2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
53 * The Regents of the University of California. All rights reserved.
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
83 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
84 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
85 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.30 2004/07/08 22:07:35 hsu Exp $
88 #include "opt_ipfw.h" /* for ipfw_fwd */
89 #include "opt_inet6.h"
90 #include "opt_ipsec.h"
91 #include "opt_tcpdebug.h"
92 #include "opt_tcp_input.h"
94 #include <sys/param.h>
95 #include <sys/systm.h>
96 #include <sys/kernel.h>
97 #include <sys/sysctl.h>
98 #include <sys/malloc.h>
100 #include <sys/proc.h> /* for proc0 declaration */
101 #include <sys/protosw.h>
102 #include <sys/socket.h>
103 #include <sys/socketvar.h>
104 #include <sys/syslog.h>
105 #include <sys/in_cksum.h>
107 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
108 #include <machine/stdarg.h>
111 #include <net/route.h>
113 #include <netinet/in.h>
114 #include <netinet/in_systm.h>
115 #include <netinet/ip.h>
116 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
117 #include <netinet/in_var.h>
118 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
119 #include <netinet/in_pcb.h>
120 #include <netinet/ip_var.h>
121 #include <netinet/ip6.h>
122 #include <netinet/icmp6.h>
123 #include <netinet6/nd6.h>
124 #include <netinet6/ip6_var.h>
125 #include <netinet6/in6_pcb.h>
126 #include <netinet/tcp.h>
127 #include <netinet/tcp_fsm.h>
128 #include <netinet/tcp_seq.h>
129 #include <netinet/tcp_timer.h>
130 #include <netinet/tcp_var.h>
131 #include <netinet6/tcp6_var.h>
132 #include <netinet/tcpip.h>
134 #include <netinet/tcp_debug.h>
136 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
137 struct tcphdr tcp_savetcp;
138 #endif /* TCPDEBUG */
141 #include <netipsec/ipsec.h>
142 #include <netipsec/ipsec6.h>
146 #include <netinet6/ipsec.h>
147 #include <netinet6/ipsec6.h>
148 #include <netproto/key/key.h>
151 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
153 static const int tcprexmtthresh = 3;
155 static int log_in_vain = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
157 &log_in_vain, 0, "Log all incoming TCP connections");
159 static int blackhole = 0;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
161 &blackhole, 0, "Do not send RST when dropping refused connections");
163 int tcp_delack_enabled = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
165 &tcp_delack_enabled, 0,
166 "Delay ACK to try and piggyback it onto a data packet");
168 #ifdef TCP_DROP_SYNFIN
169 static int drop_synfin = 0;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
171 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
174 static int tcp_do_limitedtransmit = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
176 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
178 static int tcp_do_early_retransmit = 0;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
180 &tcp_do_early_retransmit, 0, "Early retransmit");
182 static int tcp_do_rfc3390 = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
187 static int tcp_do_eifel_detect = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
189 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
191 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
192 "TCP Segment Reassembly Queue");
194 int tcp_reass_maxseg = 0;
195 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
196 &tcp_reass_maxseg, 0,
197 "Global maximum number of TCP Segments in Reassembly Queue");
199 int tcp_reass_qsize = 0;
200 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
202 "Global number of TCP Segments currently in Reassembly Queue");
204 static int tcp_reass_overflows = 0;
205 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
206 &tcp_reass_overflows, 0,
207 "Global number of TCP Segment Reassembly Queue Overflows");
209 struct inpcbinfo tcbinfo[MAXCPU];
211 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
212 static void tcp_pulloutofband(struct socket *,
213 struct tcphdr *, struct mbuf *, int);
214 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
216 static void tcp_xmit_timer(struct tcpcb *, int);
217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
219 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
221 #define ND6_HINT(tp) \
223 if ((tp) && (tp)->t_inpcb && \
224 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
225 (tp)->t_inpcb->in6p_route.ro_rt) \
226 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
233 * Indicate whether this ack should be delayed. We can delay the ack if
234 * - delayed acks are enabled and
235 * - there is no delayed ack timer in progress and
236 * - our last ack wasn't a 0-sized window. We never want to delay
237 * the ack that opens up a 0-sized window.
239 #define DELAY_ACK(tp) \
240 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
241 (tp->t_flags & TF_RXWIN0SENT) == 0)
244 tcp_reass(tp, th, tlenp, m)
251 struct tseg_qent *p = NULL;
252 struct tseg_qent *nq;
253 struct tseg_qent *te;
254 struct socket *so = tp->t_inpcb->inp_socket;
258 * Call with th==0 after become established to
259 * force pre-ESTABLISHED data up to user socket.
265 * Limit the number of segments in the reassembly queue to prevent
266 * holding on to too many segments (and thus running out of mbufs).
267 * Make sure to let the missing segment through which caused this
268 * queue. Always keep one global queue entry spare to be able to
269 * process the missing segment.
271 if (th->th_seq != tp->rcv_nxt &&
272 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
273 tcp_reass_overflows++;
274 tcpstat.tcps_rcvmemdrop++;
279 /* Allocate a new queue entry. */
280 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
281 M_INTWAIT | M_NULLOK);
283 tcpstat.tcps_rcvmemdrop++;
290 * Find a segment which begins after this one does.
292 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
293 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
299 * If there is a preceding segment, it may provide some of
300 * our data already. If so, drop the data from the incoming
301 * segment. If it provides all of our data, drop us.
305 /* conversion to int (in i) handles seq wraparound */
306 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
309 tcpstat.tcps_rcvduppack++;
310 tcpstat.tcps_rcvdupbyte += *tlenp;
315 * Try to present any queued data
316 * at the left window edge to the user.
317 * This is needed after the 3-WHS
320 goto present; /* ??? */
327 tcpstat.tcps_rcvoopack++;
328 tcpstat.tcps_rcvoobyte += *tlenp;
331 * While we overlap succeeding segments trim them or,
332 * if they are completely covered, dequeue them.
335 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
338 if (i < q->tqe_len) {
339 q->tqe_th->th_seq += i;
345 nq = LIST_NEXT(q, tqe_q);
346 LIST_REMOVE(q, tqe_q);
353 /* Insert the new segment queue entry into place. */
356 te->tqe_len = *tlenp;
359 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
361 LIST_INSERT_AFTER(p, te, tqe_q);
366 * Present data to user, advancing rcv_nxt through
367 * completed sequence space.
369 if (!TCPS_HAVEESTABLISHED(tp->t_state))
371 q = LIST_FIRST(&tp->t_segq);
372 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
375 tp->rcv_nxt += q->tqe_len;
376 flags = q->tqe_th->th_flags & TH_FIN;
377 nq = LIST_NEXT(q, tqe_q);
378 LIST_REMOVE(q, tqe_q);
379 if (so->so_state & SS_CANTRCVMORE)
382 sbappend(&so->so_rcv, q->tqe_m);
386 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
393 * TCP input routine, follows pages 65-76 of the
394 * protocol specification dated September, 1981 very closely.
398 tcp6_input(mp, offp, proto)
402 struct mbuf *m = *mp;
403 struct in6_ifaddr *ia6;
405 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
408 * draft-itojun-ipv6-tcp-to-anycast
409 * better place to put this in?
411 ia6 = ip6_getdstifaddr(m);
412 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
415 ip6 = mtod(m, struct ip6_hdr *);
416 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
417 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
418 return (IPPROTO_DONE);
421 tcp_input(m, *offp, proto);
422 return (IPPROTO_DONE);
427 tcp_input(struct mbuf *m, ...)
432 struct ip *ip = NULL;
434 struct inpcb *inp = NULL;
439 struct tcpcb *tp = NULL;
441 struct socket *so = 0;
442 int todrop, acked, ourfinisacked, needoutput = 0;
444 struct tcpopt to; /* options in this segment */
445 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
446 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
447 struct sockaddr_in *next_hop = NULL;
448 int rstreason; /* For badport_bandlim accounting purposes */
450 struct ip6_hdr *ip6 = NULL;
454 const boolean_t isipv6 = FALSE;
461 off0 = __va_arg(ap, int);
462 proto = __va_arg(ap, int);
465 tcpstat.tcps_rcvtotal++;
467 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
468 while (m->m_type == MT_TAG) {
469 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
470 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
475 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
479 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
480 ip6 = mtod(m, struct ip6_hdr *);
481 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
482 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
483 tcpstat.tcps_rcvbadsum++;
486 th = (struct tcphdr *)((caddr_t)ip6 + off0);
489 * Be proactive about unspecified IPv6 address in source.
490 * As we use all-zero to indicate unbounded/unconnected pcb,
491 * unspecified IPv6 address can be used to confuse us.
493 * Note that packets with unspecified IPv6 destination is
494 * already dropped in ip6_input.
496 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
502 * Get IP and TCP header together in first mbuf.
503 * Note: IP leaves IP header in first mbuf.
505 if (off0 > sizeof(struct ip)) {
507 off0 = sizeof(struct ip);
509 /* already checked and pulled up in ip_demux() */
510 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
511 ("TCP header not in one mbuf"));
512 ip = mtod(m, struct ip *);
513 ipov = (struct ipovly *)ip;
514 th = (struct tcphdr *)((caddr_t)ip + off0);
517 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
518 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
519 th->th_sum = m->m_pkthdr.csum_data;
521 th->th_sum = in_pseudo(ip->ip_src.s_addr,
523 htonl(m->m_pkthdr.csum_data +
526 th->th_sum ^= 0xffff;
529 * Checksum extended TCP header and data.
531 len = sizeof(struct ip) + tlen;
532 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
533 ipov->ih_len = (u_short)tlen;
534 ipov->ih_len = htons(ipov->ih_len);
535 th->th_sum = in_cksum(m, len);
538 tcpstat.tcps_rcvbadsum++;
542 /* Re-initialization for later version check */
543 ip->ip_v = IPVERSION;
548 * Check that TCP offset makes sense,
549 * pull out TCP options and adjust length. XXX
551 off = th->th_off << 2;
552 /* already checked and pulled up in ip_demux() */
553 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
554 ("bad TCP data offset"));
555 tlen -= off; /* tlen is used instead of ti->ti_len */
556 if (off > sizeof(struct tcphdr)) {
558 IP6_EXTHDR_CHECK(m, off0, off, );
559 ip6 = mtod(m, struct ip6_hdr *);
560 th = (struct tcphdr *)((caddr_t)ip6 + off0);
562 /* already pulled up in ip_demux() */
563 KASSERT(m->m_len >= sizeof(struct ip) + off,
564 ("TCP header and options not in one mbuf"));
566 optlen = off - sizeof(struct tcphdr);
567 optp = (u_char *)(th + 1);
569 thflags = th->th_flags;
571 #ifdef TCP_DROP_SYNFIN
573 * If the drop_synfin option is enabled, drop all packets with
574 * both the SYN and FIN bits set. This prevents e.g. nmap from
575 * identifying the TCP/IP stack.
577 * This is a violation of the TCP specification.
579 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
584 * Convert TCP protocol specific fields to host format.
586 th->th_seq = ntohl(th->th_seq);
587 th->th_ack = ntohl(th->th_ack);
588 th->th_win = ntohs(th->th_win);
589 th->th_urp = ntohs(th->th_urp);
592 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
593 * until after ip6_savecontrol() is called and before other functions
594 * which don't want those proto headers.
595 * Because ip6_savecontrol() is going to parse the mbuf to
596 * search for data to be passed up to user-land, it wants mbuf
597 * parameters to be unchanged.
598 * XXX: the call of ip6_savecontrol() has been obsoleted based on
599 * latest version of the advanced API (20020110).
601 drop_hdrlen = off0 + off;
604 * Locate pcb for segment.
607 /* IPFIREWALL_FORWARD section */
608 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
610 * Transparently forwarded. Pretend to be the destination.
611 * already got one like this?
613 cpu = mycpu->gd_cpuid;
614 inp = in_pcblookup_hash(&tcbinfo[cpu],
615 ip->ip_src, th->th_sport,
616 ip->ip_dst, th->th_dport,
617 0, m->m_pkthdr.rcvif);
620 * It's new. Try to find the ambushing socket.
624 * The rest of the ipfw code stores the port in
626 * (The IP address is still in network order.)
628 in_port_t dport = next_hop->sin_port ?
629 htons(next_hop->sin_port) :
632 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
633 next_hop->sin_addr.s_addr, dport);
634 inp = in_pcblookup_hash(&tcbinfo[cpu],
635 ip->ip_src, th->th_sport,
636 next_hop->sin_addr, dport,
637 1, m->m_pkthdr.rcvif);
641 inp = in6_pcblookup_hash(&tcbinfo[0],
642 &ip6->ip6_src, th->th_sport,
643 &ip6->ip6_dst, th->th_dport,
644 1, m->m_pkthdr.rcvif);
646 cpu = mycpu->gd_cpuid;
647 inp = in_pcblookup_hash(&tcbinfo[cpu],
648 ip->ip_src, th->th_sport,
649 ip->ip_dst, th->th_dport,
650 1, m->m_pkthdr.rcvif);
656 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
657 ipsec6stat.in_polvio++;
661 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
662 ipsecstat.in_polvio++;
669 if (inp != NULL && ipsec6_in_reject(m, inp)) {
673 if (inp != NULL && ipsec4_in_reject(m, inp)) {
680 * If the state is CLOSED (i.e., TCB does not exist) then
681 * all data in the incoming segment is discarded.
682 * If the TCB exists but is in CLOSED state, it is embryonic,
683 * but should either do a listen or a connect soon.
688 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
690 char dbuf[4 * sizeof "123"], sbuf[4 * sizeof "123"];
695 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
696 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
700 strcpy(dbuf, inet_ntoa(ip->ip_dst));
701 strcpy(sbuf, inet_ntoa(ip->ip_src));
703 switch (log_in_vain) {
705 if ((thflags & TH_SYN) == 0)
709 "Connection attempt to TCP %s:%d "
710 "from %s:%d flags:0x%02x\n",
711 dbuf, ntohs(th->th_dport), sbuf,
712 ntohs(th->th_sport), thflags);
721 if (thflags & TH_SYN)
730 rstreason = BANDLIM_RST_CLOSEDPORT;
735 rstreason = BANDLIM_RST_CLOSEDPORT;
738 if (tp->t_state == TCPS_CLOSED)
741 /* Unscale the window into a 32-bit value. */
742 if ((thflags & TH_SYN) == 0)
743 tiwin = th->th_win << tp->snd_scale;
747 so = inp->inp_socket;
750 if (so->so_options & SO_DEBUG) {
751 ostate = tp->t_state;
753 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
755 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
760 bzero((char *)&to, sizeof(to));
762 if (so->so_options & SO_ACCEPTCONN) {
763 struct in_conninfo inc;
766 inc.inc_isipv6 = (isipv6 == TRUE);
769 inc.inc6_faddr = ip6->ip6_src;
770 inc.inc6_laddr = ip6->ip6_dst;
771 inc.inc6_route.ro_rt = NULL; /* XXX */
773 inc.inc_faddr = ip->ip_src;
774 inc.inc_laddr = ip->ip_dst;
775 inc.inc_route.ro_rt = NULL; /* XXX */
777 inc.inc_fport = th->th_sport;
778 inc.inc_lport = th->th_dport;
781 * If the state is LISTEN then ignore segment if it contains
782 * a RST. If the segment contains an ACK then it is bad and
783 * send a RST. If it does not contain a SYN then it is not
784 * interesting; drop it.
786 * If the state is SYN_RECEIVED (syncache) and seg contains
787 * an ACK, but not for our SYN/ACK, send a RST. If the seg
788 * contains a RST, check the sequence number to see if it
789 * is a valid reset segment.
791 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
792 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
793 if (!syncache_expand(&inc, th, &so, m)) {
795 * No syncache entry, or ACK was not
796 * for our SYN/ACK. Send a RST.
798 tcpstat.tcps_badsyn++;
799 rstreason = BANDLIM_RST_OPENPORT;
804 * Could not complete 3-way handshake,
805 * connection is being closed down, and
806 * syncache will free mbuf.
810 * Socket is created in state SYN_RECEIVED.
811 * Continue processing segment.
816 * This is what would have happened in
817 * tcp_output() when the SYN,ACK was sent.
819 tp->snd_up = tp->snd_una;
820 tp->snd_max = tp->snd_nxt = tp->iss + 1;
821 tp->last_ack_sent = tp->rcv_nxt;
823 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
824 * until the _second_ ACK is received:
825 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
826 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
827 * move to ESTAB, set snd_wnd to tiwin.
829 tp->snd_wnd = tiwin; /* unscaled */
832 if (thflags & TH_RST) {
833 syncache_chkrst(&inc, th);
836 if (thflags & TH_ACK) {
837 syncache_badack(&inc);
838 tcpstat.tcps_badsyn++;
839 rstreason = BANDLIM_RST_OPENPORT;
846 * Segment's flags are (SYN) or (SYN|FIN).
850 * If deprecated address is forbidden,
851 * we do not accept SYN to deprecated interface
852 * address to prevent any new inbound connection from
853 * getting established.
854 * When we do not accept SYN, we send a TCP RST,
855 * with deprecated source address (instead of dropping
856 * it). We compromise it as it is much better for peer
857 * to send a RST, and RST will be the final packet
860 * If we do not forbid deprecated addresses, we accept
861 * the SYN packet. RFC2462 does not suggest dropping
863 * If we decipher RFC2462 5.5.4, it says like this:
864 * 1. use of deprecated addr with existing
865 * communication is okay - "SHOULD continue to be
867 * 2. use of it with new communication:
868 * (2a) "SHOULD NOT be used if alternate address
869 * with sufficient scope is available"
870 * (2b) nothing mentioned otherwise.
871 * Here we fall into (2b) case as we have no choice in
872 * our source address selection - we must obey the peer.
874 * The wording in RFC2462 is confusing, and there are
875 * multiple description text for deprecated address
876 * handling - worse, they are not exactly the same.
877 * I believe 5.5.4 is the best one, so we follow 5.5.4.
879 if (isipv6 && !ip6_use_deprecated) {
880 struct in6_ifaddr *ia6;
882 if ((ia6 = ip6_getdstifaddr(m)) &&
883 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
885 rstreason = BANDLIM_RST_OPENPORT;
891 * If it is from this socket, drop it, it must be forged.
892 * Don't bother responding if the destination was a broadcast.
894 if (th->th_dport == th->th_sport) {
896 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
900 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
905 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
907 * Note that it is quite possible to receive unicast
908 * link-layer packets with a broadcast IP address. Use
909 * in_broadcast() to find them.
911 if (m->m_flags & (M_BCAST|M_MCAST))
914 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
915 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
918 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
919 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
920 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
921 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
925 * SYN appears to be valid; create compressed TCP state
926 * for syncache, or perform t/tcp connection.
928 if (so->so_qlen <= so->so_qlimit) {
929 tcp_dooptions(&to, optp, optlen, 1);
930 if (!syncache_add(&inc, &to, th, &so, m))
934 * Entry added to syncache, mbuf used to
935 * send SYN,ACK packet.
939 * Segment passed TAO tests.
944 tp->t_starttime = ticks;
945 tp->t_state = TCPS_ESTABLISHED;
948 * If there is a FIN, or if there is data and the
949 * connection is local, then delay SYN,ACK(SYN) in
950 * the hope of piggy-backing it on a response
951 * segment. Otherwise must send ACK now in case
952 * the other side is slow starting.
955 ((thflags & TH_FIN) ||
957 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
958 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
959 callout_reset(tp->tt_delack, tcp_delacktime,
960 tcp_timer_delack, tp);
961 tp->t_flags |= TF_NEEDSYN;
963 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
965 tcpstat.tcps_connects++;
973 /* XXX temp debugging */
974 /* should not happen - syncache should pick up these connections */
975 if (tp->t_state == TCPS_LISTEN)
976 panic("tcp_input: TCPS_LISTEN");
979 * Segment received on connection.
980 * Reset idle time and keep-alive timer.
982 tp->t_rcvtime = ticks;
983 if (TCPS_HAVEESTABLISHED(tp->t_state))
984 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
988 * XXX this is tradtitional behavior, may need to be cleaned up.
990 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
991 if (thflags & TH_SYN) {
992 if (to.to_flags & TOF_SCALE) {
993 tp->t_flags |= TF_RCVD_SCALE;
994 tp->requested_s_scale = to.to_requested_s_scale;
996 if (to.to_flags & TOF_TS) {
997 tp->t_flags |= TF_RCVD_TSTMP;
998 tp->ts_recent = to.to_tsval;
999 tp->ts_recent_age = ticks;
1001 if (to.to_flags & (TOF_CC|TOF_CCNEW))
1002 tp->t_flags |= TF_RCVD_CC;
1003 if (to.to_flags & TOF_MSS)
1004 tcp_mss(tp, to.to_mss);
1008 * Header prediction: check for the two common cases
1009 * of a uni-directional data xfer. If the packet has
1010 * no control flags, is in-sequence, the window didn't
1011 * change and we're not retransmitting, it's a
1012 * candidate. If the length is zero and the ack moved
1013 * forward, we're the sender side of the xfer. Just
1014 * free the data acked & wake any higher level process
1015 * that was blocked waiting for space. If the length
1016 * is non-zero and the ack didn't move, we're the
1017 * receiver side. If we're getting packets in-order
1018 * (the reassembly queue is empty), add the data to
1019 * the socket buffer and note that we need a delayed ack.
1020 * Make sure that the hidden state-flags are also off.
1021 * Since we check for TCPS_ESTABLISHED above, it can only
1024 if (tp->t_state == TCPS_ESTABLISHED &&
1025 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1026 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1027 ((to.to_flags & TOF_TS) == 0 ||
1028 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1030 * Using the CC option is compulsory if once started:
1031 * the segment is OK if no T/TCP was negotiated or
1032 * if the segment has a CC option equal to CCrecv
1034 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1035 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1036 th->th_seq == tp->rcv_nxt &&
1037 tiwin && tiwin == tp->snd_wnd &&
1038 tp->snd_nxt == tp->snd_max) {
1041 * If last ACK falls within this segment's sequence numbers,
1042 * record the timestamp.
1043 * NOTE that the test is modified according to the latest
1044 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1046 if ((to.to_flags & TOF_TS) != 0 &&
1047 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1048 tp->ts_recent_age = ticks;
1049 tp->ts_recent = to.to_tsval;
1053 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1054 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1055 tp->snd_cwnd >= tp->snd_wnd &&
1056 ((!tcp_do_newreno &&
1057 tp->t_dupacks < tcprexmtthresh) ||
1058 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
1060 * this is a pure ack for outstanding data.
1062 ++tcpstat.tcps_predack;
1064 * "bad retransmit" recovery
1066 * If Eifel detection applies, then
1067 * it is deterministic, so use it
1068 * unconditionally over the old heuristic.
1069 * Otherwise, fall back to the old heuristic.
1071 if (tcp_do_eifel_detect &&
1072 (to.to_flags & TOF_TS) && to.to_tsecr &&
1073 (tp->t_flags & TF_FIRSTACCACK)) {
1074 /* Eifel detection applicable. */
1075 if (to.to_tsecr < tp->t_rexmtTS) {
1076 tcp_revert_congestion_state(tp);
1077 ++tcpstat.tcps_eifeldetected;
1079 } else if (tp->t_rxtshift == 1 &&
1080 ticks < tp->t_badrxtwin) {
1081 tcp_revert_congestion_state(tp);
1082 ++tcpstat.tcps_rttdetected;
1084 tp->t_flags &= ~(TF_FIRSTACCACK |
1085 TF_FASTREXMT | TF_EARLYREXMT);
1087 * Recalculate the retransmit timer / rtt.
1089 * Some machines (certain windows boxes)
1090 * send broken timestamp replies during the
1091 * SYN+ACK phase, ignore timestamps of 0.
1093 if ((to.to_flags & TOF_TS) != 0 &&
1096 ticks - to.to_tsecr + 1);
1097 } else if (tp->t_rtttime &&
1098 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1100 ticks - tp->t_rtttime);
1102 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1103 acked = th->th_ack - tp->snd_una;
1104 tcpstat.tcps_rcvackpack++;
1105 tcpstat.tcps_rcvackbyte += acked;
1106 sbdrop(&so->so_snd, acked);
1107 tp->snd_recover = th->th_ack - 1;
1108 tp->snd_una = th->th_ack;
1111 ND6_HINT(tp); /* some progress has been done */
1114 * If all outstanding data are acked, stop
1115 * retransmit timer, otherwise restart timer
1116 * using current (possibly backed-off) value.
1117 * If process is waiting for space,
1118 * wakeup/selwakeup/signal. If data
1119 * are ready to send, let tcp_output
1120 * decide between more output or persist.
1122 if (tp->snd_una == tp->snd_max)
1123 callout_stop(tp->tt_rexmt);
1124 else if (!callout_active(tp->tt_persist))
1125 callout_reset(tp->tt_rexmt,
1127 tcp_timer_rexmt, tp);
1130 if (so->so_snd.sb_cc)
1131 (void) tcp_output(tp);
1134 } else if (th->th_ack == tp->snd_una &&
1135 LIST_EMPTY(&tp->t_segq) &&
1136 tlen <= sbspace(&so->so_rcv)) {
1138 * this is a pure, in-sequence data packet
1139 * with nothing on the reassembly queue and
1140 * we have enough buffer space to take it.
1142 ++tcpstat.tcps_preddat;
1143 tp->rcv_nxt += tlen;
1144 tcpstat.tcps_rcvpack++;
1145 tcpstat.tcps_rcvbyte += tlen;
1146 ND6_HINT(tp); /* some progress has been done */
1148 * Add data to socket buffer.
1150 if (so->so_state & SS_CANTRCVMORE) {
1153 m_adj(m, drop_hdrlen); /* delayed header drop */
1154 sbappend(&so->so_rcv, m);
1157 if (DELAY_ACK(tp)) {
1158 callout_reset(tp->tt_delack, tcp_delacktime,
1159 tcp_timer_delack, tp);
1161 tp->t_flags |= TF_ACKNOW;
1169 * Calculate amount of space in receive window,
1170 * and then do TCP input processing.
1171 * Receive window is amount of space in rcv queue,
1172 * but not less than advertised window.
1176 win = sbspace(&so->so_rcv);
1179 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1182 switch (tp->t_state) {
1185 * If the state is SYN_RECEIVED:
1186 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1188 case TCPS_SYN_RECEIVED:
1189 if ((thflags & TH_ACK) &&
1190 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1191 SEQ_GT(th->th_ack, tp->snd_max))) {
1192 rstreason = BANDLIM_RST_OPENPORT;
1198 * If the state is SYN_SENT:
1199 * if seg contains an ACK, but not for our SYN, drop the input.
1200 * if seg contains a RST, then drop the connection.
1201 * if seg does not contain SYN, then drop it.
1202 * Otherwise this is an acceptable SYN segment
1203 * initialize tp->rcv_nxt and tp->irs
1204 * if seg contains ack then advance tp->snd_una
1205 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1206 * arrange for segment to be acked (eventually)
1207 * continue processing rest of data/controls, beginning with URG
1210 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1211 taop = &tao_noncached;
1212 bzero(taop, sizeof(*taop));
1215 if ((thflags & TH_ACK) &&
1216 (SEQ_LEQ(th->th_ack, tp->iss) ||
1217 SEQ_GT(th->th_ack, tp->snd_max))) {
1219 * If we have a cached CCsent for the remote host,
1220 * hence we haven't just crashed and restarted,
1221 * do not send a RST. This may be a retransmission
1222 * from the other side after our earlier ACK was lost.
1223 * Our new SYN, when it arrives, will serve as the
1226 if (taop->tao_ccsent != 0)
1229 rstreason = BANDLIM_UNLIMITED;
1233 if (thflags & TH_RST) {
1234 if (thflags & TH_ACK)
1235 tp = tcp_drop(tp, ECONNREFUSED);
1238 if ((thflags & TH_SYN) == 0)
1240 tp->snd_wnd = th->th_win; /* initial send window */
1241 tp->cc_recv = to.to_cc; /* foreign CC */
1243 tp->irs = th->th_seq;
1245 if (thflags & TH_ACK) {
1247 * Our SYN was acked. If segment contains CC.ECHO
1248 * option, check it to make sure this segment really
1249 * matches our SYN. If not, just drop it as old
1250 * duplicate, but send an RST if we're still playing
1251 * by the old rules. If no CC.ECHO option, make sure
1252 * we don't get fooled into using T/TCP.
1254 if (to.to_flags & TOF_CCECHO) {
1255 if (tp->cc_send != to.to_ccecho) {
1256 if (taop->tao_ccsent != 0)
1259 rstreason = BANDLIM_UNLIMITED;
1264 tp->t_flags &= ~TF_RCVD_CC;
1265 tcpstat.tcps_connects++;
1267 /* Do window scaling on this connection? */
1268 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1269 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1270 tp->snd_scale = tp->requested_s_scale;
1271 tp->rcv_scale = tp->request_r_scale;
1273 /* Segment is acceptable, update cache if undefined. */
1274 if (taop->tao_ccsent == 0)
1275 taop->tao_ccsent = to.to_ccecho;
1277 tp->rcv_adv += tp->rcv_wnd;
1278 tp->snd_una++; /* SYN is acked */
1280 * If there's data, delay ACK; if there's also a FIN
1281 * ACKNOW will be turned on later.
1283 if (DELAY_ACK(tp) && tlen != 0)
1284 callout_reset(tp->tt_delack, tcp_delacktime,
1285 tcp_timer_delack, tp);
1287 tp->t_flags |= TF_ACKNOW;
1289 * Received <SYN,ACK> in SYN_SENT[*] state.
1291 * SYN_SENT --> ESTABLISHED
1292 * SYN_SENT* --> FIN_WAIT_1
1294 tp->t_starttime = ticks;
1295 if (tp->t_flags & TF_NEEDFIN) {
1296 tp->t_state = TCPS_FIN_WAIT_1;
1297 tp->t_flags &= ~TF_NEEDFIN;
1300 tp->t_state = TCPS_ESTABLISHED;
1301 callout_reset(tp->tt_keep, tcp_keepidle,
1302 tcp_timer_keep, tp);
1306 * Received initial SYN in SYN-SENT[*] state =>
1307 * simultaneous open. If segment contains CC option
1308 * and there is a cached CC, apply TAO test.
1309 * If it succeeds, connection is * half-synchronized.
1310 * Otherwise, do 3-way handshake:
1311 * SYN-SENT -> SYN-RECEIVED
1312 * SYN-SENT* -> SYN-RECEIVED*
1313 * If there was no CC option, clear cached CC value.
1315 tp->t_flags |= TF_ACKNOW;
1316 callout_stop(tp->tt_rexmt);
1317 if (to.to_flags & TOF_CC) {
1318 if (taop->tao_cc != 0 &&
1319 CC_GT(to.to_cc, taop->tao_cc)) {
1321 * update cache and make transition:
1322 * SYN-SENT -> ESTABLISHED*
1323 * SYN-SENT* -> FIN-WAIT-1*
1325 taop->tao_cc = to.to_cc;
1326 tp->t_starttime = ticks;
1327 if (tp->t_flags & TF_NEEDFIN) {
1328 tp->t_state = TCPS_FIN_WAIT_1;
1329 tp->t_flags &= ~TF_NEEDFIN;
1331 tp->t_state = TCPS_ESTABLISHED;
1332 callout_reset(tp->tt_keep,
1337 tp->t_flags |= TF_NEEDSYN;
1339 tp->t_state = TCPS_SYN_RECEIVED;
1341 /* CC.NEW or no option => invalidate cache */
1343 tp->t_state = TCPS_SYN_RECEIVED;
1349 * Advance th->th_seq to correspond to first data byte.
1350 * If data, trim to stay within window,
1351 * dropping FIN if necessary.
1354 if (tlen > tp->rcv_wnd) {
1355 todrop = tlen - tp->rcv_wnd;
1359 tcpstat.tcps_rcvpackafterwin++;
1360 tcpstat.tcps_rcvbyteafterwin += todrop;
1362 tp->snd_wl1 = th->th_seq - 1;
1363 tp->rcv_up = th->th_seq;
1365 * Client side of transaction: already sent SYN and data.
1366 * If the remote host used T/TCP to validate the SYN,
1367 * our data will be ACK'd; if so, enter normal data segment
1368 * processing in the middle of step 5, ack processing.
1369 * Otherwise, goto step 6.
1371 if (thflags & TH_ACK)
1377 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1378 * if segment contains a SYN and CC [not CC.NEW] option:
1379 * if state == TIME_WAIT and connection duration > MSL,
1380 * drop packet and send RST;
1382 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1383 * ack the FIN (and data) in retransmission queue.
1384 * Complete close and delete TCPCB. Then reprocess
1385 * segment, hoping to find new TCPCB in LISTEN state;
1387 * else must be old SYN; drop it.
1388 * else do normal processing.
1392 case TCPS_TIME_WAIT:
1393 if ((thflags & TH_SYN) &&
1394 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1395 if (tp->t_state == TCPS_TIME_WAIT &&
1396 (ticks - tp->t_starttime) > tcp_msl) {
1397 rstreason = BANDLIM_UNLIMITED;
1400 if (CC_GT(to.to_cc, tp->cc_recv)) {
1407 break; /* continue normal processing */
1411 * States other than LISTEN or SYN_SENT.
1412 * First check the RST flag and sequence number since reset segments
1413 * are exempt from the timestamp and connection count tests. This
1414 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1415 * below which allowed reset segments in half the sequence space
1416 * to fall though and be processed (which gives forged reset
1417 * segments with a random sequence number a 50 percent chance of
1418 * killing a connection).
1419 * Then check timestamp, if present.
1420 * Then check the connection count, if present.
1421 * Then check that at least some bytes of segment are within
1422 * receive window. If segment begins before rcv_nxt,
1423 * drop leading data (and SYN); if nothing left, just ack.
1426 * If the RST bit is set, check the sequence number to see
1427 * if this is a valid reset segment.
1429 * In all states except SYN-SENT, all reset (RST) segments
1430 * are validated by checking their SEQ-fields. A reset is
1431 * valid if its sequence number is in the window.
1432 * Note: this does not take into account delayed ACKs, so
1433 * we should test against last_ack_sent instead of rcv_nxt.
1434 * The sequence number in the reset segment is normally an
1435 * echo of our outgoing acknowlegement numbers, but some hosts
1436 * send a reset with the sequence number at the rightmost edge
1437 * of our receive window, and we have to handle this case.
1438 * If we have multiple segments in flight, the intial reset
1439 * segment sequence numbers will be to the left of last_ack_sent,
1440 * but they will eventually catch up.
1441 * In any case, it never made sense to trim reset segments to
1442 * fit the receive window since RFC 1122 says:
1443 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1445 * A TCP SHOULD allow a received RST segment to include data.
1448 * It has been suggested that a RST segment could contain
1449 * ASCII text that encoded and explained the cause of the
1450 * RST. No standard has yet been established for such
1453 * If the reset segment passes the sequence number test examine
1455 * SYN_RECEIVED STATE:
1456 * If passive open, return to LISTEN state.
1457 * If active open, inform user that connection was refused.
1458 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1459 * Inform user that connection was reset, and close tcb.
1460 * CLOSING, LAST_ACK STATES:
1463 * Drop the segment - see Stevens, vol. 2, p. 964 and
1466 if (thflags & TH_RST) {
1467 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1468 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1469 switch (tp->t_state) {
1471 case TCPS_SYN_RECEIVED:
1472 so->so_error = ECONNREFUSED;
1475 case TCPS_ESTABLISHED:
1476 case TCPS_FIN_WAIT_1:
1477 case TCPS_FIN_WAIT_2:
1478 case TCPS_CLOSE_WAIT:
1479 so->so_error = ECONNRESET;
1481 tp->t_state = TCPS_CLOSED;
1482 tcpstat.tcps_drops++;
1491 case TCPS_TIME_WAIT:
1499 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1500 * and it's less than ts_recent, drop it.
1502 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1503 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1505 /* Check to see if ts_recent is over 24 days old. */
1506 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1508 * Invalidate ts_recent. If this segment updates
1509 * ts_recent, the age will be reset later and ts_recent
1510 * will get a valid value. If it does not, setting
1511 * ts_recent to zero will at least satisfy the
1512 * requirement that zero be placed in the timestamp
1513 * echo reply when ts_recent isn't valid. The
1514 * age isn't reset until we get a valid ts_recent
1515 * because we don't want out-of-order segments to be
1516 * dropped when ts_recent is old.
1520 tcpstat.tcps_rcvduppack++;
1521 tcpstat.tcps_rcvdupbyte += tlen;
1522 tcpstat.tcps_pawsdrop++;
1531 * If T/TCP was negotiated and the segment doesn't have CC,
1532 * or if its CC is wrong then drop the segment.
1533 * RST segments do not have to comply with this.
1535 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1536 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1540 * In the SYN-RECEIVED state, validate that the packet belongs to
1541 * this connection before trimming the data to fit the receive
1542 * window. Check the sequence number versus IRS since we know
1543 * the sequence numbers haven't wrapped. This is a partial fix
1544 * for the "LAND" DoS attack.
1546 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1547 rstreason = BANDLIM_RST_OPENPORT;
1551 todrop = tp->rcv_nxt - th->th_seq;
1553 if (thflags & TH_SYN) {
1563 * Following if statement from Stevens, vol. 2, p. 960.
1566 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1568 * Any valid FIN must be to the left of the window.
1569 * At this point the FIN must be a duplicate or out
1570 * of sequence; drop it.
1575 * Send an ACK to resynchronize and drop any data.
1576 * But keep on processing for RST or ACK.
1578 tp->t_flags |= TF_ACKNOW;
1580 tcpstat.tcps_rcvduppack++;
1581 tcpstat.tcps_rcvdupbyte += todrop;
1583 tcpstat.tcps_rcvpartduppack++;
1584 tcpstat.tcps_rcvpartdupbyte += todrop;
1586 drop_hdrlen += todrop; /* drop from the top afterwards */
1587 th->th_seq += todrop;
1589 if (th->th_urp > todrop)
1590 th->th_urp -= todrop;
1598 * If new data are received on a connection after the
1599 * user processes are gone, then RST the other end.
1601 if ((so->so_state & SS_NOFDREF) &&
1602 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1604 tcpstat.tcps_rcvafterclose++;
1605 rstreason = BANDLIM_UNLIMITED;
1610 * If segment ends after window, drop trailing data
1611 * (and PUSH and FIN); if nothing left, just ACK.
1613 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1615 tcpstat.tcps_rcvpackafterwin++;
1616 if (todrop >= tlen) {
1617 tcpstat.tcps_rcvbyteafterwin += tlen;
1619 * If a new connection request is received
1620 * while in TIME_WAIT, drop the old connection
1621 * and start over if the sequence numbers
1622 * are above the previous ones.
1624 if (thflags & TH_SYN &&
1625 tp->t_state == TCPS_TIME_WAIT &&
1626 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1631 * If window is closed can only take segments at
1632 * window edge, and have to drop data and PUSH from
1633 * incoming segments. Continue processing, but
1634 * remember to ack. Otherwise, drop segment
1637 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1638 tp->t_flags |= TF_ACKNOW;
1639 tcpstat.tcps_rcvwinprobe++;
1643 tcpstat.tcps_rcvbyteafterwin += todrop;
1646 thflags &= ~(TH_PUSH|TH_FIN);
1650 * If last ACK falls within this segment's sequence numbers,
1651 * record its timestamp.
1652 * NOTE that the test is modified according to the latest
1653 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1655 if ((to.to_flags & TOF_TS) != 0 &&
1656 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1657 tp->ts_recent_age = ticks;
1658 tp->ts_recent = to.to_tsval;
1662 * If a SYN is in the window, then this is an
1663 * error and we send an RST and drop the connection.
1665 if (thflags & TH_SYN) {
1666 tp = tcp_drop(tp, ECONNRESET);
1667 rstreason = BANDLIM_UNLIMITED;
1672 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1673 * flag is on (half-synchronized state), then queue data for
1674 * later processing; else drop segment and return.
1676 if ((thflags & TH_ACK) == 0) {
1677 if (tp->t_state == TCPS_SYN_RECEIVED ||
1678 (tp->t_flags & TF_NEEDSYN))
1687 switch (tp->t_state) {
1690 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1691 * ESTABLISHED state and continue processing.
1692 * The ACK was checked above.
1694 case TCPS_SYN_RECEIVED:
1696 tcpstat.tcps_connects++;
1698 /* Do window scaling? */
1699 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1700 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1701 tp->snd_scale = tp->requested_s_scale;
1702 tp->rcv_scale = tp->request_r_scale;
1705 * Upon successful completion of 3-way handshake,
1706 * update cache.CC if it was undefined, pass any queued
1707 * data to the user, and advance state appropriately.
1709 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1711 taop->tao_cc = tp->cc_recv;
1715 * SYN-RECEIVED -> ESTABLISHED
1716 * SYN-RECEIVED* -> FIN-WAIT-1
1718 tp->t_starttime = ticks;
1719 if (tp->t_flags & TF_NEEDFIN) {
1720 tp->t_state = TCPS_FIN_WAIT_1;
1721 tp->t_flags &= ~TF_NEEDFIN;
1723 tp->t_state = TCPS_ESTABLISHED;
1724 callout_reset(tp->tt_keep, tcp_keepidle,
1725 tcp_timer_keep, tp);
1728 * If segment contains data or ACK, will call tcp_reass()
1729 * later; if not, do so now to pass queued data to user.
1731 if (tlen == 0 && (thflags & TH_FIN) == 0)
1732 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1734 tp->snd_wl1 = th->th_seq - 1;
1738 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1739 * ACKs. If the ack is in the range
1740 * tp->snd_una < th->th_ack <= tp->snd_max
1741 * then advance tp->snd_una to th->th_ack and drop
1742 * data from the retransmission queue. If this ACK reflects
1743 * more up to date window information we update our window information.
1745 case TCPS_ESTABLISHED:
1746 case TCPS_FIN_WAIT_1:
1747 case TCPS_FIN_WAIT_2:
1748 case TCPS_CLOSE_WAIT:
1751 case TCPS_TIME_WAIT:
1753 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1754 if (tlen == 0 && tiwin == tp->snd_wnd) {
1755 tcpstat.tcps_rcvdupack++;
1757 * If we have outstanding data (other than
1758 * a window probe), this is a completely
1759 * duplicate ack (ie, window info didn't
1760 * change), the ack is the biggest we've
1761 * seen and we've seen exactly our rexmt
1762 * threshhold of them, assume a packet
1763 * has been dropped and retransmit it.
1764 * Kludge snd_nxt & the congestion
1765 * window so we send only this one
1768 * We know we're losing at the current
1769 * window size so do congestion avoidance
1770 * (set ssthresh to half the current window
1771 * and pull our congestion window back to
1772 * the new ssthresh).
1774 * Dup acks mean that packets have left the
1775 * network (they're now cached at the receiver)
1776 * so bump cwnd by the amount in the receiver
1777 * to keep a constant cwnd packets in the
1780 if (!callout_active(tp->tt_rexmt) ||
1781 th->th_ack != tp->snd_una)
1783 else if (++tp->t_dupacks > tcprexmtthresh ||
1785 IN_FASTRECOVERY(tp))) {
1786 tp->snd_cwnd += tp->t_maxseg;
1787 (void) tcp_output(tp);
1789 } else if (tp->t_dupacks == tcprexmtthresh) {
1793 if (tcp_do_newreno &&
1800 if (tcp_do_eifel_detect &&
1801 (tp->t_flags & TF_RCVD_TSTMP)) {
1802 tcp_save_congestion_state(tp);
1803 tp->t_flags |= TF_FASTREXMT;
1805 win = min(tp->snd_wnd, tp->snd_cwnd) /
1809 tp->snd_ssthresh = win * tp->t_maxseg;
1810 ENTER_FASTRECOVERY(tp);
1811 tp->snd_recover = tp->snd_max;
1812 callout_stop(tp->tt_rexmt);
1815 tp->snd_nxt = th->th_ack;
1816 tp->snd_cwnd = tp->t_maxseg;
1817 (void) tcp_output(tp);
1818 ++tcpstat.tcps_sndfastrexmit;
1819 KASSERT(tp->snd_limited <= 2,
1820 ("tp->snd_limited too big"));
1821 tp->snd_cwnd = tp->snd_ssthresh +
1823 (tp->t_dupacks - tp->snd_limited));
1824 if (SEQ_GT(onxt, tp->snd_nxt))
1827 } else if (tcp_do_limitedtransmit) {
1828 u_long oldcwnd = tp->snd_cwnd;
1829 tcp_seq oldsndmax = tp->snd_max;
1830 /* outstanding data */
1832 tp->snd_max - tp->snd_una;
1835 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1837 KASSERT(tp->t_dupacks == 1 ||
1839 ("dupacks not 1 or 2"));
1840 if (tp->t_dupacks == 1)
1841 tp->snd_limited = 0;
1842 tp->snd_cwnd = ownd +
1843 (tp->t_dupacks - tp->snd_limited) *
1845 (void) tcp_output(tp);
1846 tp->snd_cwnd = oldcwnd;
1847 sent = tp->snd_max - oldsndmax;
1848 if (sent > tp->t_maxseg) {
1849 KASSERT((tp->t_dupacks == 2 &&
1850 tp->snd_limited == 0) ||
1851 (sent == tp->t_maxseg + 1 &&
1852 tp->t_flags & TF_SENTFIN),
1856 ("sent too many segments"));
1857 tp->snd_limited = 2;
1858 tcpstat.tcps_sndlimited += 2;
1859 } else if (sent > 0) {
1861 ++tcpstat.tcps_sndlimited;
1862 } else if (tcp_do_early_retransmit &&
1863 (tcp_do_eifel_detect &&
1864 (tp->t_flags & TF_RCVD_TSTMP)) &&
1866 tp->t_dupacks + 1 >=
1867 iceildiv(ownd, tp->t_maxseg)) {
1868 ++tcpstat.tcps_sndearlyrexmit;
1869 tp->t_flags |= TF_EARLYREXMT;
1870 goto fastretransmit;
1879 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1882 * If the congestion window was inflated to account
1883 * for the other side's cached packets, retract it.
1885 if (tcp_do_newreno) {
1886 if (IN_FASTRECOVERY(tp)) {
1887 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1888 tcp_newreno_partial_ack(tp, th);
1891 * Window inflation should have left us
1892 * with approximately snd_ssthresh
1894 * But in case we would be inclined to
1895 * send a burst, better to do it via
1896 * the slow start mechanism.
1898 if (SEQ_GT(th->th_ack +
1901 tp->snd_cwnd = tp->snd_max -
1905 tp->snd_cwnd = tp->snd_ssthresh;
1909 if (tp->t_dupacks >= tcprexmtthresh &&
1910 tp->snd_cwnd > tp->snd_ssthresh)
1911 tp->snd_cwnd = tp->snd_ssthresh;
1914 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1916 * Detected optimistic ACK attack.
1917 * Force slow-start to de-synchronize attack.
1919 tp->snd_cwnd = tp->t_maxseg;
1921 tcpstat.tcps_rcvacktoomuch++;
1925 * If we reach this point, ACK is not a duplicate,
1926 * i.e., it ACKs something we sent.
1928 if (tp->t_flags & TF_NEEDSYN) {
1930 * T/TCP: Connection was half-synchronized, and our
1931 * SYN has been ACK'd (so connection is now fully
1932 * synchronized). Go to non-starred state,
1933 * increment snd_una for ACK of SYN, and check if
1934 * we can do window scaling.
1936 tp->t_flags &= ~TF_NEEDSYN;
1938 /* Do window scaling? */
1939 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1940 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1941 tp->snd_scale = tp->requested_s_scale;
1942 tp->rcv_scale = tp->request_r_scale;
1947 acked = th->th_ack - tp->snd_una;
1948 tcpstat.tcps_rcvackpack++;
1949 tcpstat.tcps_rcvackbyte += acked;
1952 * If we just performed our first retransmit, and the ACK
1953 * arrives within our recovery window, then it was a mistake
1954 * to do the retransmit in the first place. Recover our
1955 * original cwnd and ssthresh, and proceed to transmit where
1958 if (tcp_do_eifel_detect && acked &&
1959 (to.to_flags & TOF_TS) && to.to_tsecr &&
1960 (tp->t_flags & TF_FIRSTACCACK)) {
1961 /* Eifel detection applicable. */
1962 if (to.to_tsecr < tp->t_rexmtTS) {
1963 ++tcpstat.tcps_eifeldetected;
1964 tcp_revert_congestion_state(tp);
1965 if (tp->t_rxtshift == 1 &&
1966 ticks >= tp->t_badrxtwin)
1967 ++tcpstat.tcps_rttcantdetect;
1969 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1970 tcp_revert_congestion_state(tp);
1971 ++tcpstat.tcps_rttdetected;
1975 * If we have a timestamp reply, update smoothed
1976 * round trip time. If no timestamp is present but
1977 * transmit timer is running and timed sequence
1978 * number was acked, update smoothed round trip time.
1979 * Since we now have an rtt measurement, cancel the
1980 * timer backoff (cf., Phil Karn's retransmit alg.).
1981 * Recompute the initial retransmit timer.
1983 * Some machines (certain windows boxes) send broken
1984 * timestamp replies during the SYN+ACK phase, ignore
1987 if ((to.to_flags & TOF_TS) != 0 &&
1989 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1990 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1991 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1993 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1996 * If all outstanding data is acked, stop retransmit
1997 * timer and remember to restart (more output or persist).
1998 * If there is more data to be acked, restart retransmit
1999 * timer, using current (possibly backed-off) value.
2001 if (th->th_ack == tp->snd_max) {
2002 callout_stop(tp->tt_rexmt);
2004 } else if (!callout_active(tp->tt_persist))
2005 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2006 tcp_timer_rexmt, tp);
2009 * If no data (only SYN) was ACK'd,
2010 * skip rest of ACK processing.
2015 /* Stop looking for an acceptable ACK since one was received. */
2016 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2019 * When new data is acked, open the congestion window.
2020 * If the window gives us less than ssthresh packets
2021 * in flight, open exponentially (maxseg per packet).
2022 * Otherwise open linearly: maxseg per window
2023 * (maxseg^2 / cwnd per packet).
2025 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
2026 u_int cw = tp->snd_cwnd;
2027 u_int incr = tp->t_maxseg;
2028 if (cw > tp->snd_ssthresh)
2029 incr = incr * incr / cw;
2030 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2032 if (acked > so->so_snd.sb_cc) {
2033 tp->snd_wnd -= so->so_snd.sb_cc;
2034 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2037 sbdrop(&so->so_snd, acked);
2038 tp->snd_wnd -= acked;
2042 if (tcp_do_newreno) {
2043 if (IN_FASTRECOVERY(tp)) {
2044 if (SEQ_GEQ(th->th_ack, tp->snd_recover))
2045 EXIT_FASTRECOVERY(tp);
2047 tp->snd_recover = th->th_ack - 1;
2050 tp->snd_una = th->th_ack;
2051 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2052 tp->snd_nxt = tp->snd_una;
2054 switch (tp->t_state) {
2057 * In FIN_WAIT_1 STATE in addition to the processing
2058 * for the ESTABLISHED state if our FIN is now acknowledged
2059 * then enter FIN_WAIT_2.
2061 case TCPS_FIN_WAIT_1:
2062 if (ourfinisacked) {
2064 * If we can't receive any more
2065 * data, then closing user can proceed.
2066 * Starting the timer is contrary to the
2067 * specification, but if we don't get a FIN
2068 * we'll hang forever.
2070 if (so->so_state & SS_CANTRCVMORE) {
2071 soisdisconnected(so);
2072 callout_reset(tp->tt_2msl, tcp_maxidle,
2073 tcp_timer_2msl, tp);
2075 tp->t_state = TCPS_FIN_WAIT_2;
2080 * In CLOSING STATE in addition to the processing for
2081 * the ESTABLISHED state if the ACK acknowledges our FIN
2082 * then enter the TIME-WAIT state, otherwise ignore
2086 if (ourfinisacked) {
2087 tp->t_state = TCPS_TIME_WAIT;
2088 tcp_canceltimers(tp);
2089 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2090 if (tp->cc_recv != 0 &&
2091 (ticks - tp->t_starttime) < tcp_msl)
2092 callout_reset(tp->tt_2msl,
2095 tcp_timer_2msl, tp);
2097 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2098 tcp_timer_2msl, tp);
2099 soisdisconnected(so);
2104 * In LAST_ACK, we may still be waiting for data to drain
2105 * and/or to be acked, as well as for the ack of our FIN.
2106 * If our FIN is now acknowledged, delete the TCB,
2107 * enter the closed state and return.
2110 if (ourfinisacked) {
2117 * In TIME_WAIT state the only thing that should arrive
2118 * is a retransmission of the remote FIN. Acknowledge
2119 * it and restart the finack timer.
2121 case TCPS_TIME_WAIT:
2122 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2123 tcp_timer_2msl, tp);
2130 * Update window information.
2131 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2133 if ((thflags & TH_ACK) &&
2134 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2135 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2136 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2137 /* keep track of pure window updates */
2139 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2140 tcpstat.tcps_rcvwinupd++;
2141 tp->snd_wnd = tiwin;
2142 tp->snd_wl1 = th->th_seq;
2143 tp->snd_wl2 = th->th_ack;
2144 if (tp->snd_wnd > tp->max_sndwnd)
2145 tp->max_sndwnd = tp->snd_wnd;
2150 * Process segments with URG.
2152 if ((thflags & TH_URG) && th->th_urp &&
2153 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2155 * This is a kludge, but if we receive and accept
2156 * random urgent pointers, we'll crash in
2157 * soreceive. It's hard to imagine someone
2158 * actually wanting to send this much urgent data.
2160 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2161 th->th_urp = 0; /* XXX */
2162 thflags &= ~TH_URG; /* XXX */
2163 goto dodata; /* XXX */
2166 * If this segment advances the known urgent pointer,
2167 * then mark the data stream. This should not happen
2168 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2169 * a FIN has been received from the remote side.
2170 * In these states we ignore the URG.
2172 * According to RFC961 (Assigned Protocols),
2173 * the urgent pointer points to the last octet
2174 * of urgent data. We continue, however,
2175 * to consider it to indicate the first octet
2176 * of data past the urgent section as the original
2177 * spec states (in one of two places).
2179 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2180 tp->rcv_up = th->th_seq + th->th_urp;
2181 so->so_oobmark = so->so_rcv.sb_cc +
2182 (tp->rcv_up - tp->rcv_nxt) - 1;
2183 if (so->so_oobmark == 0)
2184 so->so_state |= SS_RCVATMARK;
2186 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2189 * Remove out of band data so doesn't get presented to user.
2190 * This can happen independent of advancing the URG pointer,
2191 * but if two URG's are pending at once, some out-of-band
2192 * data may creep in... ick.
2194 if (th->th_urp <= (u_long)tlen
2196 && (so->so_options & SO_OOBINLINE) == 0
2199 tcp_pulloutofband(so, th, m,
2200 drop_hdrlen); /* hdr drop is delayed */
2203 * If no out of band data is expected,
2204 * pull receive urgent pointer along
2205 * with the receive window.
2207 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2208 tp->rcv_up = tp->rcv_nxt;
2213 * Process the segment text, merging it into the TCP sequencing queue,
2214 * and arranging for acknowledgment of receipt if necessary.
2215 * This process logically involves adjusting tp->rcv_wnd as data
2216 * is presented to the user (this happens in tcp_usrreq.c,
2217 * case PRU_RCVD). If a FIN has already been received on this
2218 * connection then we just ignore the text.
2220 if ((tlen || (thflags & TH_FIN)) &&
2221 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2222 m_adj(m, drop_hdrlen); /* delayed header drop */
2224 * Insert segment which includes th into TCP reassembly queue
2225 * with control block tp. Set thflags to whether reassembly now
2226 * includes a segment with FIN. This handles the common case
2227 * inline (segment is the next to be received on an established
2228 * connection, and the queue is empty), avoiding linkage into
2229 * and removal from the queue and repetition of various
2231 * Set DELACK for segments received in order, but ack
2232 * immediately when segments are out of order (so
2233 * fast retransmit can work).
2235 if (th->th_seq == tp->rcv_nxt &&
2236 LIST_EMPTY(&tp->t_segq) &&
2237 TCPS_HAVEESTABLISHED(tp->t_state)) {
2239 callout_reset(tp->tt_delack, tcp_delacktime,
2240 tcp_timer_delack, tp);
2242 tp->t_flags |= TF_ACKNOW;
2243 tp->rcv_nxt += tlen;
2244 thflags = th->th_flags & TH_FIN;
2245 tcpstat.tcps_rcvpack++;
2246 tcpstat.tcps_rcvbyte += tlen;
2248 if (so->so_state & SS_CANTRCVMORE)
2251 sbappend(&so->so_rcv, m);
2254 thflags = tcp_reass(tp, th, &tlen, m);
2255 tp->t_flags |= TF_ACKNOW;
2259 * Note the amount of data that peer has sent into
2260 * our window, in order to estimate the sender's
2263 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2270 * If FIN is received ACK the FIN and let the user know
2271 * that the connection is closing.
2273 if (thflags & TH_FIN) {
2274 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2277 * If connection is half-synchronized
2278 * (ie NEEDSYN flag on) then delay ACK,
2279 * so it may be piggybacked when SYN is sent.
2280 * Otherwise, since we received a FIN then no
2281 * more input can be expected, send ACK now.
2283 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2284 callout_reset(tp->tt_delack, tcp_delacktime,
2285 tcp_timer_delack, tp);
2287 tp->t_flags |= TF_ACKNOW;
2290 switch (tp->t_state) {
2293 * In SYN_RECEIVED and ESTABLISHED STATES
2294 * enter the CLOSE_WAIT state.
2296 case TCPS_SYN_RECEIVED:
2297 tp->t_starttime = ticks;
2299 case TCPS_ESTABLISHED:
2300 tp->t_state = TCPS_CLOSE_WAIT;
2304 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2305 * enter the CLOSING state.
2307 case TCPS_FIN_WAIT_1:
2308 tp->t_state = TCPS_CLOSING;
2312 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2313 * starting the time-wait timer, turning off the other
2316 case TCPS_FIN_WAIT_2:
2317 tp->t_state = TCPS_TIME_WAIT;
2318 tcp_canceltimers(tp);
2319 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2320 if (tp->cc_recv != 0 &&
2321 (ticks - tp->t_starttime) < tcp_msl) {
2322 callout_reset(tp->tt_2msl,
2323 tp->t_rxtcur * TCPTV_TWTRUNC,
2324 tcp_timer_2msl, tp);
2325 /* For transaction client, force ACK now. */
2326 tp->t_flags |= TF_ACKNOW;
2329 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2330 tcp_timer_2msl, tp);
2331 soisdisconnected(so);
2335 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2337 case TCPS_TIME_WAIT:
2338 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2339 tcp_timer_2msl, tp);
2344 if (so->so_options & SO_DEBUG)
2345 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2350 * Return any desired output.
2352 if (needoutput || (tp->t_flags & TF_ACKNOW))
2353 (void) tcp_output(tp);
2358 * Generate an ACK dropping incoming segment if it occupies
2359 * sequence space, where the ACK reflects our state.
2361 * We can now skip the test for the RST flag since all
2362 * paths to this code happen after packets containing
2363 * RST have been dropped.
2365 * In the SYN-RECEIVED state, don't send an ACK unless the
2366 * segment we received passes the SYN-RECEIVED ACK test.
2367 * If it fails send a RST. This breaks the loop in the
2368 * "LAND" DoS attack, and also prevents an ACK storm
2369 * between two listening ports that have been sent forged
2370 * SYN segments, each with the source address of the other.
2372 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2373 (SEQ_GT(tp->snd_una, th->th_ack) ||
2374 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2375 rstreason = BANDLIM_RST_OPENPORT;
2379 if (so->so_options & SO_DEBUG)
2380 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2384 tp->t_flags |= TF_ACKNOW;
2385 (void) tcp_output(tp);
2390 * Generate a RST, dropping incoming segment.
2391 * Make ACK acceptable to originator of segment.
2392 * Don't bother to respond if destination was broadcast/multicast.
2394 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2397 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2398 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2401 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2402 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2403 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2404 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2407 /* IPv6 anycast check is done at tcp6_input() */
2410 * Perform bandwidth limiting.
2413 if (badport_bandlim(rstreason) < 0)
2418 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2419 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2422 if (thflags & TH_ACK)
2423 /* mtod() below is safe as long as hdr dropping is delayed */
2424 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2427 if (thflags & TH_SYN)
2429 /* mtod() below is safe as long as hdr dropping is delayed */
2430 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2431 (tcp_seq)0, TH_RST|TH_ACK);
2437 * Drop space held by incoming segment and return.
2440 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2441 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2449 * Parse TCP options and place in tcpopt.
2452 tcp_dooptions(to, cp, cnt, is_syn)
2460 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2462 if (opt == TCPOPT_EOL)
2464 if (opt == TCPOPT_NOP)
2470 if (optlen < 2 || optlen > cnt)
2475 if (optlen != TCPOLEN_MAXSEG)
2479 to->to_flags |= TOF_MSS;
2480 bcopy((char *)cp + 2,
2481 (char *)&to->to_mss, sizeof(to->to_mss));
2482 to->to_mss = ntohs(to->to_mss);
2485 if (optlen != TCPOLEN_WINDOW)
2489 to->to_flags |= TOF_SCALE;
2490 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2492 case TCPOPT_TIMESTAMP:
2493 if (optlen != TCPOLEN_TIMESTAMP)
2495 to->to_flags |= TOF_TS;
2496 bcopy((char *)cp + 2,
2497 (char *)&to->to_tsval, sizeof(to->to_tsval));
2498 to->to_tsval = ntohl(to->to_tsval);
2499 bcopy((char *)cp + 6,
2500 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2501 to->to_tsecr = ntohl(to->to_tsecr);
2504 if (optlen != TCPOLEN_CC)
2506 to->to_flags |= TOF_CC;
2507 bcopy((char *)cp + 2,
2508 (char *)&to->to_cc, sizeof(to->to_cc));
2509 to->to_cc = ntohl(to->to_cc);
2512 if (optlen != TCPOLEN_CC)
2516 to->to_flags |= TOF_CCNEW;
2517 bcopy((char *)cp + 2,
2518 (char *)&to->to_cc, sizeof(to->to_cc));
2519 to->to_cc = ntohl(to->to_cc);
2522 if (optlen != TCPOLEN_CC)
2526 to->to_flags |= TOF_CCECHO;
2527 bcopy((char *)cp + 2,
2528 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2529 to->to_ccecho = ntohl(to->to_ccecho);
2538 * Pull out of band byte out of a segment so
2539 * it doesn't appear in the user's data queue.
2540 * It is still reflected in the segment length for
2541 * sequencing purposes.
2544 tcp_pulloutofband(so, th, m, off)
2548 int off; /* delayed to be droped hdrlen */
2550 int cnt = off + th->th_urp - 1;
2553 if (m->m_len > cnt) {
2554 char *cp = mtod(m, caddr_t) + cnt;
2555 struct tcpcb *tp = sototcpcb(so);
2558 tp->t_oobflags |= TCPOOB_HAVEDATA;
2559 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2561 if (m->m_flags & M_PKTHDR)
2570 panic("tcp_pulloutofband");
2574 * Collect new round-trip time estimate
2575 * and update averages and current timeout.
2578 tcp_xmit_timer(tp, rtt)
2584 tcpstat.tcps_rttupdated++;
2586 if (tp->t_srtt != 0) {
2588 * srtt is stored as fixed point with 5 bits after the
2589 * binary point (i.e., scaled by 8). The following magic
2590 * is equivalent to the smoothing algorithm in rfc793 with
2591 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2592 * point). Adjust rtt to origin 0.
2594 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2595 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2597 if ((tp->t_srtt += delta) <= 0)
2601 * We accumulate a smoothed rtt variance (actually, a
2602 * smoothed mean difference), then set the retransmit
2603 * timer to smoothed rtt + 4 times the smoothed variance.
2604 * rttvar is stored as fixed point with 4 bits after the
2605 * binary point (scaled by 16). The following is
2606 * equivalent to rfc793 smoothing with an alpha of .75
2607 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2608 * rfc793's wired-in beta.
2612 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2613 if ((tp->t_rttvar += delta) <= 0)
2615 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2616 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2619 * No rtt measurement yet - use the unsmoothed rtt.
2620 * Set the variance to half the rtt (so our first
2621 * retransmit happens at 3*rtt).
2623 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2624 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2625 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2631 * the retransmit should happen at rtt + 4 * rttvar.
2632 * Because of the way we do the smoothing, srtt and rttvar
2633 * will each average +1/2 tick of bias. When we compute
2634 * the retransmit timer, we want 1/2 tick of rounding and
2635 * 1 extra tick because of +-1/2 tick uncertainty in the
2636 * firing of the timer. The bias will give us exactly the
2637 * 1.5 tick we need. But, because the bias is
2638 * statistical, we have to test that we don't drop below
2639 * the minimum feasible timer (which is 2 ticks).
2641 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2642 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2645 * We received an ack for a packet that wasn't retransmitted;
2646 * it is probably safe to discard any error indications we've
2647 * received recently. This isn't quite right, but close enough
2648 * for now (a route might have failed after we sent a segment,
2649 * and the return path might not be symmetrical).
2651 tp->t_softerror = 0;
2655 * Determine a reasonable value for maxseg size.
2656 * If the route is known, check route for mtu.
2657 * If none, use an mss that can be handled on the outgoing
2658 * interface without forcing IP to fragment; if bigger than
2659 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2660 * to utilize large mbufs. If no route is found, route has no mtu,
2661 * or the destination isn't local, use a default, hopefully conservative
2662 * size (usually 512 or the default IP max size, but no more than the mtu
2663 * of the interface), as we can't discover anything about intervening
2664 * gateways or networks. We also initialize the congestion/slow start
2665 * window to be a single segment if the destination isn't local.
2666 * While looking at the routing entry, we also initialize other path-dependent
2667 * parameters from pre-set or cached values in the routing entry.
2669 * Also take into account the space needed for options that we
2670 * send regularly. Make maxseg shorter by that amount to assure
2671 * that we can send maxseg amount of data even when the options
2672 * are present. Store the upper limit of the length of options plus
2675 * NOTE that this routine is only called when we process an incoming
2676 * segment, for outgoing segments only tcp_mssopt is called.
2678 * In case of T/TCP, we call this routine during implicit connection
2679 * setup as well (offer = -1), to initialize maxseg from the cached
2691 struct inpcb *inp = tp->t_inpcb;
2693 struct rmxp_tao *taop;
2694 int origoffer = offer;
2696 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2697 size_t min_protoh = isipv6 ?
2698 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2699 sizeof(struct tcpiphdr);
2701 const boolean_t isipv6 = FALSE;
2702 const size_t min_protoh = sizeof(struct tcpiphdr);
2706 rt = tcp_rtlookup6(&inp->inp_inc);
2708 rt = tcp_rtlookup(&inp->inp_inc);
2710 tp->t_maxopd = tp->t_maxseg =
2711 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2715 so = inp->inp_socket;
2717 taop = rmx_taop(rt->rt_rmx);
2719 * Offer == -1 means that we didn't receive SYN yet,
2720 * use cached value in that case;
2723 offer = taop->tao_mssopt;
2725 * Offer == 0 means that there was no MSS on the SYN segment,
2726 * in this case we use tcp_mssdflt.
2729 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2732 * Sanity check: make sure that maxopd will be large
2733 * enough to allow some data on segments even is the
2734 * all the option space is used (40bytes). Otherwise
2735 * funny things may happen in tcp_output.
2737 offer = max(offer, 64);
2738 taop->tao_mssopt = offer;
2741 * While we're here, check if there's an initial rtt
2742 * or rttvar. Convert from the route-table units
2743 * to scaled multiples of the slow timeout timer.
2745 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2747 * XXX the lock bit for RTT indicates that the value
2748 * is also a minimum value; this is subject to time.
2750 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2751 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2752 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2753 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2754 tcpstat.tcps_usedrtt++;
2755 if (rt->rt_rmx.rmx_rttvar) {
2756 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2757 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2758 tcpstat.tcps_usedrttvar++;
2760 /* default variation is +- 1 rtt */
2762 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2764 TCPT_RANGESET(tp->t_rxtcur,
2765 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2766 tp->t_rttmin, TCPTV_REXMTMAX);
2769 * if there's an mtu associated with the route, use it
2770 * else, use the link mtu.
2772 if (rt->rt_rmx.rmx_mtu)
2773 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2776 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2778 if (!in6_localaddr(&inp->in6p_faddr))
2779 mss = min(mss, tcp_v6mssdflt);
2781 mss = ifp->if_mtu - min_protoh;
2782 if (!in_localaddr(inp->inp_faddr))
2783 mss = min(mss, tcp_mssdflt);
2786 mss = min(mss, offer);
2788 * maxopd stores the maximum length of data AND options
2789 * in a segment; maxseg is the amount of data in a normal
2790 * segment. We need to store this value (maxopd) apart
2791 * from maxseg, because now every segment carries options
2792 * and thus we normally have somewhat less data in segments.
2797 * In case of T/TCP, origoffer==-1 indicates, that no segments
2798 * were received yet. In this case we just guess, otherwise
2799 * we do the same as before T/TCP.
2801 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2803 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2804 mss -= TCPOLEN_TSTAMP_APPA;
2805 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2807 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2808 mss -= TCPOLEN_CC_APPA;
2810 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2812 mss &= ~(MCLBYTES-1);
2815 mss = mss / MCLBYTES * MCLBYTES;
2818 * If there's a pipesize, change the socket buffer
2819 * to that size. Make the socket buffers an integral
2820 * number of mss units; if the mss is larger than
2821 * the socket buffer, decrease the mss.
2824 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2826 bufsize = so->so_snd.sb_hiwat;
2830 bufsize = roundup(bufsize, mss);
2831 if (bufsize > sb_max)
2833 if (bufsize > so->so_snd.sb_hiwat)
2834 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2839 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2841 bufsize = so->so_rcv.sb_hiwat;
2842 if (bufsize > mss) {
2843 bufsize = roundup(bufsize, mss);
2844 if (bufsize > sb_max)
2846 if (bufsize > so->so_rcv.sb_hiwat)
2847 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2851 * Set the slow-start flight size depending on whether this
2852 * is a local network or not.
2855 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2856 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2857 (!isipv6 && in_localaddr(inp->inp_faddr)))
2858 tp->snd_cwnd = mss * ss_fltsz_local;
2860 tp->snd_cwnd = mss * ss_fltsz;
2862 if (rt->rt_rmx.rmx_ssthresh) {
2864 * There's some sort of gateway or interface
2865 * buffer limit on the path. Use this to set
2866 * the slow start threshhold, but set the
2867 * threshold to no less than 2*mss.
2869 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2870 tcpstat.tcps_usedssthresh++;
2875 * Determine the MSS option to send on an outgoing SYN.
2884 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2885 int min_protoh = isipv6 ?
2886 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2887 sizeof(struct tcpiphdr);
2889 const boolean_t isipv6 = FALSE;
2890 const size_t min_protoh = sizeof(struct tcpiphdr);
2894 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2896 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2898 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2900 return (rt->rt_ifp->if_mtu - min_protoh);
2905 * When a partial ack arrives, force the retransmission of the
2906 * next unacknowledged segment. Do not clear tp->t_dupacks.
2907 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2911 tcp_newreno_partial_ack(tp, th)
2915 tcp_seq onxt = tp->snd_nxt;
2916 u_long ocwnd = tp->snd_cwnd;
2918 callout_stop(tp->tt_rexmt);
2920 tp->snd_nxt = th->th_ack;
2922 * Set snd_cwnd to one segment beyond acknowledged offset
2923 * (tp->snd_una has not yet been updated when this function is called.)
2925 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2926 tp->t_flags |= TF_ACKNOW;
2927 (void) tcp_output(tp);
2928 tp->snd_cwnd = ocwnd;
2929 if (SEQ_GT(onxt, tp->snd_nxt))
2932 * Partial window deflation. Relies on fact that tp->snd_una
2935 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);