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.
44 #define TCPREXMTTHRESH 3
46 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
48 /* for modulo comparisons of timestamps */
49 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0)
50 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0)
53 * Insert segment ti into reassembly queue of tcp with
54 * control block tp. Return TH_FIN if reassembly now includes
55 * a segment with FIN. The macro form does the common case inline
56 * (segment is the next to be received on an established connection,
57 * and the queue is empty), avoiding linkage into and removal
58 * from the queue and repetition of various conversions.
59 * Set DELACK for segments received in order, but ack immediately
60 * when segments are out of order (so fast retransmit can work).
63 #define TCP_REASS(tp, ti, m, so, flags) {\
64 if ((ti)->ti_seq == (tp)->rcv_nxt && \
65 tcpfrag_list_empty(tp) && \
66 (tp)->t_state == TCPS_ESTABLISHED) {\
67 if (ti->ti_flags & TH_PUSH) \
68 tp->t_flags |= TF_ACKNOW; \
70 tp->t_flags |= TF_DELACK; \
71 (tp)->rcv_nxt += (ti)->ti_len; \
72 flags = (ti)->ti_flags & TH_FIN; \
74 if (tcp_emu((so),(m))) sbappend((so), (m)); \
76 sbappend((so), (m)); \
78 (flags) = tcp_reass((tp), (ti), (m)); \
79 tp->t_flags |= TF_ACKNOW; \
83 #define TCP_REASS(tp, ti, m, so, flags) { \
84 if ((ti)->ti_seq == (tp)->rcv_nxt && \
85 tcpfrag_list_empty(tp) && \
86 (tp)->t_state == TCPS_ESTABLISHED) { \
87 tp->t_flags |= TF_DELACK; \
88 (tp)->rcv_nxt += (ti)->ti_len; \
89 flags = (ti)->ti_flags & TH_FIN; \
91 if (tcp_emu((so),(m))) sbappend(so, (m)); \
93 sbappend((so), (m)); \
95 (flags) = tcp_reass((tp), (ti), (m)); \
96 tp->t_flags |= TF_ACKNOW; \
100 static void tcp_dooptions(struct tcpcb
*tp
, u_char
*cp
, int cnt
,
101 struct tcpiphdr
*ti
);
102 static void tcp_xmit_timer(register struct tcpcb
*tp
, int rtt
);
105 tcp_reass(register struct tcpcb
*tp
, register struct tcpiphdr
*ti
,
108 register struct tcpiphdr
*q
;
109 struct socket
*so
= tp
->t_socket
;
113 * Call with ti==NULL after become established to
114 * force pre-ESTABLISHED data up to user socket.
120 * Find a segment which begins after this one does.
122 for (q
= tcpfrag_list_first(tp
); !tcpfrag_list_end(q
, tp
);
123 q
= tcpiphdr_next(q
))
124 if (SEQ_GT(q
->ti_seq
, ti
->ti_seq
))
128 * If there is a preceding segment, it may provide some of
129 * our data already. If so, drop the data from the incoming
130 * segment. If it provides all of our data, drop us.
132 if (!tcpfrag_list_end(tcpiphdr_prev(q
), tp
)) {
134 q
= tcpiphdr_prev(q
);
135 /* conversion to int (in i) handles seq wraparound */
136 i
= q
->ti_seq
+ q
->ti_len
- ti
->ti_seq
;
138 if (i
>= ti
->ti_len
) {
141 * Try to present any queued data
142 * at the left window edge to the user.
143 * This is needed after the 3-WHS
146 goto present
; /* ??? */
152 q
= tcpiphdr_next(q
);
157 * While we overlap succeeding segments trim them or,
158 * if they are completely covered, dequeue them.
160 while (!tcpfrag_list_end(q
, tp
)) {
161 register int i
= (ti
->ti_seq
+ ti
->ti_len
) - q
->ti_seq
;
167 m_adj(q
->ti_mbuf
, i
);
170 q
= tcpiphdr_next(q
);
171 m
= tcpiphdr_prev(q
)->ti_mbuf
;
172 remque(tcpiphdr2qlink(tcpiphdr_prev(q
)));
177 * Stick new segment in its place.
179 insque(tcpiphdr2qlink(ti
), tcpiphdr2qlink(tcpiphdr_prev(q
)));
183 * Present data to user, advancing rcv_nxt through
184 * completed sequence space.
186 if (!TCPS_HAVEESTABLISHED(tp
->t_state
))
188 ti
= tcpfrag_list_first(tp
);
189 if (tcpfrag_list_end(ti
, tp
) || ti
->ti_seq
!= tp
->rcv_nxt
)
191 if (tp
->t_state
== TCPS_SYN_RECEIVED
&& ti
->ti_len
)
194 tp
->rcv_nxt
+= ti
->ti_len
;
195 flags
= ti
->ti_flags
& TH_FIN
;
196 remque(tcpiphdr2qlink(ti
));
198 ti
= tcpiphdr_next(ti
);
199 if (so
->so_state
& SS_FCANTSENDMORE
)
203 if (tcp_emu(so
,m
)) sbappend(so
, m
);
207 } while (ti
!= (struct tcpiphdr
*)tp
&& ti
->ti_seq
== tp
->rcv_nxt
);
212 * TCP input routine, follows pages 65-76 of the
213 * protocol specification dated September, 1981 very closely.
216 tcp_input(struct mbuf
*m
, int iphlen
, struct socket
*inso
)
218 struct ip save_ip
, *ip
;
219 register struct tcpiphdr
*ti
;
223 register struct tcpcb
*tp
= NULL
;
224 register int tiflags
;
225 struct socket
*so
= NULL
;
226 int todrop
, acked
, ourfinisacked
, needoutput
= 0;
230 struct ex_list
*ex_ptr
;
233 DEBUG_CALL("tcp_input");
234 DEBUG_ARGS((dfd
, " m = %8lx iphlen = %2d inso = %lx\n",
235 (long )m
, iphlen
, (long )inso
));
238 * If called with m == 0, then we're continuing the connect
244 /* Re-set a few variables */
250 tiflags
= ti
->ti_flags
;
257 * Get IP and TCP header together in first mbuf.
258 * Note: IP leaves IP header in first mbuf.
260 ti
= mtod(m
, struct tcpiphdr
*);
261 if (iphlen
> sizeof(struct ip
)) {
262 ip_stripoptions(m
, (struct mbuf
*)0);
263 iphlen
=sizeof(struct ip
);
265 /* XXX Check if too short */
269 * Save a copy of the IP header in case we want restore it
270 * for sending an ICMP error message in response.
272 ip
=mtod(m
, struct ip
*);
274 save_ip
.ip_len
+= iphlen
;
277 * Checksum extended TCP header and data.
279 tlen
= ((struct ip
*)ti
)->ip_len
;
280 tcpiphdr2qlink(ti
)->next
= tcpiphdr2qlink(ti
)->prev
= NULL
;
281 memset(&ti
->ti_i
.ih_mbuf
, 0 , sizeof(struct mbuf_ptr
));
283 ti
->ti_len
= htons((uint16_t)tlen
);
284 len
= sizeof(struct ip
) + tlen
;
290 * Check that TCP offset makes sense,
291 * pull out TCP options and adjust length. XXX
293 off
= ti
->ti_off
<< 2;
294 if (off
< sizeof (struct tcphdr
) || off
> tlen
) {
299 if (off
> sizeof (struct tcphdr
)) {
300 optlen
= off
- sizeof (struct tcphdr
);
301 optp
= mtod(m
, caddr_t
) + sizeof (struct tcpiphdr
);
303 tiflags
= ti
->ti_flags
;
306 * Convert TCP protocol specific fields to host format.
314 * Drop TCP, IP headers and TCP options.
316 m
->m_data
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
317 m
->m_len
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
319 if (slirp
->restricted
) {
320 for (ex_ptr
= slirp
->exec_list
; ex_ptr
; ex_ptr
= ex_ptr
->ex_next
) {
321 if (ex_ptr
->ex_fport
== ti
->ti_dport
&&
322 ti
->ti_dst
.s_addr
== ex_ptr
->ex_addr
.s_addr
) {
330 * Locate pcb for segment.
333 so
= slirp
->tcp_last_so
;
334 if (so
->so_fport
!= ti
->ti_dport
||
335 so
->so_lport
!= ti
->ti_sport
||
336 so
->so_laddr
.s_addr
!= ti
->ti_src
.s_addr
||
337 so
->so_faddr
.s_addr
!= ti
->ti_dst
.s_addr
) {
338 so
= solookup(&slirp
->tcb
, ti
->ti_src
, ti
->ti_sport
,
339 ti
->ti_dst
, ti
->ti_dport
);
341 slirp
->tcp_last_so
= so
;
345 * If the state is CLOSED (i.e., TCB does not exist) then
346 * all data in the incoming segment is discarded.
347 * If the TCB exists but is in CLOSED state, it is embryonic,
348 * but should either do a listen or a connect soon.
350 * state == CLOSED means we've done socreate() but haven't
351 * attached it to a protocol yet...
353 * XXX If a TCB does not exist, and the TH_SYN flag is
354 * the only flag set, then create a session, mark it
355 * as if it was LISTENING, and continue...
358 if ((tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) != TH_SYN
)
361 if ((so
= socreate(slirp
)) == NULL
)
363 if (tcp_attach(so
) < 0) {
364 free(so
); /* Not sofree (if it failed, it's not insqued) */
368 sbreserve(&so
->so_snd
, TCP_SNDSPACE
);
369 sbreserve(&so
->so_rcv
, TCP_RCVSPACE
);
371 so
->so_laddr
= ti
->ti_src
;
372 so
->so_lport
= ti
->ti_sport
;
373 so
->so_faddr
= ti
->ti_dst
;
374 so
->so_fport
= ti
->ti_dport
;
376 if ((so
->so_iptos
= tcp_tos(so
)) == 0)
377 so
->so_iptos
= ((struct ip
*)ti
)->ip_tos
;
380 tp
->t_state
= TCPS_LISTEN
;
384 * If this is a still-connecting socket, this probably
385 * a retransmit of the SYN. Whether it's a retransmit SYN
386 * or something else, we nuke it.
388 if (so
->so_state
& SS_ISFCONNECTING
)
393 /* XXX Should never fail */
396 if (tp
->t_state
== TCPS_CLOSED
)
402 * Segment received on connection.
403 * Reset idle time and keep-alive timer.
407 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEPINTVL
;
409 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_IDLE
;
412 * Process options if not in LISTEN state,
413 * else do it below (after getting remote address).
415 if (optp
&& tp
->t_state
!= TCPS_LISTEN
)
416 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
419 * Header prediction: check for the two common cases
420 * of a uni-directional data xfer. If the packet has
421 * no control flags, is in-sequence, the window didn't
422 * change and we're not retransmitting, it's a
423 * candidate. If the length is zero and the ack moved
424 * forward, we're the sender side of the xfer. Just
425 * free the data acked & wake any higher level process
426 * that was blocked waiting for space. If the length
427 * is non-zero and the ack didn't move, we're the
428 * receiver side. If we're getting packets in-order
429 * (the reassembly queue is empty), add the data to
430 * the socket buffer and note that we need a delayed ack.
432 * XXX Some of these tests are not needed
433 * eg: the tiwin == tp->snd_wnd prevents many more
434 * predictions.. with no *real* advantage..
436 if (tp
->t_state
== TCPS_ESTABLISHED
&&
437 (tiflags
& (TH_SYN
|TH_FIN
|TH_RST
|TH_URG
|TH_ACK
)) == TH_ACK
&&
438 ti
->ti_seq
== tp
->rcv_nxt
&&
439 tiwin
&& tiwin
== tp
->snd_wnd
&&
440 tp
->snd_nxt
== tp
->snd_max
) {
441 if (ti
->ti_len
== 0) {
442 if (SEQ_GT(ti
->ti_ack
, tp
->snd_una
) &&
443 SEQ_LEQ(ti
->ti_ack
, tp
->snd_max
) &&
444 tp
->snd_cwnd
>= tp
->snd_wnd
) {
446 * this is a pure ack for outstanding data.
449 SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
450 tcp_xmit_timer(tp
, tp
->t_rtt
);
451 acked
= ti
->ti_ack
- tp
->snd_una
;
452 sbdrop(&so
->so_snd
, acked
);
453 tp
->snd_una
= ti
->ti_ack
;
457 * If all outstanding data are acked, stop
458 * retransmit timer, otherwise restart timer
459 * using current (possibly backed-off) value.
460 * If process is waiting for space,
461 * wakeup/selwakeup/signal. If data
462 * are ready to send, let tcp_output
463 * decide between more output or persist.
465 if (tp
->snd_una
== tp
->snd_max
)
466 tp
->t_timer
[TCPT_REXMT
] = 0;
467 else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
468 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
471 * This is called because sowwakeup might have
472 * put data into so_snd. Since we don't so sowwakeup,
473 * we don't need this.. XXX???
475 if (so
->so_snd
.sb_cc
)
476 (void) tcp_output(tp
);
480 } else if (ti
->ti_ack
== tp
->snd_una
&&
481 tcpfrag_list_empty(tp
) &&
482 ti
->ti_len
<= sbspace(&so
->so_rcv
)) {
484 * this is a pure, in-sequence data packet
485 * with nothing on the reassembly queue and
486 * we have enough buffer space to take it.
488 tp
->rcv_nxt
+= ti
->ti_len
;
490 * Add data to socket buffer.
493 if (tcp_emu(so
,m
)) sbappend(so
, m
);
498 * If this is a short packet, then ACK now - with Nagel
499 * congestion avoidance sender won't send more until
502 * It is better to not delay acks at all to maximize
503 * TCP throughput. See RFC 2581.
505 tp
->t_flags
|= TF_ACKNOW
;
509 } /* header prediction */
511 * Calculate amount of space in receive window,
512 * and then do TCP input processing.
513 * Receive window is amount of space in rcv queue,
514 * but not less than advertised window.
517 win
= sbspace(&so
->so_rcv
);
520 tp
->rcv_wnd
= max(win
, (int)(tp
->rcv_adv
- tp
->rcv_nxt
));
523 switch (tp
->t_state
) {
526 * If the state is LISTEN then ignore segment if it contains an RST.
527 * If the segment contains an ACK then it is bad and send a RST.
528 * If it does not contain a SYN then it is not interesting; drop it.
529 * Don't bother responding if the destination was a broadcast.
530 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
531 * tp->iss, and send a segment:
532 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
533 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
534 * Fill in remote peer address fields if not previously specified.
535 * Enter SYN_RECEIVED state, and process any other fields of this
536 * segment in this state.
540 if (tiflags
& TH_RST
)
542 if (tiflags
& TH_ACK
)
544 if ((tiflags
& TH_SYN
) == 0)
548 * This has way too many gotos...
549 * But a bit of spaghetti code never hurt anybody :)
553 * If this is destined for the control address, then flag to
554 * tcp_ctl once connected, otherwise connect
556 if ((so
->so_faddr
.s_addr
& slirp
->vnetwork_mask
.s_addr
) ==
557 slirp
->vnetwork_addr
.s_addr
) {
558 if (so
->so_faddr
.s_addr
!= slirp
->vhost_addr
.s_addr
&&
559 so
->so_faddr
.s_addr
!= slirp
->vnameserver_addr
.s_addr
) {
560 /* May be an add exec */
561 for (ex_ptr
= slirp
->exec_list
; ex_ptr
;
562 ex_ptr
= ex_ptr
->ex_next
) {
563 if(ex_ptr
->ex_fport
== so
->so_fport
&&
564 so
->so_faddr
.s_addr
== ex_ptr
->ex_addr
.s_addr
) {
565 so
->so_state
|= SS_CTL
;
569 if (so
->so_state
& SS_CTL
) {
573 /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
576 if (so
->so_emu
& EMU_NOCONNECT
) {
577 so
->so_emu
&= ~EMU_NOCONNECT
;
581 if((tcp_fconnect(so
) == -1) && (errno
!= EINPROGRESS
) && (errno
!= EWOULDBLOCK
)) {
582 u_char code
=ICMP_UNREACH_NET
;
583 DEBUG_MISC((dfd
, " tcp fconnect errno = %d-%s\n",
584 errno
,strerror(errno
)));
585 if(errno
== ECONNREFUSED
) {
586 /* ACK the SYN, send RST to refuse the connection */
587 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+1, (tcp_seq
)0,
590 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
591 HTONL(ti
->ti_seq
); /* restore tcp header */
595 m
->m_data
-= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
596 m
->m_len
+= sizeof(struct tcpiphdr
)+off
-sizeof(struct tcphdr
);
598 icmp_error(m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
604 * Haven't connected yet, save the current mbuf
606 * XXX Some OS's don't tell us whether the connect()
607 * succeeded or not. So we must time it out.
611 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
612 tp
->t_state
= TCPS_SYN_RECEIVED
;
619 * Check if the connect succeeded
621 if (so
->so_state
& SS_NOFDREF
) {
629 tcp_dooptions(tp
, (u_char
*)optp
, optlen
, ti
);
634 tp
->iss
= slirp
->tcp_iss
;
635 slirp
->tcp_iss
+= TCP_ISSINCR
/2;
636 tp
->irs
= ti
->ti_seq
;
639 tp
->t_flags
|= TF_ACKNOW
;
640 tp
->t_state
= TCPS_SYN_RECEIVED
;
641 tp
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
;
643 } /* case TCPS_LISTEN */
646 * If the state is SYN_SENT:
647 * if seg contains an ACK, but not for our SYN, drop the input.
648 * if seg contains a RST, then drop the connection.
649 * if seg does not contain SYN, then drop it.
650 * Otherwise this is an acceptable SYN segment
651 * initialize tp->rcv_nxt and tp->irs
652 * if seg contains ack then advance tp->snd_una
653 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
654 * arrange for segment to be acked (eventually)
655 * continue processing rest of data/controls, beginning with URG
658 if ((tiflags
& TH_ACK
) &&
659 (SEQ_LEQ(ti
->ti_ack
, tp
->iss
) ||
660 SEQ_GT(ti
->ti_ack
, tp
->snd_max
)))
663 if (tiflags
& TH_RST
) {
664 if (tiflags
& TH_ACK
) {
665 tcp_drop(tp
, 0); /* XXX Check t_softerror! */
670 if ((tiflags
& TH_SYN
) == 0)
672 if (tiflags
& TH_ACK
) {
673 tp
->snd_una
= ti
->ti_ack
;
674 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
675 tp
->snd_nxt
= tp
->snd_una
;
678 tp
->t_timer
[TCPT_REXMT
] = 0;
679 tp
->irs
= ti
->ti_seq
;
681 tp
->t_flags
|= TF_ACKNOW
;
682 if (tiflags
& TH_ACK
&& SEQ_GT(tp
->snd_una
, tp
->iss
)) {
684 tp
->t_state
= TCPS_ESTABLISHED
;
686 (void) tcp_reass(tp
, (struct tcpiphdr
*)0,
689 * if we didn't have to retransmit the SYN,
690 * use its rtt as our initial srtt & rtt var.
693 tcp_xmit_timer(tp
, tp
->t_rtt
);
695 tp
->t_state
= TCPS_SYN_RECEIVED
;
699 * Advance ti->ti_seq to correspond to first data byte.
700 * If data, trim to stay within window,
701 * dropping FIN if necessary.
704 if (ti
->ti_len
> tp
->rcv_wnd
) {
705 todrop
= ti
->ti_len
- tp
->rcv_wnd
;
707 ti
->ti_len
= tp
->rcv_wnd
;
710 tp
->snd_wl1
= ti
->ti_seq
- 1;
711 tp
->rcv_up
= ti
->ti_seq
;
713 } /* switch tp->t_state */
715 * States other than LISTEN or SYN_SENT.
716 * Check that at least some bytes of segment are within
717 * receive window. If segment begins before rcv_nxt,
718 * drop leading data (and SYN); if nothing left, just ack.
720 todrop
= tp
->rcv_nxt
- ti
->ti_seq
;
722 if (tiflags
& TH_SYN
) {
732 * Following if statement from Stevens, vol. 2, p. 960.
734 if (todrop
> ti
->ti_len
735 || (todrop
== ti
->ti_len
&& (tiflags
& TH_FIN
) == 0)) {
737 * Any valid FIN must be to the left of the window.
738 * At this point the FIN must be a duplicate or out
739 * of sequence; drop it.
744 * Send an ACK to resynchronize and drop any data.
745 * But keep on processing for RST or ACK.
747 tp
->t_flags
|= TF_ACKNOW
;
751 ti
->ti_seq
+= todrop
;
752 ti
->ti_len
-= todrop
;
753 if (ti
->ti_urp
> todrop
)
754 ti
->ti_urp
-= todrop
;
761 * If new data are received on a connection after the
762 * user processes are gone, then RST the other end.
764 if ((so
->so_state
& SS_NOFDREF
) &&
765 tp
->t_state
> TCPS_CLOSE_WAIT
&& ti
->ti_len
) {
771 * If segment ends after window, drop trailing data
772 * (and PUSH and FIN); if nothing left, just ACK.
774 todrop
= (ti
->ti_seq
+ti
->ti_len
) - (tp
->rcv_nxt
+tp
->rcv_wnd
);
776 if (todrop
>= ti
->ti_len
) {
778 * If a new connection request is received
779 * while in TIME_WAIT, drop the old connection
780 * and start over if the sequence numbers
781 * are above the previous ones.
783 if (tiflags
& TH_SYN
&&
784 tp
->t_state
== TCPS_TIME_WAIT
&&
785 SEQ_GT(ti
->ti_seq
, tp
->rcv_nxt
)) {
786 iss
= tp
->rcv_nxt
+ TCP_ISSINCR
;
791 * If window is closed can only take segments at
792 * window edge, and have to drop data and PUSH from
793 * incoming segments. Continue processing, but
794 * remember to ack. Otherwise, drop segment
797 if (tp
->rcv_wnd
== 0 && ti
->ti_seq
== tp
->rcv_nxt
) {
798 tp
->t_flags
|= TF_ACKNOW
;
804 ti
->ti_len
-= todrop
;
805 tiflags
&= ~(TH_PUSH
|TH_FIN
);
809 * If the RST bit is set examine the state:
810 * SYN_RECEIVED STATE:
811 * If passive open, return to LISTEN state.
812 * If active open, inform user that connection was refused.
813 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
814 * Inform user that connection was reset, and close tcb.
815 * CLOSING, LAST_ACK, TIME_WAIT STATES
818 if (tiflags
&TH_RST
) switch (tp
->t_state
) {
820 case TCPS_SYN_RECEIVED
:
821 case TCPS_ESTABLISHED
:
822 case TCPS_FIN_WAIT_1
:
823 case TCPS_FIN_WAIT_2
:
824 case TCPS_CLOSE_WAIT
:
825 tp
->t_state
= TCPS_CLOSED
;
837 * If a SYN is in the window, then this is an
838 * error and we send an RST and drop the connection.
840 if (tiflags
& TH_SYN
) {
846 * If the ACK bit is off we drop the segment and return.
848 if ((tiflags
& TH_ACK
) == 0) goto drop
;
853 switch (tp
->t_state
) {
855 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
856 * ESTABLISHED state and continue processing, otherwise
857 * send an RST. una<=ack<=max
859 case TCPS_SYN_RECEIVED
:
861 if (SEQ_GT(tp
->snd_una
, ti
->ti_ack
) ||
862 SEQ_GT(ti
->ti_ack
, tp
->snd_max
))
864 tp
->t_state
= TCPS_ESTABLISHED
;
866 * The sent SYN is ack'ed with our sequence number +1
867 * The first data byte already in the buffer will get
868 * lost if no correction is made. This is only needed for
869 * SS_CTL since the buffer is empty otherwise.
872 tp
->snd_una
=ti
->ti_ack
;
873 if (so
->so_state
& SS_CTL
) {
874 /* So tcp_ctl reports the right state */
878 so
->so_state
&= ~SS_CTL
; /* success XXX */
879 } else if (ret
== 2) {
880 so
->so_state
&= SS_PERSISTENT_MASK
;
881 so
->so_state
|= SS_NOFDREF
; /* CTL_CMD */
884 tp
->t_state
= TCPS_FIN_WAIT_1
;
890 (void) tcp_reass(tp
, (struct tcpiphdr
*)0, (struct mbuf
*)0);
891 tp
->snd_wl1
= ti
->ti_seq
- 1;
892 /* Avoid ack processing; snd_una==ti_ack => dup ack */
897 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
898 * ACKs. If the ack is in the range
899 * tp->snd_una < ti->ti_ack <= tp->snd_max
900 * then advance tp->snd_una to ti->ti_ack and drop
901 * data from the retransmission queue. If this ACK reflects
902 * more up to date window information we update our window information.
904 case TCPS_ESTABLISHED
:
905 case TCPS_FIN_WAIT_1
:
906 case TCPS_FIN_WAIT_2
:
907 case TCPS_CLOSE_WAIT
:
912 if (SEQ_LEQ(ti
->ti_ack
, tp
->snd_una
)) {
913 if (ti
->ti_len
== 0 && tiwin
== tp
->snd_wnd
) {
914 DEBUG_MISC((dfd
, " dup ack m = %lx so = %lx\n",
915 (long )m
, (long )so
));
917 * If we have outstanding data (other than
918 * a window probe), this is a completely
919 * duplicate ack (ie, window info didn't
920 * change), the ack is the biggest we've
921 * seen and we've seen exactly our rexmt
922 * threshold of them, assume a packet
923 * has been dropped and retransmit it.
924 * Kludge snd_nxt & the congestion
925 * window so we send only this one
928 * We know we're losing at the current
929 * window size so do congestion avoidance
930 * (set ssthresh to half the current window
931 * and pull our congestion window back to
934 * Dup acks mean that packets have left the
935 * network (they're now cached at the receiver)
936 * so bump cwnd by the amount in the receiver
937 * to keep a constant cwnd packets in the
940 if (tp
->t_timer
[TCPT_REXMT
] == 0 ||
941 ti
->ti_ack
!= tp
->snd_una
)
943 else if (++tp
->t_dupacks
== TCPREXMTTHRESH
) {
944 tcp_seq onxt
= tp
->snd_nxt
;
946 min(tp
->snd_wnd
, tp
->snd_cwnd
) / 2 /
951 tp
->snd_ssthresh
= win
* tp
->t_maxseg
;
952 tp
->t_timer
[TCPT_REXMT
] = 0;
954 tp
->snd_nxt
= ti
->ti_ack
;
955 tp
->snd_cwnd
= tp
->t_maxseg
;
956 (void) tcp_output(tp
);
957 tp
->snd_cwnd
= tp
->snd_ssthresh
+
958 tp
->t_maxseg
* tp
->t_dupacks
;
959 if (SEQ_GT(onxt
, tp
->snd_nxt
))
962 } else if (tp
->t_dupacks
> TCPREXMTTHRESH
) {
963 tp
->snd_cwnd
+= tp
->t_maxseg
;
964 (void) tcp_output(tp
);
973 * If the congestion window was inflated to account
974 * for the other side's cached packets, retract it.
976 if (tp
->t_dupacks
> TCPREXMTTHRESH
&&
977 tp
->snd_cwnd
> tp
->snd_ssthresh
)
978 tp
->snd_cwnd
= tp
->snd_ssthresh
;
980 if (SEQ_GT(ti
->ti_ack
, tp
->snd_max
)) {
983 acked
= ti
->ti_ack
- tp
->snd_una
;
986 * If transmit timer is running and timed sequence
987 * number was acked, update smoothed round trip time.
988 * Since we now have an rtt measurement, cancel the
989 * timer backoff (cf., Phil Karn's retransmit alg.).
990 * Recompute the initial retransmit timer.
992 if (tp
->t_rtt
&& SEQ_GT(ti
->ti_ack
, tp
->t_rtseq
))
993 tcp_xmit_timer(tp
,tp
->t_rtt
);
996 * If all outstanding data is acked, stop retransmit
997 * timer and remember to restart (more output or persist).
998 * If there is more data to be acked, restart retransmit
999 * timer, using current (possibly backed-off) value.
1001 if (ti
->ti_ack
== tp
->snd_max
) {
1002 tp
->t_timer
[TCPT_REXMT
] = 0;
1004 } else if (tp
->t_timer
[TCPT_PERSIST
] == 0)
1005 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
1007 * When new data is acked, open the congestion window.
1008 * If the window gives us less than ssthresh packets
1009 * in flight, open exponentially (maxseg per packet).
1010 * Otherwise open linearly: maxseg per window
1011 * (maxseg^2 / cwnd per packet).
1014 register u_int cw
= tp
->snd_cwnd
;
1015 register u_int incr
= tp
->t_maxseg
;
1017 if (cw
> tp
->snd_ssthresh
)
1018 incr
= incr
* incr
/ cw
;
1019 tp
->snd_cwnd
= min(cw
+ incr
, TCP_MAXWIN
<<tp
->snd_scale
);
1021 if (acked
> so
->so_snd
.sb_cc
) {
1022 tp
->snd_wnd
-= so
->so_snd
.sb_cc
;
1023 sbdrop(&so
->so_snd
, (int )so
->so_snd
.sb_cc
);
1026 sbdrop(&so
->so_snd
, acked
);
1027 tp
->snd_wnd
-= acked
;
1030 tp
->snd_una
= ti
->ti_ack
;
1031 if (SEQ_LT(tp
->snd_nxt
, tp
->snd_una
))
1032 tp
->snd_nxt
= tp
->snd_una
;
1034 switch (tp
->t_state
) {
1037 * In FIN_WAIT_1 STATE in addition to the processing
1038 * for the ESTABLISHED state if our FIN is now acknowledged
1039 * then enter FIN_WAIT_2.
1041 case TCPS_FIN_WAIT_1
:
1042 if (ourfinisacked
) {
1044 * If we can't receive any more
1045 * data, then closing user can proceed.
1046 * Starting the timer is contrary to the
1047 * specification, but if we don't get a FIN
1048 * we'll hang forever.
1050 if (so
->so_state
& SS_FCANTRCVMORE
) {
1051 tp
->t_timer
[TCPT_2MSL
] = TCP_MAXIDLE
;
1053 tp
->t_state
= TCPS_FIN_WAIT_2
;
1058 * In CLOSING STATE in addition to the processing for
1059 * the ESTABLISHED state if the ACK acknowledges our FIN
1060 * then enter the TIME-WAIT state, otherwise ignore
1064 if (ourfinisacked
) {
1065 tp
->t_state
= TCPS_TIME_WAIT
;
1066 tcp_canceltimers(tp
);
1067 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1072 * In LAST_ACK, we may still be waiting for data to drain
1073 * and/or to be acked, as well as for the ack of our FIN.
1074 * If our FIN is now acknowledged, delete the TCB,
1075 * enter the closed state and return.
1078 if (ourfinisacked
) {
1085 * In TIME_WAIT state the only thing that should arrive
1086 * is a retransmission of the remote FIN. Acknowledge
1087 * it and restart the finack timer.
1089 case TCPS_TIME_WAIT
:
1090 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1093 } /* switch(tp->t_state) */
1097 * Update window information.
1098 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1100 if ((tiflags
& TH_ACK
) &&
1101 (SEQ_LT(tp
->snd_wl1
, ti
->ti_seq
) ||
1102 (tp
->snd_wl1
== ti
->ti_seq
&& (SEQ_LT(tp
->snd_wl2
, ti
->ti_ack
) ||
1103 (tp
->snd_wl2
== ti
->ti_ack
&& tiwin
> tp
->snd_wnd
))))) {
1104 tp
->snd_wnd
= tiwin
;
1105 tp
->snd_wl1
= ti
->ti_seq
;
1106 tp
->snd_wl2
= ti
->ti_ack
;
1107 if (tp
->snd_wnd
> tp
->max_sndwnd
)
1108 tp
->max_sndwnd
= tp
->snd_wnd
;
1113 * Process segments with URG.
1115 if ((tiflags
& TH_URG
) && ti
->ti_urp
&&
1116 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1118 * This is a kludge, but if we receive and accept
1119 * random urgent pointers, we'll crash in
1120 * soreceive. It's hard to imagine someone
1121 * actually wanting to send this much urgent data.
1123 if (ti
->ti_urp
+ so
->so_rcv
.sb_cc
> so
->so_rcv
.sb_datalen
) {
1129 * If this segment advances the known urgent pointer,
1130 * then mark the data stream. This should not happen
1131 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1132 * a FIN has been received from the remote side.
1133 * In these states we ignore the URG.
1135 * According to RFC961 (Assigned Protocols),
1136 * the urgent pointer points to the last octet
1137 * of urgent data. We continue, however,
1138 * to consider it to indicate the first octet
1139 * of data past the urgent section as the original
1140 * spec states (in one of two places).
1142 if (SEQ_GT(ti
->ti_seq
+ti
->ti_urp
, tp
->rcv_up
)) {
1143 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1144 so
->so_urgc
= so
->so_rcv
.sb_cc
+
1145 (tp
->rcv_up
- tp
->rcv_nxt
); /* -1; */
1146 tp
->rcv_up
= ti
->ti_seq
+ ti
->ti_urp
;
1151 * If no out of band data is expected,
1152 * pull receive urgent pointer along
1153 * with the receive window.
1155 if (SEQ_GT(tp
->rcv_nxt
, tp
->rcv_up
))
1156 tp
->rcv_up
= tp
->rcv_nxt
;
1160 * If this is a small packet, then ACK now - with Nagel
1161 * congestion avoidance sender won't send more until
1164 if (ti
->ti_len
&& (unsigned)ti
->ti_len
<= 5 &&
1165 ((struct tcpiphdr_2
*)ti
)->first_char
== (char)27) {
1166 tp
->t_flags
|= TF_ACKNOW
;
1170 * Process the segment text, merging it into the TCP sequencing queue,
1171 * and arranging for acknowledgment of receipt if necessary.
1172 * This process logically involves adjusting tp->rcv_wnd as data
1173 * is presented to the user (this happens in tcp_usrreq.c,
1174 * case PRU_RCVD). If a FIN has already been received on this
1175 * connection then we just ignore the text.
1177 if ((ti
->ti_len
|| (tiflags
&TH_FIN
)) &&
1178 TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1179 TCP_REASS(tp
, ti
, m
, so
, tiflags
);
1186 * If FIN is received ACK the FIN and let the user know
1187 * that the connection is closing.
1189 if (tiflags
& TH_FIN
) {
1190 if (TCPS_HAVERCVDFIN(tp
->t_state
) == 0) {
1192 * If we receive a FIN we can't send more data,
1194 * Shutdown the socket if there is no rx data in the
1196 * soread() is called on completion of shutdown() and
1197 * will got to TCPS_LAST_ACK, and use tcp_output()
1202 tp
->t_flags
|= TF_ACKNOW
;
1205 switch (tp
->t_state
) {
1208 * In SYN_RECEIVED and ESTABLISHED STATES
1209 * enter the CLOSE_WAIT state.
1211 case TCPS_SYN_RECEIVED
:
1212 case TCPS_ESTABLISHED
:
1213 if(so
->so_emu
== EMU_CTL
) /* no shutdown on socket */
1214 tp
->t_state
= TCPS_LAST_ACK
;
1216 tp
->t_state
= TCPS_CLOSE_WAIT
;
1220 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1221 * enter the CLOSING state.
1223 case TCPS_FIN_WAIT_1
:
1224 tp
->t_state
= TCPS_CLOSING
;
1228 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1229 * starting the time-wait timer, turning off the other
1232 case TCPS_FIN_WAIT_2
:
1233 tp
->t_state
= TCPS_TIME_WAIT
;
1234 tcp_canceltimers(tp
);
1235 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1239 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1241 case TCPS_TIME_WAIT
:
1242 tp
->t_timer
[TCPT_2MSL
] = 2 * TCPTV_MSL
;
1248 * Return any desired output.
1250 if (needoutput
|| (tp
->t_flags
& TF_ACKNOW
)) {
1251 (void) tcp_output(tp
);
1257 * Generate an ACK dropping incoming segment if it occupies
1258 * sequence space, where the ACK reflects our state.
1260 if (tiflags
& TH_RST
)
1263 tp
->t_flags
|= TF_ACKNOW
;
1264 (void) tcp_output(tp
);
1268 /* reuses m if m!=NULL, m_free() unnecessary */
1269 if (tiflags
& TH_ACK
)
1270 tcp_respond(tp
, ti
, m
, (tcp_seq
)0, ti
->ti_ack
, TH_RST
);
1272 if (tiflags
& TH_SYN
) ti
->ti_len
++;
1273 tcp_respond(tp
, ti
, m
, ti
->ti_seq
+ti
->ti_len
, (tcp_seq
)0,
1281 * Drop space held by incoming segment and return.
1287 tcp_dooptions(struct tcpcb
*tp
, u_char
*cp
, int cnt
, struct tcpiphdr
*ti
)
1292 DEBUG_CALL("tcp_dooptions");
1293 DEBUG_ARGS((dfd
, " tp = %lx cnt=%i\n", (long)tp
, cnt
));
1295 for (; cnt
> 0; cnt
-= optlen
, cp
+= optlen
) {
1297 if (opt
== TCPOPT_EOL
)
1299 if (opt
== TCPOPT_NOP
)
1312 if (optlen
!= TCPOLEN_MAXSEG
)
1314 if (!(ti
->ti_flags
& TH_SYN
))
1316 memcpy((char *) &mss
, (char *) cp
+ 2, sizeof(mss
));
1318 (void) tcp_mss(tp
, mss
); /* sets t_maxseg */
1326 * Pull out of band byte out of a segment so
1327 * it doesn't appear in the user's data queue.
1328 * It is still reflected in the segment length for
1329 * sequencing purposes.
1335 tcp_pulloutofband(so
, ti
, m
)
1337 struct tcpiphdr
*ti
;
1338 register struct mbuf
*m
;
1340 int cnt
= ti
->ti_urp
- 1;
1343 if (m
->m_len
> cnt
) {
1344 char *cp
= mtod(m
, caddr_t
) + cnt
;
1345 struct tcpcb
*tp
= sototcpcb(so
);
1348 tp
->t_oobflags
|= TCPOOB_HAVEDATA
;
1349 memcpy(sp
, cp
+1, (unsigned)(m
->m_len
- cnt
- 1));
1354 m
= m
->m_next
; /* XXX WRONG! Fix it! */
1358 panic("tcp_pulloutofband");
1364 * Collect new round-trip time estimate
1365 * and update averages and current timeout.
1369 tcp_xmit_timer(register struct tcpcb
*tp
, int rtt
)
1371 register short delta
;
1373 DEBUG_CALL("tcp_xmit_timer");
1374 DEBUG_ARG("tp = %lx", (long)tp
);
1375 DEBUG_ARG("rtt = %d", rtt
);
1377 if (tp
->t_srtt
!= 0) {
1379 * srtt is stored as fixed point with 3 bits after the
1380 * binary point (i.e., scaled by 8). The following magic
1381 * is equivalent to the smoothing algorithm in rfc793 with
1382 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1383 * point). Adjust rtt to origin 0.
1385 delta
= rtt
- 1 - (tp
->t_srtt
>> TCP_RTT_SHIFT
);
1386 if ((tp
->t_srtt
+= delta
) <= 0)
1389 * We accumulate a smoothed rtt variance (actually, a
1390 * smoothed mean difference), then set the retransmit
1391 * timer to smoothed rtt + 4 times the smoothed variance.
1392 * rttvar is stored as fixed point with 2 bits after the
1393 * binary point (scaled by 4). The following is
1394 * equivalent to rfc793 smoothing with an alpha of .75
1395 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
1396 * rfc793's wired-in beta.
1400 delta
-= (tp
->t_rttvar
>> TCP_RTTVAR_SHIFT
);
1401 if ((tp
->t_rttvar
+= delta
) <= 0)
1405 * No rtt measurement yet - use the unsmoothed rtt.
1406 * Set the variance to half the rtt (so our first
1407 * retransmit happens at 3*rtt).
1409 tp
->t_srtt
= rtt
<< TCP_RTT_SHIFT
;
1410 tp
->t_rttvar
= rtt
<< (TCP_RTTVAR_SHIFT
- 1);
1416 * the retransmit should happen at rtt + 4 * rttvar.
1417 * Because of the way we do the smoothing, srtt and rttvar
1418 * will each average +1/2 tick of bias. When we compute
1419 * the retransmit timer, we want 1/2 tick of rounding and
1420 * 1 extra tick because of +-1/2 tick uncertainty in the
1421 * firing of the timer. The bias will give us exactly the
1422 * 1.5 tick we need. But, because the bias is
1423 * statistical, we have to test that we don't drop below
1424 * the minimum feasible timer (which is 2 ticks).
1426 TCPT_RANGESET(tp
->t_rxtcur
, TCP_REXMTVAL(tp
),
1427 (short)tp
->t_rttmin
, TCPTV_REXMTMAX
); /* XXX */
1430 * We received an ack for a packet that wasn't retransmitted;
1431 * it is probably safe to discard any error indications we've
1432 * received recently. This isn't quite right, but close enough
1433 * for now (a route might have failed after we sent a segment,
1434 * and the return path might not be symmetrical).
1436 tp
->t_softerror
= 0;
1440 * Determine a reasonable value for maxseg size.
1441 * If the route is known, check route for mtu.
1442 * If none, use an mss that can be handled on the outgoing
1443 * interface without forcing IP to fragment; if bigger than
1444 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1445 * to utilize large mbufs. If no route is found, route has no mtu,
1446 * or the destination isn't local, use a default, hopefully conservative
1447 * size (usually 512 or the default IP max size, but no more than the mtu
1448 * of the interface), as we can't discover anything about intervening
1449 * gateways or networks. We also initialize the congestion/slow start
1450 * window to be a single segment if the destination isn't local.
1451 * While looking at the routing entry, we also initialize other path-dependent
1452 * parameters from pre-set or cached values in the routing entry.
1456 tcp_mss(struct tcpcb
*tp
, u_int offer
)
1458 struct socket
*so
= tp
->t_socket
;
1461 DEBUG_CALL("tcp_mss");
1462 DEBUG_ARG("tp = %lx", (long)tp
);
1463 DEBUG_ARG("offer = %d", offer
);
1465 mss
= min(IF_MTU
, IF_MRU
) - sizeof(struct tcpiphdr
);
1467 mss
= min(mss
, offer
);
1469 if (mss
< tp
->t_maxseg
|| offer
!= 0)
1474 sbreserve(&so
->so_snd
, TCP_SNDSPACE
+ ((TCP_SNDSPACE
% mss
) ?
1475 (mss
- (TCP_SNDSPACE
% mss
)) :
1477 sbreserve(&so
->so_rcv
, TCP_RCVSPACE
+ ((TCP_RCVSPACE
% mss
) ?
1478 (mss
- (TCP_RCVSPACE
% mss
)) :
1481 DEBUG_MISC((dfd
, " returning mss = %d\n", mss
));