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"
40 #include <sys/param.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
44 #include <sys/proc.h> /* for proc0 declaration */
45 #include <sys/protosw.h>
46 #include <sys/signalvar.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
51 #include <sys/systm.h>
53 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
58 #include <net/route.h>
60 #include <netinet/in.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
66 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
67 #include <netinet/ip_var.h>
68 #include <netinet/ip_options.h>
69 #include <netinet/ip6.h>
70 #include <netinet/icmp6.h>
71 #include <netinet6/in6_pcb.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet6/tcp6_var.h>
80 #include <netinet/tcpip.h>
82 #include <netinet/tcp_debug.h>
86 #include <netipsec/ipsec.h>
87 #include <netipsec/ipsec6.h>
91 #include <netinet6/ipsec.h>
92 #include <netinet6/ipsec6.h>
93 #include <netkey/key.h>
96 #include <machine/in_cksum.h>
98 #include <security/mac/mac_framework.h>
100 static const int tcprexmtthresh = 3;
102 struct tcpstat tcpstat;
103 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
106 static int tcp_log_in_vain = 0;
107 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
108 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports");
110 static int blackhole = 0;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
112 &blackhole, 0, "Do not send RST on segments to closed ports");
114 int tcp_delack_enabled = 1;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
116 &tcp_delack_enabled, 0,
117 "Delay ACK to try and piggyback it onto a data packet");
119 static int drop_synfin = 0;
120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
121 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
123 static int tcp_do_rfc3042 = 1;
124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
125 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
127 static int tcp_do_rfc3390 = 1;
128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
130 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
132 static int tcp_insecure_rst = 0;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
134 &tcp_insecure_rst, 0,
135 "Follow the old (insecure) criteria for accepting RST packets");
137 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
138 "TCP Segment Reassembly Queue");
140 static int tcp_reass_maxseg = 0;
141 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
142 &tcp_reass_maxseg, 0,
143 "Global maximum number of TCP Segments in Reassembly Queue");
145 int tcp_reass_qsize = 0;
146 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
148 "Global number of TCP Segments currently in Reassembly Queue");
150 static int tcp_reass_maxqlen = 48;
151 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
152 &tcp_reass_maxqlen, 0,
153 "Maximum number of TCP Segments per individual Reassembly Queue");
155 static int tcp_reass_overflows = 0;
156 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
157 &tcp_reass_overflows, 0,
158 "Global number of TCP Segment Reassembly Queue Overflows");
160 int tcp_do_autorcvbuf = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
162 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
164 int tcp_autorcvbuf_inc = 16*1024;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
166 &tcp_autorcvbuf_inc, 0,
167 "Incrementor step size of automatic receive buffer");
169 int tcp_autorcvbuf_max = 256*1024;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
171 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
173 struct inpcbhead tcb;
174 #define tcb6 tcb /* for KAME src sync over BSD*'s */
175 struct inpcbinfo tcbinfo;
177 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
178 static int tcp_do_segment(struct mbuf *, struct tcphdr *,
179 struct socket *, struct tcpcb *, int, int);
180 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
181 struct tcpcb *, int, int);
182 static void tcp_pulloutofband(struct socket *,
183 struct tcphdr *, struct mbuf *, int);
184 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
186 static void tcp_xmit_timer(struct tcpcb *, int);
187 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
188 static int tcp_timewait(struct inpcb *, struct tcpopt *,
189 struct tcphdr *, struct mbuf *, int);
191 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
193 #define ND6_HINT(tp) \
195 if ((tp) && (tp)->t_inpcb && \
196 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
197 nd6_nud_hint(NULL, NULL, 0); \
204 * Indicate whether this ack should be delayed. We can delay the ack if
205 * - there is no delayed ack timer in progress and
206 * - our last ack wasn't a 0-sized window. We never want to delay
207 * the ack that opens up a 0-sized window and
208 * - delayed acks are enabled or
209 * - this is a half-synchronized T/TCP connection.
211 #define DELAY_ACK(tp) \
212 ((!tcp_timer_active(tp, TT_DELACK) && \
213 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
214 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
216 /* Initialize TCP reassembly queue */
218 tcp_reass_zone_change(void *tag)
221 tcp_reass_maxseg = nmbclusters / 16;
222 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
225 uma_zone_t tcp_reass_zone;
229 tcp_reass_maxseg = nmbclusters / 16;
230 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
232 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
233 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
234 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
235 EVENTHANDLER_REGISTER(nmbclusters_change,
236 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
240 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
243 struct tseg_qent *p = NULL;
244 struct tseg_qent *nq;
245 struct tseg_qent *te = NULL;
246 struct socket *so = tp->t_inpcb->inp_socket;
249 INP_LOCK_ASSERT(tp->t_inpcb);
252 * XXX: tcp_reass() is rather inefficient with its data structures
253 * and should be rewritten (see NetBSD for optimizations). While
254 * doing that it should move to its own file tcp_reass.c.
258 * Call with th==NULL after become established to
259 * force pre-ESTABLISHED data up to user socket.
265 * Limit the number of segments in the reassembly queue to prevent
266 * holding on to too many segments (and thus running out of mbufs).
267 * Make sure to let the missing segment through which caused this
268 * queue. Always keep one global queue entry spare to be able to
269 * process the missing segment.
271 if (th->th_seq != tp->rcv_nxt &&
272 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
273 tp->t_segqlen >= tcp_reass_maxqlen)) {
274 tcp_reass_overflows++;
275 tcpstat.tcps_rcvmemdrop++;
282 * Allocate a new queue entry. If we can't, or hit the zone limit
285 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
287 tcpstat.tcps_rcvmemdrop++;
296 * Find a segment which begins after this one does.
298 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
299 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
305 * If there is a preceding segment, it may provide some of
306 * our data already. If so, drop the data from the incoming
307 * segment. If it provides all of our data, drop us.
311 /* conversion to int (in i) handles seq wraparound */
312 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
315 tcpstat.tcps_rcvduppack++;
316 tcpstat.tcps_rcvdupbyte += *tlenp;
318 uma_zfree(tcp_reass_zone, te);
322 * Try to present any queued data
323 * at the left window edge to the user.
324 * This is needed after the 3-WHS
327 goto present; /* ??? */
334 tcpstat.tcps_rcvoopack++;
335 tcpstat.tcps_rcvoobyte += *tlenp;
338 * While we overlap succeeding segments trim them or,
339 * if they are completely covered, dequeue them.
342 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
345 if (i < q->tqe_len) {
346 q->tqe_th->th_seq += i;
352 nq = LIST_NEXT(q, tqe_q);
353 LIST_REMOVE(q, tqe_q);
355 uma_zfree(tcp_reass_zone, q);
361 /* Insert the new segment queue entry into place. */
364 te->tqe_len = *tlenp;
367 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
369 LIST_INSERT_AFTER(p, te, tqe_q);
374 * Present data to user, advancing rcv_nxt through
375 * completed sequence space.
377 if (!TCPS_HAVEESTABLISHED(tp->t_state))
379 q = LIST_FIRST(&tp->t_segq);
380 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
382 SOCKBUF_LOCK(&so->so_rcv);
384 tp->rcv_nxt += q->tqe_len;
385 flags = q->tqe_th->th_flags & TH_FIN;
386 nq = LIST_NEXT(q, tqe_q);
387 LIST_REMOVE(q, tqe_q);
388 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
391 sbappendstream_locked(&so->so_rcv, q->tqe_m);
392 uma_zfree(tcp_reass_zone, q);
396 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
398 sorwakeup_locked(so);
403 * TCP input routine, follows pages 65-76 of the
404 * protocol specification dated September, 1981 very closely.
408 tcp6_input(struct mbuf **mp, int *offp, int proto)
410 struct mbuf *m = *mp;
411 struct in6_ifaddr *ia6;
413 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
416 * draft-itojun-ipv6-tcp-to-anycast
417 * better place to put this in?
419 ia6 = ip6_getdstifaddr(m);
420 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
423 ip6 = mtod(m, struct ip6_hdr *);
424 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
425 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
435 tcp_input(struct mbuf *m, int off0)
438 struct ip *ip = NULL;
440 struct inpcb *inp = NULL;
441 struct tcpcb *tp = NULL;
442 struct socket *so = NULL;
448 int rstreason = 0; /* For badport_bandlim accounting purposes */
449 #ifdef IPFIREWALL_FORWARD
450 struct m_tag *fwd_tag;
453 struct ip6_hdr *ip6 = NULL;
455 char ip6buf[INET6_ADDRSTRLEN];
457 const int isipv6 = 0;
459 struct tcpopt to; /* options in this segment */
463 * The size of tcp_saveipgen must be the size of the max ip header,
466 u_char tcp_saveipgen[IP6_HDR_LEN];
467 struct tcphdr tcp_savetcp;
472 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
476 tcpstat.tcps_rcvtotal++;
480 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
481 ip6 = mtod(m, struct ip6_hdr *);
482 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
483 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
484 tcpstat.tcps_rcvbadsum++;
487 th = (struct tcphdr *)((caddr_t)ip6 + off0);
490 * Be proactive about unspecified IPv6 address in source.
491 * As we use all-zero to indicate unbounded/unconnected pcb,
492 * unspecified IPv6 address can be used to confuse us.
494 * Note that packets with unspecified IPv6 destination is
495 * already dropped in ip6_input.
497 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
502 th = NULL; /* XXX: avoid compiler warning */
506 * Get IP and TCP header together in first mbuf.
507 * Note: IP leaves IP header in first mbuf.
509 if (off0 > sizeof (struct ip)) {
510 ip_stripoptions(m, (struct mbuf *)0);
511 off0 = sizeof(struct ip);
513 if (m->m_len < sizeof (struct tcpiphdr)) {
514 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
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;
592 * If the drop_synfin option is enabled, drop all packets with
593 * both the SYN and FIN bits set. This prevents e.g. nmap from
594 * identifying the TCP/IP stack.
596 * This is a violation of the TCP specification.
598 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
602 * Convert TCP protocol specific fields to host format.
604 th->th_seq = ntohl(th->th_seq);
605 th->th_ack = ntohl(th->th_ack);
606 th->th_win = ntohs(th->th_win);
607 th->th_urp = ntohs(th->th_urp);
610 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
612 drop_hdrlen = off0 + off;
615 * Locate pcb for segment.
617 INP_INFO_WLOCK(&tcbinfo);
619 INP_INFO_WLOCK_ASSERT(&tcbinfo);
620 #ifdef IPFIREWALL_FORWARD
621 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
622 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
624 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
625 struct sockaddr_in *next_hop;
627 next_hop = (struct sockaddr_in *)(fwd_tag+1);
629 * Transparently forwarded. Pretend to be the destination.
630 * already got one like this?
632 inp = in_pcblookup_hash(&tcbinfo,
633 ip->ip_src, th->th_sport,
634 ip->ip_dst, th->th_dport,
635 0, m->m_pkthdr.rcvif);
637 /* It's new. Try to find the ambushing socket. */
638 inp = in_pcblookup_hash(&tcbinfo,
639 ip->ip_src, th->th_sport,
642 ntohs(next_hop->sin_port) :
647 /* Remove the tag from the packet. We don't need it anymore. */
648 m_tag_delete(m, fwd_tag);
650 #endif /* IPFIREWALL_FORWARD */
654 inp = in6_pcblookup_hash(&tcbinfo,
655 &ip6->ip6_src, th->th_sport,
656 &ip6->ip6_dst, th->th_dport,
661 inp = in_pcblookup_hash(&tcbinfo,
662 ip->ip_src, th->th_sport,
663 ip->ip_dst, th->th_dport,
668 #if defined(IPSEC) || defined(FAST_IPSEC)
670 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) {
672 ipsec6stat.in_polvio++;
677 if (inp != NULL && ipsec4_in_reject(m, inp)) {
679 ipsecstat.in_polvio++;
683 #endif /*IPSEC || FAST_IPSEC*/
686 * If the INPCB does not exist then all data in the incoming
687 * segment is discarded and an appropriate RST is sent back.
691 * Log communication attempts to ports that are not
694 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
695 tcp_log_in_vain == 2) {
697 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
699 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
703 ip6_sprintf(ip6buf, &ip6->ip6_dst));
707 ip6_sprintf(ip6buf, &ip6->ip6_src));
712 strcpy(dbuf, inet_ntoa(ip->ip_dst));
713 strcpy(sbuf, inet_ntoa(ip->ip_src));
716 "Connection attempt to TCP %s:%d "
717 "from %s:%d flags:0x%02x\n",
718 dbuf, ntohs(th->th_dport), sbuf,
719 ntohs(th->th_sport), thflags);
722 * When blackholing do not respond with a RST but
723 * completely ignore the segment and drop it.
725 if ((blackhole == 1 && (thflags & TH_SYN)) ||
729 rstreason = BANDLIM_RST_CLOSEDPORT;
734 /* Check the minimum TTL for socket. */
735 if (inp->inp_ip_minttl != 0) {
737 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
741 if (inp->inp_ip_minttl > ip->ip_ttl)
746 * A previous connection in TIMEWAIT state is supposed to catch
747 * stray or duplicate segments arriving late. If this segment
748 * was a legitimate new connection attempt the old INPCB gets
749 * removed and we can try again to find a listening socket.
751 if (inp->inp_vflag & INP_TIMEWAIT) {
752 if (thflags & TH_SYN)
753 tcp_dooptions(&to, optp, optlen, TO_SYN);
754 if (tcp_timewait(inp, &to, th, m, tlen))
756 /* tcp_timewait unlocks inp. */
757 INP_INFO_WUNLOCK(&tcbinfo);
761 * The TCPCB may no longer exist if the connection is winding
762 * down or it is in the CLOSED state. Either way we drop the
763 * segment and send an appropriate response.
768 rstreason = BANDLIM_RST_CLOSEDPORT;
771 if (tp->t_state == TCPS_CLOSED)
772 goto dropunlock; /* XXX: dropwithreset??? */
775 INP_LOCK_ASSERT(inp);
776 if (mac_check_inpcb_deliver(inp, m))
779 so = inp->inp_socket;
780 KASSERT(so != NULL, ("%s: so == NULL", __func__));
782 if (so->so_options & SO_DEBUG) {
783 ostate = tp->t_state;
785 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
787 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
792 * When the socket is accepting connections (the INPCB is in LISTEN
793 * state) we look into the SYN cache if this is a new connection
794 * attempt or the completion of a previous one.
796 if (so->so_options & SO_ACCEPTCONN) {
797 struct in_conninfo inc;
799 bzero(&inc, sizeof(inc));
800 inc.inc_isipv6 = isipv6;
803 inc.inc6_faddr = ip6->ip6_src;
804 inc.inc6_laddr = ip6->ip6_dst;
808 inc.inc_faddr = ip->ip_src;
809 inc.inc_laddr = ip->ip_dst;
811 inc.inc_fport = th->th_sport;
812 inc.inc_lport = th->th_dport;
815 * If the state is LISTEN then ignore segment if it contains
816 * a RST. If the segment contains an ACK then it is bad and
817 * send a RST. If it does not contain a SYN then it is not
818 * interesting; drop it.
820 * If the state is SYN_RECEIVED (syncache) and seg contains
821 * an ACK, but not for our SYN/ACK, send a RST. If the seg
822 * contains a RST, check the sequence number to see if it
823 * is a valid reset segment.
825 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
826 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
828 * Parse the TCP options here because
829 * syncookies need access to the reflected
832 tcp_dooptions(&to, optp, optlen, 0);
833 if (!syncache_expand(&inc, &to, th, &so, m)) {
835 * No syncache entry or ACK was not
836 * for our SYN/ACK. Send a RST.
838 rstreason = BANDLIM_RST_OPENPORT;
843 * We completed the 3-way handshake
844 * but could not allocate a socket
845 * either due to memory shortage,
846 * listen queue length limits or
847 * global socket limits.
849 rstreason = BANDLIM_UNLIMITED;
853 * Socket is created in state SYN_RECEIVED.
854 * Continue processing segment.
856 INP_UNLOCK(inp); /* listen socket */
858 INP_LOCK(inp); /* new connection */
861 * Process the segment and the data it
862 * contains. tcp_do_segment() consumes
863 * the mbuf chain and unlocks the inpcb.
864 * XXX: The potential return value of
865 * TIME_WAIT nuked is supposed to be
868 if (tcp_do_segment(m, th, so, tp,
870 goto findpcb; /* TIME_WAIT nuked */
873 if (thflags & TH_RST) {
874 syncache_chkrst(&inc, th);
877 if (thflags & TH_ACK) {
878 syncache_badack(&inc);
879 tcpstat.tcps_badsyn++;
880 rstreason = BANDLIM_RST_OPENPORT;
887 * Segment's flags are (SYN) or (SYN|FIN).
891 * If deprecated address is forbidden,
892 * we do not accept SYN to deprecated interface
893 * address to prevent any new inbound connection from
894 * getting established.
895 * When we do not accept SYN, we send a TCP RST,
896 * with deprecated source address (instead of dropping
897 * it). We compromise it as it is much better for peer
898 * to send a RST, and RST will be the final packet
901 * If we do not forbid deprecated addresses, we accept
902 * the SYN packet. RFC2462 does not suggest dropping
904 * If we decipher RFC2462 5.5.4, it says like this:
905 * 1. use of deprecated addr with existing
906 * communication is okay - "SHOULD continue to be
908 * 2. use of it with new communication:
909 * (2a) "SHOULD NOT be used if alternate address
910 * with sufficient scope is available"
911 * (2b) nothing mentioned otherwise.
912 * Here we fall into (2b) case as we have no choice in
913 * our source address selection - we must obey the peer.
915 * The wording in RFC2462 is confusing, and there are
916 * multiple description text for deprecated address
917 * handling - worse, they are not exactly the same.
918 * I believe 5.5.4 is the best one, so we follow 5.5.4.
920 if (isipv6 && !ip6_use_deprecated) {
921 struct in6_ifaddr *ia6;
923 if ((ia6 = ip6_getdstifaddr(m)) &&
924 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
927 rstreason = BANDLIM_RST_OPENPORT;
933 * Basic sanity checks on incoming SYN requests:
935 * Don't bother responding if the destination was a
936 * broadcast according to RFC1122 4.2.3.10, p. 104.
938 * If it is from this socket, drop it, it must be forged.
940 * Note that it is quite possible to receive unicast
941 * link-layer packets with a broadcast IP address. Use
942 * in_broadcast() to find them.
944 if (m->m_flags & (M_BCAST|M_MCAST))
948 if (th->th_dport == th->th_sport &&
949 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src))
951 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
952 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
956 if (th->th_dport == th->th_sport &&
957 ip->ip_dst.s_addr == ip->ip_src.s_addr)
959 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
960 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
961 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
962 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
966 * SYN appears to be valid. Create compressed TCP state
970 if (so->so_options & SO_DEBUG)
971 tcp_trace(TA_INPUT, ostate, tp,
972 (void *)tcp_saveipgen, &tcp_savetcp, 0);
974 tcp_dooptions(&to, optp, optlen, TO_SYN);
975 syncache_add(&inc, &to, th, inp, &so, m);
977 * Entry added to syncache and mbuf consumed.
978 * Everything unlocked already by syncache_add().
984 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or
985 * later state. tcp_do_segment() always consumes the mbuf chain
986 * and unlocks the inpcb.
988 if (tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen))
989 goto findpcb; /* XXX: TIME_WAIT was nuked. */
993 INP_INFO_WLOCK_ASSERT(&tcbinfo);
994 tcp_dropwithreset(m, th, tp, tlen, rstreason);
995 m = NULL; /* mbuf chain got consumed. */
997 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1000 INP_INFO_WUNLOCK(&tcbinfo);
1002 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
1009 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1010 struct tcpcb *tp, int drop_hdrlen, int tlen)
1012 int thflags, acked, ourfinisacked, needoutput = 0;
1014 int rstreason, todrop, win;
1020 * The size of tcp_saveipgen must be the size of the max ip header,
1023 u_char tcp_saveipgen[IP6_HDR_LEN];
1024 struct tcphdr tcp_savetcp;
1027 thflags = th->th_flags;
1029 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1030 INP_LOCK_ASSERT(tp->t_inpcb);
1031 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__));
1034 * Segment received on connection.
1035 * Reset idle time and keep-alive timer.
1037 tp->t_rcvtime = ticks;
1038 if (TCPS_HAVEESTABLISHED(tp->t_state))
1039 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1042 * Unscale the window into a 32-bit value.
1043 * This value is bogus for the TCPS_SYN_SENT state
1044 * and is overwritten later.
1046 tiwin = th->th_win << tp->snd_scale;
1049 * Parse options on any incoming segment.
1051 tcp_dooptions(&to, (u_char *)(th + 1),
1052 (th->th_off << 2) - sizeof(struct tcphdr),
1053 (thflags & TH_SYN) ? TO_SYN : 0);
1056 * If echoed timestamp is later than the current time,
1057 * fall back to non RFC1323 RTT calculation. Normalize
1058 * timestamp if syncookies were used when this connection
1061 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1062 to.to_tsecr -= tp->ts_offset;
1063 if (TSTMP_GT(to.to_tsecr, ticks))
1068 * Process options only when we get SYN/ACK back. The SYN case
1069 * for incoming connections is handled in tcp_syncache.
1070 * XXX this is traditional behavior, may need to be cleaned up.
1072 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1073 if ((to.to_flags & TOF_SCALE) &&
1074 (tp->t_flags & TF_REQ_SCALE)) {
1075 tp->t_flags |= TF_RCVD_SCALE;
1076 tp->snd_scale = to.to_wscale;
1077 tp->snd_wnd = th->th_win << tp->snd_scale;
1078 tiwin = tp->snd_wnd;
1080 if (to.to_flags & TOF_TS) {
1081 tp->t_flags |= TF_RCVD_TSTMP;
1082 tp->ts_recent = to.to_tsval;
1083 tp->ts_recent_age = ticks;
1085 /* Initial send window, already scaled. */
1086 tp->snd_wnd = th->th_win;
1087 if (to.to_flags & TOF_MSS)
1088 tcp_mss(tp, to.to_mss);
1089 if (tp->sack_enable) {
1090 if (!(to.to_flags & TOF_SACKPERM))
1091 tp->sack_enable = 0;
1093 tp->t_flags |= TF_SACK_PERMIT;
1099 * Header prediction: check for the two common cases
1100 * of a uni-directional data xfer. If the packet has
1101 * no control flags, is in-sequence, the window didn't
1102 * change and we're not retransmitting, it's a
1103 * candidate. If the length is zero and the ack moved
1104 * forward, we're the sender side of the xfer. Just
1105 * free the data acked & wake any higher level process
1106 * that was blocked waiting for space. If the length
1107 * is non-zero and the ack didn't move, we're the
1108 * receiver side. If we're getting packets in-order
1109 * (the reassembly queue is empty), add the data to
1110 * the socket buffer and note that we need a delayed ack.
1111 * Make sure that the hidden state-flags are also off.
1112 * Since we check for TCPS_ESTABLISHED above, it can only
1115 if (tp->t_state == TCPS_ESTABLISHED &&
1116 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1117 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1118 ((to.to_flags & TOF_TS) == 0 ||
1119 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1120 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1121 tp->snd_nxt == tp->snd_max) {
1124 * If last ACK falls within this segment's sequence numbers,
1125 * record the timestamp.
1126 * NOTE that the test is modified according to the latest
1127 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1129 if ((to.to_flags & TOF_TS) != 0 &&
1130 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1131 tp->ts_recent_age = ticks;
1132 tp->ts_recent = to.to_tsval;
1136 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1137 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1138 tp->snd_cwnd >= tp->snd_wnd &&
1139 ((!tcp_do_newreno && !tp->sack_enable &&
1140 tp->t_dupacks < tcprexmtthresh) ||
1141 ((tcp_do_newreno || tp->sack_enable) &&
1142 !IN_FASTRECOVERY(tp) &&
1143 (to.to_flags & TOF_SACK) == 0 &&
1144 TAILQ_EMPTY(&tp->snd_holes)))) {
1146 ("%s: headlocked", __func__));
1147 INP_INFO_WUNLOCK(&tcbinfo);
1150 * this is a pure ack for outstanding data.
1152 ++tcpstat.tcps_predack;
1154 * "bad retransmit" recovery
1156 if (tp->t_rxtshift == 1 &&
1157 ticks < tp->t_badrxtwin) {
1158 ++tcpstat.tcps_sndrexmitbad;
1159 tp->snd_cwnd = tp->snd_cwnd_prev;
1161 tp->snd_ssthresh_prev;
1162 tp->snd_recover = tp->snd_recover_prev;
1163 if (tp->t_flags & TF_WASFRECOVERY)
1164 ENTER_FASTRECOVERY(tp);
1165 tp->snd_nxt = tp->snd_max;
1166 tp->t_badrxtwin = 0;
1170 * Recalculate the transmit timer / rtt.
1172 * Some boxes send broken timestamp replies
1173 * during the SYN+ACK phase, ignore
1174 * timestamps of 0 or we could calculate a
1175 * huge RTT and blow up the retransmit timer.
1177 if ((to.to_flags & TOF_TS) != 0 &&
1179 if (!tp->t_rttlow ||
1180 tp->t_rttlow > ticks - to.to_tsecr)
1181 tp->t_rttlow = ticks - to.to_tsecr;
1183 ticks - to.to_tsecr + 1);
1184 } else if (tp->t_rtttime &&
1185 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1186 if (!tp->t_rttlow ||
1187 tp->t_rttlow > ticks - tp->t_rtttime)
1188 tp->t_rttlow = ticks - tp->t_rtttime;
1190 ticks - tp->t_rtttime);
1192 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1193 acked = th->th_ack - tp->snd_una;
1194 tcpstat.tcps_rcvackpack++;
1195 tcpstat.tcps_rcvackbyte += acked;
1196 sbdrop(&so->so_snd, acked);
1197 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1198 SEQ_LEQ(th->th_ack, tp->snd_recover))
1199 tp->snd_recover = th->th_ack - 1;
1200 tp->snd_una = th->th_ack;
1202 * pull snd_wl2 up to prevent seq wrap relative
1205 tp->snd_wl2 = th->th_ack;
1208 ND6_HINT(tp); /* some progress has been done */
1211 * If all outstanding data are acked, stop
1212 * retransmit timer, otherwise restart timer
1213 * using current (possibly backed-off) value.
1214 * If process is waiting for space,
1215 * wakeup/selwakeup/signal. If data
1216 * are ready to send, let tcp_output
1217 * decide between more output or persist.
1220 if (so->so_options & SO_DEBUG)
1221 tcp_trace(TA_INPUT, ostate, tp,
1222 (void *)tcp_saveipgen,
1226 if (tp->snd_una == tp->snd_max)
1227 tcp_timer_activate(tp, TT_REXMT, 0);
1228 else if (!tcp_timer_active(tp, TT_PERSIST))
1229 tcp_timer_activate(tp, TT_REXMT,
1233 if (so->so_snd.sb_cc)
1234 (void) tcp_output(tp);
1237 } else if (th->th_ack == tp->snd_una &&
1238 LIST_EMPTY(&tp->t_segq) &&
1239 tlen <= sbspace(&so->so_rcv)) {
1240 int newsize = 0; /* automatic sockbuf scaling */
1242 KASSERT(headlocked, ("%s: headlocked", __func__));
1243 INP_INFO_WUNLOCK(&tcbinfo);
1246 * this is a pure, in-sequence data packet
1247 * with nothing on the reassembly queue and
1248 * we have enough buffer space to take it.
1250 /* Clean receiver SACK report if present */
1251 if (tp->sack_enable && tp->rcv_numsacks)
1252 tcp_clean_sackreport(tp);
1253 ++tcpstat.tcps_preddat;
1254 tp->rcv_nxt += tlen;
1256 * Pull snd_wl1 up to prevent seq wrap relative to
1259 tp->snd_wl1 = th->th_seq;
1261 * Pull rcv_up up to prevent seq wrap relative to
1264 tp->rcv_up = tp->rcv_nxt;
1265 tcpstat.tcps_rcvpack++;
1266 tcpstat.tcps_rcvbyte += tlen;
1267 ND6_HINT(tp); /* some progress has been done */
1269 if (so->so_options & SO_DEBUG)
1270 tcp_trace(TA_INPUT, ostate, tp,
1271 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1274 * Automatic sizing of receive socket buffer. Often the send
1275 * buffer size is not optimally adjusted to the actual network
1276 * conditions at hand (delay bandwidth product). Setting the
1277 * buffer size too small limits throughput on links with high
1278 * bandwidth and high delay (eg. trans-continental/oceanic links).
1280 * On the receive side the socket buffer memory is only rarely
1281 * used to any significant extent. This allows us to be much
1282 * more aggressive in scaling the receive socket buffer. For
1283 * the case that the buffer space is actually used to a large
1284 * extent and we run out of kernel memory we can simply drop
1285 * the new segments; TCP on the sender will just retransmit it
1286 * later. Setting the buffer size too big may only consume too
1287 * much kernel memory if the application doesn't read() from
1288 * the socket or packet loss or reordering makes use of the
1291 * The criteria to step up the receive buffer one notch are:
1292 * 1. the number of bytes received during the time it takes
1293 * one timestamp to be reflected back to us (the RTT);
1294 * 2. received bytes per RTT is within seven eighth of the
1295 * current socket buffer size;
1296 * 3. receive buffer size has not hit maximal automatic size;
1298 * This algorithm does one step per RTT at most and only if
1299 * we receive a bulk stream w/o packet losses or reorderings.
1300 * Shrinking the buffer during idle times is not necessary as
1301 * it doesn't consume any memory when idle.
1303 * TODO: Only step up if the application is actually serving
1304 * the buffer to better manage the socket buffer resources.
1306 if (tcp_do_autorcvbuf &&
1308 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1309 if (to.to_tsecr > tp->rfbuf_ts &&
1310 to.to_tsecr - tp->rfbuf_ts < hz) {
1312 (so->so_rcv.sb_hiwat / 8 * 7) &&
1313 so->so_rcv.sb_hiwat <
1314 tcp_autorcvbuf_max) {
1316 min(so->so_rcv.sb_hiwat +
1318 tcp_autorcvbuf_max);
1320 /* Start over with next RTT. */
1324 tp->rfbuf_cnt += tlen; /* add up */
1327 /* Add data to socket buffer. */
1328 SOCKBUF_LOCK(&so->so_rcv);
1329 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1333 * Set new socket buffer size.
1334 * Give up when limit is reached.
1337 if (!sbreserve_locked(&so->so_rcv,
1338 newsize, so, curthread))
1339 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1340 m_adj(m, drop_hdrlen); /* delayed header drop */
1341 sbappendstream_locked(&so->so_rcv, m);
1343 sorwakeup_locked(so);
1344 if (DELAY_ACK(tp)) {
1345 tp->t_flags |= TF_DELACK;
1347 tp->t_flags |= TF_ACKNOW;
1355 * Calculate amount of space in receive window,
1356 * and then do TCP input processing.
1357 * Receive window is amount of space in rcv queue,
1358 * but not less than advertised window.
1360 win = sbspace(&so->so_rcv);
1363 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1365 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1369 switch (tp->t_state) {
1372 * If the state is SYN_RECEIVED:
1373 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1375 case TCPS_SYN_RECEIVED:
1376 if ((thflags & TH_ACK) &&
1377 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1378 SEQ_GT(th->th_ack, tp->snd_max))) {
1379 rstreason = BANDLIM_RST_OPENPORT;
1385 * If the state is SYN_SENT:
1386 * if seg contains an ACK, but not for our SYN, drop the input.
1387 * if seg contains a RST, then drop the connection.
1388 * if seg does not contain SYN, then drop it.
1389 * Otherwise this is an acceptable SYN segment
1390 * initialize tp->rcv_nxt and tp->irs
1391 * if seg contains ack then advance tp->snd_una
1392 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1393 * arrange for segment to be acked (eventually)
1394 * continue processing rest of data/controls, beginning with URG
1397 if ((thflags & TH_ACK) &&
1398 (SEQ_LEQ(th->th_ack, tp->iss) ||
1399 SEQ_GT(th->th_ack, tp->snd_max))) {
1400 rstreason = BANDLIM_UNLIMITED;
1403 if (thflags & TH_RST) {
1404 if (thflags & TH_ACK) {
1405 KASSERT(headlocked, ("%s: after_listen: "
1406 "tcp_drop.2: head not locked", __func__));
1407 tp = tcp_drop(tp, ECONNREFUSED);
1411 if ((thflags & TH_SYN) == 0)
1414 tp->irs = th->th_seq;
1416 if (thflags & TH_ACK) {
1417 tcpstat.tcps_connects++;
1421 mac_set_socket_peer_from_mbuf(m, so);
1424 /* Do window scaling on this connection? */
1425 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1426 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1427 tp->rcv_scale = tp->request_r_scale;
1429 tp->rcv_adv += tp->rcv_wnd;
1430 tp->snd_una++; /* SYN is acked */
1432 * If there's data, delay ACK; if there's also a FIN
1433 * ACKNOW will be turned on later.
1435 if (DELAY_ACK(tp) && tlen != 0)
1436 tcp_timer_activate(tp, TT_DELACK,
1439 tp->t_flags |= TF_ACKNOW;
1441 * Received <SYN,ACK> in SYN_SENT[*] state.
1443 * SYN_SENT --> ESTABLISHED
1444 * SYN_SENT* --> FIN_WAIT_1
1446 tp->t_starttime = ticks;
1447 if (tp->t_flags & TF_NEEDFIN) {
1448 tp->t_state = TCPS_FIN_WAIT_1;
1449 tp->t_flags &= ~TF_NEEDFIN;
1452 tp->t_state = TCPS_ESTABLISHED;
1453 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
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 tcp_timer_activate(tp, TT_REXMT, 0);
1468 tp->t_state = TCPS_SYN_RECEIVED;
1471 KASSERT(headlocked, ("%s: trimthenstep6: head not locked",
1473 INP_LOCK_ASSERT(tp->t_inpcb);
1476 * Advance th->th_seq to correspond to first data byte.
1477 * If data, trim to stay within window,
1478 * dropping FIN if necessary.
1481 if (tlen > tp->rcv_wnd) {
1482 todrop = tlen - tp->rcv_wnd;
1486 tcpstat.tcps_rcvpackafterwin++;
1487 tcpstat.tcps_rcvbyteafterwin += todrop;
1489 tp->snd_wl1 = th->th_seq - 1;
1490 tp->rcv_up = th->th_seq;
1492 * Client side of transaction: already sent SYN and data.
1493 * If the remote host used T/TCP to validate the SYN,
1494 * our data will be ACK'd; if so, enter normal data segment
1495 * processing in the middle of step 5, ack processing.
1496 * Otherwise, goto step 6.
1498 if (thflags & TH_ACK)
1504 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1505 * do normal processing.
1507 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1511 case TCPS_TIME_WAIT:
1512 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait",
1514 break; /* continue normal processing */
1518 * States other than LISTEN or SYN_SENT.
1519 * First check the RST flag and sequence number since reset segments
1520 * are exempt from the timestamp and connection count tests. This
1521 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1522 * below which allowed reset segments in half the sequence space
1523 * to fall though and be processed (which gives forged reset
1524 * segments with a random sequence number a 50 percent chance of
1525 * killing a connection).
1526 * Then check timestamp, if present.
1527 * Then check the connection count, if present.
1528 * Then check that at least some bytes of segment are within
1529 * receive window. If segment begins before rcv_nxt,
1530 * drop leading data (and SYN); if nothing left, just ack.
1533 * If the RST bit is set, check the sequence number to see
1534 * if this is a valid reset segment.
1536 * In all states except SYN-SENT, all reset (RST) segments
1537 * are validated by checking their SEQ-fields. A reset is
1538 * valid if its sequence number is in the window.
1539 * Note: this does not take into account delayed ACKs, so
1540 * we should test against last_ack_sent instead of rcv_nxt.
1541 * The sequence number in the reset segment is normally an
1542 * echo of our outgoing acknowlegement numbers, but some hosts
1543 * send a reset with the sequence number at the rightmost edge
1544 * of our receive window, and we have to handle this case.
1545 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1546 * that brute force RST attacks are possible. To combat this,
1547 * we use a much stricter check while in the ESTABLISHED state,
1548 * only accepting RSTs where the sequence number is equal to
1549 * last_ack_sent. In all other states (the states in which a
1550 * RST is more likely), the more permissive check is used.
1551 * If we have multiple segments in flight, the intial reset
1552 * segment sequence numbers will be to the left of last_ack_sent,
1553 * but they will eventually catch up.
1554 * In any case, it never made sense to trim reset segments to
1555 * fit the receive window since RFC 1122 says:
1556 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1558 * A TCP SHOULD allow a received RST segment to include data.
1561 * It has been suggested that a RST segment could contain
1562 * ASCII text that encoded and explained the cause of the
1563 * RST. No standard has yet been established for such
1566 * If the reset segment passes the sequence number test examine
1568 * SYN_RECEIVED STATE:
1569 * If passive open, return to LISTEN state.
1570 * If active open, inform user that connection was refused.
1571 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1572 * Inform user that connection was reset, and close tcb.
1573 * CLOSING, LAST_ACK STATES:
1576 * Drop the segment - see Stevens, vol. 2, p. 964 and
1579 if (thflags & TH_RST) {
1580 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1581 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1582 switch (tp->t_state) {
1584 case TCPS_SYN_RECEIVED:
1585 so->so_error = ECONNREFUSED;
1588 case TCPS_ESTABLISHED:
1589 if (tcp_insecure_rst == 0 &&
1590 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
1591 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
1592 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1593 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
1594 tcpstat.tcps_badrst++;
1597 case TCPS_FIN_WAIT_1:
1598 case TCPS_FIN_WAIT_2:
1599 case TCPS_CLOSE_WAIT:
1600 so->so_error = ECONNRESET;
1602 tp->t_state = TCPS_CLOSED;
1603 tcpstat.tcps_drops++;
1604 KASSERT(headlocked, ("%s: trimthenstep6: "
1605 "tcp_close: head not locked", __func__));
1611 KASSERT(headlocked, ("%s: trimthenstep6: "
1612 "tcp_close.2: head not locked", __func__));
1616 case TCPS_TIME_WAIT:
1617 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1618 ("%s: timewait", __func__));
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, ("%s: trimthenstep6: tcp_close.3: head "
1721 "not locked", __func__));
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, ("%s: timewait",
1745 if (thflags & TH_SYN &&
1746 tp->t_state == TCPS_TIME_WAIT &&
1747 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1748 KASSERT(headlocked, ("%s: trimthenstep6: "
1749 "tcp_close.4: head not locked", __func__));
1751 /* XXX: Shouldn't be possible. */
1755 * If window is closed can only take segments at
1756 * window edge, and have to drop data and PUSH from
1757 * incoming segments. Continue processing, but
1758 * remember to ack. Otherwise, drop segment
1761 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1762 tp->t_flags |= TF_ACKNOW;
1763 tcpstat.tcps_rcvwinprobe++;
1767 tcpstat.tcps_rcvbyteafterwin += todrop;
1770 thflags &= ~(TH_PUSH|TH_FIN);
1774 * If last ACK falls within this segment's sequence numbers,
1775 * record its timestamp.
1777 * 1) That the test incorporates suggestions from the latest
1778 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1779 * 2) That updating only on newer timestamps interferes with
1780 * our earlier PAWS tests, so this check should be solely
1781 * predicated on the sequence space of this segment.
1782 * 3) That we modify the segment boundary check to be
1783 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1784 * instead of RFC1323's
1785 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1786 * This modified check allows us to overcome RFC1323's
1787 * limitations as described in Stevens TCP/IP Illustrated
1788 * Vol. 2 p.869. In such cases, we can still calculate the
1789 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1791 if ((to.to_flags & TOF_TS) != 0 &&
1792 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1793 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1794 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1795 tp->ts_recent_age = ticks;
1796 tp->ts_recent = to.to_tsval;
1800 * If a SYN is in the window, then this is an
1801 * error and we send an RST and drop the connection.
1803 if (thflags & TH_SYN) {
1804 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: "
1805 "head not locked", __func__));
1806 tp = tcp_drop(tp, ECONNRESET);
1807 rstreason = BANDLIM_UNLIMITED;
1812 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1813 * flag is on (half-synchronized state), then queue data for
1814 * later processing; else drop segment and return.
1816 if ((thflags & TH_ACK) == 0) {
1817 if (tp->t_state == TCPS_SYN_RECEIVED ||
1818 (tp->t_flags & TF_NEEDSYN))
1820 else if (tp->t_flags & TF_ACKNOW)
1829 switch (tp->t_state) {
1832 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1833 * ESTABLISHED state and continue processing.
1834 * The ACK was checked above.
1836 case TCPS_SYN_RECEIVED:
1838 tcpstat.tcps_connects++;
1840 /* Do window scaling? */
1841 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1842 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1843 tp->rcv_scale = tp->request_r_scale;
1844 tp->snd_wnd = tiwin;
1848 * SYN-RECEIVED -> ESTABLISHED
1849 * SYN-RECEIVED* -> FIN-WAIT-1
1851 tp->t_starttime = ticks;
1852 if (tp->t_flags & TF_NEEDFIN) {
1853 tp->t_state = TCPS_FIN_WAIT_1;
1854 tp->t_flags &= ~TF_NEEDFIN;
1856 tp->t_state = TCPS_ESTABLISHED;
1857 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1860 * If segment contains data or ACK, will call tcp_reass()
1861 * later; if not, do so now to pass queued data to user.
1863 if (tlen == 0 && (thflags & TH_FIN) == 0)
1864 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1866 tp->snd_wl1 = th->th_seq - 1;
1870 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1871 * ACKs. If the ack is in the range
1872 * tp->snd_una < th->th_ack <= tp->snd_max
1873 * then advance tp->snd_una to th->th_ack and drop
1874 * data from the retransmission queue. If this ACK reflects
1875 * more up to date window information we update our window information.
1877 case TCPS_ESTABLISHED:
1878 case TCPS_FIN_WAIT_1:
1879 case TCPS_FIN_WAIT_2:
1880 case TCPS_CLOSE_WAIT:
1883 case TCPS_TIME_WAIT:
1884 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait",
1886 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1887 tcpstat.tcps_rcvacktoomuch++;
1890 if (tp->sack_enable &&
1891 ((to.to_flags & TOF_SACK) ||
1892 !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 (!tcp_timer_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 tcp_timer_activate(tp, TT_REXMT, 0);
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 ("%s: tp->snd_limited too big",
1992 tp->snd_cwnd = tp->snd_ssthresh +
1994 (tp->t_dupacks - tp->snd_limited);
1995 if (SEQ_GT(onxt, tp->snd_nxt))
1998 } else if (tcp_do_rfc3042) {
1999 u_long oldcwnd = tp->snd_cwnd;
2000 tcp_seq oldsndmax = tp->snd_max;
2003 KASSERT(tp->t_dupacks == 1 ||
2005 ("%s: dupacks not 1 or 2",
2007 if (tp->t_dupacks == 1)
2008 tp->snd_limited = 0;
2010 (tp->snd_nxt - tp->snd_una) +
2011 (tp->t_dupacks - tp->snd_limited) *
2013 (void) tcp_output(tp);
2014 sent = tp->snd_max - oldsndmax;
2015 if (sent > tp->t_maxseg) {
2016 KASSERT((tp->t_dupacks == 2 &&
2017 tp->snd_limited == 0) ||
2018 (sent == tp->t_maxseg + 1 &&
2019 tp->t_flags & TF_SENTFIN),
2020 ("%s: sent too much",
2022 tp->snd_limited = 2;
2023 } else if (sent > 0)
2025 tp->snd_cwnd = oldcwnd;
2033 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2034 ("%s: th_ack <= snd_una", __func__));
2037 * If the congestion window was inflated to account
2038 * for the other side's cached packets, retract it.
2040 if (tcp_do_newreno || tp->sack_enable) {
2041 if (IN_FASTRECOVERY(tp)) {
2042 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2043 if (tp->sack_enable)
2044 tcp_sack_partialack(tp, th);
2046 tcp_newreno_partial_ack(tp, th);
2049 * Out of fast recovery.
2050 * Window inflation should have left us
2051 * with approximately snd_ssthresh
2053 * But in case we would be inclined to
2054 * send a burst, better to do it via
2055 * the slow start mechanism.
2057 if (SEQ_GT(th->th_ack +
2060 tp->snd_cwnd = tp->snd_max -
2064 tp->snd_cwnd = tp->snd_ssthresh;
2068 if (tp->t_dupacks >= tcprexmtthresh &&
2069 tp->snd_cwnd > tp->snd_ssthresh)
2070 tp->snd_cwnd = tp->snd_ssthresh;
2074 * If we reach this point, ACK is not a duplicate,
2075 * i.e., it ACKs something we sent.
2077 if (tp->t_flags & TF_NEEDSYN) {
2079 * T/TCP: Connection was half-synchronized, and our
2080 * SYN has been ACK'd (so connection is now fully
2081 * synchronized). Go to non-starred state,
2082 * increment snd_una for ACK of SYN, and check if
2083 * we can do window scaling.
2085 tp->t_flags &= ~TF_NEEDSYN;
2087 /* Do window scaling? */
2088 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2089 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2090 tp->rcv_scale = tp->request_r_scale;
2091 /* Send window already scaled. */
2096 KASSERT(headlocked, ("%s: process_ACK: head not locked",
2098 INP_LOCK_ASSERT(tp->t_inpcb);
2100 acked = th->th_ack - tp->snd_una;
2101 tcpstat.tcps_rcvackpack++;
2102 tcpstat.tcps_rcvackbyte += acked;
2105 * If we just performed our first retransmit, and the ACK
2106 * arrives within our recovery window, then it was a mistake
2107 * to do the retransmit in the first place. Recover our
2108 * original cwnd and ssthresh, and proceed to transmit where
2111 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2112 ++tcpstat.tcps_sndrexmitbad;
2113 tp->snd_cwnd = tp->snd_cwnd_prev;
2114 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2115 tp->snd_recover = tp->snd_recover_prev;
2116 if (tp->t_flags & TF_WASFRECOVERY)
2117 ENTER_FASTRECOVERY(tp);
2118 tp->snd_nxt = tp->snd_max;
2119 tp->t_badrxtwin = 0; /* XXX probably not required */
2123 * If we have a timestamp reply, update smoothed
2124 * round trip time. If no timestamp is present but
2125 * transmit timer is running and timed sequence
2126 * number was acked, update smoothed round trip time.
2127 * Since we now have an rtt measurement, cancel the
2128 * timer backoff (cf., Phil Karn's retransmit alg.).
2129 * Recompute the initial retransmit timer.
2131 * Some boxes send broken timestamp replies
2132 * during the SYN+ACK phase, ignore
2133 * timestamps of 0 or we could calculate a
2134 * huge RTT and blow up the retransmit timer.
2136 if ((to.to_flags & TOF_TS) != 0 &&
2138 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2139 tp->t_rttlow = ticks - to.to_tsecr;
2140 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2141 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2142 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2143 tp->t_rttlow = ticks - tp->t_rtttime;
2144 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2146 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2149 * If all outstanding data is acked, stop retransmit
2150 * timer and remember to restart (more output or persist).
2151 * If there is more data to be acked, restart retransmit
2152 * timer, using current (possibly backed-off) value.
2154 if (th->th_ack == tp->snd_max) {
2155 tcp_timer_activate(tp, TT_REXMT, 0);
2157 } else if (!tcp_timer_active(tp, TT_PERSIST))
2158 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2161 * If no data (only SYN) was ACK'd,
2162 * skip rest of ACK processing.
2168 * When new data is acked, open the congestion window.
2169 * If the window gives us less than ssthresh packets
2170 * in flight, open exponentially (maxseg per packet).
2171 * Otherwise open linearly: maxseg per window
2172 * (maxseg^2 / cwnd per packet).
2174 if ((!tcp_do_newreno && !tp->sack_enable) ||
2175 !IN_FASTRECOVERY(tp)) {
2176 u_int cw = tp->snd_cwnd;
2177 u_int incr = tp->t_maxseg;
2178 if (cw > tp->snd_ssthresh)
2179 incr = incr * incr / cw;
2180 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2182 SOCKBUF_LOCK(&so->so_snd);
2183 if (acked > so->so_snd.sb_cc) {
2184 tp->snd_wnd -= so->so_snd.sb_cc;
2185 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2188 sbdrop_locked(&so->so_snd, acked);
2189 tp->snd_wnd -= acked;
2192 sowwakeup_locked(so);
2193 /* detect una wraparound */
2194 if ((tcp_do_newreno || tp->sack_enable) &&
2195 !IN_FASTRECOVERY(tp) &&
2196 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2197 SEQ_LEQ(th->th_ack, tp->snd_recover))
2198 tp->snd_recover = th->th_ack - 1;
2199 if ((tcp_do_newreno || tp->sack_enable) &&
2200 IN_FASTRECOVERY(tp) &&
2201 SEQ_GEQ(th->th_ack, tp->snd_recover))
2202 EXIT_FASTRECOVERY(tp);
2203 tp->snd_una = th->th_ack;
2204 if (tp->sack_enable) {
2205 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2206 tp->snd_recover = tp->snd_una;
2208 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2209 tp->snd_nxt = tp->snd_una;
2211 switch (tp->t_state) {
2214 * In FIN_WAIT_1 STATE in addition to the processing
2215 * for the ESTABLISHED state if our FIN is now acknowledged
2216 * then enter FIN_WAIT_2.
2218 case TCPS_FIN_WAIT_1:
2219 if (ourfinisacked) {
2221 * If we can't receive any more
2222 * data, then closing user can proceed.
2223 * Starting the timer is contrary to the
2224 * specification, but if we don't get a FIN
2225 * we'll hang forever.
2228 * we should release the tp also, and use a
2231 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2234 soisdisconnected(so);
2235 timeout = (tcp_fast_finwait2_recycle) ?
2236 tcp_finwait2_timeout : tcp_maxidle;
2237 tcp_timer_activate(tp, TT_2MSL, timeout);
2239 tp->t_state = TCPS_FIN_WAIT_2;
2244 * In CLOSING STATE in addition to the processing for
2245 * the ESTABLISHED state if the ACK acknowledges our FIN
2246 * then enter the TIME-WAIT state, otherwise ignore
2250 if (ourfinisacked) {
2251 KASSERT(headlocked, ("%s: process_ACK: "
2252 "head not locked", __func__));
2254 INP_INFO_WUNLOCK(&tcbinfo);
2262 * In LAST_ACK, we may still be waiting for data to drain
2263 * and/or to be acked, as well as for the ack of our FIN.
2264 * If our FIN is now acknowledged, delete the TCB,
2265 * enter the closed state and return.
2268 if (ourfinisacked) {
2269 KASSERT(headlocked, ("%s: process_ACK: "
2270 "tcp_close: head not locked", __func__));
2277 * In TIME_WAIT state the only thing that should arrive
2278 * is a retransmission of the remote FIN. Acknowledge
2279 * it and restart the finack timer.
2281 case TCPS_TIME_WAIT:
2282 KASSERT(tp->t_state != TCPS_TIME_WAIT,
2283 ("%s: timewait", __func__));
2284 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl);
2290 KASSERT(headlocked, ("%s: step6: head not locked", __func__));
2291 INP_LOCK_ASSERT(tp->t_inpcb);
2294 * Update window information.
2295 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2297 if ((thflags & TH_ACK) &&
2298 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2299 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2300 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2301 /* keep track of pure window updates */
2303 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2304 tcpstat.tcps_rcvwinupd++;
2305 tp->snd_wnd = tiwin;
2306 tp->snd_wl1 = th->th_seq;
2307 tp->snd_wl2 = th->th_ack;
2308 if (tp->snd_wnd > tp->max_sndwnd)
2309 tp->max_sndwnd = tp->snd_wnd;
2314 * Process segments with URG.
2316 if ((thflags & TH_URG) && th->th_urp &&
2317 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2319 * This is a kludge, but if we receive and accept
2320 * random urgent pointers, we'll crash in
2321 * soreceive. It's hard to imagine someone
2322 * actually wanting to send this much urgent data.
2324 SOCKBUF_LOCK(&so->so_rcv);
2325 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2326 th->th_urp = 0; /* XXX */
2327 thflags &= ~TH_URG; /* XXX */
2328 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2329 goto dodata; /* XXX */
2332 * If this segment advances the known urgent pointer,
2333 * then mark the data stream. This should not happen
2334 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2335 * a FIN has been received from the remote side.
2336 * In these states we ignore the URG.
2338 * According to RFC961 (Assigned Protocols),
2339 * the urgent pointer points to the last octet
2340 * of urgent data. We continue, however,
2341 * to consider it to indicate the first octet
2342 * of data past the urgent section as the original
2343 * spec states (in one of two places).
2345 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2346 tp->rcv_up = th->th_seq + th->th_urp;
2347 so->so_oobmark = so->so_rcv.sb_cc +
2348 (tp->rcv_up - tp->rcv_nxt) - 1;
2349 if (so->so_oobmark == 0)
2350 so->so_rcv.sb_state |= SBS_RCVATMARK;
2352 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2354 SOCKBUF_UNLOCK(&so->so_rcv);
2356 * Remove out of band data so doesn't get presented to user.
2357 * This can happen independent of advancing the URG pointer,
2358 * but if two URG's are pending at once, some out-of-band
2359 * data may creep in... ick.
2361 if (th->th_urp <= (u_long)tlen &&
2362 !(so->so_options & SO_OOBINLINE)) {
2363 /* hdr drop is delayed */
2364 tcp_pulloutofband(so, th, m, drop_hdrlen);
2368 * If no out of band data is expected,
2369 * pull receive urgent pointer along
2370 * with the receive window.
2372 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2373 tp->rcv_up = tp->rcv_nxt;
2376 KASSERT(headlocked, ("%s: dodata: head not locked", __func__));
2377 INP_LOCK_ASSERT(tp->t_inpcb);
2380 * Process the segment text, merging it into the TCP sequencing queue,
2381 * and arranging for acknowledgment of receipt if necessary.
2382 * This process logically involves adjusting tp->rcv_wnd as data
2383 * is presented to the user (this happens in tcp_usrreq.c,
2384 * case PRU_RCVD). If a FIN has already been received on this
2385 * connection then we just ignore the text.
2387 if ((tlen || (thflags & TH_FIN)) &&
2388 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2389 tcp_seq save_start = th->th_seq;
2390 tcp_seq save_end = th->th_seq + tlen;
2391 m_adj(m, drop_hdrlen); /* delayed header drop */
2393 * Insert segment which includes th into TCP reassembly queue
2394 * with control block tp. Set thflags to whether reassembly now
2395 * includes a segment with FIN. This handles the common case
2396 * inline (segment is the next to be received on an established
2397 * connection, and the queue is empty), avoiding linkage into
2398 * and removal from the queue and repetition of various
2400 * Set DELACK for segments received in order, but ack
2401 * immediately when segments are out of order (so
2402 * fast retransmit can work).
2404 if (th->th_seq == tp->rcv_nxt &&
2405 LIST_EMPTY(&tp->t_segq) &&
2406 TCPS_HAVEESTABLISHED(tp->t_state)) {
2408 tp->t_flags |= TF_DELACK;
2410 tp->t_flags |= TF_ACKNOW;
2411 tp->rcv_nxt += tlen;
2412 thflags = th->th_flags & TH_FIN;
2413 tcpstat.tcps_rcvpack++;
2414 tcpstat.tcps_rcvbyte += tlen;
2416 SOCKBUF_LOCK(&so->so_rcv);
2417 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2420 sbappendstream_locked(&so->so_rcv, m);
2421 sorwakeup_locked(so);
2423 thflags = tcp_reass(tp, th, &tlen, m);
2424 tp->t_flags |= TF_ACKNOW;
2426 if (tlen > 0 && tp->sack_enable)
2427 tcp_update_sack_list(tp, save_start, save_end);
2430 * Note the amount of data that peer has sent into
2431 * our window, in order to estimate the sender's
2435 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2443 * If FIN is received ACK the FIN and let the user know
2444 * that the connection is closing.
2446 if (thflags & TH_FIN) {
2447 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2450 * If connection is half-synchronized
2451 * (ie NEEDSYN flag on) then delay ACK,
2452 * so it may be piggybacked when SYN is sent.
2453 * Otherwise, since we received a FIN then no
2454 * more input can be expected, send ACK now.
2456 if (tp->t_flags & TF_NEEDSYN)
2457 tp->t_flags |= TF_DELACK;
2459 tp->t_flags |= TF_ACKNOW;
2462 switch (tp->t_state) {
2465 * In SYN_RECEIVED and ESTABLISHED STATES
2466 * enter the CLOSE_WAIT state.
2468 case TCPS_SYN_RECEIVED:
2469 tp->t_starttime = ticks;
2471 case TCPS_ESTABLISHED:
2472 tp->t_state = TCPS_CLOSE_WAIT;
2476 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2477 * enter the CLOSING state.
2479 case TCPS_FIN_WAIT_1:
2480 tp->t_state = TCPS_CLOSING;
2484 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2485 * starting the time-wait timer, turning off the other
2488 case TCPS_FIN_WAIT_2:
2489 KASSERT(headlocked == 1, ("%s: dodata: "
2490 "TCP_FIN_WAIT_2: head not locked", __func__));
2492 INP_INFO_WUNLOCK(&tcbinfo);
2496 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2498 case TCPS_TIME_WAIT:
2499 KASSERT(tp->t_state != TCPS_TIME_WAIT,
2500 ("%s: timewait", __func__));
2501 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl);
2505 INP_INFO_WUNLOCK(&tcbinfo);
2508 if (so->so_options & SO_DEBUG)
2509 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2514 * Return any desired output.
2516 if (needoutput || (tp->t_flags & TF_ACKNOW))
2517 (void) tcp_output(tp);
2520 KASSERT(headlocked == 0, ("%s: check_delack: head locked",
2522 INP_LOCK_ASSERT(tp->t_inpcb);
2523 if (tp->t_flags & TF_DELACK) {
2524 tp->t_flags &= ~TF_DELACK;
2525 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2527 INP_UNLOCK(tp->t_inpcb);
2531 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__));
2533 * Generate an ACK dropping incoming segment if it occupies
2534 * sequence space, where the ACK reflects our state.
2536 * We can now skip the test for the RST flag since all
2537 * paths to this code happen after packets containing
2538 * RST have been dropped.
2540 * In the SYN-RECEIVED state, don't send an ACK unless the
2541 * segment we received passes the SYN-RECEIVED ACK test.
2542 * If it fails send a RST. This breaks the loop in the
2543 * "LAND" DoS attack, and also prevents an ACK storm
2544 * between two listening ports that have been sent forged
2545 * SYN segments, each with the source address of the other.
2547 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2548 (SEQ_GT(tp->snd_una, th->th_ack) ||
2549 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2550 rstreason = BANDLIM_RST_OPENPORT;
2554 if (so->so_options & SO_DEBUG)
2555 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2558 KASSERT(headlocked, ("%s: headlocked should be 1", __func__));
2559 INP_INFO_WUNLOCK(&tcbinfo);
2560 tp->t_flags |= TF_ACKNOW;
2561 (void) tcp_output(tp);
2562 INP_UNLOCK(tp->t_inpcb);
2567 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__));
2569 tcp_dropwithreset(m, th, tp, tlen, rstreason);
2572 INP_UNLOCK(tp->t_inpcb);
2574 INP_INFO_WUNLOCK(&tcbinfo);
2579 * Drop space held by incoming segment and return.
2582 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2583 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2587 INP_UNLOCK(tp->t_inpcb);
2589 INP_INFO_WUNLOCK(&tcbinfo);
2596 * Issue RST on TCP segment. The mbuf must still include the original
2600 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
2601 int tlen, int rstreason)
2605 struct ip6_hdr *ip6;
2608 * Generate a RST, dropping incoming segment.
2609 * Make ACK acceptable to originator of segment.
2610 * Don't bother to respond if destination was broadcast/multicast.
2613 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2616 if (mtod(m, struct ip *)->ip_v == 6) {
2617 ip6 = mtod(m, struct ip6_hdr *);
2618 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2619 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2621 /* IPv6 anycast check is done at tcp6_input() */
2625 ip = mtod(m, struct ip *);
2626 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2627 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2628 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2629 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2633 /* Perform bandwidth limiting. */
2634 if (badport_bandlim(rstreason) < 0)
2637 /* tcp_respond consumes the mbuf chain. */
2638 if (th->th_flags & TH_ACK) {
2639 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
2640 th->th_ack, TH_RST);
2642 if (th->th_flags & TH_SYN)
2644 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2645 (tcp_seq)0, TH_RST|TH_ACK);
2654 * Parse TCP options and place in tcpopt.
2657 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
2662 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2664 if (opt == TCPOPT_EOL)
2666 if (opt == TCPOPT_NOP)
2672 if (optlen < 2 || optlen > cnt)
2677 if (optlen != TCPOLEN_MAXSEG)
2679 if (!(flags & TO_SYN))
2681 to->to_flags |= TOF_MSS;
2682 bcopy((char *)cp + 2,
2683 (char *)&to->to_mss, sizeof(to->to_mss));
2684 to->to_mss = ntohs(to->to_mss);
2687 if (optlen != TCPOLEN_WINDOW)
2689 if (!(flags & TO_SYN))
2691 to->to_flags |= TOF_SCALE;
2692 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
2694 case TCPOPT_TIMESTAMP:
2695 if (optlen != TCPOLEN_TIMESTAMP)
2697 to->to_flags |= TOF_TS;
2698 bcopy((char *)cp + 2,
2699 (char *)&to->to_tsval, sizeof(to->to_tsval));
2700 to->to_tsval = ntohl(to->to_tsval);
2701 bcopy((char *)cp + 6,
2702 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2703 to->to_tsecr = ntohl(to->to_tsecr);
2705 #ifdef TCP_SIGNATURE
2707 * XXX In order to reply to a host which has set the
2708 * TCP_SIGNATURE option in its initial SYN, we have to
2709 * record the fact that the option was observed here
2710 * for the syncache code to perform the correct response.
2712 case TCPOPT_SIGNATURE:
2713 if (optlen != TCPOLEN_SIGNATURE)
2715 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2718 case TCPOPT_SACK_PERMITTED:
2719 if (optlen != TCPOLEN_SACK_PERMITTED)
2721 if (!(flags & TO_SYN))
2725 to->to_flags |= TOF_SACKPERM;
2728 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2732 to->to_flags |= TOF_SACK;
2733 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2734 to->to_sacks = cp + 2;
2735 tcpstat.tcps_sack_rcv_blocks++;
2744 * Pull out of band byte out of a segment so
2745 * it doesn't appear in the user's data queue.
2746 * It is still reflected in the segment length for
2747 * sequencing purposes.
2750 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
2753 int cnt = off + th->th_urp - 1;
2756 if (m->m_len > cnt) {
2757 char *cp = mtod(m, caddr_t) + cnt;
2758 struct tcpcb *tp = sototcpcb(so);
2761 tp->t_oobflags |= TCPOOB_HAVEDATA;
2762 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2764 if (m->m_flags & M_PKTHDR)
2773 panic("tcp_pulloutofband");
2777 * Collect new round-trip time estimate
2778 * and update averages and current timeout.
2781 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2785 INP_LOCK_ASSERT(tp->t_inpcb);
2787 tcpstat.tcps_rttupdated++;
2789 if (tp->t_srtt != 0) {
2791 * srtt is stored as fixed point with 5 bits after the
2792 * binary point (i.e., scaled by 8). The following magic
2793 * is equivalent to the smoothing algorithm in rfc793 with
2794 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2795 * point). Adjust rtt to origin 0.
2797 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2798 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2800 if ((tp->t_srtt += delta) <= 0)
2804 * We accumulate a smoothed rtt variance (actually, a
2805 * smoothed mean difference), then set the retransmit
2806 * timer to smoothed rtt + 4 times the smoothed variance.
2807 * rttvar is stored as fixed point with 4 bits after the
2808 * binary point (scaled by 16). The following is
2809 * equivalent to rfc793 smoothing with an alpha of .75
2810 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2811 * rfc793's wired-in beta.
2815 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2816 if ((tp->t_rttvar += delta) <= 0)
2818 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2819 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2822 * No rtt measurement yet - use the unsmoothed rtt.
2823 * Set the variance to half the rtt (so our first
2824 * retransmit happens at 3*rtt).
2826 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2827 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2828 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2834 * the retransmit should happen at rtt + 4 * rttvar.
2835 * Because of the way we do the smoothing, srtt and rttvar
2836 * will each average +1/2 tick of bias. When we compute
2837 * the retransmit timer, we want 1/2 tick of rounding and
2838 * 1 extra tick because of +-1/2 tick uncertainty in the
2839 * firing of the timer. The bias will give us exactly the
2840 * 1.5 tick we need. But, because the bias is
2841 * statistical, we have to test that we don't drop below
2842 * the minimum feasible timer (which is 2 ticks).
2844 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2845 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2848 * We received an ack for a packet that wasn't retransmitted;
2849 * it is probably safe to discard any error indications we've
2850 * received recently. This isn't quite right, but close enough
2851 * for now (a route might have failed after we sent a segment,
2852 * and the return path might not be symmetrical).
2854 tp->t_softerror = 0;
2858 * Determine a reasonable value for maxseg size.
2859 * If the route is known, check route for mtu.
2860 * If none, use an mss that can be handled on the outgoing
2861 * interface without forcing IP to fragment; if bigger than
2862 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2863 * to utilize large mbufs. If no route is found, route has no mtu,
2864 * or the destination isn't local, use a default, hopefully conservative
2865 * size (usually 512 or the default IP max size, but no more than the mtu
2866 * of the interface), as we can't discover anything about intervening
2867 * gateways or networks. We also initialize the congestion/slow start
2868 * window to be a single segment if the destination isn't local.
2869 * While looking at the routing entry, we also initialize other path-dependent
2870 * parameters from pre-set or cached values in the routing entry.
2872 * Also take into account the space needed for options that we
2873 * send regularly. Make maxseg shorter by that amount to assure
2874 * that we can send maxseg amount of data even when the options
2875 * are present. Store the upper limit of the length of options plus
2879 * In case of T/TCP, we call this routine during implicit connection
2880 * setup as well (offer = -1), to initialize maxseg from the cached
2883 * NOTE that this routine is only called when we process an incoming
2884 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2887 tcp_mss(struct tcpcb *tp, int offer)
2892 struct inpcb *inp = tp->t_inpcb;
2894 struct hc_metrics_lite metrics;
2895 int origoffer = offer;
2898 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2899 size_t min_protoh = isipv6 ?
2900 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2901 sizeof (struct tcpiphdr);
2903 const size_t min_protoh = sizeof(struct tcpiphdr);
2909 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags);
2910 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2914 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags);
2915 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2917 so = inp->inp_socket;
2920 * no route to sender, stay with default mss and return
2925 /* what have we got? */
2929 * Offer == 0 means that there was no MSS on the SYN
2930 * segment, in this case we use tcp_mssdflt.
2934 isipv6 ? tcp_v6mssdflt :
2941 * Offer == -1 means that we didn't receive SYN yet.
2947 * Prevent DoS attack with too small MSS. Round up
2948 * to at least minmss.
2950 offer = max(offer, tcp_minmss);
2952 * Sanity check: make sure that maxopd will be large
2953 * enough to allow some data on segments even if the
2954 * all the option space is used (40bytes). Otherwise
2955 * funny things may happen in tcp_output.
2957 offer = max(offer, 64);
2961 * rmx information is now retrieved from tcp_hostcache
2963 tcp_hc_get(&inp->inp_inc, &metrics);
2966 * if there's a discovered mtu int tcp hostcache, use it
2967 * else, use the link mtu.
2969 if (metrics.rmx_mtu)
2970 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2974 mss = maxmtu - min_protoh;
2975 if (!path_mtu_discovery &&
2976 !in6_localaddr(&inp->in6p_faddr))
2977 mss = min(mss, tcp_v6mssdflt);
2981 mss = maxmtu - min_protoh;
2982 if (!path_mtu_discovery &&
2983 !in_localaddr(inp->inp_faddr))
2984 mss = min(mss, tcp_mssdflt);
2987 mss = min(mss, offer);
2990 * maxopd stores the maximum length of data AND options
2991 * in a segment; maxseg is the amount of data in a normal
2992 * segment. We need to store this value (maxopd) apart
2993 * from maxseg, because now every segment carries options
2994 * and thus we normally have somewhat less data in segments.
2999 * origoffer==-1 indicates, that no segments were received yet.
3000 * In this case we just guess.
3002 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3004 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3005 mss -= TCPOLEN_TSTAMP_APPA;
3008 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3010 mss &= ~(MCLBYTES-1);
3013 mss = mss / MCLBYTES * MCLBYTES;
3018 * If there's a pipesize, change the socket buffer to that size,
3019 * don't change if sb_hiwat is different than default (then it
3020 * has been changed on purpose with setsockopt).
3021 * Make the socket buffers an integral number of mss units;
3022 * if the mss is larger than the socket buffer, decrease the mss.
3024 SOCKBUF_LOCK(&so->so_snd);
3025 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3026 bufsize = metrics.rmx_sendpipe;
3028 bufsize = so->so_snd.sb_hiwat;
3032 bufsize = roundup(bufsize, mss);
3033 if (bufsize > sb_max)
3035 if (bufsize > so->so_snd.sb_hiwat)
3036 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3038 SOCKBUF_UNLOCK(&so->so_snd);
3041 SOCKBUF_LOCK(&so->so_rcv);
3042 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3043 bufsize = metrics.rmx_recvpipe;
3045 bufsize = so->so_rcv.sb_hiwat;
3046 if (bufsize > mss) {
3047 bufsize = roundup(bufsize, mss);
3048 if (bufsize > sb_max)
3050 if (bufsize > so->so_rcv.sb_hiwat)
3051 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3053 SOCKBUF_UNLOCK(&so->so_rcv);
3055 * While we're here, check the others too
3057 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3059 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3060 tcpstat.tcps_usedrtt++;
3061 if (metrics.rmx_rttvar) {
3062 tp->t_rttvar = metrics.rmx_rttvar;
3063 tcpstat.tcps_usedrttvar++;
3065 /* default variation is +- 1 rtt */
3067 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3069 TCPT_RANGESET(tp->t_rxtcur,
3070 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3071 tp->t_rttmin, TCPTV_REXMTMAX);
3073 if (metrics.rmx_ssthresh) {
3075 * There's some sort of gateway or interface
3076 * buffer limit on the path. Use this to set
3077 * the slow start threshhold, but set the
3078 * threshold to no less than 2*mss.
3080 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3081 tcpstat.tcps_usedssthresh++;
3083 if (metrics.rmx_bandwidth)
3084 tp->snd_bandwidth = metrics.rmx_bandwidth;
3087 * Set the slow-start flight size depending on whether this
3088 * is a local network or not.
3090 * Extend this so we cache the cwnd too and retrieve it here.
3091 * Make cwnd even bigger than RFC3390 suggests but only if we
3092 * have previous experience with the remote host. Be careful
3093 * not make cwnd bigger than remote receive window or our own
3094 * send socket buffer. Maybe put some additional upper bound
3095 * on the retrieved cwnd. Should do incremental updates to
3096 * hostcache when cwnd collapses so next connection doesn't
3097 * overloads the path again.
3099 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3100 * We currently check only in syncache_socket for that.
3102 #define TCP_METRICS_CWND
3103 #ifdef TCP_METRICS_CWND
3104 if (metrics.rmx_cwnd)
3105 tp->snd_cwnd = max(mss,
3106 min(metrics.rmx_cwnd / 2,
3107 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3111 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3113 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3114 (!isipv6 && in_localaddr(inp->inp_faddr)))
3116 else if (in_localaddr(inp->inp_faddr))
3118 tp->snd_cwnd = mss * ss_fltsz_local;
3120 tp->snd_cwnd = mss * ss_fltsz;
3122 /* Check the interface for TSO capabilities. */
3123 if (mtuflags & CSUM_TSO)
3124 tp->t_flags |= TF_TSO;
3128 * Determine the MSS option to send on an outgoing SYN.
3131 tcp_mssopt(struct in_conninfo *inc)
3138 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3141 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3145 mss = tcp_v6mssdflt;
3146 maxmtu = tcp_maxmtu6(inc, NULL);
3147 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3148 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3153 maxmtu = tcp_maxmtu(inc, NULL);
3154 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3155 min_protoh = sizeof(struct tcpiphdr);
3157 if (maxmtu && thcmtu)
3158 mss = min(maxmtu, thcmtu) - min_protoh;
3159 else if (maxmtu || thcmtu)
3160 mss = max(maxmtu, thcmtu) - min_protoh;
3167 * On a partial ack arrives, force the retransmission of the
3168 * next unacknowledged segment. Do not clear tp->t_dupacks.
3169 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3173 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3175 tcp_seq onxt = tp->snd_nxt;
3176 u_long ocwnd = tp->snd_cwnd;
3178 tcp_timer_activate(tp, TT_REXMT, 0);
3180 tp->snd_nxt = th->th_ack;
3182 * Set snd_cwnd to one segment beyond acknowledged offset.
3183 * (tp->snd_una has not yet been updated when this function is called.)
3185 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3186 tp->t_flags |= TF_ACKNOW;
3187 (void) tcp_output(tp);
3188 tp->snd_cwnd = ocwnd;
3189 if (SEQ_GT(onxt, tp->snd_nxt))
3192 * Partial window deflation. Relies on fact that tp->snd_una
3195 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3196 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3199 tp->snd_cwnd += tp->t_maxseg;
3203 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3204 * looking for a pcb in the listen state. Returns 0 otherwise.
3207 tcp_timewait(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
3208 struct mbuf *m, int tlen)
3214 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3216 const int isipv6 = 0;
3219 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */
3220 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3221 INP_LOCK_ASSERT(inp);
3224 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
3225 * still present. This is undesirable, but temporarily necessary
3226 * until we work out how to handle inpcb's who's timewait state has
3233 thflags = th->th_flags;
3236 * NOTE: for FIN_WAIT_2 (to be added later),
3237 * must validate sequence number before accepting RST
3241 * If the segment contains RST:
3242 * Drop the segment - see Stevens, vol. 2, p. 964 and
3245 if (thflags & TH_RST)
3249 /* PAWS not needed at the moment */
3251 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3252 * and it's less than ts_recent, drop it.
3254 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3255 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3256 if ((thflags & TH_ACK) == 0)
3261 * ts_recent is never updated because we never accept new segments.
3266 * If a new connection request is received
3267 * while in TIME_WAIT, drop the old connection
3268 * and start over if the sequence numbers
3269 * are above the previous ones.
3271 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3277 * Drop the the segment if it does not contain an ACK.
3279 if ((thflags & TH_ACK) == 0)
3283 * Reset the 2MSL timer if this is a duplicate FIN.
3285 if (thflags & TH_FIN) {
3286 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3287 if (seq + 1 == tw->rcv_nxt)
3288 tcp_timer_2msl_reset(tw, 1);
3292 * Acknowledge the segment if it has data or is not a duplicate ACK.
3294 if (thflags != TH_ACK || tlen != 0 ||
3295 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3296 tcp_twrespond(tw, TH_ACK);
3300 * Generate a RST, dropping incoming segment.
3301 * Make ACK acceptable to originator of segment.
3302 * Don't bother to respond if destination was broadcast/multicast.
3304 if (m->m_flags & (M_BCAST|M_MCAST))
3307 struct ip6_hdr *ip6;
3309 /* IPv6 anycast check is done at tcp6_input() */
3310 ip6 = mtod(m, struct ip6_hdr *);
3311 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3312 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3317 ip = mtod(m, struct ip *);
3318 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3319 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3320 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3321 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3324 if (thflags & TH_ACK) {
3326 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3328 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3330 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);