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 1.13 2004/02/14 21:12:39 dillon 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>
56 #include <sys/in_cksum.h>
58 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
61 #include <net/route.h>
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
67 #include <netinet/in_var.h>
68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/in6_pcb.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
82 #include <netinet/tcpip.h>
84 #include <netinet/tcp_debug.h>
86 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
87 struct tcphdr tcp_savetcp;
91 #include <netipsec/ipsec.h>
92 #include <netipsec/ipsec6.h>
96 #include <netinet6/ipsec.h>
97 #include <netinet6/ipsec6.h>
98 #include <netproto/key/key.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 static int tcp_do_rfc3390 = 1;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
136 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
138 static int tcp_do_eifel_detect = 1;
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
140 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
142 struct inpcbhead tcb;
143 #define tcb6 tcb /* for KAME src sync over BSD*'s */
144 struct inpcbinfo tcbinfo;
146 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
147 static void tcp_pulloutofband(struct socket *,
148 struct tcphdr *, struct mbuf *, int);
149 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
151 static void tcp_xmit_timer(struct tcpcb *, int);
152 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
154 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
156 #define ND6_HINT(tp) \
158 if ((tp) && (tp)->t_inpcb && \
159 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
160 (tp)->t_inpcb->in6p_route.ro_rt) \
161 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
168 * Indicate whether this ack should be delayed. We can delay the ack if
169 * - delayed acks are enabled and
170 * - there is no delayed ack timer in progress and
171 * - our last ack wasn't a 0-sized window. We never want to delay
172 * the ack that opens up a 0-sized window.
174 #define DELAY_ACK(tp) \
175 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
176 (tp->t_flags & TF_RXWIN0SENT) == 0)
179 tcp_reass(tp, th, tlenp, m)
186 struct tseg_qent *p = NULL;
187 struct tseg_qent *nq;
188 struct tseg_qent *te;
189 struct socket *so = tp->t_inpcb->inp_socket;
193 * Call with th==0 after become established to
194 * force pre-ESTABLISHED data up to user socket.
199 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
200 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
203 tcpstat.tcps_rcvmemdrop++;
209 * Find a segment which begins after this one does.
211 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
212 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
218 * If there is a preceding segment, it may provide some of
219 * our data already. If so, drop the data from the incoming
220 * segment. If it provides all of our data, drop us.
224 /* conversion to int (in i) handles seq wraparound */
225 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
228 tcpstat.tcps_rcvduppack++;
229 tcpstat.tcps_rcvdupbyte += *tlenp;
233 * Try to present any queued data
234 * at the left window edge to the user.
235 * This is needed after the 3-WHS
238 goto present; /* ??? */
245 tcpstat.tcps_rcvoopack++;
246 tcpstat.tcps_rcvoobyte += *tlenp;
249 * While we overlap succeeding segments trim them or,
250 * if they are completely covered, dequeue them.
253 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
256 if (i < q->tqe_len) {
257 q->tqe_th->th_seq += i;
263 nq = LIST_NEXT(q, tqe_q);
264 LIST_REMOVE(q, tqe_q);
270 /* Insert the new segment queue entry into place. */
273 te->tqe_len = *tlenp;
276 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
278 LIST_INSERT_AFTER(p, te, tqe_q);
283 * Present data to user, advancing rcv_nxt through
284 * completed sequence space.
286 if (!TCPS_HAVEESTABLISHED(tp->t_state))
288 q = LIST_FIRST(&tp->t_segq);
289 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
292 tp->rcv_nxt += q->tqe_len;
293 flags = q->tqe_th->th_flags & TH_FIN;
294 nq = LIST_NEXT(q, tqe_q);
295 LIST_REMOVE(q, tqe_q);
296 if (so->so_state & SS_CANTRCVMORE)
299 sbappend(&so->so_rcv, q->tqe_m);
302 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
309 * TCP input routine, follows pages 65-76 of the
310 * protocol specification dated September, 1981 very closely.
314 tcp6_input(mp, offp, proto)
318 struct mbuf *m = *mp;
319 struct in6_ifaddr *ia6;
321 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
324 * draft-itojun-ipv6-tcp-to-anycast
325 * better place to put this in?
327 ia6 = ip6_getdstifaddr(m);
328 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
331 ip6 = mtod(m, struct ip6_hdr *);
332 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
333 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
337 tcp_input(m, *offp, proto);
343 tcp_input(m, off0, proto)
348 struct ip *ip = NULL;
350 struct inpcb *inp = NULL;
355 struct tcpcb *tp = NULL;
357 struct socket *so = 0;
358 int todrop, acked, ourfinisacked, needoutput = 0;
360 struct tcpopt to; /* options in this segment */
361 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
362 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
363 struct sockaddr_in *next_hop = NULL;
364 int rstreason; /* For badport_bandlim accounting purposes */
365 struct ip6_hdr *ip6 = NULL;
369 const int isipv6 = 0;
375 /* Grab info from MT_TAG mbufs prepended to the chain. */
376 for (;m && m->m_type == MT_TAG; m = m->m_next) {
377 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
378 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
381 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
383 bzero((char *)&to, sizeof(to));
385 tcpstat.tcps_rcvtotal++;
388 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
389 ip6 = mtod(m, struct ip6_hdr *);
390 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
391 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
392 tcpstat.tcps_rcvbadsum++;
395 th = (struct tcphdr *)((caddr_t)ip6 + off0);
398 * Be proactive about unspecified IPv6 address in source.
399 * As we use all-zero to indicate unbounded/unconnected pcb,
400 * unspecified IPv6 address can be used to confuse us.
402 * Note that packets with unspecified IPv6 destination is
403 * already dropped in ip6_input.
405 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
411 * Get IP and TCP header together in first mbuf.
412 * Note: IP leaves IP header in first mbuf.
414 if (off0 > sizeof(struct ip)) {
415 ip_stripoptions(m, (struct mbuf *)0);
416 off0 = sizeof(struct ip);
418 if (m->m_len < sizeof(struct tcpiphdr)) {
419 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
420 tcpstat.tcps_rcvshort++;
424 ip = mtod(m, struct ip *);
425 ipov = (struct ipovly *)ip;
426 th = (struct tcphdr *)((caddr_t)ip + off0);
429 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
430 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
431 th->th_sum = m->m_pkthdr.csum_data;
433 th->th_sum = in_pseudo(ip->ip_src.s_addr,
435 htonl(m->m_pkthdr.csum_data +
438 th->th_sum ^= 0xffff;
441 * Checksum extended TCP header and data.
443 len = sizeof(struct ip) + tlen;
444 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
445 ipov->ih_len = (u_short)tlen;
446 ipov->ih_len = htons(ipov->ih_len);
447 th->th_sum = in_cksum(m, len);
450 tcpstat.tcps_rcvbadsum++;
454 /* Re-initialization for later version check */
455 ip->ip_v = IPVERSION;
460 * Check that TCP offset makes sense,
461 * pull out TCP options and adjust length. XXX
463 off = th->th_off << 2;
464 if (off < sizeof(struct tcphdr) || off > tlen) {
465 tcpstat.tcps_rcvbadoff++;
468 tlen -= off; /* tlen is used instead of ti->ti_len */
469 if (off > sizeof(struct tcphdr)) {
471 IP6_EXTHDR_CHECK(m, off0, off, );
472 ip6 = mtod(m, struct ip6_hdr *);
473 th = (struct tcphdr *)((caddr_t)ip6 + off0);
475 if (m->m_len < sizeof(struct ip) + off) {
476 if ((m = m_pullup(m, sizeof(struct ip) + off))
478 tcpstat.tcps_rcvshort++;
481 ip = mtod(m, struct ip *);
482 ipov = (struct ipovly *)ip;
483 th = (struct tcphdr *)((caddr_t)ip + off0);
486 optlen = off - sizeof(struct tcphdr);
487 optp = (u_char *)(th + 1);
489 thflags = th->th_flags;
491 #ifdef TCP_DROP_SYNFIN
493 * If the drop_synfin option is enabled, drop all packets with
494 * both the SYN and FIN bits set. This prevents e.g. nmap from
495 * identifying the TCP/IP stack.
497 * This is a violation of the TCP specification.
499 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
504 * Convert TCP protocol specific fields to host format.
506 th->th_seq = ntohl(th->th_seq);
507 th->th_ack = ntohl(th->th_ack);
508 th->th_win = ntohs(th->th_win);
509 th->th_urp = ntohs(th->th_urp);
512 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
513 * until after ip6_savecontrol() is called and before other functions
514 * which don't want those proto headers.
515 * Because ip6_savecontrol() is going to parse the mbuf to
516 * search for data to be passed up to user-land, it wants mbuf
517 * parameters to be unchanged.
518 * XXX: the call of ip6_savecontrol() has been obsoleted based on
519 * latest version of the advanced API (20020110).
521 drop_hdrlen = off0 + off;
524 * Locate pcb for segment.
527 /* IPFIREWALL_FORWARD section */
528 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
530 * Transparently forwarded. Pretend to be the destination.
531 * already got one like this?
533 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
534 ip->ip_dst, th->th_dport,
535 0, m->m_pkthdr.rcvif);
537 /* It's new. Try find the ambushing socket. */
538 inp = in_pcblookup_hash(&tcbinfo,
539 ip->ip_src, th->th_sport,
542 ntohs(next_hop->sin_port) :
544 1, m->m_pkthdr.rcvif);
548 inp = in6_pcblookup_hash(&tcbinfo,
549 &ip6->ip6_src, th->th_sport,
550 &ip6->ip6_dst, th->th_dport,
551 1, m->m_pkthdr.rcvif);
553 inp = in_pcblookup_hash(&tcbinfo,
554 ip->ip_src, th->th_sport,
555 ip->ip_dst, th->th_dport,
556 1, m->m_pkthdr.rcvif);
561 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
562 ipsec6stat.in_polvio++;
566 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
567 ipsecstat.in_polvio++;
574 if (inp != NULL && ipsec6_in_reject(m, inp)) {
578 if (inp != NULL && ipsec4_in_reject(m, inp)) {
585 * If the state is CLOSED (i.e., TCB does not exist) then
586 * all data in the incoming segment is discarded.
587 * If the TCB exists but is in CLOSED state, it is embryonic,
588 * but should either do a listen or a connect soon.
593 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
595 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
600 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
601 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
605 strcpy(dbuf, inet_ntoa(ip->ip_dst));
606 strcpy(sbuf, inet_ntoa(ip->ip_src));
608 switch (log_in_vain) {
610 if ((thflags & TH_SYN) == 0)
614 "Connection attempt to TCP %s:%d "
615 "from %s:%d flags:0x%02x\n",
616 dbuf, ntohs(th->th_dport), sbuf,
617 ntohs(th->th_sport), thflags);
626 if (thflags & TH_SYN)
635 rstreason = BANDLIM_RST_CLOSEDPORT;
640 rstreason = BANDLIM_RST_CLOSEDPORT;
643 if (tp->t_state == TCPS_CLOSED)
646 /* Unscale the window into a 32-bit value. */
647 if ((thflags & TH_SYN) == 0)
648 tiwin = th->th_win << tp->snd_scale;
652 so = inp->inp_socket;
655 if (so->so_options & SO_DEBUG) {
656 ostate = tp->t_state;
658 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
660 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
665 if (so->so_options & SO_ACCEPTCONN) {
666 struct in_conninfo inc;
669 inc.inc_isipv6 = isipv6;
672 inc.inc6_faddr = ip6->ip6_src;
673 inc.inc6_laddr = ip6->ip6_dst;
674 inc.inc6_route.ro_rt = NULL; /* XXX */
676 inc.inc_faddr = ip->ip_src;
677 inc.inc_laddr = ip->ip_dst;
678 inc.inc_route.ro_rt = NULL; /* XXX */
680 inc.inc_fport = th->th_sport;
681 inc.inc_lport = th->th_dport;
684 * If the state is LISTEN then ignore segment if it contains
685 * a RST. If the segment contains an ACK then it is bad and
686 * send a RST. If it does not contain a SYN then it is not
687 * interesting; drop it.
689 * If the state is SYN_RECEIVED (syncache) and seg contains
690 * an ACK, but not for our SYN/ACK, send a RST. If the seg
691 * contains a RST, check the sequence number to see if it
692 * is a valid reset segment.
694 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
695 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
696 if (!syncache_expand(&inc, th, &so, m)) {
698 * No syncache entry, or ACK was not
699 * for our SYN/ACK. Send a RST.
701 tcpstat.tcps_badsyn++;
702 rstreason = BANDLIM_RST_OPENPORT;
707 * Could not complete 3-way handshake,
708 * connection is being closed down, and
709 * syncache will free mbuf.
713 * Socket is created in state SYN_RECEIVED.
714 * Continue processing segment.
719 * This is what would have happened in
720 * tcp_output() when the SYN,ACK was sent.
722 tp->snd_up = tp->snd_una;
723 tp->snd_max = tp->snd_nxt = tp->iss + 1;
724 tp->last_ack_sent = tp->rcv_nxt;
726 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
727 * until the _second_ ACK is received:
728 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
729 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
730 * move to ESTAB, set snd_wnd to tiwin.
732 tp->snd_wnd = tiwin; /* unscaled */
735 if (thflags & TH_RST) {
736 syncache_chkrst(&inc, th);
739 if (thflags & TH_ACK) {
740 syncache_badack(&inc);
741 tcpstat.tcps_badsyn++;
742 rstreason = BANDLIM_RST_OPENPORT;
749 * Segment's flags are (SYN) or (SYN|FIN).
753 * If deprecated address is forbidden,
754 * we do not accept SYN to deprecated interface
755 * address to prevent any new inbound connection from
756 * getting established.
757 * When we do not accept SYN, we send a TCP RST,
758 * with deprecated source address (instead of dropping
759 * it). We compromise it as it is much better for peer
760 * to send a RST, and RST will be the final packet
763 * If we do not forbid deprecated addresses, we accept
764 * the SYN packet. RFC2462 does not suggest dropping
766 * If we decipher RFC2462 5.5.4, it says like this:
767 * 1. use of deprecated addr with existing
768 * communication is okay - "SHOULD continue to be
770 * 2. use of it with new communication:
771 * (2a) "SHOULD NOT be used if alternate address
772 * with sufficient scope is available"
773 * (2b) nothing mentioned otherwise.
774 * Here we fall into (2b) case as we have no choice in
775 * our source address selection - we must obey the peer.
777 * The wording in RFC2462 is confusing, and there are
778 * multiple description text for deprecated address
779 * handling - worse, they are not exactly the same.
780 * I believe 5.5.4 is the best one, so we follow 5.5.4.
782 if (isipv6 && !ip6_use_deprecated) {
783 struct in6_ifaddr *ia6;
785 if ((ia6 = ip6_getdstifaddr(m)) &&
786 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
788 rstreason = BANDLIM_RST_OPENPORT;
794 * If it is from this socket, drop it, it must be forged.
795 * Don't bother responding if the destination was a broadcast.
797 if (th->th_dport == th->th_sport) {
799 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
803 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
808 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
810 * Note that it is quite possible to receive unicast
811 * link-layer packets with a broadcast IP address. Use
812 * in_broadcast() to find them.
814 if (m->m_flags & (M_BCAST|M_MCAST))
817 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
818 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
821 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
822 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
823 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
824 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
828 * SYN appears to be valid; create compressed TCP state
829 * for syncache, or perform t/tcp connection.
831 if (so->so_qlen <= so->so_qlimit) {
832 tcp_dooptions(&to, optp, optlen, 1);
833 if (!syncache_add(&inc, &to, th, &so, m))
837 * Entry added to syncache, mbuf used to
838 * send SYN,ACK packet.
842 * Segment passed TAO tests.
847 tp->t_starttime = ticks;
848 tp->t_state = TCPS_ESTABLISHED;
851 * If there is a FIN, or if there is data and the
852 * connection is local, then delay SYN,ACK(SYN) in
853 * the hope of piggy-backing it on a response
854 * segment. Otherwise must send ACK now in case
855 * the other side is slow starting.
858 ((thflags & TH_FIN) ||
860 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
861 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
862 callout_reset(tp->tt_delack, tcp_delacktime,
863 tcp_timer_delack, tp);
864 tp->t_flags |= TF_NEEDSYN;
866 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
868 tcpstat.tcps_connects++;
876 /* XXX temp debugging */
877 /* should not happen - syncache should pick up these connections */
878 if (tp->t_state == TCPS_LISTEN)
879 panic("tcp_input: TCPS_LISTEN");
882 * Segment received on connection.
883 * Reset idle time and keep-alive timer.
885 tp->t_rcvtime = ticks;
886 if (TCPS_HAVEESTABLISHED(tp->t_state))
887 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
891 * XXX this is tradtitional behavior, may need to be cleaned up.
893 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
894 if (thflags & TH_SYN) {
895 if (to.to_flags & TOF_SCALE) {
896 tp->t_flags |= TF_RCVD_SCALE;
897 tp->requested_s_scale = to.to_requested_s_scale;
899 if (to.to_flags & TOF_TS) {
900 tp->t_flags |= TF_RCVD_TSTMP;
901 tp->ts_recent = to.to_tsval;
902 tp->ts_recent_age = ticks;
904 if (to.to_flags & (TOF_CC|TOF_CCNEW))
905 tp->t_flags |= TF_RCVD_CC;
906 if (to.to_flags & TOF_MSS)
907 tcp_mss(tp, to.to_mss);
911 * Header prediction: check for the two common cases
912 * of a uni-directional data xfer. If the packet has
913 * no control flags, is in-sequence, the window didn't
914 * change and we're not retransmitting, it's a
915 * candidate. If the length is zero and the ack moved
916 * forward, we're the sender side of the xfer. Just
917 * free the data acked & wake any higher level process
918 * that was blocked waiting for space. If the length
919 * is non-zero and the ack didn't move, we're the
920 * receiver side. If we're getting packets in-order
921 * (the reassembly queue is empty), add the data to
922 * the socket buffer and note that we need a delayed ack.
923 * Make sure that the hidden state-flags are also off.
924 * Since we check for TCPS_ESTABLISHED above, it can only
927 if (tp->t_state == TCPS_ESTABLISHED &&
928 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
929 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
930 ((to.to_flags & TOF_TS) == 0 ||
931 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
933 * Using the CC option is compulsory if once started:
934 * the segment is OK if no T/TCP was negotiated or
935 * if the segment has a CC option equal to CCrecv
937 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
938 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
939 th->th_seq == tp->rcv_nxt &&
940 tiwin && tiwin == tp->snd_wnd &&
941 tp->snd_nxt == tp->snd_max) {
944 * If last ACK falls within this segment's sequence numbers,
945 * record the timestamp.
946 * NOTE that the test is modified according to the latest
947 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
949 if ((to.to_flags & TOF_TS) != 0 &&
950 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
951 tp->ts_recent_age = ticks;
952 tp->ts_recent = to.to_tsval;
956 if (SEQ_GT(th->th_ack, tp->snd_una) &&
957 SEQ_LEQ(th->th_ack, tp->snd_max) &&
958 tp->snd_cwnd >= tp->snd_wnd &&
960 tp->t_dupacks < tcprexmtthresh) ||
961 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
963 * this is a pure ack for outstanding data.
965 ++tcpstat.tcps_predack;
967 * "bad retransmit" recovery
969 * If Eifel detection applies, then
970 * it is deterministic, so use it
971 * unconditionally over the old heuristic.
972 * Otherwise, fall back to the old heuristic.
974 if (tcp_do_eifel_detect &&
975 (to.to_flags & TOF_TS) && to.to_tsecr &&
976 (tp->t_flags & TF_FIRSTACCACK)) {
977 /* Eifel detection applicable. */
978 if (to.to_tsecr < tp->t_rexmtTS) {
979 tcp_revert_congestion_state(tp);
980 ++tcpstat.tcps_eifeldetected;
982 } else if (tp->t_rxtshift == 1 &&
983 ticks < tp->t_badrxtwin) {
984 tcp_revert_congestion_state(tp);
985 ++tcpstat.tcps_rttdetected;
987 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
989 * Recalculate the retransmit timer / rtt.
991 * Some machines (certain windows boxes)
992 * send broken timestamp replies during the
993 * SYN+ACK phase, ignore timestamps of 0.
995 if ((to.to_flags & TOF_TS) != 0 &&
998 ticks - to.to_tsecr + 1);
999 } else if (tp->t_rtttime &&
1000 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1002 ticks - tp->t_rtttime);
1004 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1005 acked = th->th_ack - tp->snd_una;
1006 tcpstat.tcps_rcvackpack++;
1007 tcpstat.tcps_rcvackbyte += acked;
1008 sbdrop(&so->so_snd, acked);
1009 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1010 SEQ_LEQ(th->th_ack, tp->snd_recover))
1011 tp->snd_recover = th->th_ack - 1;
1012 tp->snd_una = th->th_ack;
1015 ND6_HINT(tp); /* some progress has been done */
1018 * If all outstanding data are acked, stop
1019 * retransmit timer, otherwise restart timer
1020 * using current (possibly backed-off) value.
1021 * If process is waiting for space,
1022 * wakeup/selwakeup/signal. If data
1023 * are ready to send, let tcp_output
1024 * decide between more output or persist.
1026 if (tp->snd_una == tp->snd_max)
1027 callout_stop(tp->tt_rexmt);
1028 else if (!callout_active(tp->tt_persist))
1029 callout_reset(tp->tt_rexmt,
1031 tcp_timer_rexmt, tp);
1034 if (so->so_snd.sb_cc)
1035 (void) tcp_output(tp);
1038 } else if (th->th_ack == tp->snd_una &&
1039 LIST_EMPTY(&tp->t_segq) &&
1040 tlen <= sbspace(&so->so_rcv)) {
1042 * this is a pure, in-sequence data packet
1043 * with nothing on the reassembly queue and
1044 * we have enough buffer space to take it.
1046 ++tcpstat.tcps_preddat;
1047 tp->rcv_nxt += tlen;
1048 tcpstat.tcps_rcvpack++;
1049 tcpstat.tcps_rcvbyte += tlen;
1050 ND6_HINT(tp); /* some progress has been done */
1052 * Add data to socket buffer.
1054 if (so->so_state & SS_CANTRCVMORE) {
1057 m_adj(m, drop_hdrlen); /* delayed header drop */
1058 sbappend(&so->so_rcv, m);
1061 if (DELAY_ACK(tp)) {
1062 callout_reset(tp->tt_delack, tcp_delacktime,
1063 tcp_timer_delack, tp);
1065 tp->t_flags |= TF_ACKNOW;
1073 * Calculate amount of space in receive window,
1074 * and then do TCP input processing.
1075 * Receive window is amount of space in rcv queue,
1076 * but not less than advertised window.
1080 win = sbspace(&so->so_rcv);
1083 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1086 switch (tp->t_state) {
1089 * If the state is SYN_RECEIVED:
1090 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1092 case TCPS_SYN_RECEIVED:
1093 if ((thflags & TH_ACK) &&
1094 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1095 SEQ_GT(th->th_ack, tp->snd_max))) {
1096 rstreason = BANDLIM_RST_OPENPORT;
1102 * If the state is SYN_SENT:
1103 * if seg contains an ACK, but not for our SYN, drop the input.
1104 * if seg contains a RST, then drop the connection.
1105 * if seg does not contain SYN, then drop it.
1106 * Otherwise this is an acceptable SYN segment
1107 * initialize tp->rcv_nxt and tp->irs
1108 * if seg contains ack then advance tp->snd_una
1109 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1110 * arrange for segment to be acked (eventually)
1111 * continue processing rest of data/controls, beginning with URG
1114 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1115 taop = &tao_noncached;
1116 bzero(taop, sizeof(*taop));
1119 if ((thflags & TH_ACK) &&
1120 (SEQ_LEQ(th->th_ack, tp->iss) ||
1121 SEQ_GT(th->th_ack, tp->snd_max))) {
1123 * If we have a cached CCsent for the remote host,
1124 * hence we haven't just crashed and restarted,
1125 * do not send a RST. This may be a retransmission
1126 * from the other side after our earlier ACK was lost.
1127 * Our new SYN, when it arrives, will serve as the
1130 if (taop->tao_ccsent != 0)
1133 rstreason = BANDLIM_UNLIMITED;
1137 if (thflags & TH_RST) {
1138 if (thflags & TH_ACK)
1139 tp = tcp_drop(tp, ECONNREFUSED);
1142 if ((thflags & TH_SYN) == 0)
1144 tp->snd_wnd = th->th_win; /* initial send window */
1145 tp->cc_recv = to.to_cc; /* foreign CC */
1147 tp->irs = th->th_seq;
1149 if (thflags & TH_ACK) {
1151 * Our SYN was acked. If segment contains CC.ECHO
1152 * option, check it to make sure this segment really
1153 * matches our SYN. If not, just drop it as old
1154 * duplicate, but send an RST if we're still playing
1155 * by the old rules. If no CC.ECHO option, make sure
1156 * we don't get fooled into using T/TCP.
1158 if (to.to_flags & TOF_CCECHO) {
1159 if (tp->cc_send != to.to_ccecho) {
1160 if (taop->tao_ccsent != 0)
1163 rstreason = BANDLIM_UNLIMITED;
1168 tp->t_flags &= ~TF_RCVD_CC;
1169 tcpstat.tcps_connects++;
1171 /* Do window scaling on this connection? */
1172 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1173 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1174 tp->snd_scale = tp->requested_s_scale;
1175 tp->rcv_scale = tp->request_r_scale;
1177 /* Segment is acceptable, update cache if undefined. */
1178 if (taop->tao_ccsent == 0)
1179 taop->tao_ccsent = to.to_ccecho;
1181 tp->rcv_adv += tp->rcv_wnd;
1182 tp->snd_una++; /* SYN is acked */
1184 * If there's data, delay ACK; if there's also a FIN
1185 * ACKNOW will be turned on later.
1187 if (DELAY_ACK(tp) && tlen != 0)
1188 callout_reset(tp->tt_delack, tcp_delacktime,
1189 tcp_timer_delack, tp);
1191 tp->t_flags |= TF_ACKNOW;
1193 * Received <SYN,ACK> in SYN_SENT[*] state.
1195 * SYN_SENT --> ESTABLISHED
1196 * SYN_SENT* --> FIN_WAIT_1
1198 tp->t_starttime = ticks;
1199 if (tp->t_flags & TF_NEEDFIN) {
1200 tp->t_state = TCPS_FIN_WAIT_1;
1201 tp->t_flags &= ~TF_NEEDFIN;
1204 tp->t_state = TCPS_ESTABLISHED;
1205 callout_reset(tp->tt_keep, tcp_keepidle,
1206 tcp_timer_keep, tp);
1210 * Received initial SYN in SYN-SENT[*] state =>
1211 * simultaneous open. If segment contains CC option
1212 * and there is a cached CC, apply TAO test.
1213 * If it succeeds, connection is * half-synchronized.
1214 * Otherwise, do 3-way handshake:
1215 * SYN-SENT -> SYN-RECEIVED
1216 * SYN-SENT* -> SYN-RECEIVED*
1217 * If there was no CC option, clear cached CC value.
1219 tp->t_flags |= TF_ACKNOW;
1220 callout_stop(tp->tt_rexmt);
1221 if (to.to_flags & TOF_CC) {
1222 if (taop->tao_cc != 0 &&
1223 CC_GT(to.to_cc, taop->tao_cc)) {
1225 * update cache and make transition:
1226 * SYN-SENT -> ESTABLISHED*
1227 * SYN-SENT* -> FIN-WAIT-1*
1229 taop->tao_cc = to.to_cc;
1230 tp->t_starttime = ticks;
1231 if (tp->t_flags & TF_NEEDFIN) {
1232 tp->t_state = TCPS_FIN_WAIT_1;
1233 tp->t_flags &= ~TF_NEEDFIN;
1235 tp->t_state = TCPS_ESTABLISHED;
1236 callout_reset(tp->tt_keep,
1241 tp->t_flags |= TF_NEEDSYN;
1243 tp->t_state = TCPS_SYN_RECEIVED;
1245 /* CC.NEW or no option => invalidate cache */
1247 tp->t_state = TCPS_SYN_RECEIVED;
1253 * Advance th->th_seq to correspond to first data byte.
1254 * If data, trim to stay within window,
1255 * dropping FIN if necessary.
1258 if (tlen > tp->rcv_wnd) {
1259 todrop = tlen - tp->rcv_wnd;
1263 tcpstat.tcps_rcvpackafterwin++;
1264 tcpstat.tcps_rcvbyteafterwin += todrop;
1266 tp->snd_wl1 = th->th_seq - 1;
1267 tp->rcv_up = th->th_seq;
1269 * Client side of transaction: already sent SYN and data.
1270 * If the remote host used T/TCP to validate the SYN,
1271 * our data will be ACK'd; if so, enter normal data segment
1272 * processing in the middle of step 5, ack processing.
1273 * Otherwise, goto step 6.
1275 if (thflags & TH_ACK)
1281 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1282 * if segment contains a SYN and CC [not CC.NEW] option:
1283 * if state == TIME_WAIT and connection duration > MSL,
1284 * drop packet and send RST;
1286 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1287 * ack the FIN (and data) in retransmission queue.
1288 * Complete close and delete TCPCB. Then reprocess
1289 * segment, hoping to find new TCPCB in LISTEN state;
1291 * else must be old SYN; drop it.
1292 * else do normal processing.
1296 case TCPS_TIME_WAIT:
1297 if ((thflags & TH_SYN) &&
1298 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1299 if (tp->t_state == TCPS_TIME_WAIT &&
1300 (ticks - tp->t_starttime) > tcp_msl) {
1301 rstreason = BANDLIM_UNLIMITED;
1304 if (CC_GT(to.to_cc, tp->cc_recv)) {
1311 break; /* continue normal processing */
1315 * States other than LISTEN or SYN_SENT.
1316 * First check the RST flag and sequence number since reset segments
1317 * are exempt from the timestamp and connection count tests. This
1318 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1319 * below which allowed reset segments in half the sequence space
1320 * to fall though and be processed (which gives forged reset
1321 * segments with a random sequence number a 50 percent chance of
1322 * killing a connection).
1323 * Then check timestamp, if present.
1324 * Then check the connection count, if present.
1325 * Then check that at least some bytes of segment are within
1326 * receive window. If segment begins before rcv_nxt,
1327 * drop leading data (and SYN); if nothing left, just ack.
1330 * If the RST bit is set, check the sequence number to see
1331 * if this is a valid reset segment.
1333 * In all states except SYN-SENT, all reset (RST) segments
1334 * are validated by checking their SEQ-fields. A reset is
1335 * valid if its sequence number is in the window.
1336 * Note: this does not take into account delayed ACKs, so
1337 * we should test against last_ack_sent instead of rcv_nxt.
1338 * The sequence number in the reset segment is normally an
1339 * echo of our outgoing acknowlegement numbers, but some hosts
1340 * send a reset with the sequence number at the rightmost edge
1341 * of our receive window, and we have to handle this case.
1342 * If we have multiple segments in flight, the intial reset
1343 * segment sequence numbers will be to the left of last_ack_sent,
1344 * but they will eventually catch up.
1345 * In any case, it never made sense to trim reset segments to
1346 * fit the receive window since RFC 1122 says:
1347 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1349 * A TCP SHOULD allow a received RST segment to include data.
1352 * It has been suggested that a RST segment could contain
1353 * ASCII text that encoded and explained the cause of the
1354 * RST. No standard has yet been established for such
1357 * If the reset segment passes the sequence number test examine
1359 * SYN_RECEIVED STATE:
1360 * If passive open, return to LISTEN state.
1361 * If active open, inform user that connection was refused.
1362 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1363 * Inform user that connection was reset, and close tcb.
1364 * CLOSING, LAST_ACK STATES:
1367 * Drop the segment - see Stevens, vol. 2, p. 964 and
1370 if (thflags & TH_RST) {
1371 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1372 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1373 switch (tp->t_state) {
1375 case TCPS_SYN_RECEIVED:
1376 so->so_error = ECONNREFUSED;
1379 case TCPS_ESTABLISHED:
1380 case TCPS_FIN_WAIT_1:
1381 case TCPS_FIN_WAIT_2:
1382 case TCPS_CLOSE_WAIT:
1383 so->so_error = ECONNRESET;
1385 tp->t_state = TCPS_CLOSED;
1386 tcpstat.tcps_drops++;
1395 case TCPS_TIME_WAIT:
1403 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1404 * and it's less than ts_recent, drop it.
1406 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1407 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1409 /* Check to see if ts_recent is over 24 days old. */
1410 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1412 * Invalidate ts_recent. If this segment updates
1413 * ts_recent, the age will be reset later and ts_recent
1414 * will get a valid value. If it does not, setting
1415 * ts_recent to zero will at least satisfy the
1416 * requirement that zero be placed in the timestamp
1417 * echo reply when ts_recent isn't valid. The
1418 * age isn't reset until we get a valid ts_recent
1419 * because we don't want out-of-order segments to be
1420 * dropped when ts_recent is old.
1424 tcpstat.tcps_rcvduppack++;
1425 tcpstat.tcps_rcvdupbyte += tlen;
1426 tcpstat.tcps_pawsdrop++;
1435 * If T/TCP was negotiated and the segment doesn't have CC,
1436 * or if its CC is wrong then drop the segment.
1437 * RST segments do not have to comply with this.
1439 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1440 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1444 * In the SYN-RECEIVED state, validate that the packet belongs to
1445 * this connection before trimming the data to fit the receive
1446 * window. Check the sequence number versus IRS since we know
1447 * the sequence numbers haven't wrapped. This is a partial fix
1448 * for the "LAND" DoS attack.
1450 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1451 rstreason = BANDLIM_RST_OPENPORT;
1455 todrop = tp->rcv_nxt - th->th_seq;
1457 if (thflags & TH_SYN) {
1467 * Following if statement from Stevens, vol. 2, p. 960.
1470 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1472 * Any valid FIN must be to the left of the window.
1473 * At this point the FIN must be a duplicate or out
1474 * of sequence; drop it.
1479 * Send an ACK to resynchronize and drop any data.
1480 * But keep on processing for RST or ACK.
1482 tp->t_flags |= TF_ACKNOW;
1484 tcpstat.tcps_rcvduppack++;
1485 tcpstat.tcps_rcvdupbyte += todrop;
1487 tcpstat.tcps_rcvpartduppack++;
1488 tcpstat.tcps_rcvpartdupbyte += todrop;
1490 drop_hdrlen += todrop; /* drop from the top afterwards */
1491 th->th_seq += todrop;
1493 if (th->th_urp > todrop)
1494 th->th_urp -= todrop;
1502 * If new data are received on a connection after the
1503 * user processes are gone, then RST the other end.
1505 if ((so->so_state & SS_NOFDREF) &&
1506 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1508 tcpstat.tcps_rcvafterclose++;
1509 rstreason = BANDLIM_UNLIMITED;
1514 * If segment ends after window, drop trailing data
1515 * (and PUSH and FIN); if nothing left, just ACK.
1517 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1519 tcpstat.tcps_rcvpackafterwin++;
1520 if (todrop >= tlen) {
1521 tcpstat.tcps_rcvbyteafterwin += tlen;
1523 * If a new connection request is received
1524 * while in TIME_WAIT, drop the old connection
1525 * and start over if the sequence numbers
1526 * are above the previous ones.
1528 if (thflags & TH_SYN &&
1529 tp->t_state == TCPS_TIME_WAIT &&
1530 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1535 * If window is closed can only take segments at
1536 * window edge, and have to drop data and PUSH from
1537 * incoming segments. Continue processing, but
1538 * remember to ack. Otherwise, drop segment
1541 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1542 tp->t_flags |= TF_ACKNOW;
1543 tcpstat.tcps_rcvwinprobe++;
1547 tcpstat.tcps_rcvbyteafterwin += todrop;
1550 thflags &= ~(TH_PUSH|TH_FIN);
1554 * If last ACK falls within this segment's sequence numbers,
1555 * record its timestamp.
1556 * NOTE that the test is modified according to the latest
1557 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1559 if ((to.to_flags & TOF_TS) != 0 &&
1560 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1561 tp->ts_recent_age = ticks;
1562 tp->ts_recent = to.to_tsval;
1566 * If a SYN is in the window, then this is an
1567 * error and we send an RST and drop the connection.
1569 if (thflags & TH_SYN) {
1570 tp = tcp_drop(tp, ECONNRESET);
1571 rstreason = BANDLIM_UNLIMITED;
1576 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1577 * flag is on (half-synchronized state), then queue data for
1578 * later processing; else drop segment and return.
1580 if ((thflags & TH_ACK) == 0) {
1581 if (tp->t_state == TCPS_SYN_RECEIVED ||
1582 (tp->t_flags & TF_NEEDSYN))
1591 switch (tp->t_state) {
1594 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1595 * ESTABLISHED state and continue processing.
1596 * The ACK was checked above.
1598 case TCPS_SYN_RECEIVED:
1600 tcpstat.tcps_connects++;
1602 /* Do window scaling? */
1603 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1604 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1605 tp->snd_scale = tp->requested_s_scale;
1606 tp->rcv_scale = tp->request_r_scale;
1609 * Upon successful completion of 3-way handshake,
1610 * update cache.CC if it was undefined, pass any queued
1611 * data to the user, and advance state appropriately.
1613 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1615 taop->tao_cc = tp->cc_recv;
1619 * SYN-RECEIVED -> ESTABLISHED
1620 * SYN-RECEIVED* -> FIN-WAIT-1
1622 tp->t_starttime = ticks;
1623 if (tp->t_flags & TF_NEEDFIN) {
1624 tp->t_state = TCPS_FIN_WAIT_1;
1625 tp->t_flags &= ~TF_NEEDFIN;
1627 tp->t_state = TCPS_ESTABLISHED;
1628 callout_reset(tp->tt_keep, tcp_keepidle,
1629 tcp_timer_keep, tp);
1632 * If segment contains data or ACK, will call tcp_reass()
1633 * later; if not, do so now to pass queued data to user.
1635 if (tlen == 0 && (thflags & TH_FIN) == 0)
1636 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1638 tp->snd_wl1 = th->th_seq - 1;
1642 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1643 * ACKs. If the ack is in the range
1644 * tp->snd_una < th->th_ack <= tp->snd_max
1645 * then advance tp->snd_una to th->th_ack and drop
1646 * data from the retransmission queue. If this ACK reflects
1647 * more up to date window information we update our window information.
1649 case TCPS_ESTABLISHED:
1650 case TCPS_FIN_WAIT_1:
1651 case TCPS_FIN_WAIT_2:
1652 case TCPS_CLOSE_WAIT:
1655 case TCPS_TIME_WAIT:
1657 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1658 if (tlen == 0 && tiwin == tp->snd_wnd) {
1659 tcpstat.tcps_rcvdupack++;
1661 * If we have outstanding data (other than
1662 * a window probe), this is a completely
1663 * duplicate ack (ie, window info didn't
1664 * change), the ack is the biggest we've
1665 * seen and we've seen exactly our rexmt
1666 * threshhold of them, assume a packet
1667 * has been dropped and retransmit it.
1668 * Kludge snd_nxt & the congestion
1669 * window so we send only this one
1672 * We know we're losing at the current
1673 * window size so do congestion avoidance
1674 * (set ssthresh to half the current window
1675 * and pull our congestion window back to
1676 * the new ssthresh).
1678 * Dup acks mean that packets have left the
1679 * network (they're now cached at the receiver)
1680 * so bump cwnd by the amount in the receiver
1681 * to keep a constant cwnd packets in the
1684 if (!callout_active(tp->tt_rexmt) ||
1685 th->th_ack != tp->snd_una)
1687 else if (++tp->t_dupacks > tcprexmtthresh ||
1689 IN_FASTRECOVERY(tp))) {
1690 tp->snd_cwnd += tp->t_maxseg;
1691 (void) tcp_output(tp);
1693 } else if (tp->t_dupacks == tcprexmtthresh) {
1694 tcp_seq onxt = tp->snd_nxt;
1696 if (tcp_do_newreno &&
1702 if (tcp_do_eifel_detect &&
1703 (tp->t_flags & TF_RCVD_TSTMP)) {
1704 tcp_save_congestion_state(tp);
1705 tp->t_flags |= TF_FASTREXMT;
1707 win = min(tp->snd_wnd, tp->snd_cwnd) /
1711 tp->snd_ssthresh = win * tp->t_maxseg;
1712 ENTER_FASTRECOVERY(tp);
1713 tp->snd_recover = tp->snd_max;
1714 callout_stop(tp->tt_rexmt);
1716 tp->snd_nxt = th->th_ack;
1717 tp->snd_cwnd = tp->t_maxseg;
1718 (void) tcp_output(tp);
1719 KASSERT(tp->snd_limited <= 2,
1720 ("tp->snd_limited too big"));
1721 tp->snd_cwnd = tp->snd_ssthresh +
1723 (tp->t_dupacks - tp->snd_limited));
1724 if (SEQ_GT(onxt, tp->snd_nxt))
1727 } else if (tcp_do_limitedtransmit) {
1728 u_long oldcwnd = tp->snd_cwnd;
1729 tcp_seq oldsndmax = tp->snd_max;
1731 KASSERT(tp->t_dupacks == 1 ||
1733 ("dupacks not 1 or 2"));
1734 if (tp->t_dupacks == 1)
1735 tp->snd_limited = 0;
1737 (tp->snd_nxt - tp->snd_una) +
1738 (tp->t_dupacks - tp->snd_limited) *
1740 (void) tcp_output(tp);
1741 sent = tp->snd_max - oldsndmax;
1742 if (sent > tp->t_maxseg) {
1743 KASSERT(tp->snd_limited == 0 &&
1746 tp->snd_limited = 2;
1747 } else if (sent > 0)
1749 tp->snd_cwnd = oldcwnd;
1757 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1760 * If the congestion window was inflated to account
1761 * for the other side's cached packets, retract it.
1763 if (tcp_do_newreno) {
1764 if (IN_FASTRECOVERY(tp)) {
1765 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1766 tcp_newreno_partial_ack(tp, th);
1769 * Window inflation should have left us
1770 * with approximately snd_ssthresh
1772 * But in case we would be inclined to
1773 * send a burst, better to do it via
1774 * the slow start mechanism.
1776 if (SEQ_GT(th->th_ack +
1779 tp->snd_cwnd = tp->snd_max -
1783 tp->snd_cwnd = tp->snd_ssthresh;
1787 if (tp->t_dupacks >= tcprexmtthresh &&
1788 tp->snd_cwnd > tp->snd_ssthresh)
1789 tp->snd_cwnd = tp->snd_ssthresh;
1792 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1793 tcpstat.tcps_rcvacktoomuch++;
1797 * If we reach this point, ACK is not a duplicate,
1798 * i.e., it ACKs something we sent.
1800 if (tp->t_flags & TF_NEEDSYN) {
1802 * T/TCP: Connection was half-synchronized, and our
1803 * SYN has been ACK'd (so connection is now fully
1804 * synchronized). Go to non-starred state,
1805 * increment snd_una for ACK of SYN, and check if
1806 * we can do window scaling.
1808 tp->t_flags &= ~TF_NEEDSYN;
1810 /* Do window scaling? */
1811 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1812 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1813 tp->snd_scale = tp->requested_s_scale;
1814 tp->rcv_scale = tp->request_r_scale;
1819 acked = th->th_ack - tp->snd_una;
1820 tcpstat.tcps_rcvackpack++;
1821 tcpstat.tcps_rcvackbyte += acked;
1824 * If we just performed our first retransmit, and the ACK
1825 * arrives within our recovery window, then it was a mistake
1826 * to do the retransmit in the first place. Recover our
1827 * original cwnd and ssthresh, and proceed to transmit where
1830 if (tcp_do_eifel_detect && acked &&
1831 (to.to_flags & TOF_TS) && to.to_tsecr &&
1832 (tp->t_flags & TF_FIRSTACCACK)) {
1833 /* Eifel detection applicable. */
1834 if (to.to_tsecr < tp->t_rexmtTS) {
1835 tcp_revert_congestion_state(tp);
1836 ++tcpstat.tcps_eifeldetected;
1838 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1839 tcp_revert_congestion_state(tp);
1840 ++tcpstat.tcps_rttdetected;
1844 * If we have a timestamp reply, update smoothed
1845 * round trip time. If no timestamp is present but
1846 * transmit timer is running and timed sequence
1847 * number was acked, update smoothed round trip time.
1848 * Since we now have an rtt measurement, cancel the
1849 * timer backoff (cf., Phil Karn's retransmit alg.).
1850 * Recompute the initial retransmit timer.
1852 * Some machines (certain windows boxes) send broken
1853 * timestamp replies during the SYN+ACK phase, ignore
1856 if ((to.to_flags & TOF_TS) != 0 &&
1858 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1859 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1860 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1862 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1865 * If all outstanding data is acked, stop retransmit
1866 * timer and remember to restart (more output or persist).
1867 * If there is more data to be acked, restart retransmit
1868 * timer, using current (possibly backed-off) value.
1870 if (th->th_ack == tp->snd_max) {
1871 callout_stop(tp->tt_rexmt);
1873 } else if (!callout_active(tp->tt_persist))
1874 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1875 tcp_timer_rexmt, tp);
1878 * If no data (only SYN) was ACK'd,
1879 * skip rest of ACK processing.
1884 /* Stop looking for an acceptable ACK since one was received. */
1885 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
1888 * When new data is acked, open the congestion window.
1889 * If the window gives us less than ssthresh packets
1890 * in flight, open exponentially (maxseg per packet).
1891 * Otherwise open linearly: maxseg per window
1892 * (maxseg^2 / cwnd per packet).
1894 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1895 u_int cw = tp->snd_cwnd;
1896 u_int incr = tp->t_maxseg;
1897 if (cw > tp->snd_ssthresh)
1898 incr = incr * incr / cw;
1899 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1901 if (acked > so->so_snd.sb_cc) {
1902 tp->snd_wnd -= so->so_snd.sb_cc;
1903 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1906 sbdrop(&so->so_snd, acked);
1907 tp->snd_wnd -= acked;
1911 /* detect una wraparound */
1912 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1913 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1914 SEQ_LEQ(th->th_ack, tp->snd_recover))
1915 tp->snd_recover = th->th_ack - 1;
1916 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1917 SEQ_GEQ(th->th_ack, tp->snd_recover))
1918 EXIT_FASTRECOVERY(tp);
1919 tp->snd_una = th->th_ack;
1920 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1921 tp->snd_nxt = tp->snd_una;
1923 switch (tp->t_state) {
1926 * In FIN_WAIT_1 STATE in addition to the processing
1927 * for the ESTABLISHED state if our FIN is now acknowledged
1928 * then enter FIN_WAIT_2.
1930 case TCPS_FIN_WAIT_1:
1931 if (ourfinisacked) {
1933 * If we can't receive any more
1934 * data, then closing user can proceed.
1935 * Starting the timer is contrary to the
1936 * specification, but if we don't get a FIN
1937 * we'll hang forever.
1939 if (so->so_state & SS_CANTRCVMORE) {
1940 soisdisconnected(so);
1941 callout_reset(tp->tt_2msl, tcp_maxidle,
1942 tcp_timer_2msl, tp);
1944 tp->t_state = TCPS_FIN_WAIT_2;
1949 * In CLOSING STATE in addition to the processing for
1950 * the ESTABLISHED state if the ACK acknowledges our FIN
1951 * then enter the TIME-WAIT state, otherwise ignore
1955 if (ourfinisacked) {
1956 tp->t_state = TCPS_TIME_WAIT;
1957 tcp_canceltimers(tp);
1958 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1959 if (tp->cc_recv != 0 &&
1960 (ticks - tp->t_starttime) < tcp_msl)
1961 callout_reset(tp->tt_2msl,
1964 tcp_timer_2msl, tp);
1966 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1967 tcp_timer_2msl, tp);
1968 soisdisconnected(so);
1973 * In LAST_ACK, we may still be waiting for data to drain
1974 * and/or to be acked, as well as for the ack of our FIN.
1975 * If our FIN is now acknowledged, delete the TCB,
1976 * enter the closed state and return.
1979 if (ourfinisacked) {
1986 * In TIME_WAIT state the only thing that should arrive
1987 * is a retransmission of the remote FIN. Acknowledge
1988 * it and restart the finack timer.
1990 case TCPS_TIME_WAIT:
1991 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1992 tcp_timer_2msl, tp);
1999 * Update window information.
2000 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2002 if ((thflags & TH_ACK) &&
2003 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2004 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2005 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2006 /* keep track of pure window updates */
2008 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2009 tcpstat.tcps_rcvwinupd++;
2010 tp->snd_wnd = tiwin;
2011 tp->snd_wl1 = th->th_seq;
2012 tp->snd_wl2 = th->th_ack;
2013 if (tp->snd_wnd > tp->max_sndwnd)
2014 tp->max_sndwnd = tp->snd_wnd;
2019 * Process segments with URG.
2021 if ((thflags & TH_URG) && th->th_urp &&
2022 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2024 * This is a kludge, but if we receive and accept
2025 * random urgent pointers, we'll crash in
2026 * soreceive. It's hard to imagine someone
2027 * actually wanting to send this much urgent data.
2029 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2030 th->th_urp = 0; /* XXX */
2031 thflags &= ~TH_URG; /* XXX */
2032 goto dodata; /* XXX */
2035 * If this segment advances the known urgent pointer,
2036 * then mark the data stream. This should not happen
2037 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2038 * a FIN has been received from the remote side.
2039 * In these states we ignore the URG.
2041 * According to RFC961 (Assigned Protocols),
2042 * the urgent pointer points to the last octet
2043 * of urgent data. We continue, however,
2044 * to consider it to indicate the first octet
2045 * of data past the urgent section as the original
2046 * spec states (in one of two places).
2048 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2049 tp->rcv_up = th->th_seq + th->th_urp;
2050 so->so_oobmark = so->so_rcv.sb_cc +
2051 (tp->rcv_up - tp->rcv_nxt) - 1;
2052 if (so->so_oobmark == 0)
2053 so->so_state |= SS_RCVATMARK;
2055 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2058 * Remove out of band data so doesn't get presented to user.
2059 * This can happen independent of advancing the URG pointer,
2060 * but if two URG's are pending at once, some out-of-band
2061 * data may creep in... ick.
2063 if (th->th_urp <= (u_long)tlen
2065 && (so->so_options & SO_OOBINLINE) == 0
2068 tcp_pulloutofband(so, th, m,
2069 drop_hdrlen); /* hdr drop is delayed */
2072 * If no out of band data is expected,
2073 * pull receive urgent pointer along
2074 * with the receive window.
2076 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2077 tp->rcv_up = tp->rcv_nxt;
2082 * Process the segment text, merging it into the TCP sequencing queue,
2083 * and arranging for acknowledgment of receipt if necessary.
2084 * This process logically involves adjusting tp->rcv_wnd as data
2085 * is presented to the user (this happens in tcp_usrreq.c,
2086 * case PRU_RCVD). If a FIN has already been received on this
2087 * connection then we just ignore the text.
2089 if ((tlen || (thflags & TH_FIN)) &&
2090 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2091 m_adj(m, drop_hdrlen); /* delayed header drop */
2093 * Insert segment which includes th into TCP reassembly queue
2094 * with control block tp. Set thflags to whether reassembly now
2095 * includes a segment with FIN. This handles the common case
2096 * inline (segment is the next to be received on an established
2097 * connection, and the queue is empty), avoiding linkage into
2098 * and removal from the queue and repetition of various
2100 * Set DELACK for segments received in order, but ack
2101 * immediately when segments are out of order (so
2102 * fast retransmit can work).
2104 if (th->th_seq == tp->rcv_nxt &&
2105 LIST_EMPTY(&tp->t_segq) &&
2106 TCPS_HAVEESTABLISHED(tp->t_state)) {
2108 callout_reset(tp->tt_delack, tcp_delacktime,
2109 tcp_timer_delack, tp);
2111 tp->t_flags |= TF_ACKNOW;
2112 tp->rcv_nxt += tlen;
2113 thflags = th->th_flags & TH_FIN;
2114 tcpstat.tcps_rcvpack++;
2115 tcpstat.tcps_rcvbyte += tlen;
2117 if (so->so_state & SS_CANTRCVMORE)
2120 sbappend(&so->so_rcv, m);
2123 thflags = tcp_reass(tp, th, &tlen, m);
2124 tp->t_flags |= TF_ACKNOW;
2128 * Note the amount of data that peer has sent into
2129 * our window, in order to estimate the sender's
2132 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2139 * If FIN is received ACK the FIN and let the user know
2140 * that the connection is closing.
2142 if (thflags & TH_FIN) {
2143 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2146 * If connection is half-synchronized
2147 * (ie NEEDSYN flag on) then delay ACK,
2148 * so it may be piggybacked when SYN is sent.
2149 * Otherwise, since we received a FIN then no
2150 * more input can be expected, send ACK now.
2152 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2153 callout_reset(tp->tt_delack, tcp_delacktime,
2154 tcp_timer_delack, tp);
2156 tp->t_flags |= TF_ACKNOW;
2159 switch (tp->t_state) {
2162 * In SYN_RECEIVED and ESTABLISHED STATES
2163 * enter the CLOSE_WAIT state.
2165 case TCPS_SYN_RECEIVED:
2166 tp->t_starttime = ticks;
2168 case TCPS_ESTABLISHED:
2169 tp->t_state = TCPS_CLOSE_WAIT;
2173 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2174 * enter the CLOSING state.
2176 case TCPS_FIN_WAIT_1:
2177 tp->t_state = TCPS_CLOSING;
2181 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2182 * starting the time-wait timer, turning off the other
2185 case TCPS_FIN_WAIT_2:
2186 tp->t_state = TCPS_TIME_WAIT;
2187 tcp_canceltimers(tp);
2188 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2189 if (tp->cc_recv != 0 &&
2190 (ticks - tp->t_starttime) < tcp_msl) {
2191 callout_reset(tp->tt_2msl,
2192 tp->t_rxtcur * TCPTV_TWTRUNC,
2193 tcp_timer_2msl, tp);
2194 /* For transaction client, force ACK now. */
2195 tp->t_flags |= TF_ACKNOW;
2198 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2199 tcp_timer_2msl, tp);
2200 soisdisconnected(so);
2204 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2206 case TCPS_TIME_WAIT:
2207 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2208 tcp_timer_2msl, tp);
2213 if (so->so_options & SO_DEBUG)
2214 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2219 * Return any desired output.
2221 if (needoutput || (tp->t_flags & TF_ACKNOW))
2222 (void) tcp_output(tp);
2227 * Generate an ACK dropping incoming segment if it occupies
2228 * sequence space, where the ACK reflects our state.
2230 * We can now skip the test for the RST flag since all
2231 * paths to this code happen after packets containing
2232 * RST have been dropped.
2234 * In the SYN-RECEIVED state, don't send an ACK unless the
2235 * segment we received passes the SYN-RECEIVED ACK test.
2236 * If it fails send a RST. This breaks the loop in the
2237 * "LAND" DoS attack, and also prevents an ACK storm
2238 * between two listening ports that have been sent forged
2239 * SYN segments, each with the source address of the other.
2241 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2242 (SEQ_GT(tp->snd_una, th->th_ack) ||
2243 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2244 rstreason = BANDLIM_RST_OPENPORT;
2248 if (so->so_options & SO_DEBUG)
2249 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2253 tp->t_flags |= TF_ACKNOW;
2254 (void) tcp_output(tp);
2259 * Generate a RST, dropping incoming segment.
2260 * Make ACK acceptable to originator of segment.
2261 * Don't bother to respond if destination was broadcast/multicast.
2263 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2266 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2267 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2270 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2271 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2272 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2273 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2276 /* IPv6 anycast check is done at tcp6_input() */
2279 * Perform bandwidth limiting.
2282 if (badport_bandlim(rstreason) < 0)
2287 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2288 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2291 if (thflags & TH_ACK)
2292 /* mtod() below is safe as long as hdr dropping is delayed */
2293 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2296 if (thflags & TH_SYN)
2298 /* mtod() below is safe as long as hdr dropping is delayed */
2299 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2300 (tcp_seq)0, TH_RST|TH_ACK);
2306 * Drop space held by incoming segment and return.
2309 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2310 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2318 * Parse TCP options and place in tcpopt.
2321 tcp_dooptions(to, cp, cnt, is_syn)
2329 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2331 if (opt == TCPOPT_EOL)
2333 if (opt == TCPOPT_NOP)
2339 if (optlen < 2 || optlen > cnt)
2344 if (optlen != TCPOLEN_MAXSEG)
2348 to->to_flags |= TOF_MSS;
2349 bcopy((char *)cp + 2,
2350 (char *)&to->to_mss, sizeof(to->to_mss));
2351 to->to_mss = ntohs(to->to_mss);
2354 if (optlen != TCPOLEN_WINDOW)
2358 to->to_flags |= TOF_SCALE;
2359 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2361 case TCPOPT_TIMESTAMP:
2362 if (optlen != TCPOLEN_TIMESTAMP)
2364 to->to_flags |= TOF_TS;
2365 bcopy((char *)cp + 2,
2366 (char *)&to->to_tsval, sizeof(to->to_tsval));
2367 to->to_tsval = ntohl(to->to_tsval);
2368 bcopy((char *)cp + 6,
2369 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2370 to->to_tsecr = ntohl(to->to_tsecr);
2373 if (optlen != TCPOLEN_CC)
2375 to->to_flags |= TOF_CC;
2376 bcopy((char *)cp + 2,
2377 (char *)&to->to_cc, sizeof(to->to_cc));
2378 to->to_cc = ntohl(to->to_cc);
2381 if (optlen != TCPOLEN_CC)
2385 to->to_flags |= TOF_CCNEW;
2386 bcopy((char *)cp + 2,
2387 (char *)&to->to_cc, sizeof(to->to_cc));
2388 to->to_cc = ntohl(to->to_cc);
2391 if (optlen != TCPOLEN_CC)
2395 to->to_flags |= TOF_CCECHO;
2396 bcopy((char *)cp + 2,
2397 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2398 to->to_ccecho = ntohl(to->to_ccecho);
2407 * Pull out of band byte out of a segment so
2408 * it doesn't appear in the user's data queue.
2409 * It is still reflected in the segment length for
2410 * sequencing purposes.
2413 tcp_pulloutofband(so, th, m, off)
2417 int off; /* delayed to be droped hdrlen */
2419 int cnt = off + th->th_urp - 1;
2422 if (m->m_len > cnt) {
2423 char *cp = mtod(m, caddr_t) + cnt;
2424 struct tcpcb *tp = sototcpcb(so);
2427 tp->t_oobflags |= TCPOOB_HAVEDATA;
2428 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2430 if (m->m_flags & M_PKTHDR)
2439 panic("tcp_pulloutofband");
2443 * Collect new round-trip time estimate
2444 * and update averages and current timeout.
2447 tcp_xmit_timer(tp, rtt)
2453 tcpstat.tcps_rttupdated++;
2455 if (tp->t_srtt != 0) {
2457 * srtt is stored as fixed point with 5 bits after the
2458 * binary point (i.e., scaled by 8). The following magic
2459 * is equivalent to the smoothing algorithm in rfc793 with
2460 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2461 * point). Adjust rtt to origin 0.
2463 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2464 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2466 if ((tp->t_srtt += delta) <= 0)
2470 * We accumulate a smoothed rtt variance (actually, a
2471 * smoothed mean difference), then set the retransmit
2472 * timer to smoothed rtt + 4 times the smoothed variance.
2473 * rttvar is stored as fixed point with 4 bits after the
2474 * binary point (scaled by 16). The following is
2475 * equivalent to rfc793 smoothing with an alpha of .75
2476 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2477 * rfc793's wired-in beta.
2481 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2482 if ((tp->t_rttvar += delta) <= 0)
2484 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2485 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2488 * No rtt measurement yet - use the unsmoothed rtt.
2489 * Set the variance to half the rtt (so our first
2490 * retransmit happens at 3*rtt).
2492 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2493 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2494 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2500 * the retransmit should happen at rtt + 4 * rttvar.
2501 * Because of the way we do the smoothing, srtt and rttvar
2502 * will each average +1/2 tick of bias. When we compute
2503 * the retransmit timer, we want 1/2 tick of rounding and
2504 * 1 extra tick because of +-1/2 tick uncertainty in the
2505 * firing of the timer. The bias will give us exactly the
2506 * 1.5 tick we need. But, because the bias is
2507 * statistical, we have to test that we don't drop below
2508 * the minimum feasible timer (which is 2 ticks).
2510 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2511 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2514 * We received an ack for a packet that wasn't retransmitted;
2515 * it is probably safe to discard any error indications we've
2516 * received recently. This isn't quite right, but close enough
2517 * for now (a route might have failed after we sent a segment,
2518 * and the return path might not be symmetrical).
2520 tp->t_softerror = 0;
2524 * Determine a reasonable value for maxseg size.
2525 * If the route is known, check route for mtu.
2526 * If none, use an mss that can be handled on the outgoing
2527 * interface without forcing IP to fragment; if bigger than
2528 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2529 * to utilize large mbufs. If no route is found, route has no mtu,
2530 * or the destination isn't local, use a default, hopefully conservative
2531 * size (usually 512 or the default IP max size, but no more than the mtu
2532 * of the interface), as we can't discover anything about intervening
2533 * gateways or networks. We also initialize the congestion/slow start
2534 * window to be a single segment if the destination isn't local.
2535 * While looking at the routing entry, we also initialize other path-dependent
2536 * parameters from pre-set or cached values in the routing entry.
2538 * Also take into account the space needed for options that we
2539 * send regularly. Make maxseg shorter by that amount to assure
2540 * that we can send maxseg amount of data even when the options
2541 * are present. Store the upper limit of the length of options plus
2544 * NOTE that this routine is only called when we process an incoming
2545 * segment, for outgoing segments only tcp_mssopt is called.
2547 * In case of T/TCP, we call this routine during implicit connection
2548 * setup as well (offer = -1), to initialize maxseg from the cached
2560 struct inpcb *inp = tp->t_inpcb;
2562 struct rmxp_tao *taop;
2563 int origoffer = offer;
2565 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2566 size_t min_protoh = isipv6 ?
2567 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2568 sizeof(struct tcpiphdr);
2570 const int isipv6 = 0;
2571 const size_t min_protoh = sizeof(struct tcpiphdr);
2575 rt = tcp_rtlookup6(&inp->inp_inc);
2577 rt = tcp_rtlookup(&inp->inp_inc);
2579 tp->t_maxopd = tp->t_maxseg =
2580 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2584 so = inp->inp_socket;
2586 taop = rmx_taop(rt->rt_rmx);
2588 * Offer == -1 means that we didn't receive SYN yet,
2589 * use cached value in that case;
2592 offer = taop->tao_mssopt;
2594 * Offer == 0 means that there was no MSS on the SYN segment,
2595 * in this case we use tcp_mssdflt.
2598 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2601 * Sanity check: make sure that maxopd will be large
2602 * enough to allow some data on segments even is the
2603 * all the option space is used (40bytes). Otherwise
2604 * funny things may happen in tcp_output.
2606 offer = max(offer, 64);
2607 taop->tao_mssopt = offer;
2610 * While we're here, check if there's an initial rtt
2611 * or rttvar. Convert from the route-table units
2612 * to scaled multiples of the slow timeout timer.
2614 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2616 * XXX the lock bit for RTT indicates that the value
2617 * is also a minimum value; this is subject to time.
2619 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2620 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2621 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2622 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2623 tcpstat.tcps_usedrtt++;
2624 if (rt->rt_rmx.rmx_rttvar) {
2625 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2626 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2627 tcpstat.tcps_usedrttvar++;
2629 /* default variation is +- 1 rtt */
2631 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2633 TCPT_RANGESET(tp->t_rxtcur,
2634 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2635 tp->t_rttmin, TCPTV_REXMTMAX);
2638 * if there's an mtu associated with the route, use it
2639 * else, use the link mtu.
2641 if (rt->rt_rmx.rmx_mtu)
2642 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2645 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2647 if (!in6_localaddr(&inp->in6p_faddr))
2648 mss = min(mss, tcp_v6mssdflt);
2650 mss = ifp->if_mtu - min_protoh;
2651 if (!in_localaddr(inp->inp_faddr))
2652 mss = min(mss, tcp_mssdflt);
2655 mss = min(mss, offer);
2657 * maxopd stores the maximum length of data AND options
2658 * in a segment; maxseg is the amount of data in a normal
2659 * segment. We need to store this value (maxopd) apart
2660 * from maxseg, because now every segment carries options
2661 * and thus we normally have somewhat less data in segments.
2666 * In case of T/TCP, origoffer==-1 indicates, that no segments
2667 * were received yet. In this case we just guess, otherwise
2668 * we do the same as before T/TCP.
2670 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2672 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2673 mss -= TCPOLEN_TSTAMP_APPA;
2674 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2676 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2677 mss -= TCPOLEN_CC_APPA;
2679 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2681 mss &= ~(MCLBYTES-1);
2684 mss = mss / MCLBYTES * MCLBYTES;
2687 * If there's a pipesize, change the socket buffer
2688 * to that size. Make the socket buffers an integral
2689 * number of mss units; if the mss is larger than
2690 * the socket buffer, decrease the mss.
2693 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2695 bufsize = so->so_snd.sb_hiwat;
2699 bufsize = roundup(bufsize, mss);
2700 if (bufsize > sb_max)
2702 if (bufsize > so->so_snd.sb_hiwat)
2703 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2708 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2710 bufsize = so->so_rcv.sb_hiwat;
2711 if (bufsize > mss) {
2712 bufsize = roundup(bufsize, mss);
2713 if (bufsize > sb_max)
2715 if (bufsize > so->so_rcv.sb_hiwat)
2716 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2720 * Set the slow-start flight size depending on whether this
2721 * is a local network or not.
2724 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2725 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2726 (!isipv6 && in_localaddr(inp->inp_faddr)))
2727 tp->snd_cwnd = mss * ss_fltsz_local;
2729 tp->snd_cwnd = mss * ss_fltsz;
2731 if (rt->rt_rmx.rmx_ssthresh) {
2733 * There's some sort of gateway or interface
2734 * buffer limit on the path. Use this to set
2735 * the slow start threshhold, but set the
2736 * threshold to no less than 2*mss.
2738 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2739 tcpstat.tcps_usedssthresh++;
2744 * Determine the MSS option to send on an outgoing SYN.
2752 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2753 int min_protoh = isipv6 ?
2754 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2755 sizeof(struct tcpiphdr);
2757 const int isipv6 = 0;
2758 const size_t min_protoh = sizeof(struct tcpiphdr);
2762 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2764 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2766 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2768 return (rt->rt_ifp->if_mtu - min_protoh);
2773 * When a partial ack arrives, force the retransmission of the
2774 * next unacknowledged segment. Do not clear tp->t_dupacks.
2775 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2779 tcp_newreno_partial_ack(tp, th)
2783 tcp_seq onxt = tp->snd_nxt;
2784 u_long ocwnd = tp->snd_cwnd;
2786 callout_stop(tp->tt_rexmt);
2788 tp->snd_nxt = th->th_ack;
2790 * Set snd_cwnd to one segment beyond acknowledged offset
2791 * (tp->snd_una has not yet been updated when this function is called.)
2793 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2794 tp->t_flags |= TF_ACKNOW;
2795 (void) tcp_output(tp);
2796 tp->snd_cwnd = ocwnd;
2797 if (SEQ_GT(onxt, tp->snd_nxt))
2800 * Partial window deflation. Relies on fact that tp->snd_una
2803 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);