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 1.25 2004/04/13 05:23:13 dillon Exp $
39 #include "opt_ipfw.h" /* for ipfw_fwd */
40 #include "opt_inet6.h"
41 #include "opt_ipsec.h"
42 #include "opt_tcpdebug.h"
43 #include "opt_tcp_input.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/sysctl.h>
49 #include <sys/malloc.h>
51 #include <sys/proc.h> /* for proc0 declaration */
52 #include <sys/protosw.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/syslog.h>
56 #include <sys/in_cksum.h>
58 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
61 #include <net/route.h>
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
67 #include <netinet/in_var.h>
68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/in6_pcb.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
82 #include <netinet/tcpip.h>
84 #include <netinet/tcp_debug.h>
86 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
87 struct tcphdr tcp_savetcp;
91 #include <netipsec/ipsec.h>
92 #include <netipsec/ipsec6.h>
96 #include <netinet6/ipsec.h>
97 #include <netinet6/ipsec6.h>
98 #include <netproto/key/key.h>
101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
103 static const int tcprexmtthresh = 3;
105 static int log_in_vain = 0;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
107 &log_in_vain, 0, "Log all incoming TCP connections");
109 static int blackhole = 0;
110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
111 &blackhole, 0, "Do not send RST when dropping refused connections");
113 int tcp_delack_enabled = 1;
114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
115 &tcp_delack_enabled, 0,
116 "Delay ACK to try and piggyback it onto a data packet");
118 #ifdef TCP_DROP_SYNFIN
119 static int drop_synfin = 0;
120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
121 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
124 static int tcp_do_limitedtransmit = 1;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
126 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
128 static int tcp_do_early_retransmit = 0;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
130 &tcp_do_early_retransmit, 0, "Early retransmit");
132 static int tcp_do_rfc3390 = 1;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
135 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
137 static int tcp_do_eifel_detect = 1;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
139 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
142 "TCP Segment Reassembly Queue");
144 int tcp_reass_maxseg = 0;
145 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
146 &tcp_reass_maxseg, 0,
147 "Global maximum number of TCP Segments in Reassembly Queue");
149 int tcp_reass_qsize = 0;
150 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
152 "Global number of TCP Segments currently in Reassembly Queue");
154 static int tcp_reass_overflows = 0;
155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
156 &tcp_reass_overflows, 0,
157 "Global number of TCP Segment Reassembly Queue Overflows");
159 struct inpcbinfo tcbinfo[MAXCPU];
161 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
162 static void tcp_pulloutofband(struct socket *,
163 struct tcphdr *, struct mbuf *, int);
164 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
166 static void tcp_xmit_timer(struct tcpcb *, int);
167 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
169 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
171 #define ND6_HINT(tp) \
173 if ((tp) && (tp)->t_inpcb && \
174 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
175 (tp)->t_inpcb->in6p_route.ro_rt) \
176 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
183 * Indicate whether this ack should be delayed. We can delay the ack if
184 * - delayed acks are enabled and
185 * - there is no delayed ack timer in progress and
186 * - our last ack wasn't a 0-sized window. We never want to delay
187 * the ack that opens up a 0-sized window.
189 #define DELAY_ACK(tp) \
190 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
191 (tp->t_flags & TF_RXWIN0SENT) == 0)
194 tcp_reass(tp, th, tlenp, m)
201 struct tseg_qent *p = NULL;
202 struct tseg_qent *nq;
203 struct tseg_qent *te;
204 struct socket *so = tp->t_inpcb->inp_socket;
208 * Call with th==0 after become established to
209 * force pre-ESTABLISHED data up to user socket.
215 * Limit the number of segments in the reassembly queue to prevent
216 * holding on to too many segments (and thus running out of mbufs).
217 * Make sure to let the missing segment through which caused this
218 * queue. Always keep one global queue entry spare to be able to
219 * process the missing segment.
221 if (th->th_seq != tp->rcv_nxt &&
222 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
223 tcp_reass_overflows++;
224 tcpstat.tcps_rcvmemdrop++;
229 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
230 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
233 tcpstat.tcps_rcvmemdrop++;
240 * Find a segment which begins after this one does.
242 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
243 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
249 * If there is a preceding segment, it may provide some of
250 * our data already. If so, drop the data from the incoming
251 * segment. If it provides all of our data, drop us.
255 /* conversion to int (in i) handles seq wraparound */
256 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
259 tcpstat.tcps_rcvduppack++;
260 tcpstat.tcps_rcvdupbyte += *tlenp;
265 * Try to present any queued data
266 * at the left window edge to the user.
267 * This is needed after the 3-WHS
270 goto present; /* ??? */
277 tcpstat.tcps_rcvoopack++;
278 tcpstat.tcps_rcvoobyte += *tlenp;
281 * While we overlap succeeding segments trim them or,
282 * if they are completely covered, dequeue them.
285 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
288 if (i < q->tqe_len) {
289 q->tqe_th->th_seq += i;
295 nq = LIST_NEXT(q, tqe_q);
296 LIST_REMOVE(q, tqe_q);
303 /* Insert the new segment queue entry into place. */
306 te->tqe_len = *tlenp;
309 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
311 LIST_INSERT_AFTER(p, te, tqe_q);
316 * Present data to user, advancing rcv_nxt through
317 * completed sequence space.
319 if (!TCPS_HAVEESTABLISHED(tp->t_state))
321 q = LIST_FIRST(&tp->t_segq);
322 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
325 tp->rcv_nxt += q->tqe_len;
326 flags = q->tqe_th->th_flags & TH_FIN;
327 nq = LIST_NEXT(q, tqe_q);
328 LIST_REMOVE(q, tqe_q);
329 if (so->so_state & SS_CANTRCVMORE)
332 sbappend(&so->so_rcv, q->tqe_m);
336 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
343 * TCP input routine, follows pages 65-76 of the
344 * protocol specification dated September, 1981 very closely.
348 tcp6_input(mp, offp, proto)
352 struct mbuf *m = *mp;
353 struct in6_ifaddr *ia6;
355 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
358 * draft-itojun-ipv6-tcp-to-anycast
359 * better place to put this in?
361 ia6 = ip6_getdstifaddr(m);
362 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
365 ip6 = mtod(m, struct ip6_hdr *);
366 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
367 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
368 return (IPPROTO_DONE);
371 tcp_input(m, *offp, proto);
372 return (IPPROTO_DONE);
377 tcp_input(m, off0, proto)
382 struct ip *ip = NULL;
384 struct inpcb *inp = NULL;
389 struct tcpcb *tp = NULL;
391 struct socket *so = 0;
392 int todrop, acked, ourfinisacked, needoutput = 0;
394 struct tcpopt to; /* options in this segment */
395 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
396 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
397 struct sockaddr_in *next_hop = NULL;
398 int rstreason; /* For badport_bandlim accounting purposes */
400 struct ip6_hdr *ip6 = NULL;
404 const boolean_t isipv6 = FALSE;
410 tcpstat.tcps_rcvtotal++;
412 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
413 while (m->m_type == MT_TAG) {
414 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
415 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
420 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
424 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
425 ip6 = mtod(m, struct ip6_hdr *);
426 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
427 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
428 tcpstat.tcps_rcvbadsum++;
431 th = (struct tcphdr *)((caddr_t)ip6 + off0);
434 * Be proactive about unspecified IPv6 address in source.
435 * As we use all-zero to indicate unbounded/unconnected pcb,
436 * unspecified IPv6 address can be used to confuse us.
438 * Note that packets with unspecified IPv6 destination is
439 * already dropped in ip6_input.
441 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
447 * Get IP and TCP header together in first mbuf.
448 * Note: IP leaves IP header in first mbuf.
450 if (off0 > sizeof(struct ip)) {
452 off0 = sizeof(struct ip);
454 /* already checked and pulled up in ip_demux() */
455 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
456 ("TCP header not in one mbuf"));
457 ip = mtod(m, struct ip *);
458 ipov = (struct ipovly *)ip;
459 th = (struct tcphdr *)((caddr_t)ip + off0);
462 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
463 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
464 th->th_sum = m->m_pkthdr.csum_data;
466 th->th_sum = in_pseudo(ip->ip_src.s_addr,
468 htonl(m->m_pkthdr.csum_data +
471 th->th_sum ^= 0xffff;
474 * Checksum extended TCP header and data.
476 len = sizeof(struct ip) + tlen;
477 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
478 ipov->ih_len = (u_short)tlen;
479 ipov->ih_len = htons(ipov->ih_len);
480 th->th_sum = in_cksum(m, len);
483 tcpstat.tcps_rcvbadsum++;
487 /* Re-initialization for later version check */
488 ip->ip_v = IPVERSION;
493 * Check that TCP offset makes sense,
494 * pull out TCP options and adjust length. XXX
496 off = th->th_off << 2;
497 /* already checked and pulled up in ip_demux() */
498 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
499 ("bad TCP data offset"));
500 tlen -= off; /* tlen is used instead of ti->ti_len */
501 if (off > sizeof(struct tcphdr)) {
503 IP6_EXTHDR_CHECK(m, off0, off, );
504 ip6 = mtod(m, struct ip6_hdr *);
505 th = (struct tcphdr *)((caddr_t)ip6 + off0);
507 /* already pulled up in ip_demux() */
508 KASSERT(m->m_len >= sizeof(struct ip) + off,
509 ("TCP header and options not in one mbuf"));
511 optlen = off - sizeof(struct tcphdr);
512 optp = (u_char *)(th + 1);
514 thflags = th->th_flags;
516 #ifdef TCP_DROP_SYNFIN
518 * If the drop_synfin option is enabled, drop all packets with
519 * both the SYN and FIN bits set. This prevents e.g. nmap from
520 * identifying the TCP/IP stack.
522 * This is a violation of the TCP specification.
524 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
529 * Convert TCP protocol specific fields to host format.
531 th->th_seq = ntohl(th->th_seq);
532 th->th_ack = ntohl(th->th_ack);
533 th->th_win = ntohs(th->th_win);
534 th->th_urp = ntohs(th->th_urp);
537 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
538 * until after ip6_savecontrol() is called and before other functions
539 * which don't want those proto headers.
540 * Because ip6_savecontrol() is going to parse the mbuf to
541 * search for data to be passed up to user-land, it wants mbuf
542 * parameters to be unchanged.
543 * XXX: the call of ip6_savecontrol() has been obsoleted based on
544 * latest version of the advanced API (20020110).
546 drop_hdrlen = off0 + off;
549 * Locate pcb for segment.
552 /* IPFIREWALL_FORWARD section */
553 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
555 * Transparently forwarded. Pretend to be the destination.
556 * already got one like this?
558 cpu = mycpu->gd_cpuid;
559 inp = in_pcblookup_hash(&tcbinfo[cpu],
560 ip->ip_src, th->th_sport,
561 ip->ip_dst, th->th_dport,
562 0, m->m_pkthdr.rcvif);
565 * It's new. Try to find the ambushing socket.
569 * The rest of the ipfw code stores the port in
571 * (The IP address is still in network order.)
573 in_port_t dport = next_hop->sin_port ?
574 htons(next_hop->sin_port) :
577 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
578 next_hop->sin_addr.s_addr, dport);
579 inp = in_pcblookup_hash(&tcbinfo[cpu],
580 ip->ip_src, th->th_sport,
581 next_hop->sin_addr, dport,
582 1, m->m_pkthdr.rcvif);
586 inp = in6_pcblookup_hash(&tcbinfo[0],
587 &ip6->ip6_src, th->th_sport,
588 &ip6->ip6_dst, th->th_dport,
589 1, m->m_pkthdr.rcvif);
591 cpu = mycpu->gd_cpuid;
592 inp = in_pcblookup_hash(&tcbinfo[cpu],
593 ip->ip_src, th->th_sport,
594 ip->ip_dst, th->th_dport,
595 1, m->m_pkthdr.rcvif);
601 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
602 ipsec6stat.in_polvio++;
606 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
607 ipsecstat.in_polvio++;
614 if (inp != NULL && ipsec6_in_reject(m, inp)) {
618 if (inp != NULL && ipsec4_in_reject(m, inp)) {
625 * If the state is CLOSED (i.e., TCB does not exist) then
626 * all data in the incoming segment is discarded.
627 * If the TCB exists but is in CLOSED state, it is embryonic,
628 * but should either do a listen or a connect soon.
633 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
635 char dbuf[4 * sizeof "123"], sbuf[4 * sizeof "123"];
640 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
641 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
645 strcpy(dbuf, inet_ntoa(ip->ip_dst));
646 strcpy(sbuf, inet_ntoa(ip->ip_src));
648 switch (log_in_vain) {
650 if ((thflags & TH_SYN) == 0)
654 "Connection attempt to TCP %s:%d "
655 "from %s:%d flags:0x%02x\n",
656 dbuf, ntohs(th->th_dport), sbuf,
657 ntohs(th->th_sport), thflags);
666 if (thflags & TH_SYN)
675 rstreason = BANDLIM_RST_CLOSEDPORT;
680 rstreason = BANDLIM_RST_CLOSEDPORT;
683 if (tp->t_state == TCPS_CLOSED)
686 /* Unscale the window into a 32-bit value. */
687 if ((thflags & TH_SYN) == 0)
688 tiwin = th->th_win << tp->snd_scale;
692 so = inp->inp_socket;
695 if (so->so_options & SO_DEBUG) {
696 ostate = tp->t_state;
698 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
700 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
705 bzero((char *)&to, sizeof(to));
707 if (so->so_options & SO_ACCEPTCONN) {
708 struct in_conninfo inc;
711 inc.inc_isipv6 = (isipv6 == TRUE);
714 inc.inc6_faddr = ip6->ip6_src;
715 inc.inc6_laddr = ip6->ip6_dst;
716 inc.inc6_route.ro_rt = NULL; /* XXX */
718 inc.inc_faddr = ip->ip_src;
719 inc.inc_laddr = ip->ip_dst;
720 inc.inc_route.ro_rt = NULL; /* XXX */
722 inc.inc_fport = th->th_sport;
723 inc.inc_lport = th->th_dport;
726 * If the state is LISTEN then ignore segment if it contains
727 * a RST. If the segment contains an ACK then it is bad and
728 * send a RST. If it does not contain a SYN then it is not
729 * interesting; drop it.
731 * If the state is SYN_RECEIVED (syncache) and seg contains
732 * an ACK, but not for our SYN/ACK, send a RST. If the seg
733 * contains a RST, check the sequence number to see if it
734 * is a valid reset segment.
736 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
737 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
738 if (!syncache_expand(&inc, th, &so, m)) {
740 * No syncache entry, or ACK was not
741 * for our SYN/ACK. Send a RST.
743 tcpstat.tcps_badsyn++;
744 rstreason = BANDLIM_RST_OPENPORT;
749 * Could not complete 3-way handshake,
750 * connection is being closed down, and
751 * syncache will free mbuf.
755 * Socket is created in state SYN_RECEIVED.
756 * Continue processing segment.
761 * This is what would have happened in
762 * tcp_output() when the SYN,ACK was sent.
764 tp->snd_up = tp->snd_una;
765 tp->snd_max = tp->snd_nxt = tp->iss + 1;
766 tp->last_ack_sent = tp->rcv_nxt;
768 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
769 * until the _second_ ACK is received:
770 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
771 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
772 * move to ESTAB, set snd_wnd to tiwin.
774 tp->snd_wnd = tiwin; /* unscaled */
777 if (thflags & TH_RST) {
778 syncache_chkrst(&inc, th);
781 if (thflags & TH_ACK) {
782 syncache_badack(&inc);
783 tcpstat.tcps_badsyn++;
784 rstreason = BANDLIM_RST_OPENPORT;
791 * Segment's flags are (SYN) or (SYN|FIN).
795 * If deprecated address is forbidden,
796 * we do not accept SYN to deprecated interface
797 * address to prevent any new inbound connection from
798 * getting established.
799 * When we do not accept SYN, we send a TCP RST,
800 * with deprecated source address (instead of dropping
801 * it). We compromise it as it is much better for peer
802 * to send a RST, and RST will be the final packet
805 * If we do not forbid deprecated addresses, we accept
806 * the SYN packet. RFC2462 does not suggest dropping
808 * If we decipher RFC2462 5.5.4, it says like this:
809 * 1. use of deprecated addr with existing
810 * communication is okay - "SHOULD continue to be
812 * 2. use of it with new communication:
813 * (2a) "SHOULD NOT be used if alternate address
814 * with sufficient scope is available"
815 * (2b) nothing mentioned otherwise.
816 * Here we fall into (2b) case as we have no choice in
817 * our source address selection - we must obey the peer.
819 * The wording in RFC2462 is confusing, and there are
820 * multiple description text for deprecated address
821 * handling - worse, they are not exactly the same.
822 * I believe 5.5.4 is the best one, so we follow 5.5.4.
824 if (isipv6 && !ip6_use_deprecated) {
825 struct in6_ifaddr *ia6;
827 if ((ia6 = ip6_getdstifaddr(m)) &&
828 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
830 rstreason = BANDLIM_RST_OPENPORT;
836 * If it is from this socket, drop it, it must be forged.
837 * Don't bother responding if the destination was a broadcast.
839 if (th->th_dport == th->th_sport) {
841 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
845 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
850 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
852 * Note that it is quite possible to receive unicast
853 * link-layer packets with a broadcast IP address. Use
854 * in_broadcast() to find them.
856 if (m->m_flags & (M_BCAST|M_MCAST))
859 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
860 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
863 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
864 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
865 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
866 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
870 * SYN appears to be valid; create compressed TCP state
871 * for syncache, or perform t/tcp connection.
873 if (so->so_qlen <= so->so_qlimit) {
874 tcp_dooptions(&to, optp, optlen, 1);
875 if (!syncache_add(&inc, &to, th, &so, m))
879 * Entry added to syncache, mbuf used to
880 * send SYN,ACK packet.
884 * Segment passed TAO tests.
889 tp->t_starttime = ticks;
890 tp->t_state = TCPS_ESTABLISHED;
893 * If there is a FIN, or if there is data and the
894 * connection is local, then delay SYN,ACK(SYN) in
895 * the hope of piggy-backing it on a response
896 * segment. Otherwise must send ACK now in case
897 * the other side is slow starting.
900 ((thflags & TH_FIN) ||
902 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
903 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
904 callout_reset(tp->tt_delack, tcp_delacktime,
905 tcp_timer_delack, tp);
906 tp->t_flags |= TF_NEEDSYN;
908 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
910 tcpstat.tcps_connects++;
918 /* XXX temp debugging */
919 /* should not happen - syncache should pick up these connections */
920 if (tp->t_state == TCPS_LISTEN)
921 panic("tcp_input: TCPS_LISTEN");
924 * Segment received on connection.
925 * Reset idle time and keep-alive timer.
927 tp->t_rcvtime = ticks;
928 if (TCPS_HAVEESTABLISHED(tp->t_state))
929 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
933 * XXX this is tradtitional behavior, may need to be cleaned up.
935 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
936 if (thflags & TH_SYN) {
937 if (to.to_flags & TOF_SCALE) {
938 tp->t_flags |= TF_RCVD_SCALE;
939 tp->requested_s_scale = to.to_requested_s_scale;
941 if (to.to_flags & TOF_TS) {
942 tp->t_flags |= TF_RCVD_TSTMP;
943 tp->ts_recent = to.to_tsval;
944 tp->ts_recent_age = ticks;
946 if (to.to_flags & (TOF_CC|TOF_CCNEW))
947 tp->t_flags |= TF_RCVD_CC;
948 if (to.to_flags & TOF_MSS)
949 tcp_mss(tp, to.to_mss);
953 * Header prediction: check for the two common cases
954 * of a uni-directional data xfer. If the packet has
955 * no control flags, is in-sequence, the window didn't
956 * change and we're not retransmitting, it's a
957 * candidate. If the length is zero and the ack moved
958 * forward, we're the sender side of the xfer. Just
959 * free the data acked & wake any higher level process
960 * that was blocked waiting for space. If the length
961 * is non-zero and the ack didn't move, we're the
962 * receiver side. If we're getting packets in-order
963 * (the reassembly queue is empty), add the data to
964 * the socket buffer and note that we need a delayed ack.
965 * Make sure that the hidden state-flags are also off.
966 * Since we check for TCPS_ESTABLISHED above, it can only
969 if (tp->t_state == TCPS_ESTABLISHED &&
970 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
971 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
972 ((to.to_flags & TOF_TS) == 0 ||
973 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
975 * Using the CC option is compulsory if once started:
976 * the segment is OK if no T/TCP was negotiated or
977 * if the segment has a CC option equal to CCrecv
979 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
980 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
981 th->th_seq == tp->rcv_nxt &&
982 tiwin && tiwin == tp->snd_wnd &&
983 tp->snd_nxt == tp->snd_max) {
986 * If last ACK falls within this segment's sequence numbers,
987 * record the timestamp.
988 * NOTE that the test is modified according to the latest
989 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
991 if ((to.to_flags & TOF_TS) != 0 &&
992 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
993 tp->ts_recent_age = ticks;
994 tp->ts_recent = to.to_tsval;
998 if (SEQ_GT(th->th_ack, tp->snd_una) &&
999 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1000 tp->snd_cwnd >= tp->snd_wnd &&
1001 ((!tcp_do_newreno &&
1002 tp->t_dupacks < tcprexmtthresh) ||
1003 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
1005 * this is a pure ack for outstanding data.
1007 ++tcpstat.tcps_predack;
1009 * "bad retransmit" recovery
1011 * If Eifel detection applies, then
1012 * it is deterministic, so use it
1013 * unconditionally over the old heuristic.
1014 * Otherwise, fall back to the old heuristic.
1016 if (tcp_do_eifel_detect &&
1017 (to.to_flags & TOF_TS) && to.to_tsecr &&
1018 (tp->t_flags & TF_FIRSTACCACK)) {
1019 /* Eifel detection applicable. */
1020 if (to.to_tsecr < tp->t_rexmtTS) {
1021 tcp_revert_congestion_state(tp);
1022 ++tcpstat.tcps_eifeldetected;
1024 } else if (tp->t_rxtshift == 1 &&
1025 ticks < tp->t_badrxtwin) {
1026 tcp_revert_congestion_state(tp);
1027 ++tcpstat.tcps_rttdetected;
1029 tp->t_flags &= ~(TF_FIRSTACCACK |
1030 TF_FASTREXMT | TF_EARLYREXMT);
1032 * Recalculate the retransmit timer / rtt.
1034 * Some machines (certain windows boxes)
1035 * send broken timestamp replies during the
1036 * SYN+ACK phase, ignore timestamps of 0.
1038 if ((to.to_flags & TOF_TS) != 0 &&
1041 ticks - to.to_tsecr + 1);
1042 } else if (tp->t_rtttime &&
1043 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1045 ticks - tp->t_rtttime);
1047 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1048 acked = th->th_ack - tp->snd_una;
1049 tcpstat.tcps_rcvackpack++;
1050 tcpstat.tcps_rcvackbyte += acked;
1051 sbdrop(&so->so_snd, acked);
1052 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1053 SEQ_LEQ(th->th_ack, tp->snd_recover))
1054 tp->snd_recover = th->th_ack - 1;
1055 tp->snd_una = th->th_ack;
1058 ND6_HINT(tp); /* some progress has been done */
1061 * If all outstanding data are acked, stop
1062 * retransmit timer, otherwise restart timer
1063 * using current (possibly backed-off) value.
1064 * If process is waiting for space,
1065 * wakeup/selwakeup/signal. If data
1066 * are ready to send, let tcp_output
1067 * decide between more output or persist.
1069 if (tp->snd_una == tp->snd_max)
1070 callout_stop(tp->tt_rexmt);
1071 else if (!callout_active(tp->tt_persist))
1072 callout_reset(tp->tt_rexmt,
1074 tcp_timer_rexmt, tp);
1077 if (so->so_snd.sb_cc)
1078 (void) tcp_output(tp);
1081 } else if (th->th_ack == tp->snd_una &&
1082 LIST_EMPTY(&tp->t_segq) &&
1083 tlen <= sbspace(&so->so_rcv)) {
1085 * this is a pure, in-sequence data packet
1086 * with nothing on the reassembly queue and
1087 * we have enough buffer space to take it.
1089 ++tcpstat.tcps_preddat;
1090 tp->rcv_nxt += tlen;
1091 tcpstat.tcps_rcvpack++;
1092 tcpstat.tcps_rcvbyte += tlen;
1093 ND6_HINT(tp); /* some progress has been done */
1095 * Add data to socket buffer.
1097 if (so->so_state & SS_CANTRCVMORE) {
1100 m_adj(m, drop_hdrlen); /* delayed header drop */
1101 sbappend(&so->so_rcv, m);
1104 if (DELAY_ACK(tp)) {
1105 callout_reset(tp->tt_delack, tcp_delacktime,
1106 tcp_timer_delack, tp);
1108 tp->t_flags |= TF_ACKNOW;
1116 * Calculate amount of space in receive window,
1117 * and then do TCP input processing.
1118 * Receive window is amount of space in rcv queue,
1119 * but not less than advertised window.
1123 win = sbspace(&so->so_rcv);
1126 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1129 switch (tp->t_state) {
1132 * If the state is SYN_RECEIVED:
1133 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1135 case TCPS_SYN_RECEIVED:
1136 if ((thflags & TH_ACK) &&
1137 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1138 SEQ_GT(th->th_ack, tp->snd_max))) {
1139 rstreason = BANDLIM_RST_OPENPORT;
1145 * If the state is SYN_SENT:
1146 * if seg contains an ACK, but not for our SYN, drop the input.
1147 * if seg contains a RST, then drop the connection.
1148 * if seg does not contain SYN, then drop it.
1149 * Otherwise this is an acceptable SYN segment
1150 * initialize tp->rcv_nxt and tp->irs
1151 * if seg contains ack then advance tp->snd_una
1152 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1153 * arrange for segment to be acked (eventually)
1154 * continue processing rest of data/controls, beginning with URG
1157 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1158 taop = &tao_noncached;
1159 bzero(taop, sizeof(*taop));
1162 if ((thflags & TH_ACK) &&
1163 (SEQ_LEQ(th->th_ack, tp->iss) ||
1164 SEQ_GT(th->th_ack, tp->snd_max))) {
1166 * If we have a cached CCsent for the remote host,
1167 * hence we haven't just crashed and restarted,
1168 * do not send a RST. This may be a retransmission
1169 * from the other side after our earlier ACK was lost.
1170 * Our new SYN, when it arrives, will serve as the
1173 if (taop->tao_ccsent != 0)
1176 rstreason = BANDLIM_UNLIMITED;
1180 if (thflags & TH_RST) {
1181 if (thflags & TH_ACK)
1182 tp = tcp_drop(tp, ECONNREFUSED);
1185 if ((thflags & TH_SYN) == 0)
1187 tp->snd_wnd = th->th_win; /* initial send window */
1188 tp->cc_recv = to.to_cc; /* foreign CC */
1190 tp->irs = th->th_seq;
1192 if (thflags & TH_ACK) {
1194 * Our SYN was acked. If segment contains CC.ECHO
1195 * option, check it to make sure this segment really
1196 * matches our SYN. If not, just drop it as old
1197 * duplicate, but send an RST if we're still playing
1198 * by the old rules. If no CC.ECHO option, make sure
1199 * we don't get fooled into using T/TCP.
1201 if (to.to_flags & TOF_CCECHO) {
1202 if (tp->cc_send != to.to_ccecho) {
1203 if (taop->tao_ccsent != 0)
1206 rstreason = BANDLIM_UNLIMITED;
1211 tp->t_flags &= ~TF_RCVD_CC;
1212 tcpstat.tcps_connects++;
1214 /* Do window scaling on this connection? */
1215 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1216 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1217 tp->snd_scale = tp->requested_s_scale;
1218 tp->rcv_scale = tp->request_r_scale;
1220 /* Segment is acceptable, update cache if undefined. */
1221 if (taop->tao_ccsent == 0)
1222 taop->tao_ccsent = to.to_ccecho;
1224 tp->rcv_adv += tp->rcv_wnd;
1225 tp->snd_una++; /* SYN is acked */
1227 * If there's data, delay ACK; if there's also a FIN
1228 * ACKNOW will be turned on later.
1230 if (DELAY_ACK(tp) && tlen != 0)
1231 callout_reset(tp->tt_delack, tcp_delacktime,
1232 tcp_timer_delack, tp);
1234 tp->t_flags |= TF_ACKNOW;
1236 * Received <SYN,ACK> in SYN_SENT[*] state.
1238 * SYN_SENT --> ESTABLISHED
1239 * SYN_SENT* --> FIN_WAIT_1
1241 tp->t_starttime = ticks;
1242 if (tp->t_flags & TF_NEEDFIN) {
1243 tp->t_state = TCPS_FIN_WAIT_1;
1244 tp->t_flags &= ~TF_NEEDFIN;
1247 tp->t_state = TCPS_ESTABLISHED;
1248 callout_reset(tp->tt_keep, tcp_keepidle,
1249 tcp_timer_keep, tp);
1253 * Received initial SYN in SYN-SENT[*] state =>
1254 * simultaneous open. If segment contains CC option
1255 * and there is a cached CC, apply TAO test.
1256 * If it succeeds, connection is * half-synchronized.
1257 * Otherwise, do 3-way handshake:
1258 * SYN-SENT -> SYN-RECEIVED
1259 * SYN-SENT* -> SYN-RECEIVED*
1260 * If there was no CC option, clear cached CC value.
1262 tp->t_flags |= TF_ACKNOW;
1263 callout_stop(tp->tt_rexmt);
1264 if (to.to_flags & TOF_CC) {
1265 if (taop->tao_cc != 0 &&
1266 CC_GT(to.to_cc, taop->tao_cc)) {
1268 * update cache and make transition:
1269 * SYN-SENT -> ESTABLISHED*
1270 * SYN-SENT* -> FIN-WAIT-1*
1272 taop->tao_cc = to.to_cc;
1273 tp->t_starttime = ticks;
1274 if (tp->t_flags & TF_NEEDFIN) {
1275 tp->t_state = TCPS_FIN_WAIT_1;
1276 tp->t_flags &= ~TF_NEEDFIN;
1278 tp->t_state = TCPS_ESTABLISHED;
1279 callout_reset(tp->tt_keep,
1284 tp->t_flags |= TF_NEEDSYN;
1286 tp->t_state = TCPS_SYN_RECEIVED;
1288 /* CC.NEW or no option => invalidate cache */
1290 tp->t_state = TCPS_SYN_RECEIVED;
1296 * Advance th->th_seq to correspond to first data byte.
1297 * If data, trim to stay within window,
1298 * dropping FIN if necessary.
1301 if (tlen > tp->rcv_wnd) {
1302 todrop = tlen - tp->rcv_wnd;
1306 tcpstat.tcps_rcvpackafterwin++;
1307 tcpstat.tcps_rcvbyteafterwin += todrop;
1309 tp->snd_wl1 = th->th_seq - 1;
1310 tp->rcv_up = th->th_seq;
1312 * Client side of transaction: already sent SYN and data.
1313 * If the remote host used T/TCP to validate the SYN,
1314 * our data will be ACK'd; if so, enter normal data segment
1315 * processing in the middle of step 5, ack processing.
1316 * Otherwise, goto step 6.
1318 if (thflags & TH_ACK)
1324 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1325 * if segment contains a SYN and CC [not CC.NEW] option:
1326 * if state == TIME_WAIT and connection duration > MSL,
1327 * drop packet and send RST;
1329 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1330 * ack the FIN (and data) in retransmission queue.
1331 * Complete close and delete TCPCB. Then reprocess
1332 * segment, hoping to find new TCPCB in LISTEN state;
1334 * else must be old SYN; drop it.
1335 * else do normal processing.
1339 case TCPS_TIME_WAIT:
1340 if ((thflags & TH_SYN) &&
1341 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1342 if (tp->t_state == TCPS_TIME_WAIT &&
1343 (ticks - tp->t_starttime) > tcp_msl) {
1344 rstreason = BANDLIM_UNLIMITED;
1347 if (CC_GT(to.to_cc, tp->cc_recv)) {
1354 break; /* continue normal processing */
1358 * States other than LISTEN or SYN_SENT.
1359 * First check the RST flag and sequence number since reset segments
1360 * are exempt from the timestamp and connection count tests. This
1361 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1362 * below which allowed reset segments in half the sequence space
1363 * to fall though and be processed (which gives forged reset
1364 * segments with a random sequence number a 50 percent chance of
1365 * killing a connection).
1366 * Then check timestamp, if present.
1367 * Then check the connection count, if present.
1368 * Then check that at least some bytes of segment are within
1369 * receive window. If segment begins before rcv_nxt,
1370 * drop leading data (and SYN); if nothing left, just ack.
1373 * If the RST bit is set, check the sequence number to see
1374 * if this is a valid reset segment.
1376 * In all states except SYN-SENT, all reset (RST) segments
1377 * are validated by checking their SEQ-fields. A reset is
1378 * valid if its sequence number is in the window.
1379 * Note: this does not take into account delayed ACKs, so
1380 * we should test against last_ack_sent instead of rcv_nxt.
1381 * The sequence number in the reset segment is normally an
1382 * echo of our outgoing acknowlegement numbers, but some hosts
1383 * send a reset with the sequence number at the rightmost edge
1384 * of our receive window, and we have to handle this case.
1385 * If we have multiple segments in flight, the intial reset
1386 * segment sequence numbers will be to the left of last_ack_sent,
1387 * but they will eventually catch up.
1388 * In any case, it never made sense to trim reset segments to
1389 * fit the receive window since RFC 1122 says:
1390 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1392 * A TCP SHOULD allow a received RST segment to include data.
1395 * It has been suggested that a RST segment could contain
1396 * ASCII text that encoded and explained the cause of the
1397 * RST. No standard has yet been established for such
1400 * If the reset segment passes the sequence number test examine
1402 * SYN_RECEIVED STATE:
1403 * If passive open, return to LISTEN state.
1404 * If active open, inform user that connection was refused.
1405 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1406 * Inform user that connection was reset, and close tcb.
1407 * CLOSING, LAST_ACK STATES:
1410 * Drop the segment - see Stevens, vol. 2, p. 964 and
1413 if (thflags & TH_RST) {
1414 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1415 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1416 switch (tp->t_state) {
1418 case TCPS_SYN_RECEIVED:
1419 so->so_error = ECONNREFUSED;
1422 case TCPS_ESTABLISHED:
1423 case TCPS_FIN_WAIT_1:
1424 case TCPS_FIN_WAIT_2:
1425 case TCPS_CLOSE_WAIT:
1426 so->so_error = ECONNRESET;
1428 tp->t_state = TCPS_CLOSED;
1429 tcpstat.tcps_drops++;
1438 case TCPS_TIME_WAIT:
1446 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1447 * and it's less than ts_recent, drop it.
1449 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1450 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1452 /* Check to see if ts_recent is over 24 days old. */
1453 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1455 * Invalidate ts_recent. If this segment updates
1456 * ts_recent, the age will be reset later and ts_recent
1457 * will get a valid value. If it does not, setting
1458 * ts_recent to zero will at least satisfy the
1459 * requirement that zero be placed in the timestamp
1460 * echo reply when ts_recent isn't valid. The
1461 * age isn't reset until we get a valid ts_recent
1462 * because we don't want out-of-order segments to be
1463 * dropped when ts_recent is old.
1467 tcpstat.tcps_rcvduppack++;
1468 tcpstat.tcps_rcvdupbyte += tlen;
1469 tcpstat.tcps_pawsdrop++;
1478 * If T/TCP was negotiated and the segment doesn't have CC,
1479 * or if its CC is wrong then drop the segment.
1480 * RST segments do not have to comply with this.
1482 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1483 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1487 * In the SYN-RECEIVED state, validate that the packet belongs to
1488 * this connection before trimming the data to fit the receive
1489 * window. Check the sequence number versus IRS since we know
1490 * the sequence numbers haven't wrapped. This is a partial fix
1491 * for the "LAND" DoS attack.
1493 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1494 rstreason = BANDLIM_RST_OPENPORT;
1498 todrop = tp->rcv_nxt - th->th_seq;
1500 if (thflags & TH_SYN) {
1510 * Following if statement from Stevens, vol. 2, p. 960.
1513 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1515 * Any valid FIN must be to the left of the window.
1516 * At this point the FIN must be a duplicate or out
1517 * of sequence; drop it.
1522 * Send an ACK to resynchronize and drop any data.
1523 * But keep on processing for RST or ACK.
1525 tp->t_flags |= TF_ACKNOW;
1527 tcpstat.tcps_rcvduppack++;
1528 tcpstat.tcps_rcvdupbyte += todrop;
1530 tcpstat.tcps_rcvpartduppack++;
1531 tcpstat.tcps_rcvpartdupbyte += todrop;
1533 drop_hdrlen += todrop; /* drop from the top afterwards */
1534 th->th_seq += todrop;
1536 if (th->th_urp > todrop)
1537 th->th_urp -= todrop;
1545 * If new data are received on a connection after the
1546 * user processes are gone, then RST the other end.
1548 if ((so->so_state & SS_NOFDREF) &&
1549 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1551 tcpstat.tcps_rcvafterclose++;
1552 rstreason = BANDLIM_UNLIMITED;
1557 * If segment ends after window, drop trailing data
1558 * (and PUSH and FIN); if nothing left, just ACK.
1560 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1562 tcpstat.tcps_rcvpackafterwin++;
1563 if (todrop >= tlen) {
1564 tcpstat.tcps_rcvbyteafterwin += tlen;
1566 * If a new connection request is received
1567 * while in TIME_WAIT, drop the old connection
1568 * and start over if the sequence numbers
1569 * are above the previous ones.
1571 if (thflags & TH_SYN &&
1572 tp->t_state == TCPS_TIME_WAIT &&
1573 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1578 * If window is closed can only take segments at
1579 * window edge, and have to drop data and PUSH from
1580 * incoming segments. Continue processing, but
1581 * remember to ack. Otherwise, drop segment
1584 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1585 tp->t_flags |= TF_ACKNOW;
1586 tcpstat.tcps_rcvwinprobe++;
1590 tcpstat.tcps_rcvbyteafterwin += todrop;
1593 thflags &= ~(TH_PUSH|TH_FIN);
1597 * If last ACK falls within this segment's sequence numbers,
1598 * record its timestamp.
1599 * NOTE that the test is modified according to the latest
1600 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1602 if ((to.to_flags & TOF_TS) != 0 &&
1603 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1604 tp->ts_recent_age = ticks;
1605 tp->ts_recent = to.to_tsval;
1609 * If a SYN is in the window, then this is an
1610 * error and we send an RST and drop the connection.
1612 if (thflags & TH_SYN) {
1613 tp = tcp_drop(tp, ECONNRESET);
1614 rstreason = BANDLIM_UNLIMITED;
1619 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1620 * flag is on (half-synchronized state), then queue data for
1621 * later processing; else drop segment and return.
1623 if ((thflags & TH_ACK) == 0) {
1624 if (tp->t_state == TCPS_SYN_RECEIVED ||
1625 (tp->t_flags & TF_NEEDSYN))
1634 switch (tp->t_state) {
1637 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1638 * ESTABLISHED state and continue processing.
1639 * The ACK was checked above.
1641 case TCPS_SYN_RECEIVED:
1643 tcpstat.tcps_connects++;
1645 /* Do window scaling? */
1646 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1647 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1648 tp->snd_scale = tp->requested_s_scale;
1649 tp->rcv_scale = tp->request_r_scale;
1652 * Upon successful completion of 3-way handshake,
1653 * update cache.CC if it was undefined, pass any queued
1654 * data to the user, and advance state appropriately.
1656 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1658 taop->tao_cc = tp->cc_recv;
1662 * SYN-RECEIVED -> ESTABLISHED
1663 * SYN-RECEIVED* -> FIN-WAIT-1
1665 tp->t_starttime = ticks;
1666 if (tp->t_flags & TF_NEEDFIN) {
1667 tp->t_state = TCPS_FIN_WAIT_1;
1668 tp->t_flags &= ~TF_NEEDFIN;
1670 tp->t_state = TCPS_ESTABLISHED;
1671 callout_reset(tp->tt_keep, tcp_keepidle,
1672 tcp_timer_keep, tp);
1675 * If segment contains data or ACK, will call tcp_reass()
1676 * later; if not, do so now to pass queued data to user.
1678 if (tlen == 0 && (thflags & TH_FIN) == 0)
1679 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1681 tp->snd_wl1 = th->th_seq - 1;
1685 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1686 * ACKs. If the ack is in the range
1687 * tp->snd_una < th->th_ack <= tp->snd_max
1688 * then advance tp->snd_una to th->th_ack and drop
1689 * data from the retransmission queue. If this ACK reflects
1690 * more up to date window information we update our window information.
1692 case TCPS_ESTABLISHED:
1693 case TCPS_FIN_WAIT_1:
1694 case TCPS_FIN_WAIT_2:
1695 case TCPS_CLOSE_WAIT:
1698 case TCPS_TIME_WAIT:
1700 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1701 if (tlen == 0 && tiwin == tp->snd_wnd) {
1702 tcpstat.tcps_rcvdupack++;
1704 * If we have outstanding data (other than
1705 * a window probe), this is a completely
1706 * duplicate ack (ie, window info didn't
1707 * change), the ack is the biggest we've
1708 * seen and we've seen exactly our rexmt
1709 * threshhold of them, assume a packet
1710 * has been dropped and retransmit it.
1711 * Kludge snd_nxt & the congestion
1712 * window so we send only this one
1715 * We know we're losing at the current
1716 * window size so do congestion avoidance
1717 * (set ssthresh to half the current window
1718 * and pull our congestion window back to
1719 * the new ssthresh).
1721 * Dup acks mean that packets have left the
1722 * network (they're now cached at the receiver)
1723 * so bump cwnd by the amount in the receiver
1724 * to keep a constant cwnd packets in the
1727 if (!callout_active(tp->tt_rexmt) ||
1728 th->th_ack != tp->snd_una)
1730 else if (++tp->t_dupacks > tcprexmtthresh ||
1732 IN_FASTRECOVERY(tp))) {
1733 tp->snd_cwnd += tp->t_maxseg;
1734 (void) tcp_output(tp);
1736 } else if (tp->t_dupacks == tcprexmtthresh) {
1740 if (tcp_do_newreno &&
1747 if (tcp_do_eifel_detect &&
1748 (tp->t_flags & TF_RCVD_TSTMP)) {
1749 tcp_save_congestion_state(tp);
1750 tp->t_flags |= TF_FASTREXMT;
1752 win = min(tp->snd_wnd, tp->snd_cwnd) /
1756 tp->snd_ssthresh = win * tp->t_maxseg;
1757 ENTER_FASTRECOVERY(tp);
1758 tp->snd_recover = tp->snd_max;
1759 callout_stop(tp->tt_rexmt);
1762 tp->snd_nxt = th->th_ack;
1763 tp->snd_cwnd = tp->t_maxseg;
1764 (void) tcp_output(tp);
1765 ++tcpstat.tcps_sndfastrexmit;
1766 KASSERT(tp->snd_limited <= 2,
1767 ("tp->snd_limited too big"));
1768 tp->snd_cwnd = tp->snd_ssthresh +
1770 (tp->t_dupacks - tp->snd_limited));
1771 if (SEQ_GT(onxt, tp->snd_nxt))
1774 } else if (tcp_do_limitedtransmit) {
1775 u_long oldcwnd = tp->snd_cwnd;
1776 tcp_seq oldsndmax = tp->snd_max;
1777 /* outstanding data */
1779 tp->snd_max - tp->snd_una;
1782 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1784 KASSERT(tp->t_dupacks == 1 ||
1786 ("dupacks not 1 or 2"));
1787 if (tp->t_dupacks == 1)
1788 tp->snd_limited = 0;
1789 tp->snd_cwnd = ownd +
1790 (tp->t_dupacks - tp->snd_limited) *
1792 (void) tcp_output(tp);
1793 tp->snd_cwnd = oldcwnd;
1794 sent = tp->snd_max - oldsndmax;
1795 if (sent > tp->t_maxseg) {
1796 KASSERT((tp->t_dupacks == 2 &&
1797 tp->snd_limited == 0) ||
1798 (sent == tp->t_maxseg + 1 &&
1799 tp->t_flags & TF_SENTFIN),
1803 ("sent too many segments"));
1804 tp->snd_limited = 2;
1805 tcpstat.tcps_sndlimited += 2;
1806 } else if (sent > 0) {
1808 ++tcpstat.tcps_sndlimited;
1809 } else if (tcp_do_early_retransmit &&
1810 (tcp_do_eifel_detect &&
1811 (tp->t_flags & TF_RCVD_TSTMP)) &&
1813 tp->t_dupacks + 1 >=
1814 iceildiv(ownd, tp->t_maxseg)) {
1815 ++tcpstat.tcps_sndearlyrexmit;
1816 tp->t_flags |= TF_EARLYREXMT;
1817 goto fastretransmit;
1826 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1829 * If the congestion window was inflated to account
1830 * for the other side's cached packets, retract it.
1832 if (tcp_do_newreno) {
1833 if (IN_FASTRECOVERY(tp)) {
1834 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1835 tcp_newreno_partial_ack(tp, th);
1838 * Window inflation should have left us
1839 * with approximately snd_ssthresh
1841 * But in case we would be inclined to
1842 * send a burst, better to do it via
1843 * the slow start mechanism.
1845 if (SEQ_GT(th->th_ack +
1848 tp->snd_cwnd = tp->snd_max -
1852 tp->snd_cwnd = tp->snd_ssthresh;
1856 if (tp->t_dupacks >= tcprexmtthresh &&
1857 tp->snd_cwnd > tp->snd_ssthresh)
1858 tp->snd_cwnd = tp->snd_ssthresh;
1861 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1862 tcpstat.tcps_rcvacktoomuch++;
1866 * If we reach this point, ACK is not a duplicate,
1867 * i.e., it ACKs something we sent.
1869 if (tp->t_flags & TF_NEEDSYN) {
1871 * T/TCP: Connection was half-synchronized, and our
1872 * SYN has been ACK'd (so connection is now fully
1873 * synchronized). Go to non-starred state,
1874 * increment snd_una for ACK of SYN, and check if
1875 * we can do window scaling.
1877 tp->t_flags &= ~TF_NEEDSYN;
1879 /* Do window scaling? */
1880 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1881 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1882 tp->snd_scale = tp->requested_s_scale;
1883 tp->rcv_scale = tp->request_r_scale;
1888 acked = th->th_ack - tp->snd_una;
1889 tcpstat.tcps_rcvackpack++;
1890 tcpstat.tcps_rcvackbyte += acked;
1893 * If we just performed our first retransmit, and the ACK
1894 * arrives within our recovery window, then it was a mistake
1895 * to do the retransmit in the first place. Recover our
1896 * original cwnd and ssthresh, and proceed to transmit where
1899 if (tcp_do_eifel_detect && acked &&
1900 (to.to_flags & TOF_TS) && to.to_tsecr &&
1901 (tp->t_flags & TF_FIRSTACCACK)) {
1902 /* Eifel detection applicable. */
1903 if (to.to_tsecr < tp->t_rexmtTS) {
1904 ++tcpstat.tcps_eifeldetected;
1905 tcp_revert_congestion_state(tp);
1906 if (tp->t_rxtshift == 1 &&
1907 ticks >= tp->t_badrxtwin)
1908 ++tcpstat.tcps_rttcantdetect;
1910 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1911 tcp_revert_congestion_state(tp);
1912 ++tcpstat.tcps_rttdetected;
1916 * If we have a timestamp reply, update smoothed
1917 * round trip time. If no timestamp is present but
1918 * transmit timer is running and timed sequence
1919 * number was acked, update smoothed round trip time.
1920 * Since we now have an rtt measurement, cancel the
1921 * timer backoff (cf., Phil Karn's retransmit alg.).
1922 * Recompute the initial retransmit timer.
1924 * Some machines (certain windows boxes) send broken
1925 * timestamp replies during the SYN+ACK phase, ignore
1928 if ((to.to_flags & TOF_TS) != 0 &&
1930 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1931 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1932 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1934 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1937 * If all outstanding data is acked, stop retransmit
1938 * timer and remember to restart (more output or persist).
1939 * If there is more data to be acked, restart retransmit
1940 * timer, using current (possibly backed-off) value.
1942 if (th->th_ack == tp->snd_max) {
1943 callout_stop(tp->tt_rexmt);
1945 } else if (!callout_active(tp->tt_persist))
1946 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1947 tcp_timer_rexmt, tp);
1950 * If no data (only SYN) was ACK'd,
1951 * skip rest of ACK processing.
1956 /* Stop looking for an acceptable ACK since one was received. */
1957 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
1960 * When new data is acked, open the congestion window.
1961 * If the window gives us less than ssthresh packets
1962 * in flight, open exponentially (maxseg per packet).
1963 * Otherwise open linearly: maxseg per window
1964 * (maxseg^2 / cwnd per packet).
1966 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
1967 u_int cw = tp->snd_cwnd;
1968 u_int incr = tp->t_maxseg;
1969 if (cw > tp->snd_ssthresh)
1970 incr = incr * incr / cw;
1971 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1973 if (acked > so->so_snd.sb_cc) {
1974 tp->snd_wnd -= so->so_snd.sb_cc;
1975 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1978 sbdrop(&so->so_snd, acked);
1979 tp->snd_wnd -= acked;
1983 /* detect una wraparound */
1984 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1985 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1986 SEQ_LEQ(th->th_ack, tp->snd_recover))
1987 tp->snd_recover = th->th_ack - 1;
1988 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1989 SEQ_GEQ(th->th_ack, tp->snd_recover))
1990 EXIT_FASTRECOVERY(tp);
1991 tp->snd_una = th->th_ack;
1992 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1993 tp->snd_nxt = tp->snd_una;
1995 switch (tp->t_state) {
1998 * In FIN_WAIT_1 STATE in addition to the processing
1999 * for the ESTABLISHED state if our FIN is now acknowledged
2000 * then enter FIN_WAIT_2.
2002 case TCPS_FIN_WAIT_1:
2003 if (ourfinisacked) {
2005 * If we can't receive any more
2006 * data, then closing user can proceed.
2007 * Starting the timer is contrary to the
2008 * specification, but if we don't get a FIN
2009 * we'll hang forever.
2011 if (so->so_state & SS_CANTRCVMORE) {
2012 soisdisconnected(so);
2013 callout_reset(tp->tt_2msl, tcp_maxidle,
2014 tcp_timer_2msl, tp);
2016 tp->t_state = TCPS_FIN_WAIT_2;
2021 * In CLOSING STATE in addition to the processing for
2022 * the ESTABLISHED state if the ACK acknowledges our FIN
2023 * then enter the TIME-WAIT state, otherwise ignore
2027 if (ourfinisacked) {
2028 tp->t_state = TCPS_TIME_WAIT;
2029 tcp_canceltimers(tp);
2030 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2031 if (tp->cc_recv != 0 &&
2032 (ticks - tp->t_starttime) < tcp_msl)
2033 callout_reset(tp->tt_2msl,
2036 tcp_timer_2msl, tp);
2038 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2039 tcp_timer_2msl, tp);
2040 soisdisconnected(so);
2045 * In LAST_ACK, we may still be waiting for data to drain
2046 * and/or to be acked, as well as for the ack of our FIN.
2047 * If our FIN is now acknowledged, delete the TCB,
2048 * enter the closed state and return.
2051 if (ourfinisacked) {
2058 * In TIME_WAIT state the only thing that should arrive
2059 * is a retransmission of the remote FIN. Acknowledge
2060 * it and restart the finack timer.
2062 case TCPS_TIME_WAIT:
2063 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2064 tcp_timer_2msl, tp);
2071 * Update window information.
2072 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2074 if ((thflags & TH_ACK) &&
2075 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2076 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2077 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2078 /* keep track of pure window updates */
2080 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2081 tcpstat.tcps_rcvwinupd++;
2082 tp->snd_wnd = tiwin;
2083 tp->snd_wl1 = th->th_seq;
2084 tp->snd_wl2 = th->th_ack;
2085 if (tp->snd_wnd > tp->max_sndwnd)
2086 tp->max_sndwnd = tp->snd_wnd;
2091 * Process segments with URG.
2093 if ((thflags & TH_URG) && th->th_urp &&
2094 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2096 * This is a kludge, but if we receive and accept
2097 * random urgent pointers, we'll crash in
2098 * soreceive. It's hard to imagine someone
2099 * actually wanting to send this much urgent data.
2101 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2102 th->th_urp = 0; /* XXX */
2103 thflags &= ~TH_URG; /* XXX */
2104 goto dodata; /* XXX */
2107 * If this segment advances the known urgent pointer,
2108 * then mark the data stream. This should not happen
2109 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2110 * a FIN has been received from the remote side.
2111 * In these states we ignore the URG.
2113 * According to RFC961 (Assigned Protocols),
2114 * the urgent pointer points to the last octet
2115 * of urgent data. We continue, however,
2116 * to consider it to indicate the first octet
2117 * of data past the urgent section as the original
2118 * spec states (in one of two places).
2120 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2121 tp->rcv_up = th->th_seq + th->th_urp;
2122 so->so_oobmark = so->so_rcv.sb_cc +
2123 (tp->rcv_up - tp->rcv_nxt) - 1;
2124 if (so->so_oobmark == 0)
2125 so->so_state |= SS_RCVATMARK;
2127 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2130 * Remove out of band data so doesn't get presented to user.
2131 * This can happen independent of advancing the URG pointer,
2132 * but if two URG's are pending at once, some out-of-band
2133 * data may creep in... ick.
2135 if (th->th_urp <= (u_long)tlen
2137 && (so->so_options & SO_OOBINLINE) == 0
2140 tcp_pulloutofband(so, th, m,
2141 drop_hdrlen); /* hdr drop is delayed */
2144 * If no out of band data is expected,
2145 * pull receive urgent pointer along
2146 * with the receive window.
2148 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2149 tp->rcv_up = tp->rcv_nxt;
2154 * Process the segment text, merging it into the TCP sequencing queue,
2155 * and arranging for acknowledgment of receipt if necessary.
2156 * This process logically involves adjusting tp->rcv_wnd as data
2157 * is presented to the user (this happens in tcp_usrreq.c,
2158 * case PRU_RCVD). If a FIN has already been received on this
2159 * connection then we just ignore the text.
2161 if ((tlen || (thflags & TH_FIN)) &&
2162 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2163 m_adj(m, drop_hdrlen); /* delayed header drop */
2165 * Insert segment which includes th into TCP reassembly queue
2166 * with control block tp. Set thflags to whether reassembly now
2167 * includes a segment with FIN. This handles the common case
2168 * inline (segment is the next to be received on an established
2169 * connection, and the queue is empty), avoiding linkage into
2170 * and removal from the queue and repetition of various
2172 * Set DELACK for segments received in order, but ack
2173 * immediately when segments are out of order (so
2174 * fast retransmit can work).
2176 if (th->th_seq == tp->rcv_nxt &&
2177 LIST_EMPTY(&tp->t_segq) &&
2178 TCPS_HAVEESTABLISHED(tp->t_state)) {
2180 callout_reset(tp->tt_delack, tcp_delacktime,
2181 tcp_timer_delack, tp);
2183 tp->t_flags |= TF_ACKNOW;
2184 tp->rcv_nxt += tlen;
2185 thflags = th->th_flags & TH_FIN;
2186 tcpstat.tcps_rcvpack++;
2187 tcpstat.tcps_rcvbyte += tlen;
2189 if (so->so_state & SS_CANTRCVMORE)
2192 sbappend(&so->so_rcv, m);
2195 thflags = tcp_reass(tp, th, &tlen, m);
2196 tp->t_flags |= TF_ACKNOW;
2200 * Note the amount of data that peer has sent into
2201 * our window, in order to estimate the sender's
2204 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2211 * If FIN is received ACK the FIN and let the user know
2212 * that the connection is closing.
2214 if (thflags & TH_FIN) {
2215 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2218 * If connection is half-synchronized
2219 * (ie NEEDSYN flag on) then delay ACK,
2220 * so it may be piggybacked when SYN is sent.
2221 * Otherwise, since we received a FIN then no
2222 * more input can be expected, send ACK now.
2224 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2225 callout_reset(tp->tt_delack, tcp_delacktime,
2226 tcp_timer_delack, tp);
2228 tp->t_flags |= TF_ACKNOW;
2231 switch (tp->t_state) {
2234 * In SYN_RECEIVED and ESTABLISHED STATES
2235 * enter the CLOSE_WAIT state.
2237 case TCPS_SYN_RECEIVED:
2238 tp->t_starttime = ticks;
2240 case TCPS_ESTABLISHED:
2241 tp->t_state = TCPS_CLOSE_WAIT;
2245 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2246 * enter the CLOSING state.
2248 case TCPS_FIN_WAIT_1:
2249 tp->t_state = TCPS_CLOSING;
2253 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2254 * starting the time-wait timer, turning off the other
2257 case TCPS_FIN_WAIT_2:
2258 tp->t_state = TCPS_TIME_WAIT;
2259 tcp_canceltimers(tp);
2260 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2261 if (tp->cc_recv != 0 &&
2262 (ticks - tp->t_starttime) < tcp_msl) {
2263 callout_reset(tp->tt_2msl,
2264 tp->t_rxtcur * TCPTV_TWTRUNC,
2265 tcp_timer_2msl, tp);
2266 /* For transaction client, force ACK now. */
2267 tp->t_flags |= TF_ACKNOW;
2270 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2271 tcp_timer_2msl, tp);
2272 soisdisconnected(so);
2276 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2278 case TCPS_TIME_WAIT:
2279 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2280 tcp_timer_2msl, tp);
2285 if (so->so_options & SO_DEBUG)
2286 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2291 * Return any desired output.
2293 if (needoutput || (tp->t_flags & TF_ACKNOW))
2294 (void) tcp_output(tp);
2299 * Generate an ACK dropping incoming segment if it occupies
2300 * sequence space, where the ACK reflects our state.
2302 * We can now skip the test for the RST flag since all
2303 * paths to this code happen after packets containing
2304 * RST have been dropped.
2306 * In the SYN-RECEIVED state, don't send an ACK unless the
2307 * segment we received passes the SYN-RECEIVED ACK test.
2308 * If it fails send a RST. This breaks the loop in the
2309 * "LAND" DoS attack, and also prevents an ACK storm
2310 * between two listening ports that have been sent forged
2311 * SYN segments, each with the source address of the other.
2313 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2314 (SEQ_GT(tp->snd_una, th->th_ack) ||
2315 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2316 rstreason = BANDLIM_RST_OPENPORT;
2320 if (so->so_options & SO_DEBUG)
2321 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2325 tp->t_flags |= TF_ACKNOW;
2326 (void) tcp_output(tp);
2331 * Generate a RST, dropping incoming segment.
2332 * Make ACK acceptable to originator of segment.
2333 * Don't bother to respond if destination was broadcast/multicast.
2335 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2338 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2339 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2342 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2343 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2344 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2345 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2348 /* IPv6 anycast check is done at tcp6_input() */
2351 * Perform bandwidth limiting.
2354 if (badport_bandlim(rstreason) < 0)
2359 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2360 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2363 if (thflags & TH_ACK)
2364 /* mtod() below is safe as long as hdr dropping is delayed */
2365 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2368 if (thflags & TH_SYN)
2370 /* mtod() below is safe as long as hdr dropping is delayed */
2371 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2372 (tcp_seq)0, TH_RST|TH_ACK);
2378 * Drop space held by incoming segment and return.
2381 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2382 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2390 * Parse TCP options and place in tcpopt.
2393 tcp_dooptions(to, cp, cnt, is_syn)
2401 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2403 if (opt == TCPOPT_EOL)
2405 if (opt == TCPOPT_NOP)
2411 if (optlen < 2 || optlen > cnt)
2416 if (optlen != TCPOLEN_MAXSEG)
2420 to->to_flags |= TOF_MSS;
2421 bcopy((char *)cp + 2,
2422 (char *)&to->to_mss, sizeof(to->to_mss));
2423 to->to_mss = ntohs(to->to_mss);
2426 if (optlen != TCPOLEN_WINDOW)
2430 to->to_flags |= TOF_SCALE;
2431 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2433 case TCPOPT_TIMESTAMP:
2434 if (optlen != TCPOLEN_TIMESTAMP)
2436 to->to_flags |= TOF_TS;
2437 bcopy((char *)cp + 2,
2438 (char *)&to->to_tsval, sizeof(to->to_tsval));
2439 to->to_tsval = ntohl(to->to_tsval);
2440 bcopy((char *)cp + 6,
2441 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2442 to->to_tsecr = ntohl(to->to_tsecr);
2445 if (optlen != TCPOLEN_CC)
2447 to->to_flags |= TOF_CC;
2448 bcopy((char *)cp + 2,
2449 (char *)&to->to_cc, sizeof(to->to_cc));
2450 to->to_cc = ntohl(to->to_cc);
2453 if (optlen != TCPOLEN_CC)
2457 to->to_flags |= TOF_CCNEW;
2458 bcopy((char *)cp + 2,
2459 (char *)&to->to_cc, sizeof(to->to_cc));
2460 to->to_cc = ntohl(to->to_cc);
2463 if (optlen != TCPOLEN_CC)
2467 to->to_flags |= TOF_CCECHO;
2468 bcopy((char *)cp + 2,
2469 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2470 to->to_ccecho = ntohl(to->to_ccecho);
2479 * Pull out of band byte out of a segment so
2480 * it doesn't appear in the user's data queue.
2481 * It is still reflected in the segment length for
2482 * sequencing purposes.
2485 tcp_pulloutofband(so, th, m, off)
2489 int off; /* delayed to be droped hdrlen */
2491 int cnt = off + th->th_urp - 1;
2494 if (m->m_len > cnt) {
2495 char *cp = mtod(m, caddr_t) + cnt;
2496 struct tcpcb *tp = sototcpcb(so);
2499 tp->t_oobflags |= TCPOOB_HAVEDATA;
2500 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2502 if (m->m_flags & M_PKTHDR)
2511 panic("tcp_pulloutofband");
2515 * Collect new round-trip time estimate
2516 * and update averages and current timeout.
2519 tcp_xmit_timer(tp, rtt)
2525 tcpstat.tcps_rttupdated++;
2527 if (tp->t_srtt != 0) {
2529 * srtt is stored as fixed point with 5 bits after the
2530 * binary point (i.e., scaled by 8). The following magic
2531 * is equivalent to the smoothing algorithm in rfc793 with
2532 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2533 * point). Adjust rtt to origin 0.
2535 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2536 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2538 if ((tp->t_srtt += delta) <= 0)
2542 * We accumulate a smoothed rtt variance (actually, a
2543 * smoothed mean difference), then set the retransmit
2544 * timer to smoothed rtt + 4 times the smoothed variance.
2545 * rttvar is stored as fixed point with 4 bits after the
2546 * binary point (scaled by 16). The following is
2547 * equivalent to rfc793 smoothing with an alpha of .75
2548 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2549 * rfc793's wired-in beta.
2553 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2554 if ((tp->t_rttvar += delta) <= 0)
2556 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2557 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2560 * No rtt measurement yet - use the unsmoothed rtt.
2561 * Set the variance to half the rtt (so our first
2562 * retransmit happens at 3*rtt).
2564 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2565 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2566 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2572 * the retransmit should happen at rtt + 4 * rttvar.
2573 * Because of the way we do the smoothing, srtt and rttvar
2574 * will each average +1/2 tick of bias. When we compute
2575 * the retransmit timer, we want 1/2 tick of rounding and
2576 * 1 extra tick because of +-1/2 tick uncertainty in the
2577 * firing of the timer. The bias will give us exactly the
2578 * 1.5 tick we need. But, because the bias is
2579 * statistical, we have to test that we don't drop below
2580 * the minimum feasible timer (which is 2 ticks).
2582 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2583 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2586 * We received an ack for a packet that wasn't retransmitted;
2587 * it is probably safe to discard any error indications we've
2588 * received recently. This isn't quite right, but close enough
2589 * for now (a route might have failed after we sent a segment,
2590 * and the return path might not be symmetrical).
2592 tp->t_softerror = 0;
2596 * Determine a reasonable value for maxseg size.
2597 * If the route is known, check route for mtu.
2598 * If none, use an mss that can be handled on the outgoing
2599 * interface without forcing IP to fragment; if bigger than
2600 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2601 * to utilize large mbufs. If no route is found, route has no mtu,
2602 * or the destination isn't local, use a default, hopefully conservative
2603 * size (usually 512 or the default IP max size, but no more than the mtu
2604 * of the interface), as we can't discover anything about intervening
2605 * gateways or networks. We also initialize the congestion/slow start
2606 * window to be a single segment if the destination isn't local.
2607 * While looking at the routing entry, we also initialize other path-dependent
2608 * parameters from pre-set or cached values in the routing entry.
2610 * Also take into account the space needed for options that we
2611 * send regularly. Make maxseg shorter by that amount to assure
2612 * that we can send maxseg amount of data even when the options
2613 * are present. Store the upper limit of the length of options plus
2616 * NOTE that this routine is only called when we process an incoming
2617 * segment, for outgoing segments only tcp_mssopt is called.
2619 * In case of T/TCP, we call this routine during implicit connection
2620 * setup as well (offer = -1), to initialize maxseg from the cached
2632 struct inpcb *inp = tp->t_inpcb;
2634 struct rmxp_tao *taop;
2635 int origoffer = offer;
2637 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2638 size_t min_protoh = isipv6 ?
2639 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2640 sizeof(struct tcpiphdr);
2642 const boolean_t isipv6 = FALSE;
2643 const size_t min_protoh = sizeof(struct tcpiphdr);
2647 rt = tcp_rtlookup6(&inp->inp_inc);
2649 rt = tcp_rtlookup(&inp->inp_inc);
2651 tp->t_maxopd = tp->t_maxseg =
2652 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2656 so = inp->inp_socket;
2658 taop = rmx_taop(rt->rt_rmx);
2660 * Offer == -1 means that we didn't receive SYN yet,
2661 * use cached value in that case;
2664 offer = taop->tao_mssopt;
2666 * Offer == 0 means that there was no MSS on the SYN segment,
2667 * in this case we use tcp_mssdflt.
2670 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2673 * Sanity check: make sure that maxopd will be large
2674 * enough to allow some data on segments even is the
2675 * all the option space is used (40bytes). Otherwise
2676 * funny things may happen in tcp_output.
2678 offer = max(offer, 64);
2679 taop->tao_mssopt = offer;
2682 * While we're here, check if there's an initial rtt
2683 * or rttvar. Convert from the route-table units
2684 * to scaled multiples of the slow timeout timer.
2686 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2688 * XXX the lock bit for RTT indicates that the value
2689 * is also a minimum value; this is subject to time.
2691 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2692 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2693 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2694 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2695 tcpstat.tcps_usedrtt++;
2696 if (rt->rt_rmx.rmx_rttvar) {
2697 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2698 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2699 tcpstat.tcps_usedrttvar++;
2701 /* default variation is +- 1 rtt */
2703 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2705 TCPT_RANGESET(tp->t_rxtcur,
2706 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2707 tp->t_rttmin, TCPTV_REXMTMAX);
2710 * if there's an mtu associated with the route, use it
2711 * else, use the link mtu.
2713 if (rt->rt_rmx.rmx_mtu)
2714 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2717 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2719 if (!in6_localaddr(&inp->in6p_faddr))
2720 mss = min(mss, tcp_v6mssdflt);
2722 mss = ifp->if_mtu - min_protoh;
2723 if (!in_localaddr(inp->inp_faddr))
2724 mss = min(mss, tcp_mssdflt);
2727 mss = min(mss, offer);
2729 * maxopd stores the maximum length of data AND options
2730 * in a segment; maxseg is the amount of data in a normal
2731 * segment. We need to store this value (maxopd) apart
2732 * from maxseg, because now every segment carries options
2733 * and thus we normally have somewhat less data in segments.
2738 * In case of T/TCP, origoffer==-1 indicates, that no segments
2739 * were received yet. In this case we just guess, otherwise
2740 * we do the same as before T/TCP.
2742 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2744 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2745 mss -= TCPOLEN_TSTAMP_APPA;
2746 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2748 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2749 mss -= TCPOLEN_CC_APPA;
2751 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2753 mss &= ~(MCLBYTES-1);
2756 mss = mss / MCLBYTES * MCLBYTES;
2759 * If there's a pipesize, change the socket buffer
2760 * to that size. Make the socket buffers an integral
2761 * number of mss units; if the mss is larger than
2762 * the socket buffer, decrease the mss.
2765 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2767 bufsize = so->so_snd.sb_hiwat;
2771 bufsize = roundup(bufsize, mss);
2772 if (bufsize > sb_max)
2774 if (bufsize > so->so_snd.sb_hiwat)
2775 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2780 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2782 bufsize = so->so_rcv.sb_hiwat;
2783 if (bufsize > mss) {
2784 bufsize = roundup(bufsize, mss);
2785 if (bufsize > sb_max)
2787 if (bufsize > so->so_rcv.sb_hiwat)
2788 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2792 * Set the slow-start flight size depending on whether this
2793 * is a local network or not.
2796 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2797 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2798 (!isipv6 && in_localaddr(inp->inp_faddr)))
2799 tp->snd_cwnd = mss * ss_fltsz_local;
2801 tp->snd_cwnd = mss * ss_fltsz;
2803 if (rt->rt_rmx.rmx_ssthresh) {
2805 * There's some sort of gateway or interface
2806 * buffer limit on the path. Use this to set
2807 * the slow start threshhold, but set the
2808 * threshold to no less than 2*mss.
2810 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2811 tcpstat.tcps_usedssthresh++;
2816 * Determine the MSS option to send on an outgoing SYN.
2825 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2826 int min_protoh = isipv6 ?
2827 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2828 sizeof(struct tcpiphdr);
2830 const boolean_t isipv6 = FALSE;
2831 const size_t min_protoh = sizeof(struct tcpiphdr);
2835 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2837 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2839 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2841 return (rt->rt_ifp->if_mtu - min_protoh);
2846 * When a partial ack arrives, force the retransmission of the
2847 * next unacknowledged segment. Do not clear tp->t_dupacks.
2848 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2852 tcp_newreno_partial_ack(tp, th)
2856 tcp_seq onxt = tp->snd_nxt;
2857 u_long ocwnd = tp->snd_cwnd;
2859 callout_stop(tp->tt_rexmt);
2861 tp->snd_nxt = th->th_ack;
2863 * Set snd_cwnd to one segment beyond acknowledged offset
2864 * (tp->snd_una has not yet been updated when this function is called.)
2866 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2867 tp->t_flags |= TF_ACKNOW;
2868 (void) tcp_output(tp);
2869 tp->snd_cwnd = ocwnd;
2870 if (SEQ_GT(onxt, tp->snd_nxt))
2873 * Partial window deflation. Relies on fact that tp->snd_una
2876 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);