2 * Copyright (c) 2002-2003 Jeffrey Hsu
3 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
4 * The Regents of the University of California. All rights reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 * must display the following acknowledgement:
16 * This product includes software developed by the University of
17 * California, Berkeley and its contributors.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
35 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
36 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.9 2003/08/14 23:09:33 hsu Exp $
39 #include "opt_ipfw.h" /* for ipfw_fwd */
40 #include "opt_inet6.h"
41 #include "opt_ipsec.h"
42 #include "opt_tcpdebug.h"
43 #include "opt_tcp_input.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/sysctl.h>
49 #include <sys/malloc.h>
51 #include <sys/proc.h> /* for proc0 declaration */
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/syslog.h>
57 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
60 #include <net/route.h>
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
66 #include <netinet/in_var.h>
67 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
68 #include <netinet/in_pcb.h>
69 #include <netinet/ip_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/in6_pcb.h>
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_fsm.h>
77 #include <netinet/tcp_seq.h>
78 #include <netinet/tcp_timer.h>
79 #include <netinet/tcp_var.h>
80 #include <netinet6/tcp6_var.h>
81 #include <netinet/tcpip.h>
83 #include <netinet/tcp_debug.h>
85 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
86 struct tcphdr tcp_savetcp;
90 #include <netipsec/ipsec.h>
91 #include <netipsec/ipsec6.h>
95 #include <netinet6/ipsec.h>
96 #include <netinet6/ipsec6.h>
97 #include <netproto/key/key.h>
100 #include <machine/in_cksum.h>
102 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
104 static const int tcprexmtthresh = 3;
107 struct tcpstat tcpstat;
108 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
109 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
111 static int log_in_vain = 0;
112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
113 &log_in_vain, 0, "Log all incoming TCP connections");
115 static int blackhole = 0;
116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
117 &blackhole, 0, "Do not send RST when dropping refused connections");
119 int tcp_delack_enabled = 1;
120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
121 &tcp_delack_enabled, 0,
122 "Delay ACK to try and piggyback it onto a data packet");
124 #ifdef TCP_DROP_SYNFIN
125 static int drop_synfin = 0;
126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
127 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
130 static int tcp_do_limitedtransmit = 1;
131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
132 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
134 static int tcp_do_rfc3390 = 1;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
137 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
139 static int tcp_do_eifel_detect = 1;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
141 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
143 struct inpcbhead tcb;
144 #define tcb6 tcb /* for KAME src sync over BSD*'s */
145 struct inpcbinfo tcbinfo;
147 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
148 static void tcp_pulloutofband(struct socket *,
149 struct tcphdr *, struct mbuf *, int);
150 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
152 static void tcp_xmit_timer(struct tcpcb *, int);
153 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
155 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
157 #define ND6_HINT(tp) \
159 if ((tp) && (tp)->t_inpcb && \
160 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
161 (tp)->t_inpcb->in6p_route.ro_rt) \
162 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
169 * Indicate whether this ack should be delayed. We can delay the ack if
170 * - delayed acks are enabled and
171 * - there is no delayed ack timer in progress and
172 * - our last ack wasn't a 0-sized window. We never want to delay
173 * the ack that opens up a 0-sized window.
175 #define DELAY_ACK(tp) \
176 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
177 (tp->t_flags & TF_RXWIN0SENT) == 0)
180 tcp_reass(tp, th, tlenp, m)
187 struct tseg_qent *p = NULL;
188 struct tseg_qent *nq;
189 struct tseg_qent *te;
190 struct socket *so = tp->t_inpcb->inp_socket;
194 * Call with th==0 after become established to
195 * force pre-ESTABLISHED data up to user socket.
200 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
201 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
204 tcpstat.tcps_rcvmemdrop++;
210 * Find a segment which begins after this one does.
212 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
213 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
219 * If there is a preceding segment, it may provide some of
220 * our data already. If so, drop the data from the incoming
221 * segment. If it provides all of our data, drop us.
225 /* conversion to int (in i) handles seq wraparound */
226 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
229 tcpstat.tcps_rcvduppack++;
230 tcpstat.tcps_rcvdupbyte += *tlenp;
234 * Try to present any queued data
235 * at the left window edge to the user.
236 * This is needed after the 3-WHS
239 goto present; /* ??? */
246 tcpstat.tcps_rcvoopack++;
247 tcpstat.tcps_rcvoobyte += *tlenp;
250 * While we overlap succeeding segments trim them or,
251 * if they are completely covered, dequeue them.
254 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
257 if (i < q->tqe_len) {
258 q->tqe_th->th_seq += i;
264 nq = LIST_NEXT(q, tqe_q);
265 LIST_REMOVE(q, tqe_q);
271 /* Insert the new segment queue entry into place. */
274 te->tqe_len = *tlenp;
277 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
279 LIST_INSERT_AFTER(p, te, tqe_q);
284 * Present data to user, advancing rcv_nxt through
285 * completed sequence space.
287 if (!TCPS_HAVEESTABLISHED(tp->t_state))
289 q = LIST_FIRST(&tp->t_segq);
290 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
293 tp->rcv_nxt += q->tqe_len;
294 flags = q->tqe_th->th_flags & TH_FIN;
295 nq = LIST_NEXT(q, tqe_q);
296 LIST_REMOVE(q, tqe_q);
297 if (so->so_state & SS_CANTRCVMORE)
300 sbappend(&so->so_rcv, q->tqe_m);
303 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
310 * TCP input routine, follows pages 65-76 of the
311 * protocol specification dated September, 1981 very closely.
315 tcp6_input(mp, offp, proto)
319 struct mbuf *m = *mp;
320 struct in6_ifaddr *ia6;
322 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
325 * draft-itojun-ipv6-tcp-to-anycast
326 * better place to put this in?
328 ia6 = ip6_getdstifaddr(m);
329 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
332 ip6 = mtod(m, struct ip6_hdr *);
333 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
334 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
338 tcp_input(m, *offp, proto);
344 tcp_input(m, off0, proto)
349 struct ip *ip = NULL;
351 struct inpcb *inp = NULL;
356 struct tcpcb *tp = NULL;
358 struct socket *so = 0;
359 int todrop, acked, ourfinisacked, needoutput = 0;
361 struct tcpopt to; /* options in this segment */
362 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
363 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
364 struct sockaddr_in *next_hop = NULL;
365 int rstreason; /* For badport_bandlim accounting purposes */
366 int useTS; /* use timestamps in Eifel detection */
367 struct ip6_hdr *ip6 = NULL;
371 const int isipv6 = 0;
377 /* Grab info from MT_TAG mbufs prepended to the chain. */
378 for (;m && m->m_type == MT_TAG; m = m->m_next) {
379 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
380 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
383 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
385 bzero((char *)&to, sizeof(to));
387 tcpstat.tcps_rcvtotal++;
390 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
391 ip6 = mtod(m, struct ip6_hdr *);
392 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
393 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
394 tcpstat.tcps_rcvbadsum++;
397 th = (struct tcphdr *)((caddr_t)ip6 + off0);
400 * Be proactive about unspecified IPv6 address in source.
401 * As we use all-zero to indicate unbounded/unconnected pcb,
402 * unspecified IPv6 address can be used to confuse us.
404 * Note that packets with unspecified IPv6 destination is
405 * already dropped in ip6_input.
407 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
413 * Get IP and TCP header together in first mbuf.
414 * Note: IP leaves IP header in first mbuf.
416 if (off0 > sizeof(struct ip)) {
417 ip_stripoptions(m, (struct mbuf *)0);
418 off0 = sizeof(struct ip);
420 if (m->m_len < sizeof(struct tcpiphdr)) {
421 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
422 tcpstat.tcps_rcvshort++;
426 ip = mtod(m, struct ip *);
427 ipov = (struct ipovly *)ip;
428 th = (struct tcphdr *)((caddr_t)ip + off0);
431 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
432 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
433 th->th_sum = m->m_pkthdr.csum_data;
435 th->th_sum = in_pseudo(ip->ip_src.s_addr,
437 htonl(m->m_pkthdr.csum_data +
440 th->th_sum ^= 0xffff;
443 * Checksum extended TCP header and data.
445 len = sizeof(struct ip) + tlen;
446 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
447 ipov->ih_len = (u_short)tlen;
448 ipov->ih_len = htons(ipov->ih_len);
449 th->th_sum = in_cksum(m, len);
452 tcpstat.tcps_rcvbadsum++;
456 /* Re-initialization for later version check */
457 ip->ip_v = IPVERSION;
462 * Check that TCP offset makes sense,
463 * pull out TCP options and adjust length. XXX
465 off = th->th_off << 2;
466 if (off < sizeof(struct tcphdr) || off > tlen) {
467 tcpstat.tcps_rcvbadoff++;
470 tlen -= off; /* tlen is used instead of ti->ti_len */
471 if (off > sizeof(struct tcphdr)) {
473 IP6_EXTHDR_CHECK(m, off0, off, );
474 ip6 = mtod(m, struct ip6_hdr *);
475 th = (struct tcphdr *)((caddr_t)ip6 + off0);
477 if (m->m_len < sizeof(struct ip) + off) {
478 if ((m = m_pullup(m, sizeof(struct ip) + off))
480 tcpstat.tcps_rcvshort++;
483 ip = mtod(m, struct ip *);
484 ipov = (struct ipovly *)ip;
485 th = (struct tcphdr *)((caddr_t)ip + off0);
488 optlen = off - sizeof(struct tcphdr);
489 optp = (u_char *)(th + 1);
491 thflags = th->th_flags;
493 #ifdef TCP_DROP_SYNFIN
495 * If the drop_synfin option is enabled, drop all packets with
496 * both the SYN and FIN bits set. This prevents e.g. nmap from
497 * identifying the TCP/IP stack.
499 * This is a violation of the TCP specification.
501 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
506 * Convert TCP protocol specific fields to host format.
508 th->th_seq = ntohl(th->th_seq);
509 th->th_ack = ntohl(th->th_ack);
510 th->th_win = ntohs(th->th_win);
511 th->th_urp = ntohs(th->th_urp);
514 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
515 * until after ip6_savecontrol() is called and before other functions
516 * which don't want those proto headers.
517 * Because ip6_savecontrol() is going to parse the mbuf to
518 * search for data to be passed up to user-land, it wants mbuf
519 * parameters to be unchanged.
520 * XXX: the call of ip6_savecontrol() has been obsoleted based on
521 * latest version of the advanced API (20020110).
523 drop_hdrlen = off0 + off;
526 * Locate pcb for segment.
529 /* IPFIREWALL_FORWARD section */
530 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
532 * Transparently forwarded. Pretend to be the destination.
533 * already got one like this?
535 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
536 ip->ip_dst, th->th_dport,
537 0, m->m_pkthdr.rcvif);
539 /* It's new. Try find the ambushing socket. */
540 inp = in_pcblookup_hash(&tcbinfo,
541 ip->ip_src, th->th_sport,
544 ntohs(next_hop->sin_port) :
546 1, m->m_pkthdr.rcvif);
550 inp = in6_pcblookup_hash(&tcbinfo,
551 &ip6->ip6_src, th->th_sport,
552 &ip6->ip6_dst, th->th_dport,
553 1, m->m_pkthdr.rcvif);
555 inp = in_pcblookup_hash(&tcbinfo,
556 ip->ip_src, th->th_sport,
557 ip->ip_dst, th->th_dport,
558 1, m->m_pkthdr.rcvif);
563 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
564 ipsec6stat.in_polvio++;
568 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
569 ipsecstat.in_polvio++;
576 if (inp != NULL && ipsec6_in_reject(m, inp)) {
580 if (inp != NULL && ipsec4_in_reject(m, inp)) {
587 * If the state is CLOSED (i.e., TCB does not exist) then
588 * all data in the incoming segment is discarded.
589 * If the TCB exists but is in CLOSED state, it is embryonic,
590 * but should either do a listen or a connect soon.
595 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
597 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
602 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
603 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
607 strcpy(dbuf, inet_ntoa(ip->ip_dst));
608 strcpy(sbuf, inet_ntoa(ip->ip_src));
610 switch (log_in_vain) {
612 if ((thflags & TH_SYN) == 0)
616 "Connection attempt to TCP %s:%d "
617 "from %s:%d flags:0x%02x\n",
618 dbuf, ntohs(th->th_dport), sbuf,
619 ntohs(th->th_sport), thflags);
628 if (thflags & TH_SYN)
637 rstreason = BANDLIM_RST_CLOSEDPORT;
642 rstreason = BANDLIM_RST_CLOSEDPORT;
645 if (tp->t_state == TCPS_CLOSED)
648 /* Unscale the window into a 32-bit value. */
649 if ((thflags & TH_SYN) == 0)
650 tiwin = th->th_win << tp->snd_scale;
654 so = inp->inp_socket;
655 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
656 struct in_conninfo inc;
658 if (so->so_options & SO_DEBUG) {
659 ostate = tp->t_state;
661 bcopy((char *)ip6, (char *)tcp_saveipgen,
664 bcopy((char *)ip, (char *)tcp_saveipgen,
669 /* skip if this isn't a listen socket */
670 if ((so->so_options & SO_ACCEPTCONN) == 0)
673 inc.inc_isipv6 = isipv6;
676 inc.inc6_faddr = ip6->ip6_src;
677 inc.inc6_laddr = ip6->ip6_dst;
678 inc.inc6_route.ro_rt = NULL; /* XXX */
680 inc.inc_faddr = ip->ip_src;
681 inc.inc_laddr = ip->ip_dst;
682 inc.inc_route.ro_rt = NULL; /* XXX */
684 inc.inc_fport = th->th_sport;
685 inc.inc_lport = th->th_dport;
688 * If the state is LISTEN then ignore segment if it contains
689 * a RST. If the segment contains an ACK then it is bad and
690 * send a RST. If it does not contain a SYN then it is not
691 * interesting; drop it.
693 * If the state is SYN_RECEIVED (syncache) and seg contains
694 * an ACK, but not for our SYN/ACK, send a RST. If the seg
695 * contains a RST, check the sequence number to see if it
696 * is a valid reset segment.
698 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
699 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
700 if (!syncache_expand(&inc, th, &so, m)) {
702 * No syncache entry, or ACK was not
703 * for our SYN/ACK. Send a RST.
705 tcpstat.tcps_badsyn++;
706 rstreason = BANDLIM_RST_OPENPORT;
711 * Could not complete 3-way handshake,
712 * connection is being closed down, and
713 * syncache will free mbuf.
717 * Socket is created in state SYN_RECEIVED.
718 * Continue processing segment.
723 * This is what would have happened in
724 * tcp_output() when the SYN,ACK was sent.
726 tp->snd_up = tp->snd_una;
727 tp->snd_max = tp->snd_nxt = tp->iss + 1;
728 tp->last_ack_sent = tp->rcv_nxt;
730 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
731 * until the _second_ ACK is received:
732 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
733 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
734 * move to ESTAB, set snd_wnd to tiwin.
736 tp->snd_wnd = tiwin; /* unscaled */
739 if (thflags & TH_RST) {
740 syncache_chkrst(&inc, th);
743 if (thflags & TH_ACK) {
744 syncache_badack(&inc);
745 tcpstat.tcps_badsyn++;
746 rstreason = BANDLIM_RST_OPENPORT;
753 * Segment's flags are (SYN) or (SYN|FIN).
757 * If deprecated address is forbidden,
758 * we do not accept SYN to deprecated interface
759 * address to prevent any new inbound connection from
760 * getting established.
761 * When we do not accept SYN, we send a TCP RST,
762 * with deprecated source address (instead of dropping
763 * it). We compromise it as it is much better for peer
764 * to send a RST, and RST will be the final packet
767 * If we do not forbid deprecated addresses, we accept
768 * the SYN packet. RFC2462 does not suggest dropping
770 * If we decipher RFC2462 5.5.4, it says like this:
771 * 1. use of deprecated addr with existing
772 * communication is okay - "SHOULD continue to be
774 * 2. use of it with new communication:
775 * (2a) "SHOULD NOT be used if alternate address
776 * with sufficient scope is available"
777 * (2b) nothing mentioned otherwise.
778 * Here we fall into (2b) case as we have no choice in
779 * our source address selection - we must obey the peer.
781 * The wording in RFC2462 is confusing, and there are
782 * multiple description text for deprecated address
783 * handling - worse, they are not exactly the same.
784 * I believe 5.5.4 is the best one, so we follow 5.5.4.
786 if (isipv6 && !ip6_use_deprecated) {
787 struct in6_ifaddr *ia6;
789 if ((ia6 = ip6_getdstifaddr(m)) &&
790 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
792 rstreason = BANDLIM_RST_OPENPORT;
798 * If it is from this socket, drop it, it must be forged.
799 * Don't bother responding if the destination was a broadcast.
801 if (th->th_dport == th->th_sport) {
803 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
807 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
812 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
814 * Note that it is quite possible to receive unicast
815 * link-layer packets with a broadcast IP address. Use
816 * in_broadcast() to find them.
818 if (m->m_flags & (M_BCAST|M_MCAST))
821 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
822 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
825 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
826 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
827 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
828 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
832 * SYN appears to be valid; create compressed TCP state
833 * for syncache, or perform t/tcp connection.
835 if (so->so_qlen <= so->so_qlimit) {
836 tcp_dooptions(&to, optp, optlen, 1);
837 if (!syncache_add(&inc, &to, th, &so, m))
841 * Entry added to syncache, mbuf used to
842 * send SYN,ACK packet.
846 * Segment passed TAO tests.
851 tp->t_starttime = ticks;
852 tp->t_state = TCPS_ESTABLISHED;
855 * If there is a FIN, or if there is data and the
856 * connection is local, then delay SYN,ACK(SYN) in
857 * the hope of piggy-backing it on a response
858 * segment. Otherwise must send ACK now in case
859 * the other side is slow starting.
862 ((thflags & TH_FIN) ||
864 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
865 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
866 callout_reset(tp->tt_delack, tcp_delacktime,
867 tcp_timer_delack, tp);
868 tp->t_flags |= TF_NEEDSYN;
870 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
872 tcpstat.tcps_connects++;
880 /* XXX temp debugging */
881 /* should not happen - syncache should pick up these connections */
882 if (tp->t_state == TCPS_LISTEN)
883 panic("tcp_input: TCPS_LISTEN");
886 * Segment received on connection.
887 * Reset idle time and keep-alive timer.
889 tp->t_rcvtime = ticks;
890 if (TCPS_HAVEESTABLISHED(tp->t_state))
891 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
895 * XXX this is tradtitional behavior, may need to be cleaned up.
897 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
898 if (thflags & TH_SYN) {
899 if (to.to_flags & TOF_SCALE) {
900 tp->t_flags |= TF_RCVD_SCALE;
901 tp->requested_s_scale = to.to_requested_s_scale;
903 if (to.to_flags & TOF_TS) {
904 tp->t_flags |= TF_RCVD_TSTMP;
905 tp->ts_recent = to.to_tsval;
906 tp->ts_recent_age = ticks;
908 if (to.to_flags & (TOF_CC|TOF_CCNEW))
909 tp->t_flags |= TF_RCVD_CC;
910 if (to.to_flags & TOF_MSS)
911 tcp_mss(tp, to.to_mss);
915 * Header prediction: check for the two common cases
916 * of a uni-directional data xfer. If the packet has
917 * no control flags, is in-sequence, the window didn't
918 * change and we're not retransmitting, it's a
919 * candidate. If the length is zero and the ack moved
920 * forward, we're the sender side of the xfer. Just
921 * free the data acked & wake any higher level process
922 * that was blocked waiting for space. If the length
923 * is non-zero and the ack didn't move, we're the
924 * receiver side. If we're getting packets in-order
925 * (the reassembly queue is empty), add the data to
926 * the socket buffer and note that we need a delayed ack.
927 * Make sure that the hidden state-flags are also off.
928 * Since we check for TCPS_ESTABLISHED above, it can only
931 if (tp->t_state == TCPS_ESTABLISHED &&
932 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
933 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
934 ((to.to_flags & TOF_TS) == 0 ||
935 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
937 * Using the CC option is compulsory if once started:
938 * the segment is OK if no T/TCP was negotiated or
939 * if the segment has a CC option equal to CCrecv
941 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
942 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
943 th->th_seq == tp->rcv_nxt &&
944 tiwin && tiwin == tp->snd_wnd &&
945 tp->snd_nxt == tp->snd_max) {
948 * If last ACK falls within this segment's sequence numbers,
949 * record the timestamp.
950 * NOTE that the test is modified according to the latest
951 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
953 if ((to.to_flags & TOF_TS) != 0 &&
954 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
955 tp->ts_recent_age = ticks;
956 tp->ts_recent = to.to_tsval;
960 if (SEQ_GT(th->th_ack, tp->snd_una) &&
961 SEQ_LEQ(th->th_ack, tp->snd_max) &&
962 tp->snd_cwnd >= tp->snd_wnd &&
964 tp->t_dupacks < tcprexmtthresh) ||
965 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
967 * this is a pure ack for outstanding data.
969 ++tcpstat.tcps_predack;
971 * "bad retransmit" recovery
973 * If Eifel detection applies, then
974 * it is deterministic, so use it
975 * unconditionally over the old heuristic.
976 * Otherwise, fall back to the old heuristic.
978 if (tcp_do_eifel_detect &&
979 (to.to_flags & TOF_TS) && to.to_tsecr &&
980 (tp->t_flags & TF_FIRSTACCACK)) {
981 /* Eifel detection applicable. */
982 if (to.to_tsecr < tp->t_rexmtTS) {
983 tcp_revert_congestion_state(tp);
984 ++tcpstat.tcps_eifeldetected;
986 } else if (tp->t_rxtshift == 1 &&
987 ticks < tp->t_badrxtwin) {
988 tcp_revert_congestion_state(tp);
989 ++tcpstat.tcps_rttdetected;
991 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
993 * Recalculate the retransmit timer / rtt.
995 * Some machines (certain windows boxes)
996 * send broken timestamp replies during the
997 * SYN+ACK phase, ignore timestamps of 0.
999 if ((to.to_flags & TOF_TS) != 0 &&
1002 ticks - to.to_tsecr + 1);
1003 } else if (tp->t_rtttime &&
1004 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1006 ticks - tp->t_rtttime);
1008 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1009 acked = th->th_ack - tp->snd_una;
1010 tcpstat.tcps_rcvackpack++;
1011 tcpstat.tcps_rcvackbyte += acked;
1012 sbdrop(&so->so_snd, acked);
1013 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1014 SEQ_LEQ(th->th_ack, tp->snd_recover))
1015 tp->snd_recover = th->th_ack - 1;
1016 tp->snd_una = th->th_ack;
1019 ND6_HINT(tp); /* some progress has been done */
1022 * If all outstanding data are acked, stop
1023 * retransmit timer, otherwise restart timer
1024 * using current (possibly backed-off) value.
1025 * If process is waiting for space,
1026 * wakeup/selwakeup/signal. If data
1027 * are ready to send, let tcp_output
1028 * decide between more output or persist.
1030 if (tp->snd_una == tp->snd_max)
1031 callout_stop(tp->tt_rexmt);
1032 else if (!callout_active(tp->tt_persist))
1033 callout_reset(tp->tt_rexmt,
1035 tcp_timer_rexmt, tp);
1038 if (so->so_snd.sb_cc)
1039 (void) tcp_output(tp);
1042 } else if (th->th_ack == tp->snd_una &&
1043 LIST_EMPTY(&tp->t_segq) &&
1044 tlen <= sbspace(&so->so_rcv)) {
1046 * this is a pure, in-sequence data packet
1047 * with nothing on the reassembly queue and
1048 * we have enough buffer space to take it.
1050 ++tcpstat.tcps_preddat;
1051 tp->rcv_nxt += tlen;
1052 tcpstat.tcps_rcvpack++;
1053 tcpstat.tcps_rcvbyte += tlen;
1054 ND6_HINT(tp); /* some progress has been done */
1056 * Add data to socket buffer.
1058 if (so->so_state & SS_CANTRCVMORE) {
1061 m_adj(m, drop_hdrlen); /* delayed header drop */
1062 sbappend(&so->so_rcv, m);
1065 if (DELAY_ACK(tp)) {
1066 callout_reset(tp->tt_delack, tcp_delacktime,
1067 tcp_timer_delack, tp);
1069 tp->t_flags |= TF_ACKNOW;
1077 * Calculate amount of space in receive window,
1078 * and then do TCP input processing.
1079 * Receive window is amount of space in rcv queue,
1080 * but not less than advertised window.
1084 win = sbspace(&so->so_rcv);
1087 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1090 switch (tp->t_state) {
1093 * If the state is SYN_RECEIVED:
1094 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1096 case TCPS_SYN_RECEIVED:
1097 if ((thflags & TH_ACK) &&
1098 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1099 SEQ_GT(th->th_ack, tp->snd_max))) {
1100 rstreason = BANDLIM_RST_OPENPORT;
1106 * If the state is SYN_SENT:
1107 * if seg contains an ACK, but not for our SYN, drop the input.
1108 * if seg contains a RST, then drop the connection.
1109 * if seg does not contain SYN, then drop it.
1110 * Otherwise this is an acceptable SYN segment
1111 * initialize tp->rcv_nxt and tp->irs
1112 * if seg contains ack then advance tp->snd_una
1113 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1114 * arrange for segment to be acked (eventually)
1115 * continue processing rest of data/controls, beginning with URG
1118 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1119 taop = &tao_noncached;
1120 bzero(taop, sizeof(*taop));
1123 if ((thflags & TH_ACK) &&
1124 (SEQ_LEQ(th->th_ack, tp->iss) ||
1125 SEQ_GT(th->th_ack, tp->snd_max))) {
1127 * If we have a cached CCsent for the remote host,
1128 * hence we haven't just crashed and restarted,
1129 * do not send a RST. This may be a retransmission
1130 * from the other side after our earlier ACK was lost.
1131 * Our new SYN, when it arrives, will serve as the
1134 if (taop->tao_ccsent != 0)
1137 rstreason = BANDLIM_UNLIMITED;
1141 if (thflags & TH_RST) {
1142 if (thflags & TH_ACK)
1143 tp = tcp_drop(tp, ECONNREFUSED);
1146 if ((thflags & TH_SYN) == 0)
1148 tp->snd_wnd = th->th_win; /* initial send window */
1149 tp->cc_recv = to.to_cc; /* foreign CC */
1151 tp->irs = th->th_seq;
1153 if (thflags & TH_ACK) {
1155 * Our SYN was acked. If segment contains CC.ECHO
1156 * option, check it to make sure this segment really
1157 * matches our SYN. If not, just drop it as old
1158 * duplicate, but send an RST if we're still playing
1159 * by the old rules. If no CC.ECHO option, make sure
1160 * we don't get fooled into using T/TCP.
1162 if (to.to_flags & TOF_CCECHO) {
1163 if (tp->cc_send != to.to_ccecho) {
1164 if (taop->tao_ccsent != 0)
1167 rstreason = BANDLIM_UNLIMITED;
1172 tp->t_flags &= ~TF_RCVD_CC;
1173 tcpstat.tcps_connects++;
1175 /* Do window scaling on this connection? */
1176 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1177 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1178 tp->snd_scale = tp->requested_s_scale;
1179 tp->rcv_scale = tp->request_r_scale;
1181 /* Segment is acceptable, update cache if undefined. */
1182 if (taop->tao_ccsent == 0)
1183 taop->tao_ccsent = to.to_ccecho;
1185 tp->rcv_adv += tp->rcv_wnd;
1186 tp->snd_una++; /* SYN is acked */
1188 * If there's data, delay ACK; if there's also a FIN
1189 * ACKNOW will be turned on later.
1191 if (DELAY_ACK(tp) && tlen != 0)
1192 callout_reset(tp->tt_delack, tcp_delacktime,
1193 tcp_timer_delack, tp);
1195 tp->t_flags |= TF_ACKNOW;
1197 * Received <SYN,ACK> in SYN_SENT[*] state.
1199 * SYN_SENT --> ESTABLISHED
1200 * SYN_SENT* --> FIN_WAIT_1
1202 tp->t_starttime = ticks;
1203 if (tp->t_flags & TF_NEEDFIN) {
1204 tp->t_state = TCPS_FIN_WAIT_1;
1205 tp->t_flags &= ~TF_NEEDFIN;
1208 tp->t_state = TCPS_ESTABLISHED;
1209 callout_reset(tp->tt_keep, tcp_keepidle,
1210 tcp_timer_keep, tp);
1214 * Received initial SYN in SYN-SENT[*] state =>
1215 * simultaneous open. If segment contains CC option
1216 * and there is a cached CC, apply TAO test.
1217 * If it succeeds, connection is * half-synchronized.
1218 * Otherwise, do 3-way handshake:
1219 * SYN-SENT -> SYN-RECEIVED
1220 * SYN-SENT* -> SYN-RECEIVED*
1221 * If there was no CC option, clear cached CC value.
1223 tp->t_flags |= TF_ACKNOW;
1224 callout_stop(tp->tt_rexmt);
1225 if (to.to_flags & TOF_CC) {
1226 if (taop->tao_cc != 0 &&
1227 CC_GT(to.to_cc, taop->tao_cc)) {
1229 * update cache and make transition:
1230 * SYN-SENT -> ESTABLISHED*
1231 * SYN-SENT* -> FIN-WAIT-1*
1233 taop->tao_cc = to.to_cc;
1234 tp->t_starttime = ticks;
1235 if (tp->t_flags & TF_NEEDFIN) {
1236 tp->t_state = TCPS_FIN_WAIT_1;
1237 tp->t_flags &= ~TF_NEEDFIN;
1239 tp->t_state = TCPS_ESTABLISHED;
1240 callout_reset(tp->tt_keep,
1245 tp->t_flags |= TF_NEEDSYN;
1247 tp->t_state = TCPS_SYN_RECEIVED;
1249 /* CC.NEW or no option => invalidate cache */
1251 tp->t_state = TCPS_SYN_RECEIVED;
1257 * Advance th->th_seq to correspond to first data byte.
1258 * If data, trim to stay within window,
1259 * dropping FIN if necessary.
1262 if (tlen > tp->rcv_wnd) {
1263 todrop = tlen - tp->rcv_wnd;
1267 tcpstat.tcps_rcvpackafterwin++;
1268 tcpstat.tcps_rcvbyteafterwin += todrop;
1270 tp->snd_wl1 = th->th_seq - 1;
1271 tp->rcv_up = th->th_seq;
1273 * Client side of transaction: already sent SYN and data.
1274 * If the remote host used T/TCP to validate the SYN,
1275 * our data will be ACK'd; if so, enter normal data segment
1276 * processing in the middle of step 5, ack processing.
1277 * Otherwise, goto step 6.
1279 if (thflags & TH_ACK)
1285 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1286 * if segment contains a SYN and CC [not CC.NEW] option:
1287 * if state == TIME_WAIT and connection duration > MSL,
1288 * drop packet and send RST;
1290 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1291 * ack the FIN (and data) in retransmission queue.
1292 * Complete close and delete TCPCB. Then reprocess
1293 * segment, hoping to find new TCPCB in LISTEN state;
1295 * else must be old SYN; drop it.
1296 * else do normal processing.
1300 case TCPS_TIME_WAIT:
1301 if ((thflags & TH_SYN) &&
1302 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1303 if (tp->t_state == TCPS_TIME_WAIT &&
1304 (ticks - tp->t_starttime) > tcp_msl) {
1305 rstreason = BANDLIM_UNLIMITED;
1308 if (CC_GT(to.to_cc, tp->cc_recv)) {
1315 break; /* continue normal processing */
1319 * States other than LISTEN or SYN_SENT.
1320 * First check the RST flag and sequence number since reset segments
1321 * are exempt from the timestamp and connection count tests. This
1322 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1323 * below which allowed reset segments in half the sequence space
1324 * to fall though and be processed (which gives forged reset
1325 * segments with a random sequence number a 50 percent chance of
1326 * killing a connection).
1327 * Then check timestamp, if present.
1328 * Then check the connection count, if present.
1329 * Then check that at least some bytes of segment are within
1330 * receive window. If segment begins before rcv_nxt,
1331 * drop leading data (and SYN); if nothing left, just ack.
1334 * If the RST bit is set, check the sequence number to see
1335 * if this is a valid reset segment.
1337 * In all states except SYN-SENT, all reset (RST) segments
1338 * are validated by checking their SEQ-fields. A reset is
1339 * valid if its sequence number is in the window.
1340 * Note: this does not take into account delayed ACKs, so
1341 * we should test against last_ack_sent instead of rcv_nxt.
1342 * The sequence number in the reset segment is normally an
1343 * echo of our outgoing acknowlegement numbers, but some hosts
1344 * send a reset with the sequence number at the rightmost edge
1345 * of our receive window, and we have to handle this case.
1346 * If we have multiple segments in flight, the intial reset
1347 * segment sequence numbers will be to the left of last_ack_sent,
1348 * but they will eventually catch up.
1349 * In any case, it never made sense to trim reset segments to
1350 * fit the receive window since RFC 1122 says:
1351 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1353 * A TCP SHOULD allow a received RST segment to include data.
1356 * It has been suggested that a RST segment could contain
1357 * ASCII text that encoded and explained the cause of the
1358 * RST. No standard has yet been established for such
1361 * If the reset segment passes the sequence number test examine
1363 * SYN_RECEIVED STATE:
1364 * If passive open, return to LISTEN state.
1365 * If active open, inform user that connection was refused.
1366 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1367 * Inform user that connection was reset, and close tcb.
1368 * CLOSING, LAST_ACK STATES:
1371 * Drop the segment - see Stevens, vol. 2, p. 964 and
1374 if (thflags & TH_RST) {
1375 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1376 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1377 switch (tp->t_state) {
1379 case TCPS_SYN_RECEIVED:
1380 so->so_error = ECONNREFUSED;
1383 case TCPS_ESTABLISHED:
1384 case TCPS_FIN_WAIT_1:
1385 case TCPS_FIN_WAIT_2:
1386 case TCPS_CLOSE_WAIT:
1387 so->so_error = ECONNRESET;
1389 tp->t_state = TCPS_CLOSED;
1390 tcpstat.tcps_drops++;
1399 case TCPS_TIME_WAIT:
1407 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1408 * and it's less than ts_recent, drop it.
1410 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1411 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1413 /* Check to see if ts_recent is over 24 days old. */
1414 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1416 * Invalidate ts_recent. If this segment updates
1417 * ts_recent, the age will be reset later and ts_recent
1418 * will get a valid value. If it does not, setting
1419 * ts_recent to zero will at least satisfy the
1420 * requirement that zero be placed in the timestamp
1421 * echo reply when ts_recent isn't valid. The
1422 * age isn't reset until we get a valid ts_recent
1423 * because we don't want out-of-order segments to be
1424 * dropped when ts_recent is old.
1428 tcpstat.tcps_rcvduppack++;
1429 tcpstat.tcps_rcvdupbyte += tlen;
1430 tcpstat.tcps_pawsdrop++;
1439 * If T/TCP was negotiated and the segment doesn't have CC,
1440 * or if its CC is wrong then drop the segment.
1441 * RST segments do not have to comply with this.
1443 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1444 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1448 * In the SYN-RECEIVED state, validate that the packet belongs to
1449 * this connection before trimming the data to fit the receive
1450 * window. Check the sequence number versus IRS since we know
1451 * the sequence numbers haven't wrapped. This is a partial fix
1452 * for the "LAND" DoS attack.
1454 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1455 rstreason = BANDLIM_RST_OPENPORT;
1459 todrop = tp->rcv_nxt - th->th_seq;
1461 if (thflags & TH_SYN) {
1471 * Following if statement from Stevens, vol. 2, p. 960.
1474 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1476 * Any valid FIN must be to the left of the window.
1477 * At this point the FIN must be a duplicate or out
1478 * of sequence; drop it.
1483 * Send an ACK to resynchronize and drop any data.
1484 * But keep on processing for RST or ACK.
1486 tp->t_flags |= TF_ACKNOW;
1488 tcpstat.tcps_rcvduppack++;
1489 tcpstat.tcps_rcvdupbyte += todrop;
1491 tcpstat.tcps_rcvpartduppack++;
1492 tcpstat.tcps_rcvpartdupbyte += todrop;
1494 drop_hdrlen += todrop; /* drop from the top afterwards */
1495 th->th_seq += todrop;
1497 if (th->th_urp > todrop)
1498 th->th_urp -= todrop;
1506 * If new data are received on a connection after the
1507 * user processes are gone, then RST the other end.
1509 if ((so->so_state & SS_NOFDREF) &&
1510 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1512 tcpstat.tcps_rcvafterclose++;
1513 rstreason = BANDLIM_UNLIMITED;
1518 * If segment ends after window, drop trailing data
1519 * (and PUSH and FIN); if nothing left, just ACK.
1521 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1523 tcpstat.tcps_rcvpackafterwin++;
1524 if (todrop >= tlen) {
1525 tcpstat.tcps_rcvbyteafterwin += tlen;
1527 * If a new connection request is received
1528 * while in TIME_WAIT, drop the old connection
1529 * and start over if the sequence numbers
1530 * are above the previous ones.
1532 if (thflags & TH_SYN &&
1533 tp->t_state == TCPS_TIME_WAIT &&
1534 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1539 * If window is closed can only take segments at
1540 * window edge, and have to drop data and PUSH from
1541 * incoming segments. Continue processing, but
1542 * remember to ack. Otherwise, drop segment
1545 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1546 tp->t_flags |= TF_ACKNOW;
1547 tcpstat.tcps_rcvwinprobe++;
1551 tcpstat.tcps_rcvbyteafterwin += todrop;
1554 thflags &= ~(TH_PUSH|TH_FIN);
1558 * If last ACK falls within this segment's sequence numbers,
1559 * record its timestamp.
1560 * NOTE that the test is modified according to the latest
1561 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1563 if ((to.to_flags & TOF_TS) != 0 &&
1564 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1565 tp->ts_recent_age = ticks;
1566 tp->ts_recent = to.to_tsval;
1570 * If a SYN is in the window, then this is an
1571 * error and we send an RST and drop the connection.
1573 if (thflags & TH_SYN) {
1574 tp = tcp_drop(tp, ECONNRESET);
1575 rstreason = BANDLIM_UNLIMITED;
1580 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1581 * flag is on (half-synchronized state), then queue data for
1582 * later processing; else drop segment and return.
1584 if ((thflags & TH_ACK) == 0) {
1585 if (tp->t_state == TCPS_SYN_RECEIVED ||
1586 (tp->t_flags & TF_NEEDSYN))
1595 switch (tp->t_state) {
1598 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1599 * ESTABLISHED state and continue processing.
1600 * The ACK was checked above.
1602 case TCPS_SYN_RECEIVED:
1604 tcpstat.tcps_connects++;
1606 /* Do window scaling? */
1607 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1608 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1609 tp->snd_scale = tp->requested_s_scale;
1610 tp->rcv_scale = tp->request_r_scale;
1613 * Upon successful completion of 3-way handshake,
1614 * update cache.CC if it was undefined, pass any queued
1615 * data to the user, and advance state appropriately.
1617 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1619 taop->tao_cc = tp->cc_recv;
1623 * SYN-RECEIVED -> ESTABLISHED
1624 * SYN-RECEIVED* -> FIN-WAIT-1
1626 tp->t_starttime = ticks;
1627 if (tp->t_flags & TF_NEEDFIN) {
1628 tp->t_state = TCPS_FIN_WAIT_1;
1629 tp->t_flags &= ~TF_NEEDFIN;
1631 tp->t_state = TCPS_ESTABLISHED;
1632 callout_reset(tp->tt_keep, tcp_keepidle,
1633 tcp_timer_keep, tp);
1636 * If segment contains data or ACK, will call tcp_reass()
1637 * later; if not, do so now to pass queued data to user.
1639 if (tlen == 0 && (thflags & TH_FIN) == 0)
1640 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1642 tp->snd_wl1 = th->th_seq - 1;
1646 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1647 * ACKs. If the ack is in the range
1648 * tp->snd_una < th->th_ack <= tp->snd_max
1649 * then advance tp->snd_una to th->th_ack and drop
1650 * data from the retransmission queue. If this ACK reflects
1651 * more up to date window information we update our window information.
1653 case TCPS_ESTABLISHED:
1654 case TCPS_FIN_WAIT_1:
1655 case TCPS_FIN_WAIT_2:
1656 case TCPS_CLOSE_WAIT:
1659 case TCPS_TIME_WAIT:
1661 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1662 if (tlen == 0 && tiwin == tp->snd_wnd) {
1663 tcpstat.tcps_rcvdupack++;
1665 * If we have outstanding data (other than
1666 * a window probe), this is a completely
1667 * duplicate ack (ie, window info didn't
1668 * change), the ack is the biggest we've
1669 * seen and we've seen exactly our rexmt
1670 * threshhold of them, assume a packet
1671 * has been dropped and retransmit it.
1672 * Kludge snd_nxt & the congestion
1673 * window so we send only this one
1676 * We know we're losing at the current
1677 * window size so do congestion avoidance
1678 * (set ssthresh to half the current window
1679 * and pull our congestion window back to
1680 * the new ssthresh).
1682 * Dup acks mean that packets have left the
1683 * network (they're now cached at the receiver)
1684 * so bump cwnd by the amount in the receiver
1685 * to keep a constant cwnd packets in the
1688 if (!callout_active(tp->tt_rexmt) ||
1689 th->th_ack != tp->snd_una)
1691 else if (++tp->t_dupacks > tcprexmtthresh ||
1693 IN_FASTRECOVERY(tp))) {
1694 tp->snd_cwnd += tp->t_maxseg;
1695 (void) tcp_output(tp);
1697 } else if (tp->t_dupacks == tcprexmtthresh) {
1698 tcp_seq onxt = tp->snd_nxt;
1700 if (tcp_do_newreno &&
1706 if (tcp_do_eifel_detect &&
1707 (tp->t_flags & TF_RCVD_TSTMP)) {
1708 tcp_save_congestion_state(tp);
1709 tp->t_flags |= TF_FASTREXMT;
1711 win = min(tp->snd_wnd, tp->snd_cwnd) /
1715 tp->snd_ssthresh = win * tp->t_maxseg;
1716 ENTER_FASTRECOVERY(tp);
1717 tp->snd_recover = tp->snd_max;
1718 callout_stop(tp->tt_rexmt);
1720 tp->snd_nxt = th->th_ack;
1721 tp->snd_cwnd = tp->t_maxseg;
1722 (void) tcp_output(tp);
1723 KASSERT(tp->snd_limited <= 2,
1724 ("tp->snd_limited too big"));
1725 tp->snd_cwnd = tp->snd_ssthresh +
1727 (tp->t_dupacks - tp->snd_limited));
1728 if (SEQ_GT(onxt, tp->snd_nxt))
1731 } else if (tcp_do_limitedtransmit) {
1732 u_long oldcwnd = tp->snd_cwnd;
1733 tcp_seq oldsndmax = tp->snd_max;
1735 KASSERT(tp->t_dupacks == 1 ||
1737 ("dupacks not 1 or 2"));
1738 if (tp->t_dupacks == 1) {
1739 tp->snd_limited = 0;
1740 tp->snd_cwnd += tp->t_maxseg;
1745 (void) tcp_output(tp);
1746 sent = tp->snd_max - oldsndmax;
1747 if (sent > tp->t_maxseg) {
1748 KASSERT(tp->snd_limited == 0 &&
1751 tp->snd_limited = 2;
1752 } else if (sent > 0)
1754 tp->snd_cwnd = oldcwnd;
1762 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1765 * If the congestion window was inflated to account
1766 * for the other side's cached packets, retract it.
1768 if (tcp_do_newreno) {
1769 if (IN_FASTRECOVERY(tp)) {
1770 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1771 tcp_newreno_partial_ack(tp, th);
1774 * Window inflation should have left us
1775 * with approximately snd_ssthresh
1777 * But in case we would be inclined to
1778 * send a burst, better to do it via
1779 * the slow start mechanism.
1781 if (SEQ_GT(th->th_ack +
1784 tp->snd_cwnd = tp->snd_max -
1788 tp->snd_cwnd = tp->snd_ssthresh;
1792 if (tp->t_dupacks >= tcprexmtthresh &&
1793 tp->snd_cwnd > tp->snd_ssthresh)
1794 tp->snd_cwnd = tp->snd_ssthresh;
1797 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1798 tcpstat.tcps_rcvacktoomuch++;
1802 * If we reach this point, ACK is not a duplicate,
1803 * i.e., it ACKs something we sent.
1805 if (tp->t_flags & TF_NEEDSYN) {
1807 * T/TCP: Connection was half-synchronized, and our
1808 * SYN has been ACK'd (so connection is now fully
1809 * synchronized). Go to non-starred state,
1810 * increment snd_una for ACK of SYN, and check if
1811 * we can do window scaling.
1813 tp->t_flags &= ~TF_NEEDSYN;
1815 /* Do window scaling? */
1816 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1817 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1818 tp->snd_scale = tp->requested_s_scale;
1819 tp->rcv_scale = tp->request_r_scale;
1824 acked = th->th_ack - tp->snd_una;
1825 tcpstat.tcps_rcvackpack++;
1826 tcpstat.tcps_rcvackbyte += acked;
1829 * If we just performed our first retransmit, and the ACK
1830 * arrives within our recovery window, then it was a mistake
1831 * to do the retransmit in the first place. Recover our
1832 * original cwnd and ssthresh, and proceed to transmit where
1835 if (tcp_do_eifel_detect && acked &&
1836 (to.to_flags & TOF_TS) && to.to_tsecr &&
1837 (tp->t_flags & TF_FIRSTACCACK)) {
1838 /* Eifel detection applicable. */
1839 if (to.to_tsecr < tp->t_rexmtTS) {
1840 tcp_revert_congestion_state(tp);
1841 ++tcpstat.tcps_eifeldetected;
1843 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1844 tcp_revert_congestion_state(tp);
1845 ++tcpstat.tcps_rttdetected;
1849 * If we have a timestamp reply, update smoothed
1850 * round trip time. If no timestamp is present but
1851 * transmit timer is running and timed sequence
1852 * number was acked, update smoothed round trip time.
1853 * Since we now have an rtt measurement, cancel the
1854 * timer backoff (cf., Phil Karn's retransmit alg.).
1855 * Recompute the initial retransmit timer.
1857 * Some machines (certain windows boxes) send broken
1858 * timestamp replies during the SYN+ACK phase, ignore
1861 if ((to.to_flags & TOF_TS) != 0 &&
1863 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1864 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1865 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1867 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1870 * If all outstanding data is acked, stop retransmit
1871 * timer and remember to restart (more output or persist).
1872 * If there is more data to be acked, restart retransmit
1873 * timer, using current (possibly backed-off) value.
1875 if (th->th_ack == tp->snd_max) {
1876 callout_stop(tp->tt_rexmt);
1878 } else if (!callout_active(tp->tt_persist))
1879 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1880 tcp_timer_rexmt, tp);
1883 * If no data (only SYN) was ACK'd,
1884 * skip rest of ACK processing.
1889 /* Stop looking for an acceptable ACK since one was received. */
1890 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
1893 * When new data is acked, open the congestion window.
1894 * If the window gives us less than ssthresh packets
1895 * in flight, open exponentially (maxseg per packet).
1896 * Otherwise open linearly: maxseg per window
1897 * (maxseg^2 / cwnd per packet).
1899 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1900 u_int cw = tp->snd_cwnd;
1901 u_int incr = tp->t_maxseg;
1902 if (cw > tp->snd_ssthresh)
1903 incr = incr * incr / cw;
1904 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1906 if (acked > so->so_snd.sb_cc) {
1907 tp->snd_wnd -= so->so_snd.sb_cc;
1908 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1911 sbdrop(&so->so_snd, acked);
1912 tp->snd_wnd -= acked;
1916 /* detect una wraparound */
1917 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1918 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1919 SEQ_LEQ(th->th_ack, tp->snd_recover))
1920 tp->snd_recover = th->th_ack - 1;
1921 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1922 SEQ_GEQ(th->th_ack, tp->snd_recover))
1923 EXIT_FASTRECOVERY(tp);
1924 tp->snd_una = th->th_ack;
1925 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1926 tp->snd_nxt = tp->snd_una;
1928 switch (tp->t_state) {
1931 * In FIN_WAIT_1 STATE in addition to the processing
1932 * for the ESTABLISHED state if our FIN is now acknowledged
1933 * then enter FIN_WAIT_2.
1935 case TCPS_FIN_WAIT_1:
1936 if (ourfinisacked) {
1938 * If we can't receive any more
1939 * data, then closing user can proceed.
1940 * Starting the timer is contrary to the
1941 * specification, but if we don't get a FIN
1942 * we'll hang forever.
1944 if (so->so_state & SS_CANTRCVMORE) {
1945 soisdisconnected(so);
1946 callout_reset(tp->tt_2msl, tcp_maxidle,
1947 tcp_timer_2msl, tp);
1949 tp->t_state = TCPS_FIN_WAIT_2;
1954 * In CLOSING STATE in addition to the processing for
1955 * the ESTABLISHED state if the ACK acknowledges our FIN
1956 * then enter the TIME-WAIT state, otherwise ignore
1960 if (ourfinisacked) {
1961 tp->t_state = TCPS_TIME_WAIT;
1962 tcp_canceltimers(tp);
1963 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1964 if (tp->cc_recv != 0 &&
1965 (ticks - tp->t_starttime) < tcp_msl)
1966 callout_reset(tp->tt_2msl,
1969 tcp_timer_2msl, tp);
1971 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1972 tcp_timer_2msl, tp);
1973 soisdisconnected(so);
1978 * In LAST_ACK, we may still be waiting for data to drain
1979 * and/or to be acked, as well as for the ack of our FIN.
1980 * If our FIN is now acknowledged, delete the TCB,
1981 * enter the closed state and return.
1984 if (ourfinisacked) {
1991 * In TIME_WAIT state the only thing that should arrive
1992 * is a retransmission of the remote FIN. Acknowledge
1993 * it and restart the finack timer.
1995 case TCPS_TIME_WAIT:
1996 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1997 tcp_timer_2msl, tp);
2004 * Update window information.
2005 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2007 if ((thflags & TH_ACK) &&
2008 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2009 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2010 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2011 /* keep track of pure window updates */
2013 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2014 tcpstat.tcps_rcvwinupd++;
2015 tp->snd_wnd = tiwin;
2016 tp->snd_wl1 = th->th_seq;
2017 tp->snd_wl2 = th->th_ack;
2018 if (tp->snd_wnd > tp->max_sndwnd)
2019 tp->max_sndwnd = tp->snd_wnd;
2024 * Process segments with URG.
2026 if ((thflags & TH_URG) && th->th_urp &&
2027 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2029 * This is a kludge, but if we receive and accept
2030 * random urgent pointers, we'll crash in
2031 * soreceive. It's hard to imagine someone
2032 * actually wanting to send this much urgent data.
2034 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2035 th->th_urp = 0; /* XXX */
2036 thflags &= ~TH_URG; /* XXX */
2037 goto dodata; /* XXX */
2040 * If this segment advances the known urgent pointer,
2041 * then mark the data stream. This should not happen
2042 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2043 * a FIN has been received from the remote side.
2044 * In these states we ignore the URG.
2046 * According to RFC961 (Assigned Protocols),
2047 * the urgent pointer points to the last octet
2048 * of urgent data. We continue, however,
2049 * to consider it to indicate the first octet
2050 * of data past the urgent section as the original
2051 * spec states (in one of two places).
2053 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2054 tp->rcv_up = th->th_seq + th->th_urp;
2055 so->so_oobmark = so->so_rcv.sb_cc +
2056 (tp->rcv_up - tp->rcv_nxt) - 1;
2057 if (so->so_oobmark == 0)
2058 so->so_state |= SS_RCVATMARK;
2060 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2063 * Remove out of band data so doesn't get presented to user.
2064 * This can happen independent of advancing the URG pointer,
2065 * but if two URG's are pending at once, some out-of-band
2066 * data may creep in... ick.
2068 if (th->th_urp <= (u_long)tlen
2070 && (so->so_options & SO_OOBINLINE) == 0
2073 tcp_pulloutofband(so, th, m,
2074 drop_hdrlen); /* hdr drop is delayed */
2077 * If no out of band data is expected,
2078 * pull receive urgent pointer along
2079 * with the receive window.
2081 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2082 tp->rcv_up = tp->rcv_nxt;
2087 * Process the segment text, merging it into the TCP sequencing queue,
2088 * and arranging for acknowledgment of receipt if necessary.
2089 * This process logically involves adjusting tp->rcv_wnd as data
2090 * is presented to the user (this happens in tcp_usrreq.c,
2091 * case PRU_RCVD). If a FIN has already been received on this
2092 * connection then we just ignore the text.
2094 if ((tlen || (thflags & TH_FIN)) &&
2095 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2096 m_adj(m, drop_hdrlen); /* delayed header drop */
2098 * Insert segment which includes th into TCP reassembly queue
2099 * with control block tp. Set thflags to whether reassembly now
2100 * includes a segment with FIN. This handles the common case
2101 * inline (segment is the next to be received on an established
2102 * connection, and the queue is empty), avoiding linkage into
2103 * and removal from the queue and repetition of various
2105 * Set DELACK for segments received in order, but ack
2106 * immediately when segments are out of order (so
2107 * fast retransmit can work).
2109 if (th->th_seq == tp->rcv_nxt &&
2110 LIST_EMPTY(&tp->t_segq) &&
2111 TCPS_HAVEESTABLISHED(tp->t_state)) {
2113 callout_reset(tp->tt_delack, tcp_delacktime,
2114 tcp_timer_delack, tp);
2116 tp->t_flags |= TF_ACKNOW;
2117 tp->rcv_nxt += tlen;
2118 thflags = th->th_flags & TH_FIN;
2119 tcpstat.tcps_rcvpack++;
2120 tcpstat.tcps_rcvbyte += tlen;
2122 if (so->so_state & SS_CANTRCVMORE)
2125 sbappend(&so->so_rcv, m);
2128 thflags = tcp_reass(tp, th, &tlen, m);
2129 tp->t_flags |= TF_ACKNOW;
2133 * Note the amount of data that peer has sent into
2134 * our window, in order to estimate the sender's
2137 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2144 * If FIN is received ACK the FIN and let the user know
2145 * that the connection is closing.
2147 if (thflags & TH_FIN) {
2148 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2151 * If connection is half-synchronized
2152 * (ie NEEDSYN flag on) then delay ACK,
2153 * so it may be piggybacked when SYN is sent.
2154 * Otherwise, since we received a FIN then no
2155 * more input can be expected, send ACK now.
2157 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2158 callout_reset(tp->tt_delack, tcp_delacktime,
2159 tcp_timer_delack, tp);
2161 tp->t_flags |= TF_ACKNOW;
2164 switch (tp->t_state) {
2167 * In SYN_RECEIVED and ESTABLISHED STATES
2168 * enter the CLOSE_WAIT state.
2170 case TCPS_SYN_RECEIVED:
2171 tp->t_starttime = ticks;
2173 case TCPS_ESTABLISHED:
2174 tp->t_state = TCPS_CLOSE_WAIT;
2178 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2179 * enter the CLOSING state.
2181 case TCPS_FIN_WAIT_1:
2182 tp->t_state = TCPS_CLOSING;
2186 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2187 * starting the time-wait timer, turning off the other
2190 case TCPS_FIN_WAIT_2:
2191 tp->t_state = TCPS_TIME_WAIT;
2192 tcp_canceltimers(tp);
2193 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2194 if (tp->cc_recv != 0 &&
2195 (ticks - tp->t_starttime) < tcp_msl) {
2196 callout_reset(tp->tt_2msl,
2197 tp->t_rxtcur * TCPTV_TWTRUNC,
2198 tcp_timer_2msl, tp);
2199 /* For transaction client, force ACK now. */
2200 tp->t_flags |= TF_ACKNOW;
2203 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2204 tcp_timer_2msl, tp);
2205 soisdisconnected(so);
2209 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2211 case TCPS_TIME_WAIT:
2212 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2213 tcp_timer_2msl, tp);
2218 if (so->so_options & SO_DEBUG)
2219 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2224 * Return any desired output.
2226 if (needoutput || (tp->t_flags & TF_ACKNOW))
2227 (void) tcp_output(tp);
2232 * Generate an ACK dropping incoming segment if it occupies
2233 * sequence space, where the ACK reflects our state.
2235 * We can now skip the test for the RST flag since all
2236 * paths to this code happen after packets containing
2237 * RST have been dropped.
2239 * In the SYN-RECEIVED state, don't send an ACK unless the
2240 * segment we received passes the SYN-RECEIVED ACK test.
2241 * If it fails send a RST. This breaks the loop in the
2242 * "LAND" DoS attack, and also prevents an ACK storm
2243 * between two listening ports that have been sent forged
2244 * SYN segments, each with the source address of the other.
2246 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2247 (SEQ_GT(tp->snd_una, th->th_ack) ||
2248 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2249 rstreason = BANDLIM_RST_OPENPORT;
2253 if (so->so_options & SO_DEBUG)
2254 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2258 tp->t_flags |= TF_ACKNOW;
2259 (void) tcp_output(tp);
2264 * Generate a RST, dropping incoming segment.
2265 * Make ACK acceptable to originator of segment.
2266 * Don't bother to respond if destination was broadcast/multicast.
2268 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2271 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2272 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2275 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2276 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2277 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2278 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2281 /* IPv6 anycast check is done at tcp6_input() */
2284 * Perform bandwidth limiting.
2287 if (badport_bandlim(rstreason) < 0)
2292 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2293 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2296 if (thflags & TH_ACK)
2297 /* mtod() below is safe as long as hdr dropping is delayed */
2298 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2301 if (thflags & TH_SYN)
2303 /* mtod() below is safe as long as hdr dropping is delayed */
2304 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2305 (tcp_seq)0, TH_RST|TH_ACK);
2311 * Drop space held by incoming segment and return.
2314 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2315 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2323 * Parse TCP options and place in tcpopt.
2326 tcp_dooptions(to, cp, cnt, is_syn)
2334 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2336 if (opt == TCPOPT_EOL)
2338 if (opt == TCPOPT_NOP)
2344 if (optlen < 2 || optlen > cnt)
2349 if (optlen != TCPOLEN_MAXSEG)
2353 to->to_flags |= TOF_MSS;
2354 bcopy((char *)cp + 2,
2355 (char *)&to->to_mss, sizeof(to->to_mss));
2356 to->to_mss = ntohs(to->to_mss);
2359 if (optlen != TCPOLEN_WINDOW)
2363 to->to_flags |= TOF_SCALE;
2364 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2366 case TCPOPT_TIMESTAMP:
2367 if (optlen != TCPOLEN_TIMESTAMP)
2369 to->to_flags |= TOF_TS;
2370 bcopy((char *)cp + 2,
2371 (char *)&to->to_tsval, sizeof(to->to_tsval));
2372 to->to_tsval = ntohl(to->to_tsval);
2373 bcopy((char *)cp + 6,
2374 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2375 to->to_tsecr = ntohl(to->to_tsecr);
2378 if (optlen != TCPOLEN_CC)
2380 to->to_flags |= TOF_CC;
2381 bcopy((char *)cp + 2,
2382 (char *)&to->to_cc, sizeof(to->to_cc));
2383 to->to_cc = ntohl(to->to_cc);
2386 if (optlen != TCPOLEN_CC)
2390 to->to_flags |= TOF_CCNEW;
2391 bcopy((char *)cp + 2,
2392 (char *)&to->to_cc, sizeof(to->to_cc));
2393 to->to_cc = ntohl(to->to_cc);
2396 if (optlen != TCPOLEN_CC)
2400 to->to_flags |= TOF_CCECHO;
2401 bcopy((char *)cp + 2,
2402 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2403 to->to_ccecho = ntohl(to->to_ccecho);
2412 * Pull out of band byte out of a segment so
2413 * it doesn't appear in the user's data queue.
2414 * It is still reflected in the segment length for
2415 * sequencing purposes.
2418 tcp_pulloutofband(so, th, m, off)
2422 int off; /* delayed to be droped hdrlen */
2424 int cnt = off + th->th_urp - 1;
2427 if (m->m_len > cnt) {
2428 char *cp = mtod(m, caddr_t) + cnt;
2429 struct tcpcb *tp = sototcpcb(so);
2432 tp->t_oobflags |= TCPOOB_HAVEDATA;
2433 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2435 if (m->m_flags & M_PKTHDR)
2444 panic("tcp_pulloutofband");
2448 * Collect new round-trip time estimate
2449 * and update averages and current timeout.
2452 tcp_xmit_timer(tp, rtt)
2458 tcpstat.tcps_rttupdated++;
2460 if (tp->t_srtt != 0) {
2462 * srtt is stored as fixed point with 5 bits after the
2463 * binary point (i.e., scaled by 8). The following magic
2464 * is equivalent to the smoothing algorithm in rfc793 with
2465 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2466 * point). Adjust rtt to origin 0.
2468 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2469 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2471 if ((tp->t_srtt += delta) <= 0)
2475 * We accumulate a smoothed rtt variance (actually, a
2476 * smoothed mean difference), then set the retransmit
2477 * timer to smoothed rtt + 4 times the smoothed variance.
2478 * rttvar is stored as fixed point with 4 bits after the
2479 * binary point (scaled by 16). The following is
2480 * equivalent to rfc793 smoothing with an alpha of .75
2481 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2482 * rfc793's wired-in beta.
2486 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2487 if ((tp->t_rttvar += delta) <= 0)
2489 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2490 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2493 * No rtt measurement yet - use the unsmoothed rtt.
2494 * Set the variance to half the rtt (so our first
2495 * retransmit happens at 3*rtt).
2497 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2498 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2499 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2505 * the retransmit should happen at rtt + 4 * rttvar.
2506 * Because of the way we do the smoothing, srtt and rttvar
2507 * will each average +1/2 tick of bias. When we compute
2508 * the retransmit timer, we want 1/2 tick of rounding and
2509 * 1 extra tick because of +-1/2 tick uncertainty in the
2510 * firing of the timer. The bias will give us exactly the
2511 * 1.5 tick we need. But, because the bias is
2512 * statistical, we have to test that we don't drop below
2513 * the minimum feasible timer (which is 2 ticks).
2515 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2516 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2519 * We received an ack for a packet that wasn't retransmitted;
2520 * it is probably safe to discard any error indications we've
2521 * received recently. This isn't quite right, but close enough
2522 * for now (a route might have failed after we sent a segment,
2523 * and the return path might not be symmetrical).
2525 tp->t_softerror = 0;
2529 * Determine a reasonable value for maxseg size.
2530 * If the route is known, check route for mtu.
2531 * If none, use an mss that can be handled on the outgoing
2532 * interface without forcing IP to fragment; if bigger than
2533 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2534 * to utilize large mbufs. If no route is found, route has no mtu,
2535 * or the destination isn't local, use a default, hopefully conservative
2536 * size (usually 512 or the default IP max size, but no more than the mtu
2537 * of the interface), as we can't discover anything about intervening
2538 * gateways or networks. We also initialize the congestion/slow start
2539 * window to be a single segment if the destination isn't local.
2540 * While looking at the routing entry, we also initialize other path-dependent
2541 * parameters from pre-set or cached values in the routing entry.
2543 * Also take into account the space needed for options that we
2544 * send regularly. Make maxseg shorter by that amount to assure
2545 * that we can send maxseg amount of data even when the options
2546 * are present. Store the upper limit of the length of options plus
2549 * NOTE that this routine is only called when we process an incoming
2550 * segment, for outgoing segments only tcp_mssopt is called.
2552 * In case of T/TCP, we call this routine during implicit connection
2553 * setup as well (offer = -1), to initialize maxseg from the cached
2565 struct inpcb *inp = tp->t_inpcb;
2567 struct rmxp_tao *taop;
2568 int origoffer = offer;
2570 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2571 size_t min_protoh = isipv6 ?
2572 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2573 sizeof(struct tcpiphdr);
2575 const int isipv6 = 0;
2576 const size_t min_protoh = sizeof(struct tcpiphdr);
2580 rt = tcp_rtlookup6(&inp->inp_inc);
2582 rt = tcp_rtlookup(&inp->inp_inc);
2584 tp->t_maxopd = tp->t_maxseg =
2585 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2589 so = inp->inp_socket;
2591 taop = rmx_taop(rt->rt_rmx);
2593 * Offer == -1 means that we didn't receive SYN yet,
2594 * use cached value in that case;
2597 offer = taop->tao_mssopt;
2599 * Offer == 0 means that there was no MSS on the SYN segment,
2600 * in this case we use tcp_mssdflt.
2603 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2606 * Sanity check: make sure that maxopd will be large
2607 * enough to allow some data on segments even is the
2608 * all the option space is used (40bytes). Otherwise
2609 * funny things may happen in tcp_output.
2611 offer = max(offer, 64);
2612 taop->tao_mssopt = offer;
2615 * While we're here, check if there's an initial rtt
2616 * or rttvar. Convert from the route-table units
2617 * to scaled multiples of the slow timeout timer.
2619 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2621 * XXX the lock bit for RTT indicates that the value
2622 * is also a minimum value; this is subject to time.
2624 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2625 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2626 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2627 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2628 tcpstat.tcps_usedrtt++;
2629 if (rt->rt_rmx.rmx_rttvar) {
2630 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2631 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2632 tcpstat.tcps_usedrttvar++;
2634 /* default variation is +- 1 rtt */
2636 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2638 TCPT_RANGESET(tp->t_rxtcur,
2639 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2640 tp->t_rttmin, TCPTV_REXMTMAX);
2643 * if there's an mtu associated with the route, use it
2644 * else, use the link mtu.
2646 if (rt->rt_rmx.rmx_mtu)
2647 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2650 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2652 if (!in6_localaddr(&inp->in6p_faddr))
2653 mss = min(mss, tcp_v6mssdflt);
2655 mss = ifp->if_mtu - min_protoh;
2656 if (!in_localaddr(inp->inp_faddr))
2657 mss = min(mss, tcp_mssdflt);
2660 mss = min(mss, offer);
2662 * maxopd stores the maximum length of data AND options
2663 * in a segment; maxseg is the amount of data in a normal
2664 * segment. We need to store this value (maxopd) apart
2665 * from maxseg, because now every segment carries options
2666 * and thus we normally have somewhat less data in segments.
2671 * In case of T/TCP, origoffer==-1 indicates, that no segments
2672 * were received yet. In this case we just guess, otherwise
2673 * we do the same as before T/TCP.
2675 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2677 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2678 mss -= TCPOLEN_TSTAMP_APPA;
2679 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2681 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2682 mss -= TCPOLEN_CC_APPA;
2684 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2686 mss &= ~(MCLBYTES-1);
2689 mss = mss / MCLBYTES * MCLBYTES;
2692 * If there's a pipesize, change the socket buffer
2693 * to that size. Make the socket buffers an integral
2694 * number of mss units; if the mss is larger than
2695 * the socket buffer, decrease the mss.
2698 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2700 bufsize = so->so_snd.sb_hiwat;
2704 bufsize = roundup(bufsize, mss);
2705 if (bufsize > sb_max)
2707 if (bufsize > so->so_snd.sb_hiwat)
2708 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2713 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2715 bufsize = so->so_rcv.sb_hiwat;
2716 if (bufsize > mss) {
2717 bufsize = roundup(bufsize, mss);
2718 if (bufsize > sb_max)
2720 if (bufsize > so->so_rcv.sb_hiwat)
2721 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2725 * Set the slow-start flight size depending on whether this
2726 * is a local network or not.
2729 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2730 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2731 (!isipv6 && in_localaddr(inp->inp_faddr)))
2732 tp->snd_cwnd = mss * ss_fltsz_local;
2734 tp->snd_cwnd = mss * ss_fltsz;
2736 if (rt->rt_rmx.rmx_ssthresh) {
2738 * There's some sort of gateway or interface
2739 * buffer limit on the path. Use this to set
2740 * the slow start threshhold, but set the
2741 * threshold to no less than 2*mss.
2743 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2744 tcpstat.tcps_usedssthresh++;
2749 * Determine the MSS option to send on an outgoing SYN.
2757 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2758 int min_protoh = isipv6 ?
2759 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2760 sizeof(struct tcpiphdr);
2762 const int isipv6 = 0;
2763 const size_t min_protoh = sizeof(struct tcpiphdr);
2767 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2769 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2771 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2773 return (rt->rt_ifp->if_mtu - min_protoh);
2778 * When a partial ack arrives, force the retransmission of the
2779 * next unacknowledged segment. Do not clear tp->t_dupacks.
2780 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2784 tcp_newreno_partial_ack(tp, th)
2788 tcp_seq onxt = tp->snd_nxt;
2789 u_long ocwnd = tp->snd_cwnd;
2791 callout_stop(tp->tt_rexmt);
2793 tp->snd_nxt = th->th_ack;
2795 * Set snd_cwnd to one segment beyond acknowledged offset
2796 * (tp->snd_una has not yet been updated when this function is called.)
2798 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2799 tp->t_flags |= TF_ACKNOW;
2800 (void) tcp_output(tp);
2801 tp->snd_cwnd = ocwnd;
2802 if (SEQ_GT(onxt, tp->snd_nxt))
2805 * Partial window deflation. Relies on fact that tp->snd_una
2808 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);