2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
33 #include "opt_ipfw.h" /* for ipfw_fwd */
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40 #include "opt_tcp_sack.h"
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/proc.h> /* for proc0 declaration */
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <sys/systm.h>
55 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
60 #include <net/route.h>
62 #include <netinet/in.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_options.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.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>
88 #include <netipsec/ipsec.h>
89 #include <netipsec/ipsec6.h>
93 #include <netinet6/ipsec.h>
94 #include <netinet6/ipsec6.h>
95 #include <netkey/key.h>
98 #include <machine/in_cksum.h>
100 #include <security/mac/mac_framework.h>
102 static const int tcprexmtthresh = 3;
104 struct tcpstat tcpstat;
105 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
106 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
108 static int log_in_vain = 0;
109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
110 &log_in_vain, 0, "Log all incoming TCP connections");
112 static int blackhole = 0;
113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
114 &blackhole, 0, "Do not send RST when dropping refused connections");
116 int tcp_delack_enabled = 1;
117 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
118 &tcp_delack_enabled, 0,
119 "Delay ACK to try and piggyback it onto a data packet");
121 #ifdef TCP_DROP_SYNFIN
122 static int drop_synfin = 0;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
124 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
127 static int tcp_do_rfc3042 = 1;
128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
129 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
131 static int tcp_do_rfc3390 = 1;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
134 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
136 static int tcp_insecure_rst = 0;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
138 &tcp_insecure_rst, 0,
139 "Follow the old (insecure) criteria for accepting RST packets.");
141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
142 "TCP Segment Reassembly Queue");
144 static int tcp_reass_maxseg = 0;
145 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
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_maxqlen = 48;
155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
156 &tcp_reass_maxqlen, 0,
157 "Maximum number of TCP Segments per individual Reassembly Queue");
159 static int tcp_reass_overflows = 0;
160 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
161 &tcp_reass_overflows, 0,
162 "Global number of TCP Segment Reassembly Queue Overflows");
164 struct inpcbhead tcb;
165 #define tcb6 tcb /* for KAME src sync over BSD*'s */
166 struct inpcbinfo tcbinfo;
167 struct mtx *tcbinfo_mtx;
169 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
171 static void tcp_pulloutofband(struct socket *,
172 struct tcphdr *, struct mbuf *, int);
173 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
175 static void tcp_xmit_timer(struct tcpcb *, int);
176 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
177 static int tcp_timewait(struct inpcb *, struct tcpopt *,
178 struct tcphdr *, struct mbuf *, int);
180 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
182 #define ND6_HINT(tp) \
184 if ((tp) && (tp)->t_inpcb && \
185 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
186 nd6_nud_hint(NULL, NULL, 0); \
193 * Indicate whether this ack should be delayed. We can delay the ack if
194 * - there is no delayed ack timer in progress and
195 * - our last ack wasn't a 0-sized window. We never want to delay
196 * the ack that opens up a 0-sized window and
197 * - delayed acks are enabled or
198 * - this is a half-synchronized T/TCP connection.
200 #define DELAY_ACK(tp) \
201 ((!callout_active(tp->tt_delack) && \
202 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
203 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
205 /* Initialize TCP reassembly queue */
207 tcp_reass_zone_change(void *tag)
210 tcp_reass_maxseg = nmbclusters / 16;
211 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
214 uma_zone_t tcp_reass_zone;
218 tcp_reass_maxseg = nmbclusters / 16;
219 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
221 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
222 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
223 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
224 EVENTHANDLER_REGISTER(nmbclusters_change,
225 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
229 tcp_reass(tp, th, tlenp, m)
230 register struct tcpcb *tp;
231 register struct tcphdr *th;
236 struct tseg_qent *p = NULL;
237 struct tseg_qent *nq;
238 struct tseg_qent *te = NULL;
239 struct socket *so = tp->t_inpcb->inp_socket;
242 INP_LOCK_ASSERT(tp->t_inpcb);
245 * XXX: tcp_reass() is rather inefficient with its data structures
246 * and should be rewritten (see NetBSD for optimizations). While
247 * doing that it should move to its own file tcp_reass.c.
251 * Call with th==NULL after become established to
252 * force pre-ESTABLISHED data up to user socket.
258 * Limit the number of segments in the reassembly queue to prevent
259 * holding on to too many segments (and thus running out of mbufs).
260 * Make sure to let the missing segment through which caused this
261 * queue. Always keep one global queue entry spare to be able to
262 * process the missing segment.
264 if (th->th_seq != tp->rcv_nxt &&
265 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
266 tp->t_segqlen >= tcp_reass_maxqlen)) {
267 tcp_reass_overflows++;
268 tcpstat.tcps_rcvmemdrop++;
275 * Allocate a new queue entry. If we can't, or hit the zone limit
278 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
280 tcpstat.tcps_rcvmemdrop++;
289 * Find a segment which begins after this one does.
291 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
292 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
298 * If there is a preceding segment, it may provide some of
299 * our data already. If so, drop the data from the incoming
300 * segment. If it provides all of our data, drop us.
304 /* conversion to int (in i) handles seq wraparound */
305 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
308 tcpstat.tcps_rcvduppack++;
309 tcpstat.tcps_rcvdupbyte += *tlenp;
311 uma_zfree(tcp_reass_zone, te);
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 register 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);
348 uma_zfree(tcp_reass_zone, q);
354 /* Insert the new segment queue entry into place. */
357 te->tqe_len = *tlenp;
360 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
362 LIST_INSERT_AFTER(p, te, tqe_q);
367 * Present data to user, advancing rcv_nxt through
368 * completed sequence space.
370 if (!TCPS_HAVEESTABLISHED(tp->t_state))
372 q = LIST_FIRST(&tp->t_segq);
373 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
375 SOCKBUF_LOCK(&so->so_rcv);
377 tp->rcv_nxt += q->tqe_len;
378 flags = q->tqe_th->th_flags & TH_FIN;
379 nq = LIST_NEXT(q, tqe_q);
380 LIST_REMOVE(q, tqe_q);
381 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
384 sbappendstream_locked(&so->so_rcv, q->tqe_m);
385 uma_zfree(tcp_reass_zone, q);
389 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
391 sorwakeup_locked(so);
396 * TCP input routine, follows pages 65-76 of the
397 * protocol specification dated September, 1981 very closely.
401 tcp6_input(mp, offp, proto)
405 register struct mbuf *m = *mp;
406 struct in6_ifaddr *ia6;
408 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
411 * draft-itojun-ipv6-tcp-to-anycast
412 * better place to put this in?
414 ia6 = ip6_getdstifaddr(m);
415 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
418 ip6 = mtod(m, struct ip6_hdr *);
419 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
420 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
431 register struct mbuf *m;
434 register struct tcphdr *th;
435 register struct ip *ip = NULL;
436 register struct ipovly *ipov;
437 register struct inpcb *inp = NULL;
442 register struct tcpcb *tp = 0;
443 register int thflags;
444 struct socket *so = 0;
445 int todrop, acked, ourfinisacked, needoutput = 0;
447 struct tcpopt to; /* options in this segment */
449 #ifdef IPFIREWALL_FORWARD
450 struct m_tag *fwd_tag;
452 int rstreason; /* For badport_bandlim accounting purposes */
454 struct ip6_hdr *ip6 = NULL;
457 char ip6buf[INET6_ADDRSTRLEN];
459 const int isipv6 = 0;
464 * The size of tcp_saveipgen must be the size of the max ip header,
467 u_char tcp_saveipgen[40];
468 struct tcphdr tcp_savetcp;
473 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
475 bzero((char *)&to, sizeof(to));
477 tcpstat.tcps_rcvtotal++;
481 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
482 ip6 = mtod(m, struct ip6_hdr *);
483 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
484 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
485 tcpstat.tcps_rcvbadsum++;
488 th = (struct tcphdr *)((caddr_t)ip6 + off0);
491 * Be proactive about unspecified IPv6 address in source.
492 * As we use all-zero to indicate unbounded/unconnected pcb,
493 * unspecified IPv6 address can be used to confuse us.
495 * Note that packets with unspecified IPv6 destination is
496 * already dropped in ip6_input.
498 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
503 th = NULL; /* XXX: avoid compiler warning */
507 * Get IP and TCP header together in first mbuf.
508 * Note: IP leaves IP header in first mbuf.
510 if (off0 > sizeof (struct ip)) {
511 ip_stripoptions(m, (struct mbuf *)0);
512 off0 = sizeof(struct ip);
514 if (m->m_len < sizeof (struct tcpiphdr)) {
515 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
516 tcpstat.tcps_rcvshort++;
520 ip = mtod(m, struct ip *);
521 ipov = (struct ipovly *)ip;
522 th = (struct tcphdr *)((caddr_t)ip + off0);
525 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
526 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
527 th->th_sum = m->m_pkthdr.csum_data;
529 th->th_sum = in_pseudo(ip->ip_src.s_addr,
531 htonl(m->m_pkthdr.csum_data +
534 th->th_sum ^= 0xffff;
536 ipov->ih_len = (u_short)tlen;
537 ipov->ih_len = htons(ipov->ih_len);
541 * Checksum extended TCP header and data.
543 len = sizeof (struct ip) + tlen;
544 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
545 ipov->ih_len = (u_short)tlen;
546 ipov->ih_len = htons(ipov->ih_len);
547 th->th_sum = in_cksum(m, len);
550 tcpstat.tcps_rcvbadsum++;
553 /* Re-initialization for later version check */
554 ip->ip_v = IPVERSION;
558 * Check that TCP offset makes sense,
559 * pull out TCP options and adjust length. XXX
561 off = th->th_off << 2;
562 if (off < sizeof (struct tcphdr) || off > tlen) {
563 tcpstat.tcps_rcvbadoff++;
566 tlen -= off; /* tlen is used instead of ti->ti_len */
567 if (off > sizeof (struct tcphdr)) {
570 IP6_EXTHDR_CHECK(m, off0, off, );
571 ip6 = mtod(m, struct ip6_hdr *);
572 th = (struct tcphdr *)((caddr_t)ip6 + off0);
575 if (m->m_len < sizeof(struct ip) + off) {
576 if ((m = m_pullup(m, sizeof (struct ip) + off))
578 tcpstat.tcps_rcvshort++;
581 ip = mtod(m, struct ip *);
582 ipov = (struct ipovly *)ip;
583 th = (struct tcphdr *)((caddr_t)ip + off0);
586 optlen = off - sizeof (struct tcphdr);
587 optp = (u_char *)(th + 1);
589 thflags = th->th_flags;
591 #ifdef TCP_DROP_SYNFIN
593 * If the drop_synfin option is enabled, drop all packets with
594 * both the SYN and FIN bits set. This prevents e.g. nmap from
595 * identifying the TCP/IP stack.
597 * This is a violation of the TCP specification.
599 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
604 * Convert TCP protocol specific fields to host format.
606 th->th_seq = ntohl(th->th_seq);
607 th->th_ack = ntohl(th->th_ack);
608 th->th_win = ntohs(th->th_win);
609 th->th_urp = ntohs(th->th_urp);
612 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
613 * until after ip6_savecontrol() is called and before other functions
614 * which don't want those proto headers.
615 * Because ip6_savecontrol() is going to parse the mbuf to
616 * search for data to be passed up to user-land, it wants mbuf
617 * parameters to be unchanged.
618 * XXX: the call of ip6_savecontrol() has been obsoleted based on
619 * latest version of the advanced API (20020110).
621 drop_hdrlen = off0 + off;
624 * Locate pcb for segment.
626 INP_INFO_WLOCK(&tcbinfo);
629 KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
630 #ifdef IPFIREWALL_FORWARD
631 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
632 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
634 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
635 struct sockaddr_in *next_hop;
637 next_hop = (struct sockaddr_in *)(fwd_tag+1);
639 * Transparently forwarded. Pretend to be the destination.
640 * already got one like this?
642 inp = in_pcblookup_hash(&tcbinfo,
643 ip->ip_src, th->th_sport,
644 ip->ip_dst, th->th_dport,
645 0, m->m_pkthdr.rcvif);
647 /* It's new. Try to find the ambushing socket. */
648 inp = in_pcblookup_hash(&tcbinfo,
649 ip->ip_src, th->th_sport,
652 ntohs(next_hop->sin_port) :
657 /* Remove the tag from the packet. We don't need it anymore. */
658 m_tag_delete(m, fwd_tag);
660 #endif /* IPFIREWALL_FORWARD */
663 inp = in6_pcblookup_hash(&tcbinfo,
664 &ip6->ip6_src, th->th_sport,
665 &ip6->ip6_dst, th->th_dport,
670 inp = in_pcblookup_hash(&tcbinfo,
671 ip->ip_src, th->th_sport,
672 ip->ip_dst, th->th_dport,
675 #ifdef IPFIREWALL_FORWARD
677 #endif /* IPFIREWALL_FORWARD */
679 #if defined(IPSEC) || defined(FAST_IPSEC)
682 if (inp != NULL && ipsec6_in_reject(m, inp)) {
684 ipsec6stat.in_polvio++;
690 if (inp != NULL && ipsec4_in_reject(m, inp)) {
692 ipsecstat.in_polvio++;
696 #endif /*IPSEC || FAST_IPSEC*/
699 * If the state is CLOSED (i.e., TCB does not exist) then
700 * all data in the incoming segment is discarded.
701 * If the TCB exists but is in CLOSED state, it is embryonic,
702 * but should either do a listen or a connect soon.
707 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
709 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
717 ip6_sprintf(ip6buf, &ip6->ip6_dst));
719 ip6_sprintf(ip6buf, &ip6->ip6_src));
724 strcpy(dbuf, inet_ntoa(ip->ip_dst));
725 strcpy(sbuf, inet_ntoa(ip->ip_src));
727 switch (log_in_vain) {
729 if ((thflags & TH_SYN) == 0)
734 "Connection attempt to TCP %s:%d "
735 "from %s:%d flags:0x%02x\n",
736 dbuf, ntohs(th->th_dport), sbuf,
737 ntohs(th->th_sport), thflags);
746 if (thflags & TH_SYN)
755 rstreason = BANDLIM_RST_CLOSEDPORT;
760 /* Check the minimum TTL for socket. */
761 if (inp->inp_ip_minttl != 0) {
763 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
767 if (inp->inp_ip_minttl > ip->ip_ttl)
771 if (inp->inp_vflag & INP_TIMEWAIT) {
773 * The only option of relevance is TOF_CC, and only if
774 * present in a SYN segment. See tcp_timewait().
776 if (thflags & TH_SYN)
777 tcp_dooptions(&to, optp, optlen, TO_SYN);
778 if (tcp_timewait(inp, &to, th, m, tlen))
781 * tcp_timewait unlocks inp.
783 INP_INFO_WUNLOCK(&tcbinfo);
789 rstreason = BANDLIM_RST_CLOSEDPORT;
792 if (tp->t_state == TCPS_CLOSED)
796 INP_LOCK_ASSERT(inp);
797 if (mac_check_inpcb_deliver(inp, m))
800 so = inp->inp_socket;
801 KASSERT(so != NULL, ("tcp_input: so == NULL"));
803 if (so->so_options & SO_DEBUG) {
804 ostate = tp->t_state;
806 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
808 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
812 if (so->so_options & SO_ACCEPTCONN) {
813 struct in_conninfo inc;
815 bzero(&inc, sizeof(inc));
817 inc.inc_isipv6 = isipv6;
820 inc.inc6_faddr = ip6->ip6_src;
821 inc.inc6_laddr = ip6->ip6_dst;
823 inc.inc_faddr = ip->ip_src;
824 inc.inc_laddr = ip->ip_dst;
826 inc.inc_fport = th->th_sport;
827 inc.inc_lport = th->th_dport;
830 * If the state is LISTEN then ignore segment if it contains
831 * a RST. If the segment contains an ACK then it is bad and
832 * send a RST. If it does not contain a SYN then it is not
833 * interesting; drop it.
835 * If the state is SYN_RECEIVED (syncache) and seg contains
836 * an ACK, but not for our SYN/ACK, send a RST. If the seg
837 * contains a RST, check the sequence number to see if it
838 * is a valid reset segment.
840 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
841 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
843 * Parse the TCP options here because
844 * syncookies need access to the reflected
847 tcp_dooptions(&to, optp, optlen, 0);
848 if (!syncache_expand(&inc, &to, th, &so, m)) {
850 * No syncache entry, or ACK was not
851 * for our SYN/ACK. Send a RST.
853 tcpstat.tcps_badsyn++;
854 rstreason = BANDLIM_RST_OPENPORT;
859 * Could not complete 3-way handshake,
860 * connection is being closed down, and
861 * syncache has free'd mbuf.
864 INP_INFO_WUNLOCK(&tcbinfo);
868 * Socket is created in state SYN_RECEIVED.
869 * Continue processing segment.
876 * This is what would have happened in
877 * tcp_output() when the SYN,ACK was sent.
879 tp->snd_up = tp->snd_una;
880 tp->snd_max = tp->snd_nxt = tp->iss + 1;
881 tp->last_ack_sent = tp->rcv_nxt;
884 if (thflags & TH_RST) {
885 syncache_chkrst(&inc, th);
888 if (thflags & TH_ACK) {
889 syncache_badack(&inc);
890 tcpstat.tcps_badsyn++;
891 rstreason = BANDLIM_RST_OPENPORT;
898 * Segment's flags are (SYN) or (SYN|FIN).
902 * If deprecated address is forbidden,
903 * we do not accept SYN to deprecated interface
904 * address to prevent any new inbound connection from
905 * getting established.
906 * When we do not accept SYN, we send a TCP RST,
907 * with deprecated source address (instead of dropping
908 * it). We compromise it as it is much better for peer
909 * to send a RST, and RST will be the final packet
912 * If we do not forbid deprecated addresses, we accept
913 * the SYN packet. RFC2462 does not suggest dropping
915 * If we decipher RFC2462 5.5.4, it says like this:
916 * 1. use of deprecated addr with existing
917 * communication is okay - "SHOULD continue to be
919 * 2. use of it with new communication:
920 * (2a) "SHOULD NOT be used if alternate address
921 * with sufficient scope is available"
922 * (2b) nothing mentioned otherwise.
923 * Here we fall into (2b) case as we have no choice in
924 * our source address selection - we must obey the peer.
926 * The wording in RFC2462 is confusing, and there are
927 * multiple description text for deprecated address
928 * handling - worse, they are not exactly the same.
929 * I believe 5.5.4 is the best one, so we follow 5.5.4.
931 if (isipv6 && !ip6_use_deprecated) {
932 struct in6_ifaddr *ia6;
934 if ((ia6 = ip6_getdstifaddr(m)) &&
935 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
938 rstreason = BANDLIM_RST_OPENPORT;
944 * If it is from this socket, drop it, it must be forged.
945 * Don't bother responding if the destination was a broadcast.
947 if (th->th_dport == th->th_sport) {
949 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
953 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
958 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
960 * Note that it is quite possible to receive unicast
961 * link-layer packets with a broadcast IP address. Use
962 * in_broadcast() to find them.
964 if (m->m_flags & (M_BCAST|M_MCAST))
967 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
968 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
971 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
972 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
973 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
974 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
978 * SYN appears to be valid; create compressed TCP state
979 * for syncache, or perform t/tcp connection.
981 if (so->so_qlen <= so->so_qlimit) {
983 if (so->so_options & SO_DEBUG)
984 tcp_trace(TA_INPUT, ostate, tp,
985 (void *)tcp_saveipgen, &tcp_savetcp, 0);
987 tcp_dooptions(&to, optp, optlen, TO_SYN);
988 if (!syncache_add(&inc, &to, th, inp, &so, m))
989 goto drop; /* XXX: does not happen */
992 * Entry added to syncache, mbuf used to
993 * send SYN,ACK packet. Everything unlocked
998 panic("T/TCP not supported at the moment");
1001 * Segment passed TAO tests.
1002 * XXX: Can't happen at the moment.
1005 inp = sotoinpcb(so);
1007 tp = intotcpcb(inp);
1008 tp->t_starttime = ticks;
1009 tp->t_state = TCPS_ESTABLISHED;
1013 * If there is a FIN or if there is data, then
1014 * delay SYN,ACK(SYN) in the hope of piggy-backing
1015 * it on a response segment. Otherwise must send
1016 * ACK now in case the other side is slow starting.
1018 if (thflags & TH_FIN || tlen != 0)
1019 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1021 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1022 tiwin = th->th_win << tp->snd_scale;
1023 tcpstat.tcps_connects++;
1031 KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1032 INP_LOCK_ASSERT(inp);
1034 /* Syncache takes care of sockets in the listen state. */
1035 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1038 * This is the second part of the MSS DoS prevention code (after
1039 * minmss on the sending side) and it deals with too many too small
1040 * tcp packets in a too short timeframe (1 second).
1042 * For every full second we count the number of received packets
1043 * and bytes. If we get a lot of packets per second for this connection
1044 * (tcp_minmssoverload) we take a closer look at it and compute the
1045 * average packet size for the past second. If that is less than
1046 * tcp_minmss we get too many packets with very small payload which
1047 * is not good and burdens our system (and every packet generates
1048 * a wakeup to the process connected to our socket). We can reasonable
1049 * expect this to be small packet DoS attack to exhaust our CPU
1052 * Care has to be taken for the minimum packet overload value. This
1053 * value defines the minimum number of packets per second before we
1054 * start to worry. This must not be too low to avoid killing for
1055 * example interactive connections with many small packets like
1058 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1061 * Account for packet if payload packet, skip over ACK, etc.
1063 if (tcp_minmss && tcp_minmssoverload &&
1064 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1065 if ((unsigned int)(tp->rcv_second - ticks) < hz) {
1067 tp->rcv_byps += tlen + off;
1068 if (tp->rcv_pps > tcp_minmssoverload) {
1069 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1070 printf("too many small tcp packets from "
1071 "%s:%u, av. %lubyte/packet, "
1072 "dropping connection\n",
1076 &inp->inp_inc.inc6_faddr) :
1078 inet_ntoa(inp->inp_inc.inc_faddr),
1079 inp->inp_inc.inc_fport,
1080 tp->rcv_byps / tp->rcv_pps);
1081 KASSERT(headlocked, ("tcp_input: "
1082 "after_listen: tcp_drop: head "
1084 tp = tcp_drop(tp, ECONNRESET);
1085 tcpstat.tcps_minmssdrops++;
1090 tp->rcv_second = ticks + hz;
1092 tp->rcv_byps = tlen + off;
1097 * Segment received on connection.
1098 * Reset idle time and keep-alive timer.
1100 tp->t_rcvtime = ticks;
1101 if (TCPS_HAVEESTABLISHED(tp->t_state))
1102 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1105 * Unscale the window into a 32-bit value.
1106 * This value is bogus for the TCPS_SYN_SENT state
1107 * and is overwritten later.
1109 tiwin = th->th_win << tp->snd_scale;
1112 * Parse options on any incoming segment.
1114 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0);
1117 * If echoed timestamp is later than the current time,
1118 * fall back to non RFC1323 RTT calculation. Normalize
1119 * timestamp if syncookies were used when this connection
1122 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1123 to.to_tsecr -= tp->ts_offset;
1124 if (TSTMP_GT(to.to_tsecr, ticks))
1129 * Process options only when we get SYN/ACK back. The SYN case
1130 * for incoming connections is handled in tcp_syncache.
1131 * XXX this is traditional behavior, may need to be cleaned up.
1133 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1134 if ((to.to_flags & TOF_SCALE) &&
1135 (tp->t_flags & TF_REQ_SCALE)) {
1136 tp->t_flags |= TF_RCVD_SCALE;
1137 tp->snd_scale = to.to_requested_s_scale;
1138 tp->snd_wnd = th->th_win << tp->snd_scale;
1139 tiwin = tp->snd_wnd;
1141 if (to.to_flags & TOF_TS) {
1142 tp->t_flags |= TF_RCVD_TSTMP;
1143 tp->ts_recent = to.to_tsval;
1144 tp->ts_recent_age = ticks;
1146 if (to.to_flags & TOF_MSS)
1147 tcp_mss(tp, to.to_mss);
1148 if (tp->sack_enable) {
1149 if (!(to.to_flags & TOF_SACK))
1150 tp->sack_enable = 0;
1152 tp->t_flags |= TF_SACK_PERMIT;
1158 * Header prediction: check for the two common cases
1159 * of a uni-directional data xfer. If the packet has
1160 * no control flags, is in-sequence, the window didn't
1161 * change and we're not retransmitting, it's a
1162 * candidate. If the length is zero and the ack moved
1163 * forward, we're the sender side of the xfer. Just
1164 * free the data acked & wake any higher level process
1165 * that was blocked waiting for space. If the length
1166 * is non-zero and the ack didn't move, we're the
1167 * receiver side. If we're getting packets in-order
1168 * (the reassembly queue is empty), add the data to
1169 * the socket buffer and note that we need a delayed ack.
1170 * Make sure that the hidden state-flags are also off.
1171 * Since we check for TCPS_ESTABLISHED above, it can only
1174 if (tp->t_state == TCPS_ESTABLISHED &&
1175 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1176 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1177 ((to.to_flags & TOF_TS) == 0 ||
1178 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1179 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1180 tp->snd_nxt == tp->snd_max) {
1183 * If last ACK falls within this segment's sequence numbers,
1184 * record the timestamp.
1185 * NOTE that the test is modified according to the latest
1186 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1188 if ((to.to_flags & TOF_TS) != 0 &&
1189 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1190 tp->ts_recent_age = ticks;
1191 tp->ts_recent = to.to_tsval;
1195 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1196 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1197 tp->snd_cwnd >= tp->snd_wnd &&
1198 ((!tcp_do_newreno && !tp->sack_enable &&
1199 tp->t_dupacks < tcprexmtthresh) ||
1200 ((tcp_do_newreno || tp->sack_enable) &&
1201 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1202 TAILQ_EMPTY(&tp->snd_holes)))) {
1203 KASSERT(headlocked, ("headlocked"));
1204 INP_INFO_WUNLOCK(&tcbinfo);
1207 * this is a pure ack for outstanding data.
1209 ++tcpstat.tcps_predack;
1211 * "bad retransmit" recovery
1213 if (tp->t_rxtshift == 1 &&
1214 ticks < tp->t_badrxtwin) {
1215 ++tcpstat.tcps_sndrexmitbad;
1216 tp->snd_cwnd = tp->snd_cwnd_prev;
1218 tp->snd_ssthresh_prev;
1219 tp->snd_recover = tp->snd_recover_prev;
1220 if (tp->t_flags & TF_WASFRECOVERY)
1221 ENTER_FASTRECOVERY(tp);
1222 tp->snd_nxt = tp->snd_max;
1223 tp->t_badrxtwin = 0;
1227 * Recalculate the transmit timer / rtt.
1229 * Some boxes send broken timestamp replies
1230 * during the SYN+ACK phase, ignore
1231 * timestamps of 0 or we could calculate a
1232 * huge RTT and blow up the retransmit timer.
1234 if ((to.to_flags & TOF_TS) != 0 &&
1236 if (!tp->t_rttlow ||
1237 tp->t_rttlow > ticks - to.to_tsecr)
1238 tp->t_rttlow = ticks - to.to_tsecr;
1240 ticks - to.to_tsecr + 1);
1241 } else if (tp->t_rtttime &&
1242 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1243 if (!tp->t_rttlow ||
1244 tp->t_rttlow > ticks - tp->t_rtttime)
1245 tp->t_rttlow = ticks - tp->t_rtttime;
1247 ticks - tp->t_rtttime);
1249 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1250 acked = th->th_ack - tp->snd_una;
1251 tcpstat.tcps_rcvackpack++;
1252 tcpstat.tcps_rcvackbyte += acked;
1253 sbdrop(&so->so_snd, acked);
1254 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1255 SEQ_LEQ(th->th_ack, tp->snd_recover))
1256 tp->snd_recover = th->th_ack - 1;
1257 tp->snd_una = th->th_ack;
1259 * pull snd_wl2 up to prevent seq wrap relative
1262 tp->snd_wl2 = th->th_ack;
1265 ND6_HINT(tp); /* some progress has been done */
1268 * If all outstanding data are acked, stop
1269 * retransmit timer, otherwise restart timer
1270 * using current (possibly backed-off) value.
1271 * If process is waiting for space,
1272 * wakeup/selwakeup/signal. If data
1273 * are ready to send, let tcp_output
1274 * decide between more output or persist.
1277 if (so->so_options & SO_DEBUG)
1278 tcp_trace(TA_INPUT, ostate, tp,
1279 (void *)tcp_saveipgen,
1283 if (tp->snd_una == tp->snd_max)
1284 callout_stop(tp->tt_rexmt);
1285 else if (!callout_active(tp->tt_persist))
1286 callout_reset(tp->tt_rexmt,
1288 tcp_timer_rexmt, tp);
1291 if (so->so_snd.sb_cc)
1292 (void) tcp_output(tp);
1295 } else if (th->th_ack == tp->snd_una &&
1296 LIST_EMPTY(&tp->t_segq) &&
1297 tlen <= sbspace(&so->so_rcv)) {
1298 KASSERT(headlocked, ("headlocked"));
1299 INP_INFO_WUNLOCK(&tcbinfo);
1302 * this is a pure, in-sequence data packet
1303 * with nothing on the reassembly queue and
1304 * we have enough buffer space to take it.
1306 /* Clean receiver SACK report if present */
1307 if (tp->sack_enable && tp->rcv_numsacks)
1308 tcp_clean_sackreport(tp);
1309 ++tcpstat.tcps_preddat;
1310 tp->rcv_nxt += tlen;
1312 * Pull snd_wl1 up to prevent seq wrap relative to
1315 tp->snd_wl1 = th->th_seq;
1317 * Pull rcv_up up to prevent seq wrap relative to
1320 tp->rcv_up = tp->rcv_nxt;
1321 tcpstat.tcps_rcvpack++;
1322 tcpstat.tcps_rcvbyte += tlen;
1323 ND6_HINT(tp); /* some progress has been done */
1326 if (so->so_options & SO_DEBUG)
1327 tcp_trace(TA_INPUT, ostate, tp,
1328 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1330 * Add data to socket buffer.
1332 SOCKBUF_LOCK(&so->so_rcv);
1333 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1336 m_adj(m, drop_hdrlen); /* delayed header drop */
1337 sbappendstream_locked(&so->so_rcv, m);
1339 sorwakeup_locked(so);
1340 if (DELAY_ACK(tp)) {
1341 tp->t_flags |= TF_DELACK;
1343 tp->t_flags |= TF_ACKNOW;
1351 * Calculate amount of space in receive window,
1352 * and then do TCP input processing.
1353 * Receive window is amount of space in rcv queue,
1354 * but not less than advertised window.
1358 win = sbspace(&so->so_rcv);
1361 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1364 switch (tp->t_state) {
1367 * If the state is SYN_RECEIVED:
1368 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1370 case TCPS_SYN_RECEIVED:
1371 if ((thflags & TH_ACK) &&
1372 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1373 SEQ_GT(th->th_ack, tp->snd_max))) {
1374 rstreason = BANDLIM_RST_OPENPORT;
1380 * If the state is SYN_SENT:
1381 * if seg contains an ACK, but not for our SYN, drop the input.
1382 * if seg contains a RST, then drop the connection.
1383 * if seg does not contain SYN, then drop it.
1384 * Otherwise this is an acceptable SYN segment
1385 * initialize tp->rcv_nxt and tp->irs
1386 * if seg contains ack then advance tp->snd_una
1387 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1388 * arrange for segment to be acked (eventually)
1389 * continue processing rest of data/controls, beginning with URG
1392 if ((thflags & TH_ACK) &&
1393 (SEQ_LEQ(th->th_ack, tp->iss) ||
1394 SEQ_GT(th->th_ack, tp->snd_max))) {
1395 rstreason = BANDLIM_UNLIMITED;
1398 if (thflags & TH_RST) {
1399 if (thflags & TH_ACK) {
1400 KASSERT(headlocked, ("tcp_input: after_listen"
1401 ": tcp_drop.2: head not locked"));
1402 tp = tcp_drop(tp, ECONNREFUSED);
1406 if ((thflags & TH_SYN) == 0)
1409 /* Initial send window, already scaled. */
1410 tp->snd_wnd = th->th_win;
1412 tp->irs = th->th_seq;
1414 if (thflags & TH_ACK) {
1415 tcpstat.tcps_connects++;
1419 mac_set_socket_peer_from_mbuf(m, so);
1422 /* Do window scaling on this connection? */
1423 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1424 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1425 tp->rcv_scale = tp->request_r_scale;
1427 tp->rcv_adv += tp->rcv_wnd;
1428 tp->snd_una++; /* SYN is acked */
1430 * If there's data, delay ACK; if there's also a FIN
1431 * ACKNOW will be turned on later.
1433 if (DELAY_ACK(tp) && tlen != 0)
1434 callout_reset(tp->tt_delack, tcp_delacktime,
1435 tcp_timer_delack, tp);
1437 tp->t_flags |= TF_ACKNOW;
1439 * Received <SYN,ACK> in SYN_SENT[*] state.
1441 * SYN_SENT --> ESTABLISHED
1442 * SYN_SENT* --> FIN_WAIT_1
1444 tp->t_starttime = ticks;
1445 if (tp->t_flags & TF_NEEDFIN) {
1446 tp->t_state = TCPS_FIN_WAIT_1;
1447 tp->t_flags &= ~TF_NEEDFIN;
1450 tp->t_state = TCPS_ESTABLISHED;
1451 callout_reset(tp->tt_keep, tcp_keepidle,
1452 tcp_timer_keep, tp);
1456 * Received initial SYN in SYN-SENT[*] state =>
1457 * simultaneous open. If segment contains CC option
1458 * and there is a cached CC, apply TAO test.
1459 * If it succeeds, connection is * half-synchronized.
1460 * Otherwise, do 3-way handshake:
1461 * SYN-SENT -> SYN-RECEIVED
1462 * SYN-SENT* -> SYN-RECEIVED*
1463 * If there was no CC option, clear cached CC value.
1465 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1466 callout_stop(tp->tt_rexmt);
1467 tp->t_state = TCPS_SYN_RECEIVED;
1473 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1475 INP_LOCK_ASSERT(inp);
1478 * Advance th->th_seq to correspond to first data byte.
1479 * If data, trim to stay within window,
1480 * dropping FIN if necessary.
1483 if (tlen > tp->rcv_wnd) {
1484 todrop = tlen - tp->rcv_wnd;
1488 tcpstat.tcps_rcvpackafterwin++;
1489 tcpstat.tcps_rcvbyteafterwin += todrop;
1491 tp->snd_wl1 = th->th_seq - 1;
1492 tp->rcv_up = th->th_seq;
1494 * Client side of transaction: already sent SYN and data.
1495 * If the remote host used T/TCP to validate the SYN,
1496 * our data will be ACK'd; if so, enter normal data segment
1497 * processing in the middle of step 5, ack processing.
1498 * Otherwise, goto step 6.
1500 if (thflags & TH_ACK)
1506 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1507 * do normal processing.
1509 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1513 case TCPS_TIME_WAIT:
1514 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1515 break; /* continue normal processing */
1519 * States other than LISTEN or SYN_SENT.
1520 * First check the RST flag and sequence number since reset segments
1521 * are exempt from the timestamp and connection count tests. This
1522 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1523 * below which allowed reset segments in half the sequence space
1524 * to fall though and be processed (which gives forged reset
1525 * segments with a random sequence number a 50 percent chance of
1526 * killing a connection).
1527 * Then check timestamp, if present.
1528 * Then check the connection count, if present.
1529 * Then check that at least some bytes of segment are within
1530 * receive window. If segment begins before rcv_nxt,
1531 * drop leading data (and SYN); if nothing left, just ack.
1534 * If the RST bit is set, check the sequence number to see
1535 * if this is a valid reset segment.
1537 * In all states except SYN-SENT, all reset (RST) segments
1538 * are validated by checking their SEQ-fields. A reset is
1539 * valid if its sequence number is in the window.
1540 * Note: this does not take into account delayed ACKs, so
1541 * we should test against last_ack_sent instead of rcv_nxt.
1542 * The sequence number in the reset segment is normally an
1543 * echo of our outgoing acknowlegement numbers, but some hosts
1544 * send a reset with the sequence number at the rightmost edge
1545 * of our receive window, and we have to handle this case.
1546 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1547 * that brute force RST attacks are possible. To combat this,
1548 * we use a much stricter check while in the ESTABLISHED state,
1549 * only accepting RSTs where the sequence number is equal to
1550 * last_ack_sent. In all other states (the states in which a
1551 * RST is more likely), the more permissive check is used.
1552 * If we have multiple segments in flight, the intial reset
1553 * segment sequence numbers will be to the left of last_ack_sent,
1554 * but they will eventually catch up.
1555 * In any case, it never made sense to trim reset segments to
1556 * fit the receive window since RFC 1122 says:
1557 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1559 * A TCP SHOULD allow a received RST segment to include data.
1562 * It has been suggested that a RST segment could contain
1563 * ASCII text that encoded and explained the cause of the
1564 * RST. No standard has yet been established for such
1567 * If the reset segment passes the sequence number test examine
1569 * SYN_RECEIVED STATE:
1570 * If passive open, return to LISTEN state.
1571 * If active open, inform user that connection was refused.
1572 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1573 * Inform user that connection was reset, and close tcb.
1574 * CLOSING, LAST_ACK STATES:
1577 * Drop the segment - see Stevens, vol. 2, p. 964 and
1580 if (thflags & TH_RST) {
1581 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1582 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1583 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1584 switch (tp->t_state) {
1586 case TCPS_SYN_RECEIVED:
1587 so->so_error = ECONNREFUSED;
1590 case TCPS_ESTABLISHED:
1591 if (tp->last_ack_sent != th->th_seq &&
1592 tcp_insecure_rst == 0) {
1593 tcpstat.tcps_badrst++;
1596 case TCPS_FIN_WAIT_1:
1597 case TCPS_FIN_WAIT_2:
1598 case TCPS_CLOSE_WAIT:
1599 so->so_error = ECONNRESET;
1601 tp->t_state = TCPS_CLOSED;
1602 tcpstat.tcps_drops++;
1603 KASSERT(headlocked, ("tcp_input: "
1604 "trimthenstep6: tcp_close: head not "
1611 KASSERT(headlocked, ("trimthenstep6: "
1612 "tcp_close.2: head not locked"));
1616 case TCPS_TIME_WAIT:
1617 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1626 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1627 * and it's less than ts_recent, drop it.
1629 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1630 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1632 /* Check to see if ts_recent is over 24 days old. */
1633 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1635 * Invalidate ts_recent. If this segment updates
1636 * ts_recent, the age will be reset later and ts_recent
1637 * will get a valid value. If it does not, setting
1638 * ts_recent to zero will at least satisfy the
1639 * requirement that zero be placed in the timestamp
1640 * echo reply when ts_recent isn't valid. The
1641 * age isn't reset until we get a valid ts_recent
1642 * because we don't want out-of-order segments to be
1643 * dropped when ts_recent is old.
1647 tcpstat.tcps_rcvduppack++;
1648 tcpstat.tcps_rcvdupbyte += tlen;
1649 tcpstat.tcps_pawsdrop++;
1657 * In the SYN-RECEIVED state, validate that the packet belongs to
1658 * this connection before trimming the data to fit the receive
1659 * window. Check the sequence number versus IRS since we know
1660 * the sequence numbers haven't wrapped. This is a partial fix
1661 * for the "LAND" DoS attack.
1663 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1664 rstreason = BANDLIM_RST_OPENPORT;
1668 todrop = tp->rcv_nxt - th->th_seq;
1670 if (thflags & TH_SYN) {
1680 * Following if statement from Stevens, vol. 2, p. 960.
1683 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1685 * Any valid FIN must be to the left of the window.
1686 * At this point the FIN must be a duplicate or out
1687 * of sequence; drop it.
1692 * Send an ACK to resynchronize and drop any data.
1693 * But keep on processing for RST or ACK.
1695 tp->t_flags |= TF_ACKNOW;
1697 tcpstat.tcps_rcvduppack++;
1698 tcpstat.tcps_rcvdupbyte += todrop;
1700 tcpstat.tcps_rcvpartduppack++;
1701 tcpstat.tcps_rcvpartdupbyte += todrop;
1703 drop_hdrlen += todrop; /* drop from the top afterwards */
1704 th->th_seq += todrop;
1706 if (th->th_urp > todrop)
1707 th->th_urp -= todrop;
1715 * If new data are received on a connection after the
1716 * user processes are gone, then RST the other end.
1718 if ((so->so_state & SS_NOFDREF) &&
1719 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1720 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1723 tcpstat.tcps_rcvafterclose++;
1724 rstreason = BANDLIM_UNLIMITED;
1729 * If segment ends after window, drop trailing data
1730 * (and PUSH and FIN); if nothing left, just ACK.
1732 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1734 tcpstat.tcps_rcvpackafterwin++;
1735 if (todrop >= tlen) {
1736 tcpstat.tcps_rcvbyteafterwin += tlen;
1738 * If a new connection request is received
1739 * while in TIME_WAIT, drop the old connection
1740 * and start over if the sequence numbers
1741 * are above the previous ones.
1743 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1744 if (thflags & TH_SYN &&
1745 tp->t_state == TCPS_TIME_WAIT &&
1746 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1747 KASSERT(headlocked, ("trimthenstep6: "
1748 "tcp_close.4: head not locked"));
1753 * If window is closed can only take segments at
1754 * window edge, and have to drop data and PUSH from
1755 * incoming segments. Continue processing, but
1756 * remember to ack. Otherwise, drop segment
1759 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1760 tp->t_flags |= TF_ACKNOW;
1761 tcpstat.tcps_rcvwinprobe++;
1765 tcpstat.tcps_rcvbyteafterwin += todrop;
1768 thflags &= ~(TH_PUSH|TH_FIN);
1772 * If last ACK falls within this segment's sequence numbers,
1773 * record its timestamp.
1775 * 1) That the test incorporates suggestions from the latest
1776 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1777 * 2) That updating only on newer timestamps interferes with
1778 * our earlier PAWS tests, so this check should be solely
1779 * predicated on the sequence space of this segment.
1780 * 3) That we modify the segment boundary check to be
1781 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1782 * instead of RFC1323's
1783 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1784 * This modified check allows us to overcome RFC1323's
1785 * limitations as described in Stevens TCP/IP Illustrated
1786 * Vol. 2 p.869. In such cases, we can still calculate the
1787 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1789 if ((to.to_flags & TOF_TS) != 0 &&
1790 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1791 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1792 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1793 tp->ts_recent_age = ticks;
1794 tp->ts_recent = to.to_tsval;
1798 * If a SYN is in the window, then this is an
1799 * error and we send an RST and drop the connection.
1801 if (thflags & TH_SYN) {
1802 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1803 "head not locked"));
1804 tp = tcp_drop(tp, ECONNRESET);
1805 rstreason = BANDLIM_UNLIMITED;
1810 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1811 * flag is on (half-synchronized state), then queue data for
1812 * later processing; else drop segment and return.
1814 if ((thflags & TH_ACK) == 0) {
1815 if (tp->t_state == TCPS_SYN_RECEIVED ||
1816 (tp->t_flags & TF_NEEDSYN))
1818 else if (tp->t_flags & TF_ACKNOW)
1827 switch (tp->t_state) {
1830 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1831 * ESTABLISHED state and continue processing.
1832 * The ACK was checked above.
1834 case TCPS_SYN_RECEIVED:
1836 tcpstat.tcps_connects++;
1838 /* Do window scaling? */
1839 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1840 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1841 tp->rcv_scale = tp->request_r_scale;
1842 tp->snd_wnd = tiwin;
1846 * SYN-RECEIVED -> ESTABLISHED
1847 * SYN-RECEIVED* -> FIN-WAIT-1
1849 tp->t_starttime = ticks;
1850 if (tp->t_flags & TF_NEEDFIN) {
1851 tp->t_state = TCPS_FIN_WAIT_1;
1852 tp->t_flags &= ~TF_NEEDFIN;
1854 tp->t_state = TCPS_ESTABLISHED;
1855 callout_reset(tp->tt_keep, tcp_keepidle,
1856 tcp_timer_keep, tp);
1859 * If segment contains data or ACK, will call tcp_reass()
1860 * later; if not, do so now to pass queued data to user.
1862 if (tlen == 0 && (thflags & TH_FIN) == 0)
1863 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1865 tp->snd_wl1 = th->th_seq - 1;
1869 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1870 * ACKs. If the ack is in the range
1871 * tp->snd_una < th->th_ack <= tp->snd_max
1872 * then advance tp->snd_una to th->th_ack and drop
1873 * data from the retransmission queue. If this ACK reflects
1874 * more up to date window information we update our window information.
1876 case TCPS_ESTABLISHED:
1877 case TCPS_FIN_WAIT_1:
1878 case TCPS_FIN_WAIT_2:
1879 case TCPS_CLOSE_WAIT:
1882 case TCPS_TIME_WAIT:
1883 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1884 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1885 tcpstat.tcps_rcvacktoomuch++;
1888 if (tp->sack_enable &&
1889 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1890 tcp_sack_doack(tp, &to, th->th_ack);
1891 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1892 if (tlen == 0 && tiwin == tp->snd_wnd) {
1893 tcpstat.tcps_rcvdupack++;
1895 * If we have outstanding data (other than
1896 * a window probe), this is a completely
1897 * duplicate ack (ie, window info didn't
1898 * change), the ack is the biggest we've
1899 * seen and we've seen exactly our rexmt
1900 * threshhold of them, assume a packet
1901 * has been dropped and retransmit it.
1902 * Kludge snd_nxt & the congestion
1903 * window so we send only this one
1906 * We know we're losing at the current
1907 * window size so do congestion avoidance
1908 * (set ssthresh to half the current window
1909 * and pull our congestion window back to
1910 * the new ssthresh).
1912 * Dup acks mean that packets have left the
1913 * network (they're now cached at the receiver)
1914 * so bump cwnd by the amount in the receiver
1915 * to keep a constant cwnd packets in the
1918 if (!callout_active(tp->tt_rexmt) ||
1919 th->th_ack != tp->snd_una)
1921 else if (++tp->t_dupacks > tcprexmtthresh ||
1922 ((tcp_do_newreno || tp->sack_enable) &&
1923 IN_FASTRECOVERY(tp))) {
1924 if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1928 * Compute the amount of data in flight first.
1929 * We can inject new data into the pipe iff
1930 * we have less than 1/2 the original window's
1931 * worth of data in flight.
1933 awnd = (tp->snd_nxt - tp->snd_fack) +
1934 tp->sackhint.sack_bytes_rexmit;
1935 if (awnd < tp->snd_ssthresh) {
1936 tp->snd_cwnd += tp->t_maxseg;
1937 if (tp->snd_cwnd > tp->snd_ssthresh)
1938 tp->snd_cwnd = tp->snd_ssthresh;
1941 tp->snd_cwnd += tp->t_maxseg;
1942 (void) tcp_output(tp);
1944 } else if (tp->t_dupacks == tcprexmtthresh) {
1945 tcp_seq onxt = tp->snd_nxt;
1949 * If we're doing sack, check to
1950 * see if we're already in sack
1951 * recovery. If we're not doing sack,
1952 * check to see if we're in newreno
1955 if (tp->sack_enable) {
1956 if (IN_FASTRECOVERY(tp)) {
1960 } else if (tcp_do_newreno) {
1961 if (SEQ_LEQ(th->th_ack,
1967 win = min(tp->snd_wnd, tp->snd_cwnd) /
1971 tp->snd_ssthresh = win * tp->t_maxseg;
1972 ENTER_FASTRECOVERY(tp);
1973 tp->snd_recover = tp->snd_max;
1974 callout_stop(tp->tt_rexmt);
1976 if (tp->sack_enable) {
1977 tcpstat.tcps_sack_recovery_episode++;
1978 tp->sack_newdata = tp->snd_nxt;
1979 tp->snd_cwnd = tp->t_maxseg;
1980 (void) tcp_output(tp);
1983 tp->snd_nxt = th->th_ack;
1984 tp->snd_cwnd = tp->t_maxseg;
1985 (void) tcp_output(tp);
1986 KASSERT(tp->snd_limited <= 2,
1987 ("tp->snd_limited too big"));
1988 tp->snd_cwnd = tp->snd_ssthresh +
1990 (tp->t_dupacks - tp->snd_limited);
1991 if (SEQ_GT(onxt, tp->snd_nxt))
1994 } else if (tcp_do_rfc3042) {
1995 u_long oldcwnd = tp->snd_cwnd;
1996 tcp_seq oldsndmax = tp->snd_max;
1999 KASSERT(tp->t_dupacks == 1 ||
2001 ("dupacks not 1 or 2"));
2002 if (tp->t_dupacks == 1)
2003 tp->snd_limited = 0;
2005 (tp->snd_nxt - tp->snd_una) +
2006 (tp->t_dupacks - tp->snd_limited) *
2008 (void) tcp_output(tp);
2009 sent = tp->snd_max - oldsndmax;
2010 if (sent > tp->t_maxseg) {
2011 KASSERT((tp->t_dupacks == 2 &&
2012 tp->snd_limited == 0) ||
2013 (sent == tp->t_maxseg + 1 &&
2014 tp->t_flags & TF_SENTFIN),
2016 tp->snd_limited = 2;
2017 } else if (sent > 0)
2019 tp->snd_cwnd = oldcwnd;
2027 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2030 * If the congestion window was inflated to account
2031 * for the other side's cached packets, retract it.
2033 if (tcp_do_newreno || tp->sack_enable) {
2034 if (IN_FASTRECOVERY(tp)) {
2035 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2036 if (tp->sack_enable)
2037 tcp_sack_partialack(tp, th);
2039 tcp_newreno_partial_ack(tp, th);
2042 * Out of fast recovery.
2043 * Window inflation should have left us
2044 * with approximately snd_ssthresh
2046 * But in case we would be inclined to
2047 * send a burst, better to do it via
2048 * the slow start mechanism.
2050 if (SEQ_GT(th->th_ack +
2053 tp->snd_cwnd = tp->snd_max -
2057 tp->snd_cwnd = tp->snd_ssthresh;
2061 if (tp->t_dupacks >= tcprexmtthresh &&
2062 tp->snd_cwnd > tp->snd_ssthresh)
2063 tp->snd_cwnd = tp->snd_ssthresh;
2067 * If we reach this point, ACK is not a duplicate,
2068 * i.e., it ACKs something we sent.
2070 if (tp->t_flags & TF_NEEDSYN) {
2072 * T/TCP: Connection was half-synchronized, and our
2073 * SYN has been ACK'd (so connection is now fully
2074 * synchronized). Go to non-starred state,
2075 * increment snd_una for ACK of SYN, and check if
2076 * we can do window scaling.
2078 tp->t_flags &= ~TF_NEEDSYN;
2080 /* Do window scaling? */
2081 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2082 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2083 tp->rcv_scale = tp->request_r_scale;
2084 /* Send window already scaled. */
2089 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2091 INP_LOCK_ASSERT(inp);
2093 acked = th->th_ack - tp->snd_una;
2094 tcpstat.tcps_rcvackpack++;
2095 tcpstat.tcps_rcvackbyte += acked;
2098 * If we just performed our first retransmit, and the ACK
2099 * arrives within our recovery window, then it was a mistake
2100 * to do the retransmit in the first place. Recover our
2101 * original cwnd and ssthresh, and proceed to transmit where
2104 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2105 ++tcpstat.tcps_sndrexmitbad;
2106 tp->snd_cwnd = tp->snd_cwnd_prev;
2107 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2108 tp->snd_recover = tp->snd_recover_prev;
2109 if (tp->t_flags & TF_WASFRECOVERY)
2110 ENTER_FASTRECOVERY(tp);
2111 tp->snd_nxt = tp->snd_max;
2112 tp->t_badrxtwin = 0; /* XXX probably not required */
2116 * If we have a timestamp reply, update smoothed
2117 * round trip time. If no timestamp is present but
2118 * transmit timer is running and timed sequence
2119 * number was acked, update smoothed round trip time.
2120 * Since we now have an rtt measurement, cancel the
2121 * timer backoff (cf., Phil Karn's retransmit alg.).
2122 * Recompute the initial retransmit timer.
2124 * Some boxes send broken timestamp replies
2125 * during the SYN+ACK phase, ignore
2126 * timestamps of 0 or we could calculate a
2127 * huge RTT and blow up the retransmit timer.
2129 if ((to.to_flags & TOF_TS) != 0 &&
2131 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2132 tp->t_rttlow = ticks - to.to_tsecr;
2133 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2134 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2135 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2136 tp->t_rttlow = ticks - tp->t_rtttime;
2137 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2139 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2142 * If all outstanding data is acked, stop retransmit
2143 * timer and remember to restart (more output or persist).
2144 * If there is more data to be acked, restart retransmit
2145 * timer, using current (possibly backed-off) value.
2147 if (th->th_ack == tp->snd_max) {
2148 callout_stop(tp->tt_rexmt);
2150 } else if (!callout_active(tp->tt_persist))
2151 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2152 tcp_timer_rexmt, tp);
2155 * If no data (only SYN) was ACK'd,
2156 * skip rest of ACK processing.
2162 * When new data is acked, open the congestion window.
2163 * If the window gives us less than ssthresh packets
2164 * in flight, open exponentially (maxseg per packet).
2165 * Otherwise open linearly: maxseg per window
2166 * (maxseg^2 / cwnd per packet).
2168 if ((!tcp_do_newreno && !tp->sack_enable) ||
2169 !IN_FASTRECOVERY(tp)) {
2170 register u_int cw = tp->snd_cwnd;
2171 register u_int incr = tp->t_maxseg;
2172 if (cw > tp->snd_ssthresh)
2173 incr = incr * incr / cw;
2174 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2176 SOCKBUF_LOCK(&so->so_snd);
2177 if (acked > so->so_snd.sb_cc) {
2178 tp->snd_wnd -= so->so_snd.sb_cc;
2179 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2182 sbdrop_locked(&so->so_snd, acked);
2183 tp->snd_wnd -= acked;
2186 sowwakeup_locked(so);
2187 /* detect una wraparound */
2188 if ((tcp_do_newreno || tp->sack_enable) &&
2189 !IN_FASTRECOVERY(tp) &&
2190 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2191 SEQ_LEQ(th->th_ack, tp->snd_recover))
2192 tp->snd_recover = th->th_ack - 1;
2193 if ((tcp_do_newreno || tp->sack_enable) &&
2194 IN_FASTRECOVERY(tp) &&
2195 SEQ_GEQ(th->th_ack, tp->snd_recover))
2196 EXIT_FASTRECOVERY(tp);
2197 tp->snd_una = th->th_ack;
2198 if (tp->sack_enable) {
2199 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2200 tp->snd_recover = tp->snd_una;
2202 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2203 tp->snd_nxt = tp->snd_una;
2205 switch (tp->t_state) {
2208 * In FIN_WAIT_1 STATE in addition to the processing
2209 * for the ESTABLISHED state if our FIN is now acknowledged
2210 * then enter FIN_WAIT_2.
2212 case TCPS_FIN_WAIT_1:
2213 if (ourfinisacked) {
2215 * If we can't receive any more
2216 * data, then closing user can proceed.
2217 * Starting the timer is contrary to the
2218 * specification, but if we don't get a FIN
2219 * we'll hang forever.
2222 * we should release the tp also, and use a
2225 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2226 soisdisconnected(so);
2227 callout_reset(tp->tt_2msl, tcp_maxidle,
2228 tcp_timer_2msl, tp);
2230 tp->t_state = TCPS_FIN_WAIT_2;
2235 * In CLOSING STATE in addition to the processing for
2236 * the ESTABLISHED state if the ACK acknowledges our FIN
2237 * then enter the TIME-WAIT state, otherwise ignore
2241 if (ourfinisacked) {
2242 KASSERT(headlocked, ("tcp_input: process_ACK: "
2243 "head not locked"));
2245 INP_INFO_WUNLOCK(&tcbinfo);
2252 * In LAST_ACK, we may still be waiting for data to drain
2253 * and/or to be acked, as well as for the ack of our FIN.
2254 * If our FIN is now acknowledged, delete the TCB,
2255 * enter the closed state and return.
2258 if (ourfinisacked) {
2259 KASSERT(headlocked, ("tcp_input: process_ACK:"
2260 " tcp_close: head not locked"));
2267 * In TIME_WAIT state the only thing that should arrive
2268 * is a retransmission of the remote FIN. Acknowledge
2269 * it and restart the finack timer.
2271 case TCPS_TIME_WAIT:
2272 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2273 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2274 tcp_timer_2msl, tp);
2280 KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2281 INP_LOCK_ASSERT(inp);
2284 * Update window information.
2285 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2287 if ((thflags & TH_ACK) &&
2288 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2289 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2290 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2291 /* keep track of pure window updates */
2293 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2294 tcpstat.tcps_rcvwinupd++;
2295 tp->snd_wnd = tiwin;
2296 tp->snd_wl1 = th->th_seq;
2297 tp->snd_wl2 = th->th_ack;
2298 if (tp->snd_wnd > tp->max_sndwnd)
2299 tp->max_sndwnd = tp->snd_wnd;
2304 * Process segments with URG.
2306 if ((thflags & TH_URG) && th->th_urp &&
2307 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2309 * This is a kludge, but if we receive and accept
2310 * random urgent pointers, we'll crash in
2311 * soreceive. It's hard to imagine someone
2312 * actually wanting to send this much urgent data.
2314 SOCKBUF_LOCK(&so->so_rcv);
2315 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2316 th->th_urp = 0; /* XXX */
2317 thflags &= ~TH_URG; /* XXX */
2318 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2319 goto dodata; /* XXX */
2322 * If this segment advances the known urgent pointer,
2323 * then mark the data stream. This should not happen
2324 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2325 * a FIN has been received from the remote side.
2326 * In these states we ignore the URG.
2328 * According to RFC961 (Assigned Protocols),
2329 * the urgent pointer points to the last octet
2330 * of urgent data. We continue, however,
2331 * to consider it to indicate the first octet
2332 * of data past the urgent section as the original
2333 * spec states (in one of two places).
2335 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2336 tp->rcv_up = th->th_seq + th->th_urp;
2337 so->so_oobmark = so->so_rcv.sb_cc +
2338 (tp->rcv_up - tp->rcv_nxt) - 1;
2339 if (so->so_oobmark == 0)
2340 so->so_rcv.sb_state |= SBS_RCVATMARK;
2342 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2344 SOCKBUF_UNLOCK(&so->so_rcv);
2346 * Remove out of band data so doesn't get presented to user.
2347 * This can happen independent of advancing the URG pointer,
2348 * but if two URG's are pending at once, some out-of-band
2349 * data may creep in... ick.
2351 if (th->th_urp <= (u_long)tlen &&
2352 !(so->so_options & SO_OOBINLINE)) {
2353 /* hdr drop is delayed */
2354 tcp_pulloutofband(so, th, m, drop_hdrlen);
2358 * If no out of band data is expected,
2359 * pull receive urgent pointer along
2360 * with the receive window.
2362 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2363 tp->rcv_up = tp->rcv_nxt;
2366 KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2367 INP_LOCK_ASSERT(inp);
2370 * Process the segment text, merging it into the TCP sequencing queue,
2371 * and arranging for acknowledgment of receipt if necessary.
2372 * This process logically involves adjusting tp->rcv_wnd as data
2373 * is presented to the user (this happens in tcp_usrreq.c,
2374 * case PRU_RCVD). If a FIN has already been received on this
2375 * connection then we just ignore the text.
2377 if ((tlen || (thflags & TH_FIN)) &&
2378 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2379 tcp_seq save_start = th->th_seq;
2380 tcp_seq save_end = th->th_seq + tlen;
2381 m_adj(m, drop_hdrlen); /* delayed header drop */
2383 * Insert segment which includes th into TCP reassembly queue
2384 * with control block tp. Set thflags to whether reassembly now
2385 * includes a segment with FIN. This handles the common case
2386 * inline (segment is the next to be received on an established
2387 * connection, and the queue is empty), avoiding linkage into
2388 * and removal from the queue and repetition of various
2390 * Set DELACK for segments received in order, but ack
2391 * immediately when segments are out of order (so
2392 * fast retransmit can work).
2394 if (th->th_seq == tp->rcv_nxt &&
2395 LIST_EMPTY(&tp->t_segq) &&
2396 TCPS_HAVEESTABLISHED(tp->t_state)) {
2398 tp->t_flags |= TF_DELACK;
2400 tp->t_flags |= TF_ACKNOW;
2401 tp->rcv_nxt += tlen;
2402 thflags = th->th_flags & TH_FIN;
2403 tcpstat.tcps_rcvpack++;
2404 tcpstat.tcps_rcvbyte += tlen;
2406 SOCKBUF_LOCK(&so->so_rcv);
2407 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2410 sbappendstream_locked(&so->so_rcv, m);
2411 sorwakeup_locked(so);
2413 thflags = tcp_reass(tp, th, &tlen, m);
2414 tp->t_flags |= TF_ACKNOW;
2416 if (tlen > 0 && tp->sack_enable)
2417 tcp_update_sack_list(tp, save_start, save_end);
2419 * Note the amount of data that peer has sent into
2420 * our window, in order to estimate the sender's
2423 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2430 * If FIN is received ACK the FIN and let the user know
2431 * that the connection is closing.
2433 if (thflags & TH_FIN) {
2434 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2437 * If connection is half-synchronized
2438 * (ie NEEDSYN flag on) then delay ACK,
2439 * so it may be piggybacked when SYN is sent.
2440 * Otherwise, since we received a FIN then no
2441 * more input can be expected, send ACK now.
2443 if (tp->t_flags & TF_NEEDSYN)
2444 tp->t_flags |= TF_DELACK;
2446 tp->t_flags |= TF_ACKNOW;
2449 switch (tp->t_state) {
2452 * In SYN_RECEIVED and ESTABLISHED STATES
2453 * enter the CLOSE_WAIT state.
2455 case TCPS_SYN_RECEIVED:
2456 tp->t_starttime = ticks;
2458 case TCPS_ESTABLISHED:
2459 tp->t_state = TCPS_CLOSE_WAIT;
2463 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2464 * enter the CLOSING state.
2466 case TCPS_FIN_WAIT_1:
2467 tp->t_state = TCPS_CLOSING;
2471 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2472 * starting the time-wait timer, turning off the other
2475 case TCPS_FIN_WAIT_2:
2476 KASSERT(headlocked == 1, ("tcp_input: dodata: "
2477 "TCP_FIN_WAIT_2: head not locked"));
2479 INP_INFO_WUNLOCK(&tcbinfo);
2483 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2485 case TCPS_TIME_WAIT:
2486 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2487 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2488 tcp_timer_2msl, tp);
2492 INP_INFO_WUNLOCK(&tcbinfo);
2495 if (so->so_options & SO_DEBUG)
2496 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2501 * Return any desired output.
2503 if (needoutput || (tp->t_flags & TF_ACKNOW))
2504 (void) tcp_output(tp);
2507 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2508 INP_LOCK_ASSERT(inp);
2509 if (tp->t_flags & TF_DELACK) {
2510 tp->t_flags &= ~TF_DELACK;
2511 callout_reset(tp->tt_delack, tcp_delacktime,
2512 tcp_timer_delack, tp);
2518 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2520 * Generate an ACK dropping incoming segment if it occupies
2521 * sequence space, where the ACK reflects our state.
2523 * We can now skip the test for the RST flag since all
2524 * paths to this code happen after packets containing
2525 * RST have been dropped.
2527 * In the SYN-RECEIVED state, don't send an ACK unless the
2528 * segment we received passes the SYN-RECEIVED ACK test.
2529 * If it fails send a RST. This breaks the loop in the
2530 * "LAND" DoS attack, and also prevents an ACK storm
2531 * between two listening ports that have been sent forged
2532 * SYN segments, each with the source address of the other.
2534 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2535 (SEQ_GT(tp->snd_una, th->th_ack) ||
2536 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2537 rstreason = BANDLIM_RST_OPENPORT;
2541 if (so->so_options & SO_DEBUG)
2542 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2545 KASSERT(headlocked, ("headlocked should be 1"));
2546 INP_INFO_WUNLOCK(&tcbinfo);
2547 tp->t_flags |= TF_ACKNOW;
2548 (void) tcp_output(tp);
2554 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2556 * Generate a RST, dropping incoming segment.
2557 * Make ACK acceptable to originator of segment.
2558 * Don't bother to respond if destination was broadcast/multicast.
2560 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2563 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2564 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2567 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2568 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2569 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2570 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2573 /* IPv6 anycast check is done at tcp6_input() */
2576 * Perform bandwidth limiting.
2578 if (badport_bandlim(rstreason) < 0)
2582 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2583 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2587 if (thflags & TH_ACK)
2588 /* mtod() below is safe as long as hdr dropping is delayed */
2589 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2592 if (thflags & TH_SYN)
2594 /* mtod() below is safe as long as hdr dropping is delayed */
2595 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2596 (tcp_seq)0, TH_RST|TH_ACK);
2602 INP_INFO_WUNLOCK(&tcbinfo);
2607 * Drop space held by incoming segment and return.
2610 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2611 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2617 INP_INFO_WUNLOCK(&tcbinfo);
2623 * Parse TCP options and place in tcpopt.
2626 tcp_dooptions(to, cp, cnt, flags)
2635 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2637 if (opt == TCPOPT_EOL)
2639 if (opt == TCPOPT_NOP)
2645 if (optlen < 2 || optlen > cnt)
2650 if (optlen != TCPOLEN_MAXSEG)
2652 if (!(flags & TO_SYN))
2654 to->to_flags |= TOF_MSS;
2655 bcopy((char *)cp + 2,
2656 (char *)&to->to_mss, sizeof(to->to_mss));
2657 to->to_mss = ntohs(to->to_mss);
2660 if (optlen != TCPOLEN_WINDOW)
2662 if (!(flags & TO_SYN))
2664 to->to_flags |= TOF_SCALE;
2665 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2667 case TCPOPT_TIMESTAMP:
2668 if (optlen != TCPOLEN_TIMESTAMP)
2670 to->to_flags |= TOF_TS;
2671 bcopy((char *)cp + 2,
2672 (char *)&to->to_tsval, sizeof(to->to_tsval));
2673 to->to_tsval = ntohl(to->to_tsval);
2674 bcopy((char *)cp + 6,
2675 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2676 to->to_tsecr = ntohl(to->to_tsecr);
2678 #ifdef TCP_SIGNATURE
2680 * XXX In order to reply to a host which has set the
2681 * TCP_SIGNATURE option in its initial SYN, we have to
2682 * record the fact that the option was observed here
2683 * for the syncache code to perform the correct response.
2685 case TCPOPT_SIGNATURE:
2686 if (optlen != TCPOLEN_SIGNATURE)
2688 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2691 case TCPOPT_SACK_PERMITTED:
2692 if (optlen != TCPOLEN_SACK_PERMITTED)
2694 if (!(flags & TO_SYN))
2698 to->to_flags |= TOF_SACK;
2701 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2703 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2704 to->to_sacks = cp + 2;
2705 tcpstat.tcps_sack_rcv_blocks++;
2714 * Pull out of band byte out of a segment so
2715 * it doesn't appear in the user's data queue.
2716 * It is still reflected in the segment length for
2717 * sequencing purposes.
2720 tcp_pulloutofband(so, th, m, off)
2723 register struct mbuf *m;
2724 int off; /* delayed to be droped hdrlen */
2726 int cnt = off + th->th_urp - 1;
2729 if (m->m_len > cnt) {
2730 char *cp = mtod(m, caddr_t) + cnt;
2731 struct tcpcb *tp = sototcpcb(so);
2734 tp->t_oobflags |= TCPOOB_HAVEDATA;
2735 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2737 if (m->m_flags & M_PKTHDR)
2746 panic("tcp_pulloutofband");
2750 * Collect new round-trip time estimate
2751 * and update averages and current timeout.
2754 tcp_xmit_timer(tp, rtt)
2755 register struct tcpcb *tp;
2760 INP_LOCK_ASSERT(tp->t_inpcb);
2762 tcpstat.tcps_rttupdated++;
2764 if (tp->t_srtt != 0) {
2766 * srtt is stored as fixed point with 5 bits after the
2767 * binary point (i.e., scaled by 8). The following magic
2768 * is equivalent to the smoothing algorithm in rfc793 with
2769 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2770 * point). Adjust rtt to origin 0.
2772 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2773 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2775 if ((tp->t_srtt += delta) <= 0)
2779 * We accumulate a smoothed rtt variance (actually, a
2780 * smoothed mean difference), then set the retransmit
2781 * timer to smoothed rtt + 4 times the smoothed variance.
2782 * rttvar is stored as fixed point with 4 bits after the
2783 * binary point (scaled by 16). The following is
2784 * equivalent to rfc793 smoothing with an alpha of .75
2785 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2786 * rfc793's wired-in beta.
2790 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2791 if ((tp->t_rttvar += delta) <= 0)
2793 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2794 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2797 * No rtt measurement yet - use the unsmoothed rtt.
2798 * Set the variance to half the rtt (so our first
2799 * retransmit happens at 3*rtt).
2801 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2802 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2803 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2809 * the retransmit should happen at rtt + 4 * rttvar.
2810 * Because of the way we do the smoothing, srtt and rttvar
2811 * will each average +1/2 tick of bias. When we compute
2812 * the retransmit timer, we want 1/2 tick of rounding and
2813 * 1 extra tick because of +-1/2 tick uncertainty in the
2814 * firing of the timer. The bias will give us exactly the
2815 * 1.5 tick we need. But, because the bias is
2816 * statistical, we have to test that we don't drop below
2817 * the minimum feasible timer (which is 2 ticks).
2819 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2820 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2823 * We received an ack for a packet that wasn't retransmitted;
2824 * it is probably safe to discard any error indications we've
2825 * received recently. This isn't quite right, but close enough
2826 * for now (a route might have failed after we sent a segment,
2827 * and the return path might not be symmetrical).
2829 tp->t_softerror = 0;
2833 * Determine a reasonable value for maxseg size.
2834 * If the route is known, check route for mtu.
2835 * If none, use an mss that can be handled on the outgoing
2836 * interface without forcing IP to fragment; if bigger than
2837 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2838 * to utilize large mbufs. If no route is found, route has no mtu,
2839 * or the destination isn't local, use a default, hopefully conservative
2840 * size (usually 512 or the default IP max size, but no more than the mtu
2841 * of the interface), as we can't discover anything about intervening
2842 * gateways or networks. We also initialize the congestion/slow start
2843 * window to be a single segment if the destination isn't local.
2844 * While looking at the routing entry, we also initialize other path-dependent
2845 * parameters from pre-set or cached values in the routing entry.
2847 * Also take into account the space needed for options that we
2848 * send regularly. Make maxseg shorter by that amount to assure
2849 * that we can send maxseg amount of data even when the options
2850 * are present. Store the upper limit of the length of options plus
2854 * In case of T/TCP, we call this routine during implicit connection
2855 * setup as well (offer = -1), to initialize maxseg from the cached
2858 * NOTE that this routine is only called when we process an incoming
2859 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2869 struct inpcb *inp = tp->t_inpcb;
2871 struct hc_metrics_lite metrics;
2872 int origoffer = offer;
2875 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2876 size_t min_protoh = isipv6 ?
2877 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2878 sizeof (struct tcpiphdr);
2880 const size_t min_protoh = sizeof(struct tcpiphdr);
2886 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags);
2887 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2891 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags);
2892 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2894 so = inp->inp_socket;
2897 * no route to sender, stay with default mss and return
2902 /* what have we got? */
2906 * Offer == 0 means that there was no MSS on the SYN
2907 * segment, in this case we use tcp_mssdflt.
2911 isipv6 ? tcp_v6mssdflt :
2918 * Offer == -1 means that we didn't receive SYN yet.
2924 * Prevent DoS attack with too small MSS. Round up
2925 * to at least minmss.
2927 offer = max(offer, tcp_minmss);
2929 * Sanity check: make sure that maxopd will be large
2930 * enough to allow some data on segments even if the
2931 * all the option space is used (40bytes). Otherwise
2932 * funny things may happen in tcp_output.
2934 offer = max(offer, 64);
2938 * rmx information is now retrieved from tcp_hostcache
2940 tcp_hc_get(&inp->inp_inc, &metrics);
2943 * if there's a discovered mtu int tcp hostcache, use it
2944 * else, use the link mtu.
2946 if (metrics.rmx_mtu)
2947 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2951 mss = maxmtu - min_protoh;
2952 if (!path_mtu_discovery &&
2953 !in6_localaddr(&inp->in6p_faddr))
2954 mss = min(mss, tcp_v6mssdflt);
2958 mss = maxmtu - min_protoh;
2959 if (!path_mtu_discovery &&
2960 !in_localaddr(inp->inp_faddr))
2961 mss = min(mss, tcp_mssdflt);
2964 mss = min(mss, offer);
2967 * maxopd stores the maximum length of data AND options
2968 * in a segment; maxseg is the amount of data in a normal
2969 * segment. We need to store this value (maxopd) apart
2970 * from maxseg, because now every segment carries options
2971 * and thus we normally have somewhat less data in segments.
2976 * origoffer==-1 indicates, that no segments were received yet.
2977 * In this case we just guess.
2979 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2981 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2982 mss -= TCPOLEN_TSTAMP_APPA;
2985 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2987 mss &= ~(MCLBYTES-1);
2990 mss = mss / MCLBYTES * MCLBYTES;
2995 * If there's a pipesize, change the socket buffer to that size,
2996 * don't change if sb_hiwat is different than default (then it
2997 * has been changed on purpose with setsockopt).
2998 * Make the socket buffers an integral number of mss units;
2999 * if the mss is larger than the socket buffer, decrease the mss.
3001 SOCKBUF_LOCK(&so->so_snd);
3002 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3003 bufsize = metrics.rmx_sendpipe;
3005 bufsize = so->so_snd.sb_hiwat;
3009 bufsize = roundup(bufsize, mss);
3010 if (bufsize > sb_max)
3012 if (bufsize > so->so_snd.sb_hiwat)
3013 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3015 SOCKBUF_UNLOCK(&so->so_snd);
3018 SOCKBUF_LOCK(&so->so_rcv);
3019 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3020 bufsize = metrics.rmx_recvpipe;
3022 bufsize = so->so_rcv.sb_hiwat;
3023 if (bufsize > mss) {
3024 bufsize = roundup(bufsize, mss);
3025 if (bufsize > sb_max)
3027 if (bufsize > so->so_rcv.sb_hiwat)
3028 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3030 SOCKBUF_UNLOCK(&so->so_rcv);
3032 * While we're here, check the others too
3034 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3036 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3037 tcpstat.tcps_usedrtt++;
3038 if (metrics.rmx_rttvar) {
3039 tp->t_rttvar = metrics.rmx_rttvar;
3040 tcpstat.tcps_usedrttvar++;
3042 /* default variation is +- 1 rtt */
3044 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3046 TCPT_RANGESET(tp->t_rxtcur,
3047 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3048 tp->t_rttmin, TCPTV_REXMTMAX);
3050 if (metrics.rmx_ssthresh) {
3052 * There's some sort of gateway or interface
3053 * buffer limit on the path. Use this to set
3054 * the slow start threshhold, but set the
3055 * threshold to no less than 2*mss.
3057 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3058 tcpstat.tcps_usedssthresh++;
3060 if (metrics.rmx_bandwidth)
3061 tp->snd_bandwidth = metrics.rmx_bandwidth;
3064 * Set the slow-start flight size depending on whether this
3065 * is a local network or not.
3067 * Extend this so we cache the cwnd too and retrieve it here.
3068 * Make cwnd even bigger than RFC3390 suggests but only if we
3069 * have previous experience with the remote host. Be careful
3070 * not make cwnd bigger than remote receive window or our own
3071 * send socket buffer. Maybe put some additional upper bound
3072 * on the retrieved cwnd. Should do incremental updates to
3073 * hostcache when cwnd collapses so next connection doesn't
3074 * overloads the path again.
3076 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3077 * We currently check only in syncache_socket for that.
3079 #define TCP_METRICS_CWND
3080 #ifdef TCP_METRICS_CWND
3081 if (metrics.rmx_cwnd)
3082 tp->snd_cwnd = max(mss,
3083 min(metrics.rmx_cwnd / 2,
3084 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3088 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3090 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3091 (!isipv6 && in_localaddr(inp->inp_faddr)))
3093 else if (in_localaddr(inp->inp_faddr))
3095 tp->snd_cwnd = mss * ss_fltsz_local;
3097 tp->snd_cwnd = mss * ss_fltsz;
3099 /* Check the interface for TSO capabilities. */
3100 if (mtuflags & CSUM_TSO)
3101 tp->t_flags |= TF_TSO;
3105 * Determine the MSS option to send on an outgoing SYN.
3109 struct in_conninfo *inc;
3116 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3119 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3123 mss = tcp_v6mssdflt;
3124 maxmtu = tcp_maxmtu6(inc, NULL);
3125 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3126 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3131 maxmtu = tcp_maxmtu(inc, NULL);
3132 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3133 min_protoh = sizeof(struct tcpiphdr);
3135 if (maxmtu && thcmtu)
3136 mss = min(maxmtu, thcmtu) - min_protoh;
3137 else if (maxmtu || thcmtu)
3138 mss = max(maxmtu, thcmtu) - min_protoh;
3145 * On a partial ack arrives, force the retransmission of the
3146 * next unacknowledged segment. Do not clear tp->t_dupacks.
3147 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3151 tcp_newreno_partial_ack(tp, th)
3155 tcp_seq onxt = tp->snd_nxt;
3156 u_long ocwnd = tp->snd_cwnd;
3158 callout_stop(tp->tt_rexmt);
3160 tp->snd_nxt = th->th_ack;
3162 * Set snd_cwnd to one segment beyond acknowledged offset.
3163 * (tp->snd_una has not yet been updated when this function is called.)
3165 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3166 tp->t_flags |= TF_ACKNOW;
3167 (void) tcp_output(tp);
3168 tp->snd_cwnd = ocwnd;
3169 if (SEQ_GT(onxt, tp->snd_nxt))
3172 * Partial window deflation. Relies on fact that tp->snd_una
3175 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3176 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3179 tp->snd_cwnd += tp->t_maxseg;
3183 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3184 * looking for a pcb in the listen state. Returns 0 otherwise.
3187 tcp_timewait(inp, to, th, m, tlen)
3198 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3200 const int isipv6 = 0;
3203 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */
3204 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3205 INP_LOCK_ASSERT(inp);
3208 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
3209 * still present. This is undesirable, but temporarily necessary
3210 * until we work out how to handle inpcb's who's timewait state has
3217 thflags = th->th_flags;
3220 * NOTE: for FIN_WAIT_2 (to be added later),
3221 * must validate sequence number before accepting RST
3225 * If the segment contains RST:
3226 * Drop the segment - see Stevens, vol. 2, p. 964 and
3229 if (thflags & TH_RST)
3233 /* PAWS not needed at the moment */
3235 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3236 * and it's less than ts_recent, drop it.
3238 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3239 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3240 if ((thflags & TH_ACK) == 0)
3245 * ts_recent is never updated because we never accept new segments.
3250 * If a new connection request is received
3251 * while in TIME_WAIT, drop the old connection
3252 * and start over if the sequence numbers
3253 * are above the previous ones.
3255 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3261 * Drop the the segment if it does not contain an ACK.
3263 if ((thflags & TH_ACK) == 0)
3267 * Reset the 2MSL timer if this is a duplicate FIN.
3269 if (thflags & TH_FIN) {
3270 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3271 if (seq + 1 == tw->rcv_nxt)
3272 tcp_timer_2msl_reset(tw, 1);
3276 * Acknowledge the segment if it has data or is not a duplicate ACK.
3278 if (thflags != TH_ACK || tlen != 0 ||
3279 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3280 tcp_twrespond(tw, TH_ACK);
3284 * Generate a RST, dropping incoming segment.
3285 * Make ACK acceptable to originator of segment.
3286 * Don't bother to respond if destination was broadcast/multicast.
3288 if (m->m_flags & (M_BCAST|M_MCAST))
3291 struct ip6_hdr *ip6;
3293 /* IPv6 anycast check is done at tcp6_input() */
3294 ip6 = mtod(m, struct ip6_hdr *);
3295 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3296 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3301 ip = mtod(m, struct ip *);
3302 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3303 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3304 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3305 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3308 if (thflags & TH_ACK) {
3310 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3312 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3314 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);