Add softlinks so ports like sysutils/ffsrecov will compile. Eventually
[dragonfly.git] / sys / netinet / tcp_input.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
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. 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.
20 *
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
31 * SUCH DAMAGE.
32 *
33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
34 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
efd4b327 35 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.8 2003/08/13 18:34:25 hsu Exp $
984263bc
MD
36 */
37
38#include "opt_ipfw.h" /* for ipfw_fwd */
39#include "opt_inet6.h"
40#include "opt_ipsec.h"
41#include "opt_tcpdebug.h"
42#include "opt_tcp_input.h"
43
44#include <sys/param.h>
45#include <sys/systm.h>
46#include <sys/kernel.h>
47#include <sys/sysctl.h>
48#include <sys/malloc.h>
49#include <sys/mbuf.h>
50#include <sys/proc.h> /* for proc0 declaration */
51#include <sys/protosw.h>
52#include <sys/socket.h>
53#include <sys/socketvar.h>
54#include <sys/syslog.h>
55
56#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
57
58#include <net/if.h>
59#include <net/route.h>
60
61#include <netinet/in.h>
62#include <netinet/in_systm.h>
63#include <netinet/ip.h>
64#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
65#include <netinet/in_var.h>
66#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
67#include <netinet/in_pcb.h>
68#include <netinet/ip_var.h>
69#include <netinet/ip6.h>
70#include <netinet/icmp6.h>
71#include <netinet6/nd6.h>
72#include <netinet6/ip6_var.h>
73#include <netinet6/in6_pcb.h>
74#include <netinet/tcp.h>
75#include <netinet/tcp_fsm.h>
76#include <netinet/tcp_seq.h>
77#include <netinet/tcp_timer.h>
78#include <netinet/tcp_var.h>
79#include <netinet6/tcp6_var.h>
80#include <netinet/tcpip.h>
81#ifdef TCPDEBUG
82#include <netinet/tcp_debug.h>
83
84u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
85struct tcphdr tcp_savetcp;
86#endif /* TCPDEBUG */
87
88#ifdef FAST_IPSEC
89#include <netipsec/ipsec.h>
90#include <netipsec/ipsec6.h>
91#endif
92
93#ifdef IPSEC
94#include <netinet6/ipsec.h>
95#include <netinet6/ipsec6.h>
d2438d69 96#include <netproto/key/key.h>
984263bc
MD
97#endif /*IPSEC*/
98
99#include <machine/in_cksum.h>
100
101MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
102
103static const int tcprexmtthresh = 3;
104tcp_cc tcp_ccgen;
105
106struct tcpstat tcpstat;
107SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
109
110static int log_in_vain = 0;
111SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
112 &log_in_vain, 0, "Log all incoming TCP connections");
113
114static int blackhole = 0;
115SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
116 &blackhole, 0, "Do not send RST when dropping refused connections");
117
118int tcp_delack_enabled = 1;
119SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
120 &tcp_delack_enabled, 0,
121 "Delay ACK to try and piggyback it onto a data packet");
122
123#ifdef TCP_DROP_SYNFIN
124static int drop_synfin = 0;
125SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
127#endif
128
33abdd1c
MD
129static int tcp_do_limitedtransmit = 1;
130SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
131 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
132
4b52d5ee
JH
133static int tcp_do_rfc3390 = 1;
134SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
135 &tcp_do_rfc3390, 0,
136 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
137
efd4b327
JH
138static int tcp_do_eifel_detect = 1;
139SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
140 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
141
984263bc
MD
142struct inpcbhead tcb;
143#define tcb6 tcb /* for KAME src sync over BSD*'s */
144struct inpcbinfo tcbinfo;
145
146static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
147static void tcp_pulloutofband(struct socket *,
148 struct tcphdr *, struct mbuf *, int);
149static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
150 struct mbuf *);
151static void tcp_xmit_timer(struct tcpcb *, int);
152static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
153
154/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
155#ifdef INET6
156#define ND6_HINT(tp) \
157do { \
158 if ((tp) && (tp)->t_inpcb && \
159 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
160 (tp)->t_inpcb->in6p_route.ro_rt) \
161 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
162} while (0)
163#else
164#define ND6_HINT(tp)
165#endif
166
167/*
168 * Indicate whether this ack should be delayed. We can delay the ack if
169 * - delayed acks are enabled and
170 * - there is no delayed ack timer in progress and
171 * - our last ack wasn't a 0-sized window. We never want to delay
172 * the ack that opens up a 0-sized window.
173 */
174#define DELAY_ACK(tp) \
175 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
176 (tp->t_flags & TF_RXWIN0SENT) == 0)
177
178static int
179tcp_reass(tp, th, tlenp, m)
2256ba69
RG
180 struct tcpcb *tp;
181 struct tcphdr *th;
984263bc
MD
182 int *tlenp;
183 struct mbuf *m;
184{
185 struct tseg_qent *q;
186 struct tseg_qent *p = NULL;
187 struct tseg_qent *nq;
188 struct tseg_qent *te;
189 struct socket *so = tp->t_inpcb->inp_socket;
190 int flags;
191
192 /*
193 * Call with th==0 after become established to
194 * force pre-ESTABLISHED data up to user socket.
195 */
196 if (th == 0)
197 goto present;
198
199 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
200 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
201 M_NOWAIT);
202 if (te == NULL) {
203 tcpstat.tcps_rcvmemdrop++;
204 m_freem(m);
205 return (0);
206 }
207
208 /*
209 * Find a segment which begins after this one does.
210 */
211 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
212 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
213 break;
214 p = q;
215 }
216
217 /*
218 * If there is a preceding segment, it may provide some of
219 * our data already. If so, drop the data from the incoming
220 * segment. If it provides all of our data, drop us.
221 */
222 if (p != NULL) {
2256ba69 223 int i;
984263bc
MD
224 /* conversion to int (in i) handles seq wraparound */
225 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
226 if (i > 0) {
227 if (i >= *tlenp) {
228 tcpstat.tcps_rcvduppack++;
229 tcpstat.tcps_rcvdupbyte += *tlenp;
230 m_freem(m);
231 free(te, M_TSEGQ);
232 /*
233 * Try to present any queued data
234 * at the left window edge to the user.
235 * This is needed after the 3-WHS
236 * completes.
237 */
238 goto present; /* ??? */
239 }
240 m_adj(m, i);
241 *tlenp -= i;
242 th->th_seq += i;
243 }
244 }
245 tcpstat.tcps_rcvoopack++;
246 tcpstat.tcps_rcvoobyte += *tlenp;
247
248 /*
249 * While we overlap succeeding segments trim them or,
250 * if they are completely covered, dequeue them.
251 */
252 while (q) {
2256ba69 253 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
984263bc
MD
254 if (i <= 0)
255 break;
256 if (i < q->tqe_len) {
257 q->tqe_th->th_seq += i;
258 q->tqe_len -= i;
259 m_adj(q->tqe_m, i);
260 break;
261 }
262
263 nq = LIST_NEXT(q, tqe_q);
264 LIST_REMOVE(q, tqe_q);
265 m_freem(q->tqe_m);
266 free(q, M_TSEGQ);
267 q = nq;
268 }
269
270 /* Insert the new segment queue entry into place. */
271 te->tqe_m = m;
272 te->tqe_th = th;
273 te->tqe_len = *tlenp;
274
275 if (p == NULL) {
276 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
277 } else {
278 LIST_INSERT_AFTER(p, te, tqe_q);
279 }
280
281present:
282 /*
283 * Present data to user, advancing rcv_nxt through
284 * completed sequence space.
285 */
286 if (!TCPS_HAVEESTABLISHED(tp->t_state))
287 return (0);
288 q = LIST_FIRST(&tp->t_segq);
289 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
290 return (0);
291 do {
292 tp->rcv_nxt += q->tqe_len;
293 flags = q->tqe_th->th_flags & TH_FIN;
294 nq = LIST_NEXT(q, tqe_q);
295 LIST_REMOVE(q, tqe_q);
296 if (so->so_state & SS_CANTRCVMORE)
297 m_freem(q->tqe_m);
298 else
299 sbappend(&so->so_rcv, q->tqe_m);
300 free(q, M_TSEGQ);
301 q = nq;
302 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
303 ND6_HINT(tp);
304 sorwakeup(so);
305 return (flags);
306}
307
308/*
309 * TCP input routine, follows pages 65-76 of the
310 * protocol specification dated September, 1981 very closely.
311 */
312#ifdef INET6
313int
314tcp6_input(mp, offp, proto)
315 struct mbuf **mp;
316 int *offp, proto;
317{
2256ba69 318 struct mbuf *m = *mp;
984263bc
MD
319 struct in6_ifaddr *ia6;
320
321 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
322
323 /*
324 * draft-itojun-ipv6-tcp-to-anycast
325 * better place to put this in?
326 */
327 ia6 = ip6_getdstifaddr(m);
328 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
329 struct ip6_hdr *ip6;
330
331 ip6 = mtod(m, struct ip6_hdr *);
332 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
333 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
334 return IPPROTO_DONE;
335 }
336
337 tcp_input(m, *offp, proto);
338 return IPPROTO_DONE;
339}
340#endif
341
342void
343tcp_input(m, off0, proto)
2256ba69 344 struct mbuf *m;
984263bc
MD
345 int off0, proto;
346{
2256ba69
RG
347 struct tcphdr *th;
348 struct ip *ip = NULL;
349 struct ipovly *ipov;
350 struct inpcb *inp = NULL;
984263bc
MD
351 u_char *optp = NULL;
352 int optlen = 0;
353 int len, tlen, off;
354 int drop_hdrlen;
2256ba69
RG
355 struct tcpcb *tp = NULL;
356 int thflags;
984263bc
MD
357 struct socket *so = 0;
358 int todrop, acked, ourfinisacked, needoutput = 0;
359 u_long tiwin;
360 struct tcpopt to; /* options in this segment */
361 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
362 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
363 struct sockaddr_in *next_hop = NULL;
364 int rstreason; /* For badport_bandlim accounting purposes */
efd4b327 365 int useTS; /* use timestamps in Eifel detection */
984263bc
MD
366 struct ip6_hdr *ip6 = NULL;
367#ifdef INET6
368 int isipv6;
369#else
370 const int isipv6 = 0;
371#endif
372#ifdef TCPDEBUG
373 short ostate = 0;
374#endif
375
376 /* Grab info from MT_TAG mbufs prepended to the chain. */
377 for (;m && m->m_type == MT_TAG; m = m->m_next) {
378 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
379 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
380 }
381#ifdef INET6
382 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
383#endif
384 bzero((char *)&to, sizeof(to));
385
386 tcpstat.tcps_rcvtotal++;
387
388 if (isipv6) {
389 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
390 ip6 = mtod(m, struct ip6_hdr *);
391 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
392 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
393 tcpstat.tcps_rcvbadsum++;
394 goto drop;
395 }
396 th = (struct tcphdr *)((caddr_t)ip6 + off0);
397
398 /*
399 * Be proactive about unspecified IPv6 address in source.
400 * As we use all-zero to indicate unbounded/unconnected pcb,
401 * unspecified IPv6 address can be used to confuse us.
402 *
403 * Note that packets with unspecified IPv6 destination is
404 * already dropped in ip6_input.
405 */
406 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
407 /* XXX stat */
408 goto drop;
409 }
410 } else {
411 /*
412 * Get IP and TCP header together in first mbuf.
413 * Note: IP leaves IP header in first mbuf.
414 */
415 if (off0 > sizeof(struct ip)) {
416 ip_stripoptions(m, (struct mbuf *)0);
417 off0 = sizeof(struct ip);
418 }
419 if (m->m_len < sizeof(struct tcpiphdr)) {
420 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
421 tcpstat.tcps_rcvshort++;
422 return;
423 }
424 }
425 ip = mtod(m, struct ip *);
426 ipov = (struct ipovly *)ip;
427 th = (struct tcphdr *)((caddr_t)ip + off0);
428 tlen = ip->ip_len;
429
430 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
431 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
432 th->th_sum = m->m_pkthdr.csum_data;
433 else
434 th->th_sum = in_pseudo(ip->ip_src.s_addr,
435 ip->ip_dst.s_addr,
436 htonl(m->m_pkthdr.csum_data +
437 ip->ip_len +
438 IPPROTO_TCP));
439 th->th_sum ^= 0xffff;
440 } else {
441 /*
442 * Checksum extended TCP header and data.
443 */
444 len = sizeof(struct ip) + tlen;
445 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
446 ipov->ih_len = (u_short)tlen;
447 ipov->ih_len = htons(ipov->ih_len);
448 th->th_sum = in_cksum(m, len);
449 }
450 if (th->th_sum) {
451 tcpstat.tcps_rcvbadsum++;
452 goto drop;
453 }
454#ifdef INET6
455 /* Re-initialization for later version check */
456 ip->ip_v = IPVERSION;
457#endif
458 }
459
460 /*
461 * Check that TCP offset makes sense,
462 * pull out TCP options and adjust length. XXX
463 */
464 off = th->th_off << 2;
465 if (off < sizeof(struct tcphdr) || off > tlen) {
466 tcpstat.tcps_rcvbadoff++;
467 goto drop;
468 }
469 tlen -= off; /* tlen is used instead of ti->ti_len */
470 if (off > sizeof(struct tcphdr)) {
471 if (isipv6) {
472 IP6_EXTHDR_CHECK(m, off0, off, );
473 ip6 = mtod(m, struct ip6_hdr *);
474 th = (struct tcphdr *)((caddr_t)ip6 + off0);
475 } else {
476 if (m->m_len < sizeof(struct ip) + off) {
477 if ((m = m_pullup(m, sizeof(struct ip) + off))
478 == 0) {
479 tcpstat.tcps_rcvshort++;
480 return;
481 }
482 ip = mtod(m, struct ip *);
483 ipov = (struct ipovly *)ip;
484 th = (struct tcphdr *)((caddr_t)ip + off0);
485 }
486 }
487 optlen = off - sizeof(struct tcphdr);
488 optp = (u_char *)(th + 1);
489 }
490 thflags = th->th_flags;
491
492#ifdef TCP_DROP_SYNFIN
493 /*
494 * If the drop_synfin option is enabled, drop all packets with
495 * both the SYN and FIN bits set. This prevents e.g. nmap from
496 * identifying the TCP/IP stack.
497 *
498 * This is a violation of the TCP specification.
499 */
500 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
501 goto drop;
502#endif
503
504 /*
505 * Convert TCP protocol specific fields to host format.
506 */
507 th->th_seq = ntohl(th->th_seq);
508 th->th_ack = ntohl(th->th_ack);
509 th->th_win = ntohs(th->th_win);
510 th->th_urp = ntohs(th->th_urp);
511
512 /*
513 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
514 * until after ip6_savecontrol() is called and before other functions
515 * which don't want those proto headers.
516 * Because ip6_savecontrol() is going to parse the mbuf to
517 * search for data to be passed up to user-land, it wants mbuf
518 * parameters to be unchanged.
519 * XXX: the call of ip6_savecontrol() has been obsoleted based on
520 * latest version of the advanced API (20020110).
521 */
522 drop_hdrlen = off0 + off;
523
524 /*
525 * Locate pcb for segment.
526 */
527findpcb:
528 /* IPFIREWALL_FORWARD section */
529 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
530 /*
531 * Transparently forwarded. Pretend to be the destination.
532 * already got one like this?
533 */
534 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
535 ip->ip_dst, th->th_dport,
536 0, m->m_pkthdr.rcvif);
537 if (!inp) {
538 /* It's new. Try find the ambushing socket. */
539 inp = in_pcblookup_hash(&tcbinfo,
540 ip->ip_src, th->th_sport,
541 next_hop->sin_addr,
542 next_hop->sin_port ?
543 ntohs(next_hop->sin_port) :
544 th->th_dport,
545 1, m->m_pkthdr.rcvif);
546 }
547 } else {
548 if (isipv6)
549 inp = in6_pcblookup_hash(&tcbinfo,
550 &ip6->ip6_src, th->th_sport,
551 &ip6->ip6_dst, th->th_dport,
552 1, m->m_pkthdr.rcvif);
553 else
554 inp = in_pcblookup_hash(&tcbinfo,
555 ip->ip_src, th->th_sport,
556 ip->ip_dst, th->th_dport,
557 1, m->m_pkthdr.rcvif);
558 }
559
560#ifdef IPSEC
561 if (isipv6) {
562 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
563 ipsec6stat.in_polvio++;
564 goto drop;
565 }
566 } else {
567 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
568 ipsecstat.in_polvio++;
569 goto drop;
570 }
571 }
572#endif
573#ifdef FAST_IPSEC
574 if (isipv6) {
575 if (inp != NULL && ipsec6_in_reject(m, inp)) {
576 goto drop;
577 }
578 } else {
579 if (inp != NULL && ipsec4_in_reject(m, inp)) {
580 goto drop;
581 }
582 }
583#endif
584
585 /*
586 * If the state is CLOSED (i.e., TCB does not exist) then
587 * all data in the incoming segment is discarded.
588 * If the TCB exists but is in CLOSED state, it is embryonic,
589 * but should either do a listen or a connect soon.
590 */
591 if (inp == NULL) {
592 if (log_in_vain) {
593#ifdef INET6
594 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
595#else
596 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
597#endif
598 if (isipv6) {
599 strcpy(dbuf, "[");
600 strcpy(sbuf, "[");
601 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
602 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
603 strcat(dbuf, "]");
604 strcat(sbuf, "]");
605 } else {
606 strcpy(dbuf, inet_ntoa(ip->ip_dst));
607 strcpy(sbuf, inet_ntoa(ip->ip_src));
608 }
609 switch (log_in_vain) {
610 case 1:
611 if ((thflags & TH_SYN) == 0)
612 break;
613 case 2:
614 log(LOG_INFO,
615 "Connection attempt to TCP %s:%d "
616 "from %s:%d flags:0x%02x\n",
617 dbuf, ntohs(th->th_dport), sbuf,
618 ntohs(th->th_sport), thflags);
619 break;
620 default:
621 break;
622 }
623 }
624 if (blackhole) {
625 switch (blackhole) {
626 case 1:
627 if (thflags & TH_SYN)
628 goto drop;
629 break;
630 case 2:
631 goto drop;
632 default:
633 goto drop;
634 }
635 }
636 rstreason = BANDLIM_RST_CLOSEDPORT;
637 goto dropwithreset;
638 }
639 tp = intotcpcb(inp);
640 if (tp == NULL) {
641 rstreason = BANDLIM_RST_CLOSEDPORT;
642 goto dropwithreset;
643 }
644 if (tp->t_state == TCPS_CLOSED)
645 goto drop;
646
647 /* Unscale the window into a 32-bit value. */
648 if ((thflags & TH_SYN) == 0)
649 tiwin = th->th_win << tp->snd_scale;
650 else
651 tiwin = th->th_win;
652
653 so = inp->inp_socket;
654 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
655 struct in_conninfo inc;
656#ifdef TCPDEBUG
657 if (so->so_options & SO_DEBUG) {
658 ostate = tp->t_state;
659 if (isipv6)
660 bcopy((char *)ip6, (char *)tcp_saveipgen,
661 sizeof(*ip6));
662 else
663 bcopy((char *)ip, (char *)tcp_saveipgen,
664 sizeof(*ip));
665 tcp_savetcp = *th;
666 }
667#endif
668 /* skip if this isn't a listen socket */
669 if ((so->so_options & SO_ACCEPTCONN) == 0)
670 goto after_listen;
671#ifdef INET6
672 inc.inc_isipv6 = isipv6;
673#endif
674 if (isipv6) {
675 inc.inc6_faddr = ip6->ip6_src;
676 inc.inc6_laddr = ip6->ip6_dst;
677 inc.inc6_route.ro_rt = NULL; /* XXX */
678 } else {
679 inc.inc_faddr = ip->ip_src;
680 inc.inc_laddr = ip->ip_dst;
681 inc.inc_route.ro_rt = NULL; /* XXX */
682 }
683 inc.inc_fport = th->th_sport;
684 inc.inc_lport = th->th_dport;
685
686 /*
687 * If the state is LISTEN then ignore segment if it contains
688 * a RST. If the segment contains an ACK then it is bad and
689 * send a RST. If it does not contain a SYN then it is not
690 * interesting; drop it.
691 *
692 * If the state is SYN_RECEIVED (syncache) and seg contains
693 * an ACK, but not for our SYN/ACK, send a RST. If the seg
694 * contains a RST, check the sequence number to see if it
695 * is a valid reset segment.
696 */
697 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
698 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
699 if (!syncache_expand(&inc, th, &so, m)) {
700 /*
701 * No syncache entry, or ACK was not
702 * for our SYN/ACK. Send a RST.
703 */
704 tcpstat.tcps_badsyn++;
705 rstreason = BANDLIM_RST_OPENPORT;
706 goto dropwithreset;
707 }
708 if (so == NULL)
709 /*
710 * Could not complete 3-way handshake,
711 * connection is being closed down, and
712 * syncache will free mbuf.
713 */
714 return;
715 /*
716 * Socket is created in state SYN_RECEIVED.
717 * Continue processing segment.
718 */
719 inp = sotoinpcb(so);
720 tp = intotcpcb(inp);
721 /*
722 * This is what would have happened in
723 * tcp_output() when the SYN,ACK was sent.
724 */
725 tp->snd_up = tp->snd_una;
726 tp->snd_max = tp->snd_nxt = tp->iss + 1;
727 tp->last_ack_sent = tp->rcv_nxt;
728/*
729 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
730 * until the _second_ ACK is received:
731 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
732 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
733 * move to ESTAB, set snd_wnd to tiwin.
734 */
735 tp->snd_wnd = tiwin; /* unscaled */
736 goto after_listen;
737 }
738 if (thflags & TH_RST) {
739 syncache_chkrst(&inc, th);
740 goto drop;
741 }
742 if (thflags & TH_ACK) {
743 syncache_badack(&inc);
744 tcpstat.tcps_badsyn++;
745 rstreason = BANDLIM_RST_OPENPORT;
746 goto dropwithreset;
747 }
748 goto drop;
749 }
750
751 /*
752 * Segment's flags are (SYN) or (SYN|FIN).
753 */
754#ifdef INET6
755 /*
756 * If deprecated address is forbidden,
757 * we do not accept SYN to deprecated interface
758 * address to prevent any new inbound connection from
759 * getting established.
760 * When we do not accept SYN, we send a TCP RST,
761 * with deprecated source address (instead of dropping
762 * it). We compromise it as it is much better for peer
763 * to send a RST, and RST will be the final packet
764 * for the exchange.
765 *
766 * If we do not forbid deprecated addresses, we accept
767 * the SYN packet. RFC2462 does not suggest dropping
768 * SYN in this case.
769 * If we decipher RFC2462 5.5.4, it says like this:
770 * 1. use of deprecated addr with existing
771 * communication is okay - "SHOULD continue to be
772 * used"
773 * 2. use of it with new communication:
774 * (2a) "SHOULD NOT be used if alternate address
775 * with sufficient scope is available"
776 * (2b) nothing mentioned otherwise.
777 * Here we fall into (2b) case as we have no choice in
778 * our source address selection - we must obey the peer.
779 *
780 * The wording in RFC2462 is confusing, and there are
781 * multiple description text for deprecated address
782 * handling - worse, they are not exactly the same.
783 * I believe 5.5.4 is the best one, so we follow 5.5.4.
784 */
785 if (isipv6 && !ip6_use_deprecated) {
786 struct in6_ifaddr *ia6;
787
788 if ((ia6 = ip6_getdstifaddr(m)) &&
789 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
790 tp = NULL;
791 rstreason = BANDLIM_RST_OPENPORT;
792 goto dropwithreset;
793 }
794 }
795#endif
796 /*
797 * If it is from this socket, drop it, it must be forged.
798 * Don't bother responding if the destination was a broadcast.
799 */
800 if (th->th_dport == th->th_sport) {
801 if (isipv6) {
802 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
803 &ip6->ip6_src))
804 goto drop;
805 } else {
806 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
807 goto drop;
808 }
809 }
810 /*
811 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
812 *
813 * Note that it is quite possible to receive unicast
814 * link-layer packets with a broadcast IP address. Use
815 * in_broadcast() to find them.
816 */
817 if (m->m_flags & (M_BCAST|M_MCAST))
818 goto drop;
819 if (isipv6) {
820 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
821 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
822 goto drop;
823 } else {
824 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
825 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
826 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
827 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
828 goto drop;
829 }
830 /*
831 * SYN appears to be valid; create compressed TCP state
832 * for syncache, or perform t/tcp connection.
833 */
834 if (so->so_qlen <= so->so_qlimit) {
835 tcp_dooptions(&to, optp, optlen, 1);
836 if (!syncache_add(&inc, &to, th, &so, m))
837 goto drop;
838 if (so == NULL)
839 /*
840 * Entry added to syncache, mbuf used to
841 * send SYN,ACK packet.
842 */
843 return;
844 /*
845 * Segment passed TAO tests.
846 */
847 inp = sotoinpcb(so);
848 tp = intotcpcb(inp);
849 tp->snd_wnd = tiwin;
850 tp->t_starttime = ticks;
851 tp->t_state = TCPS_ESTABLISHED;
852
853 /*
854 * If there is a FIN, or if there is data and the
855 * connection is local, then delay SYN,ACK(SYN) in
856 * the hope of piggy-backing it on a response
857 * segment. Otherwise must send ACK now in case
858 * the other side is slow starting.
859 */
860 if (DELAY_ACK(tp) &&
861 ((thflags & TH_FIN) ||
862 (tlen != 0 &&
863 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
864 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
865 callout_reset(tp->tt_delack, tcp_delacktime,
866 tcp_timer_delack, tp);
867 tp->t_flags |= TF_NEEDSYN;
868 } else
869 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
870
871 tcpstat.tcps_connects++;
872 soisconnected(so);
873 goto trimthenstep6;
874 }
875 goto drop;
876 }
877after_listen:
878
879/* XXX temp debugging */
880 /* should not happen - syncache should pick up these connections */
881 if (tp->t_state == TCPS_LISTEN)
882 panic("tcp_input: TCPS_LISTEN");
883
884 /*
885 * Segment received on connection.
886 * Reset idle time and keep-alive timer.
887 */
888 tp->t_rcvtime = ticks;
889 if (TCPS_HAVEESTABLISHED(tp->t_state))
890 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
891
892 /*
893 * Process options.
894 * XXX this is tradtitional behavior, may need to be cleaned up.
895 */
896 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
897 if (thflags & TH_SYN) {
898 if (to.to_flags & TOF_SCALE) {
899 tp->t_flags |= TF_RCVD_SCALE;
900 tp->requested_s_scale = to.to_requested_s_scale;
901 }
902 if (to.to_flags & TOF_TS) {
903 tp->t_flags |= TF_RCVD_TSTMP;
904 tp->ts_recent = to.to_tsval;
905 tp->ts_recent_age = ticks;
906 }
907 if (to.to_flags & (TOF_CC|TOF_CCNEW))
908 tp->t_flags |= TF_RCVD_CC;
909 if (to.to_flags & TOF_MSS)
910 tcp_mss(tp, to.to_mss);
911 }
912
913 /*
914 * Header prediction: check for the two common cases
915 * of a uni-directional data xfer. If the packet has
916 * no control flags, is in-sequence, the window didn't
917 * change and we're not retransmitting, it's a
918 * candidate. If the length is zero and the ack moved
919 * forward, we're the sender side of the xfer. Just
920 * free the data acked & wake any higher level process
921 * that was blocked waiting for space. If the length
922 * is non-zero and the ack didn't move, we're the
923 * receiver side. If we're getting packets in-order
924 * (the reassembly queue is empty), add the data to
925 * the socket buffer and note that we need a delayed ack.
926 * Make sure that the hidden state-flags are also off.
927 * Since we check for TCPS_ESTABLISHED above, it can only
928 * be TH_NEEDSYN.
929 */
930 if (tp->t_state == TCPS_ESTABLISHED &&
931 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
932 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
933 ((to.to_flags & TOF_TS) == 0 ||
934 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
935 /*
936 * Using the CC option is compulsory if once started:
937 * the segment is OK if no T/TCP was negotiated or
938 * if the segment has a CC option equal to CCrecv
939 */
940 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
941 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
942 th->th_seq == tp->rcv_nxt &&
943 tiwin && tiwin == tp->snd_wnd &&
944 tp->snd_nxt == tp->snd_max) {
945
946 /*
947 * If last ACK falls within this segment's sequence numbers,
948 * record the timestamp.
949 * NOTE that the test is modified according to the latest
950 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
951 */
952 if ((to.to_flags & TOF_TS) != 0 &&
953 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
954 tp->ts_recent_age = ticks;
955 tp->ts_recent = to.to_tsval;
956 }
957
958 if (tlen == 0) {
959 if (SEQ_GT(th->th_ack, tp->snd_una) &&
960 SEQ_LEQ(th->th_ack, tp->snd_max) &&
961 tp->snd_cwnd >= tp->snd_wnd &&
962 ((!tcp_do_newreno &&
9845754e
MD
963 tp->t_dupacks < tcprexmtthresh) ||
964 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
984263bc
MD
965 /*
966 * this is a pure ack for outstanding data.
967 */
968 ++tcpstat.tcps_predack;
969 /*
970 * "bad retransmit" recovery
971 */
efd4b327
JH
972 useTS = tcp_do_eifel_detect &&
973 (to.to_flags & TOF_TS) &&
974 to.to_tsecr;
975 if ((useTS &&
976 (tp->t_flags & TF_FIRSTACCACK) &&
977 (to.to_tsecr < tp->t_rexmtTS)) ||
978 (!useTS &&
979 (tp->t_rxtshift == 1 &&
980 ticks < tp->t_badrxtwin))) {
984263bc
MD
981 tp->snd_cwnd = tp->snd_cwnd_prev;
982 tp->snd_ssthresh =
983 tp->snd_ssthresh_prev;
9845754e
MD
984 tp->snd_recover = tp->snd_recover_prev;
985 if (tp->t_flags & TF_WASFRECOVERY)
986 ENTER_FASTRECOVERY(tp);
984263bc
MD
987 tp->snd_nxt = tp->snd_max;
988 tp->t_badrxtwin = 0;
efd4b327
JH
989 tp->t_rxtshift = 0;
990 if (tp->t_flags & TF_FASTREXMT)
991 ++tcpstat.tcps_sndfastrexmitbad;
992 else
993 ++tcpstat.tcps_sndrtobad;
984263bc 994 }
efd4b327 995 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
984263bc
MD
996 /*
997 * Recalculate the retransmit timer / rtt.
998 *
999 * Some machines (certain windows boxes)
1000 * send broken timestamp replies during the
1001 * SYN+ACK phase, ignore timestamps of 0.
1002 */
1003 if ((to.to_flags & TOF_TS) != 0 &&
1004 to.to_tsecr) {
1005 tcp_xmit_timer(tp,
1006 ticks - to.to_tsecr + 1);
1007 } else if (tp->t_rtttime &&
1008 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1009 tcp_xmit_timer(tp,
1010 ticks - tp->t_rtttime);
1011 }
1012 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1013 acked = th->th_ack - tp->snd_una;
1014 tcpstat.tcps_rcvackpack++;
1015 tcpstat.tcps_rcvackbyte += acked;
1016 sbdrop(&so->so_snd, acked);
9845754e
MD
1017 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1018 SEQ_LEQ(th->th_ack, tp->snd_recover))
1019 tp->snd_recover = th->th_ack - 1;
1020 tp->snd_una = th->th_ack;
984263bc
MD
1021 tp->t_dupacks = 0;
1022 m_freem(m);
1023 ND6_HINT(tp); /* some progress has been done */
1024
1025 /*
1026 * If all outstanding data are acked, stop
1027 * retransmit timer, otherwise restart timer
1028 * using current (possibly backed-off) value.
1029 * If process is waiting for space,
1030 * wakeup/selwakeup/signal. If data
1031 * are ready to send, let tcp_output
1032 * decide between more output or persist.
1033 */
1034 if (tp->snd_una == tp->snd_max)
1035 callout_stop(tp->tt_rexmt);
1036 else if (!callout_active(tp->tt_persist))
1037 callout_reset(tp->tt_rexmt,
1038 tp->t_rxtcur,
1039 tcp_timer_rexmt, tp);
1040
1041 sowwakeup(so);
1042 if (so->so_snd.sb_cc)
1043 (void) tcp_output(tp);
1044 return;
1045 }
1046 } else if (th->th_ack == tp->snd_una &&
1047 LIST_EMPTY(&tp->t_segq) &&
1048 tlen <= sbspace(&so->so_rcv)) {
1049 /*
1050 * this is a pure, in-sequence data packet
1051 * with nothing on the reassembly queue and
1052 * we have enough buffer space to take it.
1053 */
1054 ++tcpstat.tcps_preddat;
1055 tp->rcv_nxt += tlen;
1056 tcpstat.tcps_rcvpack++;
1057 tcpstat.tcps_rcvbyte += tlen;
1058 ND6_HINT(tp); /* some progress has been done */
1059 /*
1060 * Add data to socket buffer.
1061 */
1062 if (so->so_state & SS_CANTRCVMORE) {
1063 m_freem(m);
1064 } else {
1065 m_adj(m, drop_hdrlen); /* delayed header drop */
1066 sbappend(&so->so_rcv, m);
1067 }
1068 sorwakeup(so);
1069 if (DELAY_ACK(tp)) {
1070 callout_reset(tp->tt_delack, tcp_delacktime,
1071 tcp_timer_delack, tp);
1072 } else {
1073 tp->t_flags |= TF_ACKNOW;
1074 tcp_output(tp);
1075 }
1076 return;
1077 }
1078 }
1079
1080 /*
1081 * Calculate amount of space in receive window,
1082 * and then do TCP input processing.
1083 * Receive window is amount of space in rcv queue,
1084 * but not less than advertised window.
1085 */
1086 { int win;
1087
1088 win = sbspace(&so->so_rcv);
1089 if (win < 0)
1090 win = 0;
1091 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1092 }
1093
1094 switch (tp->t_state) {
1095
1096 /*
1097 * If the state is SYN_RECEIVED:
1098 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1099 */
1100 case TCPS_SYN_RECEIVED:
1101 if ((thflags & TH_ACK) &&
1102 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1103 SEQ_GT(th->th_ack, tp->snd_max))) {
1104 rstreason = BANDLIM_RST_OPENPORT;
1105 goto dropwithreset;
1106 }
1107 break;
1108
1109 /*
1110 * If the state is SYN_SENT:
1111 * if seg contains an ACK, but not for our SYN, drop the input.
1112 * if seg contains a RST, then drop the connection.
1113 * if seg does not contain SYN, then drop it.
1114 * Otherwise this is an acceptable SYN segment
1115 * initialize tp->rcv_nxt and tp->irs
1116 * if seg contains ack then advance tp->snd_una
1117 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1118 * arrange for segment to be acked (eventually)
1119 * continue processing rest of data/controls, beginning with URG
1120 */
1121 case TCPS_SYN_SENT:
1122 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1123 taop = &tao_noncached;
1124 bzero(taop, sizeof(*taop));
1125 }
1126
1127 if ((thflags & TH_ACK) &&
1128 (SEQ_LEQ(th->th_ack, tp->iss) ||
1129 SEQ_GT(th->th_ack, tp->snd_max))) {
1130 /*
1131 * If we have a cached CCsent for the remote host,
1132 * hence we haven't just crashed and restarted,
1133 * do not send a RST. This may be a retransmission
1134 * from the other side after our earlier ACK was lost.
1135 * Our new SYN, when it arrives, will serve as the
1136 * needed ACK.
1137 */
1138 if (taop->tao_ccsent != 0)
1139 goto drop;
1140 else {
1141 rstreason = BANDLIM_UNLIMITED;
1142 goto dropwithreset;
1143 }
1144 }
1145 if (thflags & TH_RST) {
1146 if (thflags & TH_ACK)
1147 tp = tcp_drop(tp, ECONNREFUSED);
1148 goto drop;
1149 }
1150 if ((thflags & TH_SYN) == 0)
1151 goto drop;
1152 tp->snd_wnd = th->th_win; /* initial send window */
1153 tp->cc_recv = to.to_cc; /* foreign CC */
1154
1155 tp->irs = th->th_seq;
1156 tcp_rcvseqinit(tp);
1157 if (thflags & TH_ACK) {
1158 /*
1159 * Our SYN was acked. If segment contains CC.ECHO
1160 * option, check it to make sure this segment really
1161 * matches our SYN. If not, just drop it as old
1162 * duplicate, but send an RST if we're still playing
1163 * by the old rules. If no CC.ECHO option, make sure
1164 * we don't get fooled into using T/TCP.
1165 */
1166 if (to.to_flags & TOF_CCECHO) {
1167 if (tp->cc_send != to.to_ccecho) {
1168 if (taop->tao_ccsent != 0)
1169 goto drop;
1170 else {
1171 rstreason = BANDLIM_UNLIMITED;
1172 goto dropwithreset;
1173 }
1174 }
1175 } else
1176 tp->t_flags &= ~TF_RCVD_CC;
1177 tcpstat.tcps_connects++;
1178 soisconnected(so);
1179 /* Do window scaling on this connection? */
1180 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1181 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1182 tp->snd_scale = tp->requested_s_scale;
1183 tp->rcv_scale = tp->request_r_scale;
1184 }
1185 /* Segment is acceptable, update cache if undefined. */
1186 if (taop->tao_ccsent == 0)
1187 taop->tao_ccsent = to.to_ccecho;
1188
1189 tp->rcv_adv += tp->rcv_wnd;
1190 tp->snd_una++; /* SYN is acked */
1191 /*
1192 * If there's data, delay ACK; if there's also a FIN
1193 * ACKNOW will be turned on later.
1194 */
1195 if (DELAY_ACK(tp) && tlen != 0)
1196 callout_reset(tp->tt_delack, tcp_delacktime,
1197 tcp_timer_delack, tp);
1198 else
1199 tp->t_flags |= TF_ACKNOW;
1200 /*
1201 * Received <SYN,ACK> in SYN_SENT[*] state.
1202 * Transitions:
1203 * SYN_SENT --> ESTABLISHED
1204 * SYN_SENT* --> FIN_WAIT_1
1205 */
1206 tp->t_starttime = ticks;
1207 if (tp->t_flags & TF_NEEDFIN) {
1208 tp->t_state = TCPS_FIN_WAIT_1;
1209 tp->t_flags &= ~TF_NEEDFIN;
1210 thflags &= ~TH_SYN;
1211 } else {
1212 tp->t_state = TCPS_ESTABLISHED;
1213 callout_reset(tp->tt_keep, tcp_keepidle,
1214 tcp_timer_keep, tp);
1215 }
1216 } else {
1217 /*
1218 * Received initial SYN in SYN-SENT[*] state =>
1219 * simultaneous open. If segment contains CC option
1220 * and there is a cached CC, apply TAO test.
1221 * If it succeeds, connection is * half-synchronized.
1222 * Otherwise, do 3-way handshake:
1223 * SYN-SENT -> SYN-RECEIVED
1224 * SYN-SENT* -> SYN-RECEIVED*
1225 * If there was no CC option, clear cached CC value.
1226 */
1227 tp->t_flags |= TF_ACKNOW;
1228 callout_stop(tp->tt_rexmt);
1229 if (to.to_flags & TOF_CC) {
1230 if (taop->tao_cc != 0 &&
1231 CC_GT(to.to_cc, taop->tao_cc)) {
1232 /*
1233 * update cache and make transition:
1234 * SYN-SENT -> ESTABLISHED*
1235 * SYN-SENT* -> FIN-WAIT-1*
1236 */
1237 taop->tao_cc = to.to_cc;
1238 tp->t_starttime = ticks;
1239 if (tp->t_flags & TF_NEEDFIN) {
1240 tp->t_state = TCPS_FIN_WAIT_1;
1241 tp->t_flags &= ~TF_NEEDFIN;
1242 } else {
1243 tp->t_state = TCPS_ESTABLISHED;
1244 callout_reset(tp->tt_keep,
1245 tcp_keepidle,
1246 tcp_timer_keep,
1247 tp);
1248 }
1249 tp->t_flags |= TF_NEEDSYN;
1250 } else
1251 tp->t_state = TCPS_SYN_RECEIVED;
1252 } else {
1253 /* CC.NEW or no option => invalidate cache */
1254 taop->tao_cc = 0;
1255 tp->t_state = TCPS_SYN_RECEIVED;
1256 }
1257 }
1258
1259trimthenstep6:
1260 /*
1261 * Advance th->th_seq to correspond to first data byte.
1262 * If data, trim to stay within window,
1263 * dropping FIN if necessary.
1264 */
1265 th->th_seq++;
1266 if (tlen > tp->rcv_wnd) {
1267 todrop = tlen - tp->rcv_wnd;
1268 m_adj(m, -todrop);
1269 tlen = tp->rcv_wnd;
1270 thflags &= ~TH_FIN;
1271 tcpstat.tcps_rcvpackafterwin++;
1272 tcpstat.tcps_rcvbyteafterwin += todrop;
1273 }
1274 tp->snd_wl1 = th->th_seq - 1;
1275 tp->rcv_up = th->th_seq;
1276 /*
1277 * Client side of transaction: already sent SYN and data.
1278 * If the remote host used T/TCP to validate the SYN,
1279 * our data will be ACK'd; if so, enter normal data segment
1280 * processing in the middle of step 5, ack processing.
1281 * Otherwise, goto step 6.
1282 */
1283 if (thflags & TH_ACK)
1284 goto process_ACK;
1285
1286 goto step6;
1287
1288 /*
1289 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1290 * if segment contains a SYN and CC [not CC.NEW] option:
1291 * if state == TIME_WAIT and connection duration > MSL,
1292 * drop packet and send RST;
1293 *
1294 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1295 * ack the FIN (and data) in retransmission queue.
1296 * Complete close and delete TCPCB. Then reprocess
1297 * segment, hoping to find new TCPCB in LISTEN state;
1298 *
1299 * else must be old SYN; drop it.
1300 * else do normal processing.
1301 */
1302 case TCPS_LAST_ACK:
1303 case TCPS_CLOSING:
1304 case TCPS_TIME_WAIT:
1305 if ((thflags & TH_SYN) &&
1306 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1307 if (tp->t_state == TCPS_TIME_WAIT &&
1308 (ticks - tp->t_starttime) > tcp_msl) {
1309 rstreason = BANDLIM_UNLIMITED;
1310 goto dropwithreset;
1311 }
1312 if (CC_GT(to.to_cc, tp->cc_recv)) {
1313 tp = tcp_close(tp);
1314 goto findpcb;
1315 }
1316 else
1317 goto drop;
1318 }
1319 break; /* continue normal processing */
1320 }
1321
1322 /*
1323 * States other than LISTEN or SYN_SENT.
1324 * First check the RST flag and sequence number since reset segments
1325 * are exempt from the timestamp and connection count tests. This
1326 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1327 * below which allowed reset segments in half the sequence space
1328 * to fall though and be processed (which gives forged reset
1329 * segments with a random sequence number a 50 percent chance of
1330 * killing a connection).
1331 * Then check timestamp, if present.
1332 * Then check the connection count, if present.
1333 * Then check that at least some bytes of segment are within
1334 * receive window. If segment begins before rcv_nxt,
1335 * drop leading data (and SYN); if nothing left, just ack.
1336 *
1337 *
1338 * If the RST bit is set, check the sequence number to see
1339 * if this is a valid reset segment.
1340 * RFC 793 page 37:
1341 * In all states except SYN-SENT, all reset (RST) segments
1342 * are validated by checking their SEQ-fields. A reset is
1343 * valid if its sequence number is in the window.
1344 * Note: this does not take into account delayed ACKs, so
1345 * we should test against last_ack_sent instead of rcv_nxt.
1346 * The sequence number in the reset segment is normally an
1347 * echo of our outgoing acknowlegement numbers, but some hosts
1348 * send a reset with the sequence number at the rightmost edge
1349 * of our receive window, and we have to handle this case.
1350 * If we have multiple segments in flight, the intial reset
1351 * segment sequence numbers will be to the left of last_ack_sent,
1352 * but they will eventually catch up.
1353 * In any case, it never made sense to trim reset segments to
1354 * fit the receive window since RFC 1122 says:
1355 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1356 *
1357 * A TCP SHOULD allow a received RST segment to include data.
1358 *
1359 * DISCUSSION
1360 * It has been suggested that a RST segment could contain
1361 * ASCII text that encoded and explained the cause of the
1362 * RST. No standard has yet been established for such
1363 * data.
1364 *
1365 * If the reset segment passes the sequence number test examine
1366 * the state:
1367 * SYN_RECEIVED STATE:
1368 * If passive open, return to LISTEN state.
1369 * If active open, inform user that connection was refused.
1370 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1371 * Inform user that connection was reset, and close tcb.
1372 * CLOSING, LAST_ACK STATES:
1373 * Close the tcb.
1374 * TIME_WAIT STATE:
1375 * Drop the segment - see Stevens, vol. 2, p. 964 and
1376 * RFC 1337.
1377 */
1378 if (thflags & TH_RST) {
1379 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1380 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1381 switch (tp->t_state) {
1382
1383 case TCPS_SYN_RECEIVED:
1384 so->so_error = ECONNREFUSED;
1385 goto close;
1386
1387 case TCPS_ESTABLISHED:
1388 case TCPS_FIN_WAIT_1:
1389 case TCPS_FIN_WAIT_2:
1390 case TCPS_CLOSE_WAIT:
1391 so->so_error = ECONNRESET;
1392 close:
1393 tp->t_state = TCPS_CLOSED;
1394 tcpstat.tcps_drops++;
1395 tp = tcp_close(tp);
1396 break;
1397
1398 case TCPS_CLOSING:
1399 case TCPS_LAST_ACK:
1400 tp = tcp_close(tp);
1401 break;
1402
1403 case TCPS_TIME_WAIT:
1404 break;
1405 }
1406 }
1407 goto drop;
1408 }
1409
1410 /*
1411 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1412 * and it's less than ts_recent, drop it.
1413 */
1414 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1415 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1416
1417 /* Check to see if ts_recent is over 24 days old. */
1418 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1419 /*
1420 * Invalidate ts_recent. If this segment updates
1421 * ts_recent, the age will be reset later and ts_recent
1422 * will get a valid value. If it does not, setting
1423 * ts_recent to zero will at least satisfy the
1424 * requirement that zero be placed in the timestamp
1425 * echo reply when ts_recent isn't valid. The
1426 * age isn't reset until we get a valid ts_recent
1427 * because we don't want out-of-order segments to be
1428 * dropped when ts_recent is old.
1429 */
1430 tp->ts_recent = 0;
1431 } else {
1432 tcpstat.tcps_rcvduppack++;
1433 tcpstat.tcps_rcvdupbyte += tlen;
1434 tcpstat.tcps_pawsdrop++;
1435 if (tlen)
1436 goto dropafterack;
1437 goto drop;
1438 }
1439 }
1440
1441 /*
1442 * T/TCP mechanism
1443 * If T/TCP was negotiated and the segment doesn't have CC,
1444 * or if its CC is wrong then drop the segment.
1445 * RST segments do not have to comply with this.
1446 */
1447 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1448 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1449 goto dropafterack;
1450
1451 /*
1452 * In the SYN-RECEIVED state, validate that the packet belongs to
1453 * this connection before trimming the data to fit the receive
1454 * window. Check the sequence number versus IRS since we know
1455 * the sequence numbers haven't wrapped. This is a partial fix
1456 * for the "LAND" DoS attack.
1457 */
1458 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1459 rstreason = BANDLIM_RST_OPENPORT;
1460 goto dropwithreset;
1461 }
1462
1463 todrop = tp->rcv_nxt - th->th_seq;
1464 if (todrop > 0) {
1465 if (thflags & TH_SYN) {
1466 thflags &= ~TH_SYN;
1467 th->th_seq++;
1468 if (th->th_urp > 1)
1469 th->th_urp--;
1470 else
1471 thflags &= ~TH_URG;
1472 todrop--;
1473 }
1474 /*
1475 * Following if statement from Stevens, vol. 2, p. 960.
1476 */
1477 if (todrop > tlen
1478 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1479 /*
1480 * Any valid FIN must be to the left of the window.
1481 * At this point the FIN must be a duplicate or out
1482 * of sequence; drop it.
1483 */
1484 thflags &= ~TH_FIN;
1485
1486 /*
1487 * Send an ACK to resynchronize and drop any data.
1488 * But keep on processing for RST or ACK.
1489 */
1490 tp->t_flags |= TF_ACKNOW;
1491 todrop = tlen;
1492 tcpstat.tcps_rcvduppack++;
1493 tcpstat.tcps_rcvdupbyte += todrop;
1494 } else {
1495 tcpstat.tcps_rcvpartduppack++;
1496 tcpstat.tcps_rcvpartdupbyte += todrop;
1497 }
1498 drop_hdrlen += todrop; /* drop from the top afterwards */
1499 th->th_seq += todrop;
1500 tlen -= todrop;
1501 if (th->th_urp > todrop)
1502 th->th_urp -= todrop;
1503 else {
1504 thflags &= ~TH_URG;
1505 th->th_urp = 0;
1506 }
1507 }
1508
1509 /*
1510 * If new data are received on a connection after the
1511 * user processes are gone, then RST the other end.
1512 */
1513 if ((so->so_state & SS_NOFDREF) &&
1514 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1515 tp = tcp_close(tp);
1516 tcpstat.tcps_rcvafterclose++;
1517 rstreason = BANDLIM_UNLIMITED;
1518 goto dropwithreset;
1519 }
1520
1521 /*
1522 * If segment ends after window, drop trailing data
1523 * (and PUSH and FIN); if nothing left, just ACK.
1524 */
1525 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1526 if (todrop > 0) {
1527 tcpstat.tcps_rcvpackafterwin++;
1528 if (todrop >= tlen) {
1529 tcpstat.tcps_rcvbyteafterwin += tlen;
1530 /*
1531 * If a new connection request is received
1532 * while in TIME_WAIT, drop the old connection
1533 * and start over if the sequence numbers
1534 * are above the previous ones.
1535 */
1536 if (thflags & TH_SYN &&
1537 tp->t_state == TCPS_TIME_WAIT &&
1538 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1539 tp = tcp_close(tp);
1540 goto findpcb;
1541 }
1542 /*
1543 * If window is closed can only take segments at
1544 * window edge, and have to drop data and PUSH from
1545 * incoming segments. Continue processing, but
1546 * remember to ack. Otherwise, drop segment
1547 * and ack.
1548 */
1549 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1550 tp->t_flags |= TF_ACKNOW;
1551 tcpstat.tcps_rcvwinprobe++;
1552 } else
1553 goto dropafterack;
1554 } else
1555 tcpstat.tcps_rcvbyteafterwin += todrop;
1556 m_adj(m, -todrop);
1557 tlen -= todrop;
1558 thflags &= ~(TH_PUSH|TH_FIN);
1559 }
1560
1561 /*
1562 * If last ACK falls within this segment's sequence numbers,
1563 * record its timestamp.
1564 * NOTE that the test is modified according to the latest
1565 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1566 */
1567 if ((to.to_flags & TOF_TS) != 0 &&
1568 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1569 tp->ts_recent_age = ticks;
1570 tp->ts_recent = to.to_tsval;
1571 }
1572
1573 /*
1574 * If a SYN is in the window, then this is an
1575 * error and we send an RST and drop the connection.
1576 */
1577 if (thflags & TH_SYN) {
1578 tp = tcp_drop(tp, ECONNRESET);
1579 rstreason = BANDLIM_UNLIMITED;
1580 goto dropwithreset;
1581 }
1582
1583 /*
1584 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1585 * flag is on (half-synchronized state), then queue data for
1586 * later processing; else drop segment and return.
1587 */
1588 if ((thflags & TH_ACK) == 0) {
1589 if (tp->t_state == TCPS_SYN_RECEIVED ||
1590 (tp->t_flags & TF_NEEDSYN))
1591 goto step6;
1592 else
1593 goto drop;
1594 }
1595
1596 /*
1597 * Ack processing.
1598 */
1599 switch (tp->t_state) {
1600
1601 /*
1602 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1603 * ESTABLISHED state and continue processing.
1604 * The ACK was checked above.
1605 */
1606 case TCPS_SYN_RECEIVED:
1607
1608 tcpstat.tcps_connects++;
1609 soisconnected(so);
1610 /* Do window scaling? */
1611 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1612 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1613 tp->snd_scale = tp->requested_s_scale;
1614 tp->rcv_scale = tp->request_r_scale;
1615 }
1616 /*
1617 * Upon successful completion of 3-way handshake,
1618 * update cache.CC if it was undefined, pass any queued
1619 * data to the user, and advance state appropriately.
1620 */
1621 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1622 taop->tao_cc == 0)
1623 taop->tao_cc = tp->cc_recv;
1624
1625 /*
1626 * Make transitions:
1627 * SYN-RECEIVED -> ESTABLISHED
1628 * SYN-RECEIVED* -> FIN-WAIT-1
1629 */
1630 tp->t_starttime = ticks;
1631 if (tp->t_flags & TF_NEEDFIN) {
1632 tp->t_state = TCPS_FIN_WAIT_1;
1633 tp->t_flags &= ~TF_NEEDFIN;
1634 } else {
1635 tp->t_state = TCPS_ESTABLISHED;
1636 callout_reset(tp->tt_keep, tcp_keepidle,
1637 tcp_timer_keep, tp);
1638 }
1639 /*
1640 * If segment contains data or ACK, will call tcp_reass()
1641 * later; if not, do so now to pass queued data to user.
1642 */
1643 if (tlen == 0 && (thflags & TH_FIN) == 0)
1644 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1645 (struct mbuf *)0);
1646 tp->snd_wl1 = th->th_seq - 1;
1647 /* fall into ... */
1648
1649 /*
1650 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1651 * ACKs. If the ack is in the range
1652 * tp->snd_una < th->th_ack <= tp->snd_max
1653 * then advance tp->snd_una to th->th_ack and drop
1654 * data from the retransmission queue. If this ACK reflects
1655 * more up to date window information we update our window information.
1656 */
1657 case TCPS_ESTABLISHED:
1658 case TCPS_FIN_WAIT_1:
1659 case TCPS_FIN_WAIT_2:
1660 case TCPS_CLOSE_WAIT:
1661 case TCPS_CLOSING:
1662 case TCPS_LAST_ACK:
1663 case TCPS_TIME_WAIT:
1664
1665 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1666 if (tlen == 0 && tiwin == tp->snd_wnd) {
1667 tcpstat.tcps_rcvdupack++;
1668 /*
1669 * If we have outstanding data (other than
1670 * a window probe), this is a completely
1671 * duplicate ack (ie, window info didn't
1672 * change), the ack is the biggest we've
1673 * seen and we've seen exactly our rexmt
1674 * threshhold of them, assume a packet
1675 * has been dropped and retransmit it.
1676 * Kludge snd_nxt & the congestion
1677 * window so we send only this one
1678 * packet.
1679 *
1680 * We know we're losing at the current
1681 * window size so do congestion avoidance
1682 * (set ssthresh to half the current window
1683 * and pull our congestion window back to
1684 * the new ssthresh).
1685 *
1686 * Dup acks mean that packets have left the
1687 * network (they're now cached at the receiver)
1688 * so bump cwnd by the amount in the receiver
1689 * to keep a constant cwnd packets in the
1690 * network.
1691 */
1692 if (!callout_active(tp->tt_rexmt) ||
1693 th->th_ack != tp->snd_una)
1694 tp->t_dupacks = 0;
1695 else if (++tp->t_dupacks > tcprexmtthresh ||
1696 (tcp_do_newreno &&
9845754e 1697 IN_FASTRECOVERY(tp))) {
984263bc
MD
1698 tp->snd_cwnd += tp->t_maxseg;
1699 (void) tcp_output(tp);
1700 goto drop;
1701 } else if (tp->t_dupacks == tcprexmtthresh) {
1702 tcp_seq onxt = tp->snd_nxt;
1703 u_int win;
1704 if (tcp_do_newreno &&
9845754e
MD
1705 SEQ_LEQ(th->th_ack,
1706 tp->snd_recover)) {
984263bc
MD
1707 tp->t_dupacks = 0;
1708 break;
1709 }
efd4b327
JH
1710 if (tcp_do_eifel_detect &&
1711 (tp->t_flags & TF_RCVD_TSTMP)) {
1712 tcp_save_congestion_state(tp);
1713 tp->t_flags |= TF_FASTREXMT;
1714 }
984263bc
MD
1715 win = min(tp->snd_wnd, tp->snd_cwnd) /
1716 2 / tp->t_maxseg;
1717 if (win < 2)
1718 win = 2;
1719 tp->snd_ssthresh = win * tp->t_maxseg;
9845754e 1720 ENTER_FASTRECOVERY(tp);
984263bc
MD
1721 tp->snd_recover = tp->snd_max;
1722 callout_stop(tp->tt_rexmt);
1723 tp->t_rtttime = 0;
1724 tp->snd_nxt = th->th_ack;
1725 tp->snd_cwnd = tp->t_maxseg;
1726 (void) tcp_output(tp);
33abdd1c
MD
1727 KASSERT(tp->snd_limited <= 2,
1728 ("tp->snd_limited too big"));
984263bc 1729 tp->snd_cwnd = tp->snd_ssthresh +
33abdd1c
MD
1730 (tp->t_maxseg *
1731 (tp->t_dupacks - tp->snd_limited));
984263bc
MD
1732 if (SEQ_GT(onxt, tp->snd_nxt))
1733 tp->snd_nxt = onxt;
1734 goto drop;
33abdd1c
MD
1735 } else if (tcp_do_limitedtransmit) {
1736 u_long oldcwnd = tp->snd_cwnd;
1737 tcp_seq oldsndmax = tp->snd_max;
1738 u_int sent;
1739 KASSERT(tp->t_dupacks == 1 ||
1740 tp->t_dupacks == 2,
1741 ("dupacks not 1 or 2"));
1742 if (tp->t_dupacks == 1) {
1743 tp->snd_limited = 0;
1744 tp->snd_cwnd += tp->t_maxseg;
1745 } else {
1746 tp->snd_cwnd +=
1747 tp->t_maxseg * 2;
1748 }
1749 (void) tcp_output(tp);
1750 sent = tp->snd_max - oldsndmax;
1751 if (sent > tp->t_maxseg) {
1752 KASSERT(tp->snd_limited == 0 &&
1753 tp->t_dupacks == 2,
1754 ("sent too much"));
1755 tp->snd_limited = 2;
1756 } else if (sent > 0)
1757 ++tp->snd_limited;
1758 tp->snd_cwnd = oldcwnd;
1759 goto drop;
984263bc
MD
1760 }
1761 } else
1762 tp->t_dupacks = 0;
1763 break;
1764 }
1765
1766 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1767
1768 /*
1769 * If the congestion window was inflated to account
1770 * for the other side's cached packets, retract it.
1771 */
1772 if (tcp_do_newreno) {
9845754e 1773 if (IN_FASTRECOVERY(tp)) {
984263bc
MD
1774 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1775 tcp_newreno_partial_ack(tp, th);
1776 } else {
1777 /*
1778 * Window inflation should have left us
1779 * with approximately snd_ssthresh
1780 * outstanding data.
1781 * But in case we would be inclined to
1782 * send a burst, better to do it via
1783 * the slow start mechanism.
1784 */
1785 if (SEQ_GT(th->th_ack +
1786 tp->snd_ssthresh,
1787 tp->snd_max))
1788 tp->snd_cwnd = tp->snd_max -
1789 th->th_ack +
1790 tp->t_maxseg;
1791 else
1792 tp->snd_cwnd = tp->snd_ssthresh;
1793 }
1794 }
1795 } else {
1796 if (tp->t_dupacks >= tcprexmtthresh &&
1797 tp->snd_cwnd > tp->snd_ssthresh)
1798 tp->snd_cwnd = tp->snd_ssthresh;
1799 }
1800 tp->t_dupacks = 0;
1801 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1802 tcpstat.tcps_rcvacktoomuch++;
1803 goto dropafterack;
1804 }
1805 /*
1806 * If we reach this point, ACK is not a duplicate,
1807 * i.e., it ACKs something we sent.
1808 */
1809 if (tp->t_flags & TF_NEEDSYN) {
1810 /*
1811 * T/TCP: Connection was half-synchronized, and our
1812 * SYN has been ACK'd (so connection is now fully
1813 * synchronized). Go to non-starred state,
1814 * increment snd_una for ACK of SYN, and check if
1815 * we can do window scaling.
1816 */
1817 tp->t_flags &= ~TF_NEEDSYN;
1818 tp->snd_una++;
1819 /* Do window scaling? */
1820 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1821 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1822 tp->snd_scale = tp->requested_s_scale;
1823 tp->rcv_scale = tp->request_r_scale;
1824 }
1825 }
1826
1827process_ACK:
1828 acked = th->th_ack - tp->snd_una;
1829 tcpstat.tcps_rcvackpack++;
1830 tcpstat.tcps_rcvackbyte += acked;
1831
1832 /*
1833 * If we just performed our first retransmit, and the ACK
1834 * arrives within our recovery window, then it was a mistake
1835 * to do the retransmit in the first place. Recover our
1836 * original cwnd and ssthresh, and proceed to transmit where
1837 * we left off.
1838 */
efd4b327
JH
1839 useTS = tcp_do_eifel_detect && (to.to_flags & TOF_TS) &&
1840 to.to_tsecr;
1841 if ((useTS && (tp->t_flags & TF_FIRSTACCACK) && acked &&
1842 (to.to_tsecr < tp->t_rexmtTS)) ||
1843 (!useTS &&
1844 (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin))) {
984263bc
MD
1845 tp->snd_cwnd = tp->snd_cwnd_prev;
1846 tp->snd_ssthresh = tp->snd_ssthresh_prev;
9845754e
MD
1847 tp->snd_recover = tp->snd_recover_prev;
1848 if (tp->t_flags & TF_WASFRECOVERY)
1849 ENTER_FASTRECOVERY(tp);
984263bc 1850 tp->snd_nxt = tp->snd_max;
efd4b327
JH
1851 tp->t_badrxtwin = 0; /* XXX probably not required */
1852 tp->t_rxtshift = 0;
1853 if (tp->t_flags & TF_FASTREXMT)
1854 ++tcpstat.tcps_sndfastrexmitbad;
1855 else
1856 ++tcpstat.tcps_sndrtobad;
984263bc
MD
1857 }
1858
1859 /*
1860 * If we have a timestamp reply, update smoothed
1861 * round trip time. If no timestamp is present but
1862 * transmit timer is running and timed sequence
1863 * number was acked, update smoothed round trip time.
1864 * Since we now have an rtt measurement, cancel the
1865 * timer backoff (cf., Phil Karn's retransmit alg.).
1866 * Recompute the initial retransmit timer.
1867 *
1868 * Some machines (certain windows boxes) send broken
1869 * timestamp replies during the SYN+ACK phase, ignore
1870 * timestamps of 0.
1871 */
1872 if ((to.to_flags & TOF_TS) != 0 &&
1873 to.to_tsecr) {
1874 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1875 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1876 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1877 }
1878 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1879
1880 /*
1881 * If all outstanding data is acked, stop retransmit
1882 * timer and remember to restart (more output or persist).
1883 * If there is more data to be acked, restart retransmit
1884 * timer, using current (possibly backed-off) value.
1885 */
1886 if (th->th_ack == tp->snd_max) {
1887 callout_stop(tp->tt_rexmt);
1888 needoutput = 1;
1889 } else if (!callout_active(tp->tt_persist))
1890 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1891 tcp_timer_rexmt, tp);
1892
1893 /*
1894 * If no data (only SYN) was ACK'd,
1895 * skip rest of ACK processing.
1896 */
1897 if (acked == 0)
1898 goto step6;
1899
efd4b327
JH
1900 /* Stop looking for an acceptable ACK since one was received. */
1901 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
1902
984263bc
MD
1903 /*
1904 * When new data is acked, open the congestion window.
1905 * If the window gives us less than ssthresh packets
1906 * in flight, open exponentially (maxseg per packet).
1907 * Otherwise open linearly: maxseg per window
1908 * (maxseg^2 / cwnd per packet).
1909 */
9845754e 1910 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
2256ba69
RG
1911 u_int cw = tp->snd_cwnd;
1912 u_int incr = tp->t_maxseg;
984263bc
MD
1913 if (cw > tp->snd_ssthresh)
1914 incr = incr * incr / cw;
1915 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1916 }
1917 if (acked > so->so_snd.sb_cc) {
1918 tp->snd_wnd -= so->so_snd.sb_cc;
1919 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1920 ourfinisacked = 1;
1921 } else {
1922 sbdrop(&so->so_snd, acked);
1923 tp->snd_wnd -= acked;
1924 ourfinisacked = 0;
1925 }
1926 sowwakeup(so);
1927 /* detect una wraparound */
9845754e
MD
1928 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1929 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1930 SEQ_LEQ(th->th_ack, tp->snd_recover))
1931 tp->snd_recover = th->th_ack - 1;
1932 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1933 SEQ_GEQ(th->th_ack, tp->snd_recover))
1934 EXIT_FASTRECOVERY(tp);
984263bc
MD
1935 tp->snd_una = th->th_ack;
1936 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1937 tp->snd_nxt = tp->snd_una;
1938
1939 switch (tp->t_state) {
1940
1941 /*
1942 * In FIN_WAIT_1 STATE in addition to the processing
1943 * for the ESTABLISHED state if our FIN is now acknowledged
1944 * then enter FIN_WAIT_2.
1945 */
1946 case TCPS_FIN_WAIT_1:
1947 if (ourfinisacked) {
1948 /*
1949 * If we can't receive any more
1950 * data, then closing user can proceed.
1951 * Starting the timer is contrary to the
1952 * specification, but if we don't get a FIN
1953 * we'll hang forever.
1954 */
1955 if (so->so_state & SS_CANTRCVMORE) {
1956 soisdisconnected(so);
1957 callout_reset(tp->tt_2msl, tcp_maxidle,
1958 tcp_timer_2msl, tp);
1959 }
1960 tp->t_state = TCPS_FIN_WAIT_2;
1961 }
1962 break;
1963
1964 /*
1965 * In CLOSING STATE in addition to the processing for
1966 * the ESTABLISHED state if the ACK acknowledges our FIN
1967 * then enter the TIME-WAIT state, otherwise ignore
1968 * the segment.
1969 */
1970 case TCPS_CLOSING:
1971 if (ourfinisacked) {
1972 tp->t_state = TCPS_TIME_WAIT;
1973 tcp_canceltimers(tp);
1974 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1975 if (tp->cc_recv != 0 &&
1976 (ticks - tp->t_starttime) < tcp_msl)
1977 callout_reset(tp->tt_2msl,
1978 tp->t_rxtcur *
1979 TCPTV_TWTRUNC,
1980 tcp_timer_2msl, tp);
1981 else
1982 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1983 tcp_timer_2msl, tp);
1984 soisdisconnected(so);
1985 }
1986 break;
1987
1988 /*
1989 * In LAST_ACK, we may still be waiting for data to drain
1990 * and/or to be acked, as well as for the ack of our FIN.
1991 * If our FIN is now acknowledged, delete the TCB,
1992 * enter the closed state and return.
1993 */
1994 case TCPS_LAST_ACK:
1995 if (ourfinisacked) {
1996 tp = tcp_close(tp);
1997 goto drop;
1998 }
1999 break;
2000
2001 /*
2002 * In TIME_WAIT state the only thing that should arrive
2003 * is a retransmission of the remote FIN. Acknowledge
2004 * it and restart the finack timer.
2005 */
2006 case TCPS_TIME_WAIT:
2007 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2008 tcp_timer_2msl, tp);
2009 goto dropafterack;
2010 }
2011 }
2012
2013step6:
2014 /*
2015 * Update window information.
2016 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2017 */
2018 if ((thflags & TH_ACK) &&
2019 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2020 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2021 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2022 /* keep track of pure window updates */
2023 if (tlen == 0 &&
2024 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2025 tcpstat.tcps_rcvwinupd++;
2026 tp->snd_wnd = tiwin;
2027 tp->snd_wl1 = th->th_seq;
2028 tp->snd_wl2 = th->th_ack;
2029 if (tp->snd_wnd > tp->max_sndwnd)
2030 tp->max_sndwnd = tp->snd_wnd;
2031 needoutput = 1;
2032 }
2033
2034 /*
2035 * Process segments with URG.
2036 */
2037 if ((thflags & TH_URG) && th->th_urp &&
2038 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2039 /*
2040 * This is a kludge, but if we receive and accept
2041 * random urgent pointers, we'll crash in
2042 * soreceive. It's hard to imagine someone
2043 * actually wanting to send this much urgent data.
2044 */
2045 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2046 th->th_urp = 0; /* XXX */
2047 thflags &= ~TH_URG; /* XXX */
2048 goto dodata; /* XXX */
2049 }
2050 /*
2051 * If this segment advances the known urgent pointer,
2052 * then mark the data stream. This should not happen
2053 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2054 * a FIN has been received from the remote side.
2055 * In these states we ignore the URG.
2056 *
2057 * According to RFC961 (Assigned Protocols),
2058 * the urgent pointer points to the last octet
2059 * of urgent data. We continue, however,
2060 * to consider it to indicate the first octet
2061 * of data past the urgent section as the original
2062 * spec states (in one of two places).
2063 */
2064 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2065 tp->rcv_up = th->th_seq + th->th_urp;
2066 so->so_oobmark = so->so_rcv.sb_cc +
2067 (tp->rcv_up - tp->rcv_nxt) - 1;
2068 if (so->so_oobmark == 0)
2069 so->so_state |= SS_RCVATMARK;
2070 sohasoutofband(so);
2071 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2072 }
2073 /*
2074 * Remove out of band data so doesn't get presented to user.
2075 * This can happen independent of advancing the URG pointer,
2076 * but if two URG's are pending at once, some out-of-band
2077 * data may creep in... ick.
2078 */
2079 if (th->th_urp <= (u_long)tlen
2080#ifdef SO_OOBINLINE
2081 && (so->so_options & SO_OOBINLINE) == 0
2082#endif
2083 )
2084 tcp_pulloutofband(so, th, m,
2085 drop_hdrlen); /* hdr drop is delayed */
2086 } else {
2087 /*
2088 * If no out of band data is expected,
2089 * pull receive urgent pointer along
2090 * with the receive window.
2091 */
2092 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2093 tp->rcv_up = tp->rcv_nxt;
2094 }
2095dodata: /* XXX */
2096
2097 /*
2098 * Process the segment text, merging it into the TCP sequencing queue,
2099 * and arranging for acknowledgment of receipt if necessary.
2100 * This process logically involves adjusting tp->rcv_wnd as data
2101 * is presented to the user (this happens in tcp_usrreq.c,
2102 * case PRU_RCVD). If a FIN has already been received on this
2103 * connection then we just ignore the text.
2104 */
2105 if ((tlen || (thflags & TH_FIN)) &&
2106 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2107 m_adj(m, drop_hdrlen); /* delayed header drop */
2108 /*
2109 * Insert segment which includes th into TCP reassembly queue
2110 * with control block tp. Set thflags to whether reassembly now
2111 * includes a segment with FIN. This handles the common case
2112 * inline (segment is the next to be received on an established
2113 * connection, and the queue is empty), avoiding linkage into
2114 * and removal from the queue and repetition of various
2115 * conversions.
2116 * Set DELACK for segments received in order, but ack
2117 * immediately when segments are out of order (so
2118 * fast retransmit can work).
2119 */
2120 if (th->th_seq == tp->rcv_nxt &&
2121 LIST_EMPTY(&tp->t_segq) &&
2122 TCPS_HAVEESTABLISHED(tp->t_state)) {
2123 if (DELAY_ACK(tp))
2124 callout_reset(tp->tt_delack, tcp_delacktime,
2125 tcp_timer_delack, tp);
2126 else
2127 tp->t_flags |= TF_ACKNOW;
2128 tp->rcv_nxt += tlen;
2129 thflags = th->th_flags & TH_FIN;
2130 tcpstat.tcps_rcvpack++;
2131 tcpstat.tcps_rcvbyte += tlen;
2132 ND6_HINT(tp);
2133 if (so->so_state & SS_CANTRCVMORE)
2134 m_freem(m);
2135 else
2136 sbappend(&so->so_rcv, m);
2137 sorwakeup(so);
2138 } else {
2139 thflags = tcp_reass(tp, th, &tlen, m);
2140 tp->t_flags |= TF_ACKNOW;
2141 }
2142
2143 /*
2144 * Note the amount of data that peer has sent into
2145 * our window, in order to estimate the sender's
2146 * buffer size.
2147 */
2148 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2149 } else {
2150 m_freem(m);
2151 thflags &= ~TH_FIN;
2152 }
2153
2154 /*
2155 * If FIN is received ACK the FIN and let the user know
2156 * that the connection is closing.
2157 */
2158 if (thflags & TH_FIN) {
2159 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2160 socantrcvmore(so);
2161 /*
2162 * If connection is half-synchronized
2163 * (ie NEEDSYN flag on) then delay ACK,
2164 * so it may be piggybacked when SYN is sent.
2165 * Otherwise, since we received a FIN then no
2166 * more input can be expected, send ACK now.
2167 */
2168 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2169 callout_reset(tp->tt_delack, tcp_delacktime,
2170 tcp_timer_delack, tp);
2171 else
2172 tp->t_flags |= TF_ACKNOW;
2173 tp->rcv_nxt++;
2174 }
2175 switch (tp->t_state) {
2176
2177 /*
2178 * In SYN_RECEIVED and ESTABLISHED STATES
2179 * enter the CLOSE_WAIT state.
2180 */
2181 case TCPS_SYN_RECEIVED:
2182 tp->t_starttime = ticks;
2183 /*FALLTHROUGH*/
2184 case TCPS_ESTABLISHED:
2185 tp->t_state = TCPS_CLOSE_WAIT;
2186 break;
2187
2188 /*
2189 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2190 * enter the CLOSING state.
2191 */
2192 case TCPS_FIN_WAIT_1:
2193 tp->t_state = TCPS_CLOSING;
2194 break;
2195
2196 /*
2197 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2198 * starting the time-wait timer, turning off the other
2199 * standard timers.
2200 */
2201 case TCPS_FIN_WAIT_2:
2202 tp->t_state = TCPS_TIME_WAIT;
2203 tcp_canceltimers(tp);
2204 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2205 if (tp->cc_recv != 0 &&
2206 (ticks - tp->t_starttime) < tcp_msl) {
2207 callout_reset(tp->tt_2msl,
2208 tp->t_rxtcur * TCPTV_TWTRUNC,
2209 tcp_timer_2msl, tp);
2210 /* For transaction client, force ACK now. */
2211 tp->t_flags |= TF_ACKNOW;
2212 }
2213 else
2214 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2215 tcp_timer_2msl, tp);
2216 soisdisconnected(so);
2217 break;
2218
2219 /*
2220 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2221 */
2222 case TCPS_TIME_WAIT:
2223 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2224 tcp_timer_2msl, tp);
2225 break;
2226 }
2227 }
2228#ifdef TCPDEBUG
2229 if (so->so_options & SO_DEBUG)
2230 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2231 &tcp_savetcp, 0);
2232#endif
2233
2234 /*
2235 * Return any desired output.
2236 */
2237 if (needoutput || (tp->t_flags & TF_ACKNOW))
2238 (void) tcp_output(tp);
2239 return;
2240
2241dropafterack:
2242 /*
2243 * Generate an ACK dropping incoming segment if it occupies
2244 * sequence space, where the ACK reflects our state.
2245 *
2246 * We can now skip the test for the RST flag since all
2247 * paths to this code happen after packets containing
2248 * RST have been dropped.
2249 *
2250 * In the SYN-RECEIVED state, don't send an ACK unless the
2251 * segment we received passes the SYN-RECEIVED ACK test.
2252 * If it fails send a RST. This breaks the loop in the
2253 * "LAND" DoS attack, and also prevents an ACK storm
2254 * between two listening ports that have been sent forged
2255 * SYN segments, each with the source address of the other.
2256 */
2257 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2258 (SEQ_GT(tp->snd_una, th->th_ack) ||
2259 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2260 rstreason = BANDLIM_RST_OPENPORT;
2261 goto dropwithreset;
2262 }
2263#ifdef TCPDEBUG
2264 if (so->so_options & SO_DEBUG)
2265 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2266 &tcp_savetcp, 0);
2267#endif
2268 m_freem(m);
2269 tp->t_flags |= TF_ACKNOW;
2270 (void) tcp_output(tp);
2271 return;
2272
2273dropwithreset:
2274 /*
2275 * Generate a RST, dropping incoming segment.
2276 * Make ACK acceptable to originator of segment.
2277 * Don't bother to respond if destination was broadcast/multicast.
2278 */
2279 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2280 goto drop;
2281 if (isipv6) {
2282 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2283 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2284 goto drop;
2285 } else {
2286 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2287 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2288 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2289 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2290 goto drop;
2291 }
2292 /* IPv6 anycast check is done at tcp6_input() */
2293
2294 /*
2295 * Perform bandwidth limiting.
2296 */
2297#ifdef ICMP_BANDLIM
2298 if (badport_bandlim(rstreason) < 0)
2299 goto drop;
2300#endif
2301
2302#ifdef TCPDEBUG
2303 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2304 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2305 &tcp_savetcp, 0);
2306#endif
2307 if (thflags & TH_ACK)
2308 /* mtod() below is safe as long as hdr dropping is delayed */
2309 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2310 TH_RST);
2311 else {
2312 if (thflags & TH_SYN)
2313 tlen++;
2314 /* mtod() below is safe as long as hdr dropping is delayed */
2315 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2316 (tcp_seq)0, TH_RST|TH_ACK);
2317 }
2318 return;
2319
2320drop:
2321 /*
2322 * Drop space held by incoming segment and return.
2323 */
2324#ifdef TCPDEBUG
2325 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2326 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2327 &tcp_savetcp, 0);
2328#endif
2329 m_freem(m);
2330 return;
2331}
2332
2333/*
2334 * Parse TCP options and place in tcpopt.
2335 */
2336static void
2337tcp_dooptions(to, cp, cnt, is_syn)
2338 struct tcpopt *to;
2339 u_char *cp;
2340 int cnt;
2341{
2342 int opt, optlen;
2343
2344 to->to_flags = 0;
2345 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2346 opt = cp[0];
2347 if (opt == TCPOPT_EOL)
2348 break;
2349 if (opt == TCPOPT_NOP)
2350 optlen = 1;
2351 else {
2352 if (cnt < 2)
2353 break;
2354 optlen = cp[1];
2355 if (optlen < 2 || optlen > cnt)
2356 break;
2357 }
2358 switch (opt) {
2359 case TCPOPT_MAXSEG:
2360 if (optlen != TCPOLEN_MAXSEG)
2361 continue;
2362 if (!is_syn)
2363 continue;
2364 to->to_flags |= TOF_MSS;
2365 bcopy((char *)cp + 2,
2366 (char *)&to->to_mss, sizeof(to->to_mss));
2367 to->to_mss = ntohs(to->to_mss);
2368 break;
2369 case TCPOPT_WINDOW:
2370 if (optlen != TCPOLEN_WINDOW)
2371 continue;
2372 if (! is_syn)
2373 continue;
2374 to->to_flags |= TOF_SCALE;
2375 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2376 break;
2377 case TCPOPT_TIMESTAMP:
2378 if (optlen != TCPOLEN_TIMESTAMP)
2379 continue;
2380 to->to_flags |= TOF_TS;
2381 bcopy((char *)cp + 2,
2382 (char *)&to->to_tsval, sizeof(to->to_tsval));
2383 to->to_tsval = ntohl(to->to_tsval);
2384 bcopy((char *)cp + 6,
2385 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2386 to->to_tsecr = ntohl(to->to_tsecr);
2387 break;
2388 case TCPOPT_CC:
2389 if (optlen != TCPOLEN_CC)
2390 continue;
2391 to->to_flags |= TOF_CC;
2392 bcopy((char *)cp + 2,
2393 (char *)&to->to_cc, sizeof(to->to_cc));
2394 to->to_cc = ntohl(to->to_cc);
2395 break;
2396 case TCPOPT_CCNEW:
2397 if (optlen != TCPOLEN_CC)
2398 continue;
2399 if (!is_syn)
2400 continue;
2401 to->to_flags |= TOF_CCNEW;
2402 bcopy((char *)cp + 2,
2403 (char *)&to->to_cc, sizeof(to->to_cc));
2404 to->to_cc = ntohl(to->to_cc);
2405 break;
2406 case TCPOPT_CCECHO:
2407 if (optlen != TCPOLEN_CC)
2408 continue;
2409 if (!is_syn)
2410 continue;
2411 to->to_flags |= TOF_CCECHO;
2412 bcopy((char *)cp + 2,
2413 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2414 to->to_ccecho = ntohl(to->to_ccecho);
2415 break;
2416 default:
2417 continue;
2418 }
2419 }
2420}
2421
2422/*
2423 * Pull out of band byte out of a segment so
2424 * it doesn't appear in the user's data queue.
2425 * It is still reflected in the segment length for
2426 * sequencing purposes.
2427 */
2428static void
2429tcp_pulloutofband(so, th, m, off)
2430 struct socket *so;
2431 struct tcphdr *th;
2256ba69 2432 struct mbuf *m;
984263bc
MD
2433 int off; /* delayed to be droped hdrlen */
2434{
2435 int cnt = off + th->th_urp - 1;
2436
2437 while (cnt >= 0) {
2438 if (m->m_len > cnt) {
2439 char *cp = mtod(m, caddr_t) + cnt;
2440 struct tcpcb *tp = sototcpcb(so);
2441
2442 tp->t_iobc = *cp;
2443 tp->t_oobflags |= TCPOOB_HAVEDATA;
2444 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2445 m->m_len--;
2446 if (m->m_flags & M_PKTHDR)
2447 m->m_pkthdr.len--;
2448 return;
2449 }
2450 cnt -= m->m_len;
2451 m = m->m_next;
2452 if (m == 0)
2453 break;
2454 }
2455 panic("tcp_pulloutofband");
2456}
2457
2458/*
2459 * Collect new round-trip time estimate
2460 * and update averages and current timeout.
2461 */
2462static void
2463tcp_xmit_timer(tp, rtt)
2256ba69 2464 struct tcpcb *tp;
984263bc
MD
2465 int rtt;
2466{
2256ba69 2467 int delta;
984263bc
MD
2468
2469 tcpstat.tcps_rttupdated++;
2470 tp->t_rttupdated++;
2471 if (tp->t_srtt != 0) {
2472 /*
2473 * srtt is stored as fixed point with 5 bits after the
2474 * binary point (i.e., scaled by 8). The following magic
2475 * is equivalent to the smoothing algorithm in rfc793 with
2476 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2477 * point). Adjust rtt to origin 0.
2478 */
2479 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2480 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2481
2482 if ((tp->t_srtt += delta) <= 0)
2483 tp->t_srtt = 1;
2484
2485 /*
2486 * We accumulate a smoothed rtt variance (actually, a
2487 * smoothed mean difference), then set the retransmit
2488 * timer to smoothed rtt + 4 times the smoothed variance.
2489 * rttvar is stored as fixed point with 4 bits after the
2490 * binary point (scaled by 16). The following is
2491 * equivalent to rfc793 smoothing with an alpha of .75
2492 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2493 * rfc793's wired-in beta.
2494 */
2495 if (delta < 0)
2496 delta = -delta;
2497 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2498 if ((tp->t_rttvar += delta) <= 0)
2499 tp->t_rttvar = 1;
2500 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2501 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2502 } else {
2503 /*
2504 * No rtt measurement yet - use the unsmoothed rtt.
2505 * Set the variance to half the rtt (so our first
2506 * retransmit happens at 3*rtt).
2507 */
2508 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2509 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2510 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2511 }
2512 tp->t_rtttime = 0;
2513 tp->t_rxtshift = 0;
2514
2515 /*
2516 * the retransmit should happen at rtt + 4 * rttvar.
2517 * Because of the way we do the smoothing, srtt and rttvar
2518 * will each average +1/2 tick of bias. When we compute
2519 * the retransmit timer, we want 1/2 tick of rounding and
2520 * 1 extra tick because of +-1/2 tick uncertainty in the
2521 * firing of the timer. The bias will give us exactly the
2522 * 1.5 tick we need. But, because the bias is
2523 * statistical, we have to test that we don't drop below
2524 * the minimum feasible timer (which is 2 ticks).
2525 */
2526 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2527 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2528
2529 /*
2530 * We received an ack for a packet that wasn't retransmitted;
2531 * it is probably safe to discard any error indications we've
2532 * received recently. This isn't quite right, but close enough
2533 * for now (a route might have failed after we sent a segment,
2534 * and the return path might not be symmetrical).
2535 */
2536 tp->t_softerror = 0;
2537}
2538
2539/*
2540 * Determine a reasonable value for maxseg size.
2541 * If the route is known, check route for mtu.
2542 * If none, use an mss that can be handled on the outgoing
2543 * interface without forcing IP to fragment; if bigger than
2544 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2545 * to utilize large mbufs. If no route is found, route has no mtu,
2546 * or the destination isn't local, use a default, hopefully conservative
2547 * size (usually 512 or the default IP max size, but no more than the mtu
2548 * of the interface), as we can't discover anything about intervening
2549 * gateways or networks. We also initialize the congestion/slow start
2550 * window to be a single segment if the destination isn't local.
2551 * While looking at the routing entry, we also initialize other path-dependent
2552 * parameters from pre-set or cached values in the routing entry.
2553 *
2554 * Also take into account the space needed for options that we
2555 * send regularly. Make maxseg shorter by that amount to assure
2556 * that we can send maxseg amount of data even when the options
2557 * are present. Store the upper limit of the length of options plus
2558 * data in maxopd.
2559 *
2560 * NOTE that this routine is only called when we process an incoming
2561 * segment, for outgoing segments only tcp_mssopt is called.
2562 *
2563 * In case of T/TCP, we call this routine during implicit connection
2564 * setup as well (offer = -1), to initialize maxseg from the cached
2565 * MSS of our peer.
2566 */
2567void
2568tcp_mss(tp, offer)
2569 struct tcpcb *tp;
2570 int offer;
2571{
2256ba69 2572 struct rtentry *rt;
984263bc 2573 struct ifnet *ifp;
2256ba69 2574 int rtt, mss;
984263bc
MD
2575 u_long bufsize;
2576 struct inpcb *inp = tp->t_inpcb;
2577 struct socket *so;
2578 struct rmxp_tao *taop;
2579 int origoffer = offer;
2580#ifdef INET6
2581 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2582 size_t min_protoh = isipv6 ?
2583 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2584 sizeof(struct tcpiphdr);
2585#else
2586 const int isipv6 = 0;
2587 const size_t min_protoh = sizeof(struct tcpiphdr);
2588#endif
2589
2590 if (isipv6)
2591 rt = tcp_rtlookup6(&inp->inp_inc);
2592 else
2593 rt = tcp_rtlookup(&inp->inp_inc);
2594 if (rt == NULL) {
2595 tp->t_maxopd = tp->t_maxseg =
2596 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2597 return;
2598 }
2599 ifp = rt->rt_ifp;
2600 so = inp->inp_socket;
2601
2602 taop = rmx_taop(rt->rt_rmx);
2603 /*
2604 * Offer == -1 means that we didn't receive SYN yet,
2605 * use cached value in that case;
2606 */
2607 if (offer == -1)
2608 offer = taop->tao_mssopt;
2609 /*
2610 * Offer == 0 means that there was no MSS on the SYN segment,
2611 * in this case we use tcp_mssdflt.
2612 */
2613 if (offer == 0)
2614 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2615 else
2616 /*
2617 * Sanity check: make sure that maxopd will be large
2618 * enough to allow some data on segments even is the
2619 * all the option space is used (40bytes). Otherwise
2620 * funny things may happen in tcp_output.
2621 */
2622 offer = max(offer, 64);
2623 taop->tao_mssopt = offer;
2624
2625 /*
2626 * While we're here, check if there's an initial rtt
2627 * or rttvar. Convert from the route-table units
2628 * to scaled multiples of the slow timeout timer.
2629 */
2630 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2631 /*
2632 * XXX the lock bit for RTT indicates that the value
2633 * is also a minimum value; this is subject to time.
2634 */
2635 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2636 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2637 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2638 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2639 tcpstat.tcps_usedrtt++;
2640 if (rt->rt_rmx.rmx_rttvar) {
2641 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2642 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2643 tcpstat.tcps_usedrttvar++;
2644 } else {
2645 /* default variation is +- 1 rtt */
2646 tp->t_rttvar =
2647 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2648 }
2649 TCPT_RANGESET(tp->t_rxtcur,
2650 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2651 tp->t_rttmin, TCPTV_REXMTMAX);
2652 }
2653 /*
2654 * if there's an mtu associated with the route, use it
2655 * else, use the link mtu.
2656 */
2657 if (rt->rt_rmx.rmx_mtu)
2658 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2659 else {
2660 if (isipv6) {
2661 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2662 min_protoh;
2663 if (!in6_localaddr(&inp->in6p_faddr))
2664 mss = min(mss, tcp_v6mssdflt);
2665 } else {
2666 mss = ifp->if_mtu - min_protoh;
2667 if (!in_localaddr(inp->inp_faddr))
2668 mss = min(mss, tcp_mssdflt);
2669 }
2670 }
2671 mss = min(mss, offer);
2672 /*
2673 * maxopd stores the maximum length of data AND options
2674 * in a segment; maxseg is the amount of data in a normal
2675 * segment. We need to store this value (maxopd) apart
2676 * from maxseg, because now every segment carries options
2677 * and thus we normally have somewhat less data in segments.
2678 */
2679 tp->t_maxopd = mss;
2680
2681 /*
2682 * In case of T/TCP, origoffer==-1 indicates, that no segments
2683 * were received yet. In this case we just guess, otherwise
2684 * we do the same as before T/TCP.
2685 */
2686 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2687 (origoffer == -1 ||
2688 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2689 mss -= TCPOLEN_TSTAMP_APPA;
2690 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2691 (origoffer == -1 ||
2692 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2693 mss -= TCPOLEN_CC_APPA;
2694
2695#if (MCLBYTES & (MCLBYTES - 1)) == 0
2696 if (mss > MCLBYTES)
2697 mss &= ~(MCLBYTES-1);
2698#else
2699 if (mss > MCLBYTES)
2700 mss = mss / MCLBYTES * MCLBYTES;
2701#endif
2702 /*
2703 * If there's a pipesize, change the socket buffer
2704 * to that size. Make the socket buffers an integral
2705 * number of mss units; if the mss is larger than
2706 * the socket buffer, decrease the mss.
2707 */
2708#ifdef RTV_SPIPE
2709 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2710#endif
2711 bufsize = so->so_snd.sb_hiwat;
2712 if (bufsize < mss)
2713 mss = bufsize;
2714 else {
2715 bufsize = roundup(bufsize, mss);
2716 if (bufsize > sb_max)
2717 bufsize = sb_max;
2718 if (bufsize > so->so_snd.sb_hiwat)
2719 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2720 }
2721 tp->t_maxseg = mss;
2722
2723#ifdef RTV_RPIPE
2724 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2725#endif
2726 bufsize = so->so_rcv.sb_hiwat;
2727 if (bufsize > mss) {
2728 bufsize = roundup(bufsize, mss);
2729 if (bufsize > sb_max)
2730 bufsize = sb_max;
2731 if (bufsize > so->so_rcv.sb_hiwat)
2732 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2733 }
2734
2735 /*
2736 * Set the slow-start flight size depending on whether this
2737 * is a local network or not.
2738 */
4b52d5ee
JH
2739 if (tcp_do_rfc3390)
2740 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2741 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2742 (!isipv6 && in_localaddr(inp->inp_faddr)))
984263bc 2743 tp->snd_cwnd = mss * ss_fltsz_local;
4b52d5ee 2744 else
984263bc
MD
2745 tp->snd_cwnd = mss * ss_fltsz;
2746
2747 if (rt->rt_rmx.rmx_ssthresh) {
2748 /*
2749 * There's some sort of gateway or interface
2750 * buffer limit on the path. Use this to set
2751 * the slow start threshhold, but set the
2752 * threshold to no less than 2*mss.
2753 */
2754 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2755 tcpstat.tcps_usedssthresh++;
2756 }
2757}
2758
2759/*
2760 * Determine the MSS option to send on an outgoing SYN.
2761 */
2762int
2763tcp_mssopt(tp)
2764 struct tcpcb *tp;
2765{
2766 struct rtentry *rt;
2767#ifdef INET6
2768 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2769 int min_protoh = isipv6 ?
2770 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2771 sizeof(struct tcpiphdr);
2772#else
2773 const int isipv6 = 0;
2774 const size_t min_protoh = sizeof(struct tcpiphdr);
2775#endif
2776
2777 if (isipv6)
2778 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2779 else
2780 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2781 if (rt == NULL)
2782 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2783
2784 return (rt->rt_ifp->if_mtu - min_protoh);
2785}
2786
2787
2788/*
2789 * When a partial ack arrives, force the retransmission of the
2790 * next unacknowledged segment. Do not clear tp->t_dupacks.
2791 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2792 * be started again.
2793 */
2794static void
2795tcp_newreno_partial_ack(tp, th)
2796 struct tcpcb *tp;
2797 struct tcphdr *th;
2798{
2799 tcp_seq onxt = tp->snd_nxt;
2800 u_long ocwnd = tp->snd_cwnd;
2801
2802 callout_stop(tp->tt_rexmt);
2803 tp->t_rtttime = 0;
2804 tp->snd_nxt = th->th_ack;
2805 /*
2806 * Set snd_cwnd to one segment beyond acknowledged offset
2807 * (tp->snd_una has not yet been updated when this function is called.)
2808 */
2809 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2810 tp->t_flags |= TF_ACKNOW;
2811 (void) tcp_output(tp);
2812 tp->snd_cwnd = ocwnd;
2813 if (SEQ_GT(onxt, tp->snd_nxt))
2814 tp->snd_nxt = onxt;
2815 /*
2816 * Partial window deflation. Relies on fact that tp->snd_una
2817 * not updated yet.
2818 */
2819 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
2820}