 tp
->t_rttvar
= rtt
<< (TCP_RTTVAR_SHIFT
- 1);
1689 * the retransmit should happen at rtt + 4 * rttvar.
1690 * Because of the way we do the smoothing, srtt and rttvar
1691 * will each average +1/2 tick of bias. When we compute
1692 * the retransmit timer, we want 1/2 tick of rounding and
1693 * 1 extra tick because of +-1/2 tick uncertainty in the
1694 * firing of the timer. The bias will give us exactly the
1695 * 1.5 tick we need. But, because the bias is
1696 * statistical, we have to test that we don't drop below
1697 * the minimum feasible timer (which is 2 ticks).
1699 TCPT_RANGESET(tp
->t_rxtcur
, TCP_REXMTVAL(tp
),
1700 (short)tp
->t_rttmin
, TCPTV_REXMTMAX
); /* XXX */
1703 * We received an ack for a packet that wasn't retransmitted;
1704 * it is probably safe to discard any error indications we've
1705 * received recently. This isn't quite right, but close enough
1706 * for now (a route might have failed after we sent a segment,
1707 * and the return path might not be symmetrical).
1709 tp
->t_softerror
= 0;
1713 * Determine a reasonable value for maxseg size.
1714 * If the route is known, check route for mtu.
1715 * If none, use an mss that can be handled on the outgoing
1716 * interface without forcing IP to fragment; if bigger than
1717 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1718 * to utilize large mbufs. If no route is found, route has no mtu,
1719 * or the destination isn't local, use a default, hopefully conservative
1720 * size (usually 512 or the default IP max size, but no more than the mtu
1721 * of the interface), as we can't discover anything about intervening
1722 * gateways or networks. We also initialize the congestion/slow start
1723 * window to be a single segment if the destination isn't local.
1724 * While looking at the routing entry, we also initialize other path-dependent
1725 * parameters from pre-set or cached values in the routing entry.
1730 register struct tcpcb
*tp
;
1733 struct socket
*so
= tp
->t_socket
;
1736 DEBUG_CALL("tcp_mss");
1737 DEBUG_ARG("tp = %lx", (long)tp
);
1738 DEBUG_ARG("offer = %d", offer
);
1740 mss
= min(if_mtu
, if_mru
) - sizeof(struct tcpiphdr
);
1742 mss
= min(mss
, offer
);
1744 if (mss
< tp
->t_maxseg
|| offer
!= 0)
1749 sbreserve(&so
->so_snd
, tcp_sndspace
+((tcp_sndspace
%mss
)?(mss
-(tcp_sndspace
%mss
)):0));
1750 sbreserve(&so
->so_rcv
, tcp_rcvspace
+((tcp_rcvspace
%mss
)?(mss
-(tcp_rcvspace
%mss
)):0));
1752 DEBUG_MISC((dfd
, " returning mss = %d\n", mss
));