2 * Copyright (c) 2002-2004 Jeffrey Hsu. All rights reserved.
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 1.19 2004/03/08 19:44:32 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_early_retransmit = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
135 &tcp_do_early_retransmit, 0, "Early retransmit");
137 static int tcp_do_rfc3390 = 1;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
140 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
142 static int tcp_do_eifel_detect = 1;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
144 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
147 "TCP Segment Reassembly Queue");
149 int tcp_reass_maxseg = 0;
150 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
151 &tcp_reass_maxseg, 0,
152 "Global maximum number of TCP Segments in Reassembly Queue");
154 int tcp_reass_qsize = 0;
155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
157 "Global number of TCP Segments currently in Reassembly Queue");
159 static int tcp_reass_overflows = 0;
160 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
161 &tcp_reass_overflows, 0,
162 "Global number of TCP Segment Reassembly Queue Overflows");
164 struct inpcbinfo tcbinfo[MAXCPU];
166 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
167 static void tcp_pulloutofband(struct socket *,
168 struct tcphdr *, struct mbuf *, int);
169 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
171 static void tcp_xmit_timer(struct tcpcb *, int);
172 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
174 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
176 #define ND6_HINT(tp) \
178 if ((tp) && (tp)->t_inpcb && \
179 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
180 (tp)->t_inpcb->in6p_route.ro_rt) \
181 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
188 * Indicate whether this ack should be delayed. We can delay the ack if
189 * - delayed acks are enabled and
190 * - there is no delayed ack timer in progress and
191 * - our last ack wasn't a 0-sized window. We never want to delay
192 * the ack that opens up a 0-sized window.
194 #define DELAY_ACK(tp) \
195 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
196 (tp->t_flags & TF_RXWIN0SENT) == 0)
199 tcp_reass(tp, th, tlenp, m)
206 struct tseg_qent *p = NULL;
207 struct tseg_qent *nq;
208 struct tseg_qent *te;
209 struct socket *so = tp->t_inpcb->inp_socket;
213 * Call with th==0 after become established to
214 * force pre-ESTABLISHED data up to user socket.
220 * Limit the number of segments in the reassembly queue to prevent
221 * holding on to too many segments (and thus running out of mbufs).
222 * Make sure to let the missing segment through which caused this
223 * queue. Always keep one global queue entry spare to be able to
224 * process the missing segment.
226 if (th->th_seq != tp->rcv_nxt &&
227 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
228 tcp_reass_overflows++;
229 tcpstat.tcps_rcvmemdrop++;
234 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
235 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
238 tcpstat.tcps_rcvmemdrop++;
245 * Find a segment which begins after this one does.
247 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
248 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
254 * If there is a preceding segment, it may provide some of
255 * our data already. If so, drop the data from the incoming
256 * segment. If it provides all of our data, drop us.
260 /* conversion to int (in i) handles seq wraparound */
261 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
264 tcpstat.tcps_rcvduppack++;
265 tcpstat.tcps_rcvdupbyte += *tlenp;
270 * Try to present any queued data
271 * at the left window edge to the user.
272 * This is needed after the 3-WHS
275 goto present; /* ??? */
282 tcpstat.tcps_rcvoopack++;
283 tcpstat.tcps_rcvoobyte += *tlenp;
286 * While we overlap succeeding segments trim them or,
287 * if they are completely covered, dequeue them.
290 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
293 if (i < q->tqe_len) {
294 q->tqe_th->th_seq += i;
300 nq = LIST_NEXT(q, tqe_q);
301 LIST_REMOVE(q, tqe_q);
308 /* Insert the new segment queue entry into place. */
311 te->tqe_len = *tlenp;
314 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
316 LIST_INSERT_AFTER(p, te, tqe_q);
321 * Present data to user, advancing rcv_nxt through
322 * completed sequence space.
324 if (!TCPS_HAVEESTABLISHED(tp->t_state))
326 q = LIST_FIRST(&tp->t_segq);
327 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
330 tp->rcv_nxt += q->tqe_len;
331 flags = q->tqe_th->th_flags & TH_FIN;
332 nq = LIST_NEXT(q, tqe_q);
333 LIST_REMOVE(q, tqe_q);
334 if (so->so_state & SS_CANTRCVMORE)
337 sbappend(&so->so_rcv, q->tqe_m);
341 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
348 * TCP input routine, follows pages 65-76 of the
349 * protocol specification dated September, 1981 very closely.
353 tcp6_input(mp, offp, proto)
357 struct mbuf *m = *mp;
358 struct in6_ifaddr *ia6;
360 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
363 * draft-itojun-ipv6-tcp-to-anycast
364 * better place to put this in?
366 ia6 = ip6_getdstifaddr(m);
367 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
370 ip6 = mtod(m, struct ip6_hdr *);
371 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
372 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
376 tcp_input(m, *offp, proto);
382 tcp_input(m, off0, proto)
387 struct ip *ip = NULL;
389 struct inpcb *inp = NULL;
394 struct tcpcb *tp = NULL;
396 struct socket *so = 0;
397 int todrop, acked, ourfinisacked, needoutput = 0;
399 struct tcpopt to; /* options in this segment */
400 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
401 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
402 struct sockaddr_in *next_hop = NULL;
403 int rstreason; /* For badport_bandlim accounting purposes */
405 struct ip6_hdr *ip6 = NULL;
409 const int isipv6 = 0;
415 /* Grab info from MT_TAG mbufs prepended to the chain. */
416 for (;m && m->m_type == MT_TAG; m = m->m_next) {
417 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
418 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
421 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
423 bzero((char *)&to, sizeof(to));
425 tcpstat.tcps_rcvtotal++;
428 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
429 ip6 = mtod(m, struct ip6_hdr *);
430 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
431 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
432 tcpstat.tcps_rcvbadsum++;
435 th = (struct tcphdr *)((caddr_t)ip6 + off0);
438 * Be proactive about unspecified IPv6 address in source.
439 * As we use all-zero to indicate unbounded/unconnected pcb,
440 * unspecified IPv6 address can be used to confuse us.
442 * Note that packets with unspecified IPv6 destination is
443 * already dropped in ip6_input.
445 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
451 * Get IP and TCP header together in first mbuf.
452 * Note: IP leaves IP header in first mbuf.
454 if (off0 > sizeof(struct ip)) {
456 off0 = sizeof(struct ip);
458 if (m->m_len < sizeof(struct tcpiphdr)) {
459 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
460 tcpstat.tcps_rcvshort++;
464 ip = mtod(m, struct ip *);
465 ipov = (struct ipovly *)ip;
466 th = (struct tcphdr *)((caddr_t)ip + off0);
469 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
470 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
471 th->th_sum = m->m_pkthdr.csum_data;
473 th->th_sum = in_pseudo(ip->ip_src.s_addr,
475 htonl(m->m_pkthdr.csum_data +
478 th->th_sum ^= 0xffff;
481 * Checksum extended TCP header and data.
483 len = sizeof(struct ip) + tlen;
484 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
485 ipov->ih_len = (u_short)tlen;
486 ipov->ih_len = htons(ipov->ih_len);
487 th->th_sum = in_cksum(m, len);
490 tcpstat.tcps_rcvbadsum++;
494 /* Re-initialization for later version check */
495 ip->ip_v = IPVERSION;
500 * Check that TCP offset makes sense,
501 * pull out TCP options and adjust length. XXX
503 off = th->th_off << 2;
504 if (off < sizeof(struct tcphdr) || off > tlen) {
505 tcpstat.tcps_rcvbadoff++;
508 tlen -= off; /* tlen is used instead of ti->ti_len */
509 if (off > sizeof(struct tcphdr)) {
511 IP6_EXTHDR_CHECK(m, off0, off, );
512 ip6 = mtod(m, struct ip6_hdr *);
513 th = (struct tcphdr *)((caddr_t)ip6 + off0);
515 if (m->m_len < sizeof(struct ip) + off) {
516 if ((m = m_pullup(m, sizeof(struct ip) + off))
518 tcpstat.tcps_rcvshort++;
521 ip = mtod(m, struct ip *);
522 ipov = (struct ipovly *)ip;
523 th = (struct tcphdr *)((caddr_t)ip + off0);
526 optlen = off - sizeof(struct tcphdr);
527 optp = (u_char *)(th + 1);
529 thflags = th->th_flags;
531 #ifdef TCP_DROP_SYNFIN
533 * If the drop_synfin option is enabled, drop all packets with
534 * both the SYN and FIN bits set. This prevents e.g. nmap from
535 * identifying the TCP/IP stack.
537 * This is a violation of the TCP specification.
539 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
544 * Convert TCP protocol specific fields to host format.
546 th->th_seq = ntohl(th->th_seq);
547 th->th_ack = ntohl(th->th_ack);
548 th->th_win = ntohs(th->th_win);
549 th->th_urp = ntohs(th->th_urp);
552 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
553 * until after ip6_savecontrol() is called and before other functions
554 * which don't want those proto headers.
555 * Because ip6_savecontrol() is going to parse the mbuf to
556 * search for data to be passed up to user-land, it wants mbuf
557 * parameters to be unchanged.
558 * XXX: the call of ip6_savecontrol() has been obsoleted based on
559 * latest version of the advanced API (20020110).
561 drop_hdrlen = off0 + off;
564 * Locate pcb for segment.
567 /* IPFIREWALL_FORWARD section */
568 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
570 * Transparently forwarded. Pretend to be the destination.
571 * already got one like this?
573 inp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
574 ip->ip_src, th->th_sport,
575 ip->ip_dst, th->th_dport,
576 0, m->m_pkthdr.rcvif);
578 /* It's new. Try find the ambushing socket. */
579 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
580 next_hop->sin_addr.s_addr,
582 ntohs(next_hop->sin_port) :
584 inp = in_pcblookup_hash(&tcbinfo[cpu],
585 ip->ip_src, th->th_sport,
588 ntohs(next_hop->sin_port) :
590 1, m->m_pkthdr.rcvif);
594 inp = in6_pcblookup_hash(&tcbinfo[0],
595 &ip6->ip6_src, th->th_sport,
596 &ip6->ip6_dst, th->th_dport,
597 1, m->m_pkthdr.rcvif);
599 inp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
600 ip->ip_src, th->th_sport,
601 ip->ip_dst, th->th_dport,
602 1, m->m_pkthdr.rcvif);
607 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
608 ipsec6stat.in_polvio++;
612 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
613 ipsecstat.in_polvio++;
620 if (inp != NULL && ipsec6_in_reject(m, inp)) {
624 if (inp != NULL && ipsec4_in_reject(m, inp)) {
631 * If the state is CLOSED (i.e., TCB does not exist) then
632 * all data in the incoming segment is discarded.
633 * If the TCB exists but is in CLOSED state, it is embryonic,
634 * but should either do a listen or a connect soon.
639 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
641 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
646 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
647 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
651 strcpy(dbuf, inet_ntoa(ip->ip_dst));
652 strcpy(sbuf, inet_ntoa(ip->ip_src));
654 switch (log_in_vain) {
656 if ((thflags & TH_SYN) == 0)
660 "Connection attempt to TCP %s:%d "
661 "from %s:%d flags:0x%02x\n",
662 dbuf, ntohs(th->th_dport), sbuf,
663 ntohs(th->th_sport), thflags);
672 if (thflags & TH_SYN)
681 rstreason = BANDLIM_RST_CLOSEDPORT;
686 rstreason = BANDLIM_RST_CLOSEDPORT;
689 if (tp->t_state == TCPS_CLOSED)
692 /* Unscale the window into a 32-bit value. */
693 if ((thflags & TH_SYN) == 0)
694 tiwin = th->th_win << tp->snd_scale;
698 so = inp->inp_socket;
701 if (so->so_options & SO_DEBUG) {
702 ostate = tp->t_state;
704 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
706 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
711 if (so->so_options & SO_ACCEPTCONN) {
712 struct in_conninfo inc;
715 inc.inc_isipv6 = isipv6;
718 inc.inc6_faddr = ip6->ip6_src;
719 inc.inc6_laddr = ip6->ip6_dst;
720 inc.inc6_route.ro_rt = NULL; /* XXX */
722 inc.inc_faddr = ip->ip_src;
723 inc.inc_laddr = ip->ip_dst;
724 inc.inc_route.ro_rt = NULL; /* XXX */
726 inc.inc_fport = th->th_sport;
727 inc.inc_lport = th->th_dport;
730 * If the state is LISTEN then ignore segment if it contains
731 * a RST. If the segment contains an ACK then it is bad and
732 * send a RST. If it does not contain a SYN then it is not
733 * interesting; drop it.
735 * If the state is SYN_RECEIVED (syncache) and seg contains
736 * an ACK, but not for our SYN/ACK, send a RST. If the seg
737 * contains a RST, check the sequence number to see if it
738 * is a valid reset segment.
740 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
741 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
742 if (!syncache_expand(&inc, th, &so, m)) {
744 * No syncache entry, or ACK was not
745 * for our SYN/ACK. Send a RST.
747 tcpstat.tcps_badsyn++;
748 rstreason = BANDLIM_RST_OPENPORT;
753 * Could not complete 3-way handshake,
754 * connection is being closed down, and
755 * syncache will free mbuf.
759 * Socket is created in state SYN_RECEIVED.
760 * Continue processing segment.
765 * This is what would have happened in
766 * tcp_output() when the SYN,ACK was sent.
768 tp->snd_up = tp->snd_una;
769 tp->snd_max = tp->snd_nxt = tp->iss + 1;
770 tp->last_ack_sent = tp->rcv_nxt;
772 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
773 * until the _second_ ACK is received:
774 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
775 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
776 * move to ESTAB, set snd_wnd to tiwin.
778 tp->snd_wnd = tiwin; /* unscaled */
781 if (thflags & TH_RST) {
782 syncache_chkrst(&inc, th);
785 if (thflags & TH_ACK) {
786 syncache_badack(&inc);
787 tcpstat.tcps_badsyn++;
788 rstreason = BANDLIM_RST_OPENPORT;
795 * Segment's flags are (SYN) or (SYN|FIN).
799 * If deprecated address is forbidden,
800 * we do not accept SYN to deprecated interface
801 * address to prevent any new inbound connection from
802 * getting established.
803 * When we do not accept SYN, we send a TCP RST,
804 * with deprecated source address (instead of dropping
805 * it). We compromise it as it is much better for peer
806 * to send a RST, and RST will be the final packet
809 * If we do not forbid deprecated addresses, we accept
810 * the SYN packet. RFC2462 does not suggest dropping
812 * If we decipher RFC2462 5.5.4, it says like this:
813 * 1. use of deprecated addr with existing
814 * communication is okay - "SHOULD continue to be
816 * 2. use of it with new communication:
817 * (2a) "SHOULD NOT be used if alternate address
818 * with sufficient scope is available"
819 * (2b) nothing mentioned otherwise.
820 * Here we fall into (2b) case as we have no choice in
821 * our source address selection - we must obey the peer.
823 * The wording in RFC2462 is confusing, and there are
824 * multiple description text for deprecated address
825 * handling - worse, they are not exactly the same.
826 * I believe 5.5.4 is the best one, so we follow 5.5.4.
828 if (isipv6 && !ip6_use_deprecated) {
829 struct in6_ifaddr *ia6;
831 if ((ia6 = ip6_getdstifaddr(m)) &&
832 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
834 rstreason = BANDLIM_RST_OPENPORT;
840 * If it is from this socket, drop it, it must be forged.
841 * Don't bother responding if the destination was a broadcast.
843 if (th->th_dport == th->th_sport) {
845 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
849 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
854 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
856 * Note that it is quite possible to receive unicast
857 * link-layer packets with a broadcast IP address. Use
858 * in_broadcast() to find them.
860 if (m->m_flags & (M_BCAST|M_MCAST))
863 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
864 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
867 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
868 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
869 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
870 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
874 * SYN appears to be valid; create compressed TCP state
875 * for syncache, or perform t/tcp connection.
877 if (so->so_qlen <= so->so_qlimit) {
878 tcp_dooptions(&to, optp, optlen, 1);
879 if (!syncache_add(&inc, &to, th, &so, m))
883 * Entry added to syncache, mbuf used to
884 * send SYN,ACK packet.
888 * Segment passed TAO tests.
893 tp->t_starttime = ticks;
894 tp->t_state = TCPS_ESTABLISHED;
897 * If there is a FIN, or if there is data and the
898 * connection is local, then delay SYN,ACK(SYN) in
899 * the hope of piggy-backing it on a response
900 * segment. Otherwise must send ACK now in case
901 * the other side is slow starting.
904 ((thflags & TH_FIN) ||
906 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
907 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
908 callout_reset(tp->tt_delack, tcp_delacktime,
909 tcp_timer_delack, tp);
910 tp->t_flags |= TF_NEEDSYN;
912 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
914 tcpstat.tcps_connects++;
922 /* XXX temp debugging */
923 /* should not happen - syncache should pick up these connections */
924 if (tp->t_state == TCPS_LISTEN)
925 panic("tcp_input: TCPS_LISTEN");
928 * Segment received on connection.
929 * Reset idle time and keep-alive timer.
931 tp->t_rcvtime = ticks;
932 if (TCPS_HAVEESTABLISHED(tp->t_state))
933 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
937 * XXX this is tradtitional behavior, may need to be cleaned up.
939 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
940 if (thflags & TH_SYN) {
941 if (to.to_flags & TOF_SCALE) {
942 tp->t_flags |= TF_RCVD_SCALE;
943 tp->requested_s_scale = to.to_requested_s_scale;
945 if (to.to_flags & TOF_TS) {
946 tp->t_flags |= TF_RCVD_TSTMP;
947 tp->ts_recent = to.to_tsval;
948 tp->ts_recent_age = ticks;
950 if (to.to_flags & (TOF_CC|TOF_CCNEW))
951 tp->t_flags |= TF_RCVD_CC;
952 if (to.to_flags & TOF_MSS)
953 tcp_mss(tp, to.to_mss);
957 * Header prediction: check for the two common cases
958 * of a uni-directional data xfer. If the packet has
959 * no control flags, is in-sequence, the window didn't
960 * change and we're not retransmitting, it's a
961 * candidate. If the length is zero and the ack moved
962 * forward, we're the sender side of the xfer. Just
963 * free the data acked & wake any higher level process
964 * that was blocked waiting for space. If the length
965 * is non-zero and the ack didn't move, we're the
966 * receiver side. If we're getting packets in-order
967 * (the reassembly queue is empty), add the data to
968 * the socket buffer and note that we need a delayed ack.
969 * Make sure that the hidden state-flags are also off.
970 * Since we check for TCPS_ESTABLISHED above, it can only
973 if (tp->t_state == TCPS_ESTABLISHED &&
974 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
975 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
976 ((to.to_flags & TOF_TS) == 0 ||
977 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
979 * Using the CC option is compulsory if once started:
980 * the segment is OK if no T/TCP was negotiated or
981 * if the segment has a CC option equal to CCrecv
983 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
984 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
985 th->th_seq == tp->rcv_nxt &&
986 tiwin && tiwin == tp->snd_wnd &&
987 tp->snd_nxt == tp->snd_max) {
990 * If last ACK falls within this segment's sequence numbers,
991 * record the timestamp.
992 * NOTE that the test is modified according to the latest
993 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
995 if ((to.to_flags & TOF_TS) != 0 &&
996 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
997 tp->ts_recent_age = ticks;
998 tp->ts_recent = to.to_tsval;
1002 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1003 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1004 tp->snd_cwnd >= tp->snd_wnd &&
1005 ((!tcp_do_newreno &&
1006 tp->t_dupacks < tcprexmtthresh) ||
1007 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
1009 * this is a pure ack for outstanding data.
1011 ++tcpstat.tcps_predack;
1013 * "bad retransmit" recovery
1015 * If Eifel detection applies, then
1016 * it is deterministic, so use it
1017 * unconditionally over the old heuristic.
1018 * Otherwise, fall back to the old heuristic.
1020 if (tcp_do_eifel_detect &&
1021 (to.to_flags & TOF_TS) && to.to_tsecr &&
1022 (tp->t_flags & TF_FIRSTACCACK)) {
1023 /* Eifel detection applicable. */
1024 if (to.to_tsecr < tp->t_rexmtTS) {
1025 tcp_revert_congestion_state(tp);
1026 ++tcpstat.tcps_eifeldetected;
1028 } else if (tp->t_rxtshift == 1 &&
1029 ticks < tp->t_badrxtwin) {
1030 tcp_revert_congestion_state(tp);
1031 ++tcpstat.tcps_rttdetected;
1033 tp->t_flags &= ~(TF_FIRSTACCACK |
1034 TF_FASTREXMT | TF_EARLYREXMT);
1036 * Recalculate the retransmit timer / rtt.
1038 * Some machines (certain windows boxes)
1039 * send broken timestamp replies during the
1040 * SYN+ACK phase, ignore timestamps of 0.
1042 if ((to.to_flags & TOF_TS) != 0 &&
1045 ticks - to.to_tsecr + 1);
1046 } else if (tp->t_rtttime &&
1047 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1049 ticks - tp->t_rtttime);
1051 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1052 acked = th->th_ack - tp->snd_una;
1053 tcpstat.tcps_rcvackpack++;
1054 tcpstat.tcps_rcvackbyte += acked;
1055 sbdrop(&so->so_snd, acked);
1056 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1057 SEQ_LEQ(th->th_ack, tp->snd_recover))
1058 tp->snd_recover = th->th_ack - 1;
1059 tp->snd_una = th->th_ack;
1062 ND6_HINT(tp); /* some progress has been done */
1065 * If all outstanding data are acked, stop
1066 * retransmit timer, otherwise restart timer
1067 * using current (possibly backed-off) value.
1068 * If process is waiting for space,
1069 * wakeup/selwakeup/signal. If data
1070 * are ready to send, let tcp_output
1071 * decide between more output or persist.
1073 if (tp->snd_una == tp->snd_max)
1074 callout_stop(tp->tt_rexmt);
1075 else if (!callout_active(tp->tt_persist))
1076 callout_reset(tp->tt_rexmt,
1078 tcp_timer_rexmt, tp);
1081 if (so->so_snd.sb_cc)
1082 (void) tcp_output(tp);
1085 } else if (th->th_ack == tp->snd_una &&
1086 LIST_EMPTY(&tp->t_segq) &&
1087 tlen <= sbspace(&so->so_rcv)) {
1089 * this is a pure, in-sequence data packet
1090 * with nothing on the reassembly queue and
1091 * we have enough buffer space to take it.
1093 ++tcpstat.tcps_preddat;
1094 tp->rcv_nxt += tlen;
1095 tcpstat.tcps_rcvpack++;
1096 tcpstat.tcps_rcvbyte += tlen;
1097 ND6_HINT(tp); /* some progress has been done */
1099 * Add data to socket buffer.
1101 if (so->so_state & SS_CANTRCVMORE) {
1104 m_adj(m, drop_hdrlen); /* delayed header drop */
1105 sbappend(&so->so_rcv, m);
1108 if (DELAY_ACK(tp)) {
1109 callout_reset(tp->tt_delack, tcp_delacktime,
1110 tcp_timer_delack, tp);
1112 tp->t_flags |= TF_ACKNOW;
1120 * Calculate amount of space in receive window,
1121 * and then do TCP input processing.
1122 * Receive window is amount of space in rcv queue,
1123 * but not less than advertised window.
1127 win = sbspace(&so->so_rcv);
1130 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1133 switch (tp->t_state) {
1136 * If the state is SYN_RECEIVED:
1137 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1139 case TCPS_SYN_RECEIVED:
1140 if ((thflags & TH_ACK) &&
1141 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1142 SEQ_GT(th->th_ack, tp->snd_max))) {
1143 rstreason = BANDLIM_RST_OPENPORT;
1149 * If the state is SYN_SENT:
1150 * if seg contains an ACK, but not for our SYN, drop the input.
1151 * if seg contains a RST, then drop the connection.
1152 * if seg does not contain SYN, then drop it.
1153 * Otherwise this is an acceptable SYN segment
1154 * initialize tp->rcv_nxt and tp->irs
1155 * if seg contains ack then advance tp->snd_una
1156 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1157 * arrange for segment to be acked (eventually)
1158 * continue processing rest of data/controls, beginning with URG
1161 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1162 taop = &tao_noncached;
1163 bzero(taop, sizeof(*taop));
1166 if ((thflags & TH_ACK) &&
1167 (SEQ_LEQ(th->th_ack, tp->iss) ||
1168 SEQ_GT(th->th_ack, tp->snd_max))) {
1170 * If we have a cached CCsent for the remote host,
1171 * hence we haven't just crashed and restarted,
1172 * do not send a RST. This may be a retransmission
1173 * from the other side after our earlier ACK was lost.
1174 * Our new SYN, when it arrives, will serve as the
1177 if (taop->tao_ccsent != 0)
1180 rstreason = BANDLIM_UNLIMITED;
1184 if (thflags & TH_RST) {
1185 if (thflags & TH_ACK)
1186 tp = tcp_drop(tp, ECONNREFUSED);
1189 if ((thflags & TH_SYN) == 0)
1191 tp->snd_wnd = th->th_win; /* initial send window */
1192 tp->cc_recv = to.to_cc; /* foreign CC */
1194 tp->irs = th->th_seq;
1196 if (thflags & TH_ACK) {
1198 * Our SYN was acked. If segment contains CC.ECHO
1199 * option, check it to make sure this segment really
1200 * matches our SYN. If not, just drop it as old
1201 * duplicate, but send an RST if we're still playing
1202 * by the old rules. If no CC.ECHO option, make sure
1203 * we don't get fooled into using T/TCP.
1205 if (to.to_flags & TOF_CCECHO) {
1206 if (tp->cc_send != to.to_ccecho) {
1207 if (taop->tao_ccsent != 0)
1210 rstreason = BANDLIM_UNLIMITED;
1215 tp->t_flags &= ~TF_RCVD_CC;
1216 tcpstat.tcps_connects++;
1218 /* Do window scaling on this connection? */
1219 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1220 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1221 tp->snd_scale = tp->requested_s_scale;
1222 tp->rcv_scale = tp->request_r_scale;
1224 /* Segment is acceptable, update cache if undefined. */
1225 if (taop->tao_ccsent == 0)
1226 taop->tao_ccsent = to.to_ccecho;
1228 tp->rcv_adv += tp->rcv_wnd;
1229 tp->snd_una++; /* SYN is acked */
1231 * If there's data, delay ACK; if there's also a FIN
1232 * ACKNOW will be turned on later.
1234 if (DELAY_ACK(tp) && tlen != 0)
1235 callout_reset(tp->tt_delack, tcp_delacktime,
1236 tcp_timer_delack, tp);
1238 tp->t_flags |= TF_ACKNOW;
1240 * Received <SYN,ACK> in SYN_SENT[*] state.
1242 * SYN_SENT --> ESTABLISHED
1243 * SYN_SENT* --> FIN_WAIT_1
1245 tp->t_starttime = ticks;
1246 if (tp->t_flags & TF_NEEDFIN) {
1247 tp->t_state = TCPS_FIN_WAIT_1;
1248 tp->t_flags &= ~TF_NEEDFIN;
1251 tp->t_state = TCPS_ESTABLISHED;
1252 callout_reset(tp->tt_keep, tcp_keepidle,
1253 tcp_timer_keep, tp);
1257 * Received initial SYN in SYN-SENT[*] state =>
1258 * simultaneous open. If segment contains CC option
1259 * and there is a cached CC, apply TAO test.
1260 * If it succeeds, connection is * half-synchronized.
1261 * Otherwise, do 3-way handshake:
1262 * SYN-SENT -> SYN-RECEIVED
1263 * SYN-SENT* -> SYN-RECEIVED*
1264 * If there was no CC option, clear cached CC value.
1266 tp->t_flags |= TF_ACKNOW;
1267 callout_stop(tp->tt_rexmt);
1268 if (to.to_flags & TOF_CC) {
1269 if (taop->tao_cc != 0 &&
1270 CC_GT(to.to_cc, taop->tao_cc)) {
1272 * update cache and make transition:
1273 * SYN-SENT -> ESTABLISHED*
1274 * SYN-SENT* -> FIN-WAIT-1*
1276 taop->tao_cc = to.to_cc;
1277 tp->t_starttime = ticks;
1278 if (tp->t_flags & TF_NEEDFIN) {
1279 tp->t_state = TCPS_FIN_WAIT_1;
1280 tp->t_flags &= ~TF_NEEDFIN;
1282 tp->t_state = TCPS_ESTABLISHED;
1283 callout_reset(tp->tt_keep,
1288 tp->t_flags |= TF_NEEDSYN;
1290 tp->t_state = TCPS_SYN_RECEIVED;
1292 /* CC.NEW or no option => invalidate cache */
1294 tp->t_state = TCPS_SYN_RECEIVED;
1300 * Advance th->th_seq to correspond to first data byte.
1301 * If data, trim to stay within window,
1302 * dropping FIN if necessary.
1305 if (tlen > tp->rcv_wnd) {
1306 todrop = tlen - tp->rcv_wnd;
1310 tcpstat.tcps_rcvpackafterwin++;
1311 tcpstat.tcps_rcvbyteafterwin += todrop;
1313 tp->snd_wl1 = th->th_seq - 1;
1314 tp->rcv_up = th->th_seq;
1316 * Client side of transaction: already sent SYN and data.
1317 * If the remote host used T/TCP to validate the SYN,
1318 * our data will be ACK'd; if so, enter normal data segment
1319 * processing in the middle of step 5, ack processing.
1320 * Otherwise, goto step 6.
1322 if (thflags & TH_ACK)
1328 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1329 * if segment contains a SYN and CC [not CC.NEW] option:
1330 * if state == TIME_WAIT and connection duration > MSL,
1331 * drop packet and send RST;
1333 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1334 * ack the FIN (and data) in retransmission queue.
1335 * Complete close and delete TCPCB. Then reprocess
1336 * segment, hoping to find new TCPCB in LISTEN state;
1338 * else must be old SYN; drop it.
1339 * else do normal processing.
1343 case TCPS_TIME_WAIT:
1344 if ((thflags & TH_SYN) &&
1345 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1346 if (tp->t_state == TCPS_TIME_WAIT &&
1347 (ticks - tp->t_starttime) > tcp_msl) {
1348 rstreason = BANDLIM_UNLIMITED;
1351 if (CC_GT(to.to_cc, tp->cc_recv)) {
1358 break; /* continue normal processing */
1362 * States other than LISTEN or SYN_SENT.
1363 * First check the RST flag and sequence number since reset segments
1364 * are exempt from the timestamp and connection count tests. This
1365 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1366 * below which allowed reset segments in half the sequence space
1367 * to fall though and be processed (which gives forged reset
1368 * segments with a random sequence number a 50 percent chance of
1369 * killing a connection).
1370 * Then check timestamp, if present.
1371 * Then check the connection count, if present.
1372 * Then check that at least some bytes of segment are within
1373 * receive window. If segment begins before rcv_nxt,
1374 * drop leading data (and SYN); if nothing left, just ack.
1377 * If the RST bit is set, check the sequence number to see
1378 * if this is a valid reset segment.
1380 * In all states except SYN-SENT, all reset (RST) segments
1381 * are validated by checking their SEQ-fields. A reset is
1382 * valid if its sequence number is in the window.
1383 * Note: this does not take into account delayed ACKs, so
1384 * we should test against last_ack_sent instead of rcv_nxt.
1385 * The sequence number in the reset segment is normally an
1386 * echo of our outgoing acknowlegement numbers, but some hosts
1387 * send a reset with the sequence number at the rightmost edge
1388 * of our receive window, and we have to handle this case.
1389 * If we have multiple segments in flight, the intial reset
1390 * segment sequence numbers will be to the left of last_ack_sent,
1391 * but they will eventually catch up.
1392 * In any case, it never made sense to trim reset segments to
1393 * fit the receive window since RFC 1122 says:
1394 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1396 * A TCP SHOULD allow a received RST segment to include data.
1399 * It has been suggested that a RST segment could contain
1400 * ASCII text that encoded and explained the cause of the
1401 * RST. No standard has yet been established for such
1404 * If the reset segment passes the sequence number test examine
1406 * SYN_RECEIVED STATE:
1407 * If passive open, return to LISTEN state.
1408 * If active open, inform user that connection was refused.
1409 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1410 * Inform user that connection was reset, and close tcb.
1411 * CLOSING, LAST_ACK STATES:
1414 * Drop the segment - see Stevens, vol. 2, p. 964 and
1417 if (thflags & TH_RST) {
1418 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1419 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1420 switch (tp->t_state) {
1422 case TCPS_SYN_RECEIVED:
1423 so->so_error = ECONNREFUSED;
1426 case TCPS_ESTABLISHED:
1427 case TCPS_FIN_WAIT_1:
1428 case TCPS_FIN_WAIT_2:
1429 case TCPS_CLOSE_WAIT:
1430 so->so_error = ECONNRESET;
1432 tp->t_state = TCPS_CLOSED;
1433 tcpstat.tcps_drops++;
1442 case TCPS_TIME_WAIT:
1450 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1451 * and it's less than ts_recent, drop it.
1453 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1454 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1456 /* Check to see if ts_recent is over 24 days old. */
1457 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1459 * Invalidate ts_recent. If this segment updates
1460 * ts_recent, the age will be reset later and ts_recent
1461 * will get a valid value. If it does not, setting
1462 * ts_recent to zero will at least satisfy the
1463 * requirement that zero be placed in the timestamp
1464 * echo reply when ts_recent isn't valid. The
1465 * age isn't reset until we get a valid ts_recent
1466 * because we don't want out-of-order segments to be
1467 * dropped when ts_recent is old.
1471 tcpstat.tcps_rcvduppack++;
1472 tcpstat.tcps_rcvdupbyte += tlen;
1473 tcpstat.tcps_pawsdrop++;
1482 * If T/TCP was negotiated and the segment doesn't have CC,
1483 * or if its CC is wrong then drop the segment.
1484 * RST segments do not have to comply with this.
1486 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1487 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1491 * In the SYN-RECEIVED state, validate that the packet belongs to
1492 * this connection before trimming the data to fit the receive
1493 * window. Check the sequence number versus IRS since we know
1494 * the sequence numbers haven't wrapped. This is a partial fix
1495 * for the "LAND" DoS attack.
1497 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1498 rstreason = BANDLIM_RST_OPENPORT;
1502 todrop = tp->rcv_nxt - th->th_seq;
1504 if (thflags & TH_SYN) {
1514 * Following if statement from Stevens, vol. 2, p. 960.
1517 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1519 * Any valid FIN must be to the left of the window.
1520 * At this point the FIN must be a duplicate or out
1521 * of sequence; drop it.
1526 * Send an ACK to resynchronize and drop any data.
1527 * But keep on processing for RST or ACK.
1529 tp->t_flags |= TF_ACKNOW;
1531 tcpstat.tcps_rcvduppack++;
1532 tcpstat.tcps_rcvdupbyte += todrop;
1534 tcpstat.tcps_rcvpartduppack++;
1535 tcpstat.tcps_rcvpartdupbyte += todrop;
1537 drop_hdrlen += todrop; /* drop from the top afterwards */
1538 th->th_seq += todrop;
1540 if (th->th_urp > todrop)
1541 th->th_urp -= todrop;
1549 * If new data are received on a connection after the
1550 * user processes are gone, then RST the other end.
1552 if ((so->so_state & SS_NOFDREF) &&
1553 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1555 tcpstat.tcps_rcvafterclose++;
1556 rstreason = BANDLIM_UNLIMITED;
1561 * If segment ends after window, drop trailing data
1562 * (and PUSH and FIN); if nothing left, just ACK.
1564 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1566 tcpstat.tcps_rcvpackafterwin++;
1567 if (todrop >= tlen) {
1568 tcpstat.tcps_rcvbyteafterwin += tlen;
1570 * If a new connection request is received
1571 * while in TIME_WAIT, drop the old connection
1572 * and start over if the sequence numbers
1573 * are above the previous ones.
1575 if (thflags & TH_SYN &&
1576 tp->t_state == TCPS_TIME_WAIT &&
1577 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1582 * If window is closed can only take segments at
1583 * window edge, and have to drop data and PUSH from
1584 * incoming segments. Continue processing, but
1585 * remember to ack. Otherwise, drop segment
1588 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1589 tp->t_flags |= TF_ACKNOW;
1590 tcpstat.tcps_rcvwinprobe++;
1594 tcpstat.tcps_rcvbyteafterwin += todrop;
1597 thflags &= ~(TH_PUSH|TH_FIN);
1601 * If last ACK falls within this segment's sequence numbers,
1602 * record its timestamp.
1603 * NOTE that the test is modified according to the latest
1604 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1606 if ((to.to_flags & TOF_TS) != 0 &&
1607 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1608 tp->ts_recent_age = ticks;
1609 tp->ts_recent = to.to_tsval;
1613 * If a SYN is in the window, then this is an
1614 * error and we send an RST and drop the connection.
1616 if (thflags & TH_SYN) {
1617 tp = tcp_drop(tp, ECONNRESET);
1618 rstreason = BANDLIM_UNLIMITED;
1623 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1624 * flag is on (half-synchronized state), then queue data for
1625 * later processing; else drop segment and return.
1627 if ((thflags & TH_ACK) == 0) {
1628 if (tp->t_state == TCPS_SYN_RECEIVED ||
1629 (tp->t_flags & TF_NEEDSYN))
1638 switch (tp->t_state) {
1641 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1642 * ESTABLISHED state and continue processing.
1643 * The ACK was checked above.
1645 case TCPS_SYN_RECEIVED:
1647 tcpstat.tcps_connects++;
1649 /* Do window scaling? */
1650 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1651 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1652 tp->snd_scale = tp->requested_s_scale;
1653 tp->rcv_scale = tp->request_r_scale;
1656 * Upon successful completion of 3-way handshake,
1657 * update cache.CC if it was undefined, pass any queued
1658 * data to the user, and advance state appropriately.
1660 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1662 taop->tao_cc = tp->cc_recv;
1666 * SYN-RECEIVED -> ESTABLISHED
1667 * SYN-RECEIVED* -> FIN-WAIT-1
1669 tp->t_starttime = ticks;
1670 if (tp->t_flags & TF_NEEDFIN) {
1671 tp->t_state = TCPS_FIN_WAIT_1;
1672 tp->t_flags &= ~TF_NEEDFIN;
1674 tp->t_state = TCPS_ESTABLISHED;
1675 callout_reset(tp->tt_keep, tcp_keepidle,
1676 tcp_timer_keep, tp);
1679 * If segment contains data or ACK, will call tcp_reass()
1680 * later; if not, do so now to pass queued data to user.
1682 if (tlen == 0 && (thflags & TH_FIN) == 0)
1683 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1685 tp->snd_wl1 = th->th_seq - 1;
1689 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1690 * ACKs. If the ack is in the range
1691 * tp->snd_una < th->th_ack <= tp->snd_max
1692 * then advance tp->snd_una to th->th_ack and drop
1693 * data from the retransmission queue. If this ACK reflects
1694 * more up to date window information we update our window information.
1696 case TCPS_ESTABLISHED:
1697 case TCPS_FIN_WAIT_1:
1698 case TCPS_FIN_WAIT_2:
1699 case TCPS_CLOSE_WAIT:
1702 case TCPS_TIME_WAIT:
1704 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1705 if (tlen == 0 && tiwin == tp->snd_wnd) {
1706 tcpstat.tcps_rcvdupack++;
1708 * If we have outstanding data (other than
1709 * a window probe), this is a completely
1710 * duplicate ack (ie, window info didn't
1711 * change), the ack is the biggest we've
1712 * seen and we've seen exactly our rexmt
1713 * threshhold of them, assume a packet
1714 * has been dropped and retransmit it.
1715 * Kludge snd_nxt & the congestion
1716 * window so we send only this one
1719 * We know we're losing at the current
1720 * window size so do congestion avoidance
1721 * (set ssthresh to half the current window
1722 * and pull our congestion window back to
1723 * the new ssthresh).
1725 * Dup acks mean that packets have left the
1726 * network (they're now cached at the receiver)
1727 * so bump cwnd by the amount in the receiver
1728 * to keep a constant cwnd packets in the
1731 if (!callout_active(tp->tt_rexmt) ||
1732 th->th_ack != tp->snd_una)
1734 else if (++tp->t_dupacks > tcprexmtthresh ||
1736 IN_FASTRECOVERY(tp))) {
1737 tp->snd_cwnd += tp->t_maxseg;
1738 (void) tcp_output(tp);
1740 } else if (tp->t_dupacks == tcprexmtthresh) {
1744 if (tcp_do_newreno &&
1751 if (tcp_do_eifel_detect &&
1752 (tp->t_flags & TF_RCVD_TSTMP)) {
1753 tcp_save_congestion_state(tp);
1754 tp->t_flags |= TF_FASTREXMT;
1756 win = min(tp->snd_wnd, tp->snd_cwnd) /
1760 tp->snd_ssthresh = win * tp->t_maxseg;
1761 ENTER_FASTRECOVERY(tp);
1762 tp->snd_recover = tp->snd_max;
1763 callout_stop(tp->tt_rexmt);
1766 tp->snd_nxt = th->th_ack;
1767 tp->snd_cwnd = tp->t_maxseg;
1768 (void) tcp_output(tp);
1769 ++tcpstat.tcps_sndfastrexmit;
1770 KASSERT(tp->snd_limited <= 2,
1771 ("tp->snd_limited too big"));
1772 tp->snd_cwnd = tp->snd_ssthresh +
1774 (tp->t_dupacks - tp->snd_limited));
1775 if (SEQ_GT(onxt, tp->snd_nxt))
1778 } else if (tcp_do_limitedtransmit) {
1779 u_long oldcwnd = tp->snd_cwnd;
1780 tcp_seq oldsndmax = tp->snd_max;
1781 /* outstanding data */
1783 tp->snd_max - tp->snd_una;
1786 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1788 KASSERT(tp->t_dupacks == 1 ||
1790 ("dupacks not 1 or 2"));
1791 if (tp->t_dupacks == 1)
1792 tp->snd_limited = 0;
1793 tp->snd_cwnd = ownd +
1794 (tp->t_dupacks - tp->snd_limited) *
1796 (void) tcp_output(tp);
1797 tp->snd_cwnd = oldcwnd;
1798 sent = tp->snd_max - oldsndmax;
1799 if (sent > tp->t_maxseg) {
1800 KASSERT((tp->t_dupacks == 2 &&
1801 tp->snd_limited == 0) ||
1802 (sent == tp->t_maxseg + 1 &&
1803 tp->t_flags & TF_SENTFIN),
1807 ("sent too many segments"));
1808 tp->snd_limited = 2;
1809 tcpstat.tcps_sndlimited += 2;
1810 } else if (sent > 0) {
1812 ++tcpstat.tcps_sndlimited;
1813 } else if (tcp_do_early_retransmit &&
1814 (tcp_do_eifel_detect &&
1815 (tp->t_flags & TF_RCVD_TSTMP)) &&
1817 tp->t_dupacks + 1 >=
1818 iceildiv(ownd, tp->t_maxseg)) {
1819 ++tcpstat.tcps_sndearlyrexmit;
1820 tp->t_flags |= TF_EARLYREXMT;
1821 goto fastretransmit;
1830 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1833 * If the congestion window was inflated to account
1834 * for the other side's cached packets, retract it.
1836 if (tcp_do_newreno) {
1837 if (IN_FASTRECOVERY(tp)) {
1838 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1839 tcp_newreno_partial_ack(tp, th);
1842 * Window inflation should have left us
1843 * with approximately snd_ssthresh
1845 * But in case we would be inclined to
1846 * send a burst, better to do it via
1847 * the slow start mechanism.
1849 if (SEQ_GT(th->th_ack +
1852 tp->snd_cwnd = tp->snd_max -
1856 tp->snd_cwnd = tp->snd_ssthresh;
1860 if (tp->t_dupacks >= tcprexmtthresh &&
1861 tp->snd_cwnd > tp->snd_ssthresh)
1862 tp->snd_cwnd = tp->snd_ssthresh;
1865 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1866 tcpstat.tcps_rcvacktoomuch++;
1870 * If we reach this point, ACK is not a duplicate,
1871 * i.e., it ACKs something we sent.
1873 if (tp->t_flags & TF_NEEDSYN) {
1875 * T/TCP: Connection was half-synchronized, and our
1876 * SYN has been ACK'd (so connection is now fully
1877 * synchronized). Go to non-starred state,
1878 * increment snd_una for ACK of SYN, and check if
1879 * we can do window scaling.
1881 tp->t_flags &= ~TF_NEEDSYN;
1883 /* Do window scaling? */
1884 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1885 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1886 tp->snd_scale = tp->requested_s_scale;
1887 tp->rcv_scale = tp->request_r_scale;
1892 acked = th->th_ack - tp->snd_una;
1893 tcpstat.tcps_rcvackpack++;
1894 tcpstat.tcps_rcvackbyte += acked;
1897 * If we just performed our first retransmit, and the ACK
1898 * arrives within our recovery window, then it was a mistake
1899 * to do the retransmit in the first place. Recover our
1900 * original cwnd and ssthresh, and proceed to transmit where
1903 if (tcp_do_eifel_detect && acked &&
1904 (to.to_flags & TOF_TS) && to.to_tsecr &&
1905 (tp->t_flags & TF_FIRSTACCACK)) {
1906 /* Eifel detection applicable. */
1907 if (to.to_tsecr < tp->t_rexmtTS) {
1908 ++tcpstat.tcps_eifeldetected;
1909 tcp_revert_congestion_state(tp);
1910 if (tp->t_rxtshift == 1 &&
1911 ticks >= tp->t_badrxtwin)
1912 ++tcpstat.tcps_rttcantdetect;
1914 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1915 tcp_revert_congestion_state(tp);
1916 ++tcpstat.tcps_rttdetected;
1920 * If we have a timestamp reply, update smoothed
1921 * round trip time. If no timestamp is present but
1922 * transmit timer is running and timed sequence
1923 * number was acked, update smoothed round trip time.
1924 * Since we now have an rtt measurement, cancel the
1925 * timer backoff (cf., Phil Karn's retransmit alg.).
1926 * Recompute the initial retransmit timer.
1928 * Some machines (certain windows boxes) send broken
1929 * timestamp replies during the SYN+ACK phase, ignore
1932 if ((to.to_flags & TOF_TS) != 0 &&
1934 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1935 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1936 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1938 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1941 * If all outstanding data is acked, stop retransmit
1942 * timer and remember to restart (more output or persist).
1943 * If there is more data to be acked, restart retransmit
1944 * timer, using current (possibly backed-off) value.
1946 if (th->th_ack == tp->snd_max) {
1947 callout_stop(tp->tt_rexmt);
1949 } else if (!callout_active(tp->tt_persist))
1950 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1951 tcp_timer_rexmt, tp);
1954 * If no data (only SYN) was ACK'd,
1955 * skip rest of ACK processing.
1960 /* Stop looking for an acceptable ACK since one was received. */
1961 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
1964 * When new data is acked, open the congestion window.
1965 * If the window gives us less than ssthresh packets
1966 * in flight, open exponentially (maxseg per packet).
1967 * Otherwise open linearly: maxseg per window
1968 * (maxseg^2 / cwnd per packet).
1970 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1971 u_int cw = tp->snd_cwnd;
1972 u_int incr = tp->t_maxseg;
1973 if (cw > tp->snd_ssthresh)
1974 incr = incr * incr / cw;
1975 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1977 if (acked > so->so_snd.sb_cc) {
1978 tp->snd_wnd -= so->so_snd.sb_cc;
1979 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1982 sbdrop(&so->so_snd, acked);
1983 tp->snd_wnd -= acked;
1987 /* detect una wraparound */
1988 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1989 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1990 SEQ_LEQ(th->th_ack, tp->snd_recover))
1991 tp->snd_recover = th->th_ack - 1;
1992 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1993 SEQ_GEQ(th->th_ack, tp->snd_recover))
1994 EXIT_FASTRECOVERY(tp);
1995 tp->snd_una = th->th_ack;
1996 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1997 tp->snd_nxt = tp->snd_una;
1999 switch (tp->t_state) {
2002 * In FIN_WAIT_1 STATE in addition to the processing
2003 * for the ESTABLISHED state if our FIN is now acknowledged
2004 * then enter FIN_WAIT_2.
2006 case TCPS_FIN_WAIT_1:
2007 if (ourfinisacked) {
2009 * If we can't receive any more
2010 * data, then closing user can proceed.
2011 * Starting the timer is contrary to the
2012 * specification, but if we don't get a FIN
2013 * we'll hang forever.
2015 if (so->so_state & SS_CANTRCVMORE) {
2016 soisdisconnected(so);
2017 callout_reset(tp->tt_2msl, tcp_maxidle,
2018 tcp_timer_2msl, tp);
2020 tp->t_state = TCPS_FIN_WAIT_2;
2025 * In CLOSING STATE in addition to the processing for
2026 * the ESTABLISHED state if the ACK acknowledges our FIN
2027 * then enter the TIME-WAIT state, otherwise ignore
2031 if (ourfinisacked) {
2032 tp->t_state = TCPS_TIME_WAIT;
2033 tcp_canceltimers(tp);
2034 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2035 if (tp->cc_recv != 0 &&
2036 (ticks - tp->t_starttime) < tcp_msl)
2037 callout_reset(tp->tt_2msl,
2040 tcp_timer_2msl, tp);
2042 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2043 tcp_timer_2msl, tp);
2044 soisdisconnected(so);
2049 * In LAST_ACK, we may still be waiting for data to drain
2050 * and/or to be acked, as well as for the ack of our FIN.
2051 * If our FIN is now acknowledged, delete the TCB,
2052 * enter the closed state and return.
2055 if (ourfinisacked) {
2062 * In TIME_WAIT state the only thing that should arrive
2063 * is a retransmission of the remote FIN. Acknowledge
2064 * it and restart the finack timer.
2066 case TCPS_TIME_WAIT:
2067 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2068 tcp_timer_2msl, tp);
2075 * Update window information.
2076 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2078 if ((thflags & TH_ACK) &&
2079 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2080 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2081 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2082 /* keep track of pure window updates */
2084 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2085 tcpstat.tcps_rcvwinupd++;
2086 tp->snd_wnd = tiwin;
2087 tp->snd_wl1 = th->th_seq;
2088 tp->snd_wl2 = th->th_ack;
2089 if (tp->snd_wnd > tp->max_sndwnd)
2090 tp->max_sndwnd = tp->snd_wnd;
2095 * Process segments with URG.
2097 if ((thflags & TH_URG) && th->th_urp &&
2098 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2100 * This is a kludge, but if we receive and accept
2101 * random urgent pointers, we'll crash in
2102 * soreceive. It's hard to imagine someone
2103 * actually wanting to send this much urgent data.
2105 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2106 th->th_urp = 0; /* XXX */
2107 thflags &= ~TH_URG; /* XXX */
2108 goto dodata; /* XXX */
2111 * If this segment advances the known urgent pointer,
2112 * then mark the data stream. This should not happen
2113 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2114 * a FIN has been received from the remote side.
2115 * In these states we ignore the URG.
2117 * According to RFC961 (Assigned Protocols),
2118 * the urgent pointer points to the last octet
2119 * of urgent data. We continue, however,
2120 * to consider it to indicate the first octet
2121 * of data past the urgent section as the original
2122 * spec states (in one of two places).
2124 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2125 tp->rcv_up = th->th_seq + th->th_urp;
2126 so->so_oobmark = so->so_rcv.sb_cc +
2127 (tp->rcv_up - tp->rcv_nxt) - 1;
2128 if (so->so_oobmark == 0)
2129 so->so_state |= SS_RCVATMARK;
2131 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2134 * Remove out of band data so doesn't get presented to user.
2135 * This can happen independent of advancing the URG pointer,
2136 * but if two URG's are pending at once, some out-of-band
2137 * data may creep in... ick.
2139 if (th->th_urp <= (u_long)tlen
2141 && (so->so_options & SO_OOBINLINE) == 0
2144 tcp_pulloutofband(so, th, m,
2145 drop_hdrlen); /* hdr drop is delayed */
2148 * If no out of band data is expected,
2149 * pull receive urgent pointer along
2150 * with the receive window.
2152 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2153 tp->rcv_up = tp->rcv_nxt;
2158 * Process the segment text, merging it into the TCP sequencing queue,
2159 * and arranging for acknowledgment of receipt if necessary.
2160 * This process logically involves adjusting tp->rcv_wnd as data
2161 * is presented to the user (this happens in tcp_usrreq.c,
2162 * case PRU_RCVD). If a FIN has already been received on this
2163 * connection then we just ignore the text.
2165 if ((tlen || (thflags & TH_FIN)) &&
2166 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2167 m_adj(m, drop_hdrlen); /* delayed header drop */
2169 * Insert segment which includes th into TCP reassembly queue
2170 * with control block tp. Set thflags to whether reassembly now
2171 * includes a segment with FIN. This handles the common case
2172 * inline (segment is the next to be received on an established
2173 * connection, and the queue is empty), avoiding linkage into
2174 * and removal from the queue and repetition of various
2176 * Set DELACK for segments received in order, but ack
2177 * immediately when segments are out of order (so
2178 * fast retransmit can work).
2180 if (th->th_seq == tp->rcv_nxt &&
2181 LIST_EMPTY(&tp->t_segq) &&
2182 TCPS_HAVEESTABLISHED(tp->t_state)) {
2184 callout_reset(tp->tt_delack, tcp_delacktime,
2185 tcp_timer_delack, tp);
2187 tp->t_flags |= TF_ACKNOW;
2188 tp->rcv_nxt += tlen;
2189 thflags = th->th_flags & TH_FIN;
2190 tcpstat.tcps_rcvpack++;
2191 tcpstat.tcps_rcvbyte += tlen;
2193 if (so->so_state & SS_CANTRCVMORE)
2196 sbappend(&so->so_rcv, m);
2199 thflags = tcp_reass(tp, th, &tlen, m);
2200 tp->t_flags |= TF_ACKNOW;
2204 * Note the amount of data that peer has sent into
2205 * our window, in order to estimate the sender's
2208 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2215 * If FIN is received ACK the FIN and let the user know
2216 * that the connection is closing.
2218 if (thflags & TH_FIN) {
2219 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2222 * If connection is half-synchronized
2223 * (ie NEEDSYN flag on) then delay ACK,
2224 * so it may be piggybacked when SYN is sent.
2225 * Otherwise, since we received a FIN then no
2226 * more input can be expected, send ACK now.
2228 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2229 callout_reset(tp->tt_delack, tcp_delacktime,
2230 tcp_timer_delack, tp);
2232 tp->t_flags |= TF_ACKNOW;
2235 switch (tp->t_state) {
2238 * In SYN_RECEIVED and ESTABLISHED STATES
2239 * enter the CLOSE_WAIT state.
2241 case TCPS_SYN_RECEIVED:
2242 tp->t_starttime = ticks;
2244 case TCPS_ESTABLISHED:
2245 tp->t_state = TCPS_CLOSE_WAIT;
2249 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2250 * enter the CLOSING state.
2252 case TCPS_FIN_WAIT_1:
2253 tp->t_state = TCPS_CLOSING;
2257 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2258 * starting the time-wait timer, turning off the other
2261 case TCPS_FIN_WAIT_2:
2262 tp->t_state = TCPS_TIME_WAIT;
2263 tcp_canceltimers(tp);
2264 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2265 if (tp->cc_recv != 0 &&
2266 (ticks - tp->t_starttime) < tcp_msl) {
2267 callout_reset(tp->tt_2msl,
2268 tp->t_rxtcur * TCPTV_TWTRUNC,
2269 tcp_timer_2msl, tp);
2270 /* For transaction client, force ACK now. */
2271 tp->t_flags |= TF_ACKNOW;
2274 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2275 tcp_timer_2msl, tp);
2276 soisdisconnected(so);
2280 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2282 case TCPS_TIME_WAIT:
2283 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2284 tcp_timer_2msl, tp);
2289 if (so->so_options & SO_DEBUG)
2290 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2295 * Return any desired output.
2297 if (needoutput || (tp->t_flags & TF_ACKNOW))
2298 (void) tcp_output(tp);
2303 * Generate an ACK dropping incoming segment if it occupies
2304 * sequence space, where the ACK reflects our state.
2306 * We can now skip the test for the RST flag since all
2307 * paths to this code happen after packets containing
2308 * RST have been dropped.
2310 * In the SYN-RECEIVED state, don't send an ACK unless the
2311 * segment we received passes the SYN-RECEIVED ACK test.
2312 * If it fails send a RST. This breaks the loop in the
2313 * "LAND" DoS attack, and also prevents an ACK storm
2314 * between two listening ports that have been sent forged
2315 * SYN segments, each with the source address of the other.
2317 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2318 (SEQ_GT(tp->snd_una, th->th_ack) ||
2319 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2320 rstreason = BANDLIM_RST_OPENPORT;
2324 if (so->so_options & SO_DEBUG)
2325 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2329 tp->t_flags |= TF_ACKNOW;
2330 (void) tcp_output(tp);
2335 * Generate a RST, dropping incoming segment.
2336 * Make ACK acceptable to originator of segment.
2337 * Don't bother to respond if destination was broadcast/multicast.
2339 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2342 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2343 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2346 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2347 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2348 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2349 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2352 /* IPv6 anycast check is done at tcp6_input() */
2355 * Perform bandwidth limiting.
2358 if (badport_bandlim(rstreason) < 0)
2363 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2364 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2367 if (thflags & TH_ACK)
2368 /* mtod() below is safe as long as hdr dropping is delayed */
2369 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2372 if (thflags & TH_SYN)
2374 /* mtod() below is safe as long as hdr dropping is delayed */
2375 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2376 (tcp_seq)0, TH_RST|TH_ACK);
2382 * Drop space held by incoming segment and return.
2385 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2386 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2394 * Parse TCP options and place in tcpopt.
2397 tcp_dooptions(to, cp, cnt, is_syn)
2405 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2407 if (opt == TCPOPT_EOL)
2409 if (opt == TCPOPT_NOP)
2415 if (optlen < 2 || optlen > cnt)
2420 if (optlen != TCPOLEN_MAXSEG)
2424 to->to_flags |= TOF_MSS;
2425 bcopy((char *)cp + 2,
2426 (char *)&to->to_mss, sizeof(to->to_mss));
2427 to->to_mss = ntohs(to->to_mss);
2430 if (optlen != TCPOLEN_WINDOW)
2434 to->to_flags |= TOF_SCALE;
2435 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2437 case TCPOPT_TIMESTAMP:
2438 if (optlen != TCPOLEN_TIMESTAMP)
2440 to->to_flags |= TOF_TS;
2441 bcopy((char *)cp + 2,
2442 (char *)&to->to_tsval, sizeof(to->to_tsval));
2443 to->to_tsval = ntohl(to->to_tsval);
2444 bcopy((char *)cp + 6,
2445 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2446 to->to_tsecr = ntohl(to->to_tsecr);
2449 if (optlen != TCPOLEN_CC)
2451 to->to_flags |= TOF_CC;
2452 bcopy((char *)cp + 2,
2453 (char *)&to->to_cc, sizeof(to->to_cc));
2454 to->to_cc = ntohl(to->to_cc);
2457 if (optlen != TCPOLEN_CC)
2461 to->to_flags |= TOF_CCNEW;
2462 bcopy((char *)cp + 2,
2463 (char *)&to->to_cc, sizeof(to->to_cc));
2464 to->to_cc = ntohl(to->to_cc);
2467 if (optlen != TCPOLEN_CC)
2471 to->to_flags |= TOF_CCECHO;
2472 bcopy((char *)cp + 2,
2473 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2474 to->to_ccecho = ntohl(to->to_ccecho);
2483 * Pull out of band byte out of a segment so
2484 * it doesn't appear in the user's data queue.
2485 * It is still reflected in the segment length for
2486 * sequencing purposes.
2489 tcp_pulloutofband(so, th, m, off)
2493 int off; /* delayed to be droped hdrlen */
2495 int cnt = off + th->th_urp - 1;
2498 if (m->m_len > cnt) {
2499 char *cp = mtod(m, caddr_t) + cnt;
2500 struct tcpcb *tp = sototcpcb(so);
2503 tp->t_oobflags |= TCPOOB_HAVEDATA;
2504 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2506 if (m->m_flags & M_PKTHDR)
2515 panic("tcp_pulloutofband");
2519 * Collect new round-trip time estimate
2520 * and update averages and current timeout.
2523 tcp_xmit_timer(tp, rtt)
2529 tcpstat.tcps_rttupdated++;
2531 if (tp->t_srtt != 0) {
2533 * srtt is stored as fixed point with 5 bits after the
2534 * binary point (i.e., scaled by 8). The following magic
2535 * is equivalent to the smoothing algorithm in rfc793 with
2536 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2537 * point). Adjust rtt to origin 0.
2539 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2540 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2542 if ((tp->t_srtt += delta) <= 0)
2546 * We accumulate a smoothed rtt variance (actually, a
2547 * smoothed mean difference), then set the retransmit
2548 * timer to smoothed rtt + 4 times the smoothed variance.
2549 * rttvar is stored as fixed point with 4 bits after the
2550 * binary point (scaled by 16). The following is
2551 * equivalent to rfc793 smoothing with an alpha of .75
2552 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2553 * rfc793's wired-in beta.
2557 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2558 if ((tp->t_rttvar += delta) <= 0)
2560 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2561 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2564 * No rtt measurement yet - use the unsmoothed rtt.
2565 * Set the variance to half the rtt (so our first
2566 * retransmit happens at 3*rtt).
2568 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2569 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2570 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2576 * the retransmit should happen at rtt + 4 * rttvar.
2577 * Because of the way we do the smoothing, srtt and rttvar
2578 * will each average +1/2 tick of bias. When we compute
2579 * the retransmit timer, we want 1/2 tick of rounding and
2580 * 1 extra tick because of +-1/2 tick uncertainty in the
2581 * firing of the timer. The bias will give us exactly the
2582 * 1.5 tick we need. But, because the bias is
2583 * statistical, we have to test that we don't drop below
2584 * the minimum feasible timer (which is 2 ticks).
2586 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2587 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2590 * We received an ack for a packet that wasn't retransmitted;
2591 * it is probably safe to discard any error indications we've
2592 * received recently. This isn't quite right, but close enough
2593 * for now (a route might have failed after we sent a segment,
2594 * and the return path might not be symmetrical).
2596 tp->t_softerror = 0;
2600 * Determine a reasonable value for maxseg size.
2601 * If the route is known, check route for mtu.
2602 * If none, use an mss that can be handled on the outgoing
2603 * interface without forcing IP to fragment; if bigger than
2604 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2605 * to utilize large mbufs. If no route is found, route has no mtu,
2606 * or the destination isn't local, use a default, hopefully conservative
2607 * size (usually 512 or the default IP max size, but no more than the mtu
2608 * of the interface), as we can't discover anything about intervening
2609 * gateways or networks. We also initialize the congestion/slow start
2610 * window to be a single segment if the destination isn't local.
2611 * While looking at the routing entry, we also initialize other path-dependent
2612 * parameters from pre-set or cached values in the routing entry.
2614 * Also take into account the space needed for options that we
2615 * send regularly. Make maxseg shorter by that amount to assure
2616 * that we can send maxseg amount of data even when the options
2617 * are present. Store the upper limit of the length of options plus
2620 * NOTE that this routine is only called when we process an incoming
2621 * segment, for outgoing segments only tcp_mssopt is called.
2623 * In case of T/TCP, we call this routine during implicit connection
2624 * setup as well (offer = -1), to initialize maxseg from the cached
2636 struct inpcb *inp = tp->t_inpcb;
2638 struct rmxp_tao *taop;
2639 int origoffer = offer;
2641 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2642 size_t min_protoh = isipv6 ?
2643 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2644 sizeof(struct tcpiphdr);
2646 const int isipv6 = 0;
2647 const size_t min_protoh = sizeof(struct tcpiphdr);
2651 rt = tcp_rtlookup6(&inp->inp_inc);
2653 rt = tcp_rtlookup(&inp->inp_inc);
2655 tp->t_maxopd = tp->t_maxseg =
2656 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2660 so = inp->inp_socket;
2662 taop = rmx_taop(rt->rt_rmx);
2664 * Offer == -1 means that we didn't receive SYN yet,
2665 * use cached value in that case;
2668 offer = taop->tao_mssopt;
2670 * Offer == 0 means that there was no MSS on the SYN segment,
2671 * in this case we use tcp_mssdflt.
2674 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2677 * Sanity check: make sure that maxopd will be large
2678 * enough to allow some data on segments even is the
2679 * all the option space is used (40bytes). Otherwise
2680 * funny things may happen in tcp_output.
2682 offer = max(offer, 64);
2683 taop->tao_mssopt = offer;
2686 * While we're here, check if there's an initial rtt
2687 * or rttvar. Convert from the route-table units
2688 * to scaled multiples of the slow timeout timer.
2690 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2692 * XXX the lock bit for RTT indicates that the value
2693 * is also a minimum value; this is subject to time.
2695 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2696 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2697 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2698 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2699 tcpstat.tcps_usedrtt++;
2700 if (rt->rt_rmx.rmx_rttvar) {
2701 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2702 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2703 tcpstat.tcps_usedrttvar++;
2705 /* default variation is +- 1 rtt */
2707 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2709 TCPT_RANGESET(tp->t_rxtcur,
2710 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2711 tp->t_rttmin, TCPTV_REXMTMAX);
2714 * if there's an mtu associated with the route, use it
2715 * else, use the link mtu.
2717 if (rt->rt_rmx.rmx_mtu)
2718 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2721 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2723 if (!in6_localaddr(&inp->in6p_faddr))
2724 mss = min(mss, tcp_v6mssdflt);
2726 mss = ifp->if_mtu - min_protoh;
2727 if (!in_localaddr(inp->inp_faddr))
2728 mss = min(mss, tcp_mssdflt);
2731 mss = min(mss, offer);
2733 * maxopd stores the maximum length of data AND options
2734 * in a segment; maxseg is the amount of data in a normal
2735 * segment. We need to store this value (maxopd) apart
2736 * from maxseg, because now every segment carries options
2737 * and thus we normally have somewhat less data in segments.
2742 * In case of T/TCP, origoffer==-1 indicates, that no segments
2743 * were received yet. In this case we just guess, otherwise
2744 * we do the same as before T/TCP.
2746 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2748 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2749 mss -= TCPOLEN_TSTAMP_APPA;
2750 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2752 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2753 mss -= TCPOLEN_CC_APPA;
2755 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2757 mss &= ~(MCLBYTES-1);
2760 mss = mss / MCLBYTES * MCLBYTES;
2763 * If there's a pipesize, change the socket buffer
2764 * to that size. Make the socket buffers an integral
2765 * number of mss units; if the mss is larger than
2766 * the socket buffer, decrease the mss.
2769 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2771 bufsize = so->so_snd.sb_hiwat;
2775 bufsize = roundup(bufsize, mss);
2776 if (bufsize > sb_max)
2778 if (bufsize > so->so_snd.sb_hiwat)
2779 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2784 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2786 bufsize = so->so_rcv.sb_hiwat;
2787 if (bufsize > mss) {
2788 bufsize = roundup(bufsize, mss);
2789 if (bufsize > sb_max)
2791 if (bufsize > so->so_rcv.sb_hiwat)
2792 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2796 * Set the slow-start flight size depending on whether this
2797 * is a local network or not.
2800 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2801 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2802 (!isipv6 && in_localaddr(inp->inp_faddr)))
2803 tp->snd_cwnd = mss * ss_fltsz_local;
2805 tp->snd_cwnd = mss * ss_fltsz;
2807 if (rt->rt_rmx.rmx_ssthresh) {
2809 * There's some sort of gateway or interface
2810 * buffer limit on the path. Use this to set
2811 * the slow start threshhold, but set the
2812 * threshold to no less than 2*mss.
2814 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2815 tcpstat.tcps_usedssthresh++;
2820 * Determine the MSS option to send on an outgoing SYN.
2828 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2829 int min_protoh = isipv6 ?
2830 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2831 sizeof(struct tcpiphdr);
2833 const int isipv6 = 0;
2834 const size_t min_protoh = sizeof(struct tcpiphdr);
2838 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2840 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2842 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2844 return (rt->rt_ifp->if_mtu - min_protoh);
2849 * When a partial ack arrives, force the retransmission of the
2850 * next unacknowledged segment. Do not clear tp->t_dupacks.
2851 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2855 tcp_newreno_partial_ack(tp, th)
2859 tcp_seq onxt = tp->snd_nxt;
2860 u_long ocwnd = tp->snd_cwnd;
2862 callout_stop(tp->tt_rexmt);
2864 tp->snd_nxt = th->th_ack;
2866 * Set snd_cwnd to one segment beyond acknowledged offset
2867 * (tp->snd_una has not yet been updated when this function is called.)
2869 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2870 tp->t_flags |= TF_ACKNOW;
2871 (void) tcp_output(tp);
2872 tp->snd_cwnd = ocwnd;
2873 if (SEQ_GT(onxt, tp->snd_nxt))
2876 * Partial window deflation. Relies on fact that tp->snd_una
2879 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);