2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
33 #include "opt_ipfw.h" /* for ipfw_fwd */
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40 #include "opt_tcp_sack.h"
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/proc.h> /* for proc0 declaration */
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <sys/systm.h>
55 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
60 #include <net/route.h>
62 #include <netinet/in.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_options.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
82 #include <netinet/tcpip.h>
84 #include <netinet/tcp_debug.h>
88 #include <netipsec/ipsec.h>
89 #include <netipsec/ipsec6.h>
93 #include <netinet6/ipsec.h>
94 #include <netinet6/ipsec6.h>
95 #include <netkey/key.h>
98 #include <machine/in_cksum.h>
100 #include <security/mac/mac_framework.h>
102 static const int tcprexmtthresh = 3;
104 struct tcpstat tcpstat;
105 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
106 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
108 static int log_in_vain = 0;
109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
110 &log_in_vain, 0, "Log all incoming TCP connections");
112 static int blackhole = 0;
113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
114 &blackhole, 0, "Do not send RST when dropping refused connections");
116 int tcp_delack_enabled = 1;
117 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
118 &tcp_delack_enabled, 0,
119 "Delay ACK to try and piggyback it onto a data packet");
121 #ifdef TCP_DROP_SYNFIN
122 static int drop_synfin = 0;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
124 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
127 static int tcp_do_rfc3042 = 1;
128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
129 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
131 static int tcp_do_rfc3390 = 1;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
134 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
136 static int tcp_insecure_rst = 0;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
138 &tcp_insecure_rst, 0,
139 "Follow the old (insecure) criteria for accepting RST packets.");
141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
142 "TCP Segment Reassembly Queue");
144 static int tcp_reass_maxseg = 0;
145 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
146 &tcp_reass_maxseg, 0,
147 "Global maximum number of TCP Segments in Reassembly Queue");
149 int tcp_reass_qsize = 0;
150 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
152 "Global number of TCP Segments currently in Reassembly Queue");
154 static int tcp_reass_maxqlen = 48;
155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
156 &tcp_reass_maxqlen, 0,
157 "Maximum number of TCP Segments per individual Reassembly Queue");
159 static int tcp_reass_overflows = 0;
160 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
161 &tcp_reass_overflows, 0,
162 "Global number of TCP Segment Reassembly Queue Overflows");
164 struct inpcbhead tcb;
165 #define tcb6 tcb /* for KAME src sync over BSD*'s */
166 struct inpcbinfo tcbinfo;
167 struct mtx *tcbinfo_mtx;
169 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
171 static void tcp_pulloutofband(struct socket *,
172 struct tcphdr *, struct mbuf *, int);
173 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
175 static void tcp_xmit_timer(struct tcpcb *, int);
176 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
177 static int tcp_timewait(struct inpcb *, struct tcpopt *,
178 struct tcphdr *, struct mbuf *, int);
180 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
182 #define ND6_HINT(tp) \
184 if ((tp) && (tp)->t_inpcb && \
185 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
186 nd6_nud_hint(NULL, NULL, 0); \
193 * Indicate whether this ack should be delayed. We can delay the ack if
194 * - there is no delayed ack timer in progress and
195 * - our last ack wasn't a 0-sized window. We never want to delay
196 * the ack that opens up a 0-sized window and
197 * - delayed acks are enabled or
198 * - this is a half-synchronized T/TCP connection.
200 #define DELAY_ACK(tp) \
201 ((!callout_active(tp->tt_delack) && \
202 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
203 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
205 /* Initialize TCP reassembly queue */
207 tcp_reass_zone_change(void *tag)
210 tcp_reass_maxseg = nmbclusters / 16;
211 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
214 uma_zone_t tcp_reass_zone;
218 tcp_reass_maxseg = nmbclusters / 16;
219 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
221 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
222 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
223 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
224 EVENTHANDLER_REGISTER(nmbclusters_change,
225 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
229 tcp_reass(tp, th, tlenp, m)
230 register struct tcpcb *tp;
231 register struct tcphdr *th;
236 struct tseg_qent *p = NULL;
237 struct tseg_qent *nq;
238 struct tseg_qent *te = NULL;
239 struct socket *so = tp->t_inpcb->inp_socket;
242 INP_LOCK_ASSERT(tp->t_inpcb);
245 * XXX: tcp_reass() is rather inefficient with its data structures
246 * and should be rewritten (see NetBSD for optimizations). While
247 * doing that it should move to its own file tcp_reass.c.
251 * Call with th==NULL after become established to
252 * force pre-ESTABLISHED data up to user socket.
258 * Limit the number of segments in the reassembly queue to prevent
259 * holding on to too many segments (and thus running out of mbufs).
260 * Make sure to let the missing segment through which caused this
261 * queue. Always keep one global queue entry spare to be able to
262 * process the missing segment.
264 if (th->th_seq != tp->rcv_nxt &&
265 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
266 tp->t_segqlen >= tcp_reass_maxqlen)) {
267 tcp_reass_overflows++;
268 tcpstat.tcps_rcvmemdrop++;
275 * Allocate a new queue entry. If we can't, or hit the zone limit
278 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
280 tcpstat.tcps_rcvmemdrop++;
289 * Find a segment which begins after this one does.
291 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
292 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
298 * If there is a preceding segment, it may provide some of
299 * our data already. If so, drop the data from the incoming
300 * segment. If it provides all of our data, drop us.
304 /* conversion to int (in i) handles seq wraparound */
305 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
308 tcpstat.tcps_rcvduppack++;
309 tcpstat.tcps_rcvdupbyte += *tlenp;
311 uma_zfree(tcp_reass_zone, te);
315 * Try to present any queued data
316 * at the left window edge to the user.
317 * This is needed after the 3-WHS
320 goto present; /* ??? */
327 tcpstat.tcps_rcvoopack++;
328 tcpstat.tcps_rcvoobyte += *tlenp;
331 * While we overlap succeeding segments trim them or,
332 * if they are completely covered, dequeue them.
335 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
338 if (i < q->tqe_len) {
339 q->tqe_th->th_seq += i;
345 nq = LIST_NEXT(q, tqe_q);
346 LIST_REMOVE(q, tqe_q);
348 uma_zfree(tcp_reass_zone, q);
354 /* Insert the new segment queue entry into place. */
357 te->tqe_len = *tlenp;
360 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
362 LIST_INSERT_AFTER(p, te, tqe_q);
367 * Present data to user, advancing rcv_nxt through
368 * completed sequence space.
370 if (!TCPS_HAVEESTABLISHED(tp->t_state))
372 q = LIST_FIRST(&tp->t_segq);
373 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
375 SOCKBUF_LOCK(&so->so_rcv);
377 tp->rcv_nxt += q->tqe_len;
378 flags = q->tqe_th->th_flags & TH_FIN;
379 nq = LIST_NEXT(q, tqe_q);
380 LIST_REMOVE(q, tqe_q);
381 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
384 sbappendstream_locked(&so->so_rcv, q->tqe_m);
385 uma_zfree(tcp_reass_zone, q);
389 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
391 sorwakeup_locked(so);
396 * TCP input routine, follows pages 65-76 of the
397 * protocol specification dated September, 1981 very closely.
401 tcp6_input(mp, offp, proto)
405 register struct mbuf *m = *mp;
406 struct in6_ifaddr *ia6;
408 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
411 * draft-itojun-ipv6-tcp-to-anycast
412 * better place to put this in?
414 ia6 = ip6_getdstifaddr(m);
415 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
418 ip6 = mtod(m, struct ip6_hdr *);
419 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
420 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
431 register struct mbuf *m;
434 register struct tcphdr *th;
435 register struct ip *ip = NULL;
436 register struct ipovly *ipov;
437 register struct inpcb *inp = NULL;
442 register struct tcpcb *tp = 0;
443 register int thflags;
444 struct socket *so = 0;
445 int todrop, acked, ourfinisacked, needoutput = 0;
447 struct tcpopt to; /* options in this segment */
449 #ifdef IPFIREWALL_FORWARD
450 struct m_tag *fwd_tag;
452 int rstreason; /* For badport_bandlim accounting purposes */
454 struct ip6_hdr *ip6 = NULL;
458 const int isipv6 = 0;
463 * The size of tcp_saveipgen must be the size of the max ip header,
466 u_char tcp_saveipgen[40];
467 struct tcphdr tcp_savetcp;
472 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
474 bzero((char *)&to, sizeof(to));
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))) == 0) {
515 tcpstat.tcps_rcvshort++;
519 ip = mtod(m, struct ip *);
520 ipov = (struct ipovly *)ip;
521 th = (struct tcphdr *)((caddr_t)ip + off0);
524 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
525 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
526 th->th_sum = m->m_pkthdr.csum_data;
528 th->th_sum = in_pseudo(ip->ip_src.s_addr,
530 htonl(m->m_pkthdr.csum_data +
533 th->th_sum ^= 0xffff;
535 ipov->ih_len = (u_short)tlen;
536 ipov->ih_len = htons(ipov->ih_len);
540 * Checksum extended TCP header and data.
542 len = sizeof (struct ip) + tlen;
543 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
544 ipov->ih_len = (u_short)tlen;
545 ipov->ih_len = htons(ipov->ih_len);
546 th->th_sum = in_cksum(m, len);
549 tcpstat.tcps_rcvbadsum++;
552 /* Re-initialization for later version check */
553 ip->ip_v = IPVERSION;
557 * Check that TCP offset makes sense,
558 * pull out TCP options and adjust length. XXX
560 off = th->th_off << 2;
561 if (off < sizeof (struct tcphdr) || off > tlen) {
562 tcpstat.tcps_rcvbadoff++;
565 tlen -= off; /* tlen is used instead of ti->ti_len */
566 if (off > sizeof (struct tcphdr)) {
569 IP6_EXTHDR_CHECK(m, off0, off, );
570 ip6 = mtod(m, struct ip6_hdr *);
571 th = (struct tcphdr *)((caddr_t)ip6 + off0);
574 if (m->m_len < sizeof(struct ip) + off) {
575 if ((m = m_pullup(m, sizeof (struct ip) + off))
577 tcpstat.tcps_rcvshort++;
580 ip = mtod(m, struct ip *);
581 ipov = (struct ipovly *)ip;
582 th = (struct tcphdr *)((caddr_t)ip + off0);
585 optlen = off - sizeof (struct tcphdr);
586 optp = (u_char *)(th + 1);
588 thflags = th->th_flags;
590 #ifdef TCP_DROP_SYNFIN
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))
603 * Convert TCP protocol specific fields to host format.
605 th->th_seq = ntohl(th->th_seq);
606 th->th_ack = ntohl(th->th_ack);
607 th->th_win = ntohs(th->th_win);
608 th->th_urp = ntohs(th->th_urp);
611 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
612 * until after ip6_savecontrol() is called and before other functions
613 * which don't want those proto headers.
614 * Because ip6_savecontrol() is going to parse the mbuf to
615 * search for data to be passed up to user-land, it wants mbuf
616 * parameters to be unchanged.
617 * XXX: the call of ip6_savecontrol() has been obsoleted based on
618 * latest version of the advanced API (20020110).
620 drop_hdrlen = off0 + off;
623 * Locate pcb for segment.
625 INP_INFO_WLOCK(&tcbinfo);
628 KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
629 #ifdef IPFIREWALL_FORWARD
630 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
631 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
633 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
634 struct sockaddr_in *next_hop;
636 next_hop = (struct sockaddr_in *)(fwd_tag+1);
638 * Transparently forwarded. Pretend to be the destination.
639 * already got one like this?
641 inp = in_pcblookup_hash(&tcbinfo,
642 ip->ip_src, th->th_sport,
643 ip->ip_dst, th->th_dport,
644 0, m->m_pkthdr.rcvif);
646 /* It's new. Try to find the ambushing socket. */
647 inp = in_pcblookup_hash(&tcbinfo,
648 ip->ip_src, th->th_sport,
651 ntohs(next_hop->sin_port) :
656 /* Remove the tag from the packet. We don't need it anymore. */
657 m_tag_delete(m, fwd_tag);
659 #endif /* IPFIREWALL_FORWARD */
662 inp = in6_pcblookup_hash(&tcbinfo,
663 &ip6->ip6_src, th->th_sport,
664 &ip6->ip6_dst, th->th_dport,
669 inp = in_pcblookup_hash(&tcbinfo,
670 ip->ip_src, th->th_sport,
671 ip->ip_dst, th->th_dport,
674 #ifdef IPFIREWALL_FORWARD
676 #endif /* IPFIREWALL_FORWARD */
678 #if defined(IPSEC) || defined(FAST_IPSEC)
681 if (inp != NULL && ipsec6_in_reject(m, inp)) {
683 ipsec6stat.in_polvio++;
689 if (inp != NULL && ipsec4_in_reject(m, inp)) {
691 ipsecstat.in_polvio++;
695 #endif /*IPSEC || FAST_IPSEC*/
698 * If the state is CLOSED (i.e., TCB does not exist) then
699 * all data in the incoming segment is discarded.
700 * If the TCB exists but is in CLOSED state, it is embryonic,
701 * but should either do a listen or a connect soon.
706 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
708 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
715 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
716 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
721 strcpy(dbuf, inet_ntoa(ip->ip_dst));
722 strcpy(sbuf, inet_ntoa(ip->ip_src));
724 switch (log_in_vain) {
726 if ((thflags & TH_SYN) == 0)
731 "Connection attempt to TCP %s:%d "
732 "from %s:%d flags:0x%02x\n",
733 dbuf, ntohs(th->th_dport), sbuf,
734 ntohs(th->th_sport), thflags);
743 if (thflags & TH_SYN)
752 rstreason = BANDLIM_RST_CLOSEDPORT;
757 /* Check the minimum TTL for socket. */
758 if (inp->inp_ip_minttl != 0) {
760 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
764 if (inp->inp_ip_minttl > ip->ip_ttl)
768 if (inp->inp_vflag & INP_TIMEWAIT) {
770 * The only option of relevance is TOF_CC, and only if
771 * present in a SYN segment. See tcp_timewait().
773 if (thflags & TH_SYN)
774 tcp_dooptions(&to, optp, optlen, TO_SYN);
775 if (tcp_timewait(inp, &to, th, m, tlen))
778 * tcp_timewait unlocks inp.
780 INP_INFO_WUNLOCK(&tcbinfo);
786 rstreason = BANDLIM_RST_CLOSEDPORT;
789 if (tp->t_state == TCPS_CLOSED)
793 INP_LOCK_ASSERT(inp);
794 if (mac_check_inpcb_deliver(inp, m))
797 so = inp->inp_socket;
798 KASSERT(so != NULL, ("tcp_input: so == NULL"));
800 if (so->so_options & SO_DEBUG) {
801 ostate = tp->t_state;
803 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
805 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
809 if (so->so_options & SO_ACCEPTCONN) {
810 struct in_conninfo inc;
812 bzero(&inc, sizeof(inc));
814 inc.inc_isipv6 = isipv6;
817 inc.inc6_faddr = ip6->ip6_src;
818 inc.inc6_laddr = ip6->ip6_dst;
820 inc.inc_faddr = ip->ip_src;
821 inc.inc_laddr = ip->ip_dst;
823 inc.inc_fport = th->th_sport;
824 inc.inc_lport = th->th_dport;
827 * If the state is LISTEN then ignore segment if it contains
828 * a RST. If the segment contains an ACK then it is bad and
829 * send a RST. If it does not contain a SYN then it is not
830 * interesting; drop it.
832 * If the state is SYN_RECEIVED (syncache) and seg contains
833 * an ACK, but not for our SYN/ACK, send a RST. If the seg
834 * contains a RST, check the sequence number to see if it
835 * is a valid reset segment.
837 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
838 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
840 * Parse the TCP options here because
841 * syncookies need access to the reflected
844 tcp_dooptions(&to, optp, optlen, 0);
845 if (!syncache_expand(&inc, &to, th, &so, m)) {
847 * No syncache entry, or ACK was not
848 * for our SYN/ACK. Send a RST.
850 tcpstat.tcps_badsyn++;
851 rstreason = BANDLIM_RST_OPENPORT;
856 * Could not complete 3-way handshake,
857 * connection is being closed down, and
858 * syncache has free'd mbuf.
861 INP_INFO_WUNLOCK(&tcbinfo);
865 * Socket is created in state SYN_RECEIVED.
866 * Continue processing segment.
873 * This is what would have happened in
874 * tcp_output() when the SYN,ACK was sent.
876 tp->snd_up = tp->snd_una;
877 tp->snd_max = tp->snd_nxt = tp->iss + 1;
878 tp->last_ack_sent = tp->rcv_nxt;
881 if (thflags & TH_RST) {
882 syncache_chkrst(&inc, th);
885 if (thflags & TH_ACK) {
886 syncache_badack(&inc);
887 tcpstat.tcps_badsyn++;
888 rstreason = BANDLIM_RST_OPENPORT;
895 * Segment's flags are (SYN) or (SYN|FIN).
899 * If deprecated address is forbidden,
900 * we do not accept SYN to deprecated interface
901 * address to prevent any new inbound connection from
902 * getting established.
903 * When we do not accept SYN, we send a TCP RST,
904 * with deprecated source address (instead of dropping
905 * it). We compromise it as it is much better for peer
906 * to send a RST, and RST will be the final packet
909 * If we do not forbid deprecated addresses, we accept
910 * the SYN packet. RFC2462 does not suggest dropping
912 * If we decipher RFC2462 5.5.4, it says like this:
913 * 1. use of deprecated addr with existing
914 * communication is okay - "SHOULD continue to be
916 * 2. use of it with new communication:
917 * (2a) "SHOULD NOT be used if alternate address
918 * with sufficient scope is available"
919 * (2b) nothing mentioned otherwise.
920 * Here we fall into (2b) case as we have no choice in
921 * our source address selection - we must obey the peer.
923 * The wording in RFC2462 is confusing, and there are
924 * multiple description text for deprecated address
925 * handling - worse, they are not exactly the same.
926 * I believe 5.5.4 is the best one, so we follow 5.5.4.
928 if (isipv6 && !ip6_use_deprecated) {
929 struct in6_ifaddr *ia6;
931 if ((ia6 = ip6_getdstifaddr(m)) &&
932 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
935 rstreason = BANDLIM_RST_OPENPORT;
941 * If it is from this socket, drop it, it must be forged.
942 * Don't bother responding if the destination was a broadcast.
944 if (th->th_dport == th->th_sport) {
946 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
950 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
955 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
957 * Note that it is quite possible to receive unicast
958 * link-layer packets with a broadcast IP address. Use
959 * in_broadcast() to find them.
961 if (m->m_flags & (M_BCAST|M_MCAST))
964 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
965 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
968 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
969 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
970 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
971 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
975 * SYN appears to be valid; create compressed TCP state
976 * for syncache, or perform t/tcp connection.
978 if (so->so_qlen <= so->so_qlimit) {
980 if (so->so_options & SO_DEBUG)
981 tcp_trace(TA_INPUT, ostate, tp,
982 (void *)tcp_saveipgen, &tcp_savetcp, 0);
984 tcp_dooptions(&to, optp, optlen, TO_SYN);
985 if (!syncache_add(&inc, &to, th, inp, &so, m))
986 goto drop; /* XXX: does not happen */
989 * Entry added to syncache, mbuf used to
990 * send SYN,ACK packet. Everything unlocked
995 panic("T/TCP not supported at the moment");
998 * Segment passed TAO tests.
999 * XXX: Can't happen at the moment.
1002 inp = sotoinpcb(so);
1004 tp = intotcpcb(inp);
1005 tp->t_starttime = ticks;
1006 tp->t_state = TCPS_ESTABLISHED;
1010 * If there is a FIN or if there is data, then
1011 * delay SYN,ACK(SYN) in the hope of piggy-backing
1012 * it on a response segment. Otherwise must send
1013 * ACK now in case the other side is slow starting.
1015 if (thflags & TH_FIN || tlen != 0)
1016 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1018 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1019 tiwin = th->th_win << tp->snd_scale;
1020 tcpstat.tcps_connects++;
1028 KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1029 INP_LOCK_ASSERT(inp);
1031 /* Syncache takes care of sockets in the listen state. */
1032 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1035 * This is the second part of the MSS DoS prevention code (after
1036 * minmss on the sending side) and it deals with too many too small
1037 * tcp packets in a too short timeframe (1 second).
1039 * For every full second we count the number of received packets
1040 * and bytes. If we get a lot of packets per second for this connection
1041 * (tcp_minmssoverload) we take a closer look at it and compute the
1042 * average packet size for the past second. If that is less than
1043 * tcp_minmss we get too many packets with very small payload which
1044 * is not good and burdens our system (and every packet generates
1045 * a wakeup to the process connected to our socket). We can reasonable
1046 * expect this to be small packet DoS attack to exhaust our CPU
1049 * Care has to be taken for the minimum packet overload value. This
1050 * value defines the minimum number of packets per second before we
1051 * start to worry. This must not be too low to avoid killing for
1052 * example interactive connections with many small packets like
1055 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1058 * Account for packet if payload packet, skip over ACK, etc.
1060 if (tcp_minmss && tcp_minmssoverload &&
1061 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1062 if ((unsigned int)(tp->rcv_second - ticks) < hz) {
1064 tp->rcv_byps += tlen + off;
1065 if (tp->rcv_pps > tcp_minmssoverload) {
1066 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1067 printf("too many small tcp packets from "
1068 "%s:%u, av. %lubyte/packet, "
1069 "dropping connection\n",
1072 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1074 inet_ntoa(inp->inp_inc.inc_faddr),
1075 inp->inp_inc.inc_fport,
1076 tp->rcv_byps / tp->rcv_pps);
1077 KASSERT(headlocked, ("tcp_input: "
1078 "after_listen: tcp_drop: head "
1080 tp = tcp_drop(tp, ECONNRESET);
1081 tcpstat.tcps_minmssdrops++;
1086 tp->rcv_second = ticks + hz;
1088 tp->rcv_byps = tlen + off;
1093 * Segment received on connection.
1094 * Reset idle time and keep-alive timer.
1096 tp->t_rcvtime = ticks;
1097 if (TCPS_HAVEESTABLISHED(tp->t_state))
1098 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1101 * Unscale the window into a 32-bit value.
1102 * This value is bogus for the TCPS_SYN_SENT state
1103 * and is overwritten later.
1105 tiwin = th->th_win << tp->snd_scale;
1108 * Parse options on any incoming segment.
1110 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0);
1113 * If echoed timestamp is later than the current time,
1114 * fall back to non RFC1323 RTT calculation. Normalize
1115 * timestamp if syncookies were used when this connection
1118 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1119 to.to_tsecr -= tp->ts_offset;
1120 if (TSTMP_GT(to.to_tsecr, ticks))
1125 * Process options only when we get SYN/ACK back. The SYN case
1126 * for incoming connections is handled in tcp_syncache.
1127 * XXX this is traditional behavior, may need to be cleaned up.
1129 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1130 if ((to.to_flags & TOF_SCALE) &&
1131 (tp->t_flags & TF_REQ_SCALE)) {
1132 tp->t_flags |= TF_RCVD_SCALE;
1133 tp->snd_scale = to.to_requested_s_scale;
1134 tp->snd_wnd = th->th_win << tp->snd_scale;
1135 tiwin = tp->snd_wnd;
1137 if (to.to_flags & TOF_TS) {
1138 tp->t_flags |= TF_RCVD_TSTMP;
1139 tp->ts_recent = to.to_tsval;
1140 tp->ts_recent_age = ticks;
1142 if (to.to_flags & TOF_MSS)
1143 tcp_mss(tp, to.to_mss);
1144 if (tp->sack_enable) {
1145 if (!(to.to_flags & TOF_SACK))
1146 tp->sack_enable = 0;
1148 tp->t_flags |= TF_SACK_PERMIT;
1154 * Header prediction: check for the two common cases
1155 * of a uni-directional data xfer. If the packet has
1156 * no control flags, is in-sequence, the window didn't
1157 * change and we're not retransmitting, it's a
1158 * candidate. If the length is zero and the ack moved
1159 * forward, we're the sender side of the xfer. Just
1160 * free the data acked & wake any higher level process
1161 * that was blocked waiting for space. If the length
1162 * is non-zero and the ack didn't move, we're the
1163 * receiver side. If we're getting packets in-order
1164 * (the reassembly queue is empty), add the data to
1165 * the socket buffer and note that we need a delayed ack.
1166 * Make sure that the hidden state-flags are also off.
1167 * Since we check for TCPS_ESTABLISHED above, it can only
1170 if (tp->t_state == TCPS_ESTABLISHED &&
1171 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1172 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1173 ((to.to_flags & TOF_TS) == 0 ||
1174 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1175 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1176 tp->snd_nxt == tp->snd_max) {
1179 * If last ACK falls within this segment's sequence numbers,
1180 * record the timestamp.
1181 * NOTE that the test is modified according to the latest
1182 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1184 if ((to.to_flags & TOF_TS) != 0 &&
1185 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1186 tp->ts_recent_age = ticks;
1187 tp->ts_recent = to.to_tsval;
1191 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1192 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1193 tp->snd_cwnd >= tp->snd_wnd &&
1194 ((!tcp_do_newreno && !tp->sack_enable &&
1195 tp->t_dupacks < tcprexmtthresh) ||
1196 ((tcp_do_newreno || tp->sack_enable) &&
1197 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1198 TAILQ_EMPTY(&tp->snd_holes)))) {
1199 KASSERT(headlocked, ("headlocked"));
1200 INP_INFO_WUNLOCK(&tcbinfo);
1203 * this is a pure ack for outstanding data.
1205 ++tcpstat.tcps_predack;
1207 * "bad retransmit" recovery
1209 if (tp->t_rxtshift == 1 &&
1210 ticks < tp->t_badrxtwin) {
1211 ++tcpstat.tcps_sndrexmitbad;
1212 tp->snd_cwnd = tp->snd_cwnd_prev;
1214 tp->snd_ssthresh_prev;
1215 tp->snd_recover = tp->snd_recover_prev;
1216 if (tp->t_flags & TF_WASFRECOVERY)
1217 ENTER_FASTRECOVERY(tp);
1218 tp->snd_nxt = tp->snd_max;
1219 tp->t_badrxtwin = 0;
1223 * Recalculate the transmit timer / rtt.
1225 * Some boxes send broken timestamp replies
1226 * during the SYN+ACK phase, ignore
1227 * timestamps of 0 or we could calculate a
1228 * huge RTT and blow up the retransmit timer.
1230 if ((to.to_flags & TOF_TS) != 0 &&
1232 if (!tp->t_rttlow ||
1233 tp->t_rttlow > ticks - to.to_tsecr)
1234 tp->t_rttlow = ticks - to.to_tsecr;
1236 ticks - to.to_tsecr + 1);
1237 } else if (tp->t_rtttime &&
1238 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1239 if (!tp->t_rttlow ||
1240 tp->t_rttlow > ticks - tp->t_rtttime)
1241 tp->t_rttlow = ticks - tp->t_rtttime;
1243 ticks - tp->t_rtttime);
1245 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1246 acked = th->th_ack - tp->snd_una;
1247 tcpstat.tcps_rcvackpack++;
1248 tcpstat.tcps_rcvackbyte += acked;
1249 sbdrop(&so->so_snd, acked);
1250 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1251 SEQ_LEQ(th->th_ack, tp->snd_recover))
1252 tp->snd_recover = th->th_ack - 1;
1253 tp->snd_una = th->th_ack;
1255 * pull snd_wl2 up to prevent seq wrap relative
1258 tp->snd_wl2 = th->th_ack;
1261 ND6_HINT(tp); /* some progress has been done */
1264 * If all outstanding data are acked, stop
1265 * retransmit timer, otherwise restart timer
1266 * using current (possibly backed-off) value.
1267 * If process is waiting for space,
1268 * wakeup/selwakeup/signal. If data
1269 * are ready to send, let tcp_output
1270 * decide between more output or persist.
1273 if (so->so_options & SO_DEBUG)
1274 tcp_trace(TA_INPUT, ostate, tp,
1275 (void *)tcp_saveipgen,
1279 if (tp->snd_una == tp->snd_max)
1280 callout_stop(tp->tt_rexmt);
1281 else if (!callout_active(tp->tt_persist))
1282 callout_reset(tp->tt_rexmt,
1284 tcp_timer_rexmt, tp);
1287 if (so->so_snd.sb_cc)
1288 (void) tcp_output(tp);
1291 } else if (th->th_ack == tp->snd_una &&
1292 LIST_EMPTY(&tp->t_segq) &&
1293 tlen <= sbspace(&so->so_rcv)) {
1294 KASSERT(headlocked, ("headlocked"));
1295 INP_INFO_WUNLOCK(&tcbinfo);
1298 * this is a pure, in-sequence data packet
1299 * with nothing on the reassembly queue and
1300 * we have enough buffer space to take it.
1302 /* Clean receiver SACK report if present */
1303 if (tp->sack_enable && tp->rcv_numsacks)
1304 tcp_clean_sackreport(tp);
1305 ++tcpstat.tcps_preddat;
1306 tp->rcv_nxt += tlen;
1308 * Pull snd_wl1 up to prevent seq wrap relative to
1311 tp->snd_wl1 = th->th_seq;
1313 * Pull rcv_up up to prevent seq wrap relative to
1316 tp->rcv_up = tp->rcv_nxt;
1317 tcpstat.tcps_rcvpack++;
1318 tcpstat.tcps_rcvbyte += tlen;
1319 ND6_HINT(tp); /* some progress has been done */
1322 if (so->so_options & SO_DEBUG)
1323 tcp_trace(TA_INPUT, ostate, tp,
1324 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1326 * Add data to socket buffer.
1328 SOCKBUF_LOCK(&so->so_rcv);
1329 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1332 m_adj(m, drop_hdrlen); /* delayed header drop */
1333 sbappendstream_locked(&so->so_rcv, m);
1335 sorwakeup_locked(so);
1336 if (DELAY_ACK(tp)) {
1337 tp->t_flags |= TF_DELACK;
1339 tp->t_flags |= TF_ACKNOW;
1347 * Calculate amount of space in receive window,
1348 * and then do TCP input processing.
1349 * Receive window is amount of space in rcv queue,
1350 * but not less than advertised window.
1354 win = sbspace(&so->so_rcv);
1357 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1360 switch (tp->t_state) {
1363 * If the state is SYN_RECEIVED:
1364 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1366 case TCPS_SYN_RECEIVED:
1367 if ((thflags & TH_ACK) &&
1368 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1369 SEQ_GT(th->th_ack, tp->snd_max))) {
1370 rstreason = BANDLIM_RST_OPENPORT;
1376 * If the state is SYN_SENT:
1377 * if seg contains an ACK, but not for our SYN, drop the input.
1378 * if seg contains a RST, then drop the connection.
1379 * if seg does not contain SYN, then drop it.
1380 * Otherwise this is an acceptable SYN segment
1381 * initialize tp->rcv_nxt and tp->irs
1382 * if seg contains ack then advance tp->snd_una
1383 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1384 * arrange for segment to be acked (eventually)
1385 * continue processing rest of data/controls, beginning with URG
1388 if ((thflags & TH_ACK) &&
1389 (SEQ_LEQ(th->th_ack, tp->iss) ||
1390 SEQ_GT(th->th_ack, tp->snd_max))) {
1391 rstreason = BANDLIM_UNLIMITED;
1394 if (thflags & TH_RST) {
1395 if (thflags & TH_ACK) {
1396 KASSERT(headlocked, ("tcp_input: after_listen"
1397 ": tcp_drop.2: head not locked"));
1398 tp = tcp_drop(tp, ECONNREFUSED);
1402 if ((thflags & TH_SYN) == 0)
1405 /* Initial send window, already scaled. */
1406 tp->snd_wnd = th->th_win;
1408 tp->irs = th->th_seq;
1410 if (thflags & TH_ACK) {
1411 tcpstat.tcps_connects++;
1415 mac_set_socket_peer_from_mbuf(m, so);
1418 /* Do window scaling on this connection? */
1419 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1420 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1421 tp->rcv_scale = tp->request_r_scale;
1423 tp->rcv_adv += tp->rcv_wnd;
1424 tp->snd_una++; /* SYN is acked */
1426 * If there's data, delay ACK; if there's also a FIN
1427 * ACKNOW will be turned on later.
1429 if (DELAY_ACK(tp) && tlen != 0)
1430 callout_reset(tp->tt_delack, tcp_delacktime,
1431 tcp_timer_delack, tp);
1433 tp->t_flags |= TF_ACKNOW;
1435 * Received <SYN,ACK> in SYN_SENT[*] state.
1437 * SYN_SENT --> ESTABLISHED
1438 * SYN_SENT* --> FIN_WAIT_1
1440 tp->t_starttime = ticks;
1441 if (tp->t_flags & TF_NEEDFIN) {
1442 tp->t_state = TCPS_FIN_WAIT_1;
1443 tp->t_flags &= ~TF_NEEDFIN;
1446 tp->t_state = TCPS_ESTABLISHED;
1447 callout_reset(tp->tt_keep, tcp_keepidle,
1448 tcp_timer_keep, tp);
1452 * Received initial SYN in SYN-SENT[*] state =>
1453 * simultaneous open. If segment contains CC option
1454 * and there is a cached CC, apply TAO test.
1455 * If it succeeds, connection is * half-synchronized.
1456 * Otherwise, do 3-way handshake:
1457 * SYN-SENT -> SYN-RECEIVED
1458 * SYN-SENT* -> SYN-RECEIVED*
1459 * If there was no CC option, clear cached CC value.
1461 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1462 callout_stop(tp->tt_rexmt);
1463 tp->t_state = TCPS_SYN_RECEIVED;
1469 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1471 INP_LOCK_ASSERT(inp);
1474 * Advance th->th_seq to correspond to first data byte.
1475 * If data, trim to stay within window,
1476 * dropping FIN if necessary.
1479 if (tlen > tp->rcv_wnd) {
1480 todrop = tlen - tp->rcv_wnd;
1484 tcpstat.tcps_rcvpackafterwin++;
1485 tcpstat.tcps_rcvbyteafterwin += todrop;
1487 tp->snd_wl1 = th->th_seq - 1;
1488 tp->rcv_up = th->th_seq;
1490 * Client side of transaction: already sent SYN and data.
1491 * If the remote host used T/TCP to validate the SYN,
1492 * our data will be ACK'd; if so, enter normal data segment
1493 * processing in the middle of step 5, ack processing.
1494 * Otherwise, goto step 6.
1496 if (thflags & TH_ACK)
1502 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1503 * do normal processing.
1505 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1509 case TCPS_TIME_WAIT:
1510 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1511 break; /* continue normal processing */
1515 * States other than LISTEN or SYN_SENT.
1516 * First check the RST flag and sequence number since reset segments
1517 * are exempt from the timestamp and connection count tests. This
1518 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1519 * below which allowed reset segments in half the sequence space
1520 * to fall though and be processed (which gives forged reset
1521 * segments with a random sequence number a 50 percent chance of
1522 * killing a connection).
1523 * Then check timestamp, if present.
1524 * Then check the connection count, if present.
1525 * Then check that at least some bytes of segment are within
1526 * receive window. If segment begins before rcv_nxt,
1527 * drop leading data (and SYN); if nothing left, just ack.
1530 * If the RST bit is set, check the sequence number to see
1531 * if this is a valid reset segment.
1533 * In all states except SYN-SENT, all reset (RST) segments
1534 * are validated by checking their SEQ-fields. A reset is
1535 * valid if its sequence number is in the window.
1536 * Note: this does not take into account delayed ACKs, so
1537 * we should test against last_ack_sent instead of rcv_nxt.
1538 * The sequence number in the reset segment is normally an
1539 * echo of our outgoing acknowlegement numbers, but some hosts
1540 * send a reset with the sequence number at the rightmost edge
1541 * of our receive window, and we have to handle this case.
1542 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1543 * that brute force RST attacks are possible. To combat this,
1544 * we use a much stricter check while in the ESTABLISHED state,
1545 * only accepting RSTs where the sequence number is equal to
1546 * last_ack_sent. In all other states (the states in which a
1547 * RST is more likely), the more permissive check is used.
1548 * If we have multiple segments in flight, the intial reset
1549 * segment sequence numbers will be to the left of last_ack_sent,
1550 * but they will eventually catch up.
1551 * In any case, it never made sense to trim reset segments to
1552 * fit the receive window since RFC 1122 says:
1553 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1555 * A TCP SHOULD allow a received RST segment to include data.
1558 * It has been suggested that a RST segment could contain
1559 * ASCII text that encoded and explained the cause of the
1560 * RST. No standard has yet been established for such
1563 * If the reset segment passes the sequence number test examine
1565 * SYN_RECEIVED STATE:
1566 * If passive open, return to LISTEN state.
1567 * If active open, inform user that connection was refused.
1568 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1569 * Inform user that connection was reset, and close tcb.
1570 * CLOSING, LAST_ACK STATES:
1573 * Drop the segment - see Stevens, vol. 2, p. 964 and
1576 if (thflags & TH_RST) {
1577 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1578 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1579 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1580 switch (tp->t_state) {
1582 case TCPS_SYN_RECEIVED:
1583 so->so_error = ECONNREFUSED;
1586 case TCPS_ESTABLISHED:
1587 if (tp->last_ack_sent != th->th_seq &&
1588 tcp_insecure_rst == 0) {
1589 tcpstat.tcps_badrst++;
1592 case TCPS_FIN_WAIT_1:
1593 case TCPS_FIN_WAIT_2:
1594 case TCPS_CLOSE_WAIT:
1595 so->so_error = ECONNRESET;
1597 tp->t_state = TCPS_CLOSED;
1598 tcpstat.tcps_drops++;
1599 KASSERT(headlocked, ("tcp_input: "
1600 "trimthenstep6: tcp_close: head not "
1607 KASSERT(headlocked, ("trimthenstep6: "
1608 "tcp_close.2: head not locked"));
1612 case TCPS_TIME_WAIT:
1613 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1622 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1623 * and it's less than ts_recent, drop it.
1625 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1626 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1628 /* Check to see if ts_recent is over 24 days old. */
1629 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1631 * Invalidate ts_recent. If this segment updates
1632 * ts_recent, the age will be reset later and ts_recent
1633 * will get a valid value. If it does not, setting
1634 * ts_recent to zero will at least satisfy the
1635 * requirement that zero be placed in the timestamp
1636 * echo reply when ts_recent isn't valid. The
1637 * age isn't reset until we get a valid ts_recent
1638 * because we don't want out-of-order segments to be
1639 * dropped when ts_recent is old.
1643 tcpstat.tcps_rcvduppack++;
1644 tcpstat.tcps_rcvdupbyte += tlen;
1645 tcpstat.tcps_pawsdrop++;
1653 * In the SYN-RECEIVED state, validate that the packet belongs to
1654 * this connection before trimming the data to fit the receive
1655 * window. Check the sequence number versus IRS since we know
1656 * the sequence numbers haven't wrapped. This is a partial fix
1657 * for the "LAND" DoS attack.
1659 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1660 rstreason = BANDLIM_RST_OPENPORT;
1664 todrop = tp->rcv_nxt - th->th_seq;
1666 if (thflags & TH_SYN) {
1676 * Following if statement from Stevens, vol. 2, p. 960.
1679 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1681 * Any valid FIN must be to the left of the window.
1682 * At this point the FIN must be a duplicate or out
1683 * of sequence; drop it.
1688 * Send an ACK to resynchronize and drop any data.
1689 * But keep on processing for RST or ACK.
1691 tp->t_flags |= TF_ACKNOW;
1693 tcpstat.tcps_rcvduppack++;
1694 tcpstat.tcps_rcvdupbyte += todrop;
1696 tcpstat.tcps_rcvpartduppack++;
1697 tcpstat.tcps_rcvpartdupbyte += todrop;
1699 drop_hdrlen += todrop; /* drop from the top afterwards */
1700 th->th_seq += todrop;
1702 if (th->th_urp > todrop)
1703 th->th_urp -= todrop;
1711 * If new data are received on a connection after the
1712 * user processes are gone, then RST the other end.
1714 if ((so->so_state & SS_NOFDREF) &&
1715 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1716 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1719 tcpstat.tcps_rcvafterclose++;
1720 rstreason = BANDLIM_UNLIMITED;
1725 * If segment ends after window, drop trailing data
1726 * (and PUSH and FIN); if nothing left, just ACK.
1728 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1730 tcpstat.tcps_rcvpackafterwin++;
1731 if (todrop >= tlen) {
1732 tcpstat.tcps_rcvbyteafterwin += tlen;
1734 * If a new connection request is received
1735 * while in TIME_WAIT, drop the old connection
1736 * and start over if the sequence numbers
1737 * are above the previous ones.
1739 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1740 if (thflags & TH_SYN &&
1741 tp->t_state == TCPS_TIME_WAIT &&
1742 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1743 KASSERT(headlocked, ("trimthenstep6: "
1744 "tcp_close.4: head not locked"));
1749 * If window is closed can only take segments at
1750 * window edge, and have to drop data and PUSH from
1751 * incoming segments. Continue processing, but
1752 * remember to ack. Otherwise, drop segment
1755 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1756 tp->t_flags |= TF_ACKNOW;
1757 tcpstat.tcps_rcvwinprobe++;
1761 tcpstat.tcps_rcvbyteafterwin += todrop;
1764 thflags &= ~(TH_PUSH|TH_FIN);
1768 * If last ACK falls within this segment's sequence numbers,
1769 * record its timestamp.
1771 * 1) That the test incorporates suggestions from the latest
1772 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1773 * 2) That updating only on newer timestamps interferes with
1774 * our earlier PAWS tests, so this check should be solely
1775 * predicated on the sequence space of this segment.
1776 * 3) That we modify the segment boundary check to be
1777 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1778 * instead of RFC1323's
1779 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1780 * This modified check allows us to overcome RFC1323's
1781 * limitations as described in Stevens TCP/IP Illustrated
1782 * Vol. 2 p.869. In such cases, we can still calculate the
1783 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1785 if ((to.to_flags & TOF_TS) != 0 &&
1786 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1787 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1788 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1789 tp->ts_recent_age = ticks;
1790 tp->ts_recent = to.to_tsval;
1794 * If a SYN is in the window, then this is an
1795 * error and we send an RST and drop the connection.
1797 if (thflags & TH_SYN) {
1798 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1799 "head not locked"));
1800 tp = tcp_drop(tp, ECONNRESET);
1801 rstreason = BANDLIM_UNLIMITED;
1806 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1807 * flag is on (half-synchronized state), then queue data for
1808 * later processing; else drop segment and return.
1810 if ((thflags & TH_ACK) == 0) {
1811 if (tp->t_state == TCPS_SYN_RECEIVED ||
1812 (tp->t_flags & TF_NEEDSYN))
1814 else if (tp->t_flags & TF_ACKNOW)
1823 switch (tp->t_state) {
1826 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1827 * ESTABLISHED state and continue processing.
1828 * The ACK was checked above.
1830 case TCPS_SYN_RECEIVED:
1832 tcpstat.tcps_connects++;
1834 /* Do window scaling? */
1835 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1836 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1837 tp->rcv_scale = tp->request_r_scale;
1838 tp->snd_wnd = tiwin;
1842 * SYN-RECEIVED -> ESTABLISHED
1843 * SYN-RECEIVED* -> FIN-WAIT-1
1845 tp->t_starttime = ticks;
1846 if (tp->t_flags & TF_NEEDFIN) {
1847 tp->t_state = TCPS_FIN_WAIT_1;
1848 tp->t_flags &= ~TF_NEEDFIN;
1850 tp->t_state = TCPS_ESTABLISHED;
1851 callout_reset(tp->tt_keep, tcp_keepidle,
1852 tcp_timer_keep, tp);
1855 * If segment contains data or ACK, will call tcp_reass()
1856 * later; if not, do so now to pass queued data to user.
1858 if (tlen == 0 && (thflags & TH_FIN) == 0)
1859 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1861 tp->snd_wl1 = th->th_seq - 1;
1865 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1866 * ACKs. If the ack is in the range
1867 * tp->snd_una < th->th_ack <= tp->snd_max
1868 * then advance tp->snd_una to th->th_ack and drop
1869 * data from the retransmission queue. If this ACK reflects
1870 * more up to date window information we update our window information.
1872 case TCPS_ESTABLISHED:
1873 case TCPS_FIN_WAIT_1:
1874 case TCPS_FIN_WAIT_2:
1875 case TCPS_CLOSE_WAIT:
1878 case TCPS_TIME_WAIT:
1879 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1880 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1881 tcpstat.tcps_rcvacktoomuch++;
1884 if (tp->sack_enable &&
1885 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1886 tcp_sack_doack(tp, &to, th->th_ack);
1887 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1888 if (tlen == 0 && tiwin == tp->snd_wnd) {
1889 tcpstat.tcps_rcvdupack++;
1891 * If we have outstanding data (other than
1892 * a window probe), this is a completely
1893 * duplicate ack (ie, window info didn't
1894 * change), the ack is the biggest we've
1895 * seen and we've seen exactly our rexmt
1896 * threshhold of them, assume a packet
1897 * has been dropped and retransmit it.
1898 * Kludge snd_nxt & the congestion
1899 * window so we send only this one
1902 * We know we're losing at the current
1903 * window size so do congestion avoidance
1904 * (set ssthresh to half the current window
1905 * and pull our congestion window back to
1906 * the new ssthresh).
1908 * Dup acks mean that packets have left the
1909 * network (they're now cached at the receiver)
1910 * so bump cwnd by the amount in the receiver
1911 * to keep a constant cwnd packets in the
1914 if (!callout_active(tp->tt_rexmt) ||
1915 th->th_ack != tp->snd_una)
1917 else if (++tp->t_dupacks > tcprexmtthresh ||
1918 ((tcp_do_newreno || tp->sack_enable) &&
1919 IN_FASTRECOVERY(tp))) {
1920 if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1924 * Compute the amount of data in flight first.
1925 * We can inject new data into the pipe iff
1926 * we have less than 1/2 the original window's
1927 * worth of data in flight.
1929 awnd = (tp->snd_nxt - tp->snd_fack) +
1930 tp->sackhint.sack_bytes_rexmit;
1931 if (awnd < tp->snd_ssthresh) {
1932 tp->snd_cwnd += tp->t_maxseg;
1933 if (tp->snd_cwnd > tp->snd_ssthresh)
1934 tp->snd_cwnd = tp->snd_ssthresh;
1937 tp->snd_cwnd += tp->t_maxseg;
1938 (void) tcp_output(tp);
1940 } else if (tp->t_dupacks == tcprexmtthresh) {
1941 tcp_seq onxt = tp->snd_nxt;
1945 * If we're doing sack, check to
1946 * see if we're already in sack
1947 * recovery. If we're not doing sack,
1948 * check to see if we're in newreno
1951 if (tp->sack_enable) {
1952 if (IN_FASTRECOVERY(tp)) {
1956 } else if (tcp_do_newreno) {
1957 if (SEQ_LEQ(th->th_ack,
1963 win = min(tp->snd_wnd, tp->snd_cwnd) /
1967 tp->snd_ssthresh = win * tp->t_maxseg;
1968 ENTER_FASTRECOVERY(tp);
1969 tp->snd_recover = tp->snd_max;
1970 callout_stop(tp->tt_rexmt);
1972 if (tp->sack_enable) {
1973 tcpstat.tcps_sack_recovery_episode++;
1974 tp->sack_newdata = tp->snd_nxt;
1975 tp->snd_cwnd = tp->t_maxseg;
1976 (void) tcp_output(tp);
1979 tp->snd_nxt = th->th_ack;
1980 tp->snd_cwnd = tp->t_maxseg;
1981 (void) tcp_output(tp);
1982 KASSERT(tp->snd_limited <= 2,
1983 ("tp->snd_limited too big"));
1984 tp->snd_cwnd = tp->snd_ssthresh +
1986 (tp->t_dupacks - tp->snd_limited);
1987 if (SEQ_GT(onxt, tp->snd_nxt))
1990 } else if (tcp_do_rfc3042) {
1991 u_long oldcwnd = tp->snd_cwnd;
1992 tcp_seq oldsndmax = tp->snd_max;
1995 KASSERT(tp->t_dupacks == 1 ||
1997 ("dupacks not 1 or 2"));
1998 if (tp->t_dupacks == 1)
1999 tp->snd_limited = 0;
2001 (tp->snd_nxt - tp->snd_una) +
2002 (tp->t_dupacks - tp->snd_limited) *
2004 (void) tcp_output(tp);
2005 sent = tp->snd_max - oldsndmax;
2006 if (sent > tp->t_maxseg) {
2007 KASSERT((tp->t_dupacks == 2 &&
2008 tp->snd_limited == 0) ||
2009 (sent == tp->t_maxseg + 1 &&
2010 tp->t_flags & TF_SENTFIN),
2012 tp->snd_limited = 2;
2013 } else if (sent > 0)
2015 tp->snd_cwnd = oldcwnd;
2023 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2026 * If the congestion window was inflated to account
2027 * for the other side's cached packets, retract it.
2029 if (tcp_do_newreno || tp->sack_enable) {
2030 if (IN_FASTRECOVERY(tp)) {
2031 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2032 if (tp->sack_enable)
2033 tcp_sack_partialack(tp, th);
2035 tcp_newreno_partial_ack(tp, th);
2038 * Out of fast recovery.
2039 * Window inflation should have left us
2040 * with approximately snd_ssthresh
2042 * But in case we would be inclined to
2043 * send a burst, better to do it via
2044 * the slow start mechanism.
2046 if (SEQ_GT(th->th_ack +
2049 tp->snd_cwnd = tp->snd_max -
2053 tp->snd_cwnd = tp->snd_ssthresh;
2057 if (tp->t_dupacks >= tcprexmtthresh &&
2058 tp->snd_cwnd > tp->snd_ssthresh)
2059 tp->snd_cwnd = tp->snd_ssthresh;
2063 * If we reach this point, ACK is not a duplicate,
2064 * i.e., it ACKs something we sent.
2066 if (tp->t_flags & TF_NEEDSYN) {
2068 * T/TCP: Connection was half-synchronized, and our
2069 * SYN has been ACK'd (so connection is now fully
2070 * synchronized). Go to non-starred state,
2071 * increment snd_una for ACK of SYN, and check if
2072 * we can do window scaling.
2074 tp->t_flags &= ~TF_NEEDSYN;
2076 /* Do window scaling? */
2077 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2078 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2079 tp->rcv_scale = tp->request_r_scale;
2080 /* Send window already scaled. */
2085 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2087 INP_LOCK_ASSERT(inp);
2089 acked = th->th_ack - tp->snd_una;
2090 tcpstat.tcps_rcvackpack++;
2091 tcpstat.tcps_rcvackbyte += acked;
2094 * If we just performed our first retransmit, and the ACK
2095 * arrives within our recovery window, then it was a mistake
2096 * to do the retransmit in the first place. Recover our
2097 * original cwnd and ssthresh, and proceed to transmit where
2100 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2101 ++tcpstat.tcps_sndrexmitbad;
2102 tp->snd_cwnd = tp->snd_cwnd_prev;
2103 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2104 tp->snd_recover = tp->snd_recover_prev;
2105 if (tp->t_flags & TF_WASFRECOVERY)
2106 ENTER_FASTRECOVERY(tp);
2107 tp->snd_nxt = tp->snd_max;
2108 tp->t_badrxtwin = 0; /* XXX probably not required */
2112 * If we have a timestamp reply, update smoothed
2113 * round trip time. If no timestamp is present but
2114 * transmit timer is running and timed sequence
2115 * number was acked, update smoothed round trip time.
2116 * Since we now have an rtt measurement, cancel the
2117 * timer backoff (cf., Phil Karn's retransmit alg.).
2118 * Recompute the initial retransmit timer.
2120 * Some boxes send broken timestamp replies
2121 * during the SYN+ACK phase, ignore
2122 * timestamps of 0 or we could calculate a
2123 * huge RTT and blow up the retransmit timer.
2125 if ((to.to_flags & TOF_TS) != 0 &&
2127 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2128 tp->t_rttlow = ticks - to.to_tsecr;
2129 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2130 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2131 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2132 tp->t_rttlow = ticks - tp->t_rtttime;
2133 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2135 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2138 * If all outstanding data is acked, stop retransmit
2139 * timer and remember to restart (more output or persist).
2140 * If there is more data to be acked, restart retransmit
2141 * timer, using current (possibly backed-off) value.
2143 if (th->th_ack == tp->snd_max) {
2144 callout_stop(tp->tt_rexmt);
2146 } else if (!callout_active(tp->tt_persist))
2147 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2148 tcp_timer_rexmt, tp);
2151 * If no data (only SYN) was ACK'd,
2152 * skip rest of ACK processing.
2158 * When new data is acked, open the congestion window.
2159 * If the window gives us less than ssthresh packets
2160 * in flight, open exponentially (maxseg per packet).
2161 * Otherwise open linearly: maxseg per window
2162 * (maxseg^2 / cwnd per packet).
2164 if ((!tcp_do_newreno && !tp->sack_enable) ||
2165 !IN_FASTRECOVERY(tp)) {
2166 register u_int cw = tp->snd_cwnd;
2167 register u_int incr = tp->t_maxseg;
2168 if (cw > tp->snd_ssthresh)
2169 incr = incr * incr / cw;
2170 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2172 SOCKBUF_LOCK(&so->so_snd);
2173 if (acked > so->so_snd.sb_cc) {
2174 tp->snd_wnd -= so->so_snd.sb_cc;
2175 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2178 sbdrop_locked(&so->so_snd, acked);
2179 tp->snd_wnd -= acked;
2182 sowwakeup_locked(so);
2183 /* detect una wraparound */
2184 if ((tcp_do_newreno || tp->sack_enable) &&
2185 !IN_FASTRECOVERY(tp) &&
2186 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2187 SEQ_LEQ(th->th_ack, tp->snd_recover))
2188 tp->snd_recover = th->th_ack - 1;
2189 if ((tcp_do_newreno || tp->sack_enable) &&
2190 IN_FASTRECOVERY(tp) &&
2191 SEQ_GEQ(th->th_ack, tp->snd_recover))
2192 EXIT_FASTRECOVERY(tp);
2193 tp->snd_una = th->th_ack;
2194 if (tp->sack_enable) {
2195 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2196 tp->snd_recover = tp->snd_una;
2198 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2199 tp->snd_nxt = tp->snd_una;
2201 switch (tp->t_state) {
2204 * In FIN_WAIT_1 STATE in addition to the processing
2205 * for the ESTABLISHED state if our FIN is now acknowledged
2206 * then enter FIN_WAIT_2.
2208 case TCPS_FIN_WAIT_1:
2209 if (ourfinisacked) {
2211 * If we can't receive any more
2212 * data, then closing user can proceed.
2213 * Starting the timer is contrary to the
2214 * specification, but if we don't get a FIN
2215 * we'll hang forever.
2218 * we should release the tp also, and use a
2221 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2222 soisdisconnected(so);
2223 callout_reset(tp->tt_2msl, tcp_maxidle,
2224 tcp_timer_2msl, tp);
2226 tp->t_state = TCPS_FIN_WAIT_2;
2231 * In CLOSING STATE in addition to the processing for
2232 * the ESTABLISHED state if the ACK acknowledges our FIN
2233 * then enter the TIME-WAIT state, otherwise ignore
2237 if (ourfinisacked) {
2238 KASSERT(headlocked, ("tcp_input: process_ACK: "
2239 "head not locked"));
2241 INP_INFO_WUNLOCK(&tcbinfo);
2248 * In LAST_ACK, we may still be waiting for data to drain
2249 * and/or to be acked, as well as for the ack of our FIN.
2250 * If our FIN is now acknowledged, delete the TCB,
2251 * enter the closed state and return.
2254 if (ourfinisacked) {
2255 KASSERT(headlocked, ("tcp_input: process_ACK:"
2256 " tcp_close: head not locked"));
2263 * In TIME_WAIT state the only thing that should arrive
2264 * is a retransmission of the remote FIN. Acknowledge
2265 * it and restart the finack timer.
2267 case TCPS_TIME_WAIT:
2268 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2269 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2270 tcp_timer_2msl, tp);
2276 KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2277 INP_LOCK_ASSERT(inp);
2280 * Update window information.
2281 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2283 if ((thflags & TH_ACK) &&
2284 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2285 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2286 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2287 /* keep track of pure window updates */
2289 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2290 tcpstat.tcps_rcvwinupd++;
2291 tp->snd_wnd = tiwin;
2292 tp->snd_wl1 = th->th_seq;
2293 tp->snd_wl2 = th->th_ack;
2294 if (tp->snd_wnd > tp->max_sndwnd)
2295 tp->max_sndwnd = tp->snd_wnd;
2300 * Process segments with URG.
2302 if ((thflags & TH_URG) && th->th_urp &&
2303 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2305 * This is a kludge, but if we receive and accept
2306 * random urgent pointers, we'll crash in
2307 * soreceive. It's hard to imagine someone
2308 * actually wanting to send this much urgent data.
2310 SOCKBUF_LOCK(&so->so_rcv);
2311 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2312 th->th_urp = 0; /* XXX */
2313 thflags &= ~TH_URG; /* XXX */
2314 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2315 goto dodata; /* XXX */
2318 * If this segment advances the known urgent pointer,
2319 * then mark the data stream. This should not happen
2320 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2321 * a FIN has been received from the remote side.
2322 * In these states we ignore the URG.
2324 * According to RFC961 (Assigned Protocols),
2325 * the urgent pointer points to the last octet
2326 * of urgent data. We continue, however,
2327 * to consider it to indicate the first octet
2328 * of data past the urgent section as the original
2329 * spec states (in one of two places).
2331 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2332 tp->rcv_up = th->th_seq + th->th_urp;
2333 so->so_oobmark = so->so_rcv.sb_cc +
2334 (tp->rcv_up - tp->rcv_nxt) - 1;
2335 if (so->so_oobmark == 0)
2336 so->so_rcv.sb_state |= SBS_RCVATMARK;
2338 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2340 SOCKBUF_UNLOCK(&so->so_rcv);
2342 * Remove out of band data so doesn't get presented to user.
2343 * This can happen independent of advancing the URG pointer,
2344 * but if two URG's are pending at once, some out-of-band
2345 * data may creep in... ick.
2347 if (th->th_urp <= (u_long)tlen &&
2348 !(so->so_options & SO_OOBINLINE)) {
2349 /* hdr drop is delayed */
2350 tcp_pulloutofband(so, th, m, drop_hdrlen);
2354 * If no out of band data is expected,
2355 * pull receive urgent pointer along
2356 * with the receive window.
2358 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2359 tp->rcv_up = tp->rcv_nxt;
2362 KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2363 INP_LOCK_ASSERT(inp);
2366 * Process the segment text, merging it into the TCP sequencing queue,
2367 * and arranging for acknowledgment of receipt if necessary.
2368 * This process logically involves adjusting tp->rcv_wnd as data
2369 * is presented to the user (this happens in tcp_usrreq.c,
2370 * case PRU_RCVD). If a FIN has already been received on this
2371 * connection then we just ignore the text.
2373 if ((tlen || (thflags & TH_FIN)) &&
2374 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2375 tcp_seq save_start = th->th_seq;
2376 tcp_seq save_end = th->th_seq + tlen;
2377 m_adj(m, drop_hdrlen); /* delayed header drop */
2379 * Insert segment which includes th into TCP reassembly queue
2380 * with control block tp. Set thflags to whether reassembly now
2381 * includes a segment with FIN. This handles the common case
2382 * inline (segment is the next to be received on an established
2383 * connection, and the queue is empty), avoiding linkage into
2384 * and removal from the queue and repetition of various
2386 * Set DELACK for segments received in order, but ack
2387 * immediately when segments are out of order (so
2388 * fast retransmit can work).
2390 if (th->th_seq == tp->rcv_nxt &&
2391 LIST_EMPTY(&tp->t_segq) &&
2392 TCPS_HAVEESTABLISHED(tp->t_state)) {
2394 tp->t_flags |= TF_DELACK;
2396 tp->t_flags |= TF_ACKNOW;
2397 tp->rcv_nxt += tlen;
2398 thflags = th->th_flags & TH_FIN;
2399 tcpstat.tcps_rcvpack++;
2400 tcpstat.tcps_rcvbyte += tlen;
2402 SOCKBUF_LOCK(&so->so_rcv);
2403 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2406 sbappendstream_locked(&so->so_rcv, m);
2407 sorwakeup_locked(so);
2409 thflags = tcp_reass(tp, th, &tlen, m);
2410 tp->t_flags |= TF_ACKNOW;
2412 if (tlen > 0 && tp->sack_enable)
2413 tcp_update_sack_list(tp, save_start, save_end);
2415 * Note the amount of data that peer has sent into
2416 * our window, in order to estimate the sender's
2419 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2426 * If FIN is received ACK the FIN and let the user know
2427 * that the connection is closing.
2429 if (thflags & TH_FIN) {
2430 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2433 * If connection is half-synchronized
2434 * (ie NEEDSYN flag on) then delay ACK,
2435 * so it may be piggybacked when SYN is sent.
2436 * Otherwise, since we received a FIN then no
2437 * more input can be expected, send ACK now.
2439 if (tp->t_flags & TF_NEEDSYN)
2440 tp->t_flags |= TF_DELACK;
2442 tp->t_flags |= TF_ACKNOW;
2445 switch (tp->t_state) {
2448 * In SYN_RECEIVED and ESTABLISHED STATES
2449 * enter the CLOSE_WAIT state.
2451 case TCPS_SYN_RECEIVED:
2452 tp->t_starttime = ticks;
2454 case TCPS_ESTABLISHED:
2455 tp->t_state = TCPS_CLOSE_WAIT;
2459 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2460 * enter the CLOSING state.
2462 case TCPS_FIN_WAIT_1:
2463 tp->t_state = TCPS_CLOSING;
2467 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2468 * starting the time-wait timer, turning off the other
2471 case TCPS_FIN_WAIT_2:
2472 KASSERT(headlocked == 1, ("tcp_input: dodata: "
2473 "TCP_FIN_WAIT_2: head not locked"));
2475 INP_INFO_WUNLOCK(&tcbinfo);
2479 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2481 case TCPS_TIME_WAIT:
2482 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2483 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2484 tcp_timer_2msl, tp);
2488 INP_INFO_WUNLOCK(&tcbinfo);
2491 if (so->so_options & SO_DEBUG)
2492 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2497 * Return any desired output.
2499 if (needoutput || (tp->t_flags & TF_ACKNOW))
2500 (void) tcp_output(tp);
2503 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2504 INP_LOCK_ASSERT(inp);
2505 if (tp->t_flags & TF_DELACK) {
2506 tp->t_flags &= ~TF_DELACK;
2507 callout_reset(tp->tt_delack, tcp_delacktime,
2508 tcp_timer_delack, tp);
2514 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2516 * Generate an ACK dropping incoming segment if it occupies
2517 * sequence space, where the ACK reflects our state.
2519 * We can now skip the test for the RST flag since all
2520 * paths to this code happen after packets containing
2521 * RST have been dropped.
2523 * In the SYN-RECEIVED state, don't send an ACK unless the
2524 * segment we received passes the SYN-RECEIVED ACK test.
2525 * If it fails send a RST. This breaks the loop in the
2526 * "LAND" DoS attack, and also prevents an ACK storm
2527 * between two listening ports that have been sent forged
2528 * SYN segments, each with the source address of the other.
2530 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2531 (SEQ_GT(tp->snd_una, th->th_ack) ||
2532 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2533 rstreason = BANDLIM_RST_OPENPORT;
2537 if (so->so_options & SO_DEBUG)
2538 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2541 KASSERT(headlocked, ("headlocked should be 1"));
2542 INP_INFO_WUNLOCK(&tcbinfo);
2543 tp->t_flags |= TF_ACKNOW;
2544 (void) tcp_output(tp);
2550 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2552 * Generate a RST, dropping incoming segment.
2553 * Make ACK acceptable to originator of segment.
2554 * Don't bother to respond if destination was broadcast/multicast.
2556 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2559 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2560 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2563 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2564 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2565 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2566 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2569 /* IPv6 anycast check is done at tcp6_input() */
2572 * Perform bandwidth limiting.
2574 if (badport_bandlim(rstreason) < 0)
2578 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2579 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2583 if (thflags & TH_ACK)
2584 /* mtod() below is safe as long as hdr dropping is delayed */
2585 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2588 if (thflags & TH_SYN)
2590 /* mtod() below is safe as long as hdr dropping is delayed */
2591 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2592 (tcp_seq)0, TH_RST|TH_ACK);
2598 INP_INFO_WUNLOCK(&tcbinfo);
2603 * Drop space held by incoming segment and return.
2606 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2607 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2613 INP_INFO_WUNLOCK(&tcbinfo);
2619 * Parse TCP options and place in tcpopt.
2622 tcp_dooptions(to, cp, cnt, flags)
2631 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2633 if (opt == TCPOPT_EOL)
2635 if (opt == TCPOPT_NOP)
2641 if (optlen < 2 || optlen > cnt)
2646 if (optlen != TCPOLEN_MAXSEG)
2648 if (!(flags & TO_SYN))
2650 to->to_flags |= TOF_MSS;
2651 bcopy((char *)cp + 2,
2652 (char *)&to->to_mss, sizeof(to->to_mss));
2653 to->to_mss = ntohs(to->to_mss);
2656 if (optlen != TCPOLEN_WINDOW)
2658 if (!(flags & TO_SYN))
2660 to->to_flags |= TOF_SCALE;
2661 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2663 case TCPOPT_TIMESTAMP:
2664 if (optlen != TCPOLEN_TIMESTAMP)
2666 to->to_flags |= TOF_TS;
2667 bcopy((char *)cp + 2,
2668 (char *)&to->to_tsval, sizeof(to->to_tsval));
2669 to->to_tsval = ntohl(to->to_tsval);
2670 bcopy((char *)cp + 6,
2671 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2672 to->to_tsecr = ntohl(to->to_tsecr);
2674 #ifdef TCP_SIGNATURE
2676 * XXX In order to reply to a host which has set the
2677 * TCP_SIGNATURE option in its initial SYN, we have to
2678 * record the fact that the option was observed here
2679 * for the syncache code to perform the correct response.
2681 case TCPOPT_SIGNATURE:
2682 if (optlen != TCPOLEN_SIGNATURE)
2684 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2687 case TCPOPT_SACK_PERMITTED:
2688 if (optlen != TCPOLEN_SACK_PERMITTED)
2690 if (!(flags & TO_SYN))
2694 to->to_flags |= TOF_SACK;
2697 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2699 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2700 to->to_sacks = cp + 2;
2701 tcpstat.tcps_sack_rcv_blocks++;
2710 * Pull out of band byte out of a segment so
2711 * it doesn't appear in the user's data queue.
2712 * It is still reflected in the segment length for
2713 * sequencing purposes.
2716 tcp_pulloutofband(so, th, m, off)
2719 register struct mbuf *m;
2720 int off; /* delayed to be droped hdrlen */
2722 int cnt = off + th->th_urp - 1;
2725 if (m->m_len > cnt) {
2726 char *cp = mtod(m, caddr_t) + cnt;
2727 struct tcpcb *tp = sototcpcb(so);
2730 tp->t_oobflags |= TCPOOB_HAVEDATA;
2731 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2733 if (m->m_flags & M_PKTHDR)
2742 panic("tcp_pulloutofband");
2746 * Collect new round-trip time estimate
2747 * and update averages and current timeout.
2750 tcp_xmit_timer(tp, rtt)
2751 register struct tcpcb *tp;
2756 INP_LOCK_ASSERT(tp->t_inpcb);
2758 tcpstat.tcps_rttupdated++;
2760 if (tp->t_srtt != 0) {
2762 * srtt is stored as fixed point with 5 bits after the
2763 * binary point (i.e., scaled by 8). The following magic
2764 * is equivalent to the smoothing algorithm in rfc793 with
2765 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2766 * point). Adjust rtt to origin 0.
2768 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2769 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2771 if ((tp->t_srtt += delta) <= 0)
2775 * We accumulate a smoothed rtt variance (actually, a
2776 * smoothed mean difference), then set the retransmit
2777 * timer to smoothed rtt + 4 times the smoothed variance.
2778 * rttvar is stored as fixed point with 4 bits after the
2779 * binary point (scaled by 16). The following is
2780 * equivalent to rfc793 smoothing with an alpha of .75
2781 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2782 * rfc793's wired-in beta.
2786 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2787 if ((tp->t_rttvar += delta) <= 0)
2789 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2790 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2793 * No rtt measurement yet - use the unsmoothed rtt.
2794 * Set the variance to half the rtt (so our first
2795 * retransmit happens at 3*rtt).
2797 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2798 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2799 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2805 * the retransmit should happen at rtt + 4 * rttvar.
2806 * Because of the way we do the smoothing, srtt and rttvar
2807 * will each average +1/2 tick of bias. When we compute
2808 * the retransmit timer, we want 1/2 tick of rounding and
2809 * 1 extra tick because of +-1/2 tick uncertainty in the
2810 * firing of the timer. The bias will give us exactly the
2811 * 1.5 tick we need. But, because the bias is
2812 * statistical, we have to test that we don't drop below
2813 * the minimum feasible timer (which is 2 ticks).
2815 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2816 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2819 * We received an ack for a packet that wasn't retransmitted;
2820 * it is probably safe to discard any error indications we've
2821 * received recently. This isn't quite right, but close enough
2822 * for now (a route might have failed after we sent a segment,
2823 * and the return path might not be symmetrical).
2825 tp->t_softerror = 0;
2829 * Determine a reasonable value for maxseg size.
2830 * If the route is known, check route for mtu.
2831 * If none, use an mss that can be handled on the outgoing
2832 * interface without forcing IP to fragment; if bigger than
2833 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2834 * to utilize large mbufs. If no route is found, route has no mtu,
2835 * or the destination isn't local, use a default, hopefully conservative
2836 * size (usually 512 or the default IP max size, but no more than the mtu
2837 * of the interface), as we can't discover anything about intervening
2838 * gateways or networks. We also initialize the congestion/slow start
2839 * window to be a single segment if the destination isn't local.
2840 * While looking at the routing entry, we also initialize other path-dependent
2841 * parameters from pre-set or cached values in the routing entry.
2843 * Also take into account the space needed for options that we
2844 * send regularly. Make maxseg shorter by that amount to assure
2845 * that we can send maxseg amount of data even when the options
2846 * are present. Store the upper limit of the length of options plus
2850 * In case of T/TCP, we call this routine during implicit connection
2851 * setup as well (offer = -1), to initialize maxseg from the cached
2854 * NOTE that this routine is only called when we process an incoming
2855 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2865 struct inpcb *inp = tp->t_inpcb;
2867 struct hc_metrics_lite metrics;
2868 int origoffer = offer;
2871 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2872 size_t min_protoh = isipv6 ?
2873 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2874 sizeof (struct tcpiphdr);
2876 const size_t min_protoh = sizeof(struct tcpiphdr);
2882 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags);
2883 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2887 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags);
2888 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2890 so = inp->inp_socket;
2893 * no route to sender, stay with default mss and return
2898 /* what have we got? */
2902 * Offer == 0 means that there was no MSS on the SYN
2903 * segment, in this case we use tcp_mssdflt.
2907 isipv6 ? tcp_v6mssdflt :
2914 * Offer == -1 means that we didn't receive SYN yet.
2920 * Prevent DoS attack with too small MSS. Round up
2921 * to at least minmss.
2923 offer = max(offer, tcp_minmss);
2925 * Sanity check: make sure that maxopd will be large
2926 * enough to allow some data on segments even if the
2927 * all the option space is used (40bytes). Otherwise
2928 * funny things may happen in tcp_output.
2930 offer = max(offer, 64);
2934 * rmx information is now retrieved from tcp_hostcache
2936 tcp_hc_get(&inp->inp_inc, &metrics);
2939 * if there's a discovered mtu int tcp hostcache, use it
2940 * else, use the link mtu.
2942 if (metrics.rmx_mtu)
2943 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2947 mss = maxmtu - min_protoh;
2948 if (!path_mtu_discovery &&
2949 !in6_localaddr(&inp->in6p_faddr))
2950 mss = min(mss, tcp_v6mssdflt);
2954 mss = maxmtu - min_protoh;
2955 if (!path_mtu_discovery &&
2956 !in_localaddr(inp->inp_faddr))
2957 mss = min(mss, tcp_mssdflt);
2960 mss = min(mss, offer);
2963 * maxopd stores the maximum length of data AND options
2964 * in a segment; maxseg is the amount of data in a normal
2965 * segment. We need to store this value (maxopd) apart
2966 * from maxseg, because now every segment carries options
2967 * and thus we normally have somewhat less data in segments.
2972 * origoffer==-1 indicates, that no segments were received yet.
2973 * In this case we just guess.
2975 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2977 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2978 mss -= TCPOLEN_TSTAMP_APPA;
2981 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2983 mss &= ~(MCLBYTES-1);
2986 mss = mss / MCLBYTES * MCLBYTES;
2991 * If there's a pipesize, change the socket buffer to that size,
2992 * don't change if sb_hiwat is different than default (then it
2993 * has been changed on purpose with setsockopt).
2994 * Make the socket buffers an integral number of mss units;
2995 * if the mss is larger than the socket buffer, decrease the mss.
2997 SOCKBUF_LOCK(&so->so_snd);
2998 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2999 bufsize = metrics.rmx_sendpipe;
3001 bufsize = so->so_snd.sb_hiwat;
3005 bufsize = roundup(bufsize, mss);
3006 if (bufsize > sb_max)
3008 if (bufsize > so->so_snd.sb_hiwat)
3009 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3011 SOCKBUF_UNLOCK(&so->so_snd);
3014 SOCKBUF_LOCK(&so->so_rcv);
3015 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3016 bufsize = metrics.rmx_recvpipe;
3018 bufsize = so->so_rcv.sb_hiwat;
3019 if (bufsize > mss) {
3020 bufsize = roundup(bufsize, mss);
3021 if (bufsize > sb_max)
3023 if (bufsize > so->so_rcv.sb_hiwat)
3024 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3026 SOCKBUF_UNLOCK(&so->so_rcv);
3028 * While we're here, check the others too
3030 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3032 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3033 tcpstat.tcps_usedrtt++;
3034 if (metrics.rmx_rttvar) {
3035 tp->t_rttvar = metrics.rmx_rttvar;
3036 tcpstat.tcps_usedrttvar++;
3038 /* default variation is +- 1 rtt */
3040 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3042 TCPT_RANGESET(tp->t_rxtcur,
3043 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3044 tp->t_rttmin, TCPTV_REXMTMAX);
3046 if (metrics.rmx_ssthresh) {
3048 * There's some sort of gateway or interface
3049 * buffer limit on the path. Use this to set
3050 * the slow start threshhold, but set the
3051 * threshold to no less than 2*mss.
3053 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3054 tcpstat.tcps_usedssthresh++;
3056 if (metrics.rmx_bandwidth)
3057 tp->snd_bandwidth = metrics.rmx_bandwidth;
3060 * Set the slow-start flight size depending on whether this
3061 * is a local network or not.
3063 * Extend this so we cache the cwnd too and retrieve it here.
3064 * Make cwnd even bigger than RFC3390 suggests but only if we
3065 * have previous experience with the remote host. Be careful
3066 * not make cwnd bigger than remote receive window or our own
3067 * send socket buffer. Maybe put some additional upper bound
3068 * on the retrieved cwnd. Should do incremental updates to
3069 * hostcache when cwnd collapses so next connection doesn't
3070 * overloads the path again.
3072 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3073 * We currently check only in syncache_socket for that.
3075 #define TCP_METRICS_CWND
3076 #ifdef TCP_METRICS_CWND
3077 if (metrics.rmx_cwnd)
3078 tp->snd_cwnd = max(mss,
3079 min(metrics.rmx_cwnd / 2,
3080 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3084 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3086 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3087 (!isipv6 && in_localaddr(inp->inp_faddr)))
3089 else if (in_localaddr(inp->inp_faddr))
3091 tp->snd_cwnd = mss * ss_fltsz_local;
3093 tp->snd_cwnd = mss * ss_fltsz;
3095 /* Check the interface for TSO capabilities. */
3096 if (mtuflags & CSUM_TSO)
3097 tp->t_flags |= TF_TSO;
3101 * Determine the MSS option to send on an outgoing SYN.
3105 struct in_conninfo *inc;
3112 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3115 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3119 mss = tcp_v6mssdflt;
3120 maxmtu = tcp_maxmtu6(inc, NULL);
3121 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3122 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3127 maxmtu = tcp_maxmtu(inc, NULL);
3128 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3129 min_protoh = sizeof(struct tcpiphdr);
3131 if (maxmtu && thcmtu)
3132 mss = min(maxmtu, thcmtu) - min_protoh;
3133 else if (maxmtu || thcmtu)
3134 mss = max(maxmtu, thcmtu) - min_protoh;
3141 * On a partial ack arrives, force the retransmission of the
3142 * next unacknowledged segment. Do not clear tp->t_dupacks.
3143 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3147 tcp_newreno_partial_ack(tp, th)
3151 tcp_seq onxt = tp->snd_nxt;
3152 u_long ocwnd = tp->snd_cwnd;
3154 callout_stop(tp->tt_rexmt);
3156 tp->snd_nxt = th->th_ack;
3158 * Set snd_cwnd to one segment beyond acknowledged offset.
3159 * (tp->snd_una has not yet been updated when this function is called.)
3161 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3162 tp->t_flags |= TF_ACKNOW;
3163 (void) tcp_output(tp);
3164 tp->snd_cwnd = ocwnd;
3165 if (SEQ_GT(onxt, tp->snd_nxt))
3168 * Partial window deflation. Relies on fact that tp->snd_una
3171 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3172 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3175 tp->snd_cwnd += tp->t_maxseg;
3179 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3180 * looking for a pcb in the listen state. Returns 0 otherwise.
3183 tcp_timewait(inp, to, th, m, tlen)
3194 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3196 const int isipv6 = 0;
3199 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */
3200 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3201 INP_LOCK_ASSERT(inp);
3204 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
3205 * still present. This is undesirable, but temporarily necessary
3206 * until we work out how to handle inpcb's who's timewait state has
3213 thflags = th->th_flags;
3216 * NOTE: for FIN_WAIT_2 (to be added later),
3217 * must validate sequence number before accepting RST
3221 * If the segment contains RST:
3222 * Drop the segment - see Stevens, vol. 2, p. 964 and
3225 if (thflags & TH_RST)
3229 /* PAWS not needed at the moment */
3231 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3232 * and it's less than ts_recent, drop it.
3234 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3235 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3236 if ((thflags & TH_ACK) == 0)
3241 * ts_recent is never updated because we never accept new segments.
3246 * If a new connection request is received
3247 * while in TIME_WAIT, drop the old connection
3248 * and start over if the sequence numbers
3249 * are above the previous ones.
3251 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3257 * Drop the the segment if it does not contain an ACK.
3259 if ((thflags & TH_ACK) == 0)
3263 * Reset the 2MSL timer if this is a duplicate FIN.
3265 if (thflags & TH_FIN) {
3266 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3267 if (seq + 1 == tw->rcv_nxt)
3268 tcp_timer_2msl_reset(tw, 1);
3272 * Acknowledge the segment if it has data or is not a duplicate ACK.
3274 if (thflags != TH_ACK || tlen != 0 ||
3275 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3276 tcp_twrespond(tw, TH_ACK);
3280 * Generate a RST, dropping incoming segment.
3281 * Make ACK acceptable to originator of segment.
3282 * Don't bother to respond if destination was broadcast/multicast.
3284 if (m->m_flags & (M_BCAST|M_MCAST))
3287 struct ip6_hdr *ip6;
3289 /* IPv6 anycast check is done at tcp6_input() */
3290 ip6 = mtod(m, struct ip6_hdr *);
3291 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3292 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3297 ip = mtod(m, struct ip *);
3298 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3299 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3300 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3301 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3304 if (thflags & TH_ACK) {
3306 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3308 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3310 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
projects / FreeBSD / FreeBSD.git / blob