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. 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.
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
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
34 * Changes and additions relating to SLiRP
35 * Copyright (c) 1995 Danny Gasparovski.
37 * Please read the file COPYRIGHT for the
38 * terms and conditions of the copyright.
46 #define TCPREXMTTHRESH 3
47 struct socket
*tcp_last_so
= &tcb
;
49 tcp_seq tcp_iss
; /* tcp initial send seq # */
51 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
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)
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).
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; \
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); \
81 if (tcp_emu((so),(m))) sbappend((so), (m)); \
83 sbappend((so), (m)); \
84 /* sorwakeup(so); */ \
86 (flags) = tcp_reass((tp), (ti), (m)); \
87 tp->t_flags |= TF_ACKNOW; \
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); \
101 if (tcp_emu((so),(m))) sbappend(so, (m)); \
103 sbappend((so), (m)); \
104 /* sorwakeup(so); */ \
106 (flags) = tcp_reass((tp), (ti), (m)); \
107 tp->t_flags |= TF_ACKNOW; \
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
);
116 tcp_reass(register struct tcpcb
*tp
, register struct tcpiphdr
*ti
,
119 register struct tcpiphdr
*q
;
120 struct socket
*so
= tp
->t_socket
;
124 * Call with ti==NULL after become established to
125 * force pre-ESTABLISHED data up to user socket.
131 * Find a segment which begins after this one does.
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
))
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.
143 if (!tcpfrag_list_end(tcpiphdr_prev(q
), tp
)) {
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
;
149 if (i
>= ti
->ti_len
) {
150 STAT(tcpstat
.tcps_rcvduppack
++);
151 STAT(tcpstat
.tcps_rcvdupbyte
+= ti
->ti_len
);
154 * Try to present any queued data
155 * at the left window edge to the user.
156 * This is needed after the 3-WHS
159 goto present
; /* ??? */
165 q
= tcpiphdr_next(q
);
167 STAT(tcpstat
.tcps_rcvoopack
++);
168 STAT(tcpstat
.tcps_rcvoobyte
+= ti
->ti_len
);
172 * While we overlap succeeding segments trim them or,
173 * if they are completely covered, dequeue them.
175 while (!tcpfrag_list_end(q
, tp
)) {
176 register int i
= (ti
->ti_seq
+ ti
->ti_len
) - q
->ti_seq
;
182 m_adj(q
->ti_mbuf
, i
);
185 q
= tcpiphdr_next(q
);
186 m
= tcpiphdr_prev(q
)->ti_mbuf
;
187 remque(tcpiphdr2qlink(tcpiphdr_prev(q
)));
192 * Stick new segment in its place.
194 insque(tcpiphdr2qlink(ti
), tcpiphdr2qlink(tcpiphdr_prev(q
)));
198 * Present data to user, advancing rcv_nxt through
199 * completed sequence space.
201 if (!TCPS_HAVEESTABLISHED(tp
->t_state
))
203 ti
= tcpfrag_list_first(tp
);
204 if (tcpfrag_list_end(ti
, tp
) || ti
->ti_seq
!= tp
->rcv_nxt
)
206 if (tp
->t_state
== TCPS_SYN_RECEIVED
&& ti
->ti_len
)
209 tp
->rcv_nxt
+= ti
->ti_len
;
210 flags
= ti
->ti_flags
& TH_FIN
;
211 remque(tcpiphdr2qlink(ti
));
213 ti
= tcpiphdr_next(ti
);
214 /* if (so->so_state & SS_FCANTRCVMORE) */
215 if (so
->so_state
& SS_FCANTSENDMORE
)
219 if (tcp_emu(so
,m
)) sbappend(so
, m
);
223 } while (ti
!= (struct tcpiphdr
*)tp
&& ti
->ti_seq
== tp
->rcv_nxt
);
229 * TCP input routine, follows pages 65-76 of the
230 * protocol specification dated September, 1981 very closely.
233 tcp_input(struct mbuf
*m
, int iphlen
, struct socket
*inso
)
235 struct ip save_ip
, *ip
;
236 register struct tcpiphdr
*ti
;
240 register struct tcpcb
*tp
= NULL
;
241 register int tiflags
;
242 struct socket
*so
= NULL
;
243 int todrop
, acked
, ourfinisacked
, needoutput
= 0;
244 /* int dropsocket = 0; */
248 /* int ts_present = 0; */
249 struct ex_list
*ex_ptr
;
251 DEBUG_CALL("tcp_input");
252 DEBUG_ARGS((dfd
," m = %8lx iphlen = %2d inso = %lx\n",
253 (long )m
, iphlen
, (long )inso
));
256 * If called with m == 0, then we're continuing the connect
261 /* Re-set a few variables */
267 tiflags
= ti
->ti_flags
;
273 STAT(tcpstat
.tcps_rcvtotal
++);
275 * Get IP and TCP header together in first mbuf.
276 * Note: IP leaves IP header in first mbuf.
278 ti
= mtod(m
, struct tcpiphdr
*);
279 if (iphlen
> sizeof(struct ip
)) {
280 ip_stripoptions(m
, (struct mbuf
*)0);
281 iphlen
=sizeof(struct ip
);
283 /* XXX Check if too short */
287 * Save a copy of the IP header in case we want restore it
288 * for sending an ICMP error message in response.
290 ip
=mtod(m
, struct ip
*);
292 save_ip
.ip_len
+= iphlen
;
295 * Checksum extended TCP header and data.
297 tlen
= ((struct ip
*)ti
)->ip_len
;
298 tcpiphdr2qlink(ti
)->next
= tcpiphdr2qlink(ti
)->prev
= NULL
;
299 memset(&ti
->ti_i
.ih_mbuf
, 0 , sizeof(struct mbuf_ptr
));
301 ti
->ti_len
= htons((u_int16_t
)tlen
);
302 len
= sizeof(struct ip
) + tlen
;
303 /* keep checksum for ICMP reply
304 * ti->ti_sum = cksum(m, len);
305 * if (ti->ti_sum) { */
307 STAT(tcpstat
.tcps_rcvbadsum
++);
312 * Check that TCP offset makes sense,
313 * pull out TCP options and adjust length. XXX
315 off
= ti
->ti_off
<< 2;
316 if (off
< sizeof (struct tcphdr
) || off
> tlen
) {
317 STAT(tcpstat
.tcps_rcvbadoff
++);
322 if (off
> sizeof (struct tcphdr
)) {
323 optlen
= off
- sizeof (struct tcphdr
);
324 optp
= mtod(m
, caddr_t
) + sizeof (struct tcpiphdr
);
327 * Do quick retrieval of timestamp options ("options
328 * prediction?"). If timestamp is the only option and it's
329 * formatted as recommended in RFC 1323 appendix A, we
330 * quickly get the values now and not bother calling
331 * tcp_dooptions(), etc.
333 /* if ((optlen == TCPOLEN_TSTAMP_APPA ||
334 * (optlen > TCPOLEN_TSTAMP_APPA &&
335 * optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
336 * *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
337 * (ti->ti_flags & TH_SYN) == 0) {
339 * ts_val = ntohl(*(u_int32_t *)(optp + 4));
340 * ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
341 * optp = NULL; / * we've parsed the options * /
345 tiflags
= ti
->ti_flags
;
348 * Convert TCP protocol specific fields to host format.
356 * Drop TCP, IP headers and TCP options.
358 m
->m_data
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
359 m
->m_len
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
361 if (slirp_restrict
) {
362 for (ex_ptr
= exec_list
; ex_ptr
; ex_ptr
= ex_ptr
->ex_next
) {
363 if (ex_ptr
->ex_fport
== ti
->ti_dport
&&
364 ti
->ti_dst
.s_addr
== ex_ptr
->ex_addr
.s_addr
) {
372 * Locate pcb for segment.
376 if (so
->so_fport
!= ti
->ti_dport
||
377 so
->so_lport
!= ti
->ti_sport
||
378 so
->so_laddr
.s_addr
!= ti
->ti_src
.s_addr
||
379 so
->so_faddr
.s_addr
!= ti
->ti_dst
.s_addr
) {
380 so
= solookup(&tcb
, ti
->ti_src
, ti
->ti_sport
,
381 ti
->ti_dst
, ti
->ti_dport
);
384 STAT(tcpstat
.tcps_socachemiss
++);
388 * If the state is CLOSED (i.e., TCB does not exist) then
389 * all data in the incoming segment is discarded.
390 * If the TCB exists but is in CLOSED state, it is embryonic,
391 * but should either do a listen or a connect soon.
393 * state == CLOSED means we've done socreate() but haven't
394 * attached it to a protocol yet...
396 * XXX If a TCB does not exist, and the TH_SYN flag is
397 * the only flag set, then create a session, mark it
398 * as if it was LISTENING, and continue...
401 if ((tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) != TH_SYN
)
404 if ((so
= socreate()) == NULL
)
406 if (tcp_attach(so
) < 0) {
407 free(so
); /* Not sofree (if it failed, it's not insqued) */
411 sbreserve(&so
->so_snd
, TCP_SNDSPACE
);
412 sbreserve(&so
->so_rcv
, TCP_RCVSPACE
);
414 /* tcp_last_so = so; */ /* XXX ? */
415 /* tp = sototcpcb(so); */
417 so
->so_laddr
= ti
->ti_src
;
418 so
->so_lport
= ti
->ti_sport
;
419 so
->so_faddr
= ti
->ti_dst
;
420 so
->so_fport
= ti
->ti_dport
;
422 if ((so
->so_iptos
= tcp_tos(so
)) == 0)
423 so
->so_iptos
= ((struct ip
*)ti
)->ip_tos
;
426 tp
->t_state
= TCPS_LISTEN
;
430 * If this is a still-connecting socket, this probably
431 * a retransmit of the SYN. Whether it's a retransmit SYN
432 * or something else, we nuke it.
434 if (so
->so_state
& SS_ISFCONNECTING
)
439 /* XXX Should never fail */
442 if (tp
->t_state
== TCPS_CLOSED
)
445 /* Unscale the window into a 32-bit value. */
446 /* if ((tiflags & TH_SYN) == 0)
447 * tiwin = ti->ti_win << tp->snd_scale;
453 * Segment received on connection.
454 * Reset idle time and keep-alive timer.
458 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEPINTVL
;
460 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_IDLE
;
463 * Process options if not in LISTEN state,
464 * else do it below (after getting remote address).
466 if (optp
&& tp
->t_state
!= TCPS_LISTEN
)
467 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
469 /* &ts_present, &ts_val, &ts_ecr); */
472 * Header prediction: check for the two common cases
473 * of a uni-directional data xfer. If the packet has
474 * no control flags, is in-sequence, the window didn't
475 * change and we're not retransmitting, it's a
476 * candidate. If the length is zero and the ack moved
477 * forward, we're the sender side of the xfer. Just
478 * free the data acked & wake any higher level process
479 * that was blocked waiting for space. If the length
480 * is non-zero and the ack didn't move, we're the
481 * receiver side. If we're getting packets in-order
482 * (the reassembly queue is empty), add the data to
483 * the socket buffer and note that we need a delayed ack.
485 * XXX Some of these tests are not needed
486 * eg: the tiwin == tp->snd_wnd prevents many more
487 * predictions.. with no *real* advantage..
489 if (tp
->t_state
== TCPS_ESTABLISHED
&&
490 (tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) == TH_ACK
&&
491 /* (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && */
492 ti
->ti_seq
== tp
->rcv_nxt
&&
493 tiwin
&& tiwin
== tp
->snd_wnd
&&
494 tp
->snd_nxt
== tp
->snd_max
) {
496 * If last ACK falls within this segment's sequence numbers,
497 * record the timestamp.
499 /* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
500 * SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
501 * tp->ts_recent_age = tcp_now;
502 * tp->ts_recent = ts_val;
505 if (ti
->ti_len
== 0) {
506 if (SEQ_GT(ti
->ti_ack
, tp
->snd_una
) &&
507 SEQ_LEQ(ti
->ti_ack
, tp
->snd_max
) &&
508 tp
->snd_cwnd
>= tp
->snd_wnd
) {
510 * this is a pure ack for outstanding data.
512 STAT(tcpstat
.tcps_predack
++);
514 * tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
517 SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
518 tcp_xmit_timer(tp
, tp
->t_rtt
);
519 acked
= ti
->ti_ack
- tp
->snd_una
;
520 STAT(tcpstat
.tcps_rcvackpack
++);
521 STAT(tcpstat
.tcps_rcvackbyte
+= acked
);
522 sbdrop(&so
->so_snd
, acked
);
523 tp
->snd_una
= ti
->ti_ack
;
527 * If all outstanding data are acked, stop
528 * retransmit timer, otherwise restart timer
529 * using current (possibly backed-off) value.
530 * If process is waiting for space,
531 * wakeup/selwakeup/signal. If data
532 * are ready to send, let tcp_output
533 * decide between more output or persist.
535 if (tp
->snd_una
== tp
->snd_max
)
536 tp
->t_timer
[TCPT_REXMT
] = 0;
537 else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
538 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
541 * There's room in so_snd, sowwakup will read()
542 * from the socket if we can
544 /* if (so->so_snd.sb_flags & SB_NOTIFY)
548 * This is called because sowwakeup might have
549 * put data into so_snd. Since we don't so sowwakeup,
550 * we don't need this.. XXX???
552 if (so
->so_snd
.sb_cc
)
553 (void) tcp_output(tp
);
557 } else if (ti
->ti_ack
== tp
->snd_una
&&
558 tcpfrag_list_empty(tp
) &&
559 ti
->ti_len
<= sbspace(&so
->so_rcv
)) {
561 * this is a pure, in-sequence data packet
562 * with nothing on the reassembly queue and
563 * we have enough buffer space to take it.
565 STAT(tcpstat
.tcps_preddat
++);
566 tp
->rcv_nxt
+= ti
->ti_len
;
567 STAT(tcpstat
.tcps_rcvpack
++);
568 STAT(tcpstat
.tcps_rcvbyte
+= ti
->ti_len
);
570 * Add data to socket buffer.
573 if (tcp_emu(so
,m
)) sbappend(so
, m
);
578 * XXX This is called when data arrives. Later, check
579 * if we can actually write() to the socket
580 * XXX Need to check? It's be NON_BLOCKING
585 * If this is a short packet, then ACK now - with Nagel
586 * congestion avoidance sender won't send more until
589 * It is better to not delay acks at all to maximize
590 * TCP throughput. See RFC 2581.
592 tp
->t_flags
|= TF_ACKNOW
;
596 } /* header prediction */
598 * Calculate amount of space in receive window,
599 * and then do TCP input processing.
600 * Receive window is amount of space in rcv queue,
601 * but not less than advertised window.
604 win
= sbspace(&so
->so_rcv
);
607 tp
->rcv_wnd
= max(win
, (int)(tp
->rcv_adv
- tp
->rcv_nxt
));
610 switch (tp
->t_state
) {
613 * If the state is LISTEN then ignore segment if it contains an RST.
614 * If the segment contains an ACK then it is bad and send a RST.
615 * If it does not contain a SYN then it is not interesting; drop it.
616 * Don't bother responding if the destination was a broadcast.
617 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
618 * tp->iss, and send a segment:
619 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
620 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
621 * Fill in remote peer address fields if not previously specified.
622 * Enter SYN_RECEIVED state, and process any other fields of this
623 * segment in this state.
627 if (tiflags
& TH_RST
)
629 if (tiflags
& TH_ACK
)
631 if ((tiflags
& TH_SYN
) == 0)
635 * This has way too many gotos...
636 * But a bit of spaghetti code never hurt anybody :)
640 * If this is destined for the control address, then flag to
641 * tcp_ctl once connected, otherwise connect
643 if ((so
->so_faddr
.s_addr
& vnetwork_mask
.s_addr
) ==
644 vnetwork_addr
.s_addr
) {
645 if (so
->so_faddr
.s_addr
!= vhost_addr
.s_addr
&&
646 so
->so_faddr
.s_addr
!= vnameserver_addr
.s_addr
) {
648 if(lastbyte
==CTL_CMD
|| lastbyte
==CTL_EXEC
) {
649 /* Command or exec adress */
650 so
->so_state
|= SS_CTL
;
654 /* May be an add exec */
655 for(ex_ptr
= exec_list
; ex_ptr
; ex_ptr
= ex_ptr
->ex_next
) {
656 if(ex_ptr
->ex_fport
== so
->so_fport
&&
657 so
->so_faddr
.s_addr
== ex_ptr
->ex_addr
.s_addr
) {
658 so
->so_state
|= SS_CTL
;
663 if(so
->so_state
& SS_CTL
) goto cont_input
;
665 /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
668 if (so
->so_emu
& EMU_NOCONNECT
) {
669 so
->so_emu
&= ~EMU_NOCONNECT
;
673 if((tcp_fconnect(so
) == -1) && (errno
!= EINPROGRESS
) && (errno
!= EWOULDBLOCK
)) {
674 u_char code
=ICMP_UNREACH_NET
;
675 DEBUG_MISC((dfd
," tcp fconnect errno = %d-%s\n",
676 errno
,strerror(errno
)));
677 if(errno
== ECONNREFUSED
) {
678 /* ACK the SYN, send RST to refuse the connection */
679 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+1, (tcp_seq
)0,
682 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
683 HTONL(ti
->ti_seq
); /* restore tcp header */
687 m
->m_data
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
688 m
->m_len
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
690 icmp_error(m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
696 * Haven't connected yet, save the current mbuf
698 * XXX Some OS's don't tell us whether the connect()
699 * succeeded or not. So we must time it out.
703 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
704 tp
->t_state
= TCPS_SYN_RECEIVED
;
710 * Check if the connect succeeded
712 if (so
->so_state
& SS_NOFDREF
) {
720 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
722 /* &ts_present, &ts_val, &ts_ecr); */
728 tcp_iss
+= TCP_ISSINCR
/2;
729 tp
->irs
= ti
->ti_seq
;
732 tp
->t_flags
|= TF_ACKNOW
;
733 tp
->t_state
= TCPS_SYN_RECEIVED
;
734 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
735 STAT(tcpstat
.tcps_accepts
++);
737 } /* case TCPS_LISTEN */
740 * If the state is SYN_SENT:
741 * if seg contains an ACK, but not for our SYN, drop the input.
742 * if seg contains a RST, then drop the connection.
743 * if seg does not contain SYN, then drop it.
744 * Otherwise this is an acceptable SYN segment
745 * initialize tp->rcv_nxt and tp->irs
746 * if seg contains ack then advance tp->snd_una
747 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
748 * arrange for segment to be acked (eventually)
749 * continue processing rest of data/controls, beginning with URG
752 if ((tiflags
& TH_ACK
) &&
753 (SEQ_LEQ(ti
->ti_ack
, tp
->iss
) ||
754 SEQ_GT(ti
->ti_ack
, tp
->snd_max
)))
757 if (tiflags
& TH_RST
) {
758 if (tiflags
& TH_ACK
)
759 tp
= tcp_drop(tp
,0); /* XXX Check t_softerror! */
763 if ((tiflags
& TH_SYN
) == 0)
765 if (tiflags
& TH_ACK
) {
766 tp
->snd_una
= ti
->ti_ack
;
767 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
768 tp
->snd_nxt
= tp
->snd_una
;
771 tp
->t_timer
[TCPT_REXMT
] = 0;
772 tp
->irs
= ti
->ti_seq
;
774 tp
->t_flags
|= TF_ACKNOW
;
775 if (tiflags
& TH_ACK
&& SEQ_GT(tp
->snd_una
, tp
->iss
)) {
776 STAT(tcpstat
.tcps_connects
++);
778 tp
->t_state
= TCPS_ESTABLISHED
;
780 /* Do window scaling on this connection? */
781 /* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
782 * (TF_RCVD_SCALE|TF_REQ_SCALE)) {
783 * tp->snd_scale = tp->requested_s_scale;
784 * tp->rcv_scale = tp->request_r_scale;
787 (void) tcp_reass(tp
, (struct tcpiphdr
*)0,
790 * if we didn't have to retransmit the SYN,
791 * use its rtt as our initial srtt & rtt var.
794 tcp_xmit_timer(tp
, tp
->t_rtt
);
796 tp
->t_state
= TCPS_SYN_RECEIVED
;
800 * Advance ti->ti_seq to correspond to first data byte.
801 * If data, trim to stay within window,
802 * dropping FIN if necessary.
805 if (ti
->ti_len
> tp
->rcv_wnd
) {
806 todrop
= ti
->ti_len
- tp
->rcv_wnd
;
808 ti
->ti_len
= tp
->rcv_wnd
;
810 STAT(tcpstat
.tcps_rcvpackafterwin
++);
811 STAT(tcpstat
.tcps_rcvbyteafterwin
+= todrop
);
813 tp
->snd_wl1
= ti
->ti_seq
- 1;
814 tp
->rcv_up
= ti
->ti_seq
;
816 } /* switch tp->t_state */
818 * States other than LISTEN or SYN_SENT.
819 * First check timestamp, if present.
820 * Then check that at least some bytes of segment are within
821 * receive window. If segment begins before rcv_nxt,
822 * drop leading data (and SYN); if nothing left, just ack.
824 * RFC 1323 PAWS: If we have a timestamp reply on this segment
825 * and it's less than ts_recent, drop it.
827 /* if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
828 * TSTMP_LT(ts_val, tp->ts_recent)) {
830 */ /* Check to see if ts_recent is over 24 days old. */
831 /* if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
833 * * Invalidate ts_recent. If this segment updates
834 * * ts_recent, the age will be reset later and ts_recent
835 * * will get a valid value. If it does not, setting
836 * * ts_recent to zero will at least satisfy the
837 * * requirement that zero be placed in the timestamp
838 * * echo reply when ts_recent isn't valid. The
839 * * age isn't reset until we get a valid ts_recent
840 * * because we don't want out-of-order segments to be
841 * * dropped when ts_recent is old.
843 /* tp->ts_recent = 0;
845 * tcpstat.tcps_rcvduppack++;
846 * tcpstat.tcps_rcvdupbyte += ti->ti_len;
847 * tcpstat.tcps_pawsdrop++;
853 todrop
= tp
->rcv_nxt
- ti
->ti_seq
;
855 if (tiflags
& TH_SYN
) {
865 * Following if statement from Stevens, vol. 2, p. 960.
867 if (todrop
> ti
->ti_len
868 || (todrop
== ti
->ti_len
&& (tiflags
& TH_FIN
) == 0)) {
870 * Any valid FIN must be to the left of the window.
871 * At this point the FIN must be a duplicate or out
872 * of sequence; drop it.
877 * Send an ACK to resynchronize and drop any data.
878 * But keep on processing for RST or ACK.
880 tp
->t_flags
|= TF_ACKNOW
;
882 STAT(tcpstat
.tcps_rcvduppack
++);
883 STAT(tcpstat
.tcps_rcvdupbyte
+= todrop
);
885 STAT(tcpstat
.tcps_rcvpartduppack
++);
886 STAT(tcpstat
.tcps_rcvpartdupbyte
+= todrop
);
889 ti
->ti_seq
+= todrop
;
890 ti
->ti_len
-= todrop
;
891 if (ti
->ti_urp
> todrop
)
892 ti
->ti_urp
-= todrop
;
899 * If new data are received on a connection after the
900 * user processes are gone, then RST the other end.
902 if ((so
->so_state
& SS_NOFDREF
) &&
903 tp
->t_state
> TCPS_CLOSE_WAIT
&& ti
->ti_len
) {
905 STAT(tcpstat
.tcps_rcvafterclose
++);
910 * If segment ends after window, drop trailing data
911 * (and PUSH and FIN); if nothing left, just ACK.
913 todrop
= (ti
->ti_seq
+ti
->ti_len
) - (tp
->rcv_nxt
+tp
->rcv_wnd
);
915 STAT(tcpstat
.tcps_rcvpackafterwin
++);
916 if (todrop
>= ti
->ti_len
) {
917 STAT(tcpstat
.tcps_rcvbyteafterwin
+= ti
->ti_len
);
919 * If a new connection request is received
920 * while in TIME_WAIT, drop the old connection
921 * and start over if the sequence numbers
922 * are above the previous ones.
924 if (tiflags
& TH_SYN
&&
925 tp
->t_state
== TCPS_TIME_WAIT
&&
926 SEQ_GT(ti
->ti_seq
, tp
->rcv_nxt
)) {
927 iss
= tp
->rcv_nxt
+ TCP_ISSINCR
;
932 * If window is closed can only take segments at
933 * window edge, and have to drop data and PUSH from
934 * incoming segments. Continue processing, but
935 * remember to ack. Otherwise, drop segment
938 if (tp
->rcv_wnd
== 0 && ti
->ti_seq
== tp
->rcv_nxt
) {
939 tp
->t_flags
|= TF_ACKNOW
;
940 STAT(tcpstat
.tcps_rcvwinprobe
++);
944 STAT(tcpstat
.tcps_rcvbyteafterwin
+= todrop
);
946 ti
->ti_len
-= todrop
;
947 tiflags
&= ~(TH_PUSH
|TH_FIN
);
951 * If last ACK falls within this segment's sequence numbers,
952 * record its timestamp.
954 /* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
955 * SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
956 * ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
957 * tp->ts_recent_age = tcp_now;
958 * tp->ts_recent = ts_val;
963 * If the RST bit is set examine the state:
964 * SYN_RECEIVED STATE:
965 * If passive open, return to LISTEN state.
966 * If active open, inform user that connection was refused.
967 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
968 * Inform user that connection was reset, and close tcb.
969 * CLOSING, LAST_ACK, TIME_WAIT STATES
972 if (tiflags
&TH_RST
) switch (tp
->t_state
) {
974 case TCPS_SYN_RECEIVED
:
975 /* so->so_error = ECONNREFUSED; */
978 case TCPS_ESTABLISHED
:
979 case TCPS_FIN_WAIT_1
:
980 case TCPS_FIN_WAIT_2
:
981 case TCPS_CLOSE_WAIT
:
982 /* so->so_error = ECONNRESET; */
984 tp
->t_state
= TCPS_CLOSED
;
985 STAT(tcpstat
.tcps_drops
++);
997 * If a SYN is in the window, then this is an
998 * error and we send an RST and drop the connection.
1000 if (tiflags
& TH_SYN
) {
1001 tp
= tcp_drop(tp
,0);
1006 * If the ACK bit is off we drop the segment and return.
1008 if ((tiflags
& TH_ACK
) == 0) goto drop
;
1013 switch (tp
->t_state
) {
1015 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1016 * ESTABLISHED state and continue processing, otherwise
1017 * send an RST. una<=ack<=max
1019 case TCPS_SYN_RECEIVED
:
1021 if (SEQ_GT(tp
->snd_una
, ti
->ti_ack
) ||
1022 SEQ_GT(ti
->ti_ack
, tp
->snd_max
))
1024 STAT(tcpstat
.tcps_connects
++);
1025 tp
->t_state
= TCPS_ESTABLISHED
;
1027 * The sent SYN is ack'ed with our sequence number +1
1028 * The first data byte already in the buffer will get
1029 * lost if no correction is made. This is only needed for
1030 * SS_CTL since the buffer is empty otherwise.
1031 * tp->snd_una++; or:
1033 tp
->snd_una
=ti
->ti_ack
;
1034 if (so
->so_state
& SS_CTL
) {
1035 /* So tcp_ctl reports the right state */
1039 so
->so_state
&= ~SS_CTL
; /* success XXX */
1040 } else if (ret
== 2) {
1041 so
->so_state
= SS_NOFDREF
; /* CTL_CMD */
1044 tp
->t_state
= TCPS_FIN_WAIT_1
;
1050 /* Do window scaling? */
1051 /* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1052 * (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1053 * tp->snd_scale = tp->requested_s_scale;
1054 * tp->rcv_scale = tp->request_r_scale;
1057 (void) tcp_reass(tp
, (struct tcpiphdr
*)0, (struct mbuf
*)0);
1058 tp
->snd_wl1
= ti
->ti_seq
- 1;
1059 /* Avoid ack processing; snd_una==ti_ack => dup ack */
1064 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1065 * ACKs. If the ack is in the range
1066 * tp->snd_una < ti->ti_ack <= tp->snd_max
1067 * then advance tp->snd_una to ti->ti_ack and drop
1068 * data from the retransmission queue. If this ACK reflects
1069 * more up to date window information we update our window information.
1071 case TCPS_ESTABLISHED
:
1072 case TCPS_FIN_WAIT_1
:
1073 case TCPS_FIN_WAIT_2
:
1074 case TCPS_CLOSE_WAIT
:
1077 case TCPS_TIME_WAIT
:
1079 if (SEQ_LEQ(ti
->ti_ack
, tp
->snd_una
)) {
1080 if (ti
->ti_len
== 0 && tiwin
== tp
->snd_wnd
) {
1081 STAT(tcpstat
.tcps_rcvdupack
++);
1082 DEBUG_MISC((dfd
," dup ack m = %lx so = %lx \n",
1083 (long )m
, (long )so
));
1085 * If we have outstanding data (other than
1086 * a window probe), this is a completely
1087 * duplicate ack (ie, window info didn't
1088 * change), the ack is the biggest we've
1089 * seen and we've seen exactly our rexmt
1090 * threshold of them, assume a packet
1091 * has been dropped and retransmit it.
1092 * Kludge snd_nxt & the congestion
1093 * window so we send only this one
1096 * We know we're losing at the current
1097 * window size so do congestion avoidance
1098 * (set ssthresh to half the current window
1099 * and pull our congestion window back to
1100 * the new ssthresh).
1102 * Dup acks mean that packets have left the
1103 * network (they're now cached at the receiver)
1104 * so bump cwnd by the amount in the receiver
1105 * to keep a constant cwnd packets in the
1108 if (tp
->t_timer
[TCPT_REXMT
] == 0 ||
1109 ti
->ti_ack
!= tp
->snd_una
)
1111 else if (++tp
->t_dupacks
== TCPREXMTTHRESH
) {
1112 tcp_seq onxt
= tp
->snd_nxt
;
1114 min(tp
->snd_wnd
, tp
->snd_cwnd
) / 2 /
1119 tp
->snd_ssthresh
= win
* tp
->t_maxseg
;
1120 tp
->t_timer
[TCPT_REXMT
] = 0;
1122 tp
->snd_nxt
= ti
->ti_ack
;
1123 tp
->snd_cwnd
= tp
->t_maxseg
;
1124 (void) tcp_output(tp
);
1125 tp
->snd_cwnd
= tp
->snd_ssthresh
+
1126 tp
->t_maxseg
* tp
->t_dupacks
;
1127 if (SEQ_GT(onxt
, tp
->snd_nxt
))
1130 } else if (tp
->t_dupacks
> TCPREXMTTHRESH
) {
1131 tp
->snd_cwnd
+= tp
->t_maxseg
;
1132 (void) tcp_output(tp
);
1141 * If the congestion window was inflated to account
1142 * for the other side's cached packets, retract it.
1144 if (tp
->t_dupacks
> TCPREXMTTHRESH
&&
1145 tp
->snd_cwnd
> tp
->snd_ssthresh
)
1146 tp
->snd_cwnd
= tp
->snd_ssthresh
;
1148 if (SEQ_GT(ti
->ti_ack
, tp
->snd_max
)) {
1149 STAT(tcpstat
.tcps_rcvacktoomuch
++);
1152 acked
= ti
->ti_ack
- tp
->snd_una
;
1153 STAT(tcpstat
.tcps_rcvackpack
++);
1154 STAT(tcpstat
.tcps_rcvackbyte
+= acked
);
1157 * If we have a timestamp reply, update smoothed
1158 * round trip time. If no timestamp is present but
1159 * transmit timer is running and timed sequence
1160 * number was acked, update smoothed round trip time.
1161 * Since we now have an rtt measurement, cancel the
1162 * timer backoff (cf., Phil Karn's retransmit alg.).
1163 * Recompute the initial retransmit timer.
1166 * tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1169 if (tp
->t_rtt
&& SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
1170 tcp_xmit_timer(tp
,tp
->t_rtt
);
1173 * If all outstanding data is acked, stop retransmit
1174 * timer and remember to restart (more output or persist).
1175 * If there is more data to be acked, restart retransmit
1176 * timer, using current (possibly backed-off) value.
1178 if (ti
->ti_ack
== tp
->snd_max
) {
1179 tp
->t_timer
[TCPT_REXMT
] = 0;
1181 } else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
1182 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
1184 * When new data is acked, open the congestion window.
1185 * If the window gives us less than ssthresh packets
1186 * in flight, open exponentially (maxseg per packet).
1187 * Otherwise open linearly: maxseg per window
1188 * (maxseg^2 / cwnd per packet).
1191 register u_int cw
= tp
->snd_cwnd
;
1192 register u_int incr
= tp
->t_maxseg
;
1194 if (cw
> tp
->snd_ssthresh
)
1195 incr
= incr
* incr
/ cw
;
1196 tp
->snd_cwnd
= min(cw
+ incr
, TCP_MAXWIN
<<tp
->snd_scale
);
1198 if (acked
> so
->so_snd
.sb_cc
) {
1199 tp
->snd_wnd
-= so
->so_snd
.sb_cc
;
1200 sbdrop(&so
->so_snd
, (int )so
->so_snd
.sb_cc
);
1203 sbdrop(&so
->so_snd
, acked
);
1204 tp
->snd_wnd
-= acked
;
1208 * XXX sowwakup is called when data is acked and there's room for
1209 * for more data... it should read() the socket
1211 /* if (so->so_snd.sb_flags & SB_NOTIFY)
1214 tp
->snd_una
= ti
->ti_ack
;
1215 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
1216 tp
->snd_nxt
= tp
->snd_una
;
1218 switch (tp
->t_state
) {
1221 * In FIN_WAIT_1 STATE in addition to the processing
1222 * for the ESTABLISHED state if our FIN is now acknowledged
1223 * then enter FIN_WAIT_2.
1225 case TCPS_FIN_WAIT_1
:
1226 if (ourfinisacked
) {
1228 * If we can't receive any more
1229 * data, then closing user can proceed.
1230 * Starting the timer is contrary to the
1231 * specification, but if we don't get a FIN
1232 * we'll hang forever.
1234 if (so
->so_state
& SS_FCANTRCVMORE
) {
1235 soisfdisconnected(so
);
1236 tp
->t_timer
[TCPT_2MSL
] = TCP_MAXIDLE
;
1238 tp
->t_state
= TCPS_FIN_WAIT_2
;
1243 * In CLOSING STATE in addition to the processing for
1244 * the ESTABLISHED state if the ACK acknowledges our FIN
1245 * then enter the TIME-WAIT state, otherwise ignore
1249 if (ourfinisacked
) {
1250 tp
->t_state
= TCPS_TIME_WAIT
;
1251 tcp_canceltimers(tp
);
1252 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1253 soisfdisconnected(so
);
1258 * In LAST_ACK, we may still be waiting for data to drain
1259 * and/or to be acked, as well as for the ack of our FIN.
1260 * If our FIN is now acknowledged, delete the TCB,
1261 * enter the closed state and return.
1264 if (ourfinisacked
) {
1271 * In TIME_WAIT state the only thing that should arrive
1272 * is a retransmission of the remote FIN. Acknowledge
1273 * it and restart the finack timer.
1275 case TCPS_TIME_WAIT
:
1276 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1279 } /* switch(tp->t_state) */
1283 * Update window information.
1284 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1286 if ((tiflags
& TH_ACK
) &&
1287 (SEQ_LT(tp
->snd_wl1
, ti
->ti_seq
) ||
1288 (tp
->snd_wl1
== ti
->ti_seq
&& (SEQ_LT(tp
->snd_wl2
, ti
->ti_ack
) ||
1289 (tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
))))) {
1290 /* keep track of pure window updates */
1291 if (ti
->ti_len
== 0 &&
1292 tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
)
1293 STAT(tcpstat
.tcps_rcvwinupd
++);
1294 tp
->snd_wnd
= tiwin
;
1295 tp
->snd_wl1
= ti
->ti_seq
;
1296 tp
->snd_wl2
= ti
->ti_ack
;
1297 if (tp
->snd_wnd
> tp
->max_sndwnd
)
1298 tp
->max_sndwnd
= tp
->snd_wnd
;
1303 * Process segments with URG.
1305 if ((tiflags
& TH_URG
) && ti
->ti_urp
&&
1306 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1308 * This is a kludge, but if we receive and accept
1309 * random urgent pointers, we'll crash in
1310 * soreceive. It's hard to imagine someone
1311 * actually wanting to send this much urgent data.
1313 if (ti
->ti_urp
+ so
->so_rcv
.sb_cc
> so
->so_rcv
.sb_datalen
) {
1319 * If this segment advances the known urgent pointer,
1320 * then mark the data stream. This should not happen
1321 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1322 * a FIN has been received from the remote side.
1323 * In these states we ignore the URG.
1325 * According to RFC961 (Assigned Protocols),
1326 * the urgent pointer points to the last octet
1327 * of urgent data. We continue, however,
1328 * to consider it to indicate the first octet
1329 * of data past the urgent section as the original
1330 * spec states (in one of two places).
1332 if (SEQ_GT(ti
->ti_seq
+ti
->ti_urp
, tp
->rcv_up
)) {
1333 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1334 so
->so_urgc
= so
->so_rcv
.sb_cc
+
1335 (tp
->rcv_up
- tp
->rcv_nxt
); /* -1; */
1336 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1341 * If no out of band data is expected,
1342 * pull receive urgent pointer along
1343 * with the receive window.
1345 if (SEQ_GT(tp
->rcv_nxt
, tp
->rcv_up
))
1346 tp
->rcv_up
= tp
->rcv_nxt
;
1350 * Process the segment text, merging it into the TCP sequencing queue,
1351 * and arranging for acknowledgment of receipt if necessary.
1352 * This process logically involves adjusting tp->rcv_wnd as data
1353 * is presented to the user (this happens in tcp_usrreq.c,
1354 * case PRU_RCVD). If a FIN has already been received on this
1355 * connection then we just ignore the text.
1357 if ((ti
->ti_len
|| (tiflags
&TH_FIN
)) &&
1358 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1359 TCP_REASS(tp
, ti
, m
, so
, tiflags
);
1361 * Note the amount of data that peer has sent into
1362 * our window, in order to estimate the sender's
1365 len
= so
->so_rcv
.sb_datalen
- (tp
->rcv_adv
- tp
->rcv_nxt
);
1372 * If FIN is received ACK the FIN and let the user know
1373 * that the connection is closing.
1375 if (tiflags
& TH_FIN
) {
1376 if (TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1378 * If we receive a FIN we can't send more data,
1380 * Shutdown the socket if there is no rx data in the
1382 * soread() is called on completion of shutdown() and
1383 * will got to TCPS_LAST_ACK, and use tcp_output()
1386 /* sofcantrcvmore(so); */
1389 tp
->t_flags
|= TF_ACKNOW
;
1392 switch (tp
->t_state
) {
1395 * In SYN_RECEIVED and ESTABLISHED STATES
1396 * enter the CLOSE_WAIT state.
1398 case TCPS_SYN_RECEIVED
:
1399 case TCPS_ESTABLISHED
:
1400 if(so
->so_emu
== EMU_CTL
) /* no shutdown on socket */
1401 tp
->t_state
= TCPS_LAST_ACK
;
1403 tp
->t_state
= TCPS_CLOSE_WAIT
;
1407 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1408 * enter the CLOSING state.
1410 case TCPS_FIN_WAIT_1
:
1411 tp
->t_state
= TCPS_CLOSING
;
1415 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1416 * starting the time-wait timer, turning off the other
1419 case TCPS_FIN_WAIT_2
:
1420 tp
->t_state
= TCPS_TIME_WAIT
;
1421 tcp_canceltimers(tp
);
1422 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1423 soisfdisconnected(so
);
1427 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1429 case TCPS_TIME_WAIT
:
1430 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1436 * If this is a small packet, then ACK now - with Nagel
1437 * congestion avoidance sender won't send more until
1442 /* if (ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg) {
1444 /* if ((ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg &&
1445 * (so->so_iptos & IPTOS_LOWDELAY) == 0) ||
1446 * ((so->so_iptos & IPTOS_LOWDELAY) &&
1447 * ((struct tcpiphdr_2 *)ti)->first_char == (char)27)) {
1449 if (ti
->ti_len
&& (unsigned)ti
->ti_len
<= 5 &&
1450 ((struct tcpiphdr_2
*)ti
)->first_char
== (char)27) {
1451 tp
->t_flags
|= TF_ACKNOW
;
1455 * Return any desired output.
1457 if (needoutput
|| (tp
->t_flags
& TF_ACKNOW
)) {
1458 (void) tcp_output(tp
);
1464 * Generate an ACK dropping incoming segment if it occupies
1465 * sequence space, where the ACK reflects our state.
1467 if (tiflags
& TH_RST
)
1470 tp
->t_flags
|= TF_ACKNOW
;
1471 (void) tcp_output(tp
);
1475 /* reuses m if m!=NULL, m_free() unnecessary */
1476 if (tiflags
& TH_ACK
)
1477 tcp_respond(tp
, ti
, m
, (tcp_seq
)0, ti
->ti_ack
, TH_RST
);
1479 if (tiflags
& TH_SYN
) ti
->ti_len
++;
1480 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+ti
->ti_len
, (tcp_seq
)0,
1488 * Drop space held by incoming segment and return.
1495 /* , ts_present, ts_val, ts_ecr) */
1497 * u_int32_t *ts_val, *ts_ecr;
1500 tcp_dooptions(struct tcpcb
*tp
, u_char
*cp
, int cnt
, struct tcpiphdr
*ti
)
1505 DEBUG_CALL("tcp_dooptions");
1506 DEBUG_ARGS((dfd
," tp = %lx cnt=%i \n", (long )tp
, cnt
));
1508 for (; cnt
> 0; cnt
-= optlen
, cp
+= optlen
) {
1510 if (opt
== TCPOPT_EOL
)
1512 if (opt
== TCPOPT_NOP
)
1525 if (optlen
!= TCPOLEN_MAXSEG
)
1527 if (!(ti
->ti_flags
& TH_SYN
))
1529 memcpy((char *) &mss
, (char *) cp
+ 2, sizeof(mss
));
1531 (void) tcp_mss(tp
, mss
); /* sets t_maxseg */
1534 /* case TCPOPT_WINDOW:
1535 * if (optlen != TCPOLEN_WINDOW)
1537 * if (!(ti->ti_flags & TH_SYN))
1539 * tp->t_flags |= TF_RCVD_SCALE;
1540 * tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1543 /* case TCPOPT_TIMESTAMP:
1544 * if (optlen != TCPOLEN_TIMESTAMP)
1547 * memcpy((char *) ts_val, (char *)cp + 2, sizeof(*ts_val));
1549 * memcpy((char *) ts_ecr, (char *)cp + 6, sizeof(*ts_ecr));
1553 * * A timestamp received in a SYN makes
1554 * * it ok to send timestamp requests and replies.
1556 /* if (ti->ti_flags & TH_SYN) {
1557 * tp->t_flags |= TF_RCVD_TSTMP;
1558 * tp->ts_recent = *ts_val;
1559 * tp->ts_recent_age = tcp_now;
1568 * Pull out of band byte out of a segment so
1569 * it doesn't appear in the user's data queue.
1570 * It is still reflected in the segment length for
1571 * sequencing purposes.
1577 tcp_pulloutofband(so
, ti
, m
)
1579 struct tcpiphdr
*ti
;
1580 register struct mbuf
*m
;
1582 int cnt
= ti
->ti_urp
- 1;
1585 if (m
->m_len
> cnt
) {
1586 char *cp
= mtod(m
, caddr_t
) + cnt
;
1587 struct tcpcb
*tp
= sototcpcb(so
);
1590 tp
->t_oobflags
|= TCPOOB_HAVEDATA
;
1591 memcpy(sp
, cp
+1, (unsigned)(m
->m_len
- cnt
- 1));
1596 m
= m
->m_next
; /* XXX WRONG! Fix it! */
1600 panic("tcp_pulloutofband");
1606 * Collect new round-trip time estimate
1607 * and update averages and current timeout.
1611 tcp_xmit_timer(register struct tcpcb
*tp
, int rtt
)
1613 register short delta
;
1615 DEBUG_CALL("tcp_xmit_timer");
1616 DEBUG_ARG("tp = %lx", (long)tp
);
1617 DEBUG_ARG("rtt = %d", rtt
);
1619 STAT(tcpstat
.tcps_rttupdated
++);
1620 if (tp
->t_srtt
!= 0) {
1622 * srtt is stored as fixed point with 3 bits after the
1623 * binary point (i.e., scaled by 8). The following magic
1624 * is equivalent to the smoothing algorithm in rfc793 with
1625 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1626 * point). Adjust rtt to origin 0.
1628 delta
= rtt
- 1 - (tp
->t_srtt
>> TCP_RTT_SHIFT
);
1629 if ((tp
->t_srtt
+= delta
) <= 0)
1632 * We accumulate a smoothed rtt variance (actually, a
1633 * smoothed mean difference), then set the retransmit
1634 * timer to smoothed rtt + 4 times the smoothed variance.
1635 * rttvar is stored as fixed point with 2 bits after the
1636 * binary point (scaled by 4). The following is
1637 * equivalent to rfc793 smoothing with an alpha of .75
1638 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
1639 * rfc793's wired-in beta.
1643 delta
-= (tp
->t_rttvar
>> TCP_RTTVAR_SHIFT
);
1644 if ((tp
->t_rttvar
+= delta
) <= 0)
1648 * No rtt measurement yet - use the unsmoothed rtt.
1649 * Set the variance to half the rtt (so our first
1650 * retransmit happens at 3*rtt).
1652 tp
->t_srtt
= rtt
<< TCP_RTT_SHIFT
;
1653 tp
->t_rttvar
= rtt
<< (TCP_RTTVAR_SHIFT
- 1);
1659 * the retransmit should happen at rtt + 4 * rttvar.
1660 * Because of the way we do the smoothing, srtt and rttvar
1661 * will each average +1/2 tick of bias. When we compute
1662 * the retransmit timer, we want 1/2 tick of rounding and
1663 * 1 extra tick because of +-1/2 tick uncertainty in the
1664 * firing of the timer. The bias will give us exactly the
1665 * 1.5 tick we need. But, because the bias is
1666 * statistical, we have to test that we don't drop below
1667 * the minimum feasible timer (which is 2 ticks).
1669 TCPT_RANGESET(tp
->t_rxtcur
, TCP_REXMTVAL(tp
),
1670 (short)tp
->t_rttmin
, TCPTV_REXMTMAX
); /* XXX */
1673 * We received an ack for a packet that wasn't retransmitted;
1674 * it is probably safe to discard any error indications we've
1675 * received recently. This isn't quite right, but close enough
1676 * for now (a route might have failed after we sent a segment,
1677 * and the return path might not be symmetrical).
1679 tp
->t_softerror
= 0;
1683 * Determine a reasonable value for maxseg size.
1684 * If the route is known, check route for mtu.
1685 * If none, use an mss that can be handled on the outgoing
1686 * interface without forcing IP to fragment; if bigger than
1687 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1688 * to utilize large mbufs. If no route is found, route has no mtu,
1689 * or the destination isn't local, use a default, hopefully conservative
1690 * size (usually 512 or the default IP max size, but no more than the mtu
1691 * of the interface), as we can't discover anything about intervening
1692 * gateways or networks. We also initialize the congestion/slow start
1693 * window to be a single segment if the destination isn't local.
1694 * While looking at the routing entry, we also initialize other path-dependent
1695 * parameters from pre-set or cached values in the routing entry.
1699 tcp_mss(struct tcpcb
*tp
, u_int offer
)
1701 struct socket
*so
= tp
->t_socket
;
1704 DEBUG_CALL("tcp_mss");
1705 DEBUG_ARG("tp = %lx", (long)tp
);
1706 DEBUG_ARG("offer = %d", offer
);
1708 mss
= min(IF_MTU
, IF_MRU
) - sizeof(struct tcpiphdr
);
1710 mss
= min(mss
, offer
);
1712 if (mss
< tp
->t_maxseg
|| offer
!= 0)
1717 sbreserve(&so
->so_snd
, TCP_SNDSPACE
+ ((TCP_SNDSPACE
% mss
) ?
1718 (mss
- (TCP_SNDSPACE
% mss
)) :
1720 sbreserve(&so
->so_rcv
, TCP_RCVSPACE
+ ((TCP_RCVSPACE
% mss
) ?
1721 (mss
- (TCP_RCVSPACE
% mss
)) :
1724 DEBUG_MISC((dfd
, " returning mss = %d\n", mss
));