2
* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
3
* The Regents of the University of California. All rights reserved.
5
* Redistribution and use in source and binary forms, with or without
6
* modification, are permitted provided that the following conditions
8
* 1. Redistributions of source code must retain the above copyright
9
* notice, this list of conditions and the following disclaimer.
10
* 2. Redistributions in binary form must reproduce the above copyright
11
* notice, this list of conditions and the following disclaimer in the
12
* documentation and/or other materials provided with the distribution.
13
* 3. All advertising materials mentioning features or use of this software
14
* must display the following acknowledgement:
15
* This product includes software developed by the University of
16
* California, Berkeley and its contributors.
17
* 4. Neither the name of the University nor the names of its contributors
18
* may be used to endorse or promote products derived from this software
19
* without specific prior written permission.
21
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33
* @(#)tcp_input.c 8.5 (Berkeley) 4/10/94
34
* tcp_input.c,v 1.10 1994/10/13 18:36:32 wollman Exp
38
* Changes and additions relating to SLiRP
39
* Copyright (c) 1995 Danny Gasparovski.
41
* Please read the file COPYRIGHT for the
42
* terms and conditions of the copyright.
50
int tcprexmtthresh = 3;
51
struct socket *tcp_last_so = &tcb;
53
tcp_seq tcp_iss; /* tcp initial send seq # */
55
#define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
57
/* for modulo comparisons of timestamps */
58
#define TSTMP_LT(a,b) ((int)((a)-(b)) < 0)
59
#define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0)
62
* Insert segment ti into reassembly queue of tcp with
63
* control block tp. Return TH_FIN if reassembly now includes
64
* a segment with FIN. The macro form does the common case inline
65
* (segment is the next to be received on an established connection,
66
* and the queue is empty), avoiding linkage into and removal
67
* from the queue and repetition of various conversions.
68
* Set DELACK for segments received in order, but ack immediately
69
* when segments are out of order (so fast retransmit can work).
72
#define TCP_REASS(tp, ti, m, so, flags) {\
73
if ((ti)->ti_seq == (tp)->rcv_nxt && \
74
(tp)->seg_next == (tcpiphdrp_32)(tp) && \
75
(tp)->t_state == TCPS_ESTABLISHED) {\
76
if (ti->ti_flags & TH_PUSH) \
77
tp->t_flags |= TF_ACKNOW; \
79
tp->t_flags |= TF_DELACK; \
80
(tp)->rcv_nxt += (ti)->ti_len; \
81
flags = (ti)->ti_flags & TH_FIN; \
82
tcpstat.tcps_rcvpack++;\
83
tcpstat.tcps_rcvbyte += (ti)->ti_len;\
85
if (tcp_emu((so),(m))) sbappend((so), (m)); \
87
sbappend((so), (m)); \
88
/* sorwakeup(so); */ \
90
(flags) = tcp_reass((tp), (ti), (m)); \
91
tp->t_flags |= TF_ACKNOW; \
95
#define TCP_REASS(tp, ti, m, so, flags) { \
96
if ((ti)->ti_seq == (tp)->rcv_nxt && \
97
(tp)->seg_next == (tcpiphdrp_32)(tp) && \
98
(tp)->t_state == TCPS_ESTABLISHED) { \
99
tp->t_flags |= TF_DELACK; \
100
(tp)->rcv_nxt += (ti)->ti_len; \
101
flags = (ti)->ti_flags & TH_FIN; \
102
tcpstat.tcps_rcvpack++;\
103
tcpstat.tcps_rcvbyte += (ti)->ti_len;\
105
if (tcp_emu((so),(m))) sbappend(so, (m)); \
107
sbappend((so), (m)); \
108
/* sorwakeup(so); */ \
110
(flags) = tcp_reass((tp), (ti), (m)); \
111
tp->t_flags |= TF_ACKNOW; \
118
register struct tcpcb *tp;
119
register struct tcpiphdr *ti;
122
register struct tcpiphdr *q;
123
struct socket *so = tp->t_socket;
127
* Call with ti==0 after become established to
128
* force pre-ESTABLISHED data up to user socket.
134
* Find a segment which begins after this one does.
136
for (q = (struct tcpiphdr *)tp->seg_next; q != (struct tcpiphdr *)tp;
137
q = (struct tcpiphdr *)q->ti_next)
138
if (SEQ_GT(q->ti_seq, ti->ti_seq))
142
* If there is a preceding segment, it may provide some of
143
* our data already. If so, drop the data from the incoming
144
* segment. If it provides all of our data, drop us.
146
if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
148
q = (struct tcpiphdr *)q->ti_prev;
149
/* conversion to int (in i) handles seq wraparound */
150
i = q->ti_seq + q->ti_len - ti->ti_seq;
152
if (i >= ti->ti_len) {
153
tcpstat.tcps_rcvduppack++;
154
tcpstat.tcps_rcvdupbyte += ti->ti_len;
157
* Try to present any queued data
158
* at the left window edge to the user.
159
* This is needed after the 3-WHS
162
goto present; /* ??? */
168
q = (struct tcpiphdr *)(q->ti_next);
170
tcpstat.tcps_rcvoopack++;
171
tcpstat.tcps_rcvoobyte += ti->ti_len;
172
REASS_MBUF(ti) = (mbufp_32) m; /* XXX */
175
* While we overlap succeeding segments trim them or,
176
* if they are completely covered, dequeue them.
178
while (q != (struct tcpiphdr *)tp) {
179
register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
185
m_adj((struct mbuf *) REASS_MBUF(q), i);
188
q = (struct tcpiphdr *)q->ti_next;
189
m = (struct mbuf *) REASS_MBUF((struct tcpiphdr *)q->ti_prev);
190
remque_32((void *)(q->ti_prev));
195
* Stick new segment in its place.
197
insque_32(ti, (void *)(q->ti_prev));
201
* Present data to user, advancing rcv_nxt through
202
* completed sequence space.
204
if (!TCPS_HAVEESTABLISHED(tp->t_state))
206
ti = (struct tcpiphdr *) tp->seg_next;
207
if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
209
if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
212
tp->rcv_nxt += ti->ti_len;
213
flags = ti->ti_flags & TH_FIN;
215
m = (struct mbuf *) REASS_MBUF(ti); /* XXX */
216
ti = (struct tcpiphdr *)ti->ti_next;
217
/* if (so->so_state & SS_FCANTRCVMORE) */
218
if (so->so_state & SS_FCANTSENDMORE)
222
if (tcp_emu(so,m)) sbappend(so, m);
226
} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
232
* TCP input routine, follows pages 65-76 of the
233
* protocol specification dated September, 1981 very closely.
236
tcp_input(m, iphlen, inso)
237
register struct mbuf *m;
241
struct ip save_ip, *ip;
242
register struct tcpiphdr *ti;
246
register struct tcpcb *tp = 0;
247
register int tiflags;
248
struct socket *so = 0;
249
int todrop, acked, ourfinisacked, needoutput = 0;
250
/* int dropsocket = 0; */
254
/* int ts_present = 0; */
256
DEBUG_CALL("tcp_input");
257
DEBUG_ARGS((dfd," m = %8lx iphlen = %2d inso = %lx\n",
258
(long )m, iphlen, (long )inso ));
261
* If called with m == 0, then we're continuing the connect
266
/* Re-set a few variables */
272
tiflags = ti->ti_flags;
278
tcpstat.tcps_rcvtotal++;
280
* Get IP and TCP header together in first mbuf.
281
* Note: IP leaves IP header in first mbuf.
283
ti = mtod(m, struct tcpiphdr *);
284
if (iphlen > sizeof(struct ip )) {
285
ip_stripoptions(m, (struct mbuf *)0);
286
iphlen=sizeof(struct ip );
288
/* XXX Check if too short */
292
* Save a copy of the IP header in case we want restore it
293
* for sending an ICMP error message in response.
295
ip=mtod(m, struct ip *);
297
save_ip.ip_len+= iphlen;
300
* Checksum extended TCP header and data.
302
tlen = ((struct ip *)ti)->ip_len;
303
ti->ti_next = ti->ti_prev = 0;
305
ti->ti_len = htons((u_int16_t)tlen);
306
len = sizeof(struct ip ) + tlen;
307
/* keep checksum for ICMP reply
308
* ti->ti_sum = cksum(m, len);
309
* if (ti->ti_sum) { */
311
tcpstat.tcps_rcvbadsum++;
316
* Check that TCP offset makes sense,
317
* pull out TCP options and adjust length. XXX
319
off = ti->ti_off << 2;
320
if (off < sizeof (struct tcphdr) || off > tlen) {
321
tcpstat.tcps_rcvbadoff++;
326
if (off > sizeof (struct tcphdr)) {
327
optlen = off - sizeof (struct tcphdr);
328
optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
331
* Do quick retrieval of timestamp options ("options
332
* prediction?"). If timestamp is the only option and it's
333
* formatted as recommended in RFC 1323 appendix A, we
334
* quickly get the values now and not bother calling
335
* tcp_dooptions(), etc.
337
/* if ((optlen == TCPOLEN_TSTAMP_APPA ||
338
* (optlen > TCPOLEN_TSTAMP_APPA &&
339
* optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
340
* *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
341
* (ti->ti_flags & TH_SYN) == 0) {
343
* ts_val = ntohl(*(u_int32_t *)(optp + 4));
344
* ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
345
* optp = NULL; / * we've parsed the options * /
349
tiflags = ti->ti_flags;
352
* Convert TCP protocol specific fields to host format.
360
* Drop TCP, IP headers and TCP options.
362
m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
363
m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
366
* Locate pcb for segment.
370
if (so->so_fport != ti->ti_dport ||
371
so->so_lport != ti->ti_sport ||
372
so->so_laddr.s_addr != ti->ti_src.s_addr ||
373
so->so_faddr.s_addr != ti->ti_dst.s_addr) {
374
so = solookup(&tcb, ti->ti_src, ti->ti_sport,
375
ti->ti_dst, ti->ti_dport);
378
++tcpstat.tcps_socachemiss;
382
* If the state is CLOSED (i.e., TCB does not exist) then
383
* all data in the incoming segment is discarded.
384
* If the TCB exists but is in CLOSED state, it is embryonic,
385
* but should either do a listen or a connect soon.
387
* state == CLOSED means we've done socreate() but haven't
388
* attached it to a protocol yet...
390
* XXX If a TCB does not exist, and the TH_SYN flag is
391
* the only flag set, then create a session, mark it
392
* as if it was LISTENING, and continue...
395
if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
398
if ((so = socreate()) == NULL)
400
if (tcp_attach(so) < 0) {
401
free(so); /* Not sofree (if it failed, it's not insqued) */
405
sbreserve(&so->so_snd, tcp_sndspace);
406
sbreserve(&so->so_rcv, tcp_rcvspace);
408
/* tcp_last_so = so; */ /* XXX ? */
409
/* tp = sototcpcb(so); */
411
so->so_laddr = ti->ti_src;
412
so->so_lport = ti->ti_sport;
413
so->so_faddr = ti->ti_dst;
414
so->so_fport = ti->ti_dport;
416
if ((so->so_iptos = tcp_tos(so)) == 0)
417
so->so_iptos = ((struct ip *)ti)->ip_tos;
420
tp->t_state = TCPS_LISTEN;
424
* If this is a still-connecting socket, this probably
425
* a retransmit of the SYN. Whether it's a retransmit SYN
426
* or something else, we nuke it.
428
if (so->so_state & SS_ISFCONNECTING)
433
/* XXX Should never fail */
436
if (tp->t_state == TCPS_CLOSED)
439
/* Unscale the window into a 32-bit value. */
440
/* if ((tiflags & TH_SYN) == 0)
441
* tiwin = ti->ti_win << tp->snd_scale;
447
* Segment received on connection.
448
* Reset idle time and keep-alive timer.
452
tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
454
tp->t_timer[TCPT_KEEP] = tcp_keepidle;
457
* Process options if not in LISTEN state,
458
* else do it below (after getting remote address).
460
if (optp && tp->t_state != TCPS_LISTEN)
461
tcp_dooptions(tp, (u_char *)optp, optlen, ti);
463
/* &ts_present, &ts_val, &ts_ecr); */
466
* Header prediction: check for the two common cases
467
* of a uni-directional data xfer. If the packet has
468
* no control flags, is in-sequence, the window didn't
469
* change and we're not retransmitting, it's a
470
* candidate. If the length is zero and the ack moved
471
* forward, we're the sender side of the xfer. Just
472
* free the data acked & wake any higher level process
473
* that was blocked waiting for space. If the length
474
* is non-zero and the ack didn't move, we're the
475
* receiver side. If we're getting packets in-order
476
* (the reassembly queue is empty), add the data to
477
* the socket buffer and note that we need a delayed ack.
479
* XXX Some of these tests are not needed
480
* eg: the tiwin == tp->snd_wnd prevents many more
481
* predictions.. with no *real* advantage..
483
if (tp->t_state == TCPS_ESTABLISHED &&
484
(tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
485
/* (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && */
486
ti->ti_seq == tp->rcv_nxt &&
487
tiwin && tiwin == tp->snd_wnd &&
488
tp->snd_nxt == tp->snd_max) {
490
* If last ACK falls within this segment's sequence numbers,
491
* record the timestamp.
493
/* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
494
* SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
495
* tp->ts_recent_age = tcp_now;
496
* tp->ts_recent = ts_val;
499
if (ti->ti_len == 0) {
500
if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
501
SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
502
tp->snd_cwnd >= tp->snd_wnd) {
504
* this is a pure ack for outstanding data.
506
++tcpstat.tcps_predack;
508
* tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
511
SEQ_GT(ti->ti_ack, tp->t_rtseq))
512
tcp_xmit_timer(tp, tp->t_rtt);
513
acked = ti->ti_ack - tp->snd_una;
514
tcpstat.tcps_rcvackpack++;
515
tcpstat.tcps_rcvackbyte += acked;
516
sbdrop(&so->so_snd, acked);
517
tp->snd_una = ti->ti_ack;
521
* If all outstanding data are acked, stop
522
* retransmit timer, otherwise restart timer
523
* using current (possibly backed-off) value.
524
* If process is waiting for space,
525
* wakeup/selwakeup/signal. If data
526
* are ready to send, let tcp_output
527
* decide between more output or persist.
529
if (tp->snd_una == tp->snd_max)
530
tp->t_timer[TCPT_REXMT] = 0;
531
else if (tp->t_timer[TCPT_PERSIST] == 0)
532
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
535
* There's room in so_snd, sowwakup will read()
536
* from the socket if we can
538
/* if (so->so_snd.sb_flags & SB_NOTIFY)
542
* This is called because sowwakeup might have
543
* put data into so_snd. Since we don't so sowwakeup,
544
* we don't need this.. XXX???
546
if (so->so_snd.sb_cc)
547
(void) tcp_output(tp);
551
} else if (ti->ti_ack == tp->snd_una &&
552
tp->seg_next == (tcpiphdrp_32)tp &&
553
ti->ti_len <= sbspace(&so->so_rcv)) {
555
* this is a pure, in-sequence data packet
556
* with nothing on the reassembly queue and
557
* we have enough buffer space to take it.
559
++tcpstat.tcps_preddat;
560
tp->rcv_nxt += ti->ti_len;
561
tcpstat.tcps_rcvpack++;
562
tcpstat.tcps_rcvbyte += ti->ti_len;
564
* Add data to socket buffer.
567
if (tcp_emu(so,m)) sbappend(so, m);
572
* XXX This is called when data arrives. Later, check
573
* if we can actually write() to the socket
574
* XXX Need to check? It's be NON_BLOCKING
579
* If this is a short packet, then ACK now - with Nagel
580
* congestion avoidance sender won't send more until
583
* It is better to not delay acks at all to maximize
584
* TCP throughput. See RFC 2581.
586
tp->t_flags |= TF_ACKNOW;
590
} /* header prediction */
592
* Calculate amount of space in receive window,
593
* and then do TCP input processing.
594
* Receive window is amount of space in rcv queue,
595
* but not less than advertised window.
598
win = sbspace(&so->so_rcv);
601
tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
604
switch (tp->t_state) {
607
* If the state is LISTEN then ignore segment if it contains an RST.
608
* If the segment contains an ACK then it is bad and send a RST.
609
* If it does not contain a SYN then it is not interesting; drop it.
610
* Don't bother responding if the destination was a broadcast.
611
* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
612
* tp->iss, and send a segment:
613
* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
614
* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
615
* Fill in remote peer address fields if not previously specified.
616
* Enter SYN_RECEIVED state, and process any other fields of this
617
* segment in this state.
621
if (tiflags & TH_RST)
623
if (tiflags & TH_ACK)
625
if ((tiflags & TH_SYN) == 0)
629
* This has way too many gotos...
630
* But a bit of spaghetti code never hurt anybody :)
634
* If this is destined for the control address, then flag to
635
* tcp_ctl once connected, otherwise connect
637
if ((so->so_faddr.s_addr&htonl(0xffffff00)) == special_addr.s_addr) {
638
int lastbyte=ntohl(so->so_faddr.s_addr) & 0xff;
639
if (lastbyte!=CTL_ALIAS && lastbyte!=CTL_DNS) {
641
if(lastbyte==CTL_CMD || lastbyte==CTL_EXEC) {
642
/* Command or exec adress */
643
so->so_state |= SS_CTL;
647
/* May be an add exec */
648
struct ex_list *ex_ptr;
649
for(ex_ptr = exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next) {
650
if(ex_ptr->ex_fport == so->so_fport &&
651
lastbyte == ex_ptr->ex_addr) {
652
so->so_state |= SS_CTL;
657
if(so->so_state & SS_CTL) goto cont_input;
659
/* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
662
if (so->so_emu & EMU_NOCONNECT) {
663
so->so_emu &= ~EMU_NOCONNECT;
667
if((tcp_fconnect(so) == -1) && (errno != EINPROGRESS) && (errno != EWOULDBLOCK)) {
668
u_char code=ICMP_UNREACH_NET;
669
DEBUG_MISC((dfd," tcp fconnect errno = %d-%s\n",
670
errno,strerror(errno)));
671
if(errno == ECONNREFUSED) {
672
/* ACK the SYN, send RST to refuse the connection */
673
tcp_respond(tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
676
if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
677
HTONL(ti->ti_seq); /* restore tcp header */
681
m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
682
m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
684
icmp_error(m, ICMP_UNREACH,code, 0,strerror(errno));
690
* Haven't connected yet, save the current mbuf
692
* XXX Some OS's don't tell us whether the connect()
693
* succeeded or not. So we must time it out.
697
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
698
tp->t_state = TCPS_SYN_RECEIVED;
704
* Check if the connect succeeded
706
if (so->so_state & SS_NOFDREF) {
714
tcp_dooptions(tp, (u_char *)optp, optlen, ti);
716
/* &ts_present, &ts_val, &ts_ecr); */
722
tcp_iss += TCP_ISSINCR/2;
723
tp->irs = ti->ti_seq;
726
tp->t_flags |= TF_ACKNOW;
727
tp->t_state = TCPS_SYN_RECEIVED;
728
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
729
tcpstat.tcps_accepts++;
731
} /* case TCPS_LISTEN */
734
* If the state is SYN_SENT:
735
* if seg contains an ACK, but not for our SYN, drop the input.
736
* if seg contains a RST, then drop the connection.
737
* if seg does not contain SYN, then drop it.
738
* Otherwise this is an acceptable SYN segment
739
* initialize tp->rcv_nxt and tp->irs
740
* if seg contains ack then advance tp->snd_una
741
* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
742
* arrange for segment to be acked (eventually)
743
* continue processing rest of data/controls, beginning with URG
746
if ((tiflags & TH_ACK) &&
747
(SEQ_LEQ(ti->ti_ack, tp->iss) ||
748
SEQ_GT(ti->ti_ack, tp->snd_max)))
751
if (tiflags & TH_RST) {
752
if (tiflags & TH_ACK)
753
tp = tcp_drop(tp,0); /* XXX Check t_softerror! */
757
if ((tiflags & TH_SYN) == 0)
759
if (tiflags & TH_ACK) {
760
tp->snd_una = ti->ti_ack;
761
if (SEQ_LT(tp->snd_nxt, tp->snd_una))
762
tp->snd_nxt = tp->snd_una;
765
tp->t_timer[TCPT_REXMT] = 0;
766
tp->irs = ti->ti_seq;
768
tp->t_flags |= TF_ACKNOW;
769
if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
770
tcpstat.tcps_connects++;
772
tp->t_state = TCPS_ESTABLISHED;
774
/* Do window scaling on this connection? */
775
/* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
776
* (TF_RCVD_SCALE|TF_REQ_SCALE)) {
777
* tp->snd_scale = tp->requested_s_scale;
778
* tp->rcv_scale = tp->request_r_scale;
781
(void) tcp_reass(tp, (struct tcpiphdr *)0,
784
* if we didn't have to retransmit the SYN,
785
* use its rtt as our initial srtt & rtt var.
788
tcp_xmit_timer(tp, tp->t_rtt);
790
tp->t_state = TCPS_SYN_RECEIVED;
794
* Advance ti->ti_seq to correspond to first data byte.
795
* If data, trim to stay within window,
796
* dropping FIN if necessary.
799
if (ti->ti_len > tp->rcv_wnd) {
800
todrop = ti->ti_len - tp->rcv_wnd;
802
ti->ti_len = tp->rcv_wnd;
804
tcpstat.tcps_rcvpackafterwin++;
805
tcpstat.tcps_rcvbyteafterwin += todrop;
807
tp->snd_wl1 = ti->ti_seq - 1;
808
tp->rcv_up = ti->ti_seq;
810
} /* switch tp->t_state */
812
* States other than LISTEN or SYN_SENT.
813
* First check timestamp, if present.
814
* Then check that at least some bytes of segment are within
815
* receive window. If segment begins before rcv_nxt,
816
* drop leading data (and SYN); if nothing left, just ack.
818
* RFC 1323 PAWS: If we have a timestamp reply on this segment
819
* and it's less than ts_recent, drop it.
821
/* if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
822
* TSTMP_LT(ts_val, tp->ts_recent)) {
824
*/ /* Check to see if ts_recent is over 24 days old. */
825
/* if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
827
* * Invalidate ts_recent. If this segment updates
828
* * ts_recent, the age will be reset later and ts_recent
829
* * will get a valid value. If it does not, setting
830
* * ts_recent to zero will at least satisfy the
831
* * requirement that zero be placed in the timestamp
832
* * echo reply when ts_recent isn't valid. The
833
* * age isn't reset until we get a valid ts_recent
834
* * because we don't want out-of-order segments to be
835
* * dropped when ts_recent is old.
837
/* tp->ts_recent = 0;
839
* tcpstat.tcps_rcvduppack++;
840
* tcpstat.tcps_rcvdupbyte += ti->ti_len;
841
* tcpstat.tcps_pawsdrop++;
847
todrop = tp->rcv_nxt - ti->ti_seq;
849
if (tiflags & TH_SYN) {
859
* Following if statement from Stevens, vol. 2, p. 960.
861
if (todrop > ti->ti_len
862
|| (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
864
* Any valid FIN must be to the left of the window.
865
* At this point the FIN must be a duplicate or out
866
* of sequence; drop it.
871
* Send an ACK to resynchronize and drop any data.
872
* But keep on processing for RST or ACK.
874
tp->t_flags |= TF_ACKNOW;
876
tcpstat.tcps_rcvduppack++;
877
tcpstat.tcps_rcvdupbyte += todrop;
879
tcpstat.tcps_rcvpartduppack++;
880
tcpstat.tcps_rcvpartdupbyte += todrop;
883
ti->ti_seq += todrop;
884
ti->ti_len -= todrop;
885
if (ti->ti_urp > todrop)
886
ti->ti_urp -= todrop;
893
* If new data are received on a connection after the
894
* user processes are gone, then RST the other end.
896
if ((so->so_state & SS_NOFDREF) &&
897
tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
899
tcpstat.tcps_rcvafterclose++;
904
* If segment ends after window, drop trailing data
905
* (and PUSH and FIN); if nothing left, just ACK.
907
todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
909
tcpstat.tcps_rcvpackafterwin++;
910
if (todrop >= ti->ti_len) {
911
tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
913
* If a new connection request is received
914
* while in TIME_WAIT, drop the old connection
915
* and start over if the sequence numbers
916
* are above the previous ones.
918
if (tiflags & TH_SYN &&
919
tp->t_state == TCPS_TIME_WAIT &&
920
SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
921
iss = tp->rcv_nxt + TCP_ISSINCR;
926
* If window is closed can only take segments at
927
* window edge, and have to drop data and PUSH from
928
* incoming segments. Continue processing, but
929
* remember to ack. Otherwise, drop segment
932
if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
933
tp->t_flags |= TF_ACKNOW;
934
tcpstat.tcps_rcvwinprobe++;
938
tcpstat.tcps_rcvbyteafterwin += todrop;
940
ti->ti_len -= todrop;
941
tiflags &= ~(TH_PUSH|TH_FIN);
945
* If last ACK falls within this segment's sequence numbers,
946
* record its timestamp.
948
/* if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
949
* SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
950
* ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
951
* tp->ts_recent_age = tcp_now;
952
* tp->ts_recent = ts_val;
957
* If the RST bit is set examine the state:
958
* SYN_RECEIVED STATE:
959
* If passive open, return to LISTEN state.
960
* If active open, inform user that connection was refused.
961
* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
962
* Inform user that connection was reset, and close tcb.
963
* CLOSING, LAST_ACK, TIME_WAIT STATES
966
if (tiflags&TH_RST) switch (tp->t_state) {
968
case TCPS_SYN_RECEIVED:
969
/* so->so_error = ECONNREFUSED; */
972
case TCPS_ESTABLISHED:
973
case TCPS_FIN_WAIT_1:
974
case TCPS_FIN_WAIT_2:
975
case TCPS_CLOSE_WAIT:
976
/* so->so_error = ECONNRESET; */
978
tp->t_state = TCPS_CLOSED;
979
tcpstat.tcps_drops++;
991
* If a SYN is in the window, then this is an
992
* error and we send an RST and drop the connection.
994
if (tiflags & TH_SYN) {
1000
* If the ACK bit is off we drop the segment and return.
1002
if ((tiflags & TH_ACK) == 0) goto drop;
1007
switch (tp->t_state) {
1009
* In SYN_RECEIVED state if the ack ACKs our SYN then enter
1010
* ESTABLISHED state and continue processing, otherwise
1011
* send an RST. una<=ack<=max
1013
case TCPS_SYN_RECEIVED:
1015
if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
1016
SEQ_GT(ti->ti_ack, tp->snd_max))
1018
tcpstat.tcps_connects++;
1019
tp->t_state = TCPS_ESTABLISHED;
1021
* The sent SYN is ack'ed with our sequence number +1
1022
* The first data byte already in the buffer will get
1023
* lost if no correction is made. This is only needed for
1024
* SS_CTL since the buffer is empty otherwise.
1025
* tp->snd_una++; or:
1027
tp->snd_una=ti->ti_ack;
1028
if (so->so_state & SS_CTL) {
1029
/* So tcp_ctl reports the right state */
1033
so->so_state &= ~SS_CTL; /* success XXX */
1034
} else if (ret == 2) {
1035
so->so_state = SS_NOFDREF; /* CTL_CMD */
1038
tp->t_state = TCPS_FIN_WAIT_1;
1044
/* Do window scaling? */
1045
/* if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1046
* (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1047
* tp->snd_scale = tp->requested_s_scale;
1048
* tp->rcv_scale = tp->request_r_scale;
1051
(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
1052
tp->snd_wl1 = ti->ti_seq - 1;
1053
/* Avoid ack processing; snd_una==ti_ack => dup ack */
1058
* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1059
* ACKs. If the ack is in the range
1060
* tp->snd_una < ti->ti_ack <= tp->snd_max
1061
* then advance tp->snd_una to ti->ti_ack and drop
1062
* data from the retransmission queue. If this ACK reflects
1063
* more up to date window information we update our window information.
1065
case TCPS_ESTABLISHED:
1066
case TCPS_FIN_WAIT_1:
1067
case TCPS_FIN_WAIT_2:
1068
case TCPS_CLOSE_WAIT:
1071
case TCPS_TIME_WAIT:
1073
if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1074
if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1075
tcpstat.tcps_rcvdupack++;
1076
DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
1077
(long )m, (long )so));
1079
* If we have outstanding data (other than
1080
* a window probe), this is a completely
1081
* duplicate ack (ie, window info didn't
1082
* change), the ack is the biggest we've
1083
* seen and we've seen exactly our rexmt
1084
* threshold of them, assume a packet
1085
* has been dropped and retransmit it.
1086
* Kludge snd_nxt & the congestion
1087
* window so we send only this one
1090
* We know we're losing at the current
1091
* window size so do congestion avoidance
1092
* (set ssthresh to half the current window
1093
* and pull our congestion window back to
1094
* the new ssthresh).
1096
* Dup acks mean that packets have left the
1097
* network (they're now cached at the receiver)
1098
* so bump cwnd by the amount in the receiver
1099
* to keep a constant cwnd packets in the
1102
if (tp->t_timer[TCPT_REXMT] == 0 ||
1103
ti->ti_ack != tp->snd_una)
1105
else if (++tp->t_dupacks == tcprexmtthresh) {
1106
tcp_seq onxt = tp->snd_nxt;
1108
min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1113
tp->snd_ssthresh = win * tp->t_maxseg;
1114
tp->t_timer[TCPT_REXMT] = 0;
1116
tp->snd_nxt = ti->ti_ack;
1117
tp->snd_cwnd = tp->t_maxseg;
1118
(void) tcp_output(tp);
1119
tp->snd_cwnd = tp->snd_ssthresh +
1120
tp->t_maxseg * tp->t_dupacks;
1121
if (SEQ_GT(onxt, tp->snd_nxt))
1124
} else if (tp->t_dupacks > tcprexmtthresh) {
1125
tp->snd_cwnd += tp->t_maxseg;
1126
(void) tcp_output(tp);
1135
* If the congestion window was inflated to account
1136
* for the other side's cached packets, retract it.
1138
if (tp->t_dupacks > tcprexmtthresh &&
1139
tp->snd_cwnd > tp->snd_ssthresh)
1140
tp->snd_cwnd = tp->snd_ssthresh;
1142
if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1143
tcpstat.tcps_rcvacktoomuch++;
1146
acked = ti->ti_ack - tp->snd_una;
1147
tcpstat.tcps_rcvackpack++;
1148
tcpstat.tcps_rcvackbyte += acked;
1151
* If we have a timestamp reply, update smoothed
1152
* round trip time. If no timestamp is present but
1153
* transmit timer is running and timed sequence
1154
* number was acked, update smoothed round trip time.
1155
* Since we now have an rtt measurement, cancel the
1156
* timer backoff (cf., Phil Karn's retransmit alg.).
1157
* Recompute the initial retransmit timer.
1160
* tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1163
if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1164
tcp_xmit_timer(tp,tp->t_rtt);
1167
* If all outstanding data is acked, stop retransmit
1168
* timer and remember to restart (more output or persist).
1169
* If there is more data to be acked, restart retransmit
1170
* timer, using current (possibly backed-off) value.
1172
if (ti->ti_ack == tp->snd_max) {
1173
tp->t_timer[TCPT_REXMT] = 0;
1175
} else if (tp->t_timer[TCPT_PERSIST] == 0)
1176
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1178
* When new data is acked, open the congestion window.
1179
* If the window gives us less than ssthresh packets
1180
* in flight, open exponentially (maxseg per packet).
1181
* Otherwise open linearly: maxseg per window
1182
* (maxseg^2 / cwnd per packet).
1185
register u_int cw = tp->snd_cwnd;
1186
register u_int incr = tp->t_maxseg;
1188
if (cw > tp->snd_ssthresh)
1189
incr = incr * incr / cw;
1190
tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1192
if (acked > so->so_snd.sb_cc) {
1193
tp->snd_wnd -= so->so_snd.sb_cc;
1194
sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1197
sbdrop(&so->so_snd, acked);
1198
tp->snd_wnd -= acked;
1202
* XXX sowwakup is called when data is acked and there's room for
1203
* for more data... it should read() the socket
1205
/* if (so->so_snd.sb_flags & SB_NOTIFY)
1208
tp->snd_una = ti->ti_ack;
1209
if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1210
tp->snd_nxt = tp->snd_una;
1212
switch (tp->t_state) {
1215
* In FIN_WAIT_1 STATE in addition to the processing
1216
* for the ESTABLISHED state if our FIN is now acknowledged
1217
* then enter FIN_WAIT_2.
1219
case TCPS_FIN_WAIT_1:
1220
if (ourfinisacked) {
1222
* If we can't receive any more
1223
* data, then closing user can proceed.
1224
* Starting the timer is contrary to the
1225
* specification, but if we don't get a FIN
1226
* we'll hang forever.
1228
if (so->so_state & SS_FCANTRCVMORE) {
1229
soisfdisconnected(so);
1230
tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1232
tp->t_state = TCPS_FIN_WAIT_2;
1237
* In CLOSING STATE in addition to the processing for
1238
* the ESTABLISHED state if the ACK acknowledges our FIN
1239
* then enter the TIME-WAIT state, otherwise ignore
1243
if (ourfinisacked) {
1244
tp->t_state = TCPS_TIME_WAIT;
1245
tcp_canceltimers(tp);
1246
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1247
soisfdisconnected(so);
1252
* In LAST_ACK, we may still be waiting for data to drain
1253
* and/or to be acked, as well as for the ack of our FIN.
1254
* If our FIN is now acknowledged, delete the TCB,
1255
* enter the closed state and return.
1258
if (ourfinisacked) {
1265
* In TIME_WAIT state the only thing that should arrive
1266
* is a retransmission of the remote FIN. Acknowledge
1267
* it and restart the finack timer.
1269
case TCPS_TIME_WAIT:
1270
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1273
} /* switch(tp->t_state) */
1277
* Update window information.
1278
* Don't look at window if no ACK: TAC's send garbage on first SYN.
1280
if ((tiflags & TH_ACK) &&
1281
(SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1282
(tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1283
(tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1284
/* keep track of pure window updates */
1285
if (ti->ti_len == 0 &&
1286
tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1287
tcpstat.tcps_rcvwinupd++;
1288
tp->snd_wnd = tiwin;
1289
tp->snd_wl1 = ti->ti_seq;
1290
tp->snd_wl2 = ti->ti_ack;
1291
if (tp->snd_wnd > tp->max_sndwnd)
1292
tp->max_sndwnd = tp->snd_wnd;
1297
* Process segments with URG.
1299
if ((tiflags & TH_URG) && ti->ti_urp &&
1300
TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1302
* This is a kludge, but if we receive and accept
1303
* random urgent pointers, we'll crash in
1304
* soreceive. It's hard to imagine someone
1305
* actually wanting to send this much urgent data.
1307
if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
1313
* If this segment advances the known urgent pointer,
1314
* then mark the data stream. This should not happen
1315
* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1316
* a FIN has been received from the remote side.
1317
* In these states we ignore the URG.
1319
* According to RFC961 (Assigned Protocols),
1320
* the urgent pointer points to the last octet
1321
* of urgent data. We continue, however,
1322
* to consider it to indicate the first octet
1323
* of data past the urgent section as the original
1324
* spec states (in one of two places).
1326
if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1327
tp->rcv_up = ti->ti_seq + ti->ti_urp;
1328
so->so_urgc = so->so_rcv.sb_cc +
1329
(tp->rcv_up - tp->rcv_nxt); /* -1; */
1330
tp->rcv_up = ti->ti_seq + ti->ti_urp;
1335
* If no out of band data is expected,
1336
* pull receive urgent pointer along
1337
* with the receive window.
1339
if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1340
tp->rcv_up = tp->rcv_nxt;
1344
* Process the segment text, merging it into the TCP sequencing queue,
1345
* and arranging for acknowledgment of receipt if necessary.
1346
* This process logically involves adjusting tp->rcv_wnd as data
1347
* is presented to the user (this happens in tcp_usrreq.c,
1348
* case PRU_RCVD). If a FIN has already been received on this
1349
* connection then we just ignore the text.
1351
if ((ti->ti_len || (tiflags&TH_FIN)) &&
1352
TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1353
TCP_REASS(tp, ti, m, so, tiflags);
1355
* Note the amount of data that peer has sent into
1356
* our window, in order to estimate the sender's
1359
len = so->so_rcv.sb_datalen - (tp->rcv_adv - tp->rcv_nxt);
1366
* If FIN is received ACK the FIN and let the user know
1367
* that the connection is closing.
1369
if (tiflags & TH_FIN) {
1370
if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1372
* If we receive a FIN we can't send more data,
1374
* Shutdown the socket if there is no rx data in the
1376
* soread() is called on completion of shutdown() and
1377
* will got to TCPS_LAST_ACK, and use tcp_output()
1380
/* sofcantrcvmore(so); */
1383
tp->t_flags |= TF_ACKNOW;
1386
switch (tp->t_state) {
1389
* In SYN_RECEIVED and ESTABLISHED STATES
1390
* enter the CLOSE_WAIT state.
1392
case TCPS_SYN_RECEIVED:
1393
case TCPS_ESTABLISHED:
1394
if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1395
tp->t_state = TCPS_LAST_ACK;
1397
tp->t_state = TCPS_CLOSE_WAIT;
1401
* If still in FIN_WAIT_1 STATE FIN has not been acked so
1402
* enter the CLOSING state.
1404
case TCPS_FIN_WAIT_1:
1405
tp->t_state = TCPS_CLOSING;
1409
* In FIN_WAIT_2 state enter the TIME_WAIT state,
1410
* starting the time-wait timer, turning off the other
1413
case TCPS_FIN_WAIT_2:
1414
tp->t_state = TCPS_TIME_WAIT;
1415
tcp_canceltimers(tp);
1416
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1417
soisfdisconnected(so);
1421
* In TIME_WAIT state restart the 2 MSL time_wait timer.
1423
case TCPS_TIME_WAIT:
1424
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1430
* If this is a small packet, then ACK now - with Nagel
1431
* congestion avoidance sender won't send more until
1436
/* if (ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg) {
1438
/* if ((ti->ti_len && (unsigned)ti->ti_len < tp->t_maxseg &&
1439
* (so->so_iptos & IPTOS_LOWDELAY) == 0) ||
1440
* ((so->so_iptos & IPTOS_LOWDELAY) &&
1441
* ((struct tcpiphdr_2 *)ti)->first_char == (char)27)) {
1443
if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
1444
((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
1445
tp->t_flags |= TF_ACKNOW;
1449
* Return any desired output.
1451
if (needoutput || (tp->t_flags & TF_ACKNOW)) {
1452
(void) tcp_output(tp);
1458
* Generate an ACK dropping incoming segment if it occupies
1459
* sequence space, where the ACK reflects our state.
1461
if (tiflags & TH_RST)
1464
tp->t_flags |= TF_ACKNOW;
1465
(void) tcp_output(tp);
1469
/* reuses m if m!=NULL, m_free() unnecessary */
1470
if (tiflags & TH_ACK)
1471
tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1473
if (tiflags & TH_SYN) ti->ti_len++;
1474
tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1482
* Drop space held by incoming segment and return.
1489
/* , ts_present, ts_val, ts_ecr) */
1491
* u_int32_t *ts_val, *ts_ecr;
1494
tcp_dooptions(tp, cp, cnt, ti)
1498
struct tcpiphdr *ti;
1503
DEBUG_CALL("tcp_dooptions");
1504
DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1506
for (; cnt > 0; cnt -= optlen, cp += optlen) {
1508
if (opt == TCPOPT_EOL)
1510
if (opt == TCPOPT_NOP)
1523
if (optlen != TCPOLEN_MAXSEG)
1525
if (!(ti->ti_flags & TH_SYN))
1527
memcpy((char *) &mss, (char *) cp + 2, sizeof(mss));
1529
(void) tcp_mss(tp, mss); /* sets t_maxseg */
1532
/* case TCPOPT_WINDOW:
1533
* if (optlen != TCPOLEN_WINDOW)
1535
* if (!(ti->ti_flags & TH_SYN))
1537
* tp->t_flags |= TF_RCVD_SCALE;
1538
* tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1541
/* case TCPOPT_TIMESTAMP:
1542
* if (optlen != TCPOLEN_TIMESTAMP)
1545
* memcpy((char *) ts_val, (char *)cp + 2, sizeof(*ts_val));
1547
* memcpy((char *) ts_ecr, (char *)cp + 6, sizeof(*ts_ecr));
1551
* * A timestamp received in a SYN makes
1552
* * it ok to send timestamp requests and replies.
1554
/* if (ti->ti_flags & TH_SYN) {
1555
* tp->t_flags |= TF_RCVD_TSTMP;
1556
* tp->ts_recent = *ts_val;
1557
* tp->ts_recent_age = tcp_now;
1566
* Pull out of band byte out of a segment so
1567
* it doesn't appear in the user's data queue.
1568
* It is still reflected in the segment length for
1569
* sequencing purposes.
1575
tcp_pulloutofband(so, ti, m)
1577
struct tcpiphdr *ti;
1578
register struct mbuf *m;
1580
int cnt = ti->ti_urp - 1;
1583
if (m->m_len > cnt) {
1584
char *cp = mtod(m, caddr_t) + cnt;
1585
struct tcpcb *tp = sototcpcb(so);
1588
tp->t_oobflags |= TCPOOB_HAVEDATA;
1589
memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1594
m = m->m_next; /* XXX WRONG! Fix it! */
1598
panic("tcp_pulloutofband");
1604
* Collect new round-trip time estimate
1605
* and update averages and current timeout.
1609
tcp_xmit_timer(tp, rtt)
1610
register struct tcpcb *tp;
1613
register short delta;
1615
DEBUG_CALL("tcp_xmit_timer");
1616
DEBUG_ARG("tp = %lx", (long)tp);
1617
DEBUG_ARG("rtt = %d", rtt);
1619
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.
1700
register struct tcpcb *tp;
1703
struct socket *so = tp->t_socket;
1706
DEBUG_CALL("tcp_mss");
1707
DEBUG_ARG("tp = %lx", (long)tp);
1708
DEBUG_ARG("offer = %d", offer);
1710
mss = min(if_mtu, if_mru) - sizeof(struct tcpiphdr);
1712
mss = min(mss, offer);
1714
if (mss < tp->t_maxseg || offer != 0)
1719
sbreserve(&so->so_snd, tcp_sndspace+((tcp_sndspace%mss)?(mss-(tcp_sndspace%mss)):0));
1720
sbreserve(&so->so_rcv, tcp_rcvspace+((tcp_rcvspace%mss)?(mss-(tcp_rcvspace%mss)):0));
1722
DEBUG_MISC((dfd, " returning mss = %d\n", mss));