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 1.22 2004/03/22 06:38:17 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);
373 return (IPPROTO_DONE);
376 tcp_input(m, *offp, proto);
377 return (IPPROTO_DONE);
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 boolean_t isipv6 = FALSE;
415 tcpstat.tcps_rcvtotal++;
417 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
418 while (m->m_type == MT_TAG) {
419 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
420 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
425 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
429 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
430 ip6 = mtod(m, struct ip6_hdr *);
431 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
432 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
433 tcpstat.tcps_rcvbadsum++;
436 th = (struct tcphdr *)((caddr_t)ip6 + off0);
439 * Be proactive about unspecified IPv6 address in source.
440 * As we use all-zero to indicate unbounded/unconnected pcb,
441 * unspecified IPv6 address can be used to confuse us.
443 * Note that packets with unspecified IPv6 destination is
444 * already dropped in ip6_input.
446 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
452 * Get IP and TCP header together in first mbuf.
453 * Note: IP leaves IP header in first mbuf.
455 if (off0 > sizeof(struct ip)) {
457 off0 = sizeof(struct ip);
459 /* already checked and pulled up in ip_demux() */
460 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
461 ("TCP header not in one mbuf"));
462 ip = mtod(m, struct ip *);
463 ipov = (struct ipovly *)ip;
464 th = (struct tcphdr *)((caddr_t)ip + off0);
467 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
468 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
469 th->th_sum = m->m_pkthdr.csum_data;
471 th->th_sum = in_pseudo(ip->ip_src.s_addr,
473 htonl(m->m_pkthdr.csum_data +
476 th->th_sum ^= 0xffff;
479 * Checksum extended TCP header and data.
481 len = sizeof(struct ip) + tlen;
482 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
483 ipov->ih_len = (u_short)tlen;
484 ipov->ih_len = htons(ipov->ih_len);
485 th->th_sum = in_cksum(m, len);
488 tcpstat.tcps_rcvbadsum++;
492 /* Re-initialization for later version check */
493 ip->ip_v = IPVERSION;
498 * Check that TCP offset makes sense,
499 * pull out TCP options and adjust length. XXX
501 off = th->th_off << 2;
502 /* already checked and pulled up in ip_demux() */
503 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
504 ("bad TCP data offset"));
505 tlen -= off; /* tlen is used instead of ti->ti_len */
506 if (off > sizeof(struct tcphdr)) {
508 IP6_EXTHDR_CHECK(m, off0, off, );
509 ip6 = mtod(m, struct ip6_hdr *);
510 th = (struct tcphdr *)((caddr_t)ip6 + off0);
512 /* already pulled up in ip_demux() */
513 KASSERT(m->m_len >= sizeof(struct ip) + off,
514 ("TCP header and options not in one mbuf"));
516 optlen = off - sizeof(struct tcphdr);
517 optp = (u_char *)(th + 1);
519 thflags = th->th_flags;
521 #ifdef TCP_DROP_SYNFIN
523 * If the drop_synfin option is enabled, drop all packets with
524 * both the SYN and FIN bits set. This prevents e.g. nmap from
525 * identifying the TCP/IP stack.
527 * This is a violation of the TCP specification.
529 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
534 * Convert TCP protocol specific fields to host format.
536 th->th_seq = ntohl(th->th_seq);
537 th->th_ack = ntohl(th->th_ack);
538 th->th_win = ntohs(th->th_win);
539 th->th_urp = ntohs(th->th_urp);
542 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
543 * until after ip6_savecontrol() is called and before other functions
544 * which don't want those proto headers.
545 * Because ip6_savecontrol() is going to parse the mbuf to
546 * search for data to be passed up to user-land, it wants mbuf
547 * parameters to be unchanged.
548 * XXX: the call of ip6_savecontrol() has been obsoleted based on
549 * latest version of the advanced API (20020110).
551 drop_hdrlen = off0 + off;
554 * Locate pcb for segment.
557 /* IPFIREWALL_FORWARD section */
558 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
560 * Transparently forwarded. Pretend to be the destination.
561 * already got one like this?
563 inp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
564 ip->ip_src, th->th_sport,
565 ip->ip_dst, th->th_dport,
566 0, m->m_pkthdr.rcvif);
569 * It's new. Try to find the ambushing socket.
573 * The rest of the ipfw code stores the port in
575 * (The IP address is still in network order.)
577 in_port_t dport = next_hop->sin_port ?
578 htons(next_hop->sin_port) :
581 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
582 next_hop->sin_addr.s_addr, dport);
583 inp = in_pcblookup_hash(&tcbinfo[cpu],
584 ip->ip_src, th->th_sport,
585 next_hop->sin_addr, dport,
586 1, m->m_pkthdr.rcvif);
590 inp = in6_pcblookup_hash(&tcbinfo[0],
591 &ip6->ip6_src, th->th_sport,
592 &ip6->ip6_dst, th->th_dport,
593 1, m->m_pkthdr.rcvif);
595 inp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
596 ip->ip_src, th->th_sport,
597 ip->ip_dst, th->th_dport,
598 1, m->m_pkthdr.rcvif);
603 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
604 ipsec6stat.in_polvio++;
608 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
609 ipsecstat.in_polvio++;
616 if (inp != NULL && ipsec6_in_reject(m, inp)) {
620 if (inp != NULL && ipsec4_in_reject(m, inp)) {
627 * If the state is CLOSED (i.e., TCB does not exist) then
628 * all data in the incoming segment is discarded.
629 * If the TCB exists but is in CLOSED state, it is embryonic,
630 * but should either do a listen or a connect soon.
635 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
637 char dbuf[4 * sizeof "123"], sbuf[4 * sizeof "123"];
642 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
643 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
647 strcpy(dbuf, inet_ntoa(ip->ip_dst));
648 strcpy(sbuf, inet_ntoa(ip->ip_src));
650 switch (log_in_vain) {
652 if ((thflags & TH_SYN) == 0)
656 "Connection attempt to TCP %s:%d "
657 "from %s:%d flags:0x%02x\n",
658 dbuf, ntohs(th->th_dport), sbuf,
659 ntohs(th->th_sport), thflags);
668 if (thflags & TH_SYN)
677 rstreason = BANDLIM_RST_CLOSEDPORT;
682 rstreason = BANDLIM_RST_CLOSEDPORT;
685 if (tp->t_state == TCPS_CLOSED)
688 /* Unscale the window into a 32-bit value. */
689 if ((thflags & TH_SYN) == 0)
690 tiwin = th->th_win << tp->snd_scale;
694 so = inp->inp_socket;
697 if (so->so_options & SO_DEBUG) {
698 ostate = tp->t_state;
700 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
702 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
707 bzero((char *)&to, sizeof(to));
709 if (so->so_options & SO_ACCEPTCONN) {
710 struct in_conninfo inc;
713 inc.inc_isipv6 = (isipv6 == TRUE);
716 inc.inc6_faddr = ip6->ip6_src;
717 inc.inc6_laddr = ip6->ip6_dst;
718 inc.inc6_route.ro_rt = NULL; /* XXX */
720 inc.inc_faddr = ip->ip_src;
721 inc.inc_laddr = ip->ip_dst;
722 inc.inc_route.ro_rt = NULL; /* XXX */
724 inc.inc_fport = th->th_sport;
725 inc.inc_lport = th->th_dport;
728 * If the state is LISTEN then ignore segment if it contains
729 * a RST. If the segment contains an ACK then it is bad and
730 * send a RST. If it does not contain a SYN then it is not
731 * interesting; drop it.
733 * If the state is SYN_RECEIVED (syncache) and seg contains
734 * an ACK, but not for our SYN/ACK, send a RST. If the seg
735 * contains a RST, check the sequence number to see if it
736 * is a valid reset segment.
738 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
739 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
740 if (!syncache_expand(&inc, th, &so, m)) {
742 * No syncache entry, or ACK was not
743 * for our SYN/ACK. Send a RST.
745 tcpstat.tcps_badsyn++;
746 rstreason = BANDLIM_RST_OPENPORT;
751 * Could not complete 3-way handshake,
752 * connection is being closed down, and
753 * syncache will free mbuf.
757 * Socket is created in state SYN_RECEIVED.
758 * Continue processing segment.
763 * This is what would have happened in
764 * tcp_output() when the SYN,ACK was sent.
766 tp->snd_up = tp->snd_una;
767 tp->snd_max = tp->snd_nxt = tp->iss + 1;
768 tp->last_ack_sent = tp->rcv_nxt;
770 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
771 * until the _second_ ACK is received:
772 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
773 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
774 * move to ESTAB, set snd_wnd to tiwin.
776 tp->snd_wnd = tiwin; /* unscaled */
779 if (thflags & TH_RST) {
780 syncache_chkrst(&inc, th);
783 if (thflags & TH_ACK) {
784 syncache_badack(&inc);
785 tcpstat.tcps_badsyn++;
786 rstreason = BANDLIM_RST_OPENPORT;
793 * Segment's flags are (SYN) or (SYN|FIN).
797 * If deprecated address is forbidden,
798 * we do not accept SYN to deprecated interface
799 * address to prevent any new inbound connection from
800 * getting established.
801 * When we do not accept SYN, we send a TCP RST,
802 * with deprecated source address (instead of dropping
803 * it). We compromise it as it is much better for peer
804 * to send a RST, and RST will be the final packet
807 * If we do not forbid deprecated addresses, we accept
808 * the SYN packet. RFC2462 does not suggest dropping
810 * If we decipher RFC2462 5.5.4, it says like this:
811 * 1. use of deprecated addr with existing
812 * communication is okay - "SHOULD continue to be
814 * 2. use of it with new communication:
815 * (2a) "SHOULD NOT be used if alternate address
816 * with sufficient scope is available"
817 * (2b) nothing mentioned otherwise.
818 * Here we fall into (2b) case as we have no choice in
819 * our source address selection - we must obey the peer.
821 * The wording in RFC2462 is confusing, and there are
822 * multiple description text for deprecated address
823 * handling - worse, they are not exactly the same.
824 * I believe 5.5.4 is the best one, so we follow 5.5.4.
826 if (isipv6 && !ip6_use_deprecated) {
827 struct in6_ifaddr *ia6;
829 if ((ia6 = ip6_getdstifaddr(m)) &&
830 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
832 rstreason = BANDLIM_RST_OPENPORT;
838 * If it is from this socket, drop it, it must be forged.
839 * Don't bother responding if the destination was a broadcast.
841 if (th->th_dport == th->th_sport) {
843 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
847 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
852 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
854 * Note that it is quite possible to receive unicast
855 * link-layer packets with a broadcast IP address. Use
856 * in_broadcast() to find them.
858 if (m->m_flags & (M_BCAST|M_MCAST))
861 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
862 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
865 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
866 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
867 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
868 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
872 * SYN appears to be valid; create compressed TCP state
873 * for syncache, or perform t/tcp connection.
875 if (so->so_qlen <= so->so_qlimit) {
876 tcp_dooptions(&to, optp, optlen, 1);
877 if (!syncache_add(&inc, &to, th, &so, m))
881 * Entry added to syncache, mbuf used to
882 * send SYN,ACK packet.
886 * Segment passed TAO tests.
891 tp->t_starttime = ticks;
892 tp->t_state = TCPS_ESTABLISHED;
895 * If there is a FIN, or if there is data and the
896 * connection is local, then delay SYN,ACK(SYN) in
897 * the hope of piggy-backing it on a response
898 * segment. Otherwise must send ACK now in case
899 * the other side is slow starting.
902 ((thflags & TH_FIN) ||
904 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
905 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
906 callout_reset(tp->tt_delack, tcp_delacktime,
907 tcp_timer_delack, tp);
908 tp->t_flags |= TF_NEEDSYN;
910 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
912 tcpstat.tcps_connects++;
920 /* XXX temp debugging */
921 /* should not happen - syncache should pick up these connections */
922 if (tp->t_state == TCPS_LISTEN)
923 panic("tcp_input: TCPS_LISTEN");
926 * Segment received on connection.
927 * Reset idle time and keep-alive timer.
929 tp->t_rcvtime = ticks;
930 if (TCPS_HAVEESTABLISHED(tp->t_state))
931 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
935 * XXX this is tradtitional behavior, may need to be cleaned up.
937 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
938 if (thflags & TH_SYN) {
939 if (to.to_flags & TOF_SCALE) {
940 tp->t_flags |= TF_RCVD_SCALE;
941 tp->requested_s_scale = to.to_requested_s_scale;
943 if (to.to_flags & TOF_TS) {
944 tp->t_flags |= TF_RCVD_TSTMP;
945 tp->ts_recent = to.to_tsval;
946 tp->ts_recent_age = ticks;
948 if (to.to_flags & (TOF_CC|TOF_CCNEW))
949 tp->t_flags |= TF_RCVD_CC;
950 if (to.to_flags & TOF_MSS)
951 tcp_mss(tp, to.to_mss);
955 * Header prediction: check for the two common cases
956 * of a uni-directional data xfer. If the packet has
957 * no control flags, is in-sequence, the window didn't
958 * change and we're not retransmitting, it's a
959 * candidate. If the length is zero and the ack moved
960 * forward, we're the sender side of the xfer. Just
961 * free the data acked & wake any higher level process
962 * that was blocked waiting for space. If the length
963 * is non-zero and the ack didn't move, we're the
964 * receiver side. If we're getting packets in-order
965 * (the reassembly queue is empty), add the data to
966 * the socket buffer and note that we need a delayed ack.
967 * Make sure that the hidden state-flags are also off.
968 * Since we check for TCPS_ESTABLISHED above, it can only
971 if (tp->t_state == TCPS_ESTABLISHED &&
972 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
973 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
974 ((to.to_flags & TOF_TS) == 0 ||
975 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
977 * Using the CC option is compulsory if once started:
978 * the segment is OK if no T/TCP was negotiated or
979 * if the segment has a CC option equal to CCrecv
981 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
982 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
983 th->th_seq == tp->rcv_nxt &&
984 tiwin && tiwin == tp->snd_wnd &&
985 tp->snd_nxt == tp->snd_max) {
988 * If last ACK falls within this segment's sequence numbers,
989 * record the timestamp.
990 * NOTE that the test is modified according to the latest
991 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
993 if ((to.to_flags & TOF_TS) != 0 &&
994 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
995 tp->ts_recent_age = ticks;
996 tp->ts_recent = to.to_tsval;
1000 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1001 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1002 tp->snd_cwnd >= tp->snd_wnd &&
1003 ((!tcp_do_newreno &&
1004 tp->t_dupacks < tcprexmtthresh) ||
1005 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
1007 * this is a pure ack for outstanding data.
1009 ++tcpstat.tcps_predack;
1011 * "bad retransmit" recovery
1013 * If Eifel detection applies, then
1014 * it is deterministic, so use it
1015 * unconditionally over the old heuristic.
1016 * Otherwise, fall back to the old heuristic.
1018 if (tcp_do_eifel_detect &&
1019 (to.to_flags & TOF_TS) && to.to_tsecr &&
1020 (tp->t_flags & TF_FIRSTACCACK)) {
1021 /* Eifel detection applicable. */
1022 if (to.to_tsecr < tp->t_rexmtTS) {
1023 tcp_revert_congestion_state(tp);
1024 ++tcpstat.tcps_eifeldetected;
1026 } else if (tp->t_rxtshift == 1 &&
1027 ticks < tp->t_badrxtwin) {
1028 tcp_revert_congestion_state(tp);
1029 ++tcpstat.tcps_rttdetected;
1031 tp->t_flags &= ~(TF_FIRSTACCACK |
1032 TF_FASTREXMT | TF_EARLYREXMT);
1034 * Recalculate the retransmit timer / rtt.
1036 * Some machines (certain windows boxes)
1037 * send broken timestamp replies during the
1038 * SYN+ACK phase, ignore timestamps of 0.
1040 if ((to.to_flags & TOF_TS) != 0 &&
1043 ticks - to.to_tsecr + 1);
1044 } else if (tp->t_rtttime &&
1045 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1047 ticks - tp->t_rtttime);
1049 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1050 acked = th->th_ack - tp->snd_una;
1051 tcpstat.tcps_rcvackpack++;
1052 tcpstat.tcps_rcvackbyte += acked;
1053 sbdrop(&so->so_snd, acked);
1054 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1055 SEQ_LEQ(th->th_ack, tp->snd_recover))
1056 tp->snd_recover = th->th_ack - 1;
1057 tp->snd_una = th->th_ack;
1060 ND6_HINT(tp); /* some progress has been done */
1063 * If all outstanding data are acked, stop
1064 * retransmit timer, otherwise restart timer
1065 * using current (possibly backed-off) value.
1066 * If process is waiting for space,
1067 * wakeup/selwakeup/signal. If data
1068 * are ready to send, let tcp_output
1069 * decide between more output or persist.
1071 if (tp->snd_una == tp->snd_max)
1072 callout_stop(tp->tt_rexmt);
1073 else if (!callout_active(tp->tt_persist))
1074 callout_reset(tp->tt_rexmt,
1076 tcp_timer_rexmt, tp);
1079 if (so->so_snd.sb_cc)
1080 (void) tcp_output(tp);
1083 } else if (th->th_ack == tp->snd_una &&
1084 LIST_EMPTY(&tp->t_segq) &&
1085 tlen <= sbspace(&so->so_rcv)) {
1087 * this is a pure, in-sequence data packet
1088 * with nothing on the reassembly queue and
1089 * we have enough buffer space to take it.
1091 ++tcpstat.tcps_preddat;
1092 tp->rcv_nxt += tlen;
1093 tcpstat.tcps_rcvpack++;
1094 tcpstat.tcps_rcvbyte += tlen;
1095 ND6_HINT(tp); /* some progress has been done */
1097 * Add data to socket buffer.
1099 if (so->so_state & SS_CANTRCVMORE) {
1102 m_adj(m, drop_hdrlen); /* delayed header drop */
1103 sbappend(&so->so_rcv, m);
1106 if (DELAY_ACK(tp)) {
1107 callout_reset(tp->tt_delack, tcp_delacktime,
1108 tcp_timer_delack, tp);
1110 tp->t_flags |= TF_ACKNOW;
1118 * Calculate amount of space in receive window,
1119 * and then do TCP input processing.
1120 * Receive window is amount of space in rcv queue,
1121 * but not less than advertised window.
1125 win = sbspace(&so->so_rcv);
1128 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1131 switch (tp->t_state) {
1134 * If the state is SYN_RECEIVED:
1135 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1137 case TCPS_SYN_RECEIVED:
1138 if ((thflags & TH_ACK) &&
1139 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1140 SEQ_GT(th->th_ack, tp->snd_max))) {
1141 rstreason = BANDLIM_RST_OPENPORT;
1147 * If the state is SYN_SENT:
1148 * if seg contains an ACK, but not for our SYN, drop the input.
1149 * if seg contains a RST, then drop the connection.
1150 * if seg does not contain SYN, then drop it.
1151 * Otherwise this is an acceptable SYN segment
1152 * initialize tp->rcv_nxt and tp->irs
1153 * if seg contains ack then advance tp->snd_una
1154 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1155 * arrange for segment to be acked (eventually)
1156 * continue processing rest of data/controls, beginning with URG
1159 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1160 taop = &tao_noncached;
1161 bzero(taop, sizeof(*taop));
1164 if ((thflags & TH_ACK) &&
1165 (SEQ_LEQ(th->th_ack, tp->iss) ||
1166 SEQ_GT(th->th_ack, tp->snd_max))) {
1168 * If we have a cached CCsent for the remote host,
1169 * hence we haven't just crashed and restarted,
1170 * do not send a RST. This may be a retransmission
1171 * from the other side after our earlier ACK was lost.
1172 * Our new SYN, when it arrives, will serve as the
1175 if (taop->tao_ccsent != 0)
1178 rstreason = BANDLIM_UNLIMITED;
1182 if (thflags & TH_RST) {
1183 if (thflags & TH_ACK)
1184 tp = tcp_drop(tp, ECONNREFUSED);
1187 if ((thflags & TH_SYN) == 0)
1189 tp->snd_wnd = th->th_win; /* initial send window */
1190 tp->cc_recv = to.to_cc; /* foreign CC */
1192 tp->irs = th->th_seq;
1194 if (thflags & TH_ACK) {
1196 * Our SYN was acked. If segment contains CC.ECHO
1197 * option, check it to make sure this segment really
1198 * matches our SYN. If not, just drop it as old
1199 * duplicate, but send an RST if we're still playing
1200 * by the old rules. If no CC.ECHO option, make sure
1201 * we don't get fooled into using T/TCP.
1203 if (to.to_flags & TOF_CCECHO) {
1204 if (tp->cc_send != to.to_ccecho) {
1205 if (taop->tao_ccsent != 0)
1208 rstreason = BANDLIM_UNLIMITED;
1213 tp->t_flags &= ~TF_RCVD_CC;
1214 tcpstat.tcps_connects++;
1216 /* Do window scaling on this connection? */
1217 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1218 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1219 tp->snd_scale = tp->requested_s_scale;
1220 tp->rcv_scale = tp->request_r_scale;
1222 /* Segment is acceptable, update cache if undefined. */
1223 if (taop->tao_ccsent == 0)
1224 taop->tao_ccsent = to.to_ccecho;
1226 tp->rcv_adv += tp->rcv_wnd;
1227 tp->snd_una++; /* SYN is acked */
1229 * If there's data, delay ACK; if there's also a FIN
1230 * ACKNOW will be turned on later.
1232 if (DELAY_ACK(tp) && tlen != 0)
1233 callout_reset(tp->tt_delack, tcp_delacktime,
1234 tcp_timer_delack, tp);
1236 tp->t_flags |= TF_ACKNOW;
1238 * Received <SYN,ACK> in SYN_SENT[*] state.
1240 * SYN_SENT --> ESTABLISHED
1241 * SYN_SENT* --> FIN_WAIT_1
1243 tp->t_starttime = ticks;
1244 if (tp->t_flags & TF_NEEDFIN) {
1245 tp->t_state = TCPS_FIN_WAIT_1;
1246 tp->t_flags &= ~TF_NEEDFIN;
1249 tp->t_state = TCPS_ESTABLISHED;
1250 callout_reset(tp->tt_keep, tcp_keepidle,
1251 tcp_timer_keep, tp);
1255 * Received initial SYN in SYN-SENT[*] state =>
1256 * simultaneous open. If segment contains CC option
1257 * and there is a cached CC, apply TAO test.
1258 * If it succeeds, connection is * half-synchronized.
1259 * Otherwise, do 3-way handshake:
1260 * SYN-SENT -> SYN-RECEIVED
1261 * SYN-SENT* -> SYN-RECEIVED*
1262 * If there was no CC option, clear cached CC value.
1264 tp->t_flags |= TF_ACKNOW;
1265 callout_stop(tp->tt_rexmt);
1266 if (to.to_flags & TOF_CC) {
1267 if (taop->tao_cc != 0 &&
1268 CC_GT(to.to_cc, taop->tao_cc)) {
1270 * update cache and make transition:
1271 * SYN-SENT -> ESTABLISHED*
1272 * SYN-SENT* -> FIN-WAIT-1*
1274 taop->tao_cc = to.to_cc;
1275 tp->t_starttime = ticks;
1276 if (tp->t_flags & TF_NEEDFIN) {
1277 tp->t_state = TCPS_FIN_WAIT_1;
1278 tp->t_flags &= ~TF_NEEDFIN;
1280 tp->t_state = TCPS_ESTABLISHED;
1281 callout_reset(tp->tt_keep,
1286 tp->t_flags |= TF_NEEDSYN;
1288 tp->t_state = TCPS_SYN_RECEIVED;
1290 /* CC.NEW or no option => invalidate cache */
1292 tp->t_state = TCPS_SYN_RECEIVED;
1298 * Advance th->th_seq to correspond to first data byte.
1299 * If data, trim to stay within window,
1300 * dropping FIN if necessary.
1303 if (tlen > tp->rcv_wnd) {
1304 todrop = tlen - tp->rcv_wnd;
1308 tcpstat.tcps_rcvpackafterwin++;
1309 tcpstat.tcps_rcvbyteafterwin += todrop;
1311 tp->snd_wl1 = th->th_seq - 1;
1312 tp->rcv_up = th->th_seq;
1314 * Client side of transaction: already sent SYN and data.
1315 * If the remote host used T/TCP to validate the SYN,
1316 * our data will be ACK'd; if so, enter normal data segment
1317 * processing in the middle of step 5, ack processing.
1318 * Otherwise, goto step 6.
1320 if (thflags & TH_ACK)
1326 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1327 * if segment contains a SYN and CC [not CC.NEW] option:
1328 * if state == TIME_WAIT and connection duration > MSL,
1329 * drop packet and send RST;
1331 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1332 * ack the FIN (and data) in retransmission queue.
1333 * Complete close and delete TCPCB. Then reprocess
1334 * segment, hoping to find new TCPCB in LISTEN state;
1336 * else must be old SYN; drop it.
1337 * else do normal processing.
1341 case TCPS_TIME_WAIT:
1342 if ((thflags & TH_SYN) &&
1343 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1344 if (tp->t_state == TCPS_TIME_WAIT &&
1345 (ticks - tp->t_starttime) > tcp_msl) {
1346 rstreason = BANDLIM_UNLIMITED;
1349 if (CC_GT(to.to_cc, tp->cc_recv)) {
1356 break; /* continue normal processing */
1360 * States other than LISTEN or SYN_SENT.
1361 * First check the RST flag and sequence number since reset segments
1362 * are exempt from the timestamp and connection count tests. This
1363 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1364 * below which allowed reset segments in half the sequence space
1365 * to fall though and be processed (which gives forged reset
1366 * segments with a random sequence number a 50 percent chance of
1367 * killing a connection).
1368 * Then check timestamp, if present.
1369 * Then check the connection count, if present.
1370 * Then check that at least some bytes of segment are within
1371 * receive window. If segment begins before rcv_nxt,
1372 * drop leading data (and SYN); if nothing left, just ack.
1375 * If the RST bit is set, check the sequence number to see
1376 * if this is a valid reset segment.
1378 * In all states except SYN-SENT, all reset (RST) segments
1379 * are validated by checking their SEQ-fields. A reset is
1380 * valid if its sequence number is in the window.
1381 * Note: this does not take into account delayed ACKs, so
1382 * we should test against last_ack_sent instead of rcv_nxt.
1383 * The sequence number in the reset segment is normally an
1384 * echo of our outgoing acknowlegement numbers, but some hosts
1385 * send a reset with the sequence number at the rightmost edge
1386 * of our receive window, and we have to handle this case.
1387 * If we have multiple segments in flight, the intial reset
1388 * segment sequence numbers will be to the left of last_ack_sent,
1389 * but they will eventually catch up.
1390 * In any case, it never made sense to trim reset segments to
1391 * fit the receive window since RFC 1122 says:
1392 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1394 * A TCP SHOULD allow a received RST segment to include data.
1397 * It has been suggested that a RST segment could contain
1398 * ASCII text that encoded and explained the cause of the
1399 * RST. No standard has yet been established for such
1402 * If the reset segment passes the sequence number test examine
1404 * SYN_RECEIVED STATE:
1405 * If passive open, return to LISTEN state.
1406 * If active open, inform user that connection was refused.
1407 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1408 * Inform user that connection was reset, and close tcb.
1409 * CLOSING, LAST_ACK STATES:
1412 * Drop the segment - see Stevens, vol. 2, p. 964 and
1415 if (thflags & TH_RST) {
1416 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1417 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1418 switch (tp->t_state) {
1420 case TCPS_SYN_RECEIVED:
1421 so->so_error = ECONNREFUSED;
1424 case TCPS_ESTABLISHED:
1425 case TCPS_FIN_WAIT_1:
1426 case TCPS_FIN_WAIT_2:
1427 case TCPS_CLOSE_WAIT:
1428 so->so_error = ECONNRESET;
1430 tp->t_state = TCPS_CLOSED;
1431 tcpstat.tcps_drops++;
1440 case TCPS_TIME_WAIT:
1448 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1449 * and it's less than ts_recent, drop it.
1451 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1452 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1454 /* Check to see if ts_recent is over 24 days old. */
1455 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1457 * Invalidate ts_recent. If this segment updates
1458 * ts_recent, the age will be reset later and ts_recent
1459 * will get a valid value. If it does not, setting
1460 * ts_recent to zero will at least satisfy the
1461 * requirement that zero be placed in the timestamp
1462 * echo reply when ts_recent isn't valid. The
1463 * age isn't reset until we get a valid ts_recent
1464 * because we don't want out-of-order segments to be
1465 * dropped when ts_recent is old.
1469 tcpstat.tcps_rcvduppack++;
1470 tcpstat.tcps_rcvdupbyte += tlen;
1471 tcpstat.tcps_pawsdrop++;
1480 * If T/TCP was negotiated and the segment doesn't have CC,
1481 * or if its CC is wrong then drop the segment.
1482 * RST segments do not have to comply with this.
1484 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1485 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1489 * In the SYN-RECEIVED state, validate that the packet belongs to
1490 * this connection before trimming the data to fit the receive
1491 * window. Check the sequence number versus IRS since we know
1492 * the sequence numbers haven't wrapped. This is a partial fix
1493 * for the "LAND" DoS attack.
1495 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1496 rstreason = BANDLIM_RST_OPENPORT;
1500 todrop = tp->rcv_nxt - th->th_seq;
1502 if (thflags & TH_SYN) {
1512 * Following if statement from Stevens, vol. 2, p. 960.
1515 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1517 * Any valid FIN must be to the left of the window.
1518 * At this point the FIN must be a duplicate or out
1519 * of sequence; drop it.
1524 * Send an ACK to resynchronize and drop any data.
1525 * But keep on processing for RST or ACK.
1527 tp->t_flags |= TF_ACKNOW;
1529 tcpstat.tcps_rcvduppack++;
1530 tcpstat.tcps_rcvdupbyte += todrop;
1532 tcpstat.tcps_rcvpartduppack++;
1533 tcpstat.tcps_rcvpartdupbyte += todrop;
1535 drop_hdrlen += todrop; /* drop from the top afterwards */
1536 th->th_seq += todrop;
1538 if (th->th_urp > todrop)
1539 th->th_urp -= todrop;
1547 * If new data are received on a connection after the
1548 * user processes are gone, then RST the other end.
1550 if ((so->so_state & SS_NOFDREF) &&
1551 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1553 tcpstat.tcps_rcvafterclose++;
1554 rstreason = BANDLIM_UNLIMITED;
1559 * If segment ends after window, drop trailing data
1560 * (and PUSH and FIN); if nothing left, just ACK.
1562 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1564 tcpstat.tcps_rcvpackafterwin++;
1565 if (todrop >= tlen) {
1566 tcpstat.tcps_rcvbyteafterwin += tlen;
1568 * If a new connection request is received
1569 * while in TIME_WAIT, drop the old connection
1570 * and start over if the sequence numbers
1571 * are above the previous ones.
1573 if (thflags & TH_SYN &&
1574 tp->t_state == TCPS_TIME_WAIT &&
1575 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1580 * If window is closed can only take segments at
1581 * window edge, and have to drop data and PUSH from
1582 * incoming segments. Continue processing, but
1583 * remember to ack. Otherwise, drop segment
1586 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1587 tp->t_flags |= TF_ACKNOW;
1588 tcpstat.tcps_rcvwinprobe++;
1592 tcpstat.tcps_rcvbyteafterwin += todrop;
1595 thflags &= ~(TH_PUSH|TH_FIN);
1599 * If last ACK falls within this segment's sequence numbers,
1600 * record its timestamp.
1601 * NOTE that the test is modified according to the latest
1602 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1604 if ((to.to_flags & TOF_TS) != 0 &&
1605 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1606 tp->ts_recent_age = ticks;
1607 tp->ts_recent = to.to_tsval;
1611 * If a SYN is in the window, then this is an
1612 * error and we send an RST and drop the connection.
1614 if (thflags & TH_SYN) {
1615 tp = tcp_drop(tp, ECONNRESET);
1616 rstreason = BANDLIM_UNLIMITED;
1621 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1622 * flag is on (half-synchronized state), then queue data for
1623 * later processing; else drop segment and return.
1625 if ((thflags & TH_ACK) == 0) {
1626 if (tp->t_state == TCPS_SYN_RECEIVED ||
1627 (tp->t_flags & TF_NEEDSYN))
1636 switch (tp->t_state) {
1639 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1640 * ESTABLISHED state and continue processing.
1641 * The ACK was checked above.
1643 case TCPS_SYN_RECEIVED:
1645 tcpstat.tcps_connects++;
1647 /* Do window scaling? */
1648 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1649 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1650 tp->snd_scale = tp->requested_s_scale;
1651 tp->rcv_scale = tp->request_r_scale;
1654 * Upon successful completion of 3-way handshake,
1655 * update cache.CC if it was undefined, pass any queued
1656 * data to the user, and advance state appropriately.
1658 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1660 taop->tao_cc = tp->cc_recv;
1664 * SYN-RECEIVED -> ESTABLISHED
1665 * SYN-RECEIVED* -> FIN-WAIT-1
1667 tp->t_starttime = ticks;
1668 if (tp->t_flags & TF_NEEDFIN) {
1669 tp->t_state = TCPS_FIN_WAIT_1;
1670 tp->t_flags &= ~TF_NEEDFIN;
1672 tp->t_state = TCPS_ESTABLISHED;
1673 callout_reset(tp->tt_keep, tcp_keepidle,
1674 tcp_timer_keep, tp);
1677 * If segment contains data or ACK, will call tcp_reass()
1678 * later; if not, do so now to pass queued data to user.
1680 if (tlen == 0 && (thflags & TH_FIN) == 0)
1681 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1683 tp->snd_wl1 = th->th_seq - 1;
1687 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1688 * ACKs. If the ack is in the range
1689 * tp->snd_una < th->th_ack <= tp->snd_max
1690 * then advance tp->snd_una to th->th_ack and drop
1691 * data from the retransmission queue. If this ACK reflects
1692 * more up to date window information we update our window information.
1694 case TCPS_ESTABLISHED:
1695 case TCPS_FIN_WAIT_1:
1696 case TCPS_FIN_WAIT_2:
1697 case TCPS_CLOSE_WAIT:
1700 case TCPS_TIME_WAIT:
1702 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1703 if (tlen == 0 && tiwin == tp->snd_wnd) {
1704 tcpstat.tcps_rcvdupack++;
1706 * If we have outstanding data (other than
1707 * a window probe), this is a completely
1708 * duplicate ack (ie, window info didn't
1709 * change), the ack is the biggest we've
1710 * seen and we've seen exactly our rexmt
1711 * threshhold of them, assume a packet
1712 * has been dropped and retransmit it.
1713 * Kludge snd_nxt & the congestion
1714 * window so we send only this one
1717 * We know we're losing at the current
1718 * window size so do congestion avoidance
1719 * (set ssthresh to half the current window
1720 * and pull our congestion window back to
1721 * the new ssthresh).
1723 * Dup acks mean that packets have left the
1724 * network (they're now cached at the receiver)
1725 * so bump cwnd by the amount in the receiver
1726 * to keep a constant cwnd packets in the
1729 if (!callout_active(tp->tt_rexmt) ||
1730 th->th_ack != tp->snd_una)
1732 else if (++tp->t_dupacks > tcprexmtthresh ||
1734 IN_FASTRECOVERY(tp))) {
1735 tp->snd_cwnd += tp->t_maxseg;
1736 (void) tcp_output(tp);
1738 } else if (tp->t_dupacks == tcprexmtthresh) {
1742 if (tcp_do_newreno &&
1749 if (tcp_do_eifel_detect &&
1750 (tp->t_flags & TF_RCVD_TSTMP)) {
1751 tcp_save_congestion_state(tp);
1752 tp->t_flags |= TF_FASTREXMT;
1754 win = min(tp->snd_wnd, tp->snd_cwnd) /
1758 tp->snd_ssthresh = win * tp->t_maxseg;
1759 ENTER_FASTRECOVERY(tp);
1760 tp->snd_recover = tp->snd_max;
1761 callout_stop(tp->tt_rexmt);
1764 tp->snd_nxt = th->th_ack;
1765 tp->snd_cwnd = tp->t_maxseg;
1766 (void) tcp_output(tp);
1767 ++tcpstat.tcps_sndfastrexmit;
1768 KASSERT(tp->snd_limited <= 2,
1769 ("tp->snd_limited too big"));
1770 tp->snd_cwnd = tp->snd_ssthresh +
1772 (tp->t_dupacks - tp->snd_limited));
1773 if (SEQ_GT(onxt, tp->snd_nxt))
1776 } else if (tcp_do_limitedtransmit) {
1777 u_long oldcwnd = tp->snd_cwnd;
1778 tcp_seq oldsndmax = tp->snd_max;
1779 /* outstanding data */
1781 tp->snd_max - tp->snd_una;
1784 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1786 KASSERT(tp->t_dupacks == 1 ||
1788 ("dupacks not 1 or 2"));
1789 if (tp->t_dupacks == 1)
1790 tp->snd_limited = 0;
1791 tp->snd_cwnd = ownd +
1792 (tp->t_dupacks - tp->snd_limited) *
1794 (void) tcp_output(tp);
1795 tp->snd_cwnd = oldcwnd;
1796 sent = tp->snd_max - oldsndmax;
1797 if (sent > tp->t_maxseg) {
1798 KASSERT((tp->t_dupacks == 2 &&
1799 tp->snd_limited == 0) ||
1800 (sent == tp->t_maxseg + 1 &&
1801 tp->t_flags & TF_SENTFIN),
1805 ("sent too many segments"));
1806 tp->snd_limited = 2;
1807 tcpstat.tcps_sndlimited += 2;
1808 } else if (sent > 0) {
1810 ++tcpstat.tcps_sndlimited;
1811 } else if (tcp_do_early_retransmit &&
1812 (tcp_do_eifel_detect &&
1813 (tp->t_flags & TF_RCVD_TSTMP)) &&
1815 tp->t_dupacks + 1 >=
1816 iceildiv(ownd, tp->t_maxseg)) {
1817 ++tcpstat.tcps_sndearlyrexmit;
1818 tp->t_flags |= TF_EARLYREXMT;
1819 goto fastretransmit;
1828 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1831 * If the congestion window was inflated to account
1832 * for the other side's cached packets, retract it.
1834 if (tcp_do_newreno) {
1835 if (IN_FASTRECOVERY(tp)) {
1836 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1837 tcp_newreno_partial_ack(tp, th);
1840 * Window inflation should have left us
1841 * with approximately snd_ssthresh
1843 * But in case we would be inclined to
1844 * send a burst, better to do it via
1845 * the slow start mechanism.
1847 if (SEQ_GT(th->th_ack +
1850 tp->snd_cwnd = tp->snd_max -
1854 tp->snd_cwnd = tp->snd_ssthresh;
1858 if (tp->t_dupacks >= tcprexmtthresh &&
1859 tp->snd_cwnd > tp->snd_ssthresh)
1860 tp->snd_cwnd = tp->snd_ssthresh;
1863 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1864 tcpstat.tcps_rcvacktoomuch++;
1868 * If we reach this point, ACK is not a duplicate,
1869 * i.e., it ACKs something we sent.
1871 if (tp->t_flags & TF_NEEDSYN) {
1873 * T/TCP: Connection was half-synchronized, and our
1874 * SYN has been ACK'd (so connection is now fully
1875 * synchronized). Go to non-starred state,
1876 * increment snd_una for ACK of SYN, and check if
1877 * we can do window scaling.
1879 tp->t_flags &= ~TF_NEEDSYN;
1881 /* Do window scaling? */
1882 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1883 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1884 tp->snd_scale = tp->requested_s_scale;
1885 tp->rcv_scale = tp->request_r_scale;
1890 acked = th->th_ack - tp->snd_una;
1891 tcpstat.tcps_rcvackpack++;
1892 tcpstat.tcps_rcvackbyte += acked;
1895 * If we just performed our first retransmit, and the ACK
1896 * arrives within our recovery window, then it was a mistake
1897 * to do the retransmit in the first place. Recover our
1898 * original cwnd and ssthresh, and proceed to transmit where
1901 if (tcp_do_eifel_detect && acked &&
1902 (to.to_flags & TOF_TS) && to.to_tsecr &&
1903 (tp->t_flags & TF_FIRSTACCACK)) {
1904 /* Eifel detection applicable. */
1905 if (to.to_tsecr < tp->t_rexmtTS) {
1906 ++tcpstat.tcps_eifeldetected;
1907 tcp_revert_congestion_state(tp);
1908 if (tp->t_rxtshift == 1 &&
1909 ticks >= tp->t_badrxtwin)
1910 ++tcpstat.tcps_rttcantdetect;
1912 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1913 tcp_revert_congestion_state(tp);
1914 ++tcpstat.tcps_rttdetected;
1918 * If we have a timestamp reply, update smoothed
1919 * round trip time. If no timestamp is present but
1920 * transmit timer is running and timed sequence
1921 * number was acked, update smoothed round trip time.
1922 * Since we now have an rtt measurement, cancel the
1923 * timer backoff (cf., Phil Karn's retransmit alg.).
1924 * Recompute the initial retransmit timer.
1926 * Some machines (certain windows boxes) send broken
1927 * timestamp replies during the SYN+ACK phase, ignore
1930 if ((to.to_flags & TOF_TS) != 0 &&
1932 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1933 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1934 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1936 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1939 * If all outstanding data is acked, stop retransmit
1940 * timer and remember to restart (more output or persist).
1941 * If there is more data to be acked, restart retransmit
1942 * timer, using current (possibly backed-off) value.
1944 if (th->th_ack == tp->snd_max) {
1945 callout_stop(tp->tt_rexmt);
1947 } else if (!callout_active(tp->tt_persist))
1948 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1949 tcp_timer_rexmt, tp);
1952 * If no data (only SYN) was ACK'd,
1953 * skip rest of ACK processing.
1958 /* Stop looking for an acceptable ACK since one was received. */
1959 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
1962 * When new data is acked, open the congestion window.
1963 * If the window gives us less than ssthresh packets
1964 * in flight, open exponentially (maxseg per packet).
1965 * Otherwise open linearly: maxseg per window
1966 * (maxseg^2 / cwnd per packet).
1968 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1969 u_int cw = tp->snd_cwnd;
1970 u_int incr = tp->t_maxseg;
1971 if (cw > tp->snd_ssthresh)
1972 incr = incr * incr / cw;
1973 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1975 if (acked > so->so_snd.sb_cc) {
1976 tp->snd_wnd -= so->so_snd.sb_cc;
1977 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1980 sbdrop(&so->so_snd, acked);
1981 tp->snd_wnd -= acked;
1985 /* detect una wraparound */
1986 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1987 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1988 SEQ_LEQ(th->th_ack, tp->snd_recover))
1989 tp->snd_recover = th->th_ack - 1;
1990 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1991 SEQ_GEQ(th->th_ack, tp->snd_recover))
1992 EXIT_FASTRECOVERY(tp);
1993 tp->snd_una = th->th_ack;
1994 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1995 tp->snd_nxt = tp->snd_una;
1997 switch (tp->t_state) {
2000 * In FIN_WAIT_1 STATE in addition to the processing
2001 * for the ESTABLISHED state if our FIN is now acknowledged
2002 * then enter FIN_WAIT_2.
2004 case TCPS_FIN_WAIT_1:
2005 if (ourfinisacked) {
2007 * If we can't receive any more
2008 * data, then closing user can proceed.
2009 * Starting the timer is contrary to the
2010 * specification, but if we don't get a FIN
2011 * we'll hang forever.
2013 if (so->so_state & SS_CANTRCVMORE) {
2014 soisdisconnected(so);
2015 callout_reset(tp->tt_2msl, tcp_maxidle,
2016 tcp_timer_2msl, tp);
2018 tp->t_state = TCPS_FIN_WAIT_2;
2023 * In CLOSING STATE in addition to the processing for
2024 * the ESTABLISHED state if the ACK acknowledges our FIN
2025 * then enter the TIME-WAIT state, otherwise ignore
2029 if (ourfinisacked) {
2030 tp->t_state = TCPS_TIME_WAIT;
2031 tcp_canceltimers(tp);
2032 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2033 if (tp->cc_recv != 0 &&
2034 (ticks - tp->t_starttime) < tcp_msl)
2035 callout_reset(tp->tt_2msl,
2038 tcp_timer_2msl, tp);
2040 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2041 tcp_timer_2msl, tp);
2042 soisdisconnected(so);
2047 * In LAST_ACK, we may still be waiting for data to drain
2048 * and/or to be acked, as well as for the ack of our FIN.
2049 * If our FIN is now acknowledged, delete the TCB,
2050 * enter the closed state and return.
2053 if (ourfinisacked) {
2060 * In TIME_WAIT state the only thing that should arrive
2061 * is a retransmission of the remote FIN. Acknowledge
2062 * it and restart the finack timer.
2064 case TCPS_TIME_WAIT:
2065 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2066 tcp_timer_2msl, tp);
2073 * Update window information.
2074 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2076 if ((thflags & TH_ACK) &&
2077 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2078 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2079 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2080 /* keep track of pure window updates */
2082 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2083 tcpstat.tcps_rcvwinupd++;
2084 tp->snd_wnd = tiwin;
2085 tp->snd_wl1 = th->th_seq;
2086 tp->snd_wl2 = th->th_ack;
2087 if (tp->snd_wnd > tp->max_sndwnd)
2088 tp->max_sndwnd = tp->snd_wnd;
2093 * Process segments with URG.
2095 if ((thflags & TH_URG) && th->th_urp &&
2096 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2098 * This is a kludge, but if we receive and accept
2099 * random urgent pointers, we'll crash in
2100 * soreceive. It's hard to imagine someone
2101 * actually wanting to send this much urgent data.
2103 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2104 th->th_urp = 0; /* XXX */
2105 thflags &= ~TH_URG; /* XXX */
2106 goto dodata; /* XXX */
2109 * If this segment advances the known urgent pointer,
2110 * then mark the data stream. This should not happen
2111 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2112 * a FIN has been received from the remote side.
2113 * In these states we ignore the URG.
2115 * According to RFC961 (Assigned Protocols),
2116 * the urgent pointer points to the last octet
2117 * of urgent data. We continue, however,
2118 * to consider it to indicate the first octet
2119 * of data past the urgent section as the original
2120 * spec states (in one of two places).
2122 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2123 tp->rcv_up = th->th_seq + th->th_urp;
2124 so->so_oobmark = so->so_rcv.sb_cc +
2125 (tp->rcv_up - tp->rcv_nxt) - 1;
2126 if (so->so_oobmark == 0)
2127 so->so_state |= SS_RCVATMARK;
2129 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2132 * Remove out of band data so doesn't get presented to user.
2133 * This can happen independent of advancing the URG pointer,
2134 * but if two URG's are pending at once, some out-of-band
2135 * data may creep in... ick.
2137 if (th->th_urp <= (u_long)tlen
2139 && (so->so_options & SO_OOBINLINE) == 0
2142 tcp_pulloutofband(so, th, m,
2143 drop_hdrlen); /* hdr drop is delayed */
2146 * If no out of band data is expected,
2147 * pull receive urgent pointer along
2148 * with the receive window.
2150 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2151 tp->rcv_up = tp->rcv_nxt;
2156 * Process the segment text, merging it into the TCP sequencing queue,
2157 * and arranging for acknowledgment of receipt if necessary.
2158 * This process logically involves adjusting tp->rcv_wnd as data
2159 * is presented to the user (this happens in tcp_usrreq.c,
2160 * case PRU_RCVD). If a FIN has already been received on this
2161 * connection then we just ignore the text.
2163 if ((tlen || (thflags & TH_FIN)) &&
2164 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2165 m_adj(m, drop_hdrlen); /* delayed header drop */
2167 * Insert segment which includes th into TCP reassembly queue
2168 * with control block tp. Set thflags to whether reassembly now
2169 * includes a segment with FIN. This handles the common case
2170 * inline (segment is the next to be received on an established
2171 * connection, and the queue is empty), avoiding linkage into
2172 * and removal from the queue and repetition of various
2174 * Set DELACK for segments received in order, but ack
2175 * immediately when segments are out of order (so
2176 * fast retransmit can work).
2178 if (th->th_seq == tp->rcv_nxt &&
2179 LIST_EMPTY(&tp->t_segq) &&
2180 TCPS_HAVEESTABLISHED(tp->t_state)) {
2182 callout_reset(tp->tt_delack, tcp_delacktime,
2183 tcp_timer_delack, tp);
2185 tp->t_flags |= TF_ACKNOW;
2186 tp->rcv_nxt += tlen;
2187 thflags = th->th_flags & TH_FIN;
2188 tcpstat.tcps_rcvpack++;
2189 tcpstat.tcps_rcvbyte += tlen;
2191 if (so->so_state & SS_CANTRCVMORE)
2194 sbappend(&so->so_rcv, m);
2197 thflags = tcp_reass(tp, th, &tlen, m);
2198 tp->t_flags |= TF_ACKNOW;
2202 * Note the amount of data that peer has sent into
2203 * our window, in order to estimate the sender's
2206 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2213 * If FIN is received ACK the FIN and let the user know
2214 * that the connection is closing.
2216 if (thflags & TH_FIN) {
2217 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2220 * If connection is half-synchronized
2221 * (ie NEEDSYN flag on) then delay ACK,
2222 * so it may be piggybacked when SYN is sent.
2223 * Otherwise, since we received a FIN then no
2224 * more input can be expected, send ACK now.
2226 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2227 callout_reset(tp->tt_delack, tcp_delacktime,
2228 tcp_timer_delack, tp);
2230 tp->t_flags |= TF_ACKNOW;
2233 switch (tp->t_state) {
2236 * In SYN_RECEIVED and ESTABLISHED STATES
2237 * enter the CLOSE_WAIT state.
2239 case TCPS_SYN_RECEIVED:
2240 tp->t_starttime = ticks;
2242 case TCPS_ESTABLISHED:
2243 tp->t_state = TCPS_CLOSE_WAIT;
2247 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2248 * enter the CLOSING state.
2250 case TCPS_FIN_WAIT_1:
2251 tp->t_state = TCPS_CLOSING;
2255 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2256 * starting the time-wait timer, turning off the other
2259 case TCPS_FIN_WAIT_2:
2260 tp->t_state = TCPS_TIME_WAIT;
2261 tcp_canceltimers(tp);
2262 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2263 if (tp->cc_recv != 0 &&
2264 (ticks - tp->t_starttime) < tcp_msl) {
2265 callout_reset(tp->tt_2msl,
2266 tp->t_rxtcur * TCPTV_TWTRUNC,
2267 tcp_timer_2msl, tp);
2268 /* For transaction client, force ACK now. */
2269 tp->t_flags |= TF_ACKNOW;
2272 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2273 tcp_timer_2msl, tp);
2274 soisdisconnected(so);
2278 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2280 case TCPS_TIME_WAIT:
2281 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2282 tcp_timer_2msl, tp);
2287 if (so->so_options & SO_DEBUG)
2288 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2293 * Return any desired output.
2295 if (needoutput || (tp->t_flags & TF_ACKNOW))
2296 (void) tcp_output(tp);
2301 * Generate an ACK dropping incoming segment if it occupies
2302 * sequence space, where the ACK reflects our state.
2304 * We can now skip the test for the RST flag since all
2305 * paths to this code happen after packets containing
2306 * RST have been dropped.
2308 * In the SYN-RECEIVED state, don't send an ACK unless the
2309 * segment we received passes the SYN-RECEIVED ACK test.
2310 * If it fails send a RST. This breaks the loop in the
2311 * "LAND" DoS attack, and also prevents an ACK storm
2312 * between two listening ports that have been sent forged
2313 * SYN segments, each with the source address of the other.
2315 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2316 (SEQ_GT(tp->snd_una, th->th_ack) ||
2317 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2318 rstreason = BANDLIM_RST_OPENPORT;
2322 if (so->so_options & SO_DEBUG)
2323 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2327 tp->t_flags |= TF_ACKNOW;
2328 (void) tcp_output(tp);
2333 * Generate a RST, dropping incoming segment.
2334 * Make ACK acceptable to originator of segment.
2335 * Don't bother to respond if destination was broadcast/multicast.
2337 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2340 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2341 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2344 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2345 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2346 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2347 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2350 /* IPv6 anycast check is done at tcp6_input() */
2353 * Perform bandwidth limiting.
2356 if (badport_bandlim(rstreason) < 0)
2361 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2362 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2365 if (thflags & TH_ACK)
2366 /* mtod() below is safe as long as hdr dropping is delayed */
2367 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2370 if (thflags & TH_SYN)
2372 /* mtod() below is safe as long as hdr dropping is delayed */
2373 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2374 (tcp_seq)0, TH_RST|TH_ACK);
2380 * Drop space held by incoming segment and return.
2383 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2384 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2392 * Parse TCP options and place in tcpopt.
2395 tcp_dooptions(to, cp, cnt, is_syn)
2403 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2405 if (opt == TCPOPT_EOL)
2407 if (opt == TCPOPT_NOP)
2413 if (optlen < 2 || optlen > cnt)
2418 if (optlen != TCPOLEN_MAXSEG)
2422 to->to_flags |= TOF_MSS;
2423 bcopy((char *)cp + 2,
2424 (char *)&to->to_mss, sizeof(to->to_mss));
2425 to->to_mss = ntohs(to->to_mss);
2428 if (optlen != TCPOLEN_WINDOW)
2432 to->to_flags |= TOF_SCALE;
2433 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2435 case TCPOPT_TIMESTAMP:
2436 if (optlen != TCPOLEN_TIMESTAMP)
2438 to->to_flags |= TOF_TS;
2439 bcopy((char *)cp + 2,
2440 (char *)&to->to_tsval, sizeof(to->to_tsval));
2441 to->to_tsval = ntohl(to->to_tsval);
2442 bcopy((char *)cp + 6,
2443 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2444 to->to_tsecr = ntohl(to->to_tsecr);
2447 if (optlen != TCPOLEN_CC)
2449 to->to_flags |= TOF_CC;
2450 bcopy((char *)cp + 2,
2451 (char *)&to->to_cc, sizeof(to->to_cc));
2452 to->to_cc = ntohl(to->to_cc);
2455 if (optlen != TCPOLEN_CC)
2459 to->to_flags |= TOF_CCNEW;
2460 bcopy((char *)cp + 2,
2461 (char *)&to->to_cc, sizeof(to->to_cc));
2462 to->to_cc = ntohl(to->to_cc);
2465 if (optlen != TCPOLEN_CC)
2469 to->to_flags |= TOF_CCECHO;
2470 bcopy((char *)cp + 2,
2471 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2472 to->to_ccecho = ntohl(to->to_ccecho);
2481 * Pull out of band byte out of a segment so
2482 * it doesn't appear in the user's data queue.
2483 * It is still reflected in the segment length for
2484 * sequencing purposes.
2487 tcp_pulloutofband(so, th, m, off)
2491 int off; /* delayed to be droped hdrlen */
2493 int cnt = off + th->th_urp - 1;
2496 if (m->m_len > cnt) {
2497 char *cp = mtod(m, caddr_t) + cnt;
2498 struct tcpcb *tp = sototcpcb(so);
2501 tp->t_oobflags |= TCPOOB_HAVEDATA;
2502 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2504 if (m->m_flags & M_PKTHDR)
2513 panic("tcp_pulloutofband");
2517 * Collect new round-trip time estimate
2518 * and update averages and current timeout.
2521 tcp_xmit_timer(tp, rtt)
2527 tcpstat.tcps_rttupdated++;
2529 if (tp->t_srtt != 0) {
2531 * srtt is stored as fixed point with 5 bits after the
2532 * binary point (i.e., scaled by 8). The following magic
2533 * is equivalent to the smoothing algorithm in rfc793 with
2534 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2535 * point). Adjust rtt to origin 0.
2537 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2538 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2540 if ((tp->t_srtt += delta) <= 0)
2544 * We accumulate a smoothed rtt variance (actually, a
2545 * smoothed mean difference), then set the retransmit
2546 * timer to smoothed rtt + 4 times the smoothed variance.
2547 * rttvar is stored as fixed point with 4 bits after the
2548 * binary point (scaled by 16). The following is
2549 * equivalent to rfc793 smoothing with an alpha of .75
2550 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2551 * rfc793's wired-in beta.
2555 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2556 if ((tp->t_rttvar += delta) <= 0)
2558 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2559 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2562 * No rtt measurement yet - use the unsmoothed rtt.
2563 * Set the variance to half the rtt (so our first
2564 * retransmit happens at 3*rtt).
2566 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2567 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2568 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2574 * the retransmit should happen at rtt + 4 * rttvar.
2575 * Because of the way we do the smoothing, srtt and rttvar
2576 * will each average +1/2 tick of bias. When we compute
2577 * the retransmit timer, we want 1/2 tick of rounding and
2578 * 1 extra tick because of +-1/2 tick uncertainty in the
2579 * firing of the timer. The bias will give us exactly the
2580 * 1.5 tick we need. But, because the bias is
2581 * statistical, we have to test that we don't drop below
2582 * the minimum feasible timer (which is 2 ticks).
2584 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2585 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2588 * We received an ack for a packet that wasn't retransmitted;
2589 * it is probably safe to discard any error indications we've
2590 * received recently. This isn't quite right, but close enough
2591 * for now (a route might have failed after we sent a segment,
2592 * and the return path might not be symmetrical).
2594 tp->t_softerror = 0;
2598 * Determine a reasonable value for maxseg size.
2599 * If the route is known, check route for mtu.
2600 * If none, use an mss that can be handled on the outgoing
2601 * interface without forcing IP to fragment; if bigger than
2602 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2603 * to utilize large mbufs. If no route is found, route has no mtu,
2604 * or the destination isn't local, use a default, hopefully conservative
2605 * size (usually 512 or the default IP max size, but no more than the mtu
2606 * of the interface), as we can't discover anything about intervening
2607 * gateways or networks. We also initialize the congestion/slow start
2608 * window to be a single segment if the destination isn't local.
2609 * While looking at the routing entry, we also initialize other path-dependent
2610 * parameters from pre-set or cached values in the routing entry.
2612 * Also take into account the space needed for options that we
2613 * send regularly. Make maxseg shorter by that amount to assure
2614 * that we can send maxseg amount of data even when the options
2615 * are present. Store the upper limit of the length of options plus
2618 * NOTE that this routine is only called when we process an incoming
2619 * segment, for outgoing segments only tcp_mssopt is called.
2621 * In case of T/TCP, we call this routine during implicit connection
2622 * setup as well (offer = -1), to initialize maxseg from the cached
2634 struct inpcb *inp = tp->t_inpcb;
2636 struct rmxp_tao *taop;
2637 int origoffer = offer;
2639 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2640 size_t min_protoh = isipv6 ?
2641 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2642 sizeof(struct tcpiphdr);
2644 const boolean_t isipv6 = FALSE;
2645 const size_t min_protoh = sizeof(struct tcpiphdr);
2649 rt = tcp_rtlookup6(&inp->inp_inc);
2651 rt = tcp_rtlookup(&inp->inp_inc);
2653 tp->t_maxopd = tp->t_maxseg =
2654 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2658 so = inp->inp_socket;
2660 taop = rmx_taop(rt->rt_rmx);
2662 * Offer == -1 means that we didn't receive SYN yet,
2663 * use cached value in that case;
2666 offer = taop->tao_mssopt;
2668 * Offer == 0 means that there was no MSS on the SYN segment,
2669 * in this case we use tcp_mssdflt.
2672 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2675 * Sanity check: make sure that maxopd will be large
2676 * enough to allow some data on segments even is the
2677 * all the option space is used (40bytes). Otherwise
2678 * funny things may happen in tcp_output.
2680 offer = max(offer, 64);
2681 taop->tao_mssopt = offer;
2684 * While we're here, check if there's an initial rtt
2685 * or rttvar. Convert from the route-table units
2686 * to scaled multiples of the slow timeout timer.
2688 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2690 * XXX the lock bit for RTT indicates that the value
2691 * is also a minimum value; this is subject to time.
2693 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2694 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2695 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2696 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2697 tcpstat.tcps_usedrtt++;
2698 if (rt->rt_rmx.rmx_rttvar) {
2699 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2700 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2701 tcpstat.tcps_usedrttvar++;
2703 /* default variation is +- 1 rtt */
2705 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2707 TCPT_RANGESET(tp->t_rxtcur,
2708 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2709 tp->t_rttmin, TCPTV_REXMTMAX);
2712 * if there's an mtu associated with the route, use it
2713 * else, use the link mtu.
2715 if (rt->rt_rmx.rmx_mtu)
2716 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2719 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2721 if (!in6_localaddr(&inp->in6p_faddr))
2722 mss = min(mss, tcp_v6mssdflt);
2724 mss = ifp->if_mtu - min_protoh;
2725 if (!in_localaddr(inp->inp_faddr))
2726 mss = min(mss, tcp_mssdflt);
2729 mss = min(mss, offer);
2731 * maxopd stores the maximum length of data AND options
2732 * in a segment; maxseg is the amount of data in a normal
2733 * segment. We need to store this value (maxopd) apart
2734 * from maxseg, because now every segment carries options
2735 * and thus we normally have somewhat less data in segments.
2740 * In case of T/TCP, origoffer==-1 indicates, that no segments
2741 * were received yet. In this case we just guess, otherwise
2742 * we do the same as before T/TCP.
2744 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2746 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2747 mss -= TCPOLEN_TSTAMP_APPA;
2748 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2750 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2751 mss -= TCPOLEN_CC_APPA;
2753 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2755 mss &= ~(MCLBYTES-1);
2758 mss = mss / MCLBYTES * MCLBYTES;
2761 * If there's a pipesize, change the socket buffer
2762 * to that size. Make the socket buffers an integral
2763 * number of mss units; if the mss is larger than
2764 * the socket buffer, decrease the mss.
2767 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2769 bufsize = so->so_snd.sb_hiwat;
2773 bufsize = roundup(bufsize, mss);
2774 if (bufsize > sb_max)
2776 if (bufsize > so->so_snd.sb_hiwat)
2777 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2782 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2784 bufsize = so->so_rcv.sb_hiwat;
2785 if (bufsize > mss) {
2786 bufsize = roundup(bufsize, mss);
2787 if (bufsize > sb_max)
2789 if (bufsize > so->so_rcv.sb_hiwat)
2790 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2794 * Set the slow-start flight size depending on whether this
2795 * is a local network or not.
2798 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2799 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2800 (!isipv6 && in_localaddr(inp->inp_faddr)))
2801 tp->snd_cwnd = mss * ss_fltsz_local;
2803 tp->snd_cwnd = mss * ss_fltsz;
2805 if (rt->rt_rmx.rmx_ssthresh) {
2807 * There's some sort of gateway or interface
2808 * buffer limit on the path. Use this to set
2809 * the slow start threshhold, but set the
2810 * threshold to no less than 2*mss.
2812 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2813 tcpstat.tcps_usedssthresh++;
2818 * Determine the MSS option to send on an outgoing SYN.
2827 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2828 int min_protoh = isipv6 ?
2829 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2830 sizeof(struct tcpiphdr);
2832 const boolean_t isipv6 = FALSE;
2833 const size_t min_protoh = sizeof(struct tcpiphdr);
2837 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2839 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2841 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2843 return (rt->rt_ifp->if_mtu - min_protoh);
2848 * When a partial ack arrives, force the retransmission of the
2849 * next unacknowledged segment. Do not clear tp->t_dupacks.
2850 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2854 tcp_newreno_partial_ack(tp, th)
2858 tcp_seq onxt = tp->snd_nxt;
2859 u_long ocwnd = tp->snd_cwnd;
2861 callout_stop(tp->tt_rexmt);
2863 tp->snd_nxt = th->th_ack;
2865 * Set snd_cwnd to one segment beyond acknowledged offset
2866 * (tp->snd_una has not yet been updated when this function is called.)
2868 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2869 tp->t_flags |= TF_ACKNOW;
2870 (void) tcp_output(tp);
2871 tp->snd_cwnd = ocwnd;
2872 if (SEQ_GT(onxt, tp->snd_nxt))
2875 * Partial window deflation. Relies on fact that tp->snd_una
2878 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);