2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
34 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
35 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.5 2003/07/24 23:33:33 hsu Exp $
38 #include "opt_ipfw.h" /* for ipfw_fwd */
39 #include "opt_inet6.h"
40 #include "opt_ipsec.h"
41 #include "opt_tcpdebug.h"
42 #include "opt_tcp_input.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
48 #include <sys/malloc.h>
50 #include <sys/proc.h> /* for proc0 declaration */
51 #include <sys/protosw.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/syslog.h>
56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
59 #include <net/route.h>
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
65 #include <netinet/in_var.h>
66 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet/icmp6.h>
71 #include <netinet6/nd6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet6/tcp6_var.h>
80 #include <netinet/tcpip.h>
82 #include <netinet/tcp_debug.h>
84 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
85 struct tcphdr tcp_savetcp;
89 #include <netipsec/ipsec.h>
90 #include <netipsec/ipsec6.h>
94 #include <netinet6/ipsec.h>
95 #include <netinet6/ipsec6.h>
96 #include <netkey/key.h>
99 #include <machine/in_cksum.h>
101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
103 static const int tcprexmtthresh = 3;
106 struct tcpstat tcpstat;
107 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
110 static int log_in_vain = 0;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
112 &log_in_vain, 0, "Log all incoming TCP connections");
114 static int blackhole = 0;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
116 &blackhole, 0, "Do not send RST when dropping refused connections");
118 int tcp_delack_enabled = 1;
119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
120 &tcp_delack_enabled, 0,
121 "Delay ACK to try and piggyback it onto a data packet");
123 #ifdef TCP_DROP_SYNFIN
124 static int drop_synfin = 0;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
129 static int tcp_do_limitedtransmit = 1;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
131 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
133 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 struct inpcbhead tcb;
139 #define tcb6 tcb /* for KAME src sync over BSD*'s */
140 struct inpcbinfo tcbinfo;
142 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
143 static void tcp_pulloutofband(struct socket *,
144 struct tcphdr *, struct mbuf *, int);
145 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
147 static void tcp_xmit_timer(struct tcpcb *, int);
148 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
150 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
152 #define ND6_HINT(tp) \
154 if ((tp) && (tp)->t_inpcb && \
155 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
156 (tp)->t_inpcb->in6p_route.ro_rt) \
157 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
164 * Indicate whether this ack should be delayed. We can delay the ack if
165 * - delayed acks are enabled and
166 * - there is no delayed ack timer in progress and
167 * - our last ack wasn't a 0-sized window. We never want to delay
168 * the ack that opens up a 0-sized window.
170 #define DELAY_ACK(tp) \
171 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
172 (tp->t_flags & TF_RXWIN0SENT) == 0)
175 tcp_reass(tp, th, tlenp, m)
176 register struct tcpcb *tp;
177 register struct tcphdr *th;
182 struct tseg_qent *p = NULL;
183 struct tseg_qent *nq;
184 struct tseg_qent *te;
185 struct socket *so = tp->t_inpcb->inp_socket;
189 * Call with th==0 after become established to
190 * force pre-ESTABLISHED data up to user socket.
195 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
196 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
199 tcpstat.tcps_rcvmemdrop++;
205 * Find a segment which begins after this one does.
207 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
208 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
214 * If there is a preceding segment, it may provide some of
215 * our data already. If so, drop the data from the incoming
216 * segment. If it provides all of our data, drop us.
220 /* conversion to int (in i) handles seq wraparound */
221 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
224 tcpstat.tcps_rcvduppack++;
225 tcpstat.tcps_rcvdupbyte += *tlenp;
229 * Try to present any queued data
230 * at the left window edge to the user.
231 * This is needed after the 3-WHS
234 goto present; /* ??? */
241 tcpstat.tcps_rcvoopack++;
242 tcpstat.tcps_rcvoobyte += *tlenp;
245 * While we overlap succeeding segments trim them or,
246 * if they are completely covered, dequeue them.
249 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
252 if (i < q->tqe_len) {
253 q->tqe_th->th_seq += i;
259 nq = LIST_NEXT(q, tqe_q);
260 LIST_REMOVE(q, tqe_q);
266 /* Insert the new segment queue entry into place. */
269 te->tqe_len = *tlenp;
272 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
274 LIST_INSERT_AFTER(p, te, tqe_q);
279 * Present data to user, advancing rcv_nxt through
280 * completed sequence space.
282 if (!TCPS_HAVEESTABLISHED(tp->t_state))
284 q = LIST_FIRST(&tp->t_segq);
285 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
288 tp->rcv_nxt += q->tqe_len;
289 flags = q->tqe_th->th_flags & TH_FIN;
290 nq = LIST_NEXT(q, tqe_q);
291 LIST_REMOVE(q, tqe_q);
292 if (so->so_state & SS_CANTRCVMORE)
295 sbappend(&so->so_rcv, q->tqe_m);
298 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
305 * TCP input routine, follows pages 65-76 of the
306 * protocol specification dated September, 1981 very closely.
310 tcp6_input(mp, offp, proto)
314 register struct mbuf *m = *mp;
315 struct in6_ifaddr *ia6;
317 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
320 * draft-itojun-ipv6-tcp-to-anycast
321 * better place to put this in?
323 ia6 = ip6_getdstifaddr(m);
324 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
327 ip6 = mtod(m, struct ip6_hdr *);
328 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
329 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
333 tcp_input(m, *offp, proto);
339 tcp_input(m, off0, proto)
340 register struct mbuf *m;
343 register struct tcphdr *th;
344 register struct ip *ip = NULL;
345 register struct ipovly *ipov;
346 register struct inpcb *inp = NULL;
351 register struct tcpcb *tp = NULL;
352 register int thflags;
353 struct socket *so = 0;
354 int todrop, acked, ourfinisacked, needoutput = 0;
356 struct tcpopt to; /* options in this segment */
357 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
358 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
359 struct sockaddr_in *next_hop = NULL;
360 int rstreason; /* For badport_bandlim accounting purposes */
361 struct ip6_hdr *ip6 = NULL;
365 const int isipv6 = 0;
371 /* Grab info from MT_TAG mbufs prepended to the chain. */
372 for (;m && m->m_type == MT_TAG; m = m->m_next) {
373 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
374 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
377 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
379 bzero((char *)&to, sizeof(to));
381 tcpstat.tcps_rcvtotal++;
384 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
385 ip6 = mtod(m, struct ip6_hdr *);
386 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
387 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
388 tcpstat.tcps_rcvbadsum++;
391 th = (struct tcphdr *)((caddr_t)ip6 + off0);
394 * Be proactive about unspecified IPv6 address in source.
395 * As we use all-zero to indicate unbounded/unconnected pcb,
396 * unspecified IPv6 address can be used to confuse us.
398 * Note that packets with unspecified IPv6 destination is
399 * already dropped in ip6_input.
401 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
407 * Get IP and TCP header together in first mbuf.
408 * Note: IP leaves IP header in first mbuf.
410 if (off0 > sizeof(struct ip)) {
411 ip_stripoptions(m, (struct mbuf *)0);
412 off0 = sizeof(struct ip);
414 if (m->m_len < sizeof(struct tcpiphdr)) {
415 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
416 tcpstat.tcps_rcvshort++;
420 ip = mtod(m, struct ip *);
421 ipov = (struct ipovly *)ip;
422 th = (struct tcphdr *)((caddr_t)ip + off0);
425 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
426 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
427 th->th_sum = m->m_pkthdr.csum_data;
429 th->th_sum = in_pseudo(ip->ip_src.s_addr,
431 htonl(m->m_pkthdr.csum_data +
434 th->th_sum ^= 0xffff;
437 * Checksum extended TCP header and data.
439 len = sizeof(struct ip) + tlen;
440 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
441 ipov->ih_len = (u_short)tlen;
442 ipov->ih_len = htons(ipov->ih_len);
443 th->th_sum = in_cksum(m, len);
446 tcpstat.tcps_rcvbadsum++;
450 /* Re-initialization for later version check */
451 ip->ip_v = IPVERSION;
456 * Check that TCP offset makes sense,
457 * pull out TCP options and adjust length. XXX
459 off = th->th_off << 2;
460 if (off < sizeof(struct tcphdr) || off > tlen) {
461 tcpstat.tcps_rcvbadoff++;
464 tlen -= off; /* tlen is used instead of ti->ti_len */
465 if (off > sizeof(struct tcphdr)) {
467 IP6_EXTHDR_CHECK(m, off0, off, );
468 ip6 = mtod(m, struct ip6_hdr *);
469 th = (struct tcphdr *)((caddr_t)ip6 + off0);
471 if (m->m_len < sizeof(struct ip) + off) {
472 if ((m = m_pullup(m, sizeof(struct ip) + off))
474 tcpstat.tcps_rcvshort++;
477 ip = mtod(m, struct ip *);
478 ipov = (struct ipovly *)ip;
479 th = (struct tcphdr *)((caddr_t)ip + off0);
482 optlen = off - sizeof(struct tcphdr);
483 optp = (u_char *)(th + 1);
485 thflags = th->th_flags;
487 #ifdef TCP_DROP_SYNFIN
489 * If the drop_synfin option is enabled, drop all packets with
490 * both the SYN and FIN bits set. This prevents e.g. nmap from
491 * identifying the TCP/IP stack.
493 * This is a violation of the TCP specification.
495 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
500 * Convert TCP protocol specific fields to host format.
502 th->th_seq = ntohl(th->th_seq);
503 th->th_ack = ntohl(th->th_ack);
504 th->th_win = ntohs(th->th_win);
505 th->th_urp = ntohs(th->th_urp);
508 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
509 * until after ip6_savecontrol() is called and before other functions
510 * which don't want those proto headers.
511 * Because ip6_savecontrol() is going to parse the mbuf to
512 * search for data to be passed up to user-land, it wants mbuf
513 * parameters to be unchanged.
514 * XXX: the call of ip6_savecontrol() has been obsoleted based on
515 * latest version of the advanced API (20020110).
517 drop_hdrlen = off0 + off;
520 * Locate pcb for segment.
523 /* IPFIREWALL_FORWARD section */
524 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
526 * Transparently forwarded. Pretend to be the destination.
527 * already got one like this?
529 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
530 ip->ip_dst, th->th_dport,
531 0, m->m_pkthdr.rcvif);
533 /* It's new. Try find the ambushing socket. */
534 inp = in_pcblookup_hash(&tcbinfo,
535 ip->ip_src, th->th_sport,
538 ntohs(next_hop->sin_port) :
540 1, m->m_pkthdr.rcvif);
544 inp = in6_pcblookup_hash(&tcbinfo,
545 &ip6->ip6_src, th->th_sport,
546 &ip6->ip6_dst, th->th_dport,
547 1, m->m_pkthdr.rcvif);
549 inp = in_pcblookup_hash(&tcbinfo,
550 ip->ip_src, th->th_sport,
551 ip->ip_dst, th->th_dport,
552 1, m->m_pkthdr.rcvif);
557 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
558 ipsec6stat.in_polvio++;
562 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
563 ipsecstat.in_polvio++;
570 if (inp != NULL && ipsec6_in_reject(m, inp)) {
574 if (inp != NULL && ipsec4_in_reject(m, inp)) {
581 * If the state is CLOSED (i.e., TCB does not exist) then
582 * all data in the incoming segment is discarded.
583 * If the TCB exists but is in CLOSED state, it is embryonic,
584 * but should either do a listen or a connect soon.
589 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
591 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
596 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
597 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
601 strcpy(dbuf, inet_ntoa(ip->ip_dst));
602 strcpy(sbuf, inet_ntoa(ip->ip_src));
604 switch (log_in_vain) {
606 if ((thflags & TH_SYN) == 0)
610 "Connection attempt to TCP %s:%d "
611 "from %s:%d flags:0x%02x\n",
612 dbuf, ntohs(th->th_dport), sbuf,
613 ntohs(th->th_sport), thflags);
622 if (thflags & TH_SYN)
631 rstreason = BANDLIM_RST_CLOSEDPORT;
636 rstreason = BANDLIM_RST_CLOSEDPORT;
639 if (tp->t_state == TCPS_CLOSED)
642 /* Unscale the window into a 32-bit value. */
643 if ((thflags & TH_SYN) == 0)
644 tiwin = th->th_win << tp->snd_scale;
648 so = inp->inp_socket;
649 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
650 struct in_conninfo inc;
652 if (so->so_options & SO_DEBUG) {
653 ostate = tp->t_state;
655 bcopy((char *)ip6, (char *)tcp_saveipgen,
658 bcopy((char *)ip, (char *)tcp_saveipgen,
663 /* skip if this isn't a listen socket */
664 if ((so->so_options & SO_ACCEPTCONN) == 0)
667 inc.inc_isipv6 = isipv6;
670 inc.inc6_faddr = ip6->ip6_src;
671 inc.inc6_laddr = ip6->ip6_dst;
672 inc.inc6_route.ro_rt = NULL; /* XXX */
674 inc.inc_faddr = ip->ip_src;
675 inc.inc_laddr = ip->ip_dst;
676 inc.inc_route.ro_rt = NULL; /* XXX */
678 inc.inc_fport = th->th_sport;
679 inc.inc_lport = th->th_dport;
682 * If the state is LISTEN then ignore segment if it contains
683 * a RST. If the segment contains an ACK then it is bad and
684 * send a RST. If it does not contain a SYN then it is not
685 * interesting; drop it.
687 * If the state is SYN_RECEIVED (syncache) and seg contains
688 * an ACK, but not for our SYN/ACK, send a RST. If the seg
689 * contains a RST, check the sequence number to see if it
690 * is a valid reset segment.
692 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
693 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
694 if (!syncache_expand(&inc, th, &so, m)) {
696 * No syncache entry, or ACK was not
697 * for our SYN/ACK. Send a RST.
699 tcpstat.tcps_badsyn++;
700 rstreason = BANDLIM_RST_OPENPORT;
705 * Could not complete 3-way handshake,
706 * connection is being closed down, and
707 * syncache will free mbuf.
711 * Socket is created in state SYN_RECEIVED.
712 * Continue processing segment.
717 * This is what would have happened in
718 * tcp_output() when the SYN,ACK was sent.
720 tp->snd_up = tp->snd_una;
721 tp->snd_max = tp->snd_nxt = tp->iss + 1;
722 tp->last_ack_sent = tp->rcv_nxt;
724 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
725 * until the _second_ ACK is received:
726 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
727 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
728 * move to ESTAB, set snd_wnd to tiwin.
730 tp->snd_wnd = tiwin; /* unscaled */
733 if (thflags & TH_RST) {
734 syncache_chkrst(&inc, th);
737 if (thflags & TH_ACK) {
738 syncache_badack(&inc);
739 tcpstat.tcps_badsyn++;
740 rstreason = BANDLIM_RST_OPENPORT;
747 * Segment's flags are (SYN) or (SYN|FIN).
751 * If deprecated address is forbidden,
752 * we do not accept SYN to deprecated interface
753 * address to prevent any new inbound connection from
754 * getting established.
755 * When we do not accept SYN, we send a TCP RST,
756 * with deprecated source address (instead of dropping
757 * it). We compromise it as it is much better for peer
758 * to send a RST, and RST will be the final packet
761 * If we do not forbid deprecated addresses, we accept
762 * the SYN packet. RFC2462 does not suggest dropping
764 * If we decipher RFC2462 5.5.4, it says like this:
765 * 1. use of deprecated addr with existing
766 * communication is okay - "SHOULD continue to be
768 * 2. use of it with new communication:
769 * (2a) "SHOULD NOT be used if alternate address
770 * with sufficient scope is available"
771 * (2b) nothing mentioned otherwise.
772 * Here we fall into (2b) case as we have no choice in
773 * our source address selection - we must obey the peer.
775 * The wording in RFC2462 is confusing, and there are
776 * multiple description text for deprecated address
777 * handling - worse, they are not exactly the same.
778 * I believe 5.5.4 is the best one, so we follow 5.5.4.
780 if (isipv6 && !ip6_use_deprecated) {
781 struct in6_ifaddr *ia6;
783 if ((ia6 = ip6_getdstifaddr(m)) &&
784 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
786 rstreason = BANDLIM_RST_OPENPORT;
792 * If it is from this socket, drop it, it must be forged.
793 * Don't bother responding if the destination was a broadcast.
795 if (th->th_dport == th->th_sport) {
797 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
801 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
806 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
808 * Note that it is quite possible to receive unicast
809 * link-layer packets with a broadcast IP address. Use
810 * in_broadcast() to find them.
812 if (m->m_flags & (M_BCAST|M_MCAST))
815 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
816 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
819 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
820 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
821 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
822 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
826 * SYN appears to be valid; create compressed TCP state
827 * for syncache, or perform t/tcp connection.
829 if (so->so_qlen <= so->so_qlimit) {
830 tcp_dooptions(&to, optp, optlen, 1);
831 if (!syncache_add(&inc, &to, th, &so, m))
835 * Entry added to syncache, mbuf used to
836 * send SYN,ACK packet.
840 * Segment passed TAO tests.
845 tp->t_starttime = ticks;
846 tp->t_state = TCPS_ESTABLISHED;
849 * If there is a FIN, or if there is data and the
850 * connection is local, then delay SYN,ACK(SYN) in
851 * the hope of piggy-backing it on a response
852 * segment. Otherwise must send ACK now in case
853 * the other side is slow starting.
856 ((thflags & TH_FIN) ||
858 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
859 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
860 callout_reset(tp->tt_delack, tcp_delacktime,
861 tcp_timer_delack, tp);
862 tp->t_flags |= TF_NEEDSYN;
864 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
866 tcpstat.tcps_connects++;
874 /* XXX temp debugging */
875 /* should not happen - syncache should pick up these connections */
876 if (tp->t_state == TCPS_LISTEN)
877 panic("tcp_input: TCPS_LISTEN");
880 * Segment received on connection.
881 * Reset idle time and keep-alive timer.
883 tp->t_rcvtime = ticks;
884 if (TCPS_HAVEESTABLISHED(tp->t_state))
885 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
889 * XXX this is tradtitional behavior, may need to be cleaned up.
891 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
892 if (thflags & TH_SYN) {
893 if (to.to_flags & TOF_SCALE) {
894 tp->t_flags |= TF_RCVD_SCALE;
895 tp->requested_s_scale = to.to_requested_s_scale;
897 if (to.to_flags & TOF_TS) {
898 tp->t_flags |= TF_RCVD_TSTMP;
899 tp->ts_recent = to.to_tsval;
900 tp->ts_recent_age = ticks;
902 if (to.to_flags & (TOF_CC|TOF_CCNEW))
903 tp->t_flags |= TF_RCVD_CC;
904 if (to.to_flags & TOF_MSS)
905 tcp_mss(tp, to.to_mss);
909 * Header prediction: check for the two common cases
910 * of a uni-directional data xfer. If the packet has
911 * no control flags, is in-sequence, the window didn't
912 * change and we're not retransmitting, it's a
913 * candidate. If the length is zero and the ack moved
914 * forward, we're the sender side of the xfer. Just
915 * free the data acked & wake any higher level process
916 * that was blocked waiting for space. If the length
917 * is non-zero and the ack didn't move, we're the
918 * receiver side. If we're getting packets in-order
919 * (the reassembly queue is empty), add the data to
920 * the socket buffer and note that we need a delayed ack.
921 * Make sure that the hidden state-flags are also off.
922 * Since we check for TCPS_ESTABLISHED above, it can only
925 if (tp->t_state == TCPS_ESTABLISHED &&
926 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
927 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
928 ((to.to_flags & TOF_TS) == 0 ||
929 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
931 * Using the CC option is compulsory if once started:
932 * the segment is OK if no T/TCP was negotiated or
933 * if the segment has a CC option equal to CCrecv
935 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
936 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
937 th->th_seq == tp->rcv_nxt &&
938 tiwin && tiwin == tp->snd_wnd &&
939 tp->snd_nxt == tp->snd_max) {
942 * If last ACK falls within this segment's sequence numbers,
943 * record the timestamp.
944 * NOTE that the test is modified according to the latest
945 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
947 if ((to.to_flags & TOF_TS) != 0 &&
948 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
949 tp->ts_recent_age = ticks;
950 tp->ts_recent = to.to_tsval;
954 if (SEQ_GT(th->th_ack, tp->snd_una) &&
955 SEQ_LEQ(th->th_ack, tp->snd_max) &&
956 tp->snd_cwnd >= tp->snd_wnd &&
958 tp->t_dupacks < tcprexmtthresh) ||
959 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
961 * this is a pure ack for outstanding data.
963 ++tcpstat.tcps_predack;
965 * "bad retransmit" recovery
967 if (tp->t_rxtshift == 1 &&
968 ticks < tp->t_badrxtwin) {
969 tp->snd_cwnd = tp->snd_cwnd_prev;
971 tp->snd_ssthresh_prev;
972 tp->snd_recover = tp->snd_recover_prev;
973 if (tp->t_flags & TF_WASFRECOVERY)
974 ENTER_FASTRECOVERY(tp);
975 tp->snd_nxt = tp->snd_max;
979 * Recalculate the retransmit timer / rtt.
981 * Some machines (certain windows boxes)
982 * send broken timestamp replies during the
983 * SYN+ACK phase, ignore timestamps of 0.
985 if ((to.to_flags & TOF_TS) != 0 &&
988 ticks - to.to_tsecr + 1);
989 } else if (tp->t_rtttime &&
990 SEQ_GT(th->th_ack, tp->t_rtseq)) {
992 ticks - tp->t_rtttime);
994 tcp_xmit_bandwidth_limit(tp, th->th_ack);
995 acked = th->th_ack - tp->snd_una;
996 tcpstat.tcps_rcvackpack++;
997 tcpstat.tcps_rcvackbyte += acked;
998 sbdrop(&so->so_snd, acked);
999 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1000 SEQ_LEQ(th->th_ack, tp->snd_recover))
1001 tp->snd_recover = th->th_ack - 1;
1002 tp->snd_una = th->th_ack;
1005 ND6_HINT(tp); /* some progress has been done */
1008 * If all outstanding data are acked, stop
1009 * retransmit timer, otherwise restart timer
1010 * using current (possibly backed-off) value.
1011 * If process is waiting for space,
1012 * wakeup/selwakeup/signal. If data
1013 * are ready to send, let tcp_output
1014 * decide between more output or persist.
1016 if (tp->snd_una == tp->snd_max)
1017 callout_stop(tp->tt_rexmt);
1018 else if (!callout_active(tp->tt_persist))
1019 callout_reset(tp->tt_rexmt,
1021 tcp_timer_rexmt, tp);
1024 if (so->so_snd.sb_cc)
1025 (void) tcp_output(tp);
1028 } else if (th->th_ack == tp->snd_una &&
1029 LIST_EMPTY(&tp->t_segq) &&
1030 tlen <= sbspace(&so->so_rcv)) {
1032 * this is a pure, in-sequence data packet
1033 * with nothing on the reassembly queue and
1034 * we have enough buffer space to take it.
1036 ++tcpstat.tcps_preddat;
1037 tp->rcv_nxt += tlen;
1038 tcpstat.tcps_rcvpack++;
1039 tcpstat.tcps_rcvbyte += tlen;
1040 ND6_HINT(tp); /* some progress has been done */
1042 * Add data to socket buffer.
1044 if (so->so_state & SS_CANTRCVMORE) {
1047 m_adj(m, drop_hdrlen); /* delayed header drop */
1048 sbappend(&so->so_rcv, m);
1051 if (DELAY_ACK(tp)) {
1052 callout_reset(tp->tt_delack, tcp_delacktime,
1053 tcp_timer_delack, tp);
1055 tp->t_flags |= TF_ACKNOW;
1063 * Calculate amount of space in receive window,
1064 * and then do TCP input processing.
1065 * Receive window is amount of space in rcv queue,
1066 * but not less than advertised window.
1070 win = sbspace(&so->so_rcv);
1073 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1076 switch (tp->t_state) {
1079 * If the state is SYN_RECEIVED:
1080 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1082 case TCPS_SYN_RECEIVED:
1083 if ((thflags & TH_ACK) &&
1084 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1085 SEQ_GT(th->th_ack, tp->snd_max))) {
1086 rstreason = BANDLIM_RST_OPENPORT;
1092 * If the state is SYN_SENT:
1093 * if seg contains an ACK, but not for our SYN, drop the input.
1094 * if seg contains a RST, then drop the connection.
1095 * if seg does not contain SYN, then drop it.
1096 * Otherwise this is an acceptable SYN segment
1097 * initialize tp->rcv_nxt and tp->irs
1098 * if seg contains ack then advance tp->snd_una
1099 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1100 * arrange for segment to be acked (eventually)
1101 * continue processing rest of data/controls, beginning with URG
1104 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1105 taop = &tao_noncached;
1106 bzero(taop, sizeof(*taop));
1109 if ((thflags & TH_ACK) &&
1110 (SEQ_LEQ(th->th_ack, tp->iss) ||
1111 SEQ_GT(th->th_ack, tp->snd_max))) {
1113 * If we have a cached CCsent for the remote host,
1114 * hence we haven't just crashed and restarted,
1115 * do not send a RST. This may be a retransmission
1116 * from the other side after our earlier ACK was lost.
1117 * Our new SYN, when it arrives, will serve as the
1120 if (taop->tao_ccsent != 0)
1123 rstreason = BANDLIM_UNLIMITED;
1127 if (thflags & TH_RST) {
1128 if (thflags & TH_ACK)
1129 tp = tcp_drop(tp, ECONNREFUSED);
1132 if ((thflags & TH_SYN) == 0)
1134 tp->snd_wnd = th->th_win; /* initial send window */
1135 tp->cc_recv = to.to_cc; /* foreign CC */
1137 tp->irs = th->th_seq;
1139 if (thflags & TH_ACK) {
1141 * Our SYN was acked. If segment contains CC.ECHO
1142 * option, check it to make sure this segment really
1143 * matches our SYN. If not, just drop it as old
1144 * duplicate, but send an RST if we're still playing
1145 * by the old rules. If no CC.ECHO option, make sure
1146 * we don't get fooled into using T/TCP.
1148 if (to.to_flags & TOF_CCECHO) {
1149 if (tp->cc_send != to.to_ccecho) {
1150 if (taop->tao_ccsent != 0)
1153 rstreason = BANDLIM_UNLIMITED;
1158 tp->t_flags &= ~TF_RCVD_CC;
1159 tcpstat.tcps_connects++;
1161 /* Do window scaling on this connection? */
1162 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1163 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1164 tp->snd_scale = tp->requested_s_scale;
1165 tp->rcv_scale = tp->request_r_scale;
1167 /* Segment is acceptable, update cache if undefined. */
1168 if (taop->tao_ccsent == 0)
1169 taop->tao_ccsent = to.to_ccecho;
1171 tp->rcv_adv += tp->rcv_wnd;
1172 tp->snd_una++; /* SYN is acked */
1174 * If there's data, delay ACK; if there's also a FIN
1175 * ACKNOW will be turned on later.
1177 if (DELAY_ACK(tp) && tlen != 0)
1178 callout_reset(tp->tt_delack, tcp_delacktime,
1179 tcp_timer_delack, tp);
1181 tp->t_flags |= TF_ACKNOW;
1183 * Received <SYN,ACK> in SYN_SENT[*] state.
1185 * SYN_SENT --> ESTABLISHED
1186 * SYN_SENT* --> FIN_WAIT_1
1188 tp->t_starttime = ticks;
1189 if (tp->t_flags & TF_NEEDFIN) {
1190 tp->t_state = TCPS_FIN_WAIT_1;
1191 tp->t_flags &= ~TF_NEEDFIN;
1194 tp->t_state = TCPS_ESTABLISHED;
1195 callout_reset(tp->tt_keep, tcp_keepidle,
1196 tcp_timer_keep, tp);
1200 * Received initial SYN in SYN-SENT[*] state =>
1201 * simultaneous open. If segment contains CC option
1202 * and there is a cached CC, apply TAO test.
1203 * If it succeeds, connection is * half-synchronized.
1204 * Otherwise, do 3-way handshake:
1205 * SYN-SENT -> SYN-RECEIVED
1206 * SYN-SENT* -> SYN-RECEIVED*
1207 * If there was no CC option, clear cached CC value.
1209 tp->t_flags |= TF_ACKNOW;
1210 callout_stop(tp->tt_rexmt);
1211 if (to.to_flags & TOF_CC) {
1212 if (taop->tao_cc != 0 &&
1213 CC_GT(to.to_cc, taop->tao_cc)) {
1215 * update cache and make transition:
1216 * SYN-SENT -> ESTABLISHED*
1217 * SYN-SENT* -> FIN-WAIT-1*
1219 taop->tao_cc = to.to_cc;
1220 tp->t_starttime = ticks;
1221 if (tp->t_flags & TF_NEEDFIN) {
1222 tp->t_state = TCPS_FIN_WAIT_1;
1223 tp->t_flags &= ~TF_NEEDFIN;
1225 tp->t_state = TCPS_ESTABLISHED;
1226 callout_reset(tp->tt_keep,
1231 tp->t_flags |= TF_NEEDSYN;
1233 tp->t_state = TCPS_SYN_RECEIVED;
1235 /* CC.NEW or no option => invalidate cache */
1237 tp->t_state = TCPS_SYN_RECEIVED;
1243 * Advance th->th_seq to correspond to first data byte.
1244 * If data, trim to stay within window,
1245 * dropping FIN if necessary.
1248 if (tlen > tp->rcv_wnd) {
1249 todrop = tlen - tp->rcv_wnd;
1253 tcpstat.tcps_rcvpackafterwin++;
1254 tcpstat.tcps_rcvbyteafterwin += todrop;
1256 tp->snd_wl1 = th->th_seq - 1;
1257 tp->rcv_up = th->th_seq;
1259 * Client side of transaction: already sent SYN and data.
1260 * If the remote host used T/TCP to validate the SYN,
1261 * our data will be ACK'd; if so, enter normal data segment
1262 * processing in the middle of step 5, ack processing.
1263 * Otherwise, goto step 6.
1265 if (thflags & TH_ACK)
1271 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1272 * if segment contains a SYN and CC [not CC.NEW] option:
1273 * if state == TIME_WAIT and connection duration > MSL,
1274 * drop packet and send RST;
1276 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1277 * ack the FIN (and data) in retransmission queue.
1278 * Complete close and delete TCPCB. Then reprocess
1279 * segment, hoping to find new TCPCB in LISTEN state;
1281 * else must be old SYN; drop it.
1282 * else do normal processing.
1286 case TCPS_TIME_WAIT:
1287 if ((thflags & TH_SYN) &&
1288 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1289 if (tp->t_state == TCPS_TIME_WAIT &&
1290 (ticks - tp->t_starttime) > tcp_msl) {
1291 rstreason = BANDLIM_UNLIMITED;
1294 if (CC_GT(to.to_cc, tp->cc_recv)) {
1301 break; /* continue normal processing */
1305 * States other than LISTEN or SYN_SENT.
1306 * First check the RST flag and sequence number since reset segments
1307 * are exempt from the timestamp and connection count tests. This
1308 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1309 * below which allowed reset segments in half the sequence space
1310 * to fall though and be processed (which gives forged reset
1311 * segments with a random sequence number a 50 percent chance of
1312 * killing a connection).
1313 * Then check timestamp, if present.
1314 * Then check the connection count, if present.
1315 * Then check that at least some bytes of segment are within
1316 * receive window. If segment begins before rcv_nxt,
1317 * drop leading data (and SYN); if nothing left, just ack.
1320 * If the RST bit is set, check the sequence number to see
1321 * if this is a valid reset segment.
1323 * In all states except SYN-SENT, all reset (RST) segments
1324 * are validated by checking their SEQ-fields. A reset is
1325 * valid if its sequence number is in the window.
1326 * Note: this does not take into account delayed ACKs, so
1327 * we should test against last_ack_sent instead of rcv_nxt.
1328 * The sequence number in the reset segment is normally an
1329 * echo of our outgoing acknowlegement numbers, but some hosts
1330 * send a reset with the sequence number at the rightmost edge
1331 * of our receive window, and we have to handle this case.
1332 * If we have multiple segments in flight, the intial reset
1333 * segment sequence numbers will be to the left of last_ack_sent,
1334 * but they will eventually catch up.
1335 * In any case, it never made sense to trim reset segments to
1336 * fit the receive window since RFC 1122 says:
1337 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1339 * A TCP SHOULD allow a received RST segment to include data.
1342 * It has been suggested that a RST segment could contain
1343 * ASCII text that encoded and explained the cause of the
1344 * RST. No standard has yet been established for such
1347 * If the reset segment passes the sequence number test examine
1349 * SYN_RECEIVED STATE:
1350 * If passive open, return to LISTEN state.
1351 * If active open, inform user that connection was refused.
1352 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1353 * Inform user that connection was reset, and close tcb.
1354 * CLOSING, LAST_ACK STATES:
1357 * Drop the segment - see Stevens, vol. 2, p. 964 and
1360 if (thflags & TH_RST) {
1361 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1362 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1363 switch (tp->t_state) {
1365 case TCPS_SYN_RECEIVED:
1366 so->so_error = ECONNREFUSED;
1369 case TCPS_ESTABLISHED:
1370 case TCPS_FIN_WAIT_1:
1371 case TCPS_FIN_WAIT_2:
1372 case TCPS_CLOSE_WAIT:
1373 so->so_error = ECONNRESET;
1375 tp->t_state = TCPS_CLOSED;
1376 tcpstat.tcps_drops++;
1385 case TCPS_TIME_WAIT:
1393 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1394 * and it's less than ts_recent, drop it.
1396 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1397 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1399 /* Check to see if ts_recent is over 24 days old. */
1400 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1402 * Invalidate ts_recent. If this segment updates
1403 * ts_recent, the age will be reset later and ts_recent
1404 * will get a valid value. If it does not, setting
1405 * ts_recent to zero will at least satisfy the
1406 * requirement that zero be placed in the timestamp
1407 * echo reply when ts_recent isn't valid. The
1408 * age isn't reset until we get a valid ts_recent
1409 * because we don't want out-of-order segments to be
1410 * dropped when ts_recent is old.
1414 tcpstat.tcps_rcvduppack++;
1415 tcpstat.tcps_rcvdupbyte += tlen;
1416 tcpstat.tcps_pawsdrop++;
1425 * If T/TCP was negotiated and the segment doesn't have CC,
1426 * or if its CC is wrong then drop the segment.
1427 * RST segments do not have to comply with this.
1429 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1430 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1434 * In the SYN-RECEIVED state, validate that the packet belongs to
1435 * this connection before trimming the data to fit the receive
1436 * window. Check the sequence number versus IRS since we know
1437 * the sequence numbers haven't wrapped. This is a partial fix
1438 * for the "LAND" DoS attack.
1440 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1441 rstreason = BANDLIM_RST_OPENPORT;
1445 todrop = tp->rcv_nxt - th->th_seq;
1447 if (thflags & TH_SYN) {
1457 * Following if statement from Stevens, vol. 2, p. 960.
1460 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1462 * Any valid FIN must be to the left of the window.
1463 * At this point the FIN must be a duplicate or out
1464 * of sequence; drop it.
1469 * Send an ACK to resynchronize and drop any data.
1470 * But keep on processing for RST or ACK.
1472 tp->t_flags |= TF_ACKNOW;
1474 tcpstat.tcps_rcvduppack++;
1475 tcpstat.tcps_rcvdupbyte += todrop;
1477 tcpstat.tcps_rcvpartduppack++;
1478 tcpstat.tcps_rcvpartdupbyte += todrop;
1480 drop_hdrlen += todrop; /* drop from the top afterwards */
1481 th->th_seq += todrop;
1483 if (th->th_urp > todrop)
1484 th->th_urp -= todrop;
1492 * If new data are received on a connection after the
1493 * user processes are gone, then RST the other end.
1495 if ((so->so_state & SS_NOFDREF) &&
1496 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1498 tcpstat.tcps_rcvafterclose++;
1499 rstreason = BANDLIM_UNLIMITED;
1504 * If segment ends after window, drop trailing data
1505 * (and PUSH and FIN); if nothing left, just ACK.
1507 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1509 tcpstat.tcps_rcvpackafterwin++;
1510 if (todrop >= tlen) {
1511 tcpstat.tcps_rcvbyteafterwin += tlen;
1513 * If a new connection request is received
1514 * while in TIME_WAIT, drop the old connection
1515 * and start over if the sequence numbers
1516 * are above the previous ones.
1518 if (thflags & TH_SYN &&
1519 tp->t_state == TCPS_TIME_WAIT &&
1520 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1525 * If window is closed can only take segments at
1526 * window edge, and have to drop data and PUSH from
1527 * incoming segments. Continue processing, but
1528 * remember to ack. Otherwise, drop segment
1531 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1532 tp->t_flags |= TF_ACKNOW;
1533 tcpstat.tcps_rcvwinprobe++;
1537 tcpstat.tcps_rcvbyteafterwin += todrop;
1540 thflags &= ~(TH_PUSH|TH_FIN);
1544 * If last ACK falls within this segment's sequence numbers,
1545 * record its timestamp.
1546 * NOTE that the test is modified according to the latest
1547 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1549 if ((to.to_flags & TOF_TS) != 0 &&
1550 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1551 tp->ts_recent_age = ticks;
1552 tp->ts_recent = to.to_tsval;
1556 * If a SYN is in the window, then this is an
1557 * error and we send an RST and drop the connection.
1559 if (thflags & TH_SYN) {
1560 tp = tcp_drop(tp, ECONNRESET);
1561 rstreason = BANDLIM_UNLIMITED;
1566 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1567 * flag is on (half-synchronized state), then queue data for
1568 * later processing; else drop segment and return.
1570 if ((thflags & TH_ACK) == 0) {
1571 if (tp->t_state == TCPS_SYN_RECEIVED ||
1572 (tp->t_flags & TF_NEEDSYN))
1581 switch (tp->t_state) {
1584 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1585 * ESTABLISHED state and continue processing.
1586 * The ACK was checked above.
1588 case TCPS_SYN_RECEIVED:
1590 tcpstat.tcps_connects++;
1592 /* Do window scaling? */
1593 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1594 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1595 tp->snd_scale = tp->requested_s_scale;
1596 tp->rcv_scale = tp->request_r_scale;
1599 * Upon successful completion of 3-way handshake,
1600 * update cache.CC if it was undefined, pass any queued
1601 * data to the user, and advance state appropriately.
1603 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1605 taop->tao_cc = tp->cc_recv;
1609 * SYN-RECEIVED -> ESTABLISHED
1610 * SYN-RECEIVED* -> FIN-WAIT-1
1612 tp->t_starttime = ticks;
1613 if (tp->t_flags & TF_NEEDFIN) {
1614 tp->t_state = TCPS_FIN_WAIT_1;
1615 tp->t_flags &= ~TF_NEEDFIN;
1617 tp->t_state = TCPS_ESTABLISHED;
1618 callout_reset(tp->tt_keep, tcp_keepidle,
1619 tcp_timer_keep, tp);
1622 * If segment contains data or ACK, will call tcp_reass()
1623 * later; if not, do so now to pass queued data to user.
1625 if (tlen == 0 && (thflags & TH_FIN) == 0)
1626 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1628 tp->snd_wl1 = th->th_seq - 1;
1632 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1633 * ACKs. If the ack is in the range
1634 * tp->snd_una < th->th_ack <= tp->snd_max
1635 * then advance tp->snd_una to th->th_ack and drop
1636 * data from the retransmission queue. If this ACK reflects
1637 * more up to date window information we update our window information.
1639 case TCPS_ESTABLISHED:
1640 case TCPS_FIN_WAIT_1:
1641 case TCPS_FIN_WAIT_2:
1642 case TCPS_CLOSE_WAIT:
1645 case TCPS_TIME_WAIT:
1647 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1648 if (tlen == 0 && tiwin == tp->snd_wnd) {
1649 tcpstat.tcps_rcvdupack++;
1651 * If we have outstanding data (other than
1652 * a window probe), this is a completely
1653 * duplicate ack (ie, window info didn't
1654 * change), the ack is the biggest we've
1655 * seen and we've seen exactly our rexmt
1656 * threshhold of them, assume a packet
1657 * has been dropped and retransmit it.
1658 * Kludge snd_nxt & the congestion
1659 * window so we send only this one
1662 * We know we're losing at the current
1663 * window size so do congestion avoidance
1664 * (set ssthresh to half the current window
1665 * and pull our congestion window back to
1666 * the new ssthresh).
1668 * Dup acks mean that packets have left the
1669 * network (they're now cached at the receiver)
1670 * so bump cwnd by the amount in the receiver
1671 * to keep a constant cwnd packets in the
1674 if (!callout_active(tp->tt_rexmt) ||
1675 th->th_ack != tp->snd_una)
1677 else if (++tp->t_dupacks > tcprexmtthresh ||
1679 IN_FASTRECOVERY(tp))) {
1680 tp->snd_cwnd += tp->t_maxseg;
1681 (void) tcp_output(tp);
1683 } else if (tp->t_dupacks == tcprexmtthresh) {
1684 tcp_seq onxt = tp->snd_nxt;
1686 if (tcp_do_newreno &&
1692 win = min(tp->snd_wnd, tp->snd_cwnd) /
1696 tp->snd_ssthresh = win * tp->t_maxseg;
1697 ENTER_FASTRECOVERY(tp);
1698 tp->snd_recover = tp->snd_max;
1699 callout_stop(tp->tt_rexmt);
1701 tp->snd_nxt = th->th_ack;
1702 tp->snd_cwnd = tp->t_maxseg;
1703 (void) tcp_output(tp);
1704 KASSERT(tp->snd_limited <= 2,
1705 ("tp->snd_limited too big"));
1706 tp->snd_cwnd = tp->snd_ssthresh +
1708 (tp->t_dupacks - tp->snd_limited));
1709 if (SEQ_GT(onxt, tp->snd_nxt))
1712 } else if (tcp_do_limitedtransmit) {
1713 u_long oldcwnd = tp->snd_cwnd;
1714 tcp_seq oldsndmax = tp->snd_max;
1716 KASSERT(tp->t_dupacks == 1 ||
1718 ("dupacks not 1 or 2"));
1719 if (tp->t_dupacks == 1) {
1720 tp->snd_limited = 0;
1721 tp->snd_cwnd += tp->t_maxseg;
1726 (void) tcp_output(tp);
1727 sent = tp->snd_max - oldsndmax;
1728 if (sent > tp->t_maxseg) {
1729 KASSERT(tp->snd_limited == 0 &&
1732 tp->snd_limited = 2;
1733 } else if (sent > 0)
1735 tp->snd_cwnd = oldcwnd;
1743 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1746 * If the congestion window was inflated to account
1747 * for the other side's cached packets, retract it.
1749 if (tcp_do_newreno) {
1750 if (IN_FASTRECOVERY(tp)) {
1751 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1752 tcp_newreno_partial_ack(tp, th);
1755 * Window inflation should have left us
1756 * with approximately snd_ssthresh
1758 * But in case we would be inclined to
1759 * send a burst, better to do it via
1760 * the slow start mechanism.
1762 if (SEQ_GT(th->th_ack +
1765 tp->snd_cwnd = tp->snd_max -
1769 tp->snd_cwnd = tp->snd_ssthresh;
1773 if (tp->t_dupacks >= tcprexmtthresh &&
1774 tp->snd_cwnd > tp->snd_ssthresh)
1775 tp->snd_cwnd = tp->snd_ssthresh;
1778 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1779 tcpstat.tcps_rcvacktoomuch++;
1783 * If we reach this point, ACK is not a duplicate,
1784 * i.e., it ACKs something we sent.
1786 if (tp->t_flags & TF_NEEDSYN) {
1788 * T/TCP: Connection was half-synchronized, and our
1789 * SYN has been ACK'd (so connection is now fully
1790 * synchronized). Go to non-starred state,
1791 * increment snd_una for ACK of SYN, and check if
1792 * we can do window scaling.
1794 tp->t_flags &= ~TF_NEEDSYN;
1796 /* Do window scaling? */
1797 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1798 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1799 tp->snd_scale = tp->requested_s_scale;
1800 tp->rcv_scale = tp->request_r_scale;
1805 acked = th->th_ack - tp->snd_una;
1806 tcpstat.tcps_rcvackpack++;
1807 tcpstat.tcps_rcvackbyte += acked;
1810 * If we just performed our first retransmit, and the ACK
1811 * arrives within our recovery window, then it was a mistake
1812 * to do the retransmit in the first place. Recover our
1813 * original cwnd and ssthresh, and proceed to transmit where
1816 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1817 tp->snd_cwnd = tp->snd_cwnd_prev;
1818 tp->snd_ssthresh = tp->snd_ssthresh_prev;
1819 tp->snd_recover = tp->snd_recover_prev;
1820 if (tp->t_flags & TF_WASFRECOVERY)
1821 ENTER_FASTRECOVERY(tp);
1822 tp->snd_nxt = tp->snd_max;
1823 tp->t_badrxtwin = 0; /* XXX probably not required */
1827 * If we have a timestamp reply, update smoothed
1828 * round trip time. If no timestamp is present but
1829 * transmit timer is running and timed sequence
1830 * number was acked, update smoothed round trip time.
1831 * Since we now have an rtt measurement, cancel the
1832 * timer backoff (cf., Phil Karn's retransmit alg.).
1833 * Recompute the initial retransmit timer.
1835 * Some machines (certain windows boxes) send broken
1836 * timestamp replies during the SYN+ACK phase, ignore
1839 if ((to.to_flags & TOF_TS) != 0 &&
1841 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1842 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1843 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1845 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1848 * If all outstanding data is acked, stop retransmit
1849 * timer and remember to restart (more output or persist).
1850 * If there is more data to be acked, restart retransmit
1851 * timer, using current (possibly backed-off) value.
1853 if (th->th_ack == tp->snd_max) {
1854 callout_stop(tp->tt_rexmt);
1856 } else if (!callout_active(tp->tt_persist))
1857 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1858 tcp_timer_rexmt, tp);
1861 * If no data (only SYN) was ACK'd,
1862 * skip rest of ACK processing.
1868 * When new data is acked, open the congestion window.
1869 * If the window gives us less than ssthresh packets
1870 * in flight, open exponentially (maxseg per packet).
1871 * Otherwise open linearly: maxseg per window
1872 * (maxseg^2 / cwnd per packet).
1874 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1875 register u_int cw = tp->snd_cwnd;
1876 register u_int incr = tp->t_maxseg;
1877 if (cw > tp->snd_ssthresh)
1878 incr = incr * incr / cw;
1879 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1881 if (acked > so->so_snd.sb_cc) {
1882 tp->snd_wnd -= so->so_snd.sb_cc;
1883 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1886 sbdrop(&so->so_snd, acked);
1887 tp->snd_wnd -= acked;
1891 /* detect una wraparound */
1892 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1893 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1894 SEQ_LEQ(th->th_ack, tp->snd_recover))
1895 tp->snd_recover = th->th_ack - 1;
1896 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1897 SEQ_GEQ(th->th_ack, tp->snd_recover))
1898 EXIT_FASTRECOVERY(tp);
1899 tp->snd_una = th->th_ack;
1900 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1901 tp->snd_nxt = tp->snd_una;
1903 switch (tp->t_state) {
1906 * In FIN_WAIT_1 STATE in addition to the processing
1907 * for the ESTABLISHED state if our FIN is now acknowledged
1908 * then enter FIN_WAIT_2.
1910 case TCPS_FIN_WAIT_1:
1911 if (ourfinisacked) {
1913 * If we can't receive any more
1914 * data, then closing user can proceed.
1915 * Starting the timer is contrary to the
1916 * specification, but if we don't get a FIN
1917 * we'll hang forever.
1919 if (so->so_state & SS_CANTRCVMORE) {
1920 soisdisconnected(so);
1921 callout_reset(tp->tt_2msl, tcp_maxidle,
1922 tcp_timer_2msl, tp);
1924 tp->t_state = TCPS_FIN_WAIT_2;
1929 * In CLOSING STATE in addition to the processing for
1930 * the ESTABLISHED state if the ACK acknowledges our FIN
1931 * then enter the TIME-WAIT state, otherwise ignore
1935 if (ourfinisacked) {
1936 tp->t_state = TCPS_TIME_WAIT;
1937 tcp_canceltimers(tp);
1938 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1939 if (tp->cc_recv != 0 &&
1940 (ticks - tp->t_starttime) < tcp_msl)
1941 callout_reset(tp->tt_2msl,
1944 tcp_timer_2msl, tp);
1946 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1947 tcp_timer_2msl, tp);
1948 soisdisconnected(so);
1953 * In LAST_ACK, we may still be waiting for data to drain
1954 * and/or to be acked, as well as for the ack of our FIN.
1955 * If our FIN is now acknowledged, delete the TCB,
1956 * enter the closed state and return.
1959 if (ourfinisacked) {
1966 * In TIME_WAIT state the only thing that should arrive
1967 * is a retransmission of the remote FIN. Acknowledge
1968 * it and restart the finack timer.
1970 case TCPS_TIME_WAIT:
1971 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1972 tcp_timer_2msl, tp);
1979 * Update window information.
1980 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1982 if ((thflags & TH_ACK) &&
1983 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
1984 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
1985 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
1986 /* keep track of pure window updates */
1988 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
1989 tcpstat.tcps_rcvwinupd++;
1990 tp->snd_wnd = tiwin;
1991 tp->snd_wl1 = th->th_seq;
1992 tp->snd_wl2 = th->th_ack;
1993 if (tp->snd_wnd > tp->max_sndwnd)
1994 tp->max_sndwnd = tp->snd_wnd;
1999 * Process segments with URG.
2001 if ((thflags & TH_URG) && th->th_urp &&
2002 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2004 * This is a kludge, but if we receive and accept
2005 * random urgent pointers, we'll crash in
2006 * soreceive. It's hard to imagine someone
2007 * actually wanting to send this much urgent data.
2009 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2010 th->th_urp = 0; /* XXX */
2011 thflags &= ~TH_URG; /* XXX */
2012 goto dodata; /* XXX */
2015 * If this segment advances the known urgent pointer,
2016 * then mark the data stream. This should not happen
2017 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2018 * a FIN has been received from the remote side.
2019 * In these states we ignore the URG.
2021 * According to RFC961 (Assigned Protocols),
2022 * the urgent pointer points to the last octet
2023 * of urgent data. We continue, however,
2024 * to consider it to indicate the first octet
2025 * of data past the urgent section as the original
2026 * spec states (in one of two places).
2028 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2029 tp->rcv_up = th->th_seq + th->th_urp;
2030 so->so_oobmark = so->so_rcv.sb_cc +
2031 (tp->rcv_up - tp->rcv_nxt) - 1;
2032 if (so->so_oobmark == 0)
2033 so->so_state |= SS_RCVATMARK;
2035 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2038 * Remove out of band data so doesn't get presented to user.
2039 * This can happen independent of advancing the URG pointer,
2040 * but if two URG's are pending at once, some out-of-band
2041 * data may creep in... ick.
2043 if (th->th_urp <= (u_long)tlen
2045 && (so->so_options & SO_OOBINLINE) == 0
2048 tcp_pulloutofband(so, th, m,
2049 drop_hdrlen); /* hdr drop is delayed */
2052 * If no out of band data is expected,
2053 * pull receive urgent pointer along
2054 * with the receive window.
2056 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2057 tp->rcv_up = tp->rcv_nxt;
2062 * Process the segment text, merging it into the TCP sequencing queue,
2063 * and arranging for acknowledgment of receipt if necessary.
2064 * This process logically involves adjusting tp->rcv_wnd as data
2065 * is presented to the user (this happens in tcp_usrreq.c,
2066 * case PRU_RCVD). If a FIN has already been received on this
2067 * connection then we just ignore the text.
2069 if ((tlen || (thflags & TH_FIN)) &&
2070 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2071 m_adj(m, drop_hdrlen); /* delayed header drop */
2073 * Insert segment which includes th into TCP reassembly queue
2074 * with control block tp. Set thflags to whether reassembly now
2075 * includes a segment with FIN. This handles the common case
2076 * inline (segment is the next to be received on an established
2077 * connection, and the queue is empty), avoiding linkage into
2078 * and removal from the queue and repetition of various
2080 * Set DELACK for segments received in order, but ack
2081 * immediately when segments are out of order (so
2082 * fast retransmit can work).
2084 if (th->th_seq == tp->rcv_nxt &&
2085 LIST_EMPTY(&tp->t_segq) &&
2086 TCPS_HAVEESTABLISHED(tp->t_state)) {
2088 callout_reset(tp->tt_delack, tcp_delacktime,
2089 tcp_timer_delack, tp);
2091 tp->t_flags |= TF_ACKNOW;
2092 tp->rcv_nxt += tlen;
2093 thflags = th->th_flags & TH_FIN;
2094 tcpstat.tcps_rcvpack++;
2095 tcpstat.tcps_rcvbyte += tlen;
2097 if (so->so_state & SS_CANTRCVMORE)
2100 sbappend(&so->so_rcv, m);
2103 thflags = tcp_reass(tp, th, &tlen, m);
2104 tp->t_flags |= TF_ACKNOW;
2108 * Note the amount of data that peer has sent into
2109 * our window, in order to estimate the sender's
2112 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2119 * If FIN is received ACK the FIN and let the user know
2120 * that the connection is closing.
2122 if (thflags & TH_FIN) {
2123 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2126 * If connection is half-synchronized
2127 * (ie NEEDSYN flag on) then delay ACK,
2128 * so it may be piggybacked when SYN is sent.
2129 * Otherwise, since we received a FIN then no
2130 * more input can be expected, send ACK now.
2132 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2133 callout_reset(tp->tt_delack, tcp_delacktime,
2134 tcp_timer_delack, tp);
2136 tp->t_flags |= TF_ACKNOW;
2139 switch (tp->t_state) {
2142 * In SYN_RECEIVED and ESTABLISHED STATES
2143 * enter the CLOSE_WAIT state.
2145 case TCPS_SYN_RECEIVED:
2146 tp->t_starttime = ticks;
2148 case TCPS_ESTABLISHED:
2149 tp->t_state = TCPS_CLOSE_WAIT;
2153 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2154 * enter the CLOSING state.
2156 case TCPS_FIN_WAIT_1:
2157 tp->t_state = TCPS_CLOSING;
2161 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2162 * starting the time-wait timer, turning off the other
2165 case TCPS_FIN_WAIT_2:
2166 tp->t_state = TCPS_TIME_WAIT;
2167 tcp_canceltimers(tp);
2168 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2169 if (tp->cc_recv != 0 &&
2170 (ticks - tp->t_starttime) < tcp_msl) {
2171 callout_reset(tp->tt_2msl,
2172 tp->t_rxtcur * TCPTV_TWTRUNC,
2173 tcp_timer_2msl, tp);
2174 /* For transaction client, force ACK now. */
2175 tp->t_flags |= TF_ACKNOW;
2178 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2179 tcp_timer_2msl, tp);
2180 soisdisconnected(so);
2184 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2186 case TCPS_TIME_WAIT:
2187 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2188 tcp_timer_2msl, tp);
2193 if (so->so_options & SO_DEBUG)
2194 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2199 * Return any desired output.
2201 if (needoutput || (tp->t_flags & TF_ACKNOW))
2202 (void) tcp_output(tp);
2207 * Generate an ACK dropping incoming segment if it occupies
2208 * sequence space, where the ACK reflects our state.
2210 * We can now skip the test for the RST flag since all
2211 * paths to this code happen after packets containing
2212 * RST have been dropped.
2214 * In the SYN-RECEIVED state, don't send an ACK unless the
2215 * segment we received passes the SYN-RECEIVED ACK test.
2216 * If it fails send a RST. This breaks the loop in the
2217 * "LAND" DoS attack, and also prevents an ACK storm
2218 * between two listening ports that have been sent forged
2219 * SYN segments, each with the source address of the other.
2221 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2222 (SEQ_GT(tp->snd_una, th->th_ack) ||
2223 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2224 rstreason = BANDLIM_RST_OPENPORT;
2228 if (so->so_options & SO_DEBUG)
2229 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2233 tp->t_flags |= TF_ACKNOW;
2234 (void) tcp_output(tp);
2239 * Generate a RST, dropping incoming segment.
2240 * Make ACK acceptable to originator of segment.
2241 * Don't bother to respond if destination was broadcast/multicast.
2243 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2246 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2247 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2250 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2251 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2252 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2253 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2256 /* IPv6 anycast check is done at tcp6_input() */
2259 * Perform bandwidth limiting.
2262 if (badport_bandlim(rstreason) < 0)
2267 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2268 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2271 if (thflags & TH_ACK)
2272 /* mtod() below is safe as long as hdr dropping is delayed */
2273 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2276 if (thflags & TH_SYN)
2278 /* mtod() below is safe as long as hdr dropping is delayed */
2279 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2280 (tcp_seq)0, TH_RST|TH_ACK);
2286 * Drop space held by incoming segment and return.
2289 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2290 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2298 * Parse TCP options and place in tcpopt.
2301 tcp_dooptions(to, cp, cnt, is_syn)
2309 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2311 if (opt == TCPOPT_EOL)
2313 if (opt == TCPOPT_NOP)
2319 if (optlen < 2 || optlen > cnt)
2324 if (optlen != TCPOLEN_MAXSEG)
2328 to->to_flags |= TOF_MSS;
2329 bcopy((char *)cp + 2,
2330 (char *)&to->to_mss, sizeof(to->to_mss));
2331 to->to_mss = ntohs(to->to_mss);
2334 if (optlen != TCPOLEN_WINDOW)
2338 to->to_flags |= TOF_SCALE;
2339 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2341 case TCPOPT_TIMESTAMP:
2342 if (optlen != TCPOLEN_TIMESTAMP)
2344 to->to_flags |= TOF_TS;
2345 bcopy((char *)cp + 2,
2346 (char *)&to->to_tsval, sizeof(to->to_tsval));
2347 to->to_tsval = ntohl(to->to_tsval);
2348 bcopy((char *)cp + 6,
2349 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2350 to->to_tsecr = ntohl(to->to_tsecr);
2353 if (optlen != TCPOLEN_CC)
2355 to->to_flags |= TOF_CC;
2356 bcopy((char *)cp + 2,
2357 (char *)&to->to_cc, sizeof(to->to_cc));
2358 to->to_cc = ntohl(to->to_cc);
2361 if (optlen != TCPOLEN_CC)
2365 to->to_flags |= TOF_CCNEW;
2366 bcopy((char *)cp + 2,
2367 (char *)&to->to_cc, sizeof(to->to_cc));
2368 to->to_cc = ntohl(to->to_cc);
2371 if (optlen != TCPOLEN_CC)
2375 to->to_flags |= TOF_CCECHO;
2376 bcopy((char *)cp + 2,
2377 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2378 to->to_ccecho = ntohl(to->to_ccecho);
2387 * Pull out of band byte out of a segment so
2388 * it doesn't appear in the user's data queue.
2389 * It is still reflected in the segment length for
2390 * sequencing purposes.
2393 tcp_pulloutofband(so, th, m, off)
2396 register struct mbuf *m;
2397 int off; /* delayed to be droped hdrlen */
2399 int cnt = off + th->th_urp - 1;
2402 if (m->m_len > cnt) {
2403 char *cp = mtod(m, caddr_t) + cnt;
2404 struct tcpcb *tp = sototcpcb(so);
2407 tp->t_oobflags |= TCPOOB_HAVEDATA;
2408 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2410 if (m->m_flags & M_PKTHDR)
2419 panic("tcp_pulloutofband");
2423 * Collect new round-trip time estimate
2424 * and update averages and current timeout.
2427 tcp_xmit_timer(tp, rtt)
2428 register struct tcpcb *tp;
2433 tcpstat.tcps_rttupdated++;
2435 if (tp->t_srtt != 0) {
2437 * srtt is stored as fixed point with 5 bits after the
2438 * binary point (i.e., scaled by 8). The following magic
2439 * is equivalent to the smoothing algorithm in rfc793 with
2440 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2441 * point). Adjust rtt to origin 0.
2443 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2444 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2446 if ((tp->t_srtt += delta) <= 0)
2450 * We accumulate a smoothed rtt variance (actually, a
2451 * smoothed mean difference), then set the retransmit
2452 * timer to smoothed rtt + 4 times the smoothed variance.
2453 * rttvar is stored as fixed point with 4 bits after the
2454 * binary point (scaled by 16). The following is
2455 * equivalent to rfc793 smoothing with an alpha of .75
2456 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2457 * rfc793's wired-in beta.
2461 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2462 if ((tp->t_rttvar += delta) <= 0)
2464 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2465 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2468 * No rtt measurement yet - use the unsmoothed rtt.
2469 * Set the variance to half the rtt (so our first
2470 * retransmit happens at 3*rtt).
2472 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2473 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2474 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2480 * the retransmit should happen at rtt + 4 * rttvar.
2481 * Because of the way we do the smoothing, srtt and rttvar
2482 * will each average +1/2 tick of bias. When we compute
2483 * the retransmit timer, we want 1/2 tick of rounding and
2484 * 1 extra tick because of +-1/2 tick uncertainty in the
2485 * firing of the timer. The bias will give us exactly the
2486 * 1.5 tick we need. But, because the bias is
2487 * statistical, we have to test that we don't drop below
2488 * the minimum feasible timer (which is 2 ticks).
2490 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2491 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2494 * We received an ack for a packet that wasn't retransmitted;
2495 * it is probably safe to discard any error indications we've
2496 * received recently. This isn't quite right, but close enough
2497 * for now (a route might have failed after we sent a segment,
2498 * and the return path might not be symmetrical).
2500 tp->t_softerror = 0;
2504 * Determine a reasonable value for maxseg size.
2505 * If the route is known, check route for mtu.
2506 * If none, use an mss that can be handled on the outgoing
2507 * interface without forcing IP to fragment; if bigger than
2508 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2509 * to utilize large mbufs. If no route is found, route has no mtu,
2510 * or the destination isn't local, use a default, hopefully conservative
2511 * size (usually 512 or the default IP max size, but no more than the mtu
2512 * of the interface), as we can't discover anything about intervening
2513 * gateways or networks. We also initialize the congestion/slow start
2514 * window to be a single segment if the destination isn't local.
2515 * While looking at the routing entry, we also initialize other path-dependent
2516 * parameters from pre-set or cached values in the routing entry.
2518 * Also take into account the space needed for options that we
2519 * send regularly. Make maxseg shorter by that amount to assure
2520 * that we can send maxseg amount of data even when the options
2521 * are present. Store the upper limit of the length of options plus
2524 * NOTE that this routine is only called when we process an incoming
2525 * segment, for outgoing segments only tcp_mssopt is called.
2527 * In case of T/TCP, we call this routine during implicit connection
2528 * setup as well (offer = -1), to initialize maxseg from the cached
2536 register struct rtentry *rt;
2538 register int rtt, mss;
2540 struct inpcb *inp = tp->t_inpcb;
2542 struct rmxp_tao *taop;
2543 int origoffer = offer;
2545 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2546 size_t min_protoh = isipv6 ?
2547 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2548 sizeof(struct tcpiphdr);
2550 const int isipv6 = 0;
2551 const size_t min_protoh = sizeof(struct tcpiphdr);
2555 rt = tcp_rtlookup6(&inp->inp_inc);
2557 rt = tcp_rtlookup(&inp->inp_inc);
2559 tp->t_maxopd = tp->t_maxseg =
2560 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2564 so = inp->inp_socket;
2566 taop = rmx_taop(rt->rt_rmx);
2568 * Offer == -1 means that we didn't receive SYN yet,
2569 * use cached value in that case;
2572 offer = taop->tao_mssopt;
2574 * Offer == 0 means that there was no MSS on the SYN segment,
2575 * in this case we use tcp_mssdflt.
2578 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2581 * Sanity check: make sure that maxopd will be large
2582 * enough to allow some data on segments even is the
2583 * all the option space is used (40bytes). Otherwise
2584 * funny things may happen in tcp_output.
2586 offer = max(offer, 64);
2587 taop->tao_mssopt = offer;
2590 * While we're here, check if there's an initial rtt
2591 * or rttvar. Convert from the route-table units
2592 * to scaled multiples of the slow timeout timer.
2594 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2596 * XXX the lock bit for RTT indicates that the value
2597 * is also a minimum value; this is subject to time.
2599 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2600 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2601 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2602 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2603 tcpstat.tcps_usedrtt++;
2604 if (rt->rt_rmx.rmx_rttvar) {
2605 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2606 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2607 tcpstat.tcps_usedrttvar++;
2609 /* default variation is +- 1 rtt */
2611 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2613 TCPT_RANGESET(tp->t_rxtcur,
2614 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2615 tp->t_rttmin, TCPTV_REXMTMAX);
2618 * if there's an mtu associated with the route, use it
2619 * else, use the link mtu.
2621 if (rt->rt_rmx.rmx_mtu)
2622 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2625 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2627 if (!in6_localaddr(&inp->in6p_faddr))
2628 mss = min(mss, tcp_v6mssdflt);
2630 mss = ifp->if_mtu - min_protoh;
2631 if (!in_localaddr(inp->inp_faddr))
2632 mss = min(mss, tcp_mssdflt);
2635 mss = min(mss, offer);
2637 * maxopd stores the maximum length of data AND options
2638 * in a segment; maxseg is the amount of data in a normal
2639 * segment. We need to store this value (maxopd) apart
2640 * from maxseg, because now every segment carries options
2641 * and thus we normally have somewhat less data in segments.
2646 * In case of T/TCP, origoffer==-1 indicates, that no segments
2647 * were received yet. In this case we just guess, otherwise
2648 * we do the same as before T/TCP.
2650 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2652 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2653 mss -= TCPOLEN_TSTAMP_APPA;
2654 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2656 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2657 mss -= TCPOLEN_CC_APPA;
2659 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2661 mss &= ~(MCLBYTES-1);
2664 mss = mss / MCLBYTES * MCLBYTES;
2667 * If there's a pipesize, change the socket buffer
2668 * to that size. Make the socket buffers an integral
2669 * number of mss units; if the mss is larger than
2670 * the socket buffer, decrease the mss.
2673 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2675 bufsize = so->so_snd.sb_hiwat;
2679 bufsize = roundup(bufsize, mss);
2680 if (bufsize > sb_max)
2682 if (bufsize > so->so_snd.sb_hiwat)
2683 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2688 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2690 bufsize = so->so_rcv.sb_hiwat;
2691 if (bufsize > mss) {
2692 bufsize = roundup(bufsize, mss);
2693 if (bufsize > sb_max)
2695 if (bufsize > so->so_rcv.sb_hiwat)
2696 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2700 * Set the slow-start flight size depending on whether this
2701 * is a local network or not.
2704 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2705 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2706 (!isipv6 && in_localaddr(inp->inp_faddr)))
2707 tp->snd_cwnd = mss * ss_fltsz_local;
2709 tp->snd_cwnd = mss * ss_fltsz;
2711 if (rt->rt_rmx.rmx_ssthresh) {
2713 * There's some sort of gateway or interface
2714 * buffer limit on the path. Use this to set
2715 * the slow start threshhold, but set the
2716 * threshold to no less than 2*mss.
2718 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2719 tcpstat.tcps_usedssthresh++;
2724 * Determine the MSS option to send on an outgoing SYN.
2732 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2733 int min_protoh = isipv6 ?
2734 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2735 sizeof(struct tcpiphdr);
2737 const int isipv6 = 0;
2738 const size_t min_protoh = sizeof(struct tcpiphdr);
2742 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2744 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2746 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2748 return (rt->rt_ifp->if_mtu - min_protoh);
2753 * When a partial ack arrives, force the retransmission of the
2754 * next unacknowledged segment. Do not clear tp->t_dupacks.
2755 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2759 tcp_newreno_partial_ack(tp, th)
2763 tcp_seq onxt = tp->snd_nxt;
2764 u_long ocwnd = tp->snd_cwnd;
2766 callout_stop(tp->tt_rexmt);
2768 tp->snd_nxt = th->th_ack;
2770 * Set snd_cwnd to one segment beyond acknowledged offset
2771 * (tp->snd_una has not yet been updated when this function is called.)
2773 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2774 tp->t_flags |= TF_ACKNOW;
2775 (void) tcp_output(tp);
2776 tp->snd_cwnd = ocwnd;
2777 if (SEQ_GT(onxt, tp->snd_nxt))
2780 * Partial window deflation. Relies on fact that tp->snd_una
2783 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);