2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)tcp_input.c 8.12 (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 $
35 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.4 2003/07/24 01:31:07 dillon Exp $
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"
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>
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>
56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
59 #include <net/route.h>
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>
82 #include <netinet/tcp_debug.h>
84 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
85 struct tcphdr tcp_savetcp;
89 #include <netipsec/ipsec.h>
90 #include <netipsec/ipsec6.h>
94 #include <netinet6/ipsec.h>
95 #include <netinet6/ipsec6.h>
96 #include <netkey/key.h>
99 #include <machine/in_cksum.h>
101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
103 static const int tcprexmtthresh = 3;
106 struct tcpstat tcpstat;
107 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
110 static int log_in_vain = 0;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
112 &log_in_vain, 0, "Log all incoming TCP connections");
114 static int blackhole = 0;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
116 &blackhole, 0, "Do not send RST when dropping refused connections");
118 int tcp_delack_enabled = 1;
119 SYSCTL_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");
123 #ifdef TCP_DROP_SYNFIN
124 static int drop_synfin = 0;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
129 static int tcp_do_limitedtransmit = 1;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
131 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
133 struct inpcbhead tcb;
134 #define tcb6 tcb /* for KAME src sync over BSD*'s */
135 struct inpcbinfo tcbinfo;
137 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
138 static void tcp_pulloutofband(struct socket *,
139 struct tcphdr *, struct mbuf *, int);
140 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
142 static void tcp_xmit_timer(struct tcpcb *, int);
143 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
145 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
147 #define ND6_HINT(tp) \
149 if ((tp) && (tp)->t_inpcb && \
150 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
151 (tp)->t_inpcb->in6p_route.ro_rt) \
152 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
159 * Indicate whether this ack should be delayed. We can delay the ack if
160 * - delayed acks are enabled and
161 * - there is no delayed ack timer in progress and
162 * - our last ack wasn't a 0-sized window. We never want to delay
163 * the ack that opens up a 0-sized window.
165 #define DELAY_ACK(tp) \
166 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
167 (tp->t_flags & TF_RXWIN0SENT) == 0)
170 tcp_reass(tp, th, tlenp, m)
171 register struct tcpcb *tp;
172 register struct tcphdr *th;
177 struct tseg_qent *p = NULL;
178 struct tseg_qent *nq;
179 struct tseg_qent *te;
180 struct socket *so = tp->t_inpcb->inp_socket;
184 * Call with th==0 after become established to
185 * force pre-ESTABLISHED data up to user socket.
190 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
191 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
194 tcpstat.tcps_rcvmemdrop++;
200 * Find a segment which begins after this one does.
202 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
203 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
209 * If there is a preceding segment, it may provide some of
210 * our data already. If so, drop the data from the incoming
211 * segment. If it provides all of our data, drop us.
215 /* conversion to int (in i) handles seq wraparound */
216 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
219 tcpstat.tcps_rcvduppack++;
220 tcpstat.tcps_rcvdupbyte += *tlenp;
224 * Try to present any queued data
225 * at the left window edge to the user.
226 * This is needed after the 3-WHS
229 goto present; /* ??? */
236 tcpstat.tcps_rcvoopack++;
237 tcpstat.tcps_rcvoobyte += *tlenp;
240 * While we overlap succeeding segments trim them or,
241 * if they are completely covered, dequeue them.
244 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
247 if (i < q->tqe_len) {
248 q->tqe_th->th_seq += i;
254 nq = LIST_NEXT(q, tqe_q);
255 LIST_REMOVE(q, tqe_q);
261 /* Insert the new segment queue entry into place. */
264 te->tqe_len = *tlenp;
267 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
269 LIST_INSERT_AFTER(p, te, tqe_q);
274 * Present data to user, advancing rcv_nxt through
275 * completed sequence space.
277 if (!TCPS_HAVEESTABLISHED(tp->t_state))
279 q = LIST_FIRST(&tp->t_segq);
280 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
283 tp->rcv_nxt += q->tqe_len;
284 flags = q->tqe_th->th_flags & TH_FIN;
285 nq = LIST_NEXT(q, tqe_q);
286 LIST_REMOVE(q, tqe_q);
287 if (so->so_state & SS_CANTRCVMORE)
290 sbappend(&so->so_rcv, q->tqe_m);
293 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
300 * TCP input routine, follows pages 65-76 of the
301 * protocol specification dated September, 1981 very closely.
305 tcp6_input(mp, offp, proto)
309 register struct mbuf *m = *mp;
310 struct in6_ifaddr *ia6;
312 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
315 * draft-itojun-ipv6-tcp-to-anycast
316 * better place to put this in?
318 ia6 = ip6_getdstifaddr(m);
319 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
322 ip6 = mtod(m, struct ip6_hdr *);
323 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
324 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
328 tcp_input(m, *offp, proto);
334 tcp_input(m, off0, proto)
335 register struct mbuf *m;
338 register struct tcphdr *th;
339 register struct ip *ip = NULL;
340 register struct ipovly *ipov;
341 register struct inpcb *inp = NULL;
346 register struct tcpcb *tp = NULL;
347 register int thflags;
348 struct socket *so = 0;
349 int todrop, acked, ourfinisacked, needoutput = 0;
351 struct tcpopt to; /* options in this segment */
352 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
353 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
354 struct sockaddr_in *next_hop = NULL;
355 int rstreason; /* For badport_bandlim accounting purposes */
356 struct ip6_hdr *ip6 = NULL;
360 const int isipv6 = 0;
366 /* Grab info from MT_TAG mbufs prepended to the chain. */
367 for (;m && m->m_type == MT_TAG; m = m->m_next) {
368 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
369 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
372 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
374 bzero((char *)&to, sizeof(to));
376 tcpstat.tcps_rcvtotal++;
379 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
380 ip6 = mtod(m, struct ip6_hdr *);
381 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
382 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
383 tcpstat.tcps_rcvbadsum++;
386 th = (struct tcphdr *)((caddr_t)ip6 + off0);
389 * Be proactive about unspecified IPv6 address in source.
390 * As we use all-zero to indicate unbounded/unconnected pcb,
391 * unspecified IPv6 address can be used to confuse us.
393 * Note that packets with unspecified IPv6 destination is
394 * already dropped in ip6_input.
396 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
402 * Get IP and TCP header together in first mbuf.
403 * Note: IP leaves IP header in first mbuf.
405 if (off0 > sizeof(struct ip)) {
406 ip_stripoptions(m, (struct mbuf *)0);
407 off0 = sizeof(struct ip);
409 if (m->m_len < sizeof(struct tcpiphdr)) {
410 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
411 tcpstat.tcps_rcvshort++;
415 ip = mtod(m, struct ip *);
416 ipov = (struct ipovly *)ip;
417 th = (struct tcphdr *)((caddr_t)ip + off0);
420 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
421 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
422 th->th_sum = m->m_pkthdr.csum_data;
424 th->th_sum = in_pseudo(ip->ip_src.s_addr,
426 htonl(m->m_pkthdr.csum_data +
429 th->th_sum ^= 0xffff;
432 * Checksum extended TCP header and data.
434 len = sizeof(struct ip) + tlen;
435 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
436 ipov->ih_len = (u_short)tlen;
437 ipov->ih_len = htons(ipov->ih_len);
438 th->th_sum = in_cksum(m, len);
441 tcpstat.tcps_rcvbadsum++;
445 /* Re-initialization for later version check */
446 ip->ip_v = IPVERSION;
451 * Check that TCP offset makes sense,
452 * pull out TCP options and adjust length. XXX
454 off = th->th_off << 2;
455 if (off < sizeof(struct tcphdr) || off > tlen) {
456 tcpstat.tcps_rcvbadoff++;
459 tlen -= off; /* tlen is used instead of ti->ti_len */
460 if (off > sizeof(struct tcphdr)) {
462 IP6_EXTHDR_CHECK(m, off0, off, );
463 ip6 = mtod(m, struct ip6_hdr *);
464 th = (struct tcphdr *)((caddr_t)ip6 + off0);
466 if (m->m_len < sizeof(struct ip) + off) {
467 if ((m = m_pullup(m, sizeof(struct ip) + off))
469 tcpstat.tcps_rcvshort++;
472 ip = mtod(m, struct ip *);
473 ipov = (struct ipovly *)ip;
474 th = (struct tcphdr *)((caddr_t)ip + off0);
477 optlen = off - sizeof(struct tcphdr);
478 optp = (u_char *)(th + 1);
480 thflags = th->th_flags;
482 #ifdef TCP_DROP_SYNFIN
484 * If the drop_synfin option is enabled, drop all packets with
485 * both the SYN and FIN bits set. This prevents e.g. nmap from
486 * identifying the TCP/IP stack.
488 * This is a violation of the TCP specification.
490 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
495 * Convert TCP protocol specific fields to host format.
497 th->th_seq = ntohl(th->th_seq);
498 th->th_ack = ntohl(th->th_ack);
499 th->th_win = ntohs(th->th_win);
500 th->th_urp = ntohs(th->th_urp);
503 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
504 * until after ip6_savecontrol() is called and before other functions
505 * which don't want those proto headers.
506 * Because ip6_savecontrol() is going to parse the mbuf to
507 * search for data to be passed up to user-land, it wants mbuf
508 * parameters to be unchanged.
509 * XXX: the call of ip6_savecontrol() has been obsoleted based on
510 * latest version of the advanced API (20020110).
512 drop_hdrlen = off0 + off;
515 * Locate pcb for segment.
518 /* IPFIREWALL_FORWARD section */
519 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
521 * Transparently forwarded. Pretend to be the destination.
522 * already got one like this?
524 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
525 ip->ip_dst, th->th_dport,
526 0, m->m_pkthdr.rcvif);
528 /* It's new. Try find the ambushing socket. */
529 inp = in_pcblookup_hash(&tcbinfo,
530 ip->ip_src, th->th_sport,
533 ntohs(next_hop->sin_port) :
535 1, m->m_pkthdr.rcvif);
539 inp = in6_pcblookup_hash(&tcbinfo,
540 &ip6->ip6_src, th->th_sport,
541 &ip6->ip6_dst, th->th_dport,
542 1, m->m_pkthdr.rcvif);
544 inp = in_pcblookup_hash(&tcbinfo,
545 ip->ip_src, th->th_sport,
546 ip->ip_dst, th->th_dport,
547 1, m->m_pkthdr.rcvif);
552 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
553 ipsec6stat.in_polvio++;
557 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
558 ipsecstat.in_polvio++;
565 if (inp != NULL && ipsec6_in_reject(m, inp)) {
569 if (inp != NULL && ipsec4_in_reject(m, inp)) {
576 * If the state is CLOSED (i.e., TCB does not exist) then
577 * all data in the incoming segment is discarded.
578 * If the TCB exists but is in CLOSED state, it is embryonic,
579 * but should either do a listen or a connect soon.
584 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
586 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
591 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
592 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
596 strcpy(dbuf, inet_ntoa(ip->ip_dst));
597 strcpy(sbuf, inet_ntoa(ip->ip_src));
599 switch (log_in_vain) {
601 if ((thflags & TH_SYN) == 0)
605 "Connection attempt to TCP %s:%d "
606 "from %s:%d flags:0x%02x\n",
607 dbuf, ntohs(th->th_dport), sbuf,
608 ntohs(th->th_sport), thflags);
617 if (thflags & TH_SYN)
626 rstreason = BANDLIM_RST_CLOSEDPORT;
631 rstreason = BANDLIM_RST_CLOSEDPORT;
634 if (tp->t_state == TCPS_CLOSED)
637 /* Unscale the window into a 32-bit value. */
638 if ((thflags & TH_SYN) == 0)
639 tiwin = th->th_win << tp->snd_scale;
643 so = inp->inp_socket;
644 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
645 struct in_conninfo inc;
647 if (so->so_options & SO_DEBUG) {
648 ostate = tp->t_state;
650 bcopy((char *)ip6, (char *)tcp_saveipgen,
653 bcopy((char *)ip, (char *)tcp_saveipgen,
658 /* skip if this isn't a listen socket */
659 if ((so->so_options & SO_ACCEPTCONN) == 0)
662 inc.inc_isipv6 = isipv6;
665 inc.inc6_faddr = ip6->ip6_src;
666 inc.inc6_laddr = ip6->ip6_dst;
667 inc.inc6_route.ro_rt = NULL; /* XXX */
669 inc.inc_faddr = ip->ip_src;
670 inc.inc_laddr = ip->ip_dst;
671 inc.inc_route.ro_rt = NULL; /* XXX */
673 inc.inc_fport = th->th_sport;
674 inc.inc_lport = th->th_dport;
677 * If the state is LISTEN then ignore segment if it contains
678 * a RST. If the segment contains an ACK then it is bad and
679 * send a RST. If it does not contain a SYN then it is not
680 * interesting; drop it.
682 * If the state is SYN_RECEIVED (syncache) and seg contains
683 * an ACK, but not for our SYN/ACK, send a RST. If the seg
684 * contains a RST, check the sequence number to see if it
685 * is a valid reset segment.
687 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
688 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
689 if (!syncache_expand(&inc, th, &so, m)) {
691 * No syncache entry, or ACK was not
692 * for our SYN/ACK. Send a RST.
694 tcpstat.tcps_badsyn++;
695 rstreason = BANDLIM_RST_OPENPORT;
700 * Could not complete 3-way handshake,
701 * connection is being closed down, and
702 * syncache will free mbuf.
706 * Socket is created in state SYN_RECEIVED.
707 * Continue processing segment.
712 * This is what would have happened in
713 * tcp_output() when the SYN,ACK was sent.
715 tp->snd_up = tp->snd_una;
716 tp->snd_max = tp->snd_nxt = tp->iss + 1;
717 tp->last_ack_sent = tp->rcv_nxt;
719 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
720 * until the _second_ ACK is received:
721 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
722 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
723 * move to ESTAB, set snd_wnd to tiwin.
725 tp->snd_wnd = tiwin; /* unscaled */
728 if (thflags & TH_RST) {
729 syncache_chkrst(&inc, th);
732 if (thflags & TH_ACK) {
733 syncache_badack(&inc);
734 tcpstat.tcps_badsyn++;
735 rstreason = BANDLIM_RST_OPENPORT;
742 * Segment's flags are (SYN) or (SYN|FIN).
746 * If deprecated address is forbidden,
747 * we do not accept SYN to deprecated interface
748 * address to prevent any new inbound connection from
749 * getting established.
750 * When we do not accept SYN, we send a TCP RST,
751 * with deprecated source address (instead of dropping
752 * it). We compromise it as it is much better for peer
753 * to send a RST, and RST will be the final packet
756 * If we do not forbid deprecated addresses, we accept
757 * the SYN packet. RFC2462 does not suggest dropping
759 * If we decipher RFC2462 5.5.4, it says like this:
760 * 1. use of deprecated addr with existing
761 * communication is okay - "SHOULD continue to be
763 * 2. use of it with new communication:
764 * (2a) "SHOULD NOT be used if alternate address
765 * with sufficient scope is available"
766 * (2b) nothing mentioned otherwise.
767 * Here we fall into (2b) case as we have no choice in
768 * our source address selection - we must obey the peer.
770 * The wording in RFC2462 is confusing, and there are
771 * multiple description text for deprecated address
772 * handling - worse, they are not exactly the same.
773 * I believe 5.5.4 is the best one, so we follow 5.5.4.
775 if (isipv6 && !ip6_use_deprecated) {
776 struct in6_ifaddr *ia6;
778 if ((ia6 = ip6_getdstifaddr(m)) &&
779 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
781 rstreason = BANDLIM_RST_OPENPORT;
787 * If it is from this socket, drop it, it must be forged.
788 * Don't bother responding if the destination was a broadcast.
790 if (th->th_dport == th->th_sport) {
792 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
796 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
801 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
803 * Note that it is quite possible to receive unicast
804 * link-layer packets with a broadcast IP address. Use
805 * in_broadcast() to find them.
807 if (m->m_flags & (M_BCAST|M_MCAST))
810 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
811 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
814 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
815 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
816 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
817 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
821 * SYN appears to be valid; create compressed TCP state
822 * for syncache, or perform t/tcp connection.
824 if (so->so_qlen <= so->so_qlimit) {
825 tcp_dooptions(&to, optp, optlen, 1);
826 if (!syncache_add(&inc, &to, th, &so, m))
830 * Entry added to syncache, mbuf used to
831 * send SYN,ACK packet.
835 * Segment passed TAO tests.
840 tp->t_starttime = ticks;
841 tp->t_state = TCPS_ESTABLISHED;
844 * If there is a FIN, or if there is data and the
845 * connection is local, then delay SYN,ACK(SYN) in
846 * the hope of piggy-backing it on a response
847 * segment. Otherwise must send ACK now in case
848 * the other side is slow starting.
851 ((thflags & TH_FIN) ||
853 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
854 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
855 callout_reset(tp->tt_delack, tcp_delacktime,
856 tcp_timer_delack, tp);
857 tp->t_flags |= TF_NEEDSYN;
859 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
861 tcpstat.tcps_connects++;
869 /* XXX temp debugging */
870 /* should not happen - syncache should pick up these connections */
871 if (tp->t_state == TCPS_LISTEN)
872 panic("tcp_input: TCPS_LISTEN");
875 * Segment received on connection.
876 * Reset idle time and keep-alive timer.
878 tp->t_rcvtime = ticks;
879 if (TCPS_HAVEESTABLISHED(tp->t_state))
880 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
884 * XXX this is tradtitional behavior, may need to be cleaned up.
886 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
887 if (thflags & TH_SYN) {
888 if (to.to_flags & TOF_SCALE) {
889 tp->t_flags |= TF_RCVD_SCALE;
890 tp->requested_s_scale = to.to_requested_s_scale;
892 if (to.to_flags & TOF_TS) {
893 tp->t_flags |= TF_RCVD_TSTMP;
894 tp->ts_recent = to.to_tsval;
895 tp->ts_recent_age = ticks;
897 if (to.to_flags & (TOF_CC|TOF_CCNEW))
898 tp->t_flags |= TF_RCVD_CC;
899 if (to.to_flags & TOF_MSS)
900 tcp_mss(tp, to.to_mss);
904 * Header prediction: check for the two common cases
905 * of a uni-directional data xfer. If the packet has
906 * no control flags, is in-sequence, the window didn't
907 * change and we're not retransmitting, it's a
908 * candidate. If the length is zero and the ack moved
909 * forward, we're the sender side of the xfer. Just
910 * free the data acked & wake any higher level process
911 * that was blocked waiting for space. If the length
912 * is non-zero and the ack didn't move, we're the
913 * receiver side. If we're getting packets in-order
914 * (the reassembly queue is empty), add the data to
915 * the socket buffer and note that we need a delayed ack.
916 * Make sure that the hidden state-flags are also off.
917 * Since we check for TCPS_ESTABLISHED above, it can only
920 if (tp->t_state == TCPS_ESTABLISHED &&
921 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
922 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
923 ((to.to_flags & TOF_TS) == 0 ||
924 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
926 * Using the CC option is compulsory if once started:
927 * the segment is OK if no T/TCP was negotiated or
928 * if the segment has a CC option equal to CCrecv
930 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
931 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
932 th->th_seq == tp->rcv_nxt &&
933 tiwin && tiwin == tp->snd_wnd &&
934 tp->snd_nxt == tp->snd_max) {
937 * If last ACK falls within this segment's sequence numbers,
938 * record the timestamp.
939 * NOTE that the test is modified according to the latest
940 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
942 if ((to.to_flags & TOF_TS) != 0 &&
943 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
944 tp->ts_recent_age = ticks;
945 tp->ts_recent = to.to_tsval;
949 if (SEQ_GT(th->th_ack, tp->snd_una) &&
950 SEQ_LEQ(th->th_ack, tp->snd_max) &&
951 tp->snd_cwnd >= tp->snd_wnd &&
953 tp->t_dupacks < tcprexmtthresh) ||
954 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
956 * this is a pure ack for outstanding data.
958 ++tcpstat.tcps_predack;
960 * "bad retransmit" recovery
962 if (tp->t_rxtshift == 1 &&
963 ticks < tp->t_badrxtwin) {
964 tp->snd_cwnd = tp->snd_cwnd_prev;
966 tp->snd_ssthresh_prev;
967 tp->snd_recover = tp->snd_recover_prev;
968 if (tp->t_flags & TF_WASFRECOVERY)
969 ENTER_FASTRECOVERY(tp);
970 tp->snd_nxt = tp->snd_max;
974 * Recalculate the retransmit timer / rtt.
976 * Some machines (certain windows boxes)
977 * send broken timestamp replies during the
978 * SYN+ACK phase, ignore timestamps of 0.
980 if ((to.to_flags & TOF_TS) != 0 &&
983 ticks - to.to_tsecr + 1);
984 } else if (tp->t_rtttime &&
985 SEQ_GT(th->th_ack, tp->t_rtseq)) {
987 ticks - tp->t_rtttime);
989 tcp_xmit_bandwidth_limit(tp, th->th_ack);
990 acked = th->th_ack - tp->snd_una;
991 tcpstat.tcps_rcvackpack++;
992 tcpstat.tcps_rcvackbyte += acked;
993 sbdrop(&so->so_snd, acked);
994 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
995 SEQ_LEQ(th->th_ack, tp->snd_recover))
996 tp->snd_recover = th->th_ack - 1;
997 tp->snd_una = th->th_ack;
1000 ND6_HINT(tp); /* some progress has been done */
1003 * If all outstanding data are acked, stop
1004 * retransmit timer, otherwise restart timer
1005 * using current (possibly backed-off) value.
1006 * If process is waiting for space,
1007 * wakeup/selwakeup/signal. If data
1008 * are ready to send, let tcp_output
1009 * decide between more output or persist.
1011 if (tp->snd_una == tp->snd_max)
1012 callout_stop(tp->tt_rexmt);
1013 else if (!callout_active(tp->tt_persist))
1014 callout_reset(tp->tt_rexmt,
1016 tcp_timer_rexmt, tp);
1019 if (so->so_snd.sb_cc)
1020 (void) tcp_output(tp);
1023 } else if (th->th_ack == tp->snd_una &&
1024 LIST_EMPTY(&tp->t_segq) &&
1025 tlen <= sbspace(&so->so_rcv)) {
1027 * this is a pure, in-sequence data packet
1028 * with nothing on the reassembly queue and
1029 * we have enough buffer space to take it.
1031 ++tcpstat.tcps_preddat;
1032 tp->rcv_nxt += tlen;
1033 tcpstat.tcps_rcvpack++;
1034 tcpstat.tcps_rcvbyte += tlen;
1035 ND6_HINT(tp); /* some progress has been done */
1037 * Add data to socket buffer.
1039 if (so->so_state & SS_CANTRCVMORE) {
1042 m_adj(m, drop_hdrlen); /* delayed header drop */
1043 sbappend(&so->so_rcv, m);
1046 if (DELAY_ACK(tp)) {
1047 callout_reset(tp->tt_delack, tcp_delacktime,
1048 tcp_timer_delack, tp);
1050 tp->t_flags |= TF_ACKNOW;
1058 * Calculate amount of space in receive window,
1059 * and then do TCP input processing.
1060 * Receive window is amount of space in rcv queue,
1061 * but not less than advertised window.
1065 win = sbspace(&so->so_rcv);
1068 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1071 switch (tp->t_state) {
1074 * If the state is SYN_RECEIVED:
1075 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1077 case TCPS_SYN_RECEIVED:
1078 if ((thflags & TH_ACK) &&
1079 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1080 SEQ_GT(th->th_ack, tp->snd_max))) {
1081 rstreason = BANDLIM_RST_OPENPORT;
1087 * If the state is SYN_SENT:
1088 * if seg contains an ACK, but not for our SYN, drop the input.
1089 * if seg contains a RST, then drop the connection.
1090 * if seg does not contain SYN, then drop it.
1091 * Otherwise this is an acceptable SYN segment
1092 * initialize tp->rcv_nxt and tp->irs
1093 * if seg contains ack then advance tp->snd_una
1094 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1095 * arrange for segment to be acked (eventually)
1096 * continue processing rest of data/controls, beginning with URG
1099 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1100 taop = &tao_noncached;
1101 bzero(taop, sizeof(*taop));
1104 if ((thflags & TH_ACK) &&
1105 (SEQ_LEQ(th->th_ack, tp->iss) ||
1106 SEQ_GT(th->th_ack, tp->snd_max))) {
1108 * If we have a cached CCsent for the remote host,
1109 * hence we haven't just crashed and restarted,
1110 * do not send a RST. This may be a retransmission
1111 * from the other side after our earlier ACK was lost.
1112 * Our new SYN, when it arrives, will serve as the
1115 if (taop->tao_ccsent != 0)
1118 rstreason = BANDLIM_UNLIMITED;
1122 if (thflags & TH_RST) {
1123 if (thflags & TH_ACK)
1124 tp = tcp_drop(tp, ECONNREFUSED);
1127 if ((thflags & TH_SYN) == 0)
1129 tp->snd_wnd = th->th_win; /* initial send window */
1130 tp->cc_recv = to.to_cc; /* foreign CC */
1132 tp->irs = th->th_seq;
1134 if (thflags & TH_ACK) {
1136 * Our SYN was acked. If segment contains CC.ECHO
1137 * option, check it to make sure this segment really
1138 * matches our SYN. If not, just drop it as old
1139 * duplicate, but send an RST if we're still playing
1140 * by the old rules. If no CC.ECHO option, make sure
1141 * we don't get fooled into using T/TCP.
1143 if (to.to_flags & TOF_CCECHO) {
1144 if (tp->cc_send != to.to_ccecho) {
1145 if (taop->tao_ccsent != 0)
1148 rstreason = BANDLIM_UNLIMITED;
1153 tp->t_flags &= ~TF_RCVD_CC;
1154 tcpstat.tcps_connects++;
1156 /* Do window scaling on this connection? */
1157 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1158 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1159 tp->snd_scale = tp->requested_s_scale;
1160 tp->rcv_scale = tp->request_r_scale;
1162 /* Segment is acceptable, update cache if undefined. */
1163 if (taop->tao_ccsent == 0)
1164 taop->tao_ccsent = to.to_ccecho;
1166 tp->rcv_adv += tp->rcv_wnd;
1167 tp->snd_una++; /* SYN is acked */
1169 * If there's data, delay ACK; if there's also a FIN
1170 * ACKNOW will be turned on later.
1172 if (DELAY_ACK(tp) && tlen != 0)
1173 callout_reset(tp->tt_delack, tcp_delacktime,
1174 tcp_timer_delack, tp);
1176 tp->t_flags |= TF_ACKNOW;
1178 * Received <SYN,ACK> in SYN_SENT[*] state.
1180 * SYN_SENT --> ESTABLISHED
1181 * SYN_SENT* --> FIN_WAIT_1
1183 tp->t_starttime = ticks;
1184 if (tp->t_flags & TF_NEEDFIN) {
1185 tp->t_state = TCPS_FIN_WAIT_1;
1186 tp->t_flags &= ~TF_NEEDFIN;
1189 tp->t_state = TCPS_ESTABLISHED;
1190 callout_reset(tp->tt_keep, tcp_keepidle,
1191 tcp_timer_keep, tp);
1195 * Received initial SYN in SYN-SENT[*] state =>
1196 * simultaneous open. If segment contains CC option
1197 * and there is a cached CC, apply TAO test.
1198 * If it succeeds, connection is * half-synchronized.
1199 * Otherwise, do 3-way handshake:
1200 * SYN-SENT -> SYN-RECEIVED
1201 * SYN-SENT* -> SYN-RECEIVED*
1202 * If there was no CC option, clear cached CC value.
1204 tp->t_flags |= TF_ACKNOW;
1205 callout_stop(tp->tt_rexmt);
1206 if (to.to_flags & TOF_CC) {
1207 if (taop->tao_cc != 0 &&
1208 CC_GT(to.to_cc, taop->tao_cc)) {
1210 * update cache and make transition:
1211 * SYN-SENT -> ESTABLISHED*
1212 * SYN-SENT* -> FIN-WAIT-1*
1214 taop->tao_cc = to.to_cc;
1215 tp->t_starttime = ticks;
1216 if (tp->t_flags & TF_NEEDFIN) {
1217 tp->t_state = TCPS_FIN_WAIT_1;
1218 tp->t_flags &= ~TF_NEEDFIN;
1220 tp->t_state = TCPS_ESTABLISHED;
1221 callout_reset(tp->tt_keep,
1226 tp->t_flags |= TF_NEEDSYN;
1228 tp->t_state = TCPS_SYN_RECEIVED;
1230 /* CC.NEW or no option => invalidate cache */
1232 tp->t_state = TCPS_SYN_RECEIVED;
1238 * Advance th->th_seq to correspond to first data byte.
1239 * If data, trim to stay within window,
1240 * dropping FIN if necessary.
1243 if (tlen > tp->rcv_wnd) {
1244 todrop = tlen - tp->rcv_wnd;
1248 tcpstat.tcps_rcvpackafterwin++;
1249 tcpstat.tcps_rcvbyteafterwin += todrop;
1251 tp->snd_wl1 = th->th_seq - 1;
1252 tp->rcv_up = th->th_seq;
1254 * Client side of transaction: already sent SYN and data.
1255 * If the remote host used T/TCP to validate the SYN,
1256 * our data will be ACK'd; if so, enter normal data segment
1257 * processing in the middle of step 5, ack processing.
1258 * Otherwise, goto step 6.
1260 if (thflags & TH_ACK)
1266 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1267 * if segment contains a SYN and CC [not CC.NEW] option:
1268 * if state == TIME_WAIT and connection duration > MSL,
1269 * drop packet and send RST;
1271 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1272 * ack the FIN (and data) in retransmission queue.
1273 * Complete close and delete TCPCB. Then reprocess
1274 * segment, hoping to find new TCPCB in LISTEN state;
1276 * else must be old SYN; drop it.
1277 * else do normal processing.
1281 case TCPS_TIME_WAIT:
1282 if ((thflags & TH_SYN) &&
1283 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1284 if (tp->t_state == TCPS_TIME_WAIT &&
1285 (ticks - tp->t_starttime) > tcp_msl) {
1286 rstreason = BANDLIM_UNLIMITED;
1289 if (CC_GT(to.to_cc, tp->cc_recv)) {
1296 break; /* continue normal processing */
1300 * States other than LISTEN or SYN_SENT.
1301 * First check the RST flag and sequence number since reset segments
1302 * are exempt from the timestamp and connection count tests. This
1303 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1304 * below which allowed reset segments in half the sequence space
1305 * to fall though and be processed (which gives forged reset
1306 * segments with a random sequence number a 50 percent chance of
1307 * killing a connection).
1308 * Then check timestamp, if present.
1309 * Then check the connection count, if present.
1310 * Then check that at least some bytes of segment are within
1311 * receive window. If segment begins before rcv_nxt,
1312 * drop leading data (and SYN); if nothing left, just ack.
1315 * If the RST bit is set, check the sequence number to see
1316 * if this is a valid reset segment.
1318 * In all states except SYN-SENT, all reset (RST) segments
1319 * are validated by checking their SEQ-fields. A reset is
1320 * valid if its sequence number is in the window.
1321 * Note: this does not take into account delayed ACKs, so
1322 * we should test against last_ack_sent instead of rcv_nxt.
1323 * The sequence number in the reset segment is normally an
1324 * echo of our outgoing acknowlegement numbers, but some hosts
1325 * send a reset with the sequence number at the rightmost edge
1326 * of our receive window, and we have to handle this case.
1327 * If we have multiple segments in flight, the intial reset
1328 * segment sequence numbers will be to the left of last_ack_sent,
1329 * but they will eventually catch up.
1330 * In any case, it never made sense to trim reset segments to
1331 * fit the receive window since RFC 1122 says:
1332 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1334 * A TCP SHOULD allow a received RST segment to include data.
1337 * It has been suggested that a RST segment could contain
1338 * ASCII text that encoded and explained the cause of the
1339 * RST. No standard has yet been established for such
1342 * If the reset segment passes the sequence number test examine
1344 * SYN_RECEIVED STATE:
1345 * If passive open, return to LISTEN state.
1346 * If active open, inform user that connection was refused.
1347 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1348 * Inform user that connection was reset, and close tcb.
1349 * CLOSING, LAST_ACK STATES:
1352 * Drop the segment - see Stevens, vol. 2, p. 964 and
1355 if (thflags & TH_RST) {
1356 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1357 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1358 switch (tp->t_state) {
1360 case TCPS_SYN_RECEIVED:
1361 so->so_error = ECONNREFUSED;
1364 case TCPS_ESTABLISHED:
1365 case TCPS_FIN_WAIT_1:
1366 case TCPS_FIN_WAIT_2:
1367 case TCPS_CLOSE_WAIT:
1368 so->so_error = ECONNRESET;
1370 tp->t_state = TCPS_CLOSED;
1371 tcpstat.tcps_drops++;
1380 case TCPS_TIME_WAIT:
1388 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1389 * and it's less than ts_recent, drop it.
1391 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1392 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1394 /* Check to see if ts_recent is over 24 days old. */
1395 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1397 * Invalidate ts_recent. If this segment updates
1398 * ts_recent, the age will be reset later and ts_recent
1399 * will get a valid value. If it does not, setting
1400 * ts_recent to zero will at least satisfy the
1401 * requirement that zero be placed in the timestamp
1402 * echo reply when ts_recent isn't valid. The
1403 * age isn't reset until we get a valid ts_recent
1404 * because we don't want out-of-order segments to be
1405 * dropped when ts_recent is old.
1409 tcpstat.tcps_rcvduppack++;
1410 tcpstat.tcps_rcvdupbyte += tlen;
1411 tcpstat.tcps_pawsdrop++;
1420 * If T/TCP was negotiated and the segment doesn't have CC,
1421 * or if its CC is wrong then drop the segment.
1422 * RST segments do not have to comply with this.
1424 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1425 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1429 * In the SYN-RECEIVED state, validate that the packet belongs to
1430 * this connection before trimming the data to fit the receive
1431 * window. Check the sequence number versus IRS since we know
1432 * the sequence numbers haven't wrapped. This is a partial fix
1433 * for the "LAND" DoS attack.
1435 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1436 rstreason = BANDLIM_RST_OPENPORT;
1440 todrop = tp->rcv_nxt - th->th_seq;
1442 if (thflags & TH_SYN) {
1452 * Following if statement from Stevens, vol. 2, p. 960.
1455 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1457 * Any valid FIN must be to the left of the window.
1458 * At this point the FIN must be a duplicate or out
1459 * of sequence; drop it.
1464 * Send an ACK to resynchronize and drop any data.
1465 * But keep on processing for RST or ACK.
1467 tp->t_flags |= TF_ACKNOW;
1469 tcpstat.tcps_rcvduppack++;
1470 tcpstat.tcps_rcvdupbyte += todrop;
1472 tcpstat.tcps_rcvpartduppack++;
1473 tcpstat.tcps_rcvpartdupbyte += todrop;
1475 drop_hdrlen += todrop; /* drop from the top afterwards */
1476 th->th_seq += todrop;
1478 if (th->th_urp > todrop)
1479 th->th_urp -= todrop;
1487 * If new data are received on a connection after the
1488 * user processes are gone, then RST the other end.
1490 if ((so->so_state & SS_NOFDREF) &&
1491 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1493 tcpstat.tcps_rcvafterclose++;
1494 rstreason = BANDLIM_UNLIMITED;
1499 * If segment ends after window, drop trailing data
1500 * (and PUSH and FIN); if nothing left, just ACK.
1502 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1504 tcpstat.tcps_rcvpackafterwin++;
1505 if (todrop >= tlen) {
1506 tcpstat.tcps_rcvbyteafterwin += tlen;
1508 * If a new connection request is received
1509 * while in TIME_WAIT, drop the old connection
1510 * and start over if the sequence numbers
1511 * are above the previous ones.
1513 if (thflags & TH_SYN &&
1514 tp->t_state == TCPS_TIME_WAIT &&
1515 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1520 * If window is closed can only take segments at
1521 * window edge, and have to drop data and PUSH from
1522 * incoming segments. Continue processing, but
1523 * remember to ack. Otherwise, drop segment
1526 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1527 tp->t_flags |= TF_ACKNOW;
1528 tcpstat.tcps_rcvwinprobe++;
1532 tcpstat.tcps_rcvbyteafterwin += todrop;
1535 thflags &= ~(TH_PUSH|TH_FIN);
1539 * If last ACK falls within this segment's sequence numbers,
1540 * record its timestamp.
1541 * NOTE that the test is modified according to the latest
1542 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1544 if ((to.to_flags & TOF_TS) != 0 &&
1545 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1546 tp->ts_recent_age = ticks;
1547 tp->ts_recent = to.to_tsval;
1551 * If a SYN is in the window, then this is an
1552 * error and we send an RST and drop the connection.
1554 if (thflags & TH_SYN) {
1555 tp = tcp_drop(tp, ECONNRESET);
1556 rstreason = BANDLIM_UNLIMITED;
1561 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1562 * flag is on (half-synchronized state), then queue data for
1563 * later processing; else drop segment and return.
1565 if ((thflags & TH_ACK) == 0) {
1566 if (tp->t_state == TCPS_SYN_RECEIVED ||
1567 (tp->t_flags & TF_NEEDSYN))
1576 switch (tp->t_state) {
1579 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1580 * ESTABLISHED state and continue processing.
1581 * The ACK was checked above.
1583 case TCPS_SYN_RECEIVED:
1585 tcpstat.tcps_connects++;
1587 /* Do window scaling? */
1588 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1589 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1590 tp->snd_scale = tp->requested_s_scale;
1591 tp->rcv_scale = tp->request_r_scale;
1594 * Upon successful completion of 3-way handshake,
1595 * update cache.CC if it was undefined, pass any queued
1596 * data to the user, and advance state appropriately.
1598 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1600 taop->tao_cc = tp->cc_recv;
1604 * SYN-RECEIVED -> ESTABLISHED
1605 * SYN-RECEIVED* -> FIN-WAIT-1
1607 tp->t_starttime = ticks;
1608 if (tp->t_flags & TF_NEEDFIN) {
1609 tp->t_state = TCPS_FIN_WAIT_1;
1610 tp->t_flags &= ~TF_NEEDFIN;
1612 tp->t_state = TCPS_ESTABLISHED;
1613 callout_reset(tp->tt_keep, tcp_keepidle,
1614 tcp_timer_keep, tp);
1617 * If segment contains data or ACK, will call tcp_reass()
1618 * later; if not, do so now to pass queued data to user.
1620 if (tlen == 0 && (thflags & TH_FIN) == 0)
1621 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1623 tp->snd_wl1 = th->th_seq - 1;
1627 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1628 * ACKs. If the ack is in the range
1629 * tp->snd_una < th->th_ack <= tp->snd_max
1630 * then advance tp->snd_una to th->th_ack and drop
1631 * data from the retransmission queue. If this ACK reflects
1632 * more up to date window information we update our window information.
1634 case TCPS_ESTABLISHED:
1635 case TCPS_FIN_WAIT_1:
1636 case TCPS_FIN_WAIT_2:
1637 case TCPS_CLOSE_WAIT:
1640 case TCPS_TIME_WAIT:
1642 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1643 if (tlen == 0 && tiwin == tp->snd_wnd) {
1644 tcpstat.tcps_rcvdupack++;
1646 * If we have outstanding data (other than
1647 * a window probe), this is a completely
1648 * duplicate ack (ie, window info didn't
1649 * change), the ack is the biggest we've
1650 * seen and we've seen exactly our rexmt
1651 * threshhold of them, assume a packet
1652 * has been dropped and retransmit it.
1653 * Kludge snd_nxt & the congestion
1654 * window so we send only this one
1657 * We know we're losing at the current
1658 * window size so do congestion avoidance
1659 * (set ssthresh to half the current window
1660 * and pull our congestion window back to
1661 * the new ssthresh).
1663 * Dup acks mean that packets have left the
1664 * network (they're now cached at the receiver)
1665 * so bump cwnd by the amount in the receiver
1666 * to keep a constant cwnd packets in the
1669 if (!callout_active(tp->tt_rexmt) ||
1670 th->th_ack != tp->snd_una)
1672 else if (++tp->t_dupacks > tcprexmtthresh ||
1674 IN_FASTRECOVERY(tp))) {
1675 tp->snd_cwnd += tp->t_maxseg;
1676 (void) tcp_output(tp);
1678 } else if (tp->t_dupacks == tcprexmtthresh) {
1679 tcp_seq onxt = tp->snd_nxt;
1681 if (tcp_do_newreno &&
1687 win = min(tp->snd_wnd, tp->snd_cwnd) /
1691 tp->snd_ssthresh = win * tp->t_maxseg;
1692 ENTER_FASTRECOVERY(tp);
1693 tp->snd_recover = tp->snd_max;
1694 callout_stop(tp->tt_rexmt);
1696 tp->snd_nxt = th->th_ack;
1697 tp->snd_cwnd = tp->t_maxseg;
1698 (void) tcp_output(tp);
1699 KASSERT(tp->snd_limited <= 2,
1700 ("tp->snd_limited too big"));
1701 tp->snd_cwnd = tp->snd_ssthresh +
1703 (tp->t_dupacks - tp->snd_limited));
1704 if (SEQ_GT(onxt, tp->snd_nxt))
1707 } else if (tcp_do_limitedtransmit) {
1708 u_long oldcwnd = tp->snd_cwnd;
1709 tcp_seq oldsndmax = tp->snd_max;
1711 KASSERT(tp->t_dupacks == 1 ||
1713 ("dupacks not 1 or 2"));
1714 if (tp->t_dupacks == 1) {
1715 tp->snd_limited = 0;
1716 tp->snd_cwnd += tp->t_maxseg;
1721 (void) tcp_output(tp);
1722 sent = tp->snd_max - oldsndmax;
1723 if (sent > tp->t_maxseg) {
1724 KASSERT(tp->snd_limited == 0 &&
1727 tp->snd_limited = 2;
1728 } else if (sent > 0)
1730 tp->snd_cwnd = oldcwnd;
1738 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1741 * If the congestion window was inflated to account
1742 * for the other side's cached packets, retract it.
1744 if (tcp_do_newreno) {
1745 if (IN_FASTRECOVERY(tp)) {
1746 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1747 tcp_newreno_partial_ack(tp, th);
1750 * Window inflation should have left us
1751 * with approximately snd_ssthresh
1753 * But in case we would be inclined to
1754 * send a burst, better to do it via
1755 * the slow start mechanism.
1757 if (SEQ_GT(th->th_ack +
1760 tp->snd_cwnd = tp->snd_max -
1764 tp->snd_cwnd = tp->snd_ssthresh;
1768 if (tp->t_dupacks >= tcprexmtthresh &&
1769 tp->snd_cwnd > tp->snd_ssthresh)
1770 tp->snd_cwnd = tp->snd_ssthresh;
1773 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1774 tcpstat.tcps_rcvacktoomuch++;
1778 * If we reach this point, ACK is not a duplicate,
1779 * i.e., it ACKs something we sent.
1781 if (tp->t_flags & TF_NEEDSYN) {
1783 * T/TCP: Connection was half-synchronized, and our
1784 * SYN has been ACK'd (so connection is now fully
1785 * synchronized). Go to non-starred state,
1786 * increment snd_una for ACK of SYN, and check if
1787 * we can do window scaling.
1789 tp->t_flags &= ~TF_NEEDSYN;
1791 /* Do window scaling? */
1792 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1793 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1794 tp->snd_scale = tp->requested_s_scale;
1795 tp->rcv_scale = tp->request_r_scale;
1800 acked = th->th_ack - tp->snd_una;
1801 tcpstat.tcps_rcvackpack++;
1802 tcpstat.tcps_rcvackbyte += acked;
1805 * If we just performed our first retransmit, and the ACK
1806 * arrives within our recovery window, then it was a mistake
1807 * to do the retransmit in the first place. Recover our
1808 * original cwnd and ssthresh, and proceed to transmit where
1811 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1812 tp->snd_cwnd = tp->snd_cwnd_prev;
1813 tp->snd_ssthresh = tp->snd_ssthresh_prev;
1814 tp->snd_recover = tp->snd_recover_prev;
1815 if (tp->t_flags & TF_WASFRECOVERY)
1816 ENTER_FASTRECOVERY(tp);
1817 tp->snd_nxt = tp->snd_max;
1818 tp->t_badrxtwin = 0; /* XXX probably not required */
1822 * If we have a timestamp reply, update smoothed
1823 * round trip time. If no timestamp is present but
1824 * transmit timer is running and timed sequence
1825 * number was acked, update smoothed round trip time.
1826 * Since we now have an rtt measurement, cancel the
1827 * timer backoff (cf., Phil Karn's retransmit alg.).
1828 * Recompute the initial retransmit timer.
1830 * Some machines (certain windows boxes) send broken
1831 * timestamp replies during the SYN+ACK phase, ignore
1834 if ((to.to_flags & TOF_TS) != 0 &&
1836 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1837 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1838 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1840 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1843 * If all outstanding data is acked, stop retransmit
1844 * timer and remember to restart (more output or persist).
1845 * If there is more data to be acked, restart retransmit
1846 * timer, using current (possibly backed-off) value.
1848 if (th->th_ack == tp->snd_max) {
1849 callout_stop(tp->tt_rexmt);
1851 } else if (!callout_active(tp->tt_persist))
1852 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1853 tcp_timer_rexmt, tp);
1856 * If no data (only SYN) was ACK'd,
1857 * skip rest of ACK processing.
1863 * When new data is acked, open the congestion window.
1864 * If the window gives us less than ssthresh packets
1865 * in flight, open exponentially (maxseg per packet).
1866 * Otherwise open linearly: maxseg per window
1867 * (maxseg^2 / cwnd per packet).
1869 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1870 register u_int cw = tp->snd_cwnd;
1871 register u_int incr = tp->t_maxseg;
1872 if (cw > tp->snd_ssthresh)
1873 incr = incr * incr / cw;
1874 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1876 if (acked > so->so_snd.sb_cc) {
1877 tp->snd_wnd -= so->so_snd.sb_cc;
1878 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1881 sbdrop(&so->so_snd, acked);
1882 tp->snd_wnd -= acked;
1886 /* detect una wraparound */
1887 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1888 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1889 SEQ_LEQ(th->th_ack, tp->snd_recover))
1890 tp->snd_recover = th->th_ack - 1;
1891 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1892 SEQ_GEQ(th->th_ack, tp->snd_recover))
1893 EXIT_FASTRECOVERY(tp);
1894 tp->snd_una = th->th_ack;
1895 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1896 tp->snd_nxt = tp->snd_una;
1898 switch (tp->t_state) {
1901 * In FIN_WAIT_1 STATE in addition to the processing
1902 * for the ESTABLISHED state if our FIN is now acknowledged
1903 * then enter FIN_WAIT_2.
1905 case TCPS_FIN_WAIT_1:
1906 if (ourfinisacked) {
1908 * If we can't receive any more
1909 * data, then closing user can proceed.
1910 * Starting the timer is contrary to the
1911 * specification, but if we don't get a FIN
1912 * we'll hang forever.
1914 if (so->so_state & SS_CANTRCVMORE) {
1915 soisdisconnected(so);
1916 callout_reset(tp->tt_2msl, tcp_maxidle,
1917 tcp_timer_2msl, tp);
1919 tp->t_state = TCPS_FIN_WAIT_2;
1924 * In CLOSING STATE in addition to the processing for
1925 * the ESTABLISHED state if the ACK acknowledges our FIN
1926 * then enter the TIME-WAIT state, otherwise ignore
1930 if (ourfinisacked) {
1931 tp->t_state = TCPS_TIME_WAIT;
1932 tcp_canceltimers(tp);
1933 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1934 if (tp->cc_recv != 0 &&
1935 (ticks - tp->t_starttime) < tcp_msl)
1936 callout_reset(tp->tt_2msl,
1939 tcp_timer_2msl, tp);
1941 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1942 tcp_timer_2msl, tp);
1943 soisdisconnected(so);
1948 * In LAST_ACK, we may still be waiting for data to drain
1949 * and/or to be acked, as well as for the ack of our FIN.
1950 * If our FIN is now acknowledged, delete the TCB,
1951 * enter the closed state and return.
1954 if (ourfinisacked) {
1961 * In TIME_WAIT state the only thing that should arrive
1962 * is a retransmission of the remote FIN. Acknowledge
1963 * it and restart the finack timer.
1965 case TCPS_TIME_WAIT:
1966 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1967 tcp_timer_2msl, tp);
1974 * Update window information.
1975 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1977 if ((thflags & TH_ACK) &&
1978 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
1979 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
1980 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
1981 /* keep track of pure window updates */
1983 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
1984 tcpstat.tcps_rcvwinupd++;
1985 tp->snd_wnd = tiwin;
1986 tp->snd_wl1 = th->th_seq;
1987 tp->snd_wl2 = th->th_ack;
1988 if (tp->snd_wnd > tp->max_sndwnd)
1989 tp->max_sndwnd = tp->snd_wnd;
1994 * Process segments with URG.
1996 if ((thflags & TH_URG) && th->th_urp &&
1997 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1999 * This is a kludge, but if we receive and accept
2000 * random urgent pointers, we'll crash in
2001 * soreceive. It's hard to imagine someone
2002 * actually wanting to send this much urgent data.
2004 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2005 th->th_urp = 0; /* XXX */
2006 thflags &= ~TH_URG; /* XXX */
2007 goto dodata; /* XXX */
2010 * If this segment advances the known urgent pointer,
2011 * then mark the data stream. This should not happen
2012 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2013 * a FIN has been received from the remote side.
2014 * In these states we ignore the URG.
2016 * According to RFC961 (Assigned Protocols),
2017 * the urgent pointer points to the last octet
2018 * of urgent data. We continue, however,
2019 * to consider it to indicate the first octet
2020 * of data past the urgent section as the original
2021 * spec states (in one of two places).
2023 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2024 tp->rcv_up = th->th_seq + th->th_urp;
2025 so->so_oobmark = so->so_rcv.sb_cc +
2026 (tp->rcv_up - tp->rcv_nxt) - 1;
2027 if (so->so_oobmark == 0)
2028 so->so_state |= SS_RCVATMARK;
2030 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2033 * Remove out of band data so doesn't get presented to user.
2034 * This can happen independent of advancing the URG pointer,
2035 * but if two URG's are pending at once, some out-of-band
2036 * data may creep in... ick.
2038 if (th->th_urp <= (u_long)tlen
2040 && (so->so_options & SO_OOBINLINE) == 0
2043 tcp_pulloutofband(so, th, m,
2044 drop_hdrlen); /* hdr drop is delayed */
2047 * If no out of band data is expected,
2048 * pull receive urgent pointer along
2049 * with the receive window.
2051 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2052 tp->rcv_up = tp->rcv_nxt;
2057 * Process the segment text, merging it into the TCP sequencing queue,
2058 * and arranging for acknowledgment of receipt if necessary.
2059 * This process logically involves adjusting tp->rcv_wnd as data
2060 * is presented to the user (this happens in tcp_usrreq.c,
2061 * case PRU_RCVD). If a FIN has already been received on this
2062 * connection then we just ignore the text.
2064 if ((tlen || (thflags & TH_FIN)) &&
2065 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2066 m_adj(m, drop_hdrlen); /* delayed header drop */
2068 * Insert segment which includes th into TCP reassembly queue
2069 * with control block tp. Set thflags to whether reassembly now
2070 * includes a segment with FIN. This handles the common case
2071 * inline (segment is the next to be received on an established
2072 * connection, and the queue is empty), avoiding linkage into
2073 * and removal from the queue and repetition of various
2075 * Set DELACK for segments received in order, but ack
2076 * immediately when segments are out of order (so
2077 * fast retransmit can work).
2079 if (th->th_seq == tp->rcv_nxt &&
2080 LIST_EMPTY(&tp->t_segq) &&
2081 TCPS_HAVEESTABLISHED(tp->t_state)) {
2083 callout_reset(tp->tt_delack, tcp_delacktime,
2084 tcp_timer_delack, tp);
2086 tp->t_flags |= TF_ACKNOW;
2087 tp->rcv_nxt += tlen;
2088 thflags = th->th_flags & TH_FIN;
2089 tcpstat.tcps_rcvpack++;
2090 tcpstat.tcps_rcvbyte += tlen;
2092 if (so->so_state & SS_CANTRCVMORE)
2095 sbappend(&so->so_rcv, m);
2098 thflags = tcp_reass(tp, th, &tlen, m);
2099 tp->t_flags |= TF_ACKNOW;
2103 * Note the amount of data that peer has sent into
2104 * our window, in order to estimate the sender's
2107 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2114 * If FIN is received ACK the FIN and let the user know
2115 * that the connection is closing.
2117 if (thflags & TH_FIN) {
2118 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2121 * If connection is half-synchronized
2122 * (ie NEEDSYN flag on) then delay ACK,
2123 * so it may be piggybacked when SYN is sent.
2124 * Otherwise, since we received a FIN then no
2125 * more input can be expected, send ACK now.
2127 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2128 callout_reset(tp->tt_delack, tcp_delacktime,
2129 tcp_timer_delack, tp);
2131 tp->t_flags |= TF_ACKNOW;
2134 switch (tp->t_state) {
2137 * In SYN_RECEIVED and ESTABLISHED STATES
2138 * enter the CLOSE_WAIT state.
2140 case TCPS_SYN_RECEIVED:
2141 tp->t_starttime = ticks;
2143 case TCPS_ESTABLISHED:
2144 tp->t_state = TCPS_CLOSE_WAIT;
2148 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2149 * enter the CLOSING state.
2151 case TCPS_FIN_WAIT_1:
2152 tp->t_state = TCPS_CLOSING;
2156 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2157 * starting the time-wait timer, turning off the other
2160 case TCPS_FIN_WAIT_2:
2161 tp->t_state = TCPS_TIME_WAIT;
2162 tcp_canceltimers(tp);
2163 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2164 if (tp->cc_recv != 0 &&
2165 (ticks - tp->t_starttime) < tcp_msl) {
2166 callout_reset(tp->tt_2msl,
2167 tp->t_rxtcur * TCPTV_TWTRUNC,
2168 tcp_timer_2msl, tp);
2169 /* For transaction client, force ACK now. */
2170 tp->t_flags |= TF_ACKNOW;
2173 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2174 tcp_timer_2msl, tp);
2175 soisdisconnected(so);
2179 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2181 case TCPS_TIME_WAIT:
2182 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2183 tcp_timer_2msl, tp);
2188 if (so->so_options & SO_DEBUG)
2189 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2194 * Return any desired output.
2196 if (needoutput || (tp->t_flags & TF_ACKNOW))
2197 (void) tcp_output(tp);
2202 * Generate an ACK dropping incoming segment if it occupies
2203 * sequence space, where the ACK reflects our state.
2205 * We can now skip the test for the RST flag since all
2206 * paths to this code happen after packets containing
2207 * RST have been dropped.
2209 * In the SYN-RECEIVED state, don't send an ACK unless the
2210 * segment we received passes the SYN-RECEIVED ACK test.
2211 * If it fails send a RST. This breaks the loop in the
2212 * "LAND" DoS attack, and also prevents an ACK storm
2213 * between two listening ports that have been sent forged
2214 * SYN segments, each with the source address of the other.
2216 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2217 (SEQ_GT(tp->snd_una, th->th_ack) ||
2218 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2219 rstreason = BANDLIM_RST_OPENPORT;
2223 if (so->so_options & SO_DEBUG)
2224 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2228 tp->t_flags |= TF_ACKNOW;
2229 (void) tcp_output(tp);
2234 * Generate a RST, dropping incoming segment.
2235 * Make ACK acceptable to originator of segment.
2236 * Don't bother to respond if destination was broadcast/multicast.
2238 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2241 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2242 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2245 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2246 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2247 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2248 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2251 /* IPv6 anycast check is done at tcp6_input() */
2254 * Perform bandwidth limiting.
2257 if (badport_bandlim(rstreason) < 0)
2262 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2263 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2266 if (thflags & TH_ACK)
2267 /* mtod() below is safe as long as hdr dropping is delayed */
2268 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2271 if (thflags & TH_SYN)
2273 /* mtod() below is safe as long as hdr dropping is delayed */
2274 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2275 (tcp_seq)0, TH_RST|TH_ACK);
2281 * Drop space held by incoming segment and return.
2284 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2285 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2293 * Parse TCP options and place in tcpopt.
2296 tcp_dooptions(to, cp, cnt, is_syn)
2304 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2306 if (opt == TCPOPT_EOL)
2308 if (opt == TCPOPT_NOP)
2314 if (optlen < 2 || optlen > cnt)
2319 if (optlen != TCPOLEN_MAXSEG)
2323 to->to_flags |= TOF_MSS;
2324 bcopy((char *)cp + 2,
2325 (char *)&to->to_mss, sizeof(to->to_mss));
2326 to->to_mss = ntohs(to->to_mss);
2329 if (optlen != TCPOLEN_WINDOW)
2333 to->to_flags |= TOF_SCALE;
2334 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2336 case TCPOPT_TIMESTAMP:
2337 if (optlen != TCPOLEN_TIMESTAMP)
2339 to->to_flags |= TOF_TS;
2340 bcopy((char *)cp + 2,
2341 (char *)&to->to_tsval, sizeof(to->to_tsval));
2342 to->to_tsval = ntohl(to->to_tsval);
2343 bcopy((char *)cp + 6,
2344 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2345 to->to_tsecr = ntohl(to->to_tsecr);
2348 if (optlen != TCPOLEN_CC)
2350 to->to_flags |= TOF_CC;
2351 bcopy((char *)cp + 2,
2352 (char *)&to->to_cc, sizeof(to->to_cc));
2353 to->to_cc = ntohl(to->to_cc);
2356 if (optlen != TCPOLEN_CC)
2360 to->to_flags |= TOF_CCNEW;
2361 bcopy((char *)cp + 2,
2362 (char *)&to->to_cc, sizeof(to->to_cc));
2363 to->to_cc = ntohl(to->to_cc);
2366 if (optlen != TCPOLEN_CC)
2370 to->to_flags |= TOF_CCECHO;
2371 bcopy((char *)cp + 2,
2372 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2373 to->to_ccecho = ntohl(to->to_ccecho);
2382 * Pull out of band byte out of a segment so
2383 * it doesn't appear in the user's data queue.
2384 * It is still reflected in the segment length for
2385 * sequencing purposes.
2388 tcp_pulloutofband(so, th, m, off)
2391 register struct mbuf *m;
2392 int off; /* delayed to be droped hdrlen */
2394 int cnt = off + th->th_urp - 1;
2397 if (m->m_len > cnt) {
2398 char *cp = mtod(m, caddr_t) + cnt;
2399 struct tcpcb *tp = sototcpcb(so);
2402 tp->t_oobflags |= TCPOOB_HAVEDATA;
2403 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2405 if (m->m_flags & M_PKTHDR)
2414 panic("tcp_pulloutofband");
2418 * Collect new round-trip time estimate
2419 * and update averages and current timeout.
2422 tcp_xmit_timer(tp, rtt)
2423 register struct tcpcb *tp;
2428 tcpstat.tcps_rttupdated++;
2430 if (tp->t_srtt != 0) {
2432 * srtt is stored as fixed point with 5 bits after the
2433 * binary point (i.e., scaled by 8). The following magic
2434 * is equivalent to the smoothing algorithm in rfc793 with
2435 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2436 * point). Adjust rtt to origin 0.
2438 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2439 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2441 if ((tp->t_srtt += delta) <= 0)
2445 * We accumulate a smoothed rtt variance (actually, a
2446 * smoothed mean difference), then set the retransmit
2447 * timer to smoothed rtt + 4 times the smoothed variance.
2448 * rttvar is stored as fixed point with 4 bits after the
2449 * binary point (scaled by 16). The following is
2450 * equivalent to rfc793 smoothing with an alpha of .75
2451 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2452 * rfc793's wired-in beta.
2456 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2457 if ((tp->t_rttvar += delta) <= 0)
2459 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2460 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2463 * No rtt measurement yet - use the unsmoothed rtt.
2464 * Set the variance to half the rtt (so our first
2465 * retransmit happens at 3*rtt).
2467 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2468 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2469 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2475 * the retransmit should happen at rtt + 4 * rttvar.
2476 * Because of the way we do the smoothing, srtt and rttvar
2477 * will each average +1/2 tick of bias. When we compute
2478 * the retransmit timer, we want 1/2 tick of rounding and
2479 * 1 extra tick because of +-1/2 tick uncertainty in the
2480 * firing of the timer. The bias will give us exactly the
2481 * 1.5 tick we need. But, because the bias is
2482 * statistical, we have to test that we don't drop below
2483 * the minimum feasible timer (which is 2 ticks).
2485 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2486 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2489 * We received an ack for a packet that wasn't retransmitted;
2490 * it is probably safe to discard any error indications we've
2491 * received recently. This isn't quite right, but close enough
2492 * for now (a route might have failed after we sent a segment,
2493 * and the return path might not be symmetrical).
2495 tp->t_softerror = 0;
2499 * Determine a reasonable value for maxseg size.
2500 * If the route is known, check route for mtu.
2501 * If none, use an mss that can be handled on the outgoing
2502 * interface without forcing IP to fragment; if bigger than
2503 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2504 * to utilize large mbufs. If no route is found, route has no mtu,
2505 * or the destination isn't local, use a default, hopefully conservative
2506 * size (usually 512 or the default IP max size, but no more than the mtu
2507 * of the interface), as we can't discover anything about intervening
2508 * gateways or networks. We also initialize the congestion/slow start
2509 * window to be a single segment if the destination isn't local.
2510 * While looking at the routing entry, we also initialize other path-dependent
2511 * parameters from pre-set or cached values in the routing entry.
2513 * Also take into account the space needed for options that we
2514 * send regularly. Make maxseg shorter by that amount to assure
2515 * that we can send maxseg amount of data even when the options
2516 * are present. Store the upper limit of the length of options plus
2519 * NOTE that this routine is only called when we process an incoming
2520 * segment, for outgoing segments only tcp_mssopt is called.
2522 * In case of T/TCP, we call this routine during implicit connection
2523 * setup as well (offer = -1), to initialize maxseg from the cached
2531 register struct rtentry *rt;
2533 register int rtt, mss;
2535 struct inpcb *inp = tp->t_inpcb;
2537 struct rmxp_tao *taop;
2538 int origoffer = offer;
2540 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2541 size_t min_protoh = isipv6 ?
2542 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2543 sizeof(struct tcpiphdr);
2545 const int isipv6 = 0;
2546 const size_t min_protoh = sizeof(struct tcpiphdr);
2550 rt = tcp_rtlookup6(&inp->inp_inc);
2552 rt = tcp_rtlookup(&inp->inp_inc);
2554 tp->t_maxopd = tp->t_maxseg =
2555 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2559 so = inp->inp_socket;
2561 taop = rmx_taop(rt->rt_rmx);
2563 * Offer == -1 means that we didn't receive SYN yet,
2564 * use cached value in that case;
2567 offer = taop->tao_mssopt;
2569 * Offer == 0 means that there was no MSS on the SYN segment,
2570 * in this case we use tcp_mssdflt.
2573 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2576 * Sanity check: make sure that maxopd will be large
2577 * enough to allow some data on segments even is the
2578 * all the option space is used (40bytes). Otherwise
2579 * funny things may happen in tcp_output.
2581 offer = max(offer, 64);
2582 taop->tao_mssopt = offer;
2585 * While we're here, check if there's an initial rtt
2586 * or rttvar. Convert from the route-table units
2587 * to scaled multiples of the slow timeout timer.
2589 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2591 * XXX the lock bit for RTT indicates that the value
2592 * is also a minimum value; this is subject to time.
2594 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2595 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2596 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2597 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2598 tcpstat.tcps_usedrtt++;
2599 if (rt->rt_rmx.rmx_rttvar) {
2600 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2601 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2602 tcpstat.tcps_usedrttvar++;
2604 /* default variation is +- 1 rtt */
2606 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2608 TCPT_RANGESET(tp->t_rxtcur,
2609 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2610 tp->t_rttmin, TCPTV_REXMTMAX);
2613 * if there's an mtu associated with the route, use it
2614 * else, use the link mtu.
2616 if (rt->rt_rmx.rmx_mtu)
2617 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2620 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2622 if (!in6_localaddr(&inp->in6p_faddr))
2623 mss = min(mss, tcp_v6mssdflt);
2625 mss = ifp->if_mtu - min_protoh;
2626 if (!in_localaddr(inp->inp_faddr))
2627 mss = min(mss, tcp_mssdflt);
2630 mss = min(mss, offer);
2632 * maxopd stores the maximum length of data AND options
2633 * in a segment; maxseg is the amount of data in a normal
2634 * segment. We need to store this value (maxopd) apart
2635 * from maxseg, because now every segment carries options
2636 * and thus we normally have somewhat less data in segments.
2641 * In case of T/TCP, origoffer==-1 indicates, that no segments
2642 * were received yet. In this case we just guess, otherwise
2643 * we do the same as before T/TCP.
2645 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2647 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2648 mss -= TCPOLEN_TSTAMP_APPA;
2649 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2651 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2652 mss -= TCPOLEN_CC_APPA;
2654 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2656 mss &= ~(MCLBYTES-1);
2659 mss = mss / MCLBYTES * MCLBYTES;
2662 * If there's a pipesize, change the socket buffer
2663 * to that size. Make the socket buffers an integral
2664 * number of mss units; if the mss is larger than
2665 * the socket buffer, decrease the mss.
2668 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2670 bufsize = so->so_snd.sb_hiwat;
2674 bufsize = roundup(bufsize, mss);
2675 if (bufsize > sb_max)
2677 if (bufsize > so->so_snd.sb_hiwat)
2678 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2683 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2685 bufsize = so->so_rcv.sb_hiwat;
2686 if (bufsize > mss) {
2687 bufsize = roundup(bufsize, mss);
2688 if (bufsize > sb_max)
2690 if (bufsize > so->so_rcv.sb_hiwat)
2691 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2695 * Set the slow-start flight size depending on whether this
2696 * is a local network or not.
2698 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2699 (!isipv6 && in_localaddr(inp->inp_faddr)))
2700 tp->snd_cwnd = mss * ss_fltsz_local;
2702 tp->snd_cwnd = mss * ss_fltsz;
2704 if (rt->rt_rmx.rmx_ssthresh) {
2706 * There's some sort of gateway or interface
2707 * buffer limit on the path. Use this to set
2708 * the slow start threshhold, but set the
2709 * threshold to no less than 2*mss.
2711 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2712 tcpstat.tcps_usedssthresh++;
2717 * Determine the MSS option to send on an outgoing SYN.
2725 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2726 int min_protoh = isipv6 ?
2727 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2728 sizeof(struct tcpiphdr);
2730 const int isipv6 = 0;
2731 const size_t min_protoh = sizeof(struct tcpiphdr);
2735 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2737 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2739 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2741 return (rt->rt_ifp->if_mtu - min_protoh);
2746 * When a partial ack arrives, force the retransmission of the
2747 * next unacknowledged segment. Do not clear tp->t_dupacks.
2748 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2752 tcp_newreno_partial_ack(tp, th)
2756 tcp_seq onxt = tp->snd_nxt;
2757 u_long ocwnd = tp->snd_cwnd;
2759 callout_stop(tp->tt_rexmt);
2761 tp->snd_nxt = th->th_ack;
2763 * Set snd_cwnd to one segment beyond acknowledged offset
2764 * (tp->snd_una has not yet been updated when this function is called.)
2766 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2767 tp->t_flags |= TF_ACKNOW;
2768 (void) tcp_output(tp);
2769 tp->snd_cwnd = ocwnd;
2770 if (SEQ_GT(onxt, tp->snd_nxt))
2773 * Partial window deflation. Relies on fact that tp->snd_una
2776 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);