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 1.17 2004/03/06 07:30:43 hsu Exp $
39 #include "opt_ipfw.h" /* for ipfw_fwd */
40 #include "opt_inet6.h"
41 #include "opt_ipsec.h"
42 #include "opt_tcpdebug.h"
43 #include "opt_tcp_input.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/sysctl.h>
49 #include <sys/malloc.h>
51 #include <sys/proc.h> /* for proc0 declaration */
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/syslog.h>
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 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
143 "TCP Segment Reassembly Queue");
145 int tcp_reass_maxseg = 0;
146 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
147 &tcp_reass_maxseg, 0,
148 "Global maximum number of TCP Segments in Reassembly Queue");
150 int tcp_reass_qsize = 0;
151 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
153 "Global number of TCP Segments currently in Reassembly Queue");
155 static int tcp_reass_overflows = 0;
156 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
157 &tcp_reass_overflows, 0,
158 "Global number of TCP Segment Reassembly Queue Overflows");
160 struct inpcbinfo tcbinfo;
162 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
163 static void tcp_pulloutofband(struct socket *,
164 struct tcphdr *, struct mbuf *, int);
165 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
167 static void tcp_xmit_timer(struct tcpcb *, int);
168 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
170 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
172 #define ND6_HINT(tp) \
174 if ((tp) && (tp)->t_inpcb && \
175 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
176 (tp)->t_inpcb->in6p_route.ro_rt) \
177 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
184 * Indicate whether this ack should be delayed. We can delay the ack if
185 * - delayed acks are enabled and
186 * - there is no delayed ack timer in progress and
187 * - our last ack wasn't a 0-sized window. We never want to delay
188 * the ack that opens up a 0-sized window.
190 #define DELAY_ACK(tp) \
191 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
192 (tp->t_flags & TF_RXWIN0SENT) == 0)
195 tcp_reass(tp, th, tlenp, m)
202 struct tseg_qent *p = NULL;
203 struct tseg_qent *nq;
204 struct tseg_qent *te;
205 struct socket *so = tp->t_inpcb->inp_socket;
209 * Call with th==0 after become established to
210 * force pre-ESTABLISHED data up to user socket.
216 * Limit the number of segments in the reassembly queue to prevent
217 * holding on to too many segments (and thus running out of mbufs).
218 * Make sure to let the missing segment through which caused this
219 * queue. Always keep one global queue entry spare to be able to
220 * process the missing segment.
222 if (th->th_seq != tp->rcv_nxt &&
223 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
224 tcp_reass_overflows++;
225 tcpstat.tcps_rcvmemdrop++;
230 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
231 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
234 tcpstat.tcps_rcvmemdrop++;
241 * Find a segment which begins after this one does.
243 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
244 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
250 * If there is a preceding segment, it may provide some of
251 * our data already. If so, drop the data from the incoming
252 * segment. If it provides all of our data, drop us.
256 /* conversion to int (in i) handles seq wraparound */
257 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
260 tcpstat.tcps_rcvduppack++;
261 tcpstat.tcps_rcvdupbyte += *tlenp;
266 * Try to present any queued data
267 * at the left window edge to the user.
268 * This is needed after the 3-WHS
271 goto present; /* ??? */
278 tcpstat.tcps_rcvoopack++;
279 tcpstat.tcps_rcvoobyte += *tlenp;
282 * While we overlap succeeding segments trim them or,
283 * if they are completely covered, dequeue them.
286 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
289 if (i < q->tqe_len) {
290 q->tqe_th->th_seq += i;
296 nq = LIST_NEXT(q, tqe_q);
297 LIST_REMOVE(q, tqe_q);
304 /* Insert the new segment queue entry into place. */
307 te->tqe_len = *tlenp;
310 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
312 LIST_INSERT_AFTER(p, te, tqe_q);
317 * Present data to user, advancing rcv_nxt through
318 * completed sequence space.
320 if (!TCPS_HAVEESTABLISHED(tp->t_state))
322 q = LIST_FIRST(&tp->t_segq);
323 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
326 tp->rcv_nxt += q->tqe_len;
327 flags = q->tqe_th->th_flags & TH_FIN;
328 nq = LIST_NEXT(q, tqe_q);
329 LIST_REMOVE(q, tqe_q);
330 if (so->so_state & SS_CANTRCVMORE)
333 sbappend(&so->so_rcv, q->tqe_m);
337 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
344 * TCP input routine, follows pages 65-76 of the
345 * protocol specification dated September, 1981 very closely.
349 tcp6_input(mp, offp, proto)
353 struct mbuf *m = *mp;
354 struct in6_ifaddr *ia6;
356 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
359 * draft-itojun-ipv6-tcp-to-anycast
360 * better place to put this in?
362 ia6 = ip6_getdstifaddr(m);
363 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
366 ip6 = mtod(m, struct ip6_hdr *);
367 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
368 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
372 tcp_input(m, *offp, proto);
378 tcp_input(m, off0, proto)
383 struct ip *ip = NULL;
385 struct inpcb *inp = NULL;
390 struct tcpcb *tp = NULL;
392 struct socket *so = 0;
393 int todrop, acked, ourfinisacked, needoutput = 0;
395 struct tcpopt to; /* options in this segment */
396 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
397 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
398 struct sockaddr_in *next_hop = NULL;
399 int rstreason; /* For badport_bandlim accounting purposes */
400 struct ip6_hdr *ip6 = NULL;
404 const int isipv6 = 0;
410 /* Grab info from MT_TAG mbufs prepended to the chain. */
411 for (;m && m->m_type == MT_TAG; m = m->m_next) {
412 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
413 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
416 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
418 bzero((char *)&to, sizeof(to));
420 tcpstat.tcps_rcvtotal++;
423 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
424 ip6 = mtod(m, struct ip6_hdr *);
425 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
426 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
427 tcpstat.tcps_rcvbadsum++;
430 th = (struct tcphdr *)((caddr_t)ip6 + off0);
433 * Be proactive about unspecified IPv6 address in source.
434 * As we use all-zero to indicate unbounded/unconnected pcb,
435 * unspecified IPv6 address can be used to confuse us.
437 * Note that packets with unspecified IPv6 destination is
438 * already dropped in ip6_input.
440 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
446 * Get IP and TCP header together in first mbuf.
447 * Note: IP leaves IP header in first mbuf.
449 if (off0 > sizeof(struct ip)) {
451 off0 = sizeof(struct ip);
453 if (m->m_len < sizeof(struct tcpiphdr)) {
454 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
455 tcpstat.tcps_rcvshort++;
459 ip = mtod(m, struct ip *);
460 ipov = (struct ipovly *)ip;
461 th = (struct tcphdr *)((caddr_t)ip + off0);
464 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
465 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
466 th->th_sum = m->m_pkthdr.csum_data;
468 th->th_sum = in_pseudo(ip->ip_src.s_addr,
470 htonl(m->m_pkthdr.csum_data +
473 th->th_sum ^= 0xffff;
476 * Checksum extended TCP header and data.
478 len = sizeof(struct ip) + tlen;
479 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
480 ipov->ih_len = (u_short)tlen;
481 ipov->ih_len = htons(ipov->ih_len);
482 th->th_sum = in_cksum(m, len);
485 tcpstat.tcps_rcvbadsum++;
489 /* Re-initialization for later version check */
490 ip->ip_v = IPVERSION;
495 * Check that TCP offset makes sense,
496 * pull out TCP options and adjust length. XXX
498 off = th->th_off << 2;
499 if (off < sizeof(struct tcphdr) || off > tlen) {
500 tcpstat.tcps_rcvbadoff++;
503 tlen -= off; /* tlen is used instead of ti->ti_len */
504 if (off > sizeof(struct tcphdr)) {
506 IP6_EXTHDR_CHECK(m, off0, off, );
507 ip6 = mtod(m, struct ip6_hdr *);
508 th = (struct tcphdr *)((caddr_t)ip6 + off0);
510 if (m->m_len < sizeof(struct ip) + off) {
511 if ((m = m_pullup(m, sizeof(struct ip) + off))
513 tcpstat.tcps_rcvshort++;
516 ip = mtod(m, struct ip *);
517 ipov = (struct ipovly *)ip;
518 th = (struct tcphdr *)((caddr_t)ip + off0);
521 optlen = off - sizeof(struct tcphdr);
522 optp = (u_char *)(th + 1);
524 thflags = th->th_flags;
526 #ifdef TCP_DROP_SYNFIN
528 * If the drop_synfin option is enabled, drop all packets with
529 * both the SYN and FIN bits set. This prevents e.g. nmap from
530 * identifying the TCP/IP stack.
532 * This is a violation of the TCP specification.
534 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
539 * Convert TCP protocol specific fields to host format.
541 th->th_seq = ntohl(th->th_seq);
542 th->th_ack = ntohl(th->th_ack);
543 th->th_win = ntohs(th->th_win);
544 th->th_urp = ntohs(th->th_urp);
547 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
548 * until after ip6_savecontrol() is called and before other functions
549 * which don't want those proto headers.
550 * Because ip6_savecontrol() is going to parse the mbuf to
551 * search for data to be passed up to user-land, it wants mbuf
552 * parameters to be unchanged.
553 * XXX: the call of ip6_savecontrol() has been obsoleted based on
554 * latest version of the advanced API (20020110).
556 drop_hdrlen = off0 + off;
559 * Locate pcb for segment.
562 /* IPFIREWALL_FORWARD section */
563 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
565 * Transparently forwarded. Pretend to be the destination.
566 * already got one like this?
568 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
569 ip->ip_dst, th->th_dport,
570 0, m->m_pkthdr.rcvif);
572 /* It's new. Try find the ambushing socket. */
573 inp = in_pcblookup_hash(&tcbinfo,
574 ip->ip_src, th->th_sport,
577 ntohs(next_hop->sin_port) :
579 1, m->m_pkthdr.rcvif);
583 inp = in6_pcblookup_hash(&tcbinfo,
584 &ip6->ip6_src, th->th_sport,
585 &ip6->ip6_dst, th->th_dport,
586 1, m->m_pkthdr.rcvif);
588 inp = in_pcblookup_hash(&tcbinfo,
589 ip->ip_src, th->th_sport,
590 ip->ip_dst, th->th_dport,
591 1, m->m_pkthdr.rcvif);
596 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
597 ipsec6stat.in_polvio++;
601 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
602 ipsecstat.in_polvio++;
609 if (inp != NULL && ipsec6_in_reject(m, inp)) {
613 if (inp != NULL && ipsec4_in_reject(m, inp)) {
620 * If the state is CLOSED (i.e., TCB does not exist) then
621 * all data in the incoming segment is discarded.
622 * If the TCB exists but is in CLOSED state, it is embryonic,
623 * but should either do a listen or a connect soon.
628 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
630 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
635 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
636 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
640 strcpy(dbuf, inet_ntoa(ip->ip_dst));
641 strcpy(sbuf, inet_ntoa(ip->ip_src));
643 switch (log_in_vain) {
645 if ((thflags & TH_SYN) == 0)
649 "Connection attempt to TCP %s:%d "
650 "from %s:%d flags:0x%02x\n",
651 dbuf, ntohs(th->th_dport), sbuf,
652 ntohs(th->th_sport), thflags);
661 if (thflags & TH_SYN)
670 rstreason = BANDLIM_RST_CLOSEDPORT;
675 rstreason = BANDLIM_RST_CLOSEDPORT;
678 if (tp->t_state == TCPS_CLOSED)
681 /* Unscale the window into a 32-bit value. */
682 if ((thflags & TH_SYN) == 0)
683 tiwin = th->th_win << tp->snd_scale;
687 so = inp->inp_socket;
690 if (so->so_options & SO_DEBUG) {
691 ostate = tp->t_state;
693 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
695 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
700 if (so->so_options & SO_ACCEPTCONN) {
701 struct in_conninfo inc;
704 inc.inc_isipv6 = isipv6;
707 inc.inc6_faddr = ip6->ip6_src;
708 inc.inc6_laddr = ip6->ip6_dst;
709 inc.inc6_route.ro_rt = NULL; /* XXX */
711 inc.inc_faddr = ip->ip_src;
712 inc.inc_laddr = ip->ip_dst;
713 inc.inc_route.ro_rt = NULL; /* XXX */
715 inc.inc_fport = th->th_sport;
716 inc.inc_lport = th->th_dport;
719 * If the state is LISTEN then ignore segment if it contains
720 * a RST. If the segment contains an ACK then it is bad and
721 * send a RST. If it does not contain a SYN then it is not
722 * interesting; drop it.
724 * If the state is SYN_RECEIVED (syncache) and seg contains
725 * an ACK, but not for our SYN/ACK, send a RST. If the seg
726 * contains a RST, check the sequence number to see if it
727 * is a valid reset segment.
729 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
730 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
731 if (!syncache_expand(&inc, th, &so, m)) {
733 * No syncache entry, or ACK was not
734 * for our SYN/ACK. Send a RST.
736 tcpstat.tcps_badsyn++;
737 rstreason = BANDLIM_RST_OPENPORT;
742 * Could not complete 3-way handshake,
743 * connection is being closed down, and
744 * syncache will free mbuf.
748 * Socket is created in state SYN_RECEIVED.
749 * Continue processing segment.
754 * This is what would have happened in
755 * tcp_output() when the SYN,ACK was sent.
757 tp->snd_up = tp->snd_una;
758 tp->snd_max = tp->snd_nxt = tp->iss + 1;
759 tp->last_ack_sent = tp->rcv_nxt;
761 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
762 * until the _second_ ACK is received:
763 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
764 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
765 * move to ESTAB, set snd_wnd to tiwin.
767 tp->snd_wnd = tiwin; /* unscaled */
770 if (thflags & TH_RST) {
771 syncache_chkrst(&inc, th);
774 if (thflags & TH_ACK) {
775 syncache_badack(&inc);
776 tcpstat.tcps_badsyn++;
777 rstreason = BANDLIM_RST_OPENPORT;
784 * Segment's flags are (SYN) or (SYN|FIN).
788 * If deprecated address is forbidden,
789 * we do not accept SYN to deprecated interface
790 * address to prevent any new inbound connection from
791 * getting established.
792 * When we do not accept SYN, we send a TCP RST,
793 * with deprecated source address (instead of dropping
794 * it). We compromise it as it is much better for peer
795 * to send a RST, and RST will be the final packet
798 * If we do not forbid deprecated addresses, we accept
799 * the SYN packet. RFC2462 does not suggest dropping
801 * If we decipher RFC2462 5.5.4, it says like this:
802 * 1. use of deprecated addr with existing
803 * communication is okay - "SHOULD continue to be
805 * 2. use of it with new communication:
806 * (2a) "SHOULD NOT be used if alternate address
807 * with sufficient scope is available"
808 * (2b) nothing mentioned otherwise.
809 * Here we fall into (2b) case as we have no choice in
810 * our source address selection - we must obey the peer.
812 * The wording in RFC2462 is confusing, and there are
813 * multiple description text for deprecated address
814 * handling - worse, they are not exactly the same.
815 * I believe 5.5.4 is the best one, so we follow 5.5.4.
817 if (isipv6 && !ip6_use_deprecated) {
818 struct in6_ifaddr *ia6;
820 if ((ia6 = ip6_getdstifaddr(m)) &&
821 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
823 rstreason = BANDLIM_RST_OPENPORT;
829 * If it is from this socket, drop it, it must be forged.
830 * Don't bother responding if the destination was a broadcast.
832 if (th->th_dport == th->th_sport) {
834 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
838 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
843 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
845 * Note that it is quite possible to receive unicast
846 * link-layer packets with a broadcast IP address. Use
847 * in_broadcast() to find them.
849 if (m->m_flags & (M_BCAST|M_MCAST))
852 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
853 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
856 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
857 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
858 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
859 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
863 * SYN appears to be valid; create compressed TCP state
864 * for syncache, or perform t/tcp connection.
866 if (so->so_qlen <= so->so_qlimit) {
867 tcp_dooptions(&to, optp, optlen, 1);
868 if (!syncache_add(&inc, &to, th, &so, m))
872 * Entry added to syncache, mbuf used to
873 * send SYN,ACK packet.
877 * Segment passed TAO tests.
882 tp->t_starttime = ticks;
883 tp->t_state = TCPS_ESTABLISHED;
886 * If there is a FIN, or if there is data and the
887 * connection is local, then delay SYN,ACK(SYN) in
888 * the hope of piggy-backing it on a response
889 * segment. Otherwise must send ACK now in case
890 * the other side is slow starting.
893 ((thflags & TH_FIN) ||
895 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
896 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
897 callout_reset(tp->tt_delack, tcp_delacktime,
898 tcp_timer_delack, tp);
899 tp->t_flags |= TF_NEEDSYN;
901 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
903 tcpstat.tcps_connects++;
911 /* XXX temp debugging */
912 /* should not happen - syncache should pick up these connections */
913 if (tp->t_state == TCPS_LISTEN)
914 panic("tcp_input: TCPS_LISTEN");
917 * Segment received on connection.
918 * Reset idle time and keep-alive timer.
920 tp->t_rcvtime = ticks;
921 if (TCPS_HAVEESTABLISHED(tp->t_state))
922 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
926 * XXX this is tradtitional behavior, may need to be cleaned up.
928 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
929 if (thflags & TH_SYN) {
930 if (to.to_flags & TOF_SCALE) {
931 tp->t_flags |= TF_RCVD_SCALE;
932 tp->requested_s_scale = to.to_requested_s_scale;
934 if (to.to_flags & TOF_TS) {
935 tp->t_flags |= TF_RCVD_TSTMP;
936 tp->ts_recent = to.to_tsval;
937 tp->ts_recent_age = ticks;
939 if (to.to_flags & (TOF_CC|TOF_CCNEW))
940 tp->t_flags |= TF_RCVD_CC;
941 if (to.to_flags & TOF_MSS)
942 tcp_mss(tp, to.to_mss);
946 * Header prediction: check for the two common cases
947 * of a uni-directional data xfer. If the packet has
948 * no control flags, is in-sequence, the window didn't
949 * change and we're not retransmitting, it's a
950 * candidate. If the length is zero and the ack moved
951 * forward, we're the sender side of the xfer. Just
952 * free the data acked & wake any higher level process
953 * that was blocked waiting for space. If the length
954 * is non-zero and the ack didn't move, we're the
955 * receiver side. If we're getting packets in-order
956 * (the reassembly queue is empty), add the data to
957 * the socket buffer and note that we need a delayed ack.
958 * Make sure that the hidden state-flags are also off.
959 * Since we check for TCPS_ESTABLISHED above, it can only
962 if (tp->t_state == TCPS_ESTABLISHED &&
963 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
964 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
965 ((to.to_flags & TOF_TS) == 0 ||
966 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
968 * Using the CC option is compulsory if once started:
969 * the segment is OK if no T/TCP was negotiated or
970 * if the segment has a CC option equal to CCrecv
972 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
973 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
974 th->th_seq == tp->rcv_nxt &&
975 tiwin && tiwin == tp->snd_wnd &&
976 tp->snd_nxt == tp->snd_max) {
979 * If last ACK falls within this segment's sequence numbers,
980 * record the timestamp.
981 * NOTE that the test is modified according to the latest
982 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
984 if ((to.to_flags & TOF_TS) != 0 &&
985 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
986 tp->ts_recent_age = ticks;
987 tp->ts_recent = to.to_tsval;
991 if (SEQ_GT(th->th_ack, tp->snd_una) &&
992 SEQ_LEQ(th->th_ack, tp->snd_max) &&
993 tp->snd_cwnd >= tp->snd_wnd &&
995 tp->t_dupacks < tcprexmtthresh) ||
996 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
998 * this is a pure ack for outstanding data.
1000 ++tcpstat.tcps_predack;
1002 * "bad retransmit" recovery
1004 * If Eifel detection applies, then
1005 * it is deterministic, so use it
1006 * unconditionally over the old heuristic.
1007 * Otherwise, fall back to the old heuristic.
1009 if (tcp_do_eifel_detect &&
1010 (to.to_flags & TOF_TS) && to.to_tsecr &&
1011 (tp->t_flags & TF_FIRSTACCACK)) {
1012 /* Eifel detection applicable. */
1013 if (to.to_tsecr < tp->t_rexmtTS) {
1014 tcp_revert_congestion_state(tp);
1015 ++tcpstat.tcps_eifeldetected;
1017 } else if (tp->t_rxtshift == 1 &&
1018 ticks < tp->t_badrxtwin) {
1019 tcp_revert_congestion_state(tp);
1020 ++tcpstat.tcps_rttdetected;
1022 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
1024 * Recalculate the retransmit timer / rtt.
1026 * Some machines (certain windows boxes)
1027 * send broken timestamp replies during the
1028 * SYN+ACK phase, ignore timestamps of 0.
1030 if ((to.to_flags & TOF_TS) != 0 &&
1033 ticks - to.to_tsecr + 1);
1034 } else if (tp->t_rtttime &&
1035 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1037 ticks - tp->t_rtttime);
1039 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1040 acked = th->th_ack - tp->snd_una;
1041 tcpstat.tcps_rcvackpack++;
1042 tcpstat.tcps_rcvackbyte += acked;
1043 sbdrop(&so->so_snd, acked);
1044 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1045 SEQ_LEQ(th->th_ack, tp->snd_recover))
1046 tp->snd_recover = th->th_ack - 1;
1047 tp->snd_una = th->th_ack;
1050 ND6_HINT(tp); /* some progress has been done */
1053 * If all outstanding data are acked, stop
1054 * retransmit timer, otherwise restart timer
1055 * using current (possibly backed-off) value.
1056 * If process is waiting for space,
1057 * wakeup/selwakeup/signal. If data
1058 * are ready to send, let tcp_output
1059 * decide between more output or persist.
1061 if (tp->snd_una == tp->snd_max)
1062 callout_stop(tp->tt_rexmt);
1063 else if (!callout_active(tp->tt_persist))
1064 callout_reset(tp->tt_rexmt,
1066 tcp_timer_rexmt, tp);
1069 if (so->so_snd.sb_cc)
1070 (void) tcp_output(tp);
1073 } else if (th->th_ack == tp->snd_una &&
1074 LIST_EMPTY(&tp->t_segq) &&
1075 tlen <= sbspace(&so->so_rcv)) {
1077 * this is a pure, in-sequence data packet
1078 * with nothing on the reassembly queue and
1079 * we have enough buffer space to take it.
1081 ++tcpstat.tcps_preddat;
1082 tp->rcv_nxt += tlen;
1083 tcpstat.tcps_rcvpack++;
1084 tcpstat.tcps_rcvbyte += tlen;
1085 ND6_HINT(tp); /* some progress has been done */
1087 * Add data to socket buffer.
1089 if (so->so_state & SS_CANTRCVMORE) {
1092 m_adj(m, drop_hdrlen); /* delayed header drop */
1093 sbappend(&so->so_rcv, m);
1096 if (DELAY_ACK(tp)) {
1097 callout_reset(tp->tt_delack, tcp_delacktime,
1098 tcp_timer_delack, tp);
1100 tp->t_flags |= TF_ACKNOW;
1108 * Calculate amount of space in receive window,
1109 * and then do TCP input processing.
1110 * Receive window is amount of space in rcv queue,
1111 * but not less than advertised window.
1115 win = sbspace(&so->so_rcv);
1118 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1121 switch (tp->t_state) {
1124 * If the state is SYN_RECEIVED:
1125 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1127 case TCPS_SYN_RECEIVED:
1128 if ((thflags & TH_ACK) &&
1129 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1130 SEQ_GT(th->th_ack, tp->snd_max))) {
1131 rstreason = BANDLIM_RST_OPENPORT;
1137 * If the state is SYN_SENT:
1138 * if seg contains an ACK, but not for our SYN, drop the input.
1139 * if seg contains a RST, then drop the connection.
1140 * if seg does not contain SYN, then drop it.
1141 * Otherwise this is an acceptable SYN segment
1142 * initialize tp->rcv_nxt and tp->irs
1143 * if seg contains ack then advance tp->snd_una
1144 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1145 * arrange for segment to be acked (eventually)
1146 * continue processing rest of data/controls, beginning with URG
1149 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1150 taop = &tao_noncached;
1151 bzero(taop, sizeof(*taop));
1154 if ((thflags & TH_ACK) &&
1155 (SEQ_LEQ(th->th_ack, tp->iss) ||
1156 SEQ_GT(th->th_ack, tp->snd_max))) {
1158 * If we have a cached CCsent for the remote host,
1159 * hence we haven't just crashed and restarted,
1160 * do not send a RST. This may be a retransmission
1161 * from the other side after our earlier ACK was lost.
1162 * Our new SYN, when it arrives, will serve as the
1165 if (taop->tao_ccsent != 0)
1168 rstreason = BANDLIM_UNLIMITED;
1172 if (thflags & TH_RST) {
1173 if (thflags & TH_ACK)
1174 tp = tcp_drop(tp, ECONNREFUSED);
1177 if ((thflags & TH_SYN) == 0)
1179 tp->snd_wnd = th->th_win; /* initial send window */
1180 tp->cc_recv = to.to_cc; /* foreign CC */
1182 tp->irs = th->th_seq;
1184 if (thflags & TH_ACK) {
1186 * Our SYN was acked. If segment contains CC.ECHO
1187 * option, check it to make sure this segment really
1188 * matches our SYN. If not, just drop it as old
1189 * duplicate, but send an RST if we're still playing
1190 * by the old rules. If no CC.ECHO option, make sure
1191 * we don't get fooled into using T/TCP.
1193 if (to.to_flags & TOF_CCECHO) {
1194 if (tp->cc_send != to.to_ccecho) {
1195 if (taop->tao_ccsent != 0)
1198 rstreason = BANDLIM_UNLIMITED;
1203 tp->t_flags &= ~TF_RCVD_CC;
1204 tcpstat.tcps_connects++;
1206 /* Do window scaling on this connection? */
1207 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1208 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1209 tp->snd_scale = tp->requested_s_scale;
1210 tp->rcv_scale = tp->request_r_scale;
1212 /* Segment is acceptable, update cache if undefined. */
1213 if (taop->tao_ccsent == 0)
1214 taop->tao_ccsent = to.to_ccecho;
1216 tp->rcv_adv += tp->rcv_wnd;
1217 tp->snd_una++; /* SYN is acked */
1219 * If there's data, delay ACK; if there's also a FIN
1220 * ACKNOW will be turned on later.
1222 if (DELAY_ACK(tp) && tlen != 0)
1223 callout_reset(tp->tt_delack, tcp_delacktime,
1224 tcp_timer_delack, tp);
1226 tp->t_flags |= TF_ACKNOW;
1228 * Received <SYN,ACK> in SYN_SENT[*] state.
1230 * SYN_SENT --> ESTABLISHED
1231 * SYN_SENT* --> FIN_WAIT_1
1233 tp->t_starttime = ticks;
1234 if (tp->t_flags & TF_NEEDFIN) {
1235 tp->t_state = TCPS_FIN_WAIT_1;
1236 tp->t_flags &= ~TF_NEEDFIN;
1239 tp->t_state = TCPS_ESTABLISHED;
1240 callout_reset(tp->tt_keep, tcp_keepidle,
1241 tcp_timer_keep, tp);
1245 * Received initial SYN in SYN-SENT[*] state =>
1246 * simultaneous open. If segment contains CC option
1247 * and there is a cached CC, apply TAO test.
1248 * If it succeeds, connection is * half-synchronized.
1249 * Otherwise, do 3-way handshake:
1250 * SYN-SENT -> SYN-RECEIVED
1251 * SYN-SENT* -> SYN-RECEIVED*
1252 * If there was no CC option, clear cached CC value.
1254 tp->t_flags |= TF_ACKNOW;
1255 callout_stop(tp->tt_rexmt);
1256 if (to.to_flags & TOF_CC) {
1257 if (taop->tao_cc != 0 &&
1258 CC_GT(to.to_cc, taop->tao_cc)) {
1260 * update cache and make transition:
1261 * SYN-SENT -> ESTABLISHED*
1262 * SYN-SENT* -> FIN-WAIT-1*
1264 taop->tao_cc = to.to_cc;
1265 tp->t_starttime = ticks;
1266 if (tp->t_flags & TF_NEEDFIN) {
1267 tp->t_state = TCPS_FIN_WAIT_1;
1268 tp->t_flags &= ~TF_NEEDFIN;
1270 tp->t_state = TCPS_ESTABLISHED;
1271 callout_reset(tp->tt_keep,
1276 tp->t_flags |= TF_NEEDSYN;
1278 tp->t_state = TCPS_SYN_RECEIVED;
1280 /* CC.NEW or no option => invalidate cache */
1282 tp->t_state = TCPS_SYN_RECEIVED;
1288 * Advance th->th_seq to correspond to first data byte.
1289 * If data, trim to stay within window,
1290 * dropping FIN if necessary.
1293 if (tlen > tp->rcv_wnd) {
1294 todrop = tlen - tp->rcv_wnd;
1298 tcpstat.tcps_rcvpackafterwin++;
1299 tcpstat.tcps_rcvbyteafterwin += todrop;
1301 tp->snd_wl1 = th->th_seq - 1;
1302 tp->rcv_up = th->th_seq;
1304 * Client side of transaction: already sent SYN and data.
1305 * If the remote host used T/TCP to validate the SYN,
1306 * our data will be ACK'd; if so, enter normal data segment
1307 * processing in the middle of step 5, ack processing.
1308 * Otherwise, goto step 6.
1310 if (thflags & TH_ACK)
1316 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1317 * if segment contains a SYN and CC [not CC.NEW] option:
1318 * if state == TIME_WAIT and connection duration > MSL,
1319 * drop packet and send RST;
1321 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1322 * ack the FIN (and data) in retransmission queue.
1323 * Complete close and delete TCPCB. Then reprocess
1324 * segment, hoping to find new TCPCB in LISTEN state;
1326 * else must be old SYN; drop it.
1327 * else do normal processing.
1331 case TCPS_TIME_WAIT:
1332 if ((thflags & TH_SYN) &&
1333 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1334 if (tp->t_state == TCPS_TIME_WAIT &&
1335 (ticks - tp->t_starttime) > tcp_msl) {
1336 rstreason = BANDLIM_UNLIMITED;
1339 if (CC_GT(to.to_cc, tp->cc_recv)) {
1346 break; /* continue normal processing */
1350 * States other than LISTEN or SYN_SENT.
1351 * First check the RST flag and sequence number since reset segments
1352 * are exempt from the timestamp and connection count tests. This
1353 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1354 * below which allowed reset segments in half the sequence space
1355 * to fall though and be processed (which gives forged reset
1356 * segments with a random sequence number a 50 percent chance of
1357 * killing a connection).
1358 * Then check timestamp, if present.
1359 * Then check the connection count, if present.
1360 * Then check that at least some bytes of segment are within
1361 * receive window. If segment begins before rcv_nxt,
1362 * drop leading data (and SYN); if nothing left, just ack.
1365 * If the RST bit is set, check the sequence number to see
1366 * if this is a valid reset segment.
1368 * In all states except SYN-SENT, all reset (RST) segments
1369 * are validated by checking their SEQ-fields. A reset is
1370 * valid if its sequence number is in the window.
1371 * Note: this does not take into account delayed ACKs, so
1372 * we should test against last_ack_sent instead of rcv_nxt.
1373 * The sequence number in the reset segment is normally an
1374 * echo of our outgoing acknowlegement numbers, but some hosts
1375 * send a reset with the sequence number at the rightmost edge
1376 * of our receive window, and we have to handle this case.
1377 * If we have multiple segments in flight, the intial reset
1378 * segment sequence numbers will be to the left of last_ack_sent,
1379 * but they will eventually catch up.
1380 * In any case, it never made sense to trim reset segments to
1381 * fit the receive window since RFC 1122 says:
1382 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1384 * A TCP SHOULD allow a received RST segment to include data.
1387 * It has been suggested that a RST segment could contain
1388 * ASCII text that encoded and explained the cause of the
1389 * RST. No standard has yet been established for such
1392 * If the reset segment passes the sequence number test examine
1394 * SYN_RECEIVED STATE:
1395 * If passive open, return to LISTEN state.
1396 * If active open, inform user that connection was refused.
1397 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1398 * Inform user that connection was reset, and close tcb.
1399 * CLOSING, LAST_ACK STATES:
1402 * Drop the segment - see Stevens, vol. 2, p. 964 and
1405 if (thflags & TH_RST) {
1406 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1407 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1408 switch (tp->t_state) {
1410 case TCPS_SYN_RECEIVED:
1411 so->so_error = ECONNREFUSED;
1414 case TCPS_ESTABLISHED:
1415 case TCPS_FIN_WAIT_1:
1416 case TCPS_FIN_WAIT_2:
1417 case TCPS_CLOSE_WAIT:
1418 so->so_error = ECONNRESET;
1420 tp->t_state = TCPS_CLOSED;
1421 tcpstat.tcps_drops++;
1430 case TCPS_TIME_WAIT:
1438 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1439 * and it's less than ts_recent, drop it.
1441 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1442 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1444 /* Check to see if ts_recent is over 24 days old. */
1445 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1447 * Invalidate ts_recent. If this segment updates
1448 * ts_recent, the age will be reset later and ts_recent
1449 * will get a valid value. If it does not, setting
1450 * ts_recent to zero will at least satisfy the
1451 * requirement that zero be placed in the timestamp
1452 * echo reply when ts_recent isn't valid. The
1453 * age isn't reset until we get a valid ts_recent
1454 * because we don't want out-of-order segments to be
1455 * dropped when ts_recent is old.
1459 tcpstat.tcps_rcvduppack++;
1460 tcpstat.tcps_rcvdupbyte += tlen;
1461 tcpstat.tcps_pawsdrop++;
1470 * If T/TCP was negotiated and the segment doesn't have CC,
1471 * or if its CC is wrong then drop the segment.
1472 * RST segments do not have to comply with this.
1474 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1475 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1479 * In the SYN-RECEIVED state, validate that the packet belongs to
1480 * this connection before trimming the data to fit the receive
1481 * window. Check the sequence number versus IRS since we know
1482 * the sequence numbers haven't wrapped. This is a partial fix
1483 * for the "LAND" DoS attack.
1485 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1486 rstreason = BANDLIM_RST_OPENPORT;
1490 todrop = tp->rcv_nxt - th->th_seq;
1492 if (thflags & TH_SYN) {
1502 * Following if statement from Stevens, vol. 2, p. 960.
1505 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1507 * Any valid FIN must be to the left of the window.
1508 * At this point the FIN must be a duplicate or out
1509 * of sequence; drop it.
1514 * Send an ACK to resynchronize and drop any data.
1515 * But keep on processing for RST or ACK.
1517 tp->t_flags |= TF_ACKNOW;
1519 tcpstat.tcps_rcvduppack++;
1520 tcpstat.tcps_rcvdupbyte += todrop;
1522 tcpstat.tcps_rcvpartduppack++;
1523 tcpstat.tcps_rcvpartdupbyte += todrop;
1525 drop_hdrlen += todrop; /* drop from the top afterwards */
1526 th->th_seq += todrop;
1528 if (th->th_urp > todrop)
1529 th->th_urp -= todrop;
1537 * If new data are received on a connection after the
1538 * user processes are gone, then RST the other end.
1540 if ((so->so_state & SS_NOFDREF) &&
1541 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1543 tcpstat.tcps_rcvafterclose++;
1544 rstreason = BANDLIM_UNLIMITED;
1549 * If segment ends after window, drop trailing data
1550 * (and PUSH and FIN); if nothing left, just ACK.
1552 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1554 tcpstat.tcps_rcvpackafterwin++;
1555 if (todrop >= tlen) {
1556 tcpstat.tcps_rcvbyteafterwin += tlen;
1558 * If a new connection request is received
1559 * while in TIME_WAIT, drop the old connection
1560 * and start over if the sequence numbers
1561 * are above the previous ones.
1563 if (thflags & TH_SYN &&
1564 tp->t_state == TCPS_TIME_WAIT &&
1565 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1570 * If window is closed can only take segments at
1571 * window edge, and have to drop data and PUSH from
1572 * incoming segments. Continue processing, but
1573 * remember to ack. Otherwise, drop segment
1576 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1577 tp->t_flags |= TF_ACKNOW;
1578 tcpstat.tcps_rcvwinprobe++;
1582 tcpstat.tcps_rcvbyteafterwin += todrop;
1585 thflags &= ~(TH_PUSH|TH_FIN);
1589 * If last ACK falls within this segment's sequence numbers,
1590 * record its timestamp.
1591 * NOTE that the test is modified according to the latest
1592 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1594 if ((to.to_flags & TOF_TS) != 0 &&
1595 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1596 tp->ts_recent_age = ticks;
1597 tp->ts_recent = to.to_tsval;
1601 * If a SYN is in the window, then this is an
1602 * error and we send an RST and drop the connection.
1604 if (thflags & TH_SYN) {
1605 tp = tcp_drop(tp, ECONNRESET);
1606 rstreason = BANDLIM_UNLIMITED;
1611 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1612 * flag is on (half-synchronized state), then queue data for
1613 * later processing; else drop segment and return.
1615 if ((thflags & TH_ACK) == 0) {
1616 if (tp->t_state == TCPS_SYN_RECEIVED ||
1617 (tp->t_flags & TF_NEEDSYN))
1626 switch (tp->t_state) {
1629 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1630 * ESTABLISHED state and continue processing.
1631 * The ACK was checked above.
1633 case TCPS_SYN_RECEIVED:
1635 tcpstat.tcps_connects++;
1637 /* Do window scaling? */
1638 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1639 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1640 tp->snd_scale = tp->requested_s_scale;
1641 tp->rcv_scale = tp->request_r_scale;
1644 * Upon successful completion of 3-way handshake,
1645 * update cache.CC if it was undefined, pass any queued
1646 * data to the user, and advance state appropriately.
1648 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1650 taop->tao_cc = tp->cc_recv;
1654 * SYN-RECEIVED -> ESTABLISHED
1655 * SYN-RECEIVED* -> FIN-WAIT-1
1657 tp->t_starttime = ticks;
1658 if (tp->t_flags & TF_NEEDFIN) {
1659 tp->t_state = TCPS_FIN_WAIT_1;
1660 tp->t_flags &= ~TF_NEEDFIN;
1662 tp->t_state = TCPS_ESTABLISHED;
1663 callout_reset(tp->tt_keep, tcp_keepidle,
1664 tcp_timer_keep, tp);
1667 * If segment contains data or ACK, will call tcp_reass()
1668 * later; if not, do so now to pass queued data to user.
1670 if (tlen == 0 && (thflags & TH_FIN) == 0)
1671 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1673 tp->snd_wl1 = th->th_seq - 1;
1677 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1678 * ACKs. If the ack is in the range
1679 * tp->snd_una < th->th_ack <= tp->snd_max
1680 * then advance tp->snd_una to th->th_ack and drop
1681 * data from the retransmission queue. If this ACK reflects
1682 * more up to date window information we update our window information.
1684 case TCPS_ESTABLISHED:
1685 case TCPS_FIN_WAIT_1:
1686 case TCPS_FIN_WAIT_2:
1687 case TCPS_CLOSE_WAIT:
1690 case TCPS_TIME_WAIT:
1692 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1693 if (tlen == 0 && tiwin == tp->snd_wnd) {
1694 tcpstat.tcps_rcvdupack++;
1696 * If we have outstanding data (other than
1697 * a window probe), this is a completely
1698 * duplicate ack (ie, window info didn't
1699 * change), the ack is the biggest we've
1700 * seen and we've seen exactly our rexmt
1701 * threshhold of them, assume a packet
1702 * has been dropped and retransmit it.
1703 * Kludge snd_nxt & the congestion
1704 * window so we send only this one
1707 * We know we're losing at the current
1708 * window size so do congestion avoidance
1709 * (set ssthresh to half the current window
1710 * and pull our congestion window back to
1711 * the new ssthresh).
1713 * Dup acks mean that packets have left the
1714 * network (they're now cached at the receiver)
1715 * so bump cwnd by the amount in the receiver
1716 * to keep a constant cwnd packets in the
1719 if (!callout_active(tp->tt_rexmt) ||
1720 th->th_ack != tp->snd_una)
1722 else if (++tp->t_dupacks > tcprexmtthresh ||
1724 IN_FASTRECOVERY(tp))) {
1725 tp->snd_cwnd += tp->t_maxseg;
1726 (void) tcp_output(tp);
1728 } else if (tp->t_dupacks == tcprexmtthresh) {
1729 tcp_seq onxt = tp->snd_nxt;
1731 if (tcp_do_newreno &&
1737 if (tcp_do_eifel_detect &&
1738 (tp->t_flags & TF_RCVD_TSTMP)) {
1739 tcp_save_congestion_state(tp);
1740 tp->t_flags |= TF_FASTREXMT;
1742 win = min(tp->snd_wnd, tp->snd_cwnd) /
1746 tp->snd_ssthresh = win * tp->t_maxseg;
1747 ENTER_FASTRECOVERY(tp);
1748 tp->snd_recover = tp->snd_max;
1749 callout_stop(tp->tt_rexmt);
1751 tp->snd_nxt = th->th_ack;
1752 tp->snd_cwnd = tp->t_maxseg;
1753 (void) tcp_output(tp);
1754 KASSERT(tp->snd_limited <= 2,
1755 ("tp->snd_limited too big"));
1756 tp->snd_cwnd = tp->snd_ssthresh +
1758 (tp->t_dupacks - tp->snd_limited));
1759 if (SEQ_GT(onxt, tp->snd_nxt))
1762 } else if (tcp_do_limitedtransmit) {
1763 u_long oldcwnd = tp->snd_cwnd;
1764 tcp_seq oldsndmax = tp->snd_max;
1767 KASSERT(tp->t_dupacks == 1 ||
1769 ("dupacks not 1 or 2"));
1770 if (tp->t_dupacks == 1)
1771 tp->snd_limited = 0;
1773 (tp->snd_nxt - tp->snd_una) +
1774 (tp->t_dupacks - tp->snd_limited) *
1776 (void) tcp_output(tp);
1777 sent = tp->snd_max - oldsndmax;
1778 if (sent > tp->t_maxseg) {
1779 KASSERT((tp->t_dupacks == 2 &&
1780 tp->snd_limited == 0) ||
1781 (sent == tp->t_maxseg + 1 &&
1782 tp->t_flags & TF_SENTFIN),
1784 tp->snd_limited = 2;
1785 } else if (sent > 0)
1787 tp->snd_cwnd = oldcwnd;
1795 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1798 * If the congestion window was inflated to account
1799 * for the other side's cached packets, retract it.
1801 if (tcp_do_newreno) {
1802 if (IN_FASTRECOVERY(tp)) {
1803 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1804 tcp_newreno_partial_ack(tp, th);
1807 * Window inflation should have left us
1808 * with approximately snd_ssthresh
1810 * But in case we would be inclined to
1811 * send a burst, better to do it via
1812 * the slow start mechanism.
1814 if (SEQ_GT(th->th_ack +
1817 tp->snd_cwnd = tp->snd_max -
1821 tp->snd_cwnd = tp->snd_ssthresh;
1825 if (tp->t_dupacks >= tcprexmtthresh &&
1826 tp->snd_cwnd > tp->snd_ssthresh)
1827 tp->snd_cwnd = tp->snd_ssthresh;
1830 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1831 tcpstat.tcps_rcvacktoomuch++;
1835 * If we reach this point, ACK is not a duplicate,
1836 * i.e., it ACKs something we sent.
1838 if (tp->t_flags & TF_NEEDSYN) {
1840 * T/TCP: Connection was half-synchronized, and our
1841 * SYN has been ACK'd (so connection is now fully
1842 * synchronized). Go to non-starred state,
1843 * increment snd_una for ACK of SYN, and check if
1844 * we can do window scaling.
1846 tp->t_flags &= ~TF_NEEDSYN;
1848 /* Do window scaling? */
1849 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1850 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1851 tp->snd_scale = tp->requested_s_scale;
1852 tp->rcv_scale = tp->request_r_scale;
1857 acked = th->th_ack - tp->snd_una;
1858 tcpstat.tcps_rcvackpack++;
1859 tcpstat.tcps_rcvackbyte += acked;
1862 * If we just performed our first retransmit, and the ACK
1863 * arrives within our recovery window, then it was a mistake
1864 * to do the retransmit in the first place. Recover our
1865 * original cwnd and ssthresh, and proceed to transmit where
1868 if (tcp_do_eifel_detect && acked &&
1869 (to.to_flags & TOF_TS) && to.to_tsecr &&
1870 (tp->t_flags & TF_FIRSTACCACK)) {
1871 /* Eifel detection applicable. */
1872 if (to.to_tsecr < tp->t_rexmtTS) {
1873 tcp_revert_congestion_state(tp);
1874 ++tcpstat.tcps_eifeldetected;
1876 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1877 tcp_revert_congestion_state(tp);
1878 ++tcpstat.tcps_rttdetected;
1882 * If we have a timestamp reply, update smoothed
1883 * round trip time. If no timestamp is present but
1884 * transmit timer is running and timed sequence
1885 * number was acked, update smoothed round trip time.
1886 * Since we now have an rtt measurement, cancel the
1887 * timer backoff (cf., Phil Karn's retransmit alg.).
1888 * Recompute the initial retransmit timer.
1890 * Some machines (certain windows boxes) send broken
1891 * timestamp replies during the SYN+ACK phase, ignore
1894 if ((to.to_flags & TOF_TS) != 0 &&
1896 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1897 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1898 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1900 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1903 * If all outstanding data is acked, stop retransmit
1904 * timer and remember to restart (more output or persist).
1905 * If there is more data to be acked, restart retransmit
1906 * timer, using current (possibly backed-off) value.
1908 if (th->th_ack == tp->snd_max) {
1909 callout_stop(tp->tt_rexmt);
1911 } else if (!callout_active(tp->tt_persist))
1912 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1913 tcp_timer_rexmt, tp);
1916 * If no data (only SYN) was ACK'd,
1917 * skip rest of ACK processing.
1922 /* Stop looking for an acceptable ACK since one was received. */
1923 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT);
1926 * When new data is acked, open the congestion window.
1927 * If the window gives us less than ssthresh packets
1928 * in flight, open exponentially (maxseg per packet).
1929 * Otherwise open linearly: maxseg per window
1930 * (maxseg^2 / cwnd per packet).
1932 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1933 u_int cw = tp->snd_cwnd;
1934 u_int incr = tp->t_maxseg;
1935 if (cw > tp->snd_ssthresh)
1936 incr = incr * incr / cw;
1937 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1939 if (acked > so->so_snd.sb_cc) {
1940 tp->snd_wnd -= so->so_snd.sb_cc;
1941 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1944 sbdrop(&so->so_snd, acked);
1945 tp->snd_wnd -= acked;
1949 /* detect una wraparound */
1950 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1951 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1952 SEQ_LEQ(th->th_ack, tp->snd_recover))
1953 tp->snd_recover = th->th_ack - 1;
1954 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1955 SEQ_GEQ(th->th_ack, tp->snd_recover))
1956 EXIT_FASTRECOVERY(tp);
1957 tp->snd_una = th->th_ack;
1958 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1959 tp->snd_nxt = tp->snd_una;
1961 switch (tp->t_state) {
1964 * In FIN_WAIT_1 STATE in addition to the processing
1965 * for the ESTABLISHED state if our FIN is now acknowledged
1966 * then enter FIN_WAIT_2.
1968 case TCPS_FIN_WAIT_1:
1969 if (ourfinisacked) {
1971 * If we can't receive any more
1972 * data, then closing user can proceed.
1973 * Starting the timer is contrary to the
1974 * specification, but if we don't get a FIN
1975 * we'll hang forever.
1977 if (so->so_state & SS_CANTRCVMORE) {
1978 soisdisconnected(so);
1979 callout_reset(tp->tt_2msl, tcp_maxidle,
1980 tcp_timer_2msl, tp);
1982 tp->t_state = TCPS_FIN_WAIT_2;
1987 * In CLOSING STATE in addition to the processing for
1988 * the ESTABLISHED state if the ACK acknowledges our FIN
1989 * then enter the TIME-WAIT state, otherwise ignore
1993 if (ourfinisacked) {
1994 tp->t_state = TCPS_TIME_WAIT;
1995 tcp_canceltimers(tp);
1996 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1997 if (tp->cc_recv != 0 &&
1998 (ticks - tp->t_starttime) < tcp_msl)
1999 callout_reset(tp->tt_2msl,
2002 tcp_timer_2msl, tp);
2004 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2005 tcp_timer_2msl, tp);
2006 soisdisconnected(so);
2011 * In LAST_ACK, we may still be waiting for data to drain
2012 * and/or to be acked, as well as for the ack of our FIN.
2013 * If our FIN is now acknowledged, delete the TCB,
2014 * enter the closed state and return.
2017 if (ourfinisacked) {
2024 * In TIME_WAIT state the only thing that should arrive
2025 * is a retransmission of the remote FIN. Acknowledge
2026 * it and restart the finack timer.
2028 case TCPS_TIME_WAIT:
2029 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2030 tcp_timer_2msl, tp);
2037 * Update window information.
2038 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2040 if ((thflags & TH_ACK) &&
2041 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2042 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2043 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2044 /* keep track of pure window updates */
2046 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2047 tcpstat.tcps_rcvwinupd++;
2048 tp->snd_wnd = tiwin;
2049 tp->snd_wl1 = th->th_seq;
2050 tp->snd_wl2 = th->th_ack;
2051 if (tp->snd_wnd > tp->max_sndwnd)
2052 tp->max_sndwnd = tp->snd_wnd;
2057 * Process segments with URG.
2059 if ((thflags & TH_URG) && th->th_urp &&
2060 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2062 * This is a kludge, but if we receive and accept
2063 * random urgent pointers, we'll crash in
2064 * soreceive. It's hard to imagine someone
2065 * actually wanting to send this much urgent data.
2067 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2068 th->th_urp = 0; /* XXX */
2069 thflags &= ~TH_URG; /* XXX */
2070 goto dodata; /* XXX */
2073 * If this segment advances the known urgent pointer,
2074 * then mark the data stream. This should not happen
2075 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2076 * a FIN has been received from the remote side.
2077 * In these states we ignore the URG.
2079 * According to RFC961 (Assigned Protocols),
2080 * the urgent pointer points to the last octet
2081 * of urgent data. We continue, however,
2082 * to consider it to indicate the first octet
2083 * of data past the urgent section as the original
2084 * spec states (in one of two places).
2086 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2087 tp->rcv_up = th->th_seq + th->th_urp;
2088 so->so_oobmark = so->so_rcv.sb_cc +
2089 (tp->rcv_up - tp->rcv_nxt) - 1;
2090 if (so->so_oobmark == 0)
2091 so->so_state |= SS_RCVATMARK;
2093 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2096 * Remove out of band data so doesn't get presented to user.
2097 * This can happen independent of advancing the URG pointer,
2098 * but if two URG's are pending at once, some out-of-band
2099 * data may creep in... ick.
2101 if (th->th_urp <= (u_long)tlen
2103 && (so->so_options & SO_OOBINLINE) == 0
2106 tcp_pulloutofband(so, th, m,
2107 drop_hdrlen); /* hdr drop is delayed */
2110 * If no out of band data is expected,
2111 * pull receive urgent pointer along
2112 * with the receive window.
2114 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2115 tp->rcv_up = tp->rcv_nxt;
2120 * Process the segment text, merging it into the TCP sequencing queue,
2121 * and arranging for acknowledgment of receipt if necessary.
2122 * This process logically involves adjusting tp->rcv_wnd as data
2123 * is presented to the user (this happens in tcp_usrreq.c,
2124 * case PRU_RCVD). If a FIN has already been received on this
2125 * connection then we just ignore the text.
2127 if ((tlen || (thflags & TH_FIN)) &&
2128 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2129 m_adj(m, drop_hdrlen); /* delayed header drop */
2131 * Insert segment which includes th into TCP reassembly queue
2132 * with control block tp. Set thflags to whether reassembly now
2133 * includes a segment with FIN. This handles the common case
2134 * inline (segment is the next to be received on an established
2135 * connection, and the queue is empty), avoiding linkage into
2136 * and removal from the queue and repetition of various
2138 * Set DELACK for segments received in order, but ack
2139 * immediately when segments are out of order (so
2140 * fast retransmit can work).
2142 if (th->th_seq == tp->rcv_nxt &&
2143 LIST_EMPTY(&tp->t_segq) &&
2144 TCPS_HAVEESTABLISHED(tp->t_state)) {
2146 callout_reset(tp->tt_delack, tcp_delacktime,
2147 tcp_timer_delack, tp);
2149 tp->t_flags |= TF_ACKNOW;
2150 tp->rcv_nxt += tlen;
2151 thflags = th->th_flags & TH_FIN;
2152 tcpstat.tcps_rcvpack++;
2153 tcpstat.tcps_rcvbyte += tlen;
2155 if (so->so_state & SS_CANTRCVMORE)
2158 sbappend(&so->so_rcv, m);
2161 thflags = tcp_reass(tp, th, &tlen, m);
2162 tp->t_flags |= TF_ACKNOW;
2166 * Note the amount of data that peer has sent into
2167 * our window, in order to estimate the sender's
2170 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2177 * If FIN is received ACK the FIN and let the user know
2178 * that the connection is closing.
2180 if (thflags & TH_FIN) {
2181 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2184 * If connection is half-synchronized
2185 * (ie NEEDSYN flag on) then delay ACK,
2186 * so it may be piggybacked when SYN is sent.
2187 * Otherwise, since we received a FIN then no
2188 * more input can be expected, send ACK now.
2190 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2191 callout_reset(tp->tt_delack, tcp_delacktime,
2192 tcp_timer_delack, tp);
2194 tp->t_flags |= TF_ACKNOW;
2197 switch (tp->t_state) {
2200 * In SYN_RECEIVED and ESTABLISHED STATES
2201 * enter the CLOSE_WAIT state.
2203 case TCPS_SYN_RECEIVED:
2204 tp->t_starttime = ticks;
2206 case TCPS_ESTABLISHED:
2207 tp->t_state = TCPS_CLOSE_WAIT;
2211 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2212 * enter the CLOSING state.
2214 case TCPS_FIN_WAIT_1:
2215 tp->t_state = TCPS_CLOSING;
2219 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2220 * starting the time-wait timer, turning off the other
2223 case TCPS_FIN_WAIT_2:
2224 tp->t_state = TCPS_TIME_WAIT;
2225 tcp_canceltimers(tp);
2226 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2227 if (tp->cc_recv != 0 &&
2228 (ticks - tp->t_starttime) < tcp_msl) {
2229 callout_reset(tp->tt_2msl,
2230 tp->t_rxtcur * TCPTV_TWTRUNC,
2231 tcp_timer_2msl, tp);
2232 /* For transaction client, force ACK now. */
2233 tp->t_flags |= TF_ACKNOW;
2236 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2237 tcp_timer_2msl, tp);
2238 soisdisconnected(so);
2242 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2244 case TCPS_TIME_WAIT:
2245 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2246 tcp_timer_2msl, tp);
2251 if (so->so_options & SO_DEBUG)
2252 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2257 * Return any desired output.
2259 if (needoutput || (tp->t_flags & TF_ACKNOW))
2260 (void) tcp_output(tp);
2265 * Generate an ACK dropping incoming segment if it occupies
2266 * sequence space, where the ACK reflects our state.
2268 * We can now skip the test for the RST flag since all
2269 * paths to this code happen after packets containing
2270 * RST have been dropped.
2272 * In the SYN-RECEIVED state, don't send an ACK unless the
2273 * segment we received passes the SYN-RECEIVED ACK test.
2274 * If it fails send a RST. This breaks the loop in the
2275 * "LAND" DoS attack, and also prevents an ACK storm
2276 * between two listening ports that have been sent forged
2277 * SYN segments, each with the source address of the other.
2279 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2280 (SEQ_GT(tp->snd_una, th->th_ack) ||
2281 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2282 rstreason = BANDLIM_RST_OPENPORT;
2286 if (so->so_options & SO_DEBUG)
2287 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2291 tp->t_flags |= TF_ACKNOW;
2292 (void) tcp_output(tp);
2297 * Generate a RST, dropping incoming segment.
2298 * Make ACK acceptable to originator of segment.
2299 * Don't bother to respond if destination was broadcast/multicast.
2301 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2304 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2305 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2308 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2309 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2310 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2311 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2314 /* IPv6 anycast check is done at tcp6_input() */
2317 * Perform bandwidth limiting.
2320 if (badport_bandlim(rstreason) < 0)
2325 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2326 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2329 if (thflags & TH_ACK)
2330 /* mtod() below is safe as long as hdr dropping is delayed */
2331 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2334 if (thflags & TH_SYN)
2336 /* mtod() below is safe as long as hdr dropping is delayed */
2337 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2338 (tcp_seq)0, TH_RST|TH_ACK);
2344 * Drop space held by incoming segment and return.
2347 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2348 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2356 * Parse TCP options and place in tcpopt.
2359 tcp_dooptions(to, cp, cnt, is_syn)
2367 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2369 if (opt == TCPOPT_EOL)
2371 if (opt == TCPOPT_NOP)
2377 if (optlen < 2 || optlen > cnt)
2382 if (optlen != TCPOLEN_MAXSEG)
2386 to->to_flags |= TOF_MSS;
2387 bcopy((char *)cp + 2,
2388 (char *)&to->to_mss, sizeof(to->to_mss));
2389 to->to_mss = ntohs(to->to_mss);
2392 if (optlen != TCPOLEN_WINDOW)
2396 to->to_flags |= TOF_SCALE;
2397 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2399 case TCPOPT_TIMESTAMP:
2400 if (optlen != TCPOLEN_TIMESTAMP)
2402 to->to_flags |= TOF_TS;
2403 bcopy((char *)cp + 2,
2404 (char *)&to->to_tsval, sizeof(to->to_tsval));
2405 to->to_tsval = ntohl(to->to_tsval);
2406 bcopy((char *)cp + 6,
2407 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2408 to->to_tsecr = ntohl(to->to_tsecr);
2411 if (optlen != TCPOLEN_CC)
2413 to->to_flags |= TOF_CC;
2414 bcopy((char *)cp + 2,
2415 (char *)&to->to_cc, sizeof(to->to_cc));
2416 to->to_cc = ntohl(to->to_cc);
2419 if (optlen != TCPOLEN_CC)
2423 to->to_flags |= TOF_CCNEW;
2424 bcopy((char *)cp + 2,
2425 (char *)&to->to_cc, sizeof(to->to_cc));
2426 to->to_cc = ntohl(to->to_cc);
2429 if (optlen != TCPOLEN_CC)
2433 to->to_flags |= TOF_CCECHO;
2434 bcopy((char *)cp + 2,
2435 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2436 to->to_ccecho = ntohl(to->to_ccecho);
2445 * Pull out of band byte out of a segment so
2446 * it doesn't appear in the user's data queue.
2447 * It is still reflected in the segment length for
2448 * sequencing purposes.
2451 tcp_pulloutofband(so, th, m, off)
2455 int off; /* delayed to be droped hdrlen */
2457 int cnt = off + th->th_urp - 1;
2460 if (m->m_len > cnt) {
2461 char *cp = mtod(m, caddr_t) + cnt;
2462 struct tcpcb *tp = sototcpcb(so);
2465 tp->t_oobflags |= TCPOOB_HAVEDATA;
2466 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2468 if (m->m_flags & M_PKTHDR)
2477 panic("tcp_pulloutofband");
2481 * Collect new round-trip time estimate
2482 * and update averages and current timeout.
2485 tcp_xmit_timer(tp, rtt)
2491 tcpstat.tcps_rttupdated++;
2493 if (tp->t_srtt != 0) {
2495 * srtt is stored as fixed point with 5 bits after the
2496 * binary point (i.e., scaled by 8). The following magic
2497 * is equivalent to the smoothing algorithm in rfc793 with
2498 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2499 * point). Adjust rtt to origin 0.
2501 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2502 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2504 if ((tp->t_srtt += delta) <= 0)
2508 * We accumulate a smoothed rtt variance (actually, a
2509 * smoothed mean difference), then set the retransmit
2510 * timer to smoothed rtt + 4 times the smoothed variance.
2511 * rttvar is stored as fixed point with 4 bits after the
2512 * binary point (scaled by 16). The following is
2513 * equivalent to rfc793 smoothing with an alpha of .75
2514 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2515 * rfc793's wired-in beta.
2519 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2520 if ((tp->t_rttvar += delta) <= 0)
2522 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2523 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2526 * No rtt measurement yet - use the unsmoothed rtt.
2527 * Set the variance to half the rtt (so our first
2528 * retransmit happens at 3*rtt).
2530 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2531 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2532 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2538 * the retransmit should happen at rtt + 4 * rttvar.
2539 * Because of the way we do the smoothing, srtt and rttvar
2540 * will each average +1/2 tick of bias. When we compute
2541 * the retransmit timer, we want 1/2 tick of rounding and
2542 * 1 extra tick because of +-1/2 tick uncertainty in the
2543 * firing of the timer. The bias will give us exactly the
2544 * 1.5 tick we need. But, because the bias is
2545 * statistical, we have to test that we don't drop below
2546 * the minimum feasible timer (which is 2 ticks).
2548 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2549 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2552 * We received an ack for a packet that wasn't retransmitted;
2553 * it is probably safe to discard any error indications we've
2554 * received recently. This isn't quite right, but close enough
2555 * for now (a route might have failed after we sent a segment,
2556 * and the return path might not be symmetrical).
2558 tp->t_softerror = 0;
2562 * Determine a reasonable value for maxseg size.
2563 * If the route is known, check route for mtu.
2564 * If none, use an mss that can be handled on the outgoing
2565 * interface without forcing IP to fragment; if bigger than
2566 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2567 * to utilize large mbufs. If no route is found, route has no mtu,
2568 * or the destination isn't local, use a default, hopefully conservative
2569 * size (usually 512 or the default IP max size, but no more than the mtu
2570 * of the interface), as we can't discover anything about intervening
2571 * gateways or networks. We also initialize the congestion/slow start
2572 * window to be a single segment if the destination isn't local.
2573 * While looking at the routing entry, we also initialize other path-dependent
2574 * parameters from pre-set or cached values in the routing entry.
2576 * Also take into account the space needed for options that we
2577 * send regularly. Make maxseg shorter by that amount to assure
2578 * that we can send maxseg amount of data even when the options
2579 * are present. Store the upper limit of the length of options plus
2582 * NOTE that this routine is only called when we process an incoming
2583 * segment, for outgoing segments only tcp_mssopt is called.
2585 * In case of T/TCP, we call this routine during implicit connection
2586 * setup as well (offer = -1), to initialize maxseg from the cached
2598 struct inpcb *inp = tp->t_inpcb;
2600 struct rmxp_tao *taop;
2601 int origoffer = offer;
2603 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2604 size_t min_protoh = isipv6 ?
2605 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2606 sizeof(struct tcpiphdr);
2608 const int isipv6 = 0;
2609 const size_t min_protoh = sizeof(struct tcpiphdr);
2613 rt = tcp_rtlookup6(&inp->inp_inc);
2615 rt = tcp_rtlookup(&inp->inp_inc);
2617 tp->t_maxopd = tp->t_maxseg =
2618 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2622 so = inp->inp_socket;
2624 taop = rmx_taop(rt->rt_rmx);
2626 * Offer == -1 means that we didn't receive SYN yet,
2627 * use cached value in that case;
2630 offer = taop->tao_mssopt;
2632 * Offer == 0 means that there was no MSS on the SYN segment,
2633 * in this case we use tcp_mssdflt.
2636 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2639 * Sanity check: make sure that maxopd will be large
2640 * enough to allow some data on segments even is the
2641 * all the option space is used (40bytes). Otherwise
2642 * funny things may happen in tcp_output.
2644 offer = max(offer, 64);
2645 taop->tao_mssopt = offer;
2648 * While we're here, check if there's an initial rtt
2649 * or rttvar. Convert from the route-table units
2650 * to scaled multiples of the slow timeout timer.
2652 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2654 * XXX the lock bit for RTT indicates that the value
2655 * is also a minimum value; this is subject to time.
2657 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2658 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2659 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2660 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2661 tcpstat.tcps_usedrtt++;
2662 if (rt->rt_rmx.rmx_rttvar) {
2663 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2664 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2665 tcpstat.tcps_usedrttvar++;
2667 /* default variation is +- 1 rtt */
2669 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2671 TCPT_RANGESET(tp->t_rxtcur,
2672 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2673 tp->t_rttmin, TCPTV_REXMTMAX);
2676 * if there's an mtu associated with the route, use it
2677 * else, use the link mtu.
2679 if (rt->rt_rmx.rmx_mtu)
2680 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2683 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2685 if (!in6_localaddr(&inp->in6p_faddr))
2686 mss = min(mss, tcp_v6mssdflt);
2688 mss = ifp->if_mtu - min_protoh;
2689 if (!in_localaddr(inp->inp_faddr))
2690 mss = min(mss, tcp_mssdflt);
2693 mss = min(mss, offer);
2695 * maxopd stores the maximum length of data AND options
2696 * in a segment; maxseg is the amount of data in a normal
2697 * segment. We need to store this value (maxopd) apart
2698 * from maxseg, because now every segment carries options
2699 * and thus we normally have somewhat less data in segments.
2704 * In case of T/TCP, origoffer==-1 indicates, that no segments
2705 * were received yet. In this case we just guess, otherwise
2706 * we do the same as before T/TCP.
2708 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2710 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2711 mss -= TCPOLEN_TSTAMP_APPA;
2712 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2714 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2715 mss -= TCPOLEN_CC_APPA;
2717 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2719 mss &= ~(MCLBYTES-1);
2722 mss = mss / MCLBYTES * MCLBYTES;
2725 * If there's a pipesize, change the socket buffer
2726 * to that size. Make the socket buffers an integral
2727 * number of mss units; if the mss is larger than
2728 * the socket buffer, decrease the mss.
2731 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2733 bufsize = so->so_snd.sb_hiwat;
2737 bufsize = roundup(bufsize, mss);
2738 if (bufsize > sb_max)
2740 if (bufsize > so->so_snd.sb_hiwat)
2741 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2746 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2748 bufsize = so->so_rcv.sb_hiwat;
2749 if (bufsize > mss) {
2750 bufsize = roundup(bufsize, mss);
2751 if (bufsize > sb_max)
2753 if (bufsize > so->so_rcv.sb_hiwat)
2754 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2758 * Set the slow-start flight size depending on whether this
2759 * is a local network or not.
2762 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2763 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2764 (!isipv6 && in_localaddr(inp->inp_faddr)))
2765 tp->snd_cwnd = mss * ss_fltsz_local;
2767 tp->snd_cwnd = mss * ss_fltsz;
2769 if (rt->rt_rmx.rmx_ssthresh) {
2771 * There's some sort of gateway or interface
2772 * buffer limit on the path. Use this to set
2773 * the slow start threshhold, but set the
2774 * threshold to no less than 2*mss.
2776 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2777 tcpstat.tcps_usedssthresh++;
2782 * Determine the MSS option to send on an outgoing SYN.
2790 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2791 int min_protoh = isipv6 ?
2792 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2793 sizeof(struct tcpiphdr);
2795 const int isipv6 = 0;
2796 const size_t min_protoh = sizeof(struct tcpiphdr);
2800 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2802 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2804 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2806 return (rt->rt_ifp->if_mtu - min_protoh);
2811 * When a partial ack arrives, force the retransmission of the
2812 * next unacknowledged segment. Do not clear tp->t_dupacks.
2813 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2817 tcp_newreno_partial_ack(tp, th)
2821 tcp_seq onxt = tp->snd_nxt;
2822 u_long ocwnd = tp->snd_cwnd;
2824 callout_stop(tp->tt_rexmt);
2826 tp->snd_nxt = th->th_ack;
2828 * Set snd_cwnd to one segment beyond acknowledged offset
2829 * (tp->snd_una has not yet been updated when this function is called.)
2831 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2832 tp->t_flags |= TF_ACKNOW;
2833 (void) tcp_output(tp);
2834 tp->snd_cwnd = ocwnd;
2835 if (SEQ_GT(onxt, tp->snd_nxt))
2838 * Partial window deflation. Relies on fact that tp->snd_una
2841 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);