acl: fix memory leak
[qemu.git] / slirp / tcp_input.c
1 /*
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 *
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.
16 *
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.
28 *
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
31 */
32
33 /*
34 * Changes and additions relating to SLiRP
35 * Copyright (c) 1995 Danny Gasparovski.
36 *
37 * Please read the file COPYRIGHT for the
38 * terms and conditions of the copyright.
39 */
40
41 #include <slirp.h>
42 #include "ip_icmp.h"
43
44 #define TCPREXMTTHRESH 3
45
46 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ)
47
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)
51
52 /*
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).
61 */
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 } \
81 }
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 } \
98 }
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);
103
104 static int
105 tcp_reass(register struct tcpcb *tp, register struct tcpiphdr *ti,
106 struct mbuf *m)
107 {
108 register struct tcpiphdr *q;
109 struct socket *so = tp->t_socket;
110 int flags;
111
112 /*
113 * Call with ti==NULL after become established to
114 * force pre-ESTABLISHED data up to user socket.
115 */
116 if (ti == NULL)
117 goto present;
118
119 /*
120 * Find a segment which begins after this one does.
121 */
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;
126
127 /*
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.
131 */
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);
140 /*
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.
145 */
146 goto present; /* ??? */
147 }
148 m_adj(m, i);
149 ti->ti_len -= i;
150 ti->ti_seq += i;
151 }
152 q = tcpiphdr_next(q);
153 }
154 ti->ti_mbuf = m;
155
156 /*
157 * While we overlap succeeding segments trim them or,
158 * if they are completely covered, dequeue them.
159 */
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;
169 }
170 q = tcpiphdr_next(q);
171 m = tcpiphdr_prev(q)->ti_mbuf;
172 remque(tcpiphdr2qlink(tcpiphdr_prev(q)));
173 m_free(m);
174 }
175
176 /*
177 * Stick new segment in its place.
178 */
179 insque(tcpiphdr2qlink(ti), tcpiphdr2qlink(tcpiphdr_prev(q)));
180
181 present:
182 /*
183 * Present data to user, advancing rcv_nxt through
184 * completed sequence space.
185 */
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);
206 }
207 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
208 return (flags);
209 }
210
211 /*
212 * TCP input routine, follows pages 65-76 of the
213 * protocol specification dated September, 1981 very closely.
214 */
215 void
216 tcp_input(struct mbuf *m, int iphlen, struct socket *inso)
217 {
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;
232
233 DEBUG_CALL("tcp_input");
234 DEBUG_ARGS((dfd, " m = %8lx iphlen = %2d inso = %lx\n",
235 (long )m, iphlen, (long )inso ));
236
237 /*
238 * If called with m == 0, then we're continuing the connect
239 */
240 if (m == NULL) {
241 so = inso;
242 slirp = so->slirp;
243
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;
251
252 goto cont_conn;
253 }
254 slirp = m->slirp;
255
256 /*
257 * Get IP and TCP header together in first mbuf.
258 * Note: IP leaves IP header in first mbuf.
259 */
260 ti = mtod(m, struct tcpiphdr *);
261 if (iphlen > sizeof(struct ip )) {
262 ip_stripoptions(m, (struct mbuf *)0);
263 iphlen=sizeof(struct ip );
264 }
265 /* XXX Check if too short */
266
267
268 /*
269 * Save a copy of the IP header in case we want restore it
270 * for sending an ICMP error message in response.
271 */
272 ip=mtod(m, struct ip *);
273 save_ip = *ip;
274 save_ip.ip_len+= iphlen;
275
276 /*
277 * Checksum extended TCP header and data.
278 */
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;
287 }
288
289 /*
290 * Check that TCP offset makes sense,
291 * pull out TCP options and adjust length. XXX
292 */
293 off = ti->ti_off << 2;
294 if (off < sizeof (struct tcphdr) || off > tlen) {
295 goto drop;
296 }
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);
302 }
303 tiflags = ti->ti_flags;
304
305 /*
306 * Convert TCP protocol specific fields to host format.
307 */
308 NTOHL(ti->ti_seq);
309 NTOHL(ti->ti_ack);
310 NTOHS(ti->ti_win);
311 NTOHS(ti->ti_urp);
312
313 /*
314 * Drop TCP, IP headers and TCP options.
315 */
316 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
317 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
318
319 /*
320 * Locate pcb for segment.
321 */
322 findso:
323 so = slirp->tcp_last_so;
324 if (so->so_fport != ti->ti_dport ||
325 so->so_lport != ti->ti_sport ||
326 so->so_laddr.s_addr != ti->ti_src.s_addr ||
327 so->so_faddr.s_addr != ti->ti_dst.s_addr) {
328 so = solookup(&slirp->tcb, ti->ti_src, ti->ti_sport,
329 ti->ti_dst, ti->ti_dport);
330 if (so)
331 slirp->tcp_last_so = so;
332 }
333
334 /*
335 * If the state is CLOSED (i.e., TCB does not exist) then
336 * all data in the incoming segment is discarded.
337 * If the TCB exists but is in CLOSED state, it is embryonic,
338 * but should either do a listen or a connect soon.
339 *
340 * state == CLOSED means we've done socreate() but haven't
341 * attached it to a protocol yet...
342 *
343 * XXX If a TCB does not exist, and the TH_SYN flag is
344 * the only flag set, then create a session, mark it
345 * as if it was LISTENING, and continue...
346 */
347 if (so == NULL) {
348 if (slirp->restricted) {
349 /* Any hostfwds will have an existing socket, so we only get here
350 * for non-hostfwd connections. These should be dropped, unless it
351 * happens to be a guestfwd.
352 */
353 for (ex_ptr = slirp->exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next) {
354 if (ex_ptr->ex_fport == ti->ti_dport &&
355 ti->ti_dst.s_addr == ex_ptr->ex_addr.s_addr) {
356 break;
357 }
358 }
359 if (!ex_ptr) {
360 goto dropwithreset;
361 }
362 }
363
364 if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
365 goto dropwithreset;
366
367 if ((so = socreate(slirp)) == NULL)
368 goto dropwithreset;
369 if (tcp_attach(so) < 0) {
370 free(so); /* Not sofree (if it failed, it's not insqued) */
371 goto dropwithreset;
372 }
373
374 sbreserve(&so->so_snd, TCP_SNDSPACE);
375 sbreserve(&so->so_rcv, TCP_RCVSPACE);
376
377 so->so_laddr = ti->ti_src;
378 so->so_lport = ti->ti_sport;
379 so->so_faddr = ti->ti_dst;
380 so->so_fport = ti->ti_dport;
381
382 if ((so->so_iptos = tcp_tos(so)) == 0)
383 so->so_iptos = ((struct ip *)ti)->ip_tos;
384
385 tp = sototcpcb(so);
386 tp->t_state = TCPS_LISTEN;
387 }
388
389 /*
390 * If this is a still-connecting socket, this probably
391 * a retransmit of the SYN. Whether it's a retransmit SYN
392 * or something else, we nuke it.
393 */
394 if (so->so_state & SS_ISFCONNECTING)
395 goto drop;
396
397 tp = sototcpcb(so);
398
399 /* XXX Should never fail */
400 if (tp == NULL)
401 goto dropwithreset;
402 if (tp->t_state == TCPS_CLOSED)
403 goto drop;
404
405 tiwin = ti->ti_win;
406
407 /*
408 * Segment received on connection.
409 * Reset idle time and keep-alive timer.
410 */
411 tp->t_idle = 0;
412 if (SO_OPTIONS)
413 tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
414 else
415 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
416
417 /*
418 * Process options if not in LISTEN state,
419 * else do it below (after getting remote address).
420 */
421 if (optp && tp->t_state != TCPS_LISTEN)
422 tcp_dooptions(tp, (u_char *)optp, optlen, ti);
423
424 /*
425 * Header prediction: check for the two common cases
426 * of a uni-directional data xfer. If the packet has
427 * no control flags, is in-sequence, the window didn't
428 * change and we're not retransmitting, it's a
429 * candidate. If the length is zero and the ack moved
430 * forward, we're the sender side of the xfer. Just
431 * free the data acked & wake any higher level process
432 * that was blocked waiting for space. If the length
433 * is non-zero and the ack didn't move, we're the
434 * receiver side. If we're getting packets in-order
435 * (the reassembly queue is empty), add the data to
436 * the socket buffer and note that we need a delayed ack.
437 *
438 * XXX Some of these tests are not needed
439 * eg: the tiwin == tp->snd_wnd prevents many more
440 * predictions.. with no *real* advantage..
441 */
442 if (tp->t_state == TCPS_ESTABLISHED &&
443 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
444 ti->ti_seq == tp->rcv_nxt &&
445 tiwin && tiwin == tp->snd_wnd &&
446 tp->snd_nxt == tp->snd_max) {
447 if (ti->ti_len == 0) {
448 if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
449 SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
450 tp->snd_cwnd >= tp->snd_wnd) {
451 /*
452 * this is a pure ack for outstanding data.
453 */
454 if (tp->t_rtt &&
455 SEQ_GT(ti->ti_ack, tp->t_rtseq))
456 tcp_xmit_timer(tp, tp->t_rtt);
457 acked = ti->ti_ack - tp->snd_una;
458 sbdrop(&so->so_snd, acked);
459 tp->snd_una = ti->ti_ack;
460 m_free(m);
461
462 /*
463 * If all outstanding data are acked, stop
464 * retransmit timer, otherwise restart timer
465 * using current (possibly backed-off) value.
466 * If process is waiting for space,
467 * wakeup/selwakeup/signal. If data
468 * are ready to send, let tcp_output
469 * decide between more output or persist.
470 */
471 if (tp->snd_una == tp->snd_max)
472 tp->t_timer[TCPT_REXMT] = 0;
473 else if (tp->t_timer[TCPT_PERSIST] == 0)
474 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
475
476 /*
477 * This is called because sowwakeup might have
478 * put data into so_snd. Since we don't so sowwakeup,
479 * we don't need this.. XXX???
480 */
481 if (so->so_snd.sb_cc)
482 (void) tcp_output(tp);
483
484 return;
485 }
486 } else if (ti->ti_ack == tp->snd_una &&
487 tcpfrag_list_empty(tp) &&
488 ti->ti_len <= sbspace(&so->so_rcv)) {
489 /*
490 * this is a pure, in-sequence data packet
491 * with nothing on the reassembly queue and
492 * we have enough buffer space to take it.
493 */
494 tp->rcv_nxt += ti->ti_len;
495 /*
496 * Add data to socket buffer.
497 */
498 if (so->so_emu) {
499 if (tcp_emu(so,m)) sbappend(so, m);
500 } else
501 sbappend(so, m);
502
503 /*
504 * If this is a short packet, then ACK now - with Nagel
505 * congestion avoidance sender won't send more until
506 * he gets an ACK.
507 *
508 * It is better to not delay acks at all to maximize
509 * TCP throughput. See RFC 2581.
510 */
511 tp->t_flags |= TF_ACKNOW;
512 tcp_output(tp);
513 return;
514 }
515 } /* header prediction */
516 /*
517 * Calculate amount of space in receive window,
518 * and then do TCP input processing.
519 * Receive window is amount of space in rcv queue,
520 * but not less than advertised window.
521 */
522 { int win;
523 win = sbspace(&so->so_rcv);
524 if (win < 0)
525 win = 0;
526 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
527 }
528
529 switch (tp->t_state) {
530
531 /*
532 * If the state is LISTEN then ignore segment if it contains an RST.
533 * If the segment contains an ACK then it is bad and send a RST.
534 * If it does not contain a SYN then it is not interesting; drop it.
535 * Don't bother responding if the destination was a broadcast.
536 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
537 * tp->iss, and send a segment:
538 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
539 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
540 * Fill in remote peer address fields if not previously specified.
541 * Enter SYN_RECEIVED state, and process any other fields of this
542 * segment in this state.
543 */
544 case TCPS_LISTEN: {
545
546 if (tiflags & TH_RST)
547 goto drop;
548 if (tiflags & TH_ACK)
549 goto dropwithreset;
550 if ((tiflags & TH_SYN) == 0)
551 goto drop;
552
553 /*
554 * This has way too many gotos...
555 * But a bit of spaghetti code never hurt anybody :)
556 */
557
558 /*
559 * If this is destined for the control address, then flag to
560 * tcp_ctl once connected, otherwise connect
561 */
562 if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
563 slirp->vnetwork_addr.s_addr) {
564 if (so->so_faddr.s_addr != slirp->vhost_addr.s_addr &&
565 so->so_faddr.s_addr != slirp->vnameserver_addr.s_addr) {
566 /* May be an add exec */
567 for (ex_ptr = slirp->exec_list; ex_ptr;
568 ex_ptr = ex_ptr->ex_next) {
569 if(ex_ptr->ex_fport == so->so_fport &&
570 so->so_faddr.s_addr == ex_ptr->ex_addr.s_addr) {
571 so->so_state |= SS_CTL;
572 break;
573 }
574 }
575 if (so->so_state & SS_CTL) {
576 goto cont_input;
577 }
578 }
579 /* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
580 }
581
582 if (so->so_emu & EMU_NOCONNECT) {
583 so->so_emu &= ~EMU_NOCONNECT;
584 goto cont_input;
585 }
586
587 if((tcp_fconnect(so) == -1) && (errno != EINPROGRESS) && (errno != EWOULDBLOCK)) {
588 u_char code=ICMP_UNREACH_NET;
589 DEBUG_MISC((dfd, " tcp fconnect errno = %d-%s\n",
590 errno,strerror(errno)));
591 if(errno == ECONNREFUSED) {
592 /* ACK the SYN, send RST to refuse the connection */
593 tcp_respond(tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
594 TH_RST|TH_ACK);
595 } else {
596 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
597 HTONL(ti->ti_seq); /* restore tcp header */
598 HTONL(ti->ti_ack);
599 HTONS(ti->ti_win);
600 HTONS(ti->ti_urp);
601 m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
602 m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
603 *ip=save_ip;
604 icmp_error(m, ICMP_UNREACH,code, 0,strerror(errno));
605 }
606 tcp_close(tp);
607 m_free(m);
608 } else {
609 /*
610 * Haven't connected yet, save the current mbuf
611 * and ti, and return
612 * XXX Some OS's don't tell us whether the connect()
613 * succeeded or not. So we must time it out.
614 */
615 so->so_m = m;
616 so->so_ti = ti;
617 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
618 tp->t_state = TCPS_SYN_RECEIVED;
619 tcp_template(tp);
620 }
621 return;
622
623 cont_conn:
624 /* m==NULL
625 * Check if the connect succeeded
626 */
627 if (so->so_state & SS_NOFDREF) {
628 tp = tcp_close(tp);
629 goto dropwithreset;
630 }
631 cont_input:
632 tcp_template(tp);
633
634 if (optp)
635 tcp_dooptions(tp, (u_char *)optp, optlen, ti);
636
637 if (iss)
638 tp->iss = iss;
639 else
640 tp->iss = slirp->tcp_iss;
641 slirp->tcp_iss += TCP_ISSINCR/2;
642 tp->irs = ti->ti_seq;
643 tcp_sendseqinit(tp);
644 tcp_rcvseqinit(tp);
645 tp->t_flags |= TF_ACKNOW;
646 tp->t_state = TCPS_SYN_RECEIVED;
647 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
648 goto trimthenstep6;
649 } /* case TCPS_LISTEN */
650
651 /*
652 * If the state is SYN_SENT:
653 * if seg contains an ACK, but not for our SYN, drop the input.
654 * if seg contains a RST, then drop the connection.
655 * if seg does not contain SYN, then drop it.
656 * Otherwise this is an acceptable SYN segment
657 * initialize tp->rcv_nxt and tp->irs
658 * if seg contains ack then advance tp->snd_una
659 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
660 * arrange for segment to be acked (eventually)
661 * continue processing rest of data/controls, beginning with URG
662 */
663 case TCPS_SYN_SENT:
664 if ((tiflags & TH_ACK) &&
665 (SEQ_LEQ(ti->ti_ack, tp->iss) ||
666 SEQ_GT(ti->ti_ack, tp->snd_max)))
667 goto dropwithreset;
668
669 if (tiflags & TH_RST) {
670 if (tiflags & TH_ACK) {
671 tcp_drop(tp, 0); /* XXX Check t_softerror! */
672 }
673 goto drop;
674 }
675
676 if ((tiflags & TH_SYN) == 0)
677 goto drop;
678 if (tiflags & TH_ACK) {
679 tp->snd_una = ti->ti_ack;
680 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
681 tp->snd_nxt = tp->snd_una;
682 }
683
684 tp->t_timer[TCPT_REXMT] = 0;
685 tp->irs = ti->ti_seq;
686 tcp_rcvseqinit(tp);
687 tp->t_flags |= TF_ACKNOW;
688 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
689 soisfconnected(so);
690 tp->t_state = TCPS_ESTABLISHED;
691
692 (void) tcp_reass(tp, (struct tcpiphdr *)0,
693 (struct mbuf *)0);
694 /*
695 * if we didn't have to retransmit the SYN,
696 * use its rtt as our initial srtt & rtt var.
697 */
698 if (tp->t_rtt)
699 tcp_xmit_timer(tp, tp->t_rtt);
700 } else
701 tp->t_state = TCPS_SYN_RECEIVED;
702
703 trimthenstep6:
704 /*
705 * Advance ti->ti_seq to correspond to first data byte.
706 * If data, trim to stay within window,
707 * dropping FIN if necessary.
708 */
709 ti->ti_seq++;
710 if (ti->ti_len > tp->rcv_wnd) {
711 todrop = ti->ti_len - tp->rcv_wnd;
712 m_adj(m, -todrop);
713 ti->ti_len = tp->rcv_wnd;
714 tiflags &= ~TH_FIN;
715 }
716 tp->snd_wl1 = ti->ti_seq - 1;
717 tp->rcv_up = ti->ti_seq;
718 goto step6;
719 } /* switch tp->t_state */
720 /*
721 * States other than LISTEN or SYN_SENT.
722 * Check that at least some bytes of segment are within
723 * receive window. If segment begins before rcv_nxt,
724 * drop leading data (and SYN); if nothing left, just ack.
725 */
726 todrop = tp->rcv_nxt - ti->ti_seq;
727 if (todrop > 0) {
728 if (tiflags & TH_SYN) {
729 tiflags &= ~TH_SYN;
730 ti->ti_seq++;
731 if (ti->ti_urp > 1)
732 ti->ti_urp--;
733 else
734 tiflags &= ~TH_URG;
735 todrop--;
736 }
737 /*
738 * Following if statement from Stevens, vol. 2, p. 960.
739 */
740 if (todrop > ti->ti_len
741 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
742 /*
743 * Any valid FIN must be to the left of the window.
744 * At this point the FIN must be a duplicate or out
745 * of sequence; drop it.
746 */
747 tiflags &= ~TH_FIN;
748
749 /*
750 * Send an ACK to resynchronize and drop any data.
751 * But keep on processing for RST or ACK.
752 */
753 tp->t_flags |= TF_ACKNOW;
754 todrop = ti->ti_len;
755 }
756 m_adj(m, todrop);
757 ti->ti_seq += todrop;
758 ti->ti_len -= todrop;
759 if (ti->ti_urp > todrop)
760 ti->ti_urp -= todrop;
761 else {
762 tiflags &= ~TH_URG;
763 ti->ti_urp = 0;
764 }
765 }
766 /*
767 * If new data are received on a connection after the
768 * user processes are gone, then RST the other end.
769 */
770 if ((so->so_state & SS_NOFDREF) &&
771 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
772 tp = tcp_close(tp);
773 goto dropwithreset;
774 }
775
776 /*
777 * If segment ends after window, drop trailing data
778 * (and PUSH and FIN); if nothing left, just ACK.
779 */
780 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
781 if (todrop > 0) {
782 if (todrop >= ti->ti_len) {
783 /*
784 * If a new connection request is received
785 * while in TIME_WAIT, drop the old connection
786 * and start over if the sequence numbers
787 * are above the previous ones.
788 */
789 if (tiflags & TH_SYN &&
790 tp->t_state == TCPS_TIME_WAIT &&
791 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
792 iss = tp->rcv_nxt + TCP_ISSINCR;
793 tp = tcp_close(tp);
794 goto findso;
795 }
796 /*
797 * If window is closed can only take segments at
798 * window edge, and have to drop data and PUSH from
799 * incoming segments. Continue processing, but
800 * remember to ack. Otherwise, drop segment
801 * and ack.
802 */
803 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
804 tp->t_flags |= TF_ACKNOW;
805 } else {
806 goto dropafterack;
807 }
808 }
809 m_adj(m, -todrop);
810 ti->ti_len -= todrop;
811 tiflags &= ~(TH_PUSH|TH_FIN);
812 }
813
814 /*
815 * If the RST bit is set examine the state:
816 * SYN_RECEIVED STATE:
817 * If passive open, return to LISTEN state.
818 * If active open, inform user that connection was refused.
819 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
820 * Inform user that connection was reset, and close tcb.
821 * CLOSING, LAST_ACK, TIME_WAIT STATES
822 * Close the tcb.
823 */
824 if (tiflags&TH_RST) switch (tp->t_state) {
825
826 case TCPS_SYN_RECEIVED:
827 case TCPS_ESTABLISHED:
828 case TCPS_FIN_WAIT_1:
829 case TCPS_FIN_WAIT_2:
830 case TCPS_CLOSE_WAIT:
831 tp->t_state = TCPS_CLOSED;
832 tcp_close(tp);
833 goto drop;
834
835 case TCPS_CLOSING:
836 case TCPS_LAST_ACK:
837 case TCPS_TIME_WAIT:
838 tcp_close(tp);
839 goto drop;
840 }
841
842 /*
843 * If a SYN is in the window, then this is an
844 * error and we send an RST and drop the connection.
845 */
846 if (tiflags & TH_SYN) {
847 tp = tcp_drop(tp,0);
848 goto dropwithreset;
849 }
850
851 /*
852 * If the ACK bit is off we drop the segment and return.
853 */
854 if ((tiflags & TH_ACK) == 0) goto drop;
855
856 /*
857 * Ack processing.
858 */
859 switch (tp->t_state) {
860 /*
861 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
862 * ESTABLISHED state and continue processing, otherwise
863 * send an RST. una<=ack<=max
864 */
865 case TCPS_SYN_RECEIVED:
866
867 if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
868 SEQ_GT(ti->ti_ack, tp->snd_max))
869 goto dropwithreset;
870 tp->t_state = TCPS_ESTABLISHED;
871 /*
872 * The sent SYN is ack'ed with our sequence number +1
873 * The first data byte already in the buffer will get
874 * lost if no correction is made. This is only needed for
875 * SS_CTL since the buffer is empty otherwise.
876 * tp->snd_una++; or:
877 */
878 tp->snd_una=ti->ti_ack;
879 if (so->so_state & SS_CTL) {
880 /* So tcp_ctl reports the right state */
881 ret = tcp_ctl(so);
882 if (ret == 1) {
883 soisfconnected(so);
884 so->so_state &= ~SS_CTL; /* success XXX */
885 } else if (ret == 2) {
886 so->so_state &= SS_PERSISTENT_MASK;
887 so->so_state |= SS_NOFDREF; /* CTL_CMD */
888 } else {
889 needoutput = 1;
890 tp->t_state = TCPS_FIN_WAIT_1;
891 }
892 } else {
893 soisfconnected(so);
894 }
895
896 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
897 tp->snd_wl1 = ti->ti_seq - 1;
898 /* Avoid ack processing; snd_una==ti_ack => dup ack */
899 goto synrx_to_est;
900 /* fall into ... */
901
902 /*
903 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
904 * ACKs. If the ack is in the range
905 * tp->snd_una < ti->ti_ack <= tp->snd_max
906 * then advance tp->snd_una to ti->ti_ack and drop
907 * data from the retransmission queue. If this ACK reflects
908 * more up to date window information we update our window information.
909 */
910 case TCPS_ESTABLISHED:
911 case TCPS_FIN_WAIT_1:
912 case TCPS_FIN_WAIT_2:
913 case TCPS_CLOSE_WAIT:
914 case TCPS_CLOSING:
915 case TCPS_LAST_ACK:
916 case TCPS_TIME_WAIT:
917
918 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
919 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
920 DEBUG_MISC((dfd, " dup ack m = %lx so = %lx\n",
921 (long )m, (long )so));
922 /*
923 * If we have outstanding data (other than
924 * a window probe), this is a completely
925 * duplicate ack (ie, window info didn't
926 * change), the ack is the biggest we've
927 * seen and we've seen exactly our rexmt
928 * threshold of them, assume a packet
929 * has been dropped and retransmit it.
930 * Kludge snd_nxt & the congestion
931 * window so we send only this one
932 * packet.
933 *
934 * We know we're losing at the current
935 * window size so do congestion avoidance
936 * (set ssthresh to half the current window
937 * and pull our congestion window back to
938 * the new ssthresh).
939 *
940 * Dup acks mean that packets have left the
941 * network (they're now cached at the receiver)
942 * so bump cwnd by the amount in the receiver
943 * to keep a constant cwnd packets in the
944 * network.
945 */
946 if (tp->t_timer[TCPT_REXMT] == 0 ||
947 ti->ti_ack != tp->snd_una)
948 tp->t_dupacks = 0;
949 else if (++tp->t_dupacks == TCPREXMTTHRESH) {
950 tcp_seq onxt = tp->snd_nxt;
951 u_int win =
952 min(tp->snd_wnd, tp->snd_cwnd) / 2 /
953 tp->t_maxseg;
954
955 if (win < 2)
956 win = 2;
957 tp->snd_ssthresh = win * tp->t_maxseg;
958 tp->t_timer[TCPT_REXMT] = 0;
959 tp->t_rtt = 0;
960 tp->snd_nxt = ti->ti_ack;
961 tp->snd_cwnd = tp->t_maxseg;
962 (void) tcp_output(tp);
963 tp->snd_cwnd = tp->snd_ssthresh +
964 tp->t_maxseg * tp->t_dupacks;
965 if (SEQ_GT(onxt, tp->snd_nxt))
966 tp->snd_nxt = onxt;
967 goto drop;
968 } else if (tp->t_dupacks > TCPREXMTTHRESH) {
969 tp->snd_cwnd += tp->t_maxseg;
970 (void) tcp_output(tp);
971 goto drop;
972 }
973 } else
974 tp->t_dupacks = 0;
975 break;
976 }
977 synrx_to_est:
978 /*
979 * If the congestion window was inflated to account
980 * for the other side's cached packets, retract it.
981 */
982 if (tp->t_dupacks > TCPREXMTTHRESH &&
983 tp->snd_cwnd > tp->snd_ssthresh)
984 tp->snd_cwnd = tp->snd_ssthresh;
985 tp->t_dupacks = 0;
986 if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
987 goto dropafterack;
988 }
989 acked = ti->ti_ack - tp->snd_una;
990
991 /*
992 * If transmit timer is running and timed sequence
993 * number was acked, update smoothed round trip time.
994 * Since we now have an rtt measurement, cancel the
995 * timer backoff (cf., Phil Karn's retransmit alg.).
996 * Recompute the initial retransmit timer.
997 */
998 if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
999 tcp_xmit_timer(tp,tp->t_rtt);
1000
1001 /*
1002 * If all outstanding data is acked, stop retransmit
1003 * timer and remember to restart (more output or persist).
1004 * If there is more data to be acked, restart retransmit
1005 * timer, using current (possibly backed-off) value.
1006 */
1007 if (ti->ti_ack == tp->snd_max) {
1008 tp->t_timer[TCPT_REXMT] = 0;
1009 needoutput = 1;
1010 } else if (tp->t_timer[TCPT_PERSIST] == 0)
1011 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1012 /*
1013 * When new data is acked, open the congestion window.
1014 * If the window gives us less than ssthresh packets
1015 * in flight, open exponentially (maxseg per packet).
1016 * Otherwise open linearly: maxseg per window
1017 * (maxseg^2 / cwnd per packet).
1018 */
1019 {
1020 register u_int cw = tp->snd_cwnd;
1021 register u_int incr = tp->t_maxseg;
1022
1023 if (cw > tp->snd_ssthresh)
1024 incr = incr * incr / cw;
1025 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1026 }
1027 if (acked > so->so_snd.sb_cc) {
1028 tp->snd_wnd -= so->so_snd.sb_cc;
1029 sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
1030 ourfinisacked = 1;
1031 } else {
1032 sbdrop(&so->so_snd, acked);
1033 tp->snd_wnd -= acked;
1034 ourfinisacked = 0;
1035 }
1036 tp->snd_una = ti->ti_ack;
1037 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1038 tp->snd_nxt = tp->snd_una;
1039
1040 switch (tp->t_state) {
1041
1042 /*
1043 * In FIN_WAIT_1 STATE in addition to the processing
1044 * for the ESTABLISHED state if our FIN is now acknowledged
1045 * then enter FIN_WAIT_2.
1046 */
1047 case TCPS_FIN_WAIT_1:
1048 if (ourfinisacked) {
1049 /*
1050 * If we can't receive any more
1051 * data, then closing user can proceed.
1052 * Starting the timer is contrary to the
1053 * specification, but if we don't get a FIN
1054 * we'll hang forever.
1055 */
1056 if (so->so_state & SS_FCANTRCVMORE) {
1057 tp->t_timer[TCPT_2MSL] = TCP_MAXIDLE;
1058 }
1059 tp->t_state = TCPS_FIN_WAIT_2;
1060 }
1061 break;
1062
1063 /*
1064 * In CLOSING STATE in addition to the processing for
1065 * the ESTABLISHED state if the ACK acknowledges our FIN
1066 * then enter the TIME-WAIT state, otherwise ignore
1067 * the segment.
1068 */
1069 case TCPS_CLOSING:
1070 if (ourfinisacked) {
1071 tp->t_state = TCPS_TIME_WAIT;
1072 tcp_canceltimers(tp);
1073 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1074 }
1075 break;
1076
1077 /*
1078 * In LAST_ACK, we may still be waiting for data to drain
1079 * and/or to be acked, as well as for the ack of our FIN.
1080 * If our FIN is now acknowledged, delete the TCB,
1081 * enter the closed state and return.
1082 */
1083 case TCPS_LAST_ACK:
1084 if (ourfinisacked) {
1085 tcp_close(tp);
1086 goto drop;
1087 }
1088 break;
1089
1090 /*
1091 * In TIME_WAIT state the only thing that should arrive
1092 * is a retransmission of the remote FIN. Acknowledge
1093 * it and restart the finack timer.
1094 */
1095 case TCPS_TIME_WAIT:
1096 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1097 goto dropafterack;
1098 }
1099 } /* switch(tp->t_state) */
1100
1101 step6:
1102 /*
1103 * Update window information.
1104 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1105 */
1106 if ((tiflags & TH_ACK) &&
1107 (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1108 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1109 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1110 tp->snd_wnd = tiwin;
1111 tp->snd_wl1 = ti->ti_seq;
1112 tp->snd_wl2 = ti->ti_ack;
1113 if (tp->snd_wnd > tp->max_sndwnd)
1114 tp->max_sndwnd = tp->snd_wnd;
1115 needoutput = 1;
1116 }
1117
1118 /*
1119 * Process segments with URG.
1120 */
1121 if ((tiflags & TH_URG) && ti->ti_urp &&
1122 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1123 /*
1124 * This is a kludge, but if we receive and accept
1125 * random urgent pointers, we'll crash in
1126 * soreceive. It's hard to imagine someone
1127 * actually wanting to send this much urgent data.
1128 */
1129 if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
1130 ti->ti_urp = 0;
1131 tiflags &= ~TH_URG;
1132 goto dodata;
1133 }
1134 /*
1135 * If this segment advances the known urgent pointer,
1136 * then mark the data stream. This should not happen
1137 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1138 * a FIN has been received from the remote side.
1139 * In these states we ignore the URG.
1140 *
1141 * According to RFC961 (Assigned Protocols),
1142 * the urgent pointer points to the last octet
1143 * of urgent data. We continue, however,
1144 * to consider it to indicate the first octet
1145 * of data past the urgent section as the original
1146 * spec states (in one of two places).
1147 */
1148 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1149 tp->rcv_up = ti->ti_seq + ti->ti_urp;
1150 so->so_urgc = so->so_rcv.sb_cc +
1151 (tp->rcv_up - tp->rcv_nxt); /* -1; */
1152 tp->rcv_up = ti->ti_seq + ti->ti_urp;
1153
1154 }
1155 } else
1156 /*
1157 * If no out of band data is expected,
1158 * pull receive urgent pointer along
1159 * with the receive window.
1160 */
1161 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1162 tp->rcv_up = tp->rcv_nxt;
1163 dodata:
1164
1165 /*
1166 * If this is a small packet, then ACK now - with Nagel
1167 * congestion avoidance sender won't send more until
1168 * he gets an ACK.
1169 */
1170 if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
1171 ((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
1172 tp->t_flags |= TF_ACKNOW;
1173 }
1174
1175 /*
1176 * Process the segment text, merging it into the TCP sequencing queue,
1177 * and arranging for acknowledgment of receipt if necessary.
1178 * This process logically involves adjusting tp->rcv_wnd as data
1179 * is presented to the user (this happens in tcp_usrreq.c,
1180 * case PRU_RCVD). If a FIN has already been received on this
1181 * connection then we just ignore the text.
1182 */
1183 if ((ti->ti_len || (tiflags&TH_FIN)) &&
1184 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1185 TCP_REASS(tp, ti, m, so, tiflags);
1186 } else {
1187 m_free(m);
1188 tiflags &= ~TH_FIN;
1189 }
1190
1191 /*
1192 * If FIN is received ACK the FIN and let the user know
1193 * that the connection is closing.
1194 */
1195 if (tiflags & TH_FIN) {
1196 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1197 /*
1198 * If we receive a FIN we can't send more data,
1199 * set it SS_FDRAIN
1200 * Shutdown the socket if there is no rx data in the
1201 * buffer.
1202 * soread() is called on completion of shutdown() and
1203 * will got to TCPS_LAST_ACK, and use tcp_output()
1204 * to send the FIN.
1205 */
1206 sofwdrain(so);
1207
1208 tp->t_flags |= TF_ACKNOW;
1209 tp->rcv_nxt++;
1210 }
1211 switch (tp->t_state) {
1212
1213 /*
1214 * In SYN_RECEIVED and ESTABLISHED STATES
1215 * enter the CLOSE_WAIT state.
1216 */
1217 case TCPS_SYN_RECEIVED:
1218 case TCPS_ESTABLISHED:
1219 if(so->so_emu == EMU_CTL) /* no shutdown on socket */
1220 tp->t_state = TCPS_LAST_ACK;
1221 else
1222 tp->t_state = TCPS_CLOSE_WAIT;
1223 break;
1224
1225 /*
1226 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1227 * enter the CLOSING state.
1228 */
1229 case TCPS_FIN_WAIT_1:
1230 tp->t_state = TCPS_CLOSING;
1231 break;
1232
1233 /*
1234 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1235 * starting the time-wait timer, turning off the other
1236 * standard timers.
1237 */
1238 case TCPS_FIN_WAIT_2:
1239 tp->t_state = TCPS_TIME_WAIT;
1240 tcp_canceltimers(tp);
1241 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1242 break;
1243
1244 /*
1245 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1246 */
1247 case TCPS_TIME_WAIT:
1248 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1249 break;
1250 }
1251 }
1252
1253 /*
1254 * Return any desired output.
1255 */
1256 if (needoutput || (tp->t_flags & TF_ACKNOW)) {
1257 (void) tcp_output(tp);
1258 }
1259 return;
1260
1261 dropafterack:
1262 /*
1263 * Generate an ACK dropping incoming segment if it occupies
1264 * sequence space, where the ACK reflects our state.
1265 */
1266 if (tiflags & TH_RST)
1267 goto drop;
1268 m_free(m);
1269 tp->t_flags |= TF_ACKNOW;
1270 (void) tcp_output(tp);
1271 return;
1272
1273 dropwithreset:
1274 /* reuses m if m!=NULL, m_free() unnecessary */
1275 if (tiflags & TH_ACK)
1276 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1277 else {
1278 if (tiflags & TH_SYN) ti->ti_len++;
1279 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1280 TH_RST|TH_ACK);
1281 }
1282
1283 return;
1284
1285 drop:
1286 /*
1287 * Drop space held by incoming segment and return.
1288 */
1289 m_free(m);
1290 }
1291
1292 static void
1293 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcpiphdr *ti)
1294 {
1295 uint16_t mss;
1296 int opt, optlen;
1297
1298 DEBUG_CALL("tcp_dooptions");
1299 DEBUG_ARGS((dfd, " tp = %lx cnt=%i\n", (long)tp, cnt));
1300
1301 for (; cnt > 0; cnt -= optlen, cp += optlen) {
1302 opt = cp[0];
1303 if (opt == TCPOPT_EOL)
1304 break;
1305 if (opt == TCPOPT_NOP)
1306 optlen = 1;
1307 else {
1308 optlen = cp[1];
1309 if (optlen <= 0)
1310 break;
1311 }
1312 switch (opt) {
1313
1314 default:
1315 continue;
1316
1317 case TCPOPT_MAXSEG:
1318 if (optlen != TCPOLEN_MAXSEG)
1319 continue;
1320 if (!(ti->ti_flags & TH_SYN))
1321 continue;
1322 memcpy((char *) &mss, (char *) cp + 2, sizeof(mss));
1323 NTOHS(mss);
1324 (void) tcp_mss(tp, mss); /* sets t_maxseg */
1325 break;
1326 }
1327 }
1328 }
1329
1330
1331 /*
1332 * Pull out of band byte out of a segment so
1333 * it doesn't appear in the user's data queue.
1334 * It is still reflected in the segment length for
1335 * sequencing purposes.
1336 */
1337
1338 #ifdef notdef
1339
1340 void
1341 tcp_pulloutofband(so, ti, m)
1342 struct socket *so;
1343 struct tcpiphdr *ti;
1344 register struct mbuf *m;
1345 {
1346 int cnt = ti->ti_urp - 1;
1347
1348 while (cnt >= 0) {
1349 if (m->m_len > cnt) {
1350 char *cp = mtod(m, caddr_t) + cnt;
1351 struct tcpcb *tp = sototcpcb(so);
1352
1353 tp->t_iobc = *cp;
1354 tp->t_oobflags |= TCPOOB_HAVEDATA;
1355 memcpy(sp, cp+1, (unsigned)(m->m_len - cnt - 1));
1356 m->m_len--;
1357 return;
1358 }
1359 cnt -= m->m_len;
1360 m = m->m_next; /* XXX WRONG! Fix it! */
1361 if (m == 0)
1362 break;
1363 }
1364 panic("tcp_pulloutofband");
1365 }
1366
1367 #endif /* notdef */
1368
1369 /*
1370 * Collect new round-trip time estimate
1371 * and update averages and current timeout.
1372 */
1373
1374 static void
1375 tcp_xmit_timer(register struct tcpcb *tp, int rtt)
1376 {
1377 register short delta;
1378
1379 DEBUG_CALL("tcp_xmit_timer");
1380 DEBUG_ARG("tp = %lx", (long)tp);
1381 DEBUG_ARG("rtt = %d", rtt);
1382
1383 if (tp->t_srtt != 0) {
1384 /*
1385 * srtt is stored as fixed point with 3 bits after the
1386 * binary point (i.e., scaled by 8). The following magic
1387 * is equivalent to the smoothing algorithm in rfc793 with
1388 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1389 * point). Adjust rtt to origin 0.
1390 */
1391 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1392 if ((tp->t_srtt += delta) <= 0)
1393 tp->t_srtt = 1;
1394 /*
1395 * We accumulate a smoothed rtt variance (actually, a
1396 * smoothed mean difference), then set the retransmit
1397 * timer to smoothed rtt + 4 times the smoothed variance.
1398 * rttvar is stored as fixed point with 2 bits after the
1399 * binary point (scaled by 4). The following is
1400 * equivalent to rfc793 smoothing with an alpha of .75
1401 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
1402 * rfc793's wired-in beta.
1403 */
1404 if (delta < 0)
1405 delta = -delta;
1406 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1407 if ((tp->t_rttvar += delta) <= 0)
1408 tp->t_rttvar = 1;
1409 } else {
1410 /*
1411 * No rtt measurement yet - use the unsmoothed rtt.
1412 * Set the variance to half the rtt (so our first
1413 * retransmit happens at 3*rtt).
1414 */
1415 tp->t_srtt = rtt << TCP_RTT_SHIFT;
1416 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1417 }
1418 tp->t_rtt = 0;
1419 tp->t_rxtshift = 0;
1420
1421 /*
1422 * the retransmit should happen at rtt + 4 * rttvar.
1423 * Because of the way we do the smoothing, srtt and rttvar
1424 * will each average +1/2 tick of bias. When we compute
1425 * the retransmit timer, we want 1/2 tick of rounding and
1426 * 1 extra tick because of +-1/2 tick uncertainty in the
1427 * firing of the timer. The bias will give us exactly the
1428 * 1.5 tick we need. But, because the bias is
1429 * statistical, we have to test that we don't drop below
1430 * the minimum feasible timer (which is 2 ticks).
1431 */
1432 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1433 (short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
1434
1435 /*
1436 * We received an ack for a packet that wasn't retransmitted;
1437 * it is probably safe to discard any error indications we've
1438 * received recently. This isn't quite right, but close enough
1439 * for now (a route might have failed after we sent a segment,
1440 * and the return path might not be symmetrical).
1441 */
1442 tp->t_softerror = 0;
1443 }
1444
1445 /*
1446 * Determine a reasonable value for maxseg size.
1447 * If the route is known, check route for mtu.
1448 * If none, use an mss that can be handled on the outgoing
1449 * interface without forcing IP to fragment; if bigger than
1450 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1451 * to utilize large mbufs. If no route is found, route has no mtu,
1452 * or the destination isn't local, use a default, hopefully conservative
1453 * size (usually 512 or the default IP max size, but no more than the mtu
1454 * of the interface), as we can't discover anything about intervening
1455 * gateways or networks. We also initialize the congestion/slow start
1456 * window to be a single segment if the destination isn't local.
1457 * While looking at the routing entry, we also initialize other path-dependent
1458 * parameters from pre-set or cached values in the routing entry.
1459 */
1460
1461 int
1462 tcp_mss(struct tcpcb *tp, u_int offer)
1463 {
1464 struct socket *so = tp->t_socket;
1465 int mss;
1466
1467 DEBUG_CALL("tcp_mss");
1468 DEBUG_ARG("tp = %lx", (long)tp);
1469 DEBUG_ARG("offer = %d", offer);
1470
1471 mss = min(IF_MTU, IF_MRU) - sizeof(struct tcpiphdr);
1472 if (offer)
1473 mss = min(mss, offer);
1474 mss = max(mss, 32);
1475 if (mss < tp->t_maxseg || offer != 0)
1476 tp->t_maxseg = mss;
1477
1478 tp->snd_cwnd = mss;
1479
1480 sbreserve(&so->so_snd, TCP_SNDSPACE + ((TCP_SNDSPACE % mss) ?
1481 (mss - (TCP_SNDSPACE % mss)) :
1482 0));
1483 sbreserve(&so->so_rcv, TCP_RCVSPACE + ((TCP_RCVSPACE % mss) ?
1484 (mss - (TCP_RCVSPACE % mss)) :
1485 0));
1486
1487 DEBUG_MISC((dfd, " returning mss = %d\n", mss));
1488
1489 return mss;
1490 }