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_sack.h"
41 #include <sys/param.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
45 #include <sys/proc.h> /* for proc0 declaration */
46 #include <sys/protosw.h>
47 #include <sys/signalvar.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/systm.h>
54 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
59 #include <net/route.h>
61 #include <netinet/in.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
67 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_options.h>
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/in6_pcb.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_fsm.h>
77 #include <netinet/tcp_seq.h>
78 #include <netinet/tcp_timer.h>
79 #include <netinet/tcp_var.h>
80 #include <netinet6/tcp6_var.h>
81 #include <netinet/tcpip.h>
83 #include <netinet/tcp_debug.h>
87 #include <netipsec/ipsec.h>
88 #include <netipsec/ipsec6.h>
92 #include <netinet6/ipsec.h>
93 #include <netinet6/ipsec6.h>
94 #include <netkey/key.h>
97 #include <machine/in_cksum.h>
99 #include <security/mac/mac_framework.h>
101 static const int tcprexmtthresh = 3;
103 struct tcpstat tcpstat;
104 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
105 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
107 static int tcp_log_in_vain = 0;
108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
109 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports");
111 static int blackhole = 0;
112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
113 &blackhole, 0, "Do not send RST on segments to closed ports");
115 int tcp_delack_enabled = 1;
116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
117 &tcp_delack_enabled, 0,
118 "Delay ACK to try and piggyback it onto a data packet");
120 static int drop_synfin = 0;
121 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
124 static int tcp_do_rfc3042 = 1;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
126 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
128 static int tcp_do_rfc3390 = 1;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
131 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
133 static int tcp_insecure_rst = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
135 &tcp_insecure_rst, 0,
136 "Follow the old (insecure) criteria for accepting RST packets");
138 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
139 "TCP Segment Reassembly Queue");
141 static int tcp_reass_maxseg = 0;
142 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
143 &tcp_reass_maxseg, 0,
144 "Global maximum number of TCP Segments in Reassembly Queue");
146 int tcp_reass_qsize = 0;
147 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
149 "Global number of TCP Segments currently in Reassembly Queue");
151 static int tcp_reass_maxqlen = 48;
152 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
153 &tcp_reass_maxqlen, 0,
154 "Maximum number of TCP Segments per individual Reassembly Queue");
156 static int tcp_reass_overflows = 0;
157 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
158 &tcp_reass_overflows, 0,
159 "Global number of TCP Segment Reassembly Queue Overflows");
161 int tcp_do_autorcvbuf = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
163 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
165 int tcp_autorcvbuf_inc = 16*1024;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
167 &tcp_autorcvbuf_inc, 0,
168 "Incrementor step size of automatic receive buffer");
170 int tcp_autorcvbuf_max = 256*1024;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
172 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
174 struct inpcbhead tcb;
175 #define tcb6 tcb /* for KAME src sync over BSD*'s */
176 struct inpcbinfo tcbinfo;
177 struct mtx *tcbinfo_mtx;
179 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
181 static void tcp_pulloutofband(struct socket *,
182 struct tcphdr *, struct mbuf *, int);
183 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
185 static void tcp_xmit_timer(struct tcpcb *, int);
186 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
187 static int tcp_timewait(struct inpcb *, struct tcpopt *,
188 struct tcphdr *, struct mbuf *, int);
190 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
192 #define ND6_HINT(tp) \
194 if ((tp) && (tp)->t_inpcb && \
195 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
196 nd6_nud_hint(NULL, NULL, 0); \
203 * Indicate whether this ack should be delayed. We can delay the ack if
204 * - there is no delayed ack timer in progress and
205 * - our last ack wasn't a 0-sized window. We never want to delay
206 * the ack that opens up a 0-sized window and
207 * - delayed acks are enabled or
208 * - this is a half-synchronized T/TCP connection.
210 #define DELAY_ACK(tp) \
211 ((!callout_active(tp->tt_delack) && \
212 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
213 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
215 /* Initialize TCP reassembly queue */
217 tcp_reass_zone_change(void *tag)
220 tcp_reass_maxseg = nmbclusters / 16;
221 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
224 uma_zone_t tcp_reass_zone;
228 tcp_reass_maxseg = nmbclusters / 16;
229 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
231 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
232 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
233 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
234 EVENTHANDLER_REGISTER(nmbclusters_change,
235 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
239 tcp_reass(tp, th, tlenp, m)
240 register struct tcpcb *tp;
241 register struct tcphdr *th;
246 struct tseg_qent *p = NULL;
247 struct tseg_qent *nq;
248 struct tseg_qent *te = NULL;
249 struct socket *so = tp->t_inpcb->inp_socket;
252 INP_LOCK_ASSERT(tp->t_inpcb);
255 * XXX: tcp_reass() is rather inefficient with its data structures
256 * and should be rewritten (see NetBSD for optimizations). While
257 * doing that it should move to its own file tcp_reass.c.
261 * Call with th==NULL after become established to
262 * force pre-ESTABLISHED data up to user socket.
268 * Limit the number of segments in the reassembly queue to prevent
269 * holding on to too many segments (and thus running out of mbufs).
270 * Make sure to let the missing segment through which caused this
271 * queue. Always keep one global queue entry spare to be able to
272 * process the missing segment.
274 if (th->th_seq != tp->rcv_nxt &&
275 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
276 tp->t_segqlen >= tcp_reass_maxqlen)) {
277 tcp_reass_overflows++;
278 tcpstat.tcps_rcvmemdrop++;
285 * Allocate a new queue entry. If we can't, or hit the zone limit
288 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
290 tcpstat.tcps_rcvmemdrop++;
299 * Find a segment which begins after this one does.
301 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
302 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
308 * If there is a preceding segment, it may provide some of
309 * our data already. If so, drop the data from the incoming
310 * segment. If it provides all of our data, drop us.
314 /* conversion to int (in i) handles seq wraparound */
315 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
318 tcpstat.tcps_rcvduppack++;
319 tcpstat.tcps_rcvdupbyte += *tlenp;
321 uma_zfree(tcp_reass_zone, te);
325 * Try to present any queued data
326 * at the left window edge to the user.
327 * This is needed after the 3-WHS
330 goto present; /* ??? */
337 tcpstat.tcps_rcvoopack++;
338 tcpstat.tcps_rcvoobyte += *tlenp;
341 * While we overlap succeeding segments trim them or,
342 * if they are completely covered, dequeue them.
345 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
348 if (i < q->tqe_len) {
349 q->tqe_th->th_seq += i;
355 nq = LIST_NEXT(q, tqe_q);
356 LIST_REMOVE(q, tqe_q);
358 uma_zfree(tcp_reass_zone, q);
364 /* Insert the new segment queue entry into place. */
367 te->tqe_len = *tlenp;
370 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
372 LIST_INSERT_AFTER(p, te, tqe_q);
377 * Present data to user, advancing rcv_nxt through
378 * completed sequence space.
380 if (!TCPS_HAVEESTABLISHED(tp->t_state))
382 q = LIST_FIRST(&tp->t_segq);
383 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
385 SOCKBUF_LOCK(&so->so_rcv);
387 tp->rcv_nxt += q->tqe_len;
388 flags = q->tqe_th->th_flags & TH_FIN;
389 nq = LIST_NEXT(q, tqe_q);
390 LIST_REMOVE(q, tqe_q);
391 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
394 sbappendstream_locked(&so->so_rcv, q->tqe_m);
395 uma_zfree(tcp_reass_zone, q);
399 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
401 sorwakeup_locked(so);
406 * TCP input routine, follows pages 65-76 of the
407 * protocol specification dated September, 1981 very closely.
411 tcp6_input(mp, offp, proto)
415 register struct mbuf *m = *mp;
416 struct in6_ifaddr *ia6;
418 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
421 * draft-itojun-ipv6-tcp-to-anycast
422 * better place to put this in?
424 ia6 = ip6_getdstifaddr(m);
425 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
428 ip6 = mtod(m, struct ip6_hdr *);
429 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
430 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
441 register struct mbuf *m;
444 register struct tcphdr *th;
445 register struct ip *ip = NULL;
446 register struct ipovly *ipov;
447 register struct inpcb *inp = NULL;
452 register struct tcpcb *tp = 0;
453 register int thflags;
454 struct socket *so = 0;
455 int todrop, acked, ourfinisacked, needoutput = 0;
457 struct tcpopt to; /* options in this segment */
459 #ifdef IPFIREWALL_FORWARD
460 struct m_tag *fwd_tag;
462 int rstreason; /* For badport_bandlim accounting purposes */
464 struct ip6_hdr *ip6 = NULL;
467 char ip6buf[INET6_ADDRSTRLEN];
469 const int isipv6 = 0;
474 * The size of tcp_saveipgen must be the size of the max ip header,
477 u_char tcp_saveipgen[40];
478 struct tcphdr tcp_savetcp;
483 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
485 bzero((char *)&to, sizeof(to));
487 tcpstat.tcps_rcvtotal++;
491 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
492 ip6 = mtod(m, struct ip6_hdr *);
493 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
494 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
495 tcpstat.tcps_rcvbadsum++;
498 th = (struct tcphdr *)((caddr_t)ip6 + off0);
501 * Be proactive about unspecified IPv6 address in source.
502 * As we use all-zero to indicate unbounded/unconnected pcb,
503 * unspecified IPv6 address can be used to confuse us.
505 * Note that packets with unspecified IPv6 destination is
506 * already dropped in ip6_input.
508 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
513 th = NULL; /* XXX: avoid compiler warning */
517 * Get IP and TCP header together in first mbuf.
518 * Note: IP leaves IP header in first mbuf.
520 if (off0 > sizeof (struct ip)) {
521 ip_stripoptions(m, (struct mbuf *)0);
522 off0 = sizeof(struct ip);
524 if (m->m_len < sizeof (struct tcpiphdr)) {
525 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
526 tcpstat.tcps_rcvshort++;
530 ip = mtod(m, struct ip *);
531 ipov = (struct ipovly *)ip;
532 th = (struct tcphdr *)((caddr_t)ip + off0);
535 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
536 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
537 th->th_sum = m->m_pkthdr.csum_data;
539 th->th_sum = in_pseudo(ip->ip_src.s_addr,
541 htonl(m->m_pkthdr.csum_data +
544 th->th_sum ^= 0xffff;
546 ipov->ih_len = (u_short)tlen;
547 ipov->ih_len = htons(ipov->ih_len);
551 * Checksum extended TCP header and data.
553 len = sizeof (struct ip) + tlen;
554 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
555 ipov->ih_len = (u_short)tlen;
556 ipov->ih_len = htons(ipov->ih_len);
557 th->th_sum = in_cksum(m, len);
560 tcpstat.tcps_rcvbadsum++;
563 /* Re-initialization for later version check */
564 ip->ip_v = IPVERSION;
568 * Check that TCP offset makes sense,
569 * pull out TCP options and adjust length. XXX
571 off = th->th_off << 2;
572 if (off < sizeof (struct tcphdr) || off > tlen) {
573 tcpstat.tcps_rcvbadoff++;
576 tlen -= off; /* tlen is used instead of ti->ti_len */
577 if (off > sizeof (struct tcphdr)) {
580 IP6_EXTHDR_CHECK(m, off0, off, );
581 ip6 = mtod(m, struct ip6_hdr *);
582 th = (struct tcphdr *)((caddr_t)ip6 + off0);
585 if (m->m_len < sizeof(struct ip) + off) {
586 if ((m = m_pullup(m, sizeof (struct ip) + off))
588 tcpstat.tcps_rcvshort++;
591 ip = mtod(m, struct ip *);
592 ipov = (struct ipovly *)ip;
593 th = (struct tcphdr *)((caddr_t)ip + off0);
596 optlen = off - sizeof (struct tcphdr);
597 optp = (u_char *)(th + 1);
599 thflags = th->th_flags;
602 * If the drop_synfin option is enabled, drop all packets with
603 * both the SYN and FIN bits set. This prevents e.g. nmap from
604 * identifying the TCP/IP stack.
606 * This is a violation of the TCP specification.
608 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
612 * Convert TCP protocol specific fields to host format.
614 th->th_seq = ntohl(th->th_seq);
615 th->th_ack = ntohl(th->th_ack);
616 th->th_win = ntohs(th->th_win);
617 th->th_urp = ntohs(th->th_urp);
620 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
621 * until after ip6_savecontrol() is called and before other functions
622 * which don't want those proto headers.
623 * Because ip6_savecontrol() is going to parse the mbuf to
624 * search for data to be passed up to user-land, it wants mbuf
625 * parameters to be unchanged.
626 * XXX: the call of ip6_savecontrol() has been obsoleted based on
627 * latest version of the advanced API (20020110).
629 drop_hdrlen = off0 + off;
632 * Locate pcb for segment.
634 INP_INFO_WLOCK(&tcbinfo);
637 KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
638 #ifdef IPFIREWALL_FORWARD
639 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
640 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
642 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
643 struct sockaddr_in *next_hop;
645 next_hop = (struct sockaddr_in *)(fwd_tag+1);
647 * Transparently forwarded. Pretend to be the destination.
648 * already got one like this?
650 inp = in_pcblookup_hash(&tcbinfo,
651 ip->ip_src, th->th_sport,
652 ip->ip_dst, th->th_dport,
653 0, m->m_pkthdr.rcvif);
655 /* It's new. Try to find the ambushing socket. */
656 inp = in_pcblookup_hash(&tcbinfo,
657 ip->ip_src, th->th_sport,
660 ntohs(next_hop->sin_port) :
665 /* Remove the tag from the packet. We don't need it anymore. */
666 m_tag_delete(m, fwd_tag);
668 #endif /* IPFIREWALL_FORWARD */
671 inp = in6_pcblookup_hash(&tcbinfo,
672 &ip6->ip6_src, th->th_sport,
673 &ip6->ip6_dst, th->th_dport,
678 inp = in_pcblookup_hash(&tcbinfo,
679 ip->ip_src, th->th_sport,
680 ip->ip_dst, th->th_dport,
683 #ifdef IPFIREWALL_FORWARD
685 #endif /* IPFIREWALL_FORWARD */
687 #if defined(IPSEC) || defined(FAST_IPSEC)
690 if (inp != NULL && ipsec6_in_reject(m, inp)) {
692 ipsec6stat.in_polvio++;
698 if (inp != NULL && ipsec4_in_reject(m, inp)) {
700 ipsecstat.in_polvio++;
704 #endif /*IPSEC || FAST_IPSEC*/
707 * If the INPCB does not exist then all data in the incoming
708 * segment is discarded and an appropriate RST is sent back.
712 * Log communication attempts to ports that are not
715 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
716 tcp_log_in_vain == 2) {
718 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
720 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
724 ip6_sprintf(ip6buf, &ip6->ip6_dst));
728 ip6_sprintf(ip6buf, &ip6->ip6_src));
733 strcpy(dbuf, inet_ntoa(ip->ip_dst));
734 strcpy(sbuf, inet_ntoa(ip->ip_src));
737 "Connection attempt to TCP %s:%d "
738 "from %s:%d flags:0x%02x\n",
739 dbuf, ntohs(th->th_dport), sbuf,
740 ntohs(th->th_sport), thflags);
743 * When blackholing do not respond with a RST but
744 * completely ignore the segment and drop it.
746 if ((blackhole == 1 && (thflags & TH_SYN)) ||
750 rstreason = BANDLIM_RST_CLOSEDPORT;
755 /* Check the minimum TTL for socket. */
756 if (inp->inp_ip_minttl != 0) {
758 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
762 if (inp->inp_ip_minttl > ip->ip_ttl)
766 if (inp->inp_vflag & INP_TIMEWAIT) {
768 * The only option of relevance is TOF_CC, and only if
769 * present in a SYN segment. See tcp_timewait().
771 if (thflags & TH_SYN)
772 tcp_dooptions(&to, optp, optlen, TO_SYN);
773 if (tcp_timewait(inp, &to, th, m, tlen))
776 * tcp_timewait unlocks inp.
778 INP_INFO_WUNLOCK(&tcbinfo);
784 rstreason = BANDLIM_RST_CLOSEDPORT;
787 if (tp->t_state == TCPS_CLOSED)
791 INP_LOCK_ASSERT(inp);
792 if (mac_check_inpcb_deliver(inp, m))
795 so = inp->inp_socket;
796 KASSERT(so != NULL, ("tcp_input: so == NULL"));
798 if (so->so_options & SO_DEBUG) {
799 ostate = tp->t_state;
801 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
803 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
807 if (so->so_options & SO_ACCEPTCONN) {
808 struct in_conninfo inc;
810 bzero(&inc, sizeof(inc));
812 inc.inc_isipv6 = isipv6;
815 inc.inc6_faddr = ip6->ip6_src;
816 inc.inc6_laddr = ip6->ip6_dst;
818 inc.inc_faddr = ip->ip_src;
819 inc.inc_laddr = ip->ip_dst;
821 inc.inc_fport = th->th_sport;
822 inc.inc_lport = th->th_dport;
825 * If the state is LISTEN then ignore segment if it contains
826 * a RST. If the segment contains an ACK then it is bad and
827 * send a RST. If it does not contain a SYN then it is not
828 * interesting; drop it.
830 * If the state is SYN_RECEIVED (syncache) and seg contains
831 * an ACK, but not for our SYN/ACK, send a RST. If the seg
832 * contains a RST, check the sequence number to see if it
833 * is a valid reset segment.
835 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
836 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
838 * Parse the TCP options here because
839 * syncookies need access to the reflected
842 tcp_dooptions(&to, optp, optlen, 0);
843 if (!syncache_expand(&inc, &to, th, &so, m)) {
845 * No syncache entry, or ACK was not
846 * for our SYN/ACK. Send a RST.
848 tcpstat.tcps_badsyn++;
849 rstreason = BANDLIM_RST_OPENPORT;
854 * Could not complete 3-way handshake,
855 * connection is being closed down, and
856 * syncache has free'd mbuf.
859 INP_INFO_WUNLOCK(&tcbinfo);
863 * Socket is created in state SYN_RECEIVED.
864 * Continue processing segment.
871 * This is what would have happened in
872 * tcp_output() when the SYN,ACK was sent.
874 tp->snd_up = tp->snd_una;
875 tp->snd_max = tp->snd_nxt = tp->iss + 1;
876 tp->last_ack_sent = tp->rcv_nxt;
879 if (thflags & TH_RST) {
880 syncache_chkrst(&inc, th);
883 if (thflags & TH_ACK) {
884 syncache_badack(&inc);
885 tcpstat.tcps_badsyn++;
886 rstreason = BANDLIM_RST_OPENPORT;
893 * Segment's flags are (SYN) or (SYN|FIN).
897 * If deprecated address is forbidden,
898 * we do not accept SYN to deprecated interface
899 * address to prevent any new inbound connection from
900 * getting established.
901 * When we do not accept SYN, we send a TCP RST,
902 * with deprecated source address (instead of dropping
903 * it). We compromise it as it is much better for peer
904 * to send a RST, and RST will be the final packet
907 * If we do not forbid deprecated addresses, we accept
908 * the SYN packet. RFC2462 does not suggest dropping
910 * If we decipher RFC2462 5.5.4, it says like this:
911 * 1. use of deprecated addr with existing
912 * communication is okay - "SHOULD continue to be
914 * 2. use of it with new communication:
915 * (2a) "SHOULD NOT be used if alternate address
916 * with sufficient scope is available"
917 * (2b) nothing mentioned otherwise.
918 * Here we fall into (2b) case as we have no choice in
919 * our source address selection - we must obey the peer.
921 * The wording in RFC2462 is confusing, and there are
922 * multiple description text for deprecated address
923 * handling - worse, they are not exactly the same.
924 * I believe 5.5.4 is the best one, so we follow 5.5.4.
926 if (isipv6 && !ip6_use_deprecated) {
927 struct in6_ifaddr *ia6;
929 if ((ia6 = ip6_getdstifaddr(m)) &&
930 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
933 rstreason = BANDLIM_RST_OPENPORT;
939 * If it is from this socket, drop it, it must be forged.
940 * Don't bother responding if the destination was a broadcast.
942 if (th->th_dport == th->th_sport) {
944 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
948 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
953 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
955 * Note that it is quite possible to receive unicast
956 * link-layer packets with a broadcast IP address. Use
957 * in_broadcast() to find them.
959 if (m->m_flags & (M_BCAST|M_MCAST))
962 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
963 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
966 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
967 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
968 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
969 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
973 * SYN appears to be valid; create compressed TCP state
974 * for syncache, or perform t/tcp connection.
976 if (so->so_qlen <= so->so_qlimit) {
978 if (so->so_options & SO_DEBUG)
979 tcp_trace(TA_INPUT, ostate, tp,
980 (void *)tcp_saveipgen, &tcp_savetcp, 0);
982 tcp_dooptions(&to, optp, optlen, TO_SYN);
983 if (!syncache_add(&inc, &to, th, inp, &so, m))
984 goto drop; /* XXX: does not happen */
987 * Entry added to syncache, mbuf used to
988 * send SYN,ACK packet. Everything unlocked
993 panic("T/TCP not supported at the moment");
996 * Segment passed TAO tests.
997 * XXX: Can't happen at the moment.
1000 inp = sotoinpcb(so);
1002 tp = intotcpcb(inp);
1003 tp->t_starttime = ticks;
1004 tp->t_state = TCPS_ESTABLISHED;
1008 * If there is a FIN or if there is data, then
1009 * delay SYN,ACK(SYN) in the hope of piggy-backing
1010 * it on a response segment. Otherwise must send
1011 * ACK now in case the other side is slow starting.
1013 if (thflags & TH_FIN || tlen != 0)
1014 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1016 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1017 tiwin = th->th_win << tp->snd_scale;
1018 tcpstat.tcps_connects++;
1026 KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1027 INP_LOCK_ASSERT(inp);
1029 /* Syncache takes care of sockets in the listen state. */
1030 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1033 * Segment received on connection.
1034 * Reset idle time and keep-alive timer.
1036 tp->t_rcvtime = ticks;
1037 if (TCPS_HAVEESTABLISHED(tp->t_state))
1038 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1041 * Unscale the window into a 32-bit value.
1042 * This value is bogus for the TCPS_SYN_SENT state
1043 * and is overwritten later.
1045 tiwin = th->th_win << tp->snd_scale;
1048 * Parse options on any incoming segment.
1050 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0);
1053 * If echoed timestamp is later than the current time,
1054 * fall back to non RFC1323 RTT calculation. Normalize
1055 * timestamp if syncookies were used when this connection
1058 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1059 to.to_tsecr -= tp->ts_offset;
1060 if (TSTMP_GT(to.to_tsecr, ticks))
1065 * Process options only when we get SYN/ACK back. The SYN case
1066 * for incoming connections is handled in tcp_syncache.
1067 * XXX this is traditional behavior, may need to be cleaned up.
1069 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1070 if ((to.to_flags & TOF_SCALE) &&
1071 (tp->t_flags & TF_REQ_SCALE)) {
1072 tp->t_flags |= TF_RCVD_SCALE;
1073 tp->snd_scale = to.to_wscale;
1074 tp->snd_wnd = th->th_win << tp->snd_scale;
1075 tiwin = tp->snd_wnd;
1077 if (to.to_flags & TOF_TS) {
1078 tp->t_flags |= TF_RCVD_TSTMP;
1079 tp->ts_recent = to.to_tsval;
1080 tp->ts_recent_age = ticks;
1082 /* Initial send window, already scaled. */
1083 tp->snd_wnd = th->th_win;
1084 if (to.to_flags & TOF_MSS)
1085 tcp_mss(tp, to.to_mss);
1086 if (tp->sack_enable) {
1087 if (!(to.to_flags & TOF_SACK))
1088 tp->sack_enable = 0;
1090 tp->t_flags |= TF_SACK_PERMIT;
1096 * Header prediction: check for the two common cases
1097 * of a uni-directional data xfer. If the packet has
1098 * no control flags, is in-sequence, the window didn't
1099 * change and we're not retransmitting, it's a
1100 * candidate. If the length is zero and the ack moved
1101 * forward, we're the sender side of the xfer. Just
1102 * free the data acked & wake any higher level process
1103 * that was blocked waiting for space. If the length
1104 * is non-zero and the ack didn't move, we're the
1105 * receiver side. If we're getting packets in-order
1106 * (the reassembly queue is empty), add the data to
1107 * the socket buffer and note that we need a delayed ack.
1108 * Make sure that the hidden state-flags are also off.
1109 * Since we check for TCPS_ESTABLISHED above, it can only
1112 if (tp->t_state == TCPS_ESTABLISHED &&
1113 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1114 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1115 ((to.to_flags & TOF_TS) == 0 ||
1116 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1117 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1118 tp->snd_nxt == tp->snd_max) {
1121 * If last ACK falls within this segment's sequence numbers,
1122 * record the timestamp.
1123 * NOTE that the test is modified according to the latest
1124 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1126 if ((to.to_flags & TOF_TS) != 0 &&
1127 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1128 tp->ts_recent_age = ticks;
1129 tp->ts_recent = to.to_tsval;
1133 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1134 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1135 tp->snd_cwnd >= tp->snd_wnd &&
1136 ((!tcp_do_newreno && !tp->sack_enable &&
1137 tp->t_dupacks < tcprexmtthresh) ||
1138 ((tcp_do_newreno || tp->sack_enable) &&
1139 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1140 TAILQ_EMPTY(&tp->snd_holes)))) {
1141 KASSERT(headlocked, ("headlocked"));
1142 INP_INFO_WUNLOCK(&tcbinfo);
1145 * this is a pure ack for outstanding data.
1147 ++tcpstat.tcps_predack;
1149 * "bad retransmit" recovery
1151 if (tp->t_rxtshift == 1 &&
1152 ticks < tp->t_badrxtwin) {
1153 ++tcpstat.tcps_sndrexmitbad;
1154 tp->snd_cwnd = tp->snd_cwnd_prev;
1156 tp->snd_ssthresh_prev;
1157 tp->snd_recover = tp->snd_recover_prev;
1158 if (tp->t_flags & TF_WASFRECOVERY)
1159 ENTER_FASTRECOVERY(tp);
1160 tp->snd_nxt = tp->snd_max;
1161 tp->t_badrxtwin = 0;
1165 * Recalculate the transmit timer / rtt.
1167 * Some boxes send broken timestamp replies
1168 * during the SYN+ACK phase, ignore
1169 * timestamps of 0 or we could calculate a
1170 * huge RTT and blow up the retransmit timer.
1172 if ((to.to_flags & TOF_TS) != 0 &&
1174 if (!tp->t_rttlow ||
1175 tp->t_rttlow > ticks - to.to_tsecr)
1176 tp->t_rttlow = ticks - to.to_tsecr;
1178 ticks - to.to_tsecr + 1);
1179 } else if (tp->t_rtttime &&
1180 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1181 if (!tp->t_rttlow ||
1182 tp->t_rttlow > ticks - tp->t_rtttime)
1183 tp->t_rttlow = ticks - tp->t_rtttime;
1185 ticks - tp->t_rtttime);
1187 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1188 acked = th->th_ack - tp->snd_una;
1189 tcpstat.tcps_rcvackpack++;
1190 tcpstat.tcps_rcvackbyte += acked;
1191 sbdrop(&so->so_snd, acked);
1192 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1193 SEQ_LEQ(th->th_ack, tp->snd_recover))
1194 tp->snd_recover = th->th_ack - 1;
1195 tp->snd_una = th->th_ack;
1197 * pull snd_wl2 up to prevent seq wrap relative
1200 tp->snd_wl2 = th->th_ack;
1203 ND6_HINT(tp); /* some progress has been done */
1206 * If all outstanding data are acked, stop
1207 * retransmit timer, otherwise restart timer
1208 * using current (possibly backed-off) value.
1209 * If process is waiting for space,
1210 * wakeup/selwakeup/signal. If data
1211 * are ready to send, let tcp_output
1212 * decide between more output or persist.
1215 if (so->so_options & SO_DEBUG)
1216 tcp_trace(TA_INPUT, ostate, tp,
1217 (void *)tcp_saveipgen,
1221 if (tp->snd_una == tp->snd_max)
1222 callout_stop(tp->tt_rexmt);
1223 else if (!callout_active(tp->tt_persist))
1224 callout_reset(tp->tt_rexmt,
1226 tcp_timer_rexmt, tp);
1229 if (so->so_snd.sb_cc)
1230 (void) tcp_output(tp);
1233 } else if (th->th_ack == tp->snd_una &&
1234 LIST_EMPTY(&tp->t_segq) &&
1235 tlen <= sbspace(&so->so_rcv)) {
1236 int newsize = 0; /* automatic sockbuf scaling */
1238 KASSERT(headlocked, ("headlocked"));
1239 INP_INFO_WUNLOCK(&tcbinfo);
1242 * this is a pure, in-sequence data packet
1243 * with nothing on the reassembly queue and
1244 * we have enough buffer space to take it.
1246 /* Clean receiver SACK report if present */
1247 if (tp->sack_enable && tp->rcv_numsacks)
1248 tcp_clean_sackreport(tp);
1249 ++tcpstat.tcps_preddat;
1250 tp->rcv_nxt += tlen;
1252 * Pull snd_wl1 up to prevent seq wrap relative to
1255 tp->snd_wl1 = th->th_seq;
1257 * Pull rcv_up up to prevent seq wrap relative to
1260 tp->rcv_up = tp->rcv_nxt;
1261 tcpstat.tcps_rcvpack++;
1262 tcpstat.tcps_rcvbyte += tlen;
1263 ND6_HINT(tp); /* some progress has been done */
1265 if (so->so_options & SO_DEBUG)
1266 tcp_trace(TA_INPUT, ostate, tp,
1267 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1270 * Automatic sizing of receive socket buffer. Often the send
1271 * buffer size is not optimally adjusted to the actual network
1272 * conditions at hand (delay bandwidth product). Setting the
1273 * buffer size too small limits throughput on links with high
1274 * bandwidth and high delay (eg. trans-continental/oceanic links).
1276 * On the receive side the socket buffer memory is only rarely
1277 * used to any significant extent. This allows us to be much
1278 * more aggressive in scaling the receive socket buffer. For
1279 * the case that the buffer space is actually used to a large
1280 * extent and we run out of kernel memory we can simply drop
1281 * the new segments; TCP on the sender will just retransmit it
1282 * later. Setting the buffer size too big may only consume too
1283 * much kernel memory if the application doesn't read() from
1284 * the socket or packet loss or reordering makes use of the
1287 * The criteria to step up the receive buffer one notch are:
1288 * 1. the number of bytes received during the time it takes
1289 * one timestamp to be reflected back to us (the RTT);
1290 * 2. received bytes per RTT is within seven eighth of the
1291 * current socket buffer size;
1292 * 3. receive buffer size has not hit maximal automatic size;
1294 * This algorithm does one step per RTT at most and only if
1295 * we receive a bulk stream w/o packet losses or reorderings.
1296 * Shrinking the buffer during idle times is not necessary as
1297 * it doesn't consume any memory when idle.
1299 * TODO: Only step up if the application is actually serving
1300 * the buffer to better manage the socket buffer resources.
1302 if (tcp_do_autorcvbuf &&
1304 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1305 if (to.to_tsecr > tp->rfbuf_ts &&
1306 to.to_tsecr - tp->rfbuf_ts < hz) {
1308 (so->so_rcv.sb_hiwat / 8 * 7) &&
1309 so->so_rcv.sb_hiwat <
1310 tcp_autorcvbuf_max) {
1312 min(so->so_rcv.sb_hiwat +
1314 tcp_autorcvbuf_max);
1316 /* Start over with next RTT. */
1320 tp->rfbuf_cnt += tlen; /* add up */
1323 /* Add data to socket buffer. */
1324 SOCKBUF_LOCK(&so->so_rcv);
1325 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1329 * Set new socket buffer size.
1330 * Give up when limit is reached.
1333 if (!sbreserve_locked(&so->so_rcv,
1334 newsize, so, curthread))
1335 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
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 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1368 switch (tp->t_state) {
1371 * If the state is SYN_RECEIVED:
1372 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1374 case TCPS_SYN_RECEIVED:
1375 if ((thflags & TH_ACK) &&
1376 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1377 SEQ_GT(th->th_ack, tp->snd_max))) {
1378 rstreason = BANDLIM_RST_OPENPORT;
1384 * If the state is SYN_SENT:
1385 * if seg contains an ACK, but not for our SYN, drop the input.
1386 * if seg contains a RST, then drop the connection.
1387 * if seg does not contain SYN, then drop it.
1388 * Otherwise this is an acceptable SYN segment
1389 * initialize tp->rcv_nxt and tp->irs
1390 * if seg contains ack then advance tp->snd_una
1391 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1392 * arrange for segment to be acked (eventually)
1393 * continue processing rest of data/controls, beginning with URG
1396 if ((thflags & TH_ACK) &&
1397 (SEQ_LEQ(th->th_ack, tp->iss) ||
1398 SEQ_GT(th->th_ack, tp->snd_max))) {
1399 rstreason = BANDLIM_UNLIMITED;
1402 if (thflags & TH_RST) {
1403 if (thflags & TH_ACK) {
1404 KASSERT(headlocked, ("tcp_input: after_listen"
1405 ": tcp_drop.2: head not locked"));
1406 tp = tcp_drop(tp, ECONNREFUSED);
1410 if ((thflags & TH_SYN) == 0)
1413 tp->irs = th->th_seq;
1415 if (thflags & TH_ACK) {
1416 tcpstat.tcps_connects++;
1420 mac_set_socket_peer_from_mbuf(m, so);
1423 /* Do window scaling on this connection? */
1424 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1425 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1426 tp->rcv_scale = tp->request_r_scale;
1428 tp->rcv_adv += tp->rcv_wnd;
1429 tp->snd_una++; /* SYN is acked */
1431 * If there's data, delay ACK; if there's also a FIN
1432 * ACKNOW will be turned on later.
1434 if (DELAY_ACK(tp) && tlen != 0)
1435 callout_reset(tp->tt_delack, tcp_delacktime,
1436 tcp_timer_delack, tp);
1438 tp->t_flags |= TF_ACKNOW;
1440 * Received <SYN,ACK> in SYN_SENT[*] state.
1442 * SYN_SENT --> ESTABLISHED
1443 * SYN_SENT* --> FIN_WAIT_1
1445 tp->t_starttime = ticks;
1446 if (tp->t_flags & TF_NEEDFIN) {
1447 tp->t_state = TCPS_FIN_WAIT_1;
1448 tp->t_flags &= ~TF_NEEDFIN;
1451 tp->t_state = TCPS_ESTABLISHED;
1452 callout_reset(tp->tt_keep, tcp_keepidle,
1453 tcp_timer_keep, tp);
1457 * Received initial SYN in SYN-SENT[*] state =>
1458 * simultaneous open. If segment contains CC option
1459 * and there is a cached CC, apply TAO test.
1460 * If it succeeds, connection is * half-synchronized.
1461 * Otherwise, do 3-way handshake:
1462 * SYN-SENT -> SYN-RECEIVED
1463 * SYN-SENT* -> SYN-RECEIVED*
1464 * If there was no CC option, clear cached CC value.
1466 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1467 callout_stop(tp->tt_rexmt);
1468 tp->t_state = TCPS_SYN_RECEIVED;
1474 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1476 INP_LOCK_ASSERT(inp);
1479 * Advance th->th_seq to correspond to first data byte.
1480 * If data, trim to stay within window,
1481 * dropping FIN if necessary.
1484 if (tlen > tp->rcv_wnd) {
1485 todrop = tlen - tp->rcv_wnd;
1489 tcpstat.tcps_rcvpackafterwin++;
1490 tcpstat.tcps_rcvbyteafterwin += todrop;
1492 tp->snd_wl1 = th->th_seq - 1;
1493 tp->rcv_up = th->th_seq;
1495 * Client side of transaction: already sent SYN and data.
1496 * If the remote host used T/TCP to validate the SYN,
1497 * our data will be ACK'd; if so, enter normal data segment
1498 * processing in the middle of step 5, ack processing.
1499 * Otherwise, goto step 6.
1501 if (thflags & TH_ACK)
1507 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1508 * do normal processing.
1510 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1514 case TCPS_TIME_WAIT:
1515 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1516 break; /* continue normal processing */
1520 * States other than LISTEN or SYN_SENT.
1521 * First check the RST flag and sequence number since reset segments
1522 * are exempt from the timestamp and connection count tests. This
1523 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1524 * below which allowed reset segments in half the sequence space
1525 * to fall though and be processed (which gives forged reset
1526 * segments with a random sequence number a 50 percent chance of
1527 * killing a connection).
1528 * Then check timestamp, if present.
1529 * Then check the connection count, if present.
1530 * Then check that at least some bytes of segment are within
1531 * receive window. If segment begins before rcv_nxt,
1532 * drop leading data (and SYN); if nothing left, just ack.
1535 * If the RST bit is set, check the sequence number to see
1536 * if this is a valid reset segment.
1538 * In all states except SYN-SENT, all reset (RST) segments
1539 * are validated by checking their SEQ-fields. A reset is
1540 * valid if its sequence number is in the window.
1541 * Note: this does not take into account delayed ACKs, so
1542 * we should test against last_ack_sent instead of rcv_nxt.
1543 * The sequence number in the reset segment is normally an
1544 * echo of our outgoing acknowlegement numbers, but some hosts
1545 * send a reset with the sequence number at the rightmost edge
1546 * of our receive window, and we have to handle this case.
1547 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1548 * that brute force RST attacks are possible. To combat this,
1549 * we use a much stricter check while in the ESTABLISHED state,
1550 * only accepting RSTs where the sequence number is equal to
1551 * last_ack_sent. In all other states (the states in which a
1552 * RST is more likely), the more permissive check is used.
1553 * If we have multiple segments in flight, the intial reset
1554 * segment sequence numbers will be to the left of last_ack_sent,
1555 * but they will eventually catch up.
1556 * In any case, it never made sense to trim reset segments to
1557 * fit the receive window since RFC 1122 says:
1558 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1560 * A TCP SHOULD allow a received RST segment to include data.
1563 * It has been suggested that a RST segment could contain
1564 * ASCII text that encoded and explained the cause of the
1565 * RST. No standard has yet been established for such
1568 * If the reset segment passes the sequence number test examine
1570 * SYN_RECEIVED STATE:
1571 * If passive open, return to LISTEN state.
1572 * If active open, inform user that connection was refused.
1573 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1574 * Inform user that connection was reset, and close tcb.
1575 * CLOSING, LAST_ACK STATES:
1578 * Drop the segment - see Stevens, vol. 2, p. 964 and
1581 if (thflags & TH_RST) {
1582 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1583 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1584 switch (tp->t_state) {
1586 case TCPS_SYN_RECEIVED:
1587 so->so_error = ECONNREFUSED;
1590 case TCPS_ESTABLISHED:
1591 if (tcp_insecure_rst == 0 &&
1592 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
1593 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
1594 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1595 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
1596 tcpstat.tcps_badrst++;
1599 case TCPS_FIN_WAIT_1:
1600 case TCPS_FIN_WAIT_2:
1601 case TCPS_CLOSE_WAIT:
1602 so->so_error = ECONNRESET;
1604 tp->t_state = TCPS_CLOSED;
1605 tcpstat.tcps_drops++;
1606 KASSERT(headlocked, ("tcp_input: "
1607 "trimthenstep6: tcp_close: head not "
1614 KASSERT(headlocked, ("trimthenstep6: "
1615 "tcp_close.2: head not locked"));
1619 case TCPS_TIME_WAIT:
1620 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1629 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1630 * and it's less than ts_recent, drop it.
1632 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1633 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1635 /* Check to see if ts_recent is over 24 days old. */
1636 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1638 * Invalidate ts_recent. If this segment updates
1639 * ts_recent, the age will be reset later and ts_recent
1640 * will get a valid value. If it does not, setting
1641 * ts_recent to zero will at least satisfy the
1642 * requirement that zero be placed in the timestamp
1643 * echo reply when ts_recent isn't valid. The
1644 * age isn't reset until we get a valid ts_recent
1645 * because we don't want out-of-order segments to be
1646 * dropped when ts_recent is old.
1650 tcpstat.tcps_rcvduppack++;
1651 tcpstat.tcps_rcvdupbyte += tlen;
1652 tcpstat.tcps_pawsdrop++;
1660 * In the SYN-RECEIVED state, validate that the packet belongs to
1661 * this connection before trimming the data to fit the receive
1662 * window. Check the sequence number versus IRS since we know
1663 * the sequence numbers haven't wrapped. This is a partial fix
1664 * for the "LAND" DoS attack.
1666 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1667 rstreason = BANDLIM_RST_OPENPORT;
1671 todrop = tp->rcv_nxt - th->th_seq;
1673 if (thflags & TH_SYN) {
1683 * Following if statement from Stevens, vol. 2, p. 960.
1686 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1688 * Any valid FIN must be to the left of the window.
1689 * At this point the FIN must be a duplicate or out
1690 * of sequence; drop it.
1695 * Send an ACK to resynchronize and drop any data.
1696 * But keep on processing for RST or ACK.
1698 tp->t_flags |= TF_ACKNOW;
1700 tcpstat.tcps_rcvduppack++;
1701 tcpstat.tcps_rcvdupbyte += todrop;
1703 tcpstat.tcps_rcvpartduppack++;
1704 tcpstat.tcps_rcvpartdupbyte += todrop;
1706 drop_hdrlen += todrop; /* drop from the top afterwards */
1707 th->th_seq += todrop;
1709 if (th->th_urp > todrop)
1710 th->th_urp -= todrop;
1718 * If new data are received on a connection after the
1719 * user processes are gone, then RST the other end.
1721 if ((so->so_state & SS_NOFDREF) &&
1722 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1723 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1726 tcpstat.tcps_rcvafterclose++;
1727 rstreason = BANDLIM_UNLIMITED;
1732 * If segment ends after window, drop trailing data
1733 * (and PUSH and FIN); if nothing left, just ACK.
1735 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1737 tcpstat.tcps_rcvpackafterwin++;
1738 if (todrop >= tlen) {
1739 tcpstat.tcps_rcvbyteafterwin += tlen;
1741 * If a new connection request is received
1742 * while in TIME_WAIT, drop the old connection
1743 * and start over if the sequence numbers
1744 * are above the previous ones.
1746 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1747 if (thflags & TH_SYN &&
1748 tp->t_state == TCPS_TIME_WAIT &&
1749 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1750 KASSERT(headlocked, ("trimthenstep6: "
1751 "tcp_close.4: head not locked"));
1756 * If window is closed can only take segments at
1757 * window edge, and have to drop data and PUSH from
1758 * incoming segments. Continue processing, but
1759 * remember to ack. Otherwise, drop segment
1762 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1763 tp->t_flags |= TF_ACKNOW;
1764 tcpstat.tcps_rcvwinprobe++;
1768 tcpstat.tcps_rcvbyteafterwin += todrop;
1771 thflags &= ~(TH_PUSH|TH_FIN);
1775 * If last ACK falls within this segment's sequence numbers,
1776 * record its timestamp.
1778 * 1) That the test incorporates suggestions from the latest
1779 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1780 * 2) That updating only on newer timestamps interferes with
1781 * our earlier PAWS tests, so this check should be solely
1782 * predicated on the sequence space of this segment.
1783 * 3) That we modify the segment boundary check to be
1784 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1785 * instead of RFC1323's
1786 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1787 * This modified check allows us to overcome RFC1323's
1788 * limitations as described in Stevens TCP/IP Illustrated
1789 * Vol. 2 p.869. In such cases, we can still calculate the
1790 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1792 if ((to.to_flags & TOF_TS) != 0 &&
1793 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1794 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1795 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1796 tp->ts_recent_age = ticks;
1797 tp->ts_recent = to.to_tsval;
1801 * If a SYN is in the window, then this is an
1802 * error and we send an RST and drop the connection.
1804 if (thflags & TH_SYN) {
1805 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1806 "head not locked"));
1807 tp = tcp_drop(tp, ECONNRESET);
1808 rstreason = BANDLIM_UNLIMITED;
1813 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1814 * flag is on (half-synchronized state), then queue data for
1815 * later processing; else drop segment and return.
1817 if ((thflags & TH_ACK) == 0) {
1818 if (tp->t_state == TCPS_SYN_RECEIVED ||
1819 (tp->t_flags & TF_NEEDSYN))
1821 else if (tp->t_flags & TF_ACKNOW)
1830 switch (tp->t_state) {
1833 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1834 * ESTABLISHED state and continue processing.
1835 * The ACK was checked above.
1837 case TCPS_SYN_RECEIVED:
1839 tcpstat.tcps_connects++;
1841 /* Do window scaling? */
1842 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1843 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1844 tp->rcv_scale = tp->request_r_scale;
1845 tp->snd_wnd = tiwin;
1849 * SYN-RECEIVED -> ESTABLISHED
1850 * SYN-RECEIVED* -> FIN-WAIT-1
1852 tp->t_starttime = ticks;
1853 if (tp->t_flags & TF_NEEDFIN) {
1854 tp->t_state = TCPS_FIN_WAIT_1;
1855 tp->t_flags &= ~TF_NEEDFIN;
1857 tp->t_state = TCPS_ESTABLISHED;
1858 callout_reset(tp->tt_keep, tcp_keepidle,
1859 tcp_timer_keep, tp);
1862 * If segment contains data or ACK, will call tcp_reass()
1863 * later; if not, do so now to pass queued data to user.
1865 if (tlen == 0 && (thflags & TH_FIN) == 0)
1866 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1868 tp->snd_wl1 = th->th_seq - 1;
1872 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1873 * ACKs. If the ack is in the range
1874 * tp->snd_una < th->th_ack <= tp->snd_max
1875 * then advance tp->snd_una to th->th_ack and drop
1876 * data from the retransmission queue. If this ACK reflects
1877 * more up to date window information we update our window information.
1879 case TCPS_ESTABLISHED:
1880 case TCPS_FIN_WAIT_1:
1881 case TCPS_FIN_WAIT_2:
1882 case TCPS_CLOSE_WAIT:
1885 case TCPS_TIME_WAIT:
1886 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1887 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1888 tcpstat.tcps_rcvacktoomuch++;
1891 if (tp->sack_enable &&
1892 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1893 tcp_sack_doack(tp, &to, th->th_ack);
1894 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1895 if (tlen == 0 && tiwin == tp->snd_wnd) {
1896 tcpstat.tcps_rcvdupack++;
1898 * If we have outstanding data (other than
1899 * a window probe), this is a completely
1900 * duplicate ack (ie, window info didn't
1901 * change), the ack is the biggest we've
1902 * seen and we've seen exactly our rexmt
1903 * threshhold of them, assume a packet
1904 * has been dropped and retransmit it.
1905 * Kludge snd_nxt & the congestion
1906 * window so we send only this one
1909 * We know we're losing at the current
1910 * window size so do congestion avoidance
1911 * (set ssthresh to half the current window
1912 * and pull our congestion window back to
1913 * the new ssthresh).
1915 * Dup acks mean that packets have left the
1916 * network (they're now cached at the receiver)
1917 * so bump cwnd by the amount in the receiver
1918 * to keep a constant cwnd packets in the
1921 if (!callout_active(tp->tt_rexmt) ||
1922 th->th_ack != tp->snd_una)
1924 else if (++tp->t_dupacks > tcprexmtthresh ||
1925 ((tcp_do_newreno || tp->sack_enable) &&
1926 IN_FASTRECOVERY(tp))) {
1927 if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1931 * Compute the amount of data in flight first.
1932 * We can inject new data into the pipe iff
1933 * we have less than 1/2 the original window's
1934 * worth of data in flight.
1936 awnd = (tp->snd_nxt - tp->snd_fack) +
1937 tp->sackhint.sack_bytes_rexmit;
1938 if (awnd < tp->snd_ssthresh) {
1939 tp->snd_cwnd += tp->t_maxseg;
1940 if (tp->snd_cwnd > tp->snd_ssthresh)
1941 tp->snd_cwnd = tp->snd_ssthresh;
1944 tp->snd_cwnd += tp->t_maxseg;
1945 (void) tcp_output(tp);
1947 } else if (tp->t_dupacks == tcprexmtthresh) {
1948 tcp_seq onxt = tp->snd_nxt;
1952 * If we're doing sack, check to
1953 * see if we're already in sack
1954 * recovery. If we're not doing sack,
1955 * check to see if we're in newreno
1958 if (tp->sack_enable) {
1959 if (IN_FASTRECOVERY(tp)) {
1963 } else if (tcp_do_newreno) {
1964 if (SEQ_LEQ(th->th_ack,
1970 win = min(tp->snd_wnd, tp->snd_cwnd) /
1974 tp->snd_ssthresh = win * tp->t_maxseg;
1975 ENTER_FASTRECOVERY(tp);
1976 tp->snd_recover = tp->snd_max;
1977 callout_stop(tp->tt_rexmt);
1979 if (tp->sack_enable) {
1980 tcpstat.tcps_sack_recovery_episode++;
1981 tp->sack_newdata = tp->snd_nxt;
1982 tp->snd_cwnd = tp->t_maxseg;
1983 (void) tcp_output(tp);
1986 tp->snd_nxt = th->th_ack;
1987 tp->snd_cwnd = tp->t_maxseg;
1988 (void) tcp_output(tp);
1989 KASSERT(tp->snd_limited <= 2,
1990 ("tp->snd_limited too big"));
1991 tp->snd_cwnd = tp->snd_ssthresh +
1993 (tp->t_dupacks - tp->snd_limited);
1994 if (SEQ_GT(onxt, tp->snd_nxt))
1997 } else if (tcp_do_rfc3042) {
1998 u_long oldcwnd = tp->snd_cwnd;
1999 tcp_seq oldsndmax = tp->snd_max;
2002 KASSERT(tp->t_dupacks == 1 ||
2004 ("dupacks not 1 or 2"));
2005 if (tp->t_dupacks == 1)
2006 tp->snd_limited = 0;
2008 (tp->snd_nxt - tp->snd_una) +
2009 (tp->t_dupacks - tp->snd_limited) *
2011 (void) tcp_output(tp);
2012 sent = tp->snd_max - oldsndmax;
2013 if (sent > tp->t_maxseg) {
2014 KASSERT((tp->t_dupacks == 2 &&
2015 tp->snd_limited == 0) ||
2016 (sent == tp->t_maxseg + 1 &&
2017 tp->t_flags & TF_SENTFIN),
2019 tp->snd_limited = 2;
2020 } else if (sent > 0)
2022 tp->snd_cwnd = oldcwnd;
2030 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2033 * If the congestion window was inflated to account
2034 * for the other side's cached packets, retract it.
2036 if (tcp_do_newreno || tp->sack_enable) {
2037 if (IN_FASTRECOVERY(tp)) {
2038 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2039 if (tp->sack_enable)
2040 tcp_sack_partialack(tp, th);
2042 tcp_newreno_partial_ack(tp, th);
2045 * Out of fast recovery.
2046 * Window inflation should have left us
2047 * with approximately snd_ssthresh
2049 * But in case we would be inclined to
2050 * send a burst, better to do it via
2051 * the slow start mechanism.
2053 if (SEQ_GT(th->th_ack +
2056 tp->snd_cwnd = tp->snd_max -
2060 tp->snd_cwnd = tp->snd_ssthresh;
2064 if (tp->t_dupacks >= tcprexmtthresh &&
2065 tp->snd_cwnd > tp->snd_ssthresh)
2066 tp->snd_cwnd = tp->snd_ssthresh;
2070 * If we reach this point, ACK is not a duplicate,
2071 * i.e., it ACKs something we sent.
2073 if (tp->t_flags & TF_NEEDSYN) {
2075 * T/TCP: Connection was half-synchronized, and our
2076 * SYN has been ACK'd (so connection is now fully
2077 * synchronized). Go to non-starred state,
2078 * increment snd_una for ACK of SYN, and check if
2079 * we can do window scaling.
2081 tp->t_flags &= ~TF_NEEDSYN;
2083 /* Do window scaling? */
2084 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2085 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2086 tp->rcv_scale = tp->request_r_scale;
2087 /* Send window already scaled. */
2092 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2094 INP_LOCK_ASSERT(inp);
2096 acked = th->th_ack - tp->snd_una;
2097 tcpstat.tcps_rcvackpack++;
2098 tcpstat.tcps_rcvackbyte += acked;
2101 * If we just performed our first retransmit, and the ACK
2102 * arrives within our recovery window, then it was a mistake
2103 * to do the retransmit in the first place. Recover our
2104 * original cwnd and ssthresh, and proceed to transmit where
2107 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2108 ++tcpstat.tcps_sndrexmitbad;
2109 tp->snd_cwnd = tp->snd_cwnd_prev;
2110 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2111 tp->snd_recover = tp->snd_recover_prev;
2112 if (tp->t_flags & TF_WASFRECOVERY)
2113 ENTER_FASTRECOVERY(tp);
2114 tp->snd_nxt = tp->snd_max;
2115 tp->t_badrxtwin = 0; /* XXX probably not required */
2119 * If we have a timestamp reply, update smoothed
2120 * round trip time. If no timestamp is present but
2121 * transmit timer is running and timed sequence
2122 * number was acked, update smoothed round trip time.
2123 * Since we now have an rtt measurement, cancel the
2124 * timer backoff (cf., Phil Karn's retransmit alg.).
2125 * Recompute the initial retransmit timer.
2127 * Some boxes send broken timestamp replies
2128 * during the SYN+ACK phase, ignore
2129 * timestamps of 0 or we could calculate a
2130 * huge RTT and blow up the retransmit timer.
2132 if ((to.to_flags & TOF_TS) != 0 &&
2134 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2135 tp->t_rttlow = ticks - to.to_tsecr;
2136 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2137 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2138 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2139 tp->t_rttlow = ticks - tp->t_rtttime;
2140 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2142 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2145 * If all outstanding data is acked, stop retransmit
2146 * timer and remember to restart (more output or persist).
2147 * If there is more data to be acked, restart retransmit
2148 * timer, using current (possibly backed-off) value.
2150 if (th->th_ack == tp->snd_max) {
2151 callout_stop(tp->tt_rexmt);
2153 } else if (!callout_active(tp->tt_persist))
2154 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2155 tcp_timer_rexmt, tp);
2158 * If no data (only SYN) was ACK'd,
2159 * skip rest of ACK processing.
2165 * When new data is acked, open the congestion window.
2166 * If the window gives us less than ssthresh packets
2167 * in flight, open exponentially (maxseg per packet).
2168 * Otherwise open linearly: maxseg per window
2169 * (maxseg^2 / cwnd per packet).
2171 if ((!tcp_do_newreno && !tp->sack_enable) ||
2172 !IN_FASTRECOVERY(tp)) {
2173 register u_int cw = tp->snd_cwnd;
2174 register u_int incr = tp->t_maxseg;
2175 if (cw > tp->snd_ssthresh)
2176 incr = incr * incr / cw;
2177 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2179 SOCKBUF_LOCK(&so->so_snd);
2180 if (acked > so->so_snd.sb_cc) {
2181 tp->snd_wnd -= so->so_snd.sb_cc;
2182 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2185 sbdrop_locked(&so->so_snd, acked);
2186 tp->snd_wnd -= acked;
2189 sowwakeup_locked(so);
2190 /* detect una wraparound */
2191 if ((tcp_do_newreno || tp->sack_enable) &&
2192 !IN_FASTRECOVERY(tp) &&
2193 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2194 SEQ_LEQ(th->th_ack, tp->snd_recover))
2195 tp->snd_recover = th->th_ack - 1;
2196 if ((tcp_do_newreno || tp->sack_enable) &&
2197 IN_FASTRECOVERY(tp) &&
2198 SEQ_GEQ(th->th_ack, tp->snd_recover))
2199 EXIT_FASTRECOVERY(tp);
2200 tp->snd_una = th->th_ack;
2201 if (tp->sack_enable) {
2202 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2203 tp->snd_recover = tp->snd_una;
2205 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2206 tp->snd_nxt = tp->snd_una;
2208 switch (tp->t_state) {
2211 * In FIN_WAIT_1 STATE in addition to the processing
2212 * for the ESTABLISHED state if our FIN is now acknowledged
2213 * then enter FIN_WAIT_2.
2215 case TCPS_FIN_WAIT_1:
2216 if (ourfinisacked) {
2218 * If we can't receive any more
2219 * data, then closing user can proceed.
2220 * Starting the timer is contrary to the
2221 * specification, but if we don't get a FIN
2222 * we'll hang forever.
2225 * we should release the tp also, and use a
2228 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2231 soisdisconnected(so);
2232 timeout = (tcp_fast_finwait2_recycle) ?
2233 tcp_finwait2_timeout : tcp_maxidle;
2234 callout_reset(tp->tt_2msl, timeout,
2235 tcp_timer_2msl, tp);
2237 tp->t_state = TCPS_FIN_WAIT_2;
2242 * In CLOSING STATE in addition to the processing for
2243 * the ESTABLISHED state if the ACK acknowledges our FIN
2244 * then enter the TIME-WAIT state, otherwise ignore
2248 if (ourfinisacked) {
2249 KASSERT(headlocked, ("tcp_input: process_ACK: "
2250 "head not locked"));
2252 INP_INFO_WUNLOCK(&tcbinfo);
2259 * In LAST_ACK, we may still be waiting for data to drain
2260 * and/or to be acked, as well as for the ack of our FIN.
2261 * If our FIN is now acknowledged, delete the TCB,
2262 * enter the closed state and return.
2265 if (ourfinisacked) {
2266 KASSERT(headlocked, ("tcp_input: process_ACK:"
2267 " tcp_close: head not locked"));
2274 * In TIME_WAIT state the only thing that should arrive
2275 * is a retransmission of the remote FIN. Acknowledge
2276 * it and restart the finack timer.
2278 case TCPS_TIME_WAIT:
2279 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2280 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2281 tcp_timer_2msl, tp);
2287 KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2288 INP_LOCK_ASSERT(inp);
2291 * Update window information.
2292 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2294 if ((thflags & TH_ACK) &&
2295 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2296 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2297 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2298 /* keep track of pure window updates */
2300 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2301 tcpstat.tcps_rcvwinupd++;
2302 tp->snd_wnd = tiwin;
2303 tp->snd_wl1 = th->th_seq;
2304 tp->snd_wl2 = th->th_ack;
2305 if (tp->snd_wnd > tp->max_sndwnd)
2306 tp->max_sndwnd = tp->snd_wnd;
2311 * Process segments with URG.
2313 if ((thflags & TH_URG) && th->th_urp &&
2314 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2316 * This is a kludge, but if we receive and accept
2317 * random urgent pointers, we'll crash in
2318 * soreceive. It's hard to imagine someone
2319 * actually wanting to send this much urgent data.
2321 SOCKBUF_LOCK(&so->so_rcv);
2322 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2323 th->th_urp = 0; /* XXX */
2324 thflags &= ~TH_URG; /* XXX */
2325 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2326 goto dodata; /* XXX */
2329 * If this segment advances the known urgent pointer,
2330 * then mark the data stream. This should not happen
2331 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2332 * a FIN has been received from the remote side.
2333 * In these states we ignore the URG.
2335 * According to RFC961 (Assigned Protocols),
2336 * the urgent pointer points to the last octet
2337 * of urgent data. We continue, however,
2338 * to consider it to indicate the first octet
2339 * of data past the urgent section as the original
2340 * spec states (in one of two places).
2342 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2343 tp->rcv_up = th->th_seq + th->th_urp;
2344 so->so_oobmark = so->so_rcv.sb_cc +
2345 (tp->rcv_up - tp->rcv_nxt) - 1;
2346 if (so->so_oobmark == 0)
2347 so->so_rcv.sb_state |= SBS_RCVATMARK;
2349 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2351 SOCKBUF_UNLOCK(&so->so_rcv);
2353 * Remove out of band data so doesn't get presented to user.
2354 * This can happen independent of advancing the URG pointer,
2355 * but if two URG's are pending at once, some out-of-band
2356 * data may creep in... ick.
2358 if (th->th_urp <= (u_long)tlen &&
2359 !(so->so_options & SO_OOBINLINE)) {
2360 /* hdr drop is delayed */
2361 tcp_pulloutofband(so, th, m, drop_hdrlen);
2365 * If no out of band data is expected,
2366 * pull receive urgent pointer along
2367 * with the receive window.
2369 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2370 tp->rcv_up = tp->rcv_nxt;
2373 KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2374 INP_LOCK_ASSERT(inp);
2377 * Process the segment text, merging it into the TCP sequencing queue,
2378 * and arranging for acknowledgment of receipt if necessary.
2379 * This process logically involves adjusting tp->rcv_wnd as data
2380 * is presented to the user (this happens in tcp_usrreq.c,
2381 * case PRU_RCVD). If a FIN has already been received on this
2382 * connection then we just ignore the text.
2384 if ((tlen || (thflags & TH_FIN)) &&
2385 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2386 tcp_seq save_start = th->th_seq;
2387 tcp_seq save_end = th->th_seq + tlen;
2388 m_adj(m, drop_hdrlen); /* delayed header drop */
2390 * Insert segment which includes th into TCP reassembly queue
2391 * with control block tp. Set thflags to whether reassembly now
2392 * includes a segment with FIN. This handles the common case
2393 * inline (segment is the next to be received on an established
2394 * connection, and the queue is empty), avoiding linkage into
2395 * and removal from the queue and repetition of various
2397 * Set DELACK for segments received in order, but ack
2398 * immediately when segments are out of order (so
2399 * fast retransmit can work).
2401 if (th->th_seq == tp->rcv_nxt &&
2402 LIST_EMPTY(&tp->t_segq) &&
2403 TCPS_HAVEESTABLISHED(tp->t_state)) {
2405 tp->t_flags |= TF_DELACK;
2407 tp->t_flags |= TF_ACKNOW;
2408 tp->rcv_nxt += tlen;
2409 thflags = th->th_flags & TH_FIN;
2410 tcpstat.tcps_rcvpack++;
2411 tcpstat.tcps_rcvbyte += tlen;
2413 SOCKBUF_LOCK(&so->so_rcv);
2414 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2417 sbappendstream_locked(&so->so_rcv, m);
2418 sorwakeup_locked(so);
2420 thflags = tcp_reass(tp, th, &tlen, m);
2421 tp->t_flags |= TF_ACKNOW;
2423 if (tlen > 0 && tp->sack_enable)
2424 tcp_update_sack_list(tp, save_start, save_end);
2426 * Note the amount of data that peer has sent into
2427 * our window, in order to estimate the sender's
2430 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2437 * If FIN is received ACK the FIN and let the user know
2438 * that the connection is closing.
2440 if (thflags & TH_FIN) {
2441 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2444 * If connection is half-synchronized
2445 * (ie NEEDSYN flag on) then delay ACK,
2446 * so it may be piggybacked when SYN is sent.
2447 * Otherwise, since we received a FIN then no
2448 * more input can be expected, send ACK now.
2450 if (tp->t_flags & TF_NEEDSYN)
2451 tp->t_flags |= TF_DELACK;
2453 tp->t_flags |= TF_ACKNOW;
2456 switch (tp->t_state) {
2459 * In SYN_RECEIVED and ESTABLISHED STATES
2460 * enter the CLOSE_WAIT state.
2462 case TCPS_SYN_RECEIVED:
2463 tp->t_starttime = ticks;
2465 case TCPS_ESTABLISHED:
2466 tp->t_state = TCPS_CLOSE_WAIT;
2470 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2471 * enter the CLOSING state.
2473 case TCPS_FIN_WAIT_1:
2474 tp->t_state = TCPS_CLOSING;
2478 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2479 * starting the time-wait timer, turning off the other
2482 case TCPS_FIN_WAIT_2:
2483 KASSERT(headlocked == 1, ("tcp_input: dodata: "
2484 "TCP_FIN_WAIT_2: head not locked"));
2486 INP_INFO_WUNLOCK(&tcbinfo);
2490 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2492 case TCPS_TIME_WAIT:
2493 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2494 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2495 tcp_timer_2msl, tp);
2499 INP_INFO_WUNLOCK(&tcbinfo);
2502 if (so->so_options & SO_DEBUG)
2503 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2508 * Return any desired output.
2510 if (needoutput || (tp->t_flags & TF_ACKNOW))
2511 (void) tcp_output(tp);
2514 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2515 INP_LOCK_ASSERT(inp);
2516 if (tp->t_flags & TF_DELACK) {
2517 tp->t_flags &= ~TF_DELACK;
2518 callout_reset(tp->tt_delack, tcp_delacktime,
2519 tcp_timer_delack, tp);
2525 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2527 * Generate an ACK dropping incoming segment if it occupies
2528 * sequence space, where the ACK reflects our state.
2530 * We can now skip the test for the RST flag since all
2531 * paths to this code happen after packets containing
2532 * RST have been dropped.
2534 * In the SYN-RECEIVED state, don't send an ACK unless the
2535 * segment we received passes the SYN-RECEIVED ACK test.
2536 * If it fails send a RST. This breaks the loop in the
2537 * "LAND" DoS attack, and also prevents an ACK storm
2538 * between two listening ports that have been sent forged
2539 * SYN segments, each with the source address of the other.
2541 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2542 (SEQ_GT(tp->snd_una, th->th_ack) ||
2543 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2544 rstreason = BANDLIM_RST_OPENPORT;
2548 if (so->so_options & SO_DEBUG)
2549 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2552 KASSERT(headlocked, ("headlocked should be 1"));
2553 INP_INFO_WUNLOCK(&tcbinfo);
2554 tp->t_flags |= TF_ACKNOW;
2555 (void) tcp_output(tp);
2561 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2563 * Generate a RST, dropping incoming segment.
2564 * Make ACK acceptable to originator of segment.
2565 * Don't bother to respond if destination was broadcast/multicast.
2567 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2570 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2571 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2574 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2575 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2576 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2577 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2580 /* IPv6 anycast check is done at tcp6_input() */
2583 * Perform bandwidth limiting.
2585 if (badport_bandlim(rstreason) < 0)
2589 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2590 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2594 if (thflags & TH_ACK)
2595 /* mtod() below is safe as long as hdr dropping is delayed */
2596 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2599 if (thflags & TH_SYN)
2601 /* mtod() below is safe as long as hdr dropping is delayed */
2602 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2603 (tcp_seq)0, TH_RST|TH_ACK);
2609 INP_INFO_WUNLOCK(&tcbinfo);
2614 * Drop space held by incoming segment and return.
2617 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2618 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2624 INP_INFO_WUNLOCK(&tcbinfo);
2630 * Parse TCP options and place in tcpopt.
2633 tcp_dooptions(to, cp, cnt, flags)
2642 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2644 if (opt == TCPOPT_EOL)
2646 if (opt == TCPOPT_NOP)
2652 if (optlen < 2 || optlen > cnt)
2657 if (optlen != TCPOLEN_MAXSEG)
2659 if (!(flags & TO_SYN))
2661 to->to_flags |= TOF_MSS;
2662 bcopy((char *)cp + 2,
2663 (char *)&to->to_mss, sizeof(to->to_mss));
2664 to->to_mss = ntohs(to->to_mss);
2667 if (optlen != TCPOLEN_WINDOW)
2669 if (!(flags & TO_SYN))
2671 to->to_flags |= TOF_SCALE;
2672 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
2674 case TCPOPT_TIMESTAMP:
2675 if (optlen != TCPOLEN_TIMESTAMP)
2677 to->to_flags |= TOF_TS;
2678 bcopy((char *)cp + 2,
2679 (char *)&to->to_tsval, sizeof(to->to_tsval));
2680 to->to_tsval = ntohl(to->to_tsval);
2681 bcopy((char *)cp + 6,
2682 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2683 to->to_tsecr = ntohl(to->to_tsecr);
2685 #ifdef TCP_SIGNATURE
2687 * XXX In order to reply to a host which has set the
2688 * TCP_SIGNATURE option in its initial SYN, we have to
2689 * record the fact that the option was observed here
2690 * for the syncache code to perform the correct response.
2692 case TCPOPT_SIGNATURE:
2693 if (optlen != TCPOLEN_SIGNATURE)
2695 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2698 case TCPOPT_SACK_PERMITTED:
2699 if (optlen != TCPOLEN_SACK_PERMITTED)
2701 if (!(flags & TO_SYN))
2705 to->to_flags |= TOF_SACK;
2708 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2710 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2711 to->to_sacks = cp + 2;
2712 tcpstat.tcps_sack_rcv_blocks++;
2721 * Pull out of band byte out of a segment so
2722 * it doesn't appear in the user's data queue.
2723 * It is still reflected in the segment length for
2724 * sequencing purposes.
2727 tcp_pulloutofband(so, th, m, off)
2730 register struct mbuf *m;
2731 int off; /* delayed to be droped hdrlen */
2733 int cnt = off + th->th_urp - 1;
2736 if (m->m_len > cnt) {
2737 char *cp = mtod(m, caddr_t) + cnt;
2738 struct tcpcb *tp = sototcpcb(so);
2741 tp->t_oobflags |= TCPOOB_HAVEDATA;
2742 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2744 if (m->m_flags & M_PKTHDR)
2753 panic("tcp_pulloutofband");
2757 * Collect new round-trip time estimate
2758 * and update averages and current timeout.
2761 tcp_xmit_timer(tp, rtt)
2762 register struct tcpcb *tp;
2767 INP_LOCK_ASSERT(tp->t_inpcb);
2769 tcpstat.tcps_rttupdated++;
2771 if (tp->t_srtt != 0) {
2773 * srtt is stored as fixed point with 5 bits after the
2774 * binary point (i.e., scaled by 8). The following magic
2775 * is equivalent to the smoothing algorithm in rfc793 with
2776 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2777 * point). Adjust rtt to origin 0.
2779 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2780 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2782 if ((tp->t_srtt += delta) <= 0)
2786 * We accumulate a smoothed rtt variance (actually, a
2787 * smoothed mean difference), then set the retransmit
2788 * timer to smoothed rtt + 4 times the smoothed variance.
2789 * rttvar is stored as fixed point with 4 bits after the
2790 * binary point (scaled by 16). The following is
2791 * equivalent to rfc793 smoothing with an alpha of .75
2792 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2793 * rfc793's wired-in beta.
2797 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2798 if ((tp->t_rttvar += delta) <= 0)
2800 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2801 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2804 * No rtt measurement yet - use the unsmoothed rtt.
2805 * Set the variance to half the rtt (so our first
2806 * retransmit happens at 3*rtt).
2808 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2809 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2810 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2816 * the retransmit should happen at rtt + 4 * rttvar.
2817 * Because of the way we do the smoothing, srtt and rttvar
2818 * will each average +1/2 tick of bias. When we compute
2819 * the retransmit timer, we want 1/2 tick of rounding and
2820 * 1 extra tick because of +-1/2 tick uncertainty in the
2821 * firing of the timer. The bias will give us exactly the
2822 * 1.5 tick we need. But, because the bias is
2823 * statistical, we have to test that we don't drop below
2824 * the minimum feasible timer (which is 2 ticks).
2826 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2827 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2830 * We received an ack for a packet that wasn't retransmitted;
2831 * it is probably safe to discard any error indications we've
2832 * received recently. This isn't quite right, but close enough
2833 * for now (a route might have failed after we sent a segment,
2834 * and the return path might not be symmetrical).
2836 tp->t_softerror = 0;
2840 * Determine a reasonable value for maxseg size.
2841 * If the route is known, check route for mtu.
2842 * If none, use an mss that can be handled on the outgoing
2843 * interface without forcing IP to fragment; if bigger than
2844 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2845 * to utilize large mbufs. If no route is found, route has no mtu,
2846 * or the destination isn't local, use a default, hopefully conservative
2847 * size (usually 512 or the default IP max size, but no more than the mtu
2848 * of the interface), as we can't discover anything about intervening
2849 * gateways or networks. We also initialize the congestion/slow start
2850 * window to be a single segment if the destination isn't local.
2851 * While looking at the routing entry, we also initialize other path-dependent
2852 * parameters from pre-set or cached values in the routing entry.
2854 * Also take into account the space needed for options that we
2855 * send regularly. Make maxseg shorter by that amount to assure
2856 * that we can send maxseg amount of data even when the options
2857 * are present. Store the upper limit of the length of options plus
2861 * In case of T/TCP, we call this routine during implicit connection
2862 * setup as well (offer = -1), to initialize maxseg from the cached
2865 * NOTE that this routine is only called when we process an incoming
2866 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2876 struct inpcb *inp = tp->t_inpcb;
2878 struct hc_metrics_lite metrics;
2879 int origoffer = offer;
2882 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2883 size_t min_protoh = isipv6 ?
2884 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2885 sizeof (struct tcpiphdr);
2887 const size_t min_protoh = sizeof(struct tcpiphdr);
2893 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags);
2894 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2898 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags);
2899 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2901 so = inp->inp_socket;
2904 * no route to sender, stay with default mss and return
2909 /* what have we got? */
2913 * Offer == 0 means that there was no MSS on the SYN
2914 * segment, in this case we use tcp_mssdflt.
2918 isipv6 ? tcp_v6mssdflt :
2925 * Offer == -1 means that we didn't receive SYN yet.
2931 * Prevent DoS attack with too small MSS. Round up
2932 * to at least minmss.
2934 offer = max(offer, tcp_minmss);
2936 * Sanity check: make sure that maxopd will be large
2937 * enough to allow some data on segments even if the
2938 * all the option space is used (40bytes). Otherwise
2939 * funny things may happen in tcp_output.
2941 offer = max(offer, 64);
2945 * rmx information is now retrieved from tcp_hostcache
2947 tcp_hc_get(&inp->inp_inc, &metrics);
2950 * if there's a discovered mtu int tcp hostcache, use it
2951 * else, use the link mtu.
2953 if (metrics.rmx_mtu)
2954 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2958 mss = maxmtu - min_protoh;
2959 if (!path_mtu_discovery &&
2960 !in6_localaddr(&inp->in6p_faddr))
2961 mss = min(mss, tcp_v6mssdflt);
2965 mss = maxmtu - min_protoh;
2966 if (!path_mtu_discovery &&
2967 !in_localaddr(inp->inp_faddr))
2968 mss = min(mss, tcp_mssdflt);
2971 mss = min(mss, offer);
2974 * maxopd stores the maximum length of data AND options
2975 * in a segment; maxseg is the amount of data in a normal
2976 * segment. We need to store this value (maxopd) apart
2977 * from maxseg, because now every segment carries options
2978 * and thus we normally have somewhat less data in segments.
2983 * origoffer==-1 indicates, that no segments were received yet.
2984 * In this case we just guess.
2986 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2988 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2989 mss -= TCPOLEN_TSTAMP_APPA;
2992 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2994 mss &= ~(MCLBYTES-1);
2997 mss = mss / MCLBYTES * MCLBYTES;
3002 * If there's a pipesize, change the socket buffer to that size,
3003 * don't change if sb_hiwat is different than default (then it
3004 * has been changed on purpose with setsockopt).
3005 * Make the socket buffers an integral number of mss units;
3006 * if the mss is larger than the socket buffer, decrease the mss.
3008 SOCKBUF_LOCK(&so->so_snd);
3009 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3010 bufsize = metrics.rmx_sendpipe;
3012 bufsize = so->so_snd.sb_hiwat;
3016 bufsize = roundup(bufsize, mss);
3017 if (bufsize > sb_max)
3019 if (bufsize > so->so_snd.sb_hiwat)
3020 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3022 SOCKBUF_UNLOCK(&so->so_snd);
3025 SOCKBUF_LOCK(&so->so_rcv);
3026 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3027 bufsize = metrics.rmx_recvpipe;
3029 bufsize = so->so_rcv.sb_hiwat;
3030 if (bufsize > mss) {
3031 bufsize = roundup(bufsize, mss);
3032 if (bufsize > sb_max)
3034 if (bufsize > so->so_rcv.sb_hiwat)
3035 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3037 SOCKBUF_UNLOCK(&so->so_rcv);
3039 * While we're here, check the others too
3041 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3043 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3044 tcpstat.tcps_usedrtt++;
3045 if (metrics.rmx_rttvar) {
3046 tp->t_rttvar = metrics.rmx_rttvar;
3047 tcpstat.tcps_usedrttvar++;
3049 /* default variation is +- 1 rtt */
3051 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3053 TCPT_RANGESET(tp->t_rxtcur,
3054 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3055 tp->t_rttmin, TCPTV_REXMTMAX);
3057 if (metrics.rmx_ssthresh) {
3059 * There's some sort of gateway or interface
3060 * buffer limit on the path. Use this to set
3061 * the slow start threshhold, but set the
3062 * threshold to no less than 2*mss.
3064 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3065 tcpstat.tcps_usedssthresh++;
3067 if (metrics.rmx_bandwidth)
3068 tp->snd_bandwidth = metrics.rmx_bandwidth;
3071 * Set the slow-start flight size depending on whether this
3072 * is a local network or not.
3074 * Extend this so we cache the cwnd too and retrieve it here.
3075 * Make cwnd even bigger than RFC3390 suggests but only if we
3076 * have previous experience with the remote host. Be careful
3077 * not make cwnd bigger than remote receive window or our own
3078 * send socket buffer. Maybe put some additional upper bound
3079 * on the retrieved cwnd. Should do incremental updates to
3080 * hostcache when cwnd collapses so next connection doesn't
3081 * overloads the path again.
3083 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3084 * We currently check only in syncache_socket for that.
3086 #define TCP_METRICS_CWND
3087 #ifdef TCP_METRICS_CWND
3088 if (metrics.rmx_cwnd)
3089 tp->snd_cwnd = max(mss,
3090 min(metrics.rmx_cwnd / 2,
3091 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3095 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3097 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3098 (!isipv6 && in_localaddr(inp->inp_faddr)))
3100 else if (in_localaddr(inp->inp_faddr))
3102 tp->snd_cwnd = mss * ss_fltsz_local;
3104 tp->snd_cwnd = mss * ss_fltsz;
3106 /* Check the interface for TSO capabilities. */
3107 if (mtuflags & CSUM_TSO)
3108 tp->t_flags |= TF_TSO;
3112 * Determine the MSS option to send on an outgoing SYN.
3116 struct in_conninfo *inc;
3123 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3126 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3130 mss = tcp_v6mssdflt;
3131 maxmtu = tcp_maxmtu6(inc, NULL);
3132 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3133 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3138 maxmtu = tcp_maxmtu(inc, NULL);
3139 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3140 min_protoh = sizeof(struct tcpiphdr);
3142 if (maxmtu && thcmtu)
3143 mss = min(maxmtu, thcmtu) - min_protoh;
3144 else if (maxmtu || thcmtu)
3145 mss = max(maxmtu, thcmtu) - min_protoh;
3152 * On a partial ack arrives, force the retransmission of the
3153 * next unacknowledged segment. Do not clear tp->t_dupacks.
3154 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3158 tcp_newreno_partial_ack(tp, th)
3162 tcp_seq onxt = tp->snd_nxt;
3163 u_long ocwnd = tp->snd_cwnd;
3165 callout_stop(tp->tt_rexmt);
3167 tp->snd_nxt = th->th_ack;
3169 * Set snd_cwnd to one segment beyond acknowledged offset.
3170 * (tp->snd_una has not yet been updated when this function is called.)
3172 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3173 tp->t_flags |= TF_ACKNOW;
3174 (void) tcp_output(tp);
3175 tp->snd_cwnd = ocwnd;
3176 if (SEQ_GT(onxt, tp->snd_nxt))
3179 * Partial window deflation. Relies on fact that tp->snd_una
3182 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3183 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3186 tp->snd_cwnd += tp->t_maxseg;
3190 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3191 * looking for a pcb in the listen state. Returns 0 otherwise.
3194 tcp_timewait(inp, to, th, m, tlen)
3205 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3207 const int isipv6 = 0;
3210 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */
3211 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3212 INP_LOCK_ASSERT(inp);
3215 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
3216 * still present. This is undesirable, but temporarily necessary
3217 * until we work out how to handle inpcb's who's timewait state has
3224 thflags = th->th_flags;
3227 * NOTE: for FIN_WAIT_2 (to be added later),
3228 * must validate sequence number before accepting RST
3232 * If the segment contains RST:
3233 * Drop the segment - see Stevens, vol. 2, p. 964 and
3236 if (thflags & TH_RST)
3240 /* PAWS not needed at the moment */
3242 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3243 * and it's less than ts_recent, drop it.
3245 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3246 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3247 if ((thflags & TH_ACK) == 0)
3252 * ts_recent is never updated because we never accept new segments.
3257 * If a new connection request is received
3258 * while in TIME_WAIT, drop the old connection
3259 * and start over if the sequence numbers
3260 * are above the previous ones.
3262 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3268 * Drop the the segment if it does not contain an ACK.
3270 if ((thflags & TH_ACK) == 0)
3274 * Reset the 2MSL timer if this is a duplicate FIN.
3276 if (thflags & TH_FIN) {
3277 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3278 if (seq + 1 == tw->rcv_nxt)
3279 tcp_timer_2msl_reset(tw, 1);
3283 * Acknowledge the segment if it has data or is not a duplicate ACK.
3285 if (thflags != TH_ACK || tlen != 0 ||
3286 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3287 tcp_twrespond(tw, TH_ACK);
3291 * Generate a RST, dropping incoming segment.
3292 * Make ACK acceptable to originator of segment.
3293 * Don't bother to respond if destination was broadcast/multicast.
3295 if (m->m_flags & (M_BCAST|M_MCAST))
3298 struct ip6_hdr *ip6;
3300 /* IPv6 anycast check is done at tcp6_input() */
3301 ip6 = mtod(m, struct ip6_hdr *);
3302 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3303 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3308 ip = mtod(m, struct ip *);
3309 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3310 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3311 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3312 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3315 if (thflags & TH_ACK) {
3317 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3319 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3321 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);