Replace the VMSTOP macros with a proper state type
[qemu.git] / slirp / tcp_input.c
blobc1214c06591417cd5ea8b85372d625422fa56d3b
1 /*
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
7 * are met:
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
27 * SUCH DAMAGE.
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.
41 #include <slirp.h>
42 #include "ip_icmp.h"
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).
62 #ifdef TCP_ACK_HACK
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; \
69 else \
70 tp->t_flags |= TF_DELACK; \
71 (tp)->rcv_nxt += (ti)->ti_len; \
72 flags = (ti)->ti_flags & TH_FIN; \
73 if (so->so_emu) { \
74 if (tcp_emu((so),(m))) sbappend((so), (m)); \
75 } else \
76 sbappend((so), (m)); \
77 } else {\
78 (flags) = tcp_reass((tp), (ti), (m)); \
79 tp->t_flags |= TF_ACKNOW; \
80 } \
82 #else
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; \
90 if (so->so_emu) { \
91 if (tcp_emu((so),(m))) sbappend(so, (m)); \
92 } else \
93 sbappend((so), (m)); \
94 } else { \
95 (flags) = tcp_reass((tp), (ti), (m)); \
96 tp->t_flags |= TF_ACKNOW; \
97 } \
99 #endif
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);
104 static int
105 tcp_reass(register struct tcpcb *tp, register struct tcpiphdr *ti,
106 struct mbuf *m)
108 register struct tcpiphdr *q;
109 struct socket *so = tp->t_socket;
110 int flags;
113 * Call with ti==NULL after become established to
114 * force pre-ESTABLISHED data up to user socket.
116 if (ti == NULL)
117 goto present;
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))
125 break;
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)) {
133 register int i;
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;
137 if (i > 0) {
138 if (i >= ti->ti_len) {
139 m_free(m);
141 * Try to present any queued data
142 * at the left window edge to the user.
143 * This is needed after the 3-WHS
144 * completes.
146 goto present; /* ??? */
148 m_adj(m, i);
149 ti->ti_len -= i;
150 ti->ti_seq += i;
152 q = tcpiphdr_next(q);
154 ti->ti_mbuf = m;
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;
162 if (i <= 0)
163 break;
164 if (i < q->ti_len) {
165 q->ti_seq += i;
166 q->ti_len -= i;
167 m_adj(q->ti_mbuf, i);
168 break;
170 q = tcpiphdr_next(q);
171 m = tcpiphdr_prev(q)->ti_mbuf;
172 remque(tcpiphdr2qlink(tcpiphdr_prev(q)));
173 m_free(m);
177 * Stick new segment in its place.
179 insque(tcpiphdr2qlink(ti), tcpiphdr2qlink(tcpiphdr_prev(q)));
181 present:
183 * Present data to user, advancing rcv_nxt through
184 * completed sequence space.
186 if (!TCPS_HAVEESTABLISHED(tp->t_state))
187 return (0);
188 ti = tcpfrag_list_first(tp);
189 if (tcpfrag_list_end(ti, tp) || ti->ti_seq != tp->rcv_nxt)
190 return (0);
191 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
192 return (0);
193 do {
194 tp->rcv_nxt += ti->ti_len;
195 flags = ti->ti_flags & TH_FIN;
196 remque(tcpiphdr2qlink(ti));
197 m = ti->ti_mbuf;
198 ti = tcpiphdr_next(ti);
199 if (so->so_state & SS_FCANTSENDMORE)
200 m_free(m);
201 else {
202 if (so->so_emu) {
203 if (tcp_emu(so,m)) sbappend(so, m);
204 } else
205 sbappend(so, m);
207 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
208 return (flags);
212 * TCP input routine, follows pages 65-76 of the
213 * protocol specification dated September, 1981 very closely.
215 void
216 tcp_input(struct mbuf *m, int iphlen, struct socket *inso)
218 struct ip save_ip, *ip;
219 register struct tcpiphdr *ti;
220 caddr_t optp = NULL;
221 int optlen = 0;
222 int len, tlen, off;
223 register struct tcpcb *tp = NULL;
224 register int tiflags;
225 struct socket *so = NULL;
226 int todrop, acked, ourfinisacked, needoutput = 0;
227 int iss = 0;
228 u_long tiwin;
229 int ret;
230 struct ex_list *ex_ptr;
231 Slirp *slirp;
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
240 if (m == NULL) {
241 so = inso;
242 slirp = so->slirp;
244 /* Re-set a few variables */
245 tp = sototcpcb(so);
246 m = so->so_m;
247 so->so_m = NULL;
248 ti = so->so_ti;
249 tiwin = ti->ti_win;
250 tiflags = ti->ti_flags;
252 goto cont_conn;
254 slirp = m->slirp;
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 *);
273 save_ip = *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));
282 ti->ti_x1 = 0;
283 ti->ti_len = htons((uint16_t)tlen);
284 len = sizeof(struct ip ) + tlen;
285 if(cksum(m, len)) {
286 goto drop;
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) {
295 goto drop;
297 tlen -= off;
298 ti->ti_len = 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.
308 NTOHL(ti->ti_seq);
309 NTOHL(ti->ti_ack);
310 NTOHS(ti->ti_win);
311 NTOHS(ti->ti_urp);
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) {
323 break;
326 if (!ex_ptr)
327 goto drop;
330 * Locate pcb for segment.
332 findso:
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);
340 if (so)
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...
357 if (so == NULL) {
358 if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
359 goto dropwithreset;
361 if ((so = socreate(slirp)) == NULL)
362 goto dropwithreset;
363 if (tcp_attach(so) < 0) {
364 free(so); /* Not sofree (if it failed, it's not insqued) */
365 goto dropwithreset;
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;
379 tp = sototcpcb(so);
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)
389 goto drop;
391 tp = sototcpcb(so);
393 /* XXX Should never fail */
394 if (tp == NULL)
395 goto dropwithreset;
396 if (tp->t_state == TCPS_CLOSED)
397 goto drop;
399 tiwin = ti->ti_win;
402 * Segment received on connection.
403 * Reset idle time and keep-alive timer.
405 tp->t_idle = 0;
406 if (SO_OPTIONS)
407 tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
408 else
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.
448 if (tp->t_rtt &&
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;
454 m_free(m);
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);
478 return;
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.
492 if (so->so_emu) {
493 if (tcp_emu(so,m)) sbappend(so, m);
494 } else
495 sbappend(so, m);
498 * If this is a short packet, then ACK now - with Nagel
499 * congestion avoidance sender won't send more until
500 * he gets an ACK.
502 * It is better to not delay acks at all to maximize
503 * TCP throughput. See RFC 2581.
505 tp->t_flags |= TF_ACKNOW;
506 tcp_output(tp);
507 return;
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.
516 { int win;
517 win = sbspace(&so->so_rcv);
518 if (win < 0)
519 win = 0;
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.
538 case TCPS_LISTEN: {
540 if (tiflags & TH_RST)
541 goto drop;
542 if (tiflags & TH_ACK)
543 goto dropwithreset;
544 if ((tiflags & TH_SYN) == 0)
545 goto drop;
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;
566 break;
569 if (so->so_state & SS_CTL) {
570 goto cont_input;
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;
578 goto cont_input;
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,
588 TH_RST|TH_ACK);
589 } else {
590 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
591 HTONL(ti->ti_seq); /* restore tcp header */
592 HTONL(ti->ti_ack);
593 HTONS(ti->ti_win);
594 HTONS(ti->ti_urp);
595 m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
596 m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
597 *ip=save_ip;
598 icmp_error(m, ICMP_UNREACH,code, 0,strerror(errno));
600 tcp_close(tp);
601 m_free(m);
602 } else {
604 * Haven't connected yet, save the current mbuf
605 * and ti, and return
606 * XXX Some OS's don't tell us whether the connect()
607 * succeeded or not. So we must time it out.
609 so->so_m = m;
610 so->so_ti = ti;
611 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
612 tp->t_state = TCPS_SYN_RECEIVED;
614 return;
616 cont_conn:
617 /* m==NULL
618 * Check if the connect succeeded
620 if (so->so_state & SS_NOFDREF) {
621 tp = tcp_close(tp);
622 goto dropwithreset;
624 cont_input:
625 tcp_template(tp);
627 if (optp)
628 tcp_dooptions(tp, (u_char *)optp, optlen, ti);
630 if (iss)
631 tp->iss = iss;
632 else
633 tp->iss = slirp->tcp_iss;
634 slirp->tcp_iss += TCP_ISSINCR/2;
635 tp->irs = ti->ti_seq;
636 tcp_sendseqinit(tp);
637 tcp_rcvseqinit(tp);
638 tp->t_flags |= TF_ACKNOW;
639 tp->t_state = TCPS_SYN_RECEIVED;
640 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
641 goto trimthenstep6;
642 } /* case TCPS_LISTEN */
645 * If the state is SYN_SENT:
646 * if seg contains an ACK, but not for our SYN, drop the input.
647 * if seg contains a RST, then drop the connection.
648 * if seg does not contain SYN, then drop it.
649 * Otherwise this is an acceptable SYN segment
650 * initialize tp->rcv_nxt and tp->irs
651 * if seg contains ack then advance tp->snd_una
652 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
653 * arrange for segment to be acked (eventually)
654 * continue processing rest of data/controls, beginning with URG
656 case TCPS_SYN_SENT:
657 if ((tiflags & TH_ACK) &&
658 (SEQ_LEQ(ti->ti_ack, tp->iss) ||
659 SEQ_GT(ti->ti_ack, tp->snd_max)))
660 goto dropwithreset;
662 if (tiflags & TH_RST) {
663 if (tiflags & TH_ACK) {
664 tcp_drop(tp, 0); /* XXX Check t_softerror! */
666 goto drop;
669 if ((tiflags & TH_SYN) == 0)
670 goto drop;
671 if (tiflags & TH_ACK) {
672 tp->snd_una = ti->ti_ack;
673 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
674 tp->snd_nxt = tp->snd_una;
677 tp->t_timer[TCPT_REXMT] = 0;
678 tp->irs = ti->ti_seq;
679 tcp_rcvseqinit(tp);
680 tp->t_flags |= TF_ACKNOW;
681 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
682 soisfconnected(so);
683 tp->t_state = TCPS_ESTABLISHED;
685 (void) tcp_reass(tp, (struct tcpiphdr *)0,
686 (struct mbuf *)0);
688 * if we didn't have to retransmit the SYN,
689 * use its rtt as our initial srtt & rtt var.
691 if (tp->t_rtt)
692 tcp_xmit_timer(tp, tp->t_rtt);
693 } else
694 tp->t_state = TCPS_SYN_RECEIVED;
696 trimthenstep6:
698 * Advance ti->ti_seq to correspond to first data byte.
699 * If data, trim to stay within window,
700 * dropping FIN if necessary.
702 ti->ti_seq++;
703 if (ti->ti_len > tp->rcv_wnd) {
704 todrop = ti->ti_len - tp->rcv_wnd;
705 m_adj(m, -todrop);
706 ti->ti_len = tp->rcv_wnd;
707 tiflags &= ~TH_FIN;
709 tp->snd_wl1 = ti->ti_seq - 1;
710 tp->rcv_up = ti->ti_seq;
711 goto step6;
712 } /* switch tp->t_state */
714 * States other than LISTEN or SYN_SENT.
715 * Check that at least some bytes of segment are within
716 * receive window. If segment begins before rcv_nxt,
717 * drop leading data (and SYN); if nothing left, just ack.
719 todrop = tp->rcv_nxt - ti->ti_seq;
720 if (todrop > 0) {
721 if (tiflags & TH_SYN) {
722 tiflags &= ~TH_SYN;
723 ti->ti_seq++;
724 if (ti->ti_urp > 1)
725 ti->ti_urp--;
726 else
727 tiflags &= ~TH_URG;
728 todrop--;
731 * Following if statement from Stevens, vol. 2, p. 960.
733 if (todrop > ti->ti_len
734 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
736 * Any valid FIN must be to the left of the window.
737 * At this point the FIN must be a duplicate or out
738 * of sequence; drop it.
740 tiflags &= ~TH_FIN;
743 * Send an ACK to resynchronize and drop any data.
744 * But keep on processing for RST or ACK.
746 tp->t_flags |= TF_ACKNOW;
747 todrop = ti->ti_len;
749 m_adj(m, todrop);
750 ti->ti_seq += todrop;
751 ti->ti_len -= todrop;
752 if (ti->ti_urp > todrop)
753 ti->ti_urp -= todrop;
754 else {
755 tiflags &= ~TH_URG;
756 ti->ti_urp = 0;
760 * If new data are received on a connection after the
761 * user processes are gone, then RST the other end.
763 if ((so->so_state & SS_NOFDREF) &&
764 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
765 tp = tcp_close(tp);
766 goto dropwithreset;
770 * If segment ends after window, drop trailing data
771 * (and PUSH and FIN); if nothing left, just ACK.
773 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
774 if (todrop > 0) {
775 if (todrop >= ti->ti_len) {
777 * If a new connection request is received
778 * while in TIME_WAIT, drop the old connection
779 * and start over if the sequence numbers
780 * are above the previous ones.
782 if (tiflags & TH_SYN &&
783 tp->t_state == TCPS_TIME_WAIT &&
784 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
785 iss = tp->rcv_nxt + TCP_ISSINCR;
786 tp = tcp_close(tp);
787 goto findso;
790 * If window is closed can only take segments at
791 * window edge, and have to drop data and PUSH from
792 * incoming segments. Continue processing, but
793 * remember to ack. Otherwise, drop segment
794 * and ack.
796 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
797 tp->t_flags |= TF_ACKNOW;
798 } else {
799 goto dropafterack;
802 m_adj(m, -todrop);
803 ti->ti_len -= todrop;
804 tiflags &= ~(TH_PUSH|TH_FIN);
808 * If the RST bit is set examine the state:
809 * SYN_RECEIVED STATE:
810 * If passive open, return to LISTEN state.
811 * If active open, inform user that connection was refused.
812 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
813 * Inform user that connection was reset, and close tcb.
814 * CLOSING, LAST_ACK, TIME_WAIT STATES
815 * Close the tcb.
817 if (tiflags&TH_RST) switch (tp->t_state) {
819 case TCPS_SYN_RECEIVED:
820 case TCPS_ESTABLISHED:
821 case TCPS_FIN_WAIT_1:
822 case TCPS_FIN_WAIT_2:
823 case TCPS_CLOSE_WAIT:
824 tp->t_state = TCPS_CLOSED;
825 tcp_close(tp);
826 goto drop;
828 case TCPS_CLOSING:
829 case TCPS_LAST_ACK:
830 case TCPS_TIME_WAIT:
831 tcp_close(tp);
832 goto drop;
836 * If a SYN is in the window, then this is an
837 * error and we send an RST and drop the connection.
839 if (tiflags & TH_SYN) {
840 tp = tcp_drop(tp,0);
841 goto dropwithreset;
845 * If the ACK bit is off we drop the segment and return.
847 if ((tiflags & TH_ACK) == 0) goto drop;
850 * Ack processing.
852 switch (tp->t_state) {
854 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
855 * ESTABLISHED state and continue processing, otherwise
856 * send an RST. una<=ack<=max
858 case TCPS_SYN_RECEIVED:
860 if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
861 SEQ_GT(ti->ti_ack, tp->snd_max))
862 goto dropwithreset;
863 tp->t_state = TCPS_ESTABLISHED;
865 * The sent SYN is ack'ed with our sequence number +1
866 * The first data byte already in the buffer will get
867 * lost if no correction is made. This is only needed for
868 * SS_CTL since the buffer is empty otherwise.
869 * tp->snd_una++; or:
871 tp->snd_una=ti->ti_ack;
872 if (so->so_state & SS_CTL) {
873 /* So tcp_ctl reports the right state */
874 ret = tcp_ctl(so);
875 if (ret == 1) {
876 soisfconnected(so);
877 so->so_state &= ~SS_CTL; /* success XXX */
878 } else if (ret == 2) {
879 so->so_state &= SS_PERSISTENT_MASK;
880 so->so_state |= SS_NOFDREF; /* CTL_CMD */
881 } else {
882 needoutput = 1;
883 tp->t_state = TCPS_FIN_WAIT_1;
885 } else {
886 soisfconnected(so);
889 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
890 tp->snd_wl1 = ti->ti_seq - 1;
891 /* Avoid ack processing; snd_una==ti_ack => dup ack */
892 goto synrx_to_est;
893 /* fall into ... */
896 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
897 * ACKs. If the ack is in the range
898 * tp->snd_una < ti->ti_ack <= tp->snd_max
899 * then advance tp->snd_una to ti->ti_ack and drop
900 * data from the retransmission queue. If this ACK reflects
901 * more up to date window information we update our window information.
903 case TCPS_ESTABLISHED:
904 case TCPS_FIN_WAIT_1:
905 case TCPS_FIN_WAIT_2:
906 case TCPS_CLOSE_WAIT:
907 case TCPS_CLOSING:
908 case TCPS_LAST_ACK:
909 case TCPS_TIME_WAIT:
911 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
912 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
913 DEBUG_MISC((dfd," dup ack m = %lx so = %lx \n",
914 (long )m, (long )so));
916 * If we have outstanding data (other than
917 * a window probe), this is a completely
918 * duplicate ack (ie, window info didn't
919 * change), the ack is the biggest we've
920 * seen and we've seen exactly our rexmt
921 * threshold of them, assume a packet
922 * has been dropped and retransmit it.
923 * Kludge snd_nxt & the congestion
924 * window so we send only this one
925 * packet.
927 * We know we're losing at the current
928 * window size so do congestion avoidance
929 * (set ssthresh to half the current window
930 * and pull our congestion window back to
931 * the new ssthresh).
933 * Dup acks mean that packets have left the
934 * network (they're now cached at the receiver)
935 * so bump cwnd by the amount in the receiver
936 * to keep a constant cwnd packets in the
937 * network.
939 if (tp->t_timer[TCPT_REXMT] == 0 ||
940 ti->ti_ack != tp->snd_una)
941 tp->t_dupacks = 0;
942 else if (++tp->t_dupacks == TCPREXMTTHRESH) {
943 tcp_seq onxt = tp->snd_nxt;
944 u_int win =
945 min(tp->snd_wnd, tp->snd_cwnd) / 2 /
946 tp->t_maxseg;
948 if (win < 2)
949 win = 2;
950 tp->snd_ssthresh = win * tp->t_maxseg;
951 tp->t_timer[TCPT_REXMT] = 0;
952 tp->t_rtt = 0;
953 tp->snd_nxt = ti->ti_ack;
954 tp->snd_cwnd = tp->t_maxseg;
955 (void) tcp_output(tp);
956 tp->snd_cwnd = tp->snd_ssthresh +
957 tp->t_maxseg * tp->t_dupacks;
958 if (SEQ_GT(onxt, tp->snd_nxt))
959 tp->snd_nxt = onxt;
960 goto drop;
961 } else if (tp->t_dupacks > TCPREXMTTHRESH) {
962 tp->snd_cwnd += tp->t_maxseg;
963 (void) tcp_output(tp);
964 goto drop;
966 } else
967 tp->t_dupacks = 0;
968 break;
970 synrx_to_est:
972 * If the congestion window was inflated to account
973 * for the other side's cached packets, retract it.
975 if (tp->t_dupacks > TCPREXMTTHRESH &&
976 tp->snd_cwnd > tp->snd_ssthresh)
977 tp->snd_cwnd = tp->snd_ssthresh;
978 tp->t_dupacks = 0;
979 if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
980 goto dropafterack;
982 acked = ti->ti_ack - tp->snd_una;
985 * If transmit timer is running and timed sequence
986 * number was acked, update smoothed round trip time.
987 * Since we now have an rtt measurement, cancel the
988 * timer backoff (cf., Phil Karn's retransmit alg.).
989 * Recompute the initial retransmit timer.
991 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
992 tcp_xmit_timer(tp,tp->t_rtt);
995 * If all outstanding data is acked, stop retransmit
996 * timer and remember to restart (more output or persist).
997 * If there is more data to be acked, restart retransmit
998 * timer, using current (possibly backed-off) value.
1000 if (ti->ti_ack == tp->snd_max) {
1001 tp->t_timer[TCPT_REXMT] = 0;
1002 needoutput = 1;
1003 } else if (tp->t_timer[TCPT_PERSIST] == 0)
1004 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1006 * When new data is acked, open the congestion window.
1007 * If the window gives us less than ssthresh packets
1008 * in flight, open exponentially (maxseg per packet).
1009 * Otherwise open linearly: maxseg per window
1010 * (maxseg^2 / cwnd per packet).
1013 register u_int cw = tp->snd_cwnd;
1014 register u_int incr = tp->t_maxseg;
1016 if (cw > tp->snd_ssthresh)
1017 incr = incr * incr / cw;
1018 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1020 if (acked > so->so_snd.sb_cc) {
1021 tp->snd_wnd -= so->so_snd.sb_cc;
1022 sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1023 ourfinisacked = 1;
1024 } else {
1025 sbdrop(&so->so_snd, acked);
1026 tp->snd_wnd -= acked;
1027 ourfinisacked = 0;
1029 tp->snd_una = ti->ti_ack;
1030 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1031 tp->snd_nxt = tp->snd_una;
1033 switch (tp->t_state) {
1036 * In FIN_WAIT_1 STATE in addition to the processing
1037 * for the ESTABLISHED state if our FIN is now acknowledged
1038 * then enter FIN_WAIT_2.
1040 case TCPS_FIN_WAIT_1:
1041 if (ourfinisacked) {
1043 * If we can't receive any more
1044 * data, then closing user can proceed.
1045 * Starting the timer is contrary to the
1046 * specification, but if we don't get a FIN
1047 * we'll hang forever.
1049 if (so->so_state & SS_FCANTRCVMORE) {
1050 tp->t_timer[TCPT_2MSL] = TCP_MAXIDLE;
1052 tp->t_state = TCPS_FIN_WAIT_2;
1054 break;
1057 * In CLOSING STATE in addition to the processing for
1058 * the ESTABLISHED state if the ACK acknowledges our FIN
1059 * then enter the TIME-WAIT state, otherwise ignore
1060 * the segment.
1062 case TCPS_CLOSING:
1063 if (ourfinisacked) {
1064 tp->t_state = TCPS_TIME_WAIT;
1065 tcp_canceltimers(tp);
1066 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1068 break;
1071 * In LAST_ACK, we may still be waiting for data to drain
1072 * and/or to be acked, as well as for the ack of our FIN.
1073 * If our FIN is now acknowledged, delete the TCB,
1074 * enter the closed state and return.
1076 case TCPS_LAST_ACK:
1077 if (ourfinisacked) {
1078 tcp_close(tp);
1079 goto drop;
1081 break;
1084 * In TIME_WAIT state the only thing that should arrive
1085 * is a retransmission of the remote FIN. Acknowledge
1086 * it and restart the finack timer.
1088 case TCPS_TIME_WAIT:
1089 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1090 goto dropafterack;
1092 } /* switch(tp->t_state) */
1094 step6:
1096 * Update window information.
1097 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1099 if ((tiflags & TH_ACK) &&
1100 (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1101 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1102 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1103 tp->snd_wnd = tiwin;
1104 tp->snd_wl1 = ti->ti_seq;
1105 tp->snd_wl2 = ti->ti_ack;
1106 if (tp->snd_wnd > tp->max_sndwnd)
1107 tp->max_sndwnd = tp->snd_wnd;
1108 needoutput = 1;
1112 * Process segments with URG.
1114 if ((tiflags & TH_URG) && ti->ti_urp &&
1115 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1117 * This is a kludge, but if we receive and accept
1118 * random urgent pointers, we'll crash in
1119 * soreceive. It's hard to imagine someone
1120 * actually wanting to send this much urgent data.
1122 if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
1123 ti->ti_urp = 0;
1124 tiflags &= ~TH_URG;
1125 goto dodata;
1128 * If this segment advances the known urgent pointer,
1129 * then mark the data stream. This should not happen
1130 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1131 * a FIN has been received from the remote side.
1132 * In these states we ignore the URG.
1134 * According to RFC961 (Assigned Protocols),
1135 * the urgent pointer points to the last octet
1136 * of urgent data. We continue, however,
1137 * to consider it to indicate the first octet
1138 * of data past the urgent section as the original
1139 * spec states (in one of two places).
1141 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1142 tp->rcv_up = ti->ti_seq + ti->ti_urp;
1143 so->so_urgc = so->so_rcv.sb_cc +
1144 (tp->rcv_up - tp->rcv_nxt); /* -1; */
1145 tp->rcv_up = ti->ti_seq + ti->ti_urp;
1148 } else
1150 * If no out of band data is expected,
1151 * pull receive urgent pointer along
1152 * with the receive window.
1154 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1155 tp->rcv_up = tp->rcv_nxt;
1156 dodata:
1159 * Process the segment text, merging it into the TCP sequencing queue,
1160 * and arranging for acknowledgment of receipt if necessary.
1161 * This process logically involves adjusting tp->rcv_wnd as data
1162 * is presented to the user (this happens in tcp_usrreq.c,
1163 * case PRU_RCVD). If a FIN has already been received on this
1164 * connection then we just ignore the text.
1166 if ((ti->ti_len || (tiflags&TH_FIN)) &&
1167 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1168 TCP_REASS(tp, ti, m, so, tiflags);
1169 } else {
1170 m_free(m);
1171 tiflags &= ~TH_FIN;
1175 * If FIN is received ACK the FIN and let the user know
1176 * that the connection is closing.
1178 if (tiflags & TH_FIN) {
1179 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1181 * If we receive a FIN we can't send more data,
1182 * set it SS_FDRAIN
1183 * Shutdown the socket if there is no rx data in the
1184 * buffer.
1185 * soread() is called on completion of shutdown() and
1186 * will got to TCPS_LAST_ACK, and use tcp_output()
1187 * to send the FIN.
1189 sofwdrain(so);
1191 tp->t_flags |= TF_ACKNOW;
1192 tp->rcv_nxt++;
1194 switch (tp->t_state) {
1197 * In SYN_RECEIVED and ESTABLISHED STATES
1198 * enter the CLOSE_WAIT state.
1200 case TCPS_SYN_RECEIVED:
1201 case TCPS_ESTABLISHED:
1202 if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1203 tp->t_state = TCPS_LAST_ACK;
1204 else
1205 tp->t_state = TCPS_CLOSE_WAIT;
1206 break;
1209 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1210 * enter the CLOSING state.
1212 case TCPS_FIN_WAIT_1:
1213 tp->t_state = TCPS_CLOSING;
1214 break;
1217 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1218 * starting the time-wait timer, turning off the other
1219 * standard timers.
1221 case TCPS_FIN_WAIT_2:
1222 tp->t_state = TCPS_TIME_WAIT;
1223 tcp_canceltimers(tp);
1224 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1225 break;
1228 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1230 case TCPS_TIME_WAIT:
1231 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1232 break;
1237 * If this is a small packet, then ACK now - with Nagel
1238 * congestion avoidance sender won't send more until
1239 * he gets an ACK.
1241 * See above.
1243 if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
1244 ((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
1245 tp->t_flags |= TF_ACKNOW;
1249 * Return any desired output.
1251 if (needoutput || (tp->t_flags & TF_ACKNOW)) {
1252 (void) tcp_output(tp);
1254 return;
1256 dropafterack:
1258 * Generate an ACK dropping incoming segment if it occupies
1259 * sequence space, where the ACK reflects our state.
1261 if (tiflags & TH_RST)
1262 goto drop;
1263 m_free(m);
1264 tp->t_flags |= TF_ACKNOW;
1265 (void) tcp_output(tp);
1266 return;
1268 dropwithreset:
1269 /* reuses m if m!=NULL, m_free() unnecessary */
1270 if (tiflags & TH_ACK)
1271 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1272 else {
1273 if (tiflags & TH_SYN) ti->ti_len++;
1274 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1275 TH_RST|TH_ACK);
1278 return;
1280 drop:
1282 * Drop space held by incoming segment and return.
1284 m_free(m);
1286 return;
1289 static void
1290 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcpiphdr *ti)
1292 uint16_t mss;
1293 int opt, optlen;
1295 DEBUG_CALL("tcp_dooptions");
1296 DEBUG_ARGS((dfd," tp = %lx cnt=%i \n", (long )tp, cnt));
1298 for (; cnt > 0; cnt -= optlen, cp += optlen) {
1299 opt = cp[0];
1300 if (opt == TCPOPT_EOL)
1301 break;
1302 if (opt == TCPOPT_NOP)
1303 optlen = 1;
1304 else {
1305 optlen = cp[1];
1306 if (optlen <= 0)
1307 break;
1309 switch (opt) {
1311 default:
1312 continue;
1314 case TCPOPT_MAXSEG:
1315 if (optlen != TCPOLEN_MAXSEG)
1316 continue;
1317 if (!(ti->ti_flags & TH_SYN))
1318 continue;
1319 memcpy((char *) &mss, (char *) cp + 2, sizeof(mss));
1320 NTOHS(mss);
1321 (void) tcp_mss(tp, mss); /* sets t_maxseg */
1322 break;
1329 * Pull out of band byte out of a segment so
1330 * it doesn't appear in the user's data queue.
1331 * It is still reflected in the segment length for
1332 * sequencing purposes.
1335 #ifdef notdef
1337 void
1338 tcp_pulloutofband(so, ti, m)
1339 struct socket *so;
1340 struct tcpiphdr *ti;
1341 register struct mbuf *m;
1343 int cnt = ti->ti_urp - 1;
1345 while (cnt >= 0) {
1346 if (m->m_len > cnt) {
1347 char *cp = mtod(m, caddr_t) + cnt;
1348 struct tcpcb *tp = sototcpcb(so);
1350 tp->t_iobc = *cp;
1351 tp->t_oobflags |= TCPOOB_HAVEDATA;
1352 memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1353 m->m_len--;
1354 return;
1356 cnt -= m->m_len;
1357 m = m->m_next; /* XXX WRONG! Fix it! */
1358 if (m == 0)
1359 break;
1361 panic("tcp_pulloutofband");
1364 #endif /* notdef */
1367 * Collect new round-trip time estimate
1368 * and update averages and current timeout.
1371 static void
1372 tcp_xmit_timer(register struct tcpcb *tp, int rtt)
1374 register short delta;
1376 DEBUG_CALL("tcp_xmit_timer");
1377 DEBUG_ARG("tp = %lx", (long)tp);
1378 DEBUG_ARG("rtt = %d", rtt);
1380 if (tp->t_srtt != 0) {
1382 * srtt is stored as fixed point with 3 bits after the
1383 * binary point (i.e., scaled by 8). The following magic
1384 * is equivalent to the smoothing algorithm in rfc793 with
1385 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1386 * point). Adjust rtt to origin 0.
1388 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1389 if ((tp->t_srtt += delta) <= 0)
1390 tp->t_srtt = 1;
1392 * We accumulate a smoothed rtt variance (actually, a
1393 * smoothed mean difference), then set the retransmit
1394 * timer to smoothed rtt + 4 times the smoothed variance.
1395 * rttvar is stored as fixed point with 2 bits after the
1396 * binary point (scaled by 4). The following is
1397 * equivalent to rfc793 smoothing with an alpha of .75
1398 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
1399 * rfc793's wired-in beta.
1401 if (delta < 0)
1402 delta = -delta;
1403 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1404 if ((tp->t_rttvar += delta) <= 0)
1405 tp->t_rttvar = 1;
1406 } else {
1408 * No rtt measurement yet - use the unsmoothed rtt.
1409 * Set the variance to half the rtt (so our first
1410 * retransmit happens at 3*rtt).
1412 tp->t_srtt = rtt << TCP_RTT_SHIFT;
1413 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1415 tp->t_rtt = 0;
1416 tp->t_rxtshift = 0;
1419 * the retransmit should happen at rtt + 4 * rttvar.
1420 * Because of the way we do the smoothing, srtt and rttvar
1421 * will each average +1/2 tick of bias. When we compute
1422 * the retransmit timer, we want 1/2 tick of rounding and
1423 * 1 extra tick because of +-1/2 tick uncertainty in the
1424 * firing of the timer. The bias will give us exactly the
1425 * 1.5 tick we need. But, because the bias is
1426 * statistical, we have to test that we don't drop below
1427 * the minimum feasible timer (which is 2 ticks).
1429 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1430 (short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1433 * We received an ack for a packet that wasn't retransmitted;
1434 * it is probably safe to discard any error indications we've
1435 * received recently. This isn't quite right, but close enough
1436 * for now (a route might have failed after we sent a segment,
1437 * and the return path might not be symmetrical).
1439 tp->t_softerror = 0;
1443 * Determine a reasonable value for maxseg size.
1444 * If the route is known, check route for mtu.
1445 * If none, use an mss that can be handled on the outgoing
1446 * interface without forcing IP to fragment; if bigger than
1447 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1448 * to utilize large mbufs. If no route is found, route has no mtu,
1449 * or the destination isn't local, use a default, hopefully conservative
1450 * size (usually 512 or the default IP max size, but no more than the mtu
1451 * of the interface), as we can't discover anything about intervening
1452 * gateways or networks. We also initialize the congestion/slow start
1453 * window to be a single segment if the destination isn't local.
1454 * While looking at the routing entry, we also initialize other path-dependent
1455 * parameters from pre-set or cached values in the routing entry.
1459 tcp_mss(struct tcpcb *tp, u_int offer)
1461 struct socket *so = tp->t_socket;
1462 int mss;
1464 DEBUG_CALL("tcp_mss");
1465 DEBUG_ARG("tp = %lx", (long)tp);
1466 DEBUG_ARG("offer = %d", offer);
1468 mss = min(IF_MTU, IF_MRU) - sizeof(struct tcpiphdr);
1469 if (offer)
1470 mss = min(mss, offer);
1471 mss = max(mss, 32);
1472 if (mss < tp->t_maxseg || offer != 0)
1473 tp->t_maxseg = mss;
1475 tp->snd_cwnd = mss;
1477 sbreserve(&so->so_snd, TCP_SNDSPACE + ((TCP_SNDSPACE % mss) ?
1478 (mss - (TCP_SNDSPACE % mss)) :
1479 0));
1480 sbreserve(&so->so_rcv, TCP_RCVSPACE + ((TCP_RCVSPACE % mss) ?
1481 (mss - (TCP_RCVSPACE % mss)) :
1482 0));
1484 DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1486 return mss;