xref: /csrg-svn/sys/netinet/tcp_input.c (revision 5307)
1*5307Sroot /*	tcp_input.c	1.45	81/12/23	*/
24601Swnj 
34601Swnj #include "../h/param.h"
44601Swnj #include "../h/systm.h"
54663Swnj #include "../h/mbuf.h"
65085Swnj #include "../h/protosw.h"
74663Swnj #include "../h/socket.h"
84803Swnj #include "../h/socketvar.h"
95085Swnj #include "../net/in.h"
105085Swnj #include "../net/in_pcb.h"
115085Swnj #include "../net/in_systm.h"
125085Swnj #include "../net/if.h"
134803Swnj #include "../net/ip.h"
144899Swnj #include "../net/ip_var.h"
154803Swnj #include "../net/tcp.h"
164803Swnj #include "../net/tcp_fsm.h"
175085Swnj #include "../net/tcp_seq.h"
185085Swnj #include "../net/tcp_timer.h"
194803Swnj #include "../net/tcp_var.h"
205085Swnj #include "../net/tcpip.h"
215267Sroot #include "../net/tcp_debug.h"
225109Swnj #include "../errno.h"
234601Swnj 
245300Sroot int	tcpprintfs = 0;
254679Swnj int	tcpcksum = 1;
265244Sroot struct	sockaddr_in tcp_in = { AF_INET };
275267Sroot struct	tcpiphdr tcp_saveti;
284601Swnj 
295267Sroot struct	tcpcb *tcp_newtcpcb();
305065Swnj /*
315065Swnj  * TCP input routine, follows pages 65-76 of the
325065Swnj  * protocol specification dated September, 1981 very closely.
335065Swnj  */
344924Swnj tcp_input(m0)
354924Swnj 	struct mbuf *m0;
364601Swnj {
374924Swnj 	register struct tcpiphdr *ti;
384924Swnj 	struct inpcb *inp;
394924Swnj 	register struct mbuf *m;
404924Swnj 	int len, tlen, off;
414924Swnj 	register struct tcpcb *tp;
424924Swnj 	register int tiflags;
434803Swnj 	struct socket *so;
445109Swnj 	int todrop, acked;
455267Sroot 	short ostate;
464924Swnj 
474601Swnj COUNT(TCP_INPUT);
484924Swnj 	/*
495244Sroot 	 * Get IP and TCP header together in first mbuf.
505244Sroot 	 * Note: IP leaves IP header in first mbuf.
514924Swnj 	 */
524924Swnj 	m = m0;
535020Sroot 	ti = mtod(m, struct tcpiphdr *);
545244Sroot 	if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2))
555208Swnj 		ip_stripoptions((struct ip *)ti, (struct mbuf *)0);
56*5307Sroot 	if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) {
57*5307Sroot 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
585085Swnj 			tcpstat.tcps_hdrops++;
59*5307Sroot 			return;
605085Swnj 		}
615085Swnj 		ti = mtod(m, struct tcpiphdr *);
625085Swnj 	}
634601Swnj 
644601Swnj 	/*
655244Sroot 	 * Checksum extended TCP header and data.
664601Swnj 	 */
674924Swnj 	tlen = ((struct ip *)ti)->ip_len;
684924Swnj 	len = sizeof (struct ip) + tlen;
694679Swnj 	if (tcpcksum) {
704924Swnj 		ti->ti_next = ti->ti_prev = 0;
714924Swnj 		ti->ti_x1 = 0;
725223Swnj 		ti->ti_len = (u_short)tlen;
735223Swnj #if vax
745223Swnj 		ti->ti_len = htons(ti->ti_len);
755223Swnj #endif
765231Swnj 		if (ti->ti_sum = in_cksum(m, len)) {
774924Swnj 			tcpstat.tcps_badsum++;
785065Swnj 			printf("tcp cksum %x\n", ti->ti_sum);
795085Swnj 			goto drop;
804601Swnj 		}
814601Swnj 	}
824601Swnj 
834601Swnj 	/*
845244Sroot 	 * Check that TCP offset makes sense,
855244Sroot 	 * process TCP options and adjust length.
864601Swnj 	 */
874924Swnj 	off = ti->ti_off << 2;
885231Swnj 	if (off < sizeof (struct tcphdr) || off > tlen) {
894924Swnj 		tcpstat.tcps_badoff++;
905085Swnj 		goto drop;
914924Swnj 	}
924924Swnj 	ti->ti_len = tlen - off;
935085Swnj #if 0
945231Swnj 	if (off > sizeof (struct tcphdr))
955085Swnj 		tcp_options(ti);
965085Swnj #endif
975065Swnj 	tiflags = ti->ti_flags;
984924Swnj 
995231Swnj #if vax
1004924Swnj 	/*
1015244Sroot 	 * Convert TCP protocol specific fields to host format.
1025085Swnj 	 */
1035085Swnj 	ti->ti_seq = ntohl(ti->ti_seq);
1045085Swnj 	ti->ti_ack = ntohl(ti->ti_ack);
1055085Swnj 	ti->ti_win = ntohs(ti->ti_win);
1065085Swnj 	ti->ti_urp = ntohs(ti->ti_urp);
1075231Swnj #endif
1085085Swnj 
1095085Swnj 	/*
1104924Swnj 	 * Locate pcb for segment.
1114924Swnj 	 */
1125065Swnj 	inp = in_pcblookup
1135065Swnj 		(&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport);
1145065Swnj 
1155065Swnj 	/*
1165065Swnj 	 * If the state is CLOSED (i.e., TCB does not exist) then
1175244Sroot 	 * all data in the incoming segment is discarded.
1185065Swnj 	 */
1195300Sroot 	if (inp == 0)
1205085Swnj 		goto dropwithreset;
1215065Swnj 	tp = intotcpcb(inp);
1225300Sroot 	if (tp == 0)
1235085Swnj 		goto dropwithreset;
1245109Swnj 	so = inp->inp_socket;
1255267Sroot 	if (so->so_options & SO_DEBUG) {
1265267Sroot 		ostate = tp->t_state;
1275267Sroot 		tcp_saveti = *ti;
1285267Sroot 	}
1294601Swnj 
1304601Swnj 	/*
1315162Swnj 	 * Segment received on connection.
1325162Swnj 	 * Reset idle time and keep-alive timer.
1335162Swnj 	 */
1345162Swnj 	tp->t_idle = 0;
1355162Swnj 	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
1365162Swnj 
1375162Swnj 	/*
1385085Swnj 	 * Calculate amount of space in receive window,
1395085Swnj 	 * and then do TCP input processing.
1404601Swnj 	 */
1415085Swnj 	tp->rcv_wnd = sbspace(&so->so_rcv);
1425231Swnj 	if (tp->rcv_wnd < 0)
1435231Swnj 		tp->rcv_wnd = 0;
1444601Swnj 
1454601Swnj 	switch (tp->t_state) {
1464601Swnj 
1475065Swnj 	/*
1485065Swnj 	 * If the state is LISTEN then ignore segment if it contains an RST.
1495065Swnj 	 * If the segment contains an ACK then it is bad and send a RST.
1505065Swnj 	 * If it does not contain a SYN then it is not interesting; drop it.
1515085Swnj 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
1525065Swnj 	 * tp->iss, and send a segment:
1535085Swnj 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1545065Swnj 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
1555065Swnj 	 * Fill in remote peer address fields if not previously specified.
1565065Swnj 	 * Enter SYN_RECEIVED state, and process any other fields of this
1575244Sroot 	 * segment in this state.
1585065Swnj 	 */
1595065Swnj 	case TCPS_LISTEN:
1605065Swnj 		if (tiflags & TH_RST)
1615065Swnj 			goto drop;
1625300Sroot 		if (tiflags & TH_ACK)
1635085Swnj 			goto dropwithreset;
1645300Sroot 		if ((tiflags & TH_SYN) == 0)
1655065Swnj 			goto drop;
1665244Sroot 		tcp_in.sin_addr = ti->ti_src;
1675244Sroot 		tcp_in.sin_port = ti->ti_sport;
1685300Sroot 		if (in_pcbconnect(inp, (struct sockaddr *)&tcp_in))
1695244Sroot 			goto drop;
1705244Sroot 		tp->t_template = tcp_template(tp);
1715244Sroot 		if (tp->t_template == 0) {
1725244Sroot 			in_pcbdisconnect(inp);
1735244Sroot 			goto drop;
1745244Sroot 		}
1755085Swnj 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
1765065Swnj 		tp->irs = ti->ti_seq;
1775085Swnj 		tcp_sendseqinit(tp);
1785085Swnj 		tcp_rcvseqinit(tp);
1795065Swnj 		tp->t_state = TCPS_SYN_RECEIVED;
1805244Sroot 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
1815085Swnj 		goto trimthenstep6;
1824601Swnj 
1835065Swnj 	/*
1845065Swnj 	 * If the state is SYN_SENT:
1855065Swnj 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1865065Swnj 	 *	if seg contains a RST, then drop the connection.
1875065Swnj 	 *	if seg does not contain SYN, then drop it.
1885065Swnj 	 * Otherwise this is an acceptable SYN segment
1895065Swnj 	 *	initialize tp->rcv_nxt and tp->irs
1905065Swnj 	 *	if seg contains ack then advance tp->snd_una
1915065Swnj 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1925065Swnj 	 *	arrange for segment to be acked (eventually)
1935065Swnj 	 *	continue processing rest of data/controls, beginning with URG
1945065Swnj 	 */
1955065Swnj 	case TCPS_SYN_SENT:
1965065Swnj 		if ((tiflags & TH_ACK) &&
1975300Sroot /* this should be SEQ_LT; is SEQ_LEQ for BBN vax TCP only */
1985300Sroot 		    (SEQ_LT(ti->ti_ack, tp->iss) ||
1995231Swnj 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
2005085Swnj 			goto dropwithreset;
2015065Swnj 		if (tiflags & TH_RST) {
2025065Swnj 			if (tiflags & TH_ACK)
2035267Sroot 				tcp_drop(tp, ECONNREFUSED);
2045065Swnj 			goto drop;
2054601Swnj 		}
2065065Swnj 		if ((tiflags & TH_SYN) == 0)
2075065Swnj 			goto drop;
2085231Swnj 		tp->snd_una = ti->ti_ack;
2095244Sroot 		tp->t_timer[TCPT_REXMT] = 0;
2105065Swnj 		tp->irs = ti->ti_seq;
2115085Swnj 		tcp_rcvseqinit(tp);
2125085Swnj 		tp->t_flags |= TF_ACKNOW;
2135162Swnj 		if (SEQ_GT(tp->snd_una, tp->iss)) {
2145263Swnj 			so->so_state |= SS_CONNAWAITING;
2155244Sroot 			soisconnected(so);
2165065Swnj 			tp->t_state = TCPS_ESTABLISHED;
2175162Swnj 			(void) tcp_reass(tp, (struct tcpiphdr *)0);
2185162Swnj 		} else
2195085Swnj 			tp->t_state = TCPS_SYN_RECEIVED;
2205085Swnj 		goto trimthenstep6;
2215085Swnj 
2225085Swnj trimthenstep6:
2235085Swnj 		/*
2245231Swnj 		 * Advance ti->ti_seq to correspond to first data byte.
2255085Swnj 		 * If data, trim to stay within window,
2265085Swnj 		 * dropping FIN if necessary.
2275085Swnj 		 */
2285231Swnj 		ti->ti_seq++;
2295085Swnj 		if (ti->ti_len > tp->rcv_wnd) {
2305085Swnj 			todrop = ti->ti_len - tp->rcv_wnd;
2315085Swnj 			m_adj(m, -todrop);
2325085Swnj 			ti->ti_len = tp->rcv_wnd;
2335085Swnj 			ti->ti_flags &= ~TH_FIN;
2345065Swnj 		}
2355263Swnj 		tp->snd_wl1 = ti->ti_seq - 1;
2365085Swnj 		goto step6;
2375065Swnj 	}
2384601Swnj 
2395065Swnj 	/*
2405065Swnj 	 * States other than LISTEN or SYN_SENT.
2415065Swnj 	 * First check that at least some bytes of segment are within
2425065Swnj 	 * receive window.
2435065Swnj 	 */
2445065Swnj 	if (tp->rcv_wnd == 0) {
2455065Swnj 		/*
2465065Swnj 		 * If window is closed can only take segments at
2475231Swnj 		 * window edge, and have to drop data and PUSH from
2485065Swnj 		 * incoming segments.
2495065Swnj 		 */
2505300Sroot 		if (tp->rcv_nxt != ti->ti_seq)
2515065Swnj 			goto dropafterack;
2525085Swnj 		if (ti->ti_len > 0) {
2535085Swnj 			ti->ti_len = 0;
2545085Swnj 			ti->ti_flags &= ~(TH_PUSH|TH_FIN);
2555065Swnj 		}
2565065Swnj 	} else {
2575065Swnj 		/*
2585231Swnj 		 * If segment begins before rcv_nxt, drop leading
2595065Swnj 		 * data (and SYN); if nothing left, just ack.
2605065Swnj 		 */
2615065Swnj 		if (SEQ_GT(tp->rcv_nxt, ti->ti_seq)) {
2625085Swnj 			todrop = tp->rcv_nxt - ti->ti_seq;
2635085Swnj 			if (tiflags & TH_SYN) {
2645300Sroot 				tiflags &= ~TH_SYN;
2655085Swnj 				ti->ti_seq++;
2665085Swnj 				if (ti->ti_urp > 1)
2675085Swnj 					ti->ti_urp--;
2685085Swnj 				else
2695085Swnj 					tiflags &= ~TH_URG;
2705085Swnj 				todrop--;
2715085Swnj 			}
2725300Sroot 			if (todrop > ti->ti_len)
2735065Swnj 				goto dropafterack;
2745065Swnj 			m_adj(m, todrop);
2755065Swnj 			ti->ti_seq += todrop;
2765065Swnj 			ti->ti_len -= todrop;
2775085Swnj 			if (ti->ti_urp > todrop)
2785085Swnj 				ti->ti_urp -= todrop;
2795085Swnj 			else {
2805085Swnj 				tiflags &= ~TH_URG;
2815085Swnj 				/* ti->ti_flags &= ~TH_URG; */
2825085Swnj 				/* ti->ti_urp = 0; */
2835085Swnj 			}
2845085Swnj 			/* tiflags &= ~TH_SYN; */
2855085Swnj 			/* ti->ti_flags &= ~TH_SYN; */
2865065Swnj 		}
2875065Swnj 		/*
2885065Swnj 		 * If segment ends after window, drop trailing data
2895085Swnj 		 * (and PUSH and FIN); if nothing left, just ACK.
2905065Swnj 		 */
2915065Swnj 		if (SEQ_GT(ti->ti_seq+ti->ti_len, tp->rcv_nxt+tp->rcv_wnd)) {
2925085Swnj 			todrop =
2935065Swnj 			     ti->ti_seq+ti->ti_len - (tp->rcv_nxt+tp->rcv_wnd);
2945300Sroot 			if (todrop > ti->ti_len)
2955065Swnj 				goto dropafterack;
2965065Swnj 			m_adj(m, -todrop);
2975065Swnj 			ti->ti_len -= todrop;
2985085Swnj 			ti->ti_flags &= ~(TH_PUSH|TH_FIN);
2995065Swnj 		}
3005065Swnj 	}
3014601Swnj 
3025065Swnj 	/*
3035065Swnj 	 * If the RST bit is set examine the state:
3045065Swnj 	 *    SYN_RECEIVED STATE:
3055065Swnj 	 *	If passive open, return to LISTEN state.
3065065Swnj 	 *	If active open, inform user that connection was refused.
3075065Swnj 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
3085065Swnj 	 *	Inform user that connection was reset, and close tcb.
3095065Swnj 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
3105065Swnj 	 *	Close the tcb.
3115065Swnj 	 */
3125065Swnj 	if (tiflags&TH_RST) switch (tp->t_state) {
3135267Sroot 
3145065Swnj 	case TCPS_SYN_RECEIVED:
3155065Swnj 		if (inp->inp_socket->so_options & SO_ACCEPTCONN) {
3165267Sroot 			/* a miniature tcp_close, but invisible to user */
3175267Sroot 			(void) m_free(dtom(tp->t_template));
3185267Sroot 			(void) m_free(dtom(tp));
3195267Sroot 			inp->inp_ppcb = 0;
3205267Sroot 			tp = tcp_newtcpcb(inp);
3215085Swnj 			tp->t_state = TCPS_LISTEN;
3225065Swnj 			goto drop;
3234601Swnj 		}
3245085Swnj 		tcp_drop(tp, ECONNREFUSED);
3255065Swnj 		goto drop;
3264601Swnj 
3275065Swnj 	case TCPS_ESTABLISHED:
3285065Swnj 	case TCPS_FIN_WAIT_1:
3295065Swnj 	case TCPS_FIN_WAIT_2:
3305065Swnj 	case TCPS_CLOSE_WAIT:
3315065Swnj 		tcp_drop(tp, ECONNRESET);
3325065Swnj 		goto drop;
3335065Swnj 
3345065Swnj 	case TCPS_CLOSING:
3355065Swnj 	case TCPS_LAST_ACK:
3365065Swnj 	case TCPS_TIME_WAIT:
3375065Swnj 		tcp_close(tp);
3385065Swnj 		goto drop;
3394601Swnj 	}
3404601Swnj 
3414601Swnj 	/*
3425065Swnj 	 * If a SYN is in the window, then this is an
3435065Swnj 	 * error and we send an RST and drop the connection.
3444601Swnj 	 */
3455065Swnj 	if (tiflags & TH_SYN) {
3465231Swnj 		tcp_drop(tp, ECONNRESET);
3475085Swnj 		goto dropwithreset;
3484601Swnj 	}
3494601Swnj 
3504601Swnj 	/*
3515065Swnj 	 * If the ACK bit is off we drop the segment and return.
3524601Swnj 	 */
3535085Swnj 	if ((tiflags & TH_ACK) == 0)
3545065Swnj 		goto drop;
3555065Swnj 
3565065Swnj 	/*
3575065Swnj 	 * Ack processing.
3585065Swnj 	 */
3594601Swnj 	switch (tp->t_state) {
3604601Swnj 
3615065Swnj 	/*
3625065Swnj 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
3635065Swnj 	 * ESTABLISHED state and continue processing, othewise
3645065Swnj 	 * send an RST.
3655065Swnj 	 */
3665065Swnj 	case TCPS_SYN_RECEIVED:
3675085Swnj 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
3685231Swnj 		    SEQ_GT(ti->ti_ack, tp->snd_max))
3695085Swnj 			goto dropwithreset;
3705244Sroot 		tp->snd_una++;			/* SYN acked */
3715244Sroot 		tp->t_timer[TCPT_REXMT] = 0;
3725263Swnj 		so->so_state |= SS_CONNAWAITING;
3735085Swnj 		soisconnected(so);
3745085Swnj 		tp->t_state = TCPS_ESTABLISHED;
3755162Swnj 		(void) tcp_reass(tp, (struct tcpiphdr *)0);
3765244Sroot 		tp->snd_wl1 = ti->ti_seq - 1;
3775085Swnj 		/* fall into ... */
3784601Swnj 
3795065Swnj 	/*
3805065Swnj 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
3815065Swnj 	 * ACKs.  If the ack is in the range
3825231Swnj 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
3835065Swnj 	 * then advance tp->snd_una to ti->ti_ack and drop
3845065Swnj 	 * data from the retransmission queue.  If this ACK reflects
3855065Swnj 	 * more up to date window information we update our window information.
3865065Swnj 	 */
3875065Swnj 	case TCPS_ESTABLISHED:
3885065Swnj 	case TCPS_FIN_WAIT_1:
3895065Swnj 	case TCPS_FIN_WAIT_2:
3905065Swnj 	case TCPS_CLOSE_WAIT:
3915065Swnj 	case TCPS_CLOSING:
3925244Sroot 	case TCPS_LAST_ACK:
3935244Sroot 	case TCPS_TIME_WAIT:
3945085Swnj #define	ourfinisacked	(acked > 0)
3955085Swnj 
3965244Sroot 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
3975065Swnj 			break;
3985300Sroot 		if (SEQ_GT(ti->ti_ack, tp->snd_max))
3995065Swnj 			goto dropafterack;
4005085Swnj 		acked = ti->ti_ack - tp->snd_una;
401*5307Sroot 		if (ti->ti_ack == tp->snd_max)
4025244Sroot 			tp->t_timer[TCPT_REXMT] = 0;
403*5307Sroot 		else {
4045244Sroot 			TCPT_RANGESET(tp->t_timer[TCPT_REXMT],
4055244Sroot 			    tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX);
4065300Sroot 			tp->t_rtt = 0;
4075300Sroot 			tp->t_rxtshift = 0;
4085085Swnj 		}
409*5307Sroot 		if (acked > so->so_snd.sb_cc) {
410*5307Sroot 			sbdrop(&so->so_snd, so->so_snd.sb_cc);
411*5307Sroot 			tp->snd_wnd -= so->so_snd.sb_cc;
412*5307Sroot 		} else {
413*5307Sroot 			sbdrop(&so->so_snd.sb_cc, acked);
414*5307Sroot 			tp->snd_wnd -= acked;
415*5307Sroot 			acked = 0;
416*5307Sroot 		}
4175300Sroot 		if (so->so_snd.sb_flags & SB_WAIT)
4185300Sroot 			sowwakeup(so);
4195231Swnj 		tp->snd_una = ti->ti_ack;
4205162Swnj 
4215162Swnj 		/*
4225162Swnj 		 * If transmit timer is running and timed sequence
4235162Swnj 		 * number was acked, update smoothed round trip time.
4245162Swnj 		 */
4255162Swnj 		if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) {
4265244Sroot 			if (tp->t_srtt == 0)
4275244Sroot 				tp->t_srtt = tp->t_rtt;
4285244Sroot 			else
4295244Sroot 				tp->t_srtt =
4305244Sroot 				    tcp_alpha * tp->t_srtt +
4315244Sroot 				    (1 - tcp_alpha) * tp->t_rtt;
4325162Swnj 			tp->t_rtt = 0;
4335162Swnj 		}
4345162Swnj 
4354601Swnj 		switch (tp->t_state) {
4364601Swnj 
4375065Swnj 		/*
4385065Swnj 		 * In FIN_WAIT_1 STATE in addition to the processing
4395065Swnj 		 * for the ESTABLISHED state if our FIN is now acknowledged
4405085Swnj 		 * then enter FIN_WAIT_2.
4415065Swnj 		 */
4425065Swnj 		case TCPS_FIN_WAIT_1:
4435085Swnj 			if (ourfinisacked)
4445085Swnj 				tp->t_state = TCPS_FIN_WAIT_2;
4454601Swnj 			break;
4464601Swnj 
4475065Swnj 	 	/*
4485065Swnj 		 * In CLOSING STATE in addition to the processing for
4495065Swnj 		 * the ESTABLISHED state if the ACK acknowledges our FIN
4505065Swnj 		 * then enter the TIME-WAIT state, otherwise ignore
4515065Swnj 		 * the segment.
4525065Swnj 		 */
4535065Swnj 		case TCPS_CLOSING:
4545244Sroot 			if (ourfinisacked) {
4555065Swnj 				tp->t_state = TCPS_TIME_WAIT;
4565244Sroot 				tcp_canceltimers(tp);
4575244Sroot 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
4585244Sroot 				soisdisconnected(so);
4595244Sroot 			}
4605244Sroot 			break;
4614601Swnj 
4625065Swnj 		/*
4635085Swnj 		 * The only thing that can arrive in  LAST_ACK state
4645085Swnj 		 * is an acknowledgment of our FIN.  If our FIN is now
4655085Swnj 		 * acknowledged, delete the TCB, enter the closed state
4665085Swnj 		 * and return.
4675065Swnj 		 */
4685065Swnj 		case TCPS_LAST_ACK:
4695251Sroot 			if (ourfinisacked)
4705065Swnj 				tcp_close(tp);
4715065Swnj 			goto drop;
4724601Swnj 
4735065Swnj 		/*
4745065Swnj 		 * In TIME_WAIT state the only thing that should arrive
4755065Swnj 		 * is a retransmission of the remote FIN.  Acknowledge
4765065Swnj 		 * it and restart the finack timer.
4775065Swnj 		 */
4785065Swnj 		case TCPS_TIME_WAIT:
4795162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
4805065Swnj 			goto dropafterack;
4814601Swnj 		}
4825085Swnj #undef ourfinisacked
4835085Swnj 	}
4844601Swnj 
4855065Swnj step6:
4865065Swnj 	/*
4875244Sroot 	 * Update window information.
4885244Sroot 	 */
4895300Sroot 	if (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq &&
4905300Sroot 	    (SEQ_LEQ(tp->snd_wl2, ti->ti_ack) ||
4915300Sroot 	     tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd)) {
4925244Sroot 		tp->snd_wnd = ti->ti_win;
4935244Sroot 		tp->snd_wl1 = ti->ti_seq;
4945244Sroot 		tp->snd_wl2 = ti->ti_ack;
4955244Sroot 		if (tp->snd_wnd > 0)
4965244Sroot 			tp->t_timer[TCPT_PERSIST] = 0;
4975244Sroot 	}
4985244Sroot 
4995244Sroot 	/*
5005065Swnj 	 * If an URG bit is set in the segment and is greater than the
5015065Swnj 	 * current known urgent pointer, then signal the user that the
5025244Sroot 	 * remote side has out of band data.  This should not happen
5035065Swnj 	 * in CLOSE_WAIT, CLOSING, LAST-ACK or TIME_WAIT STATES since
5045065Swnj 	 * a FIN has been received from the remote side.  In these states
5055065Swnj 	 * we ignore the URG.
5065065Swnj 	 */
5075085Swnj 	if ((tiflags & TH_URG) == 0 && TCPS_HAVERCVDFIN(tp->t_state) == 0)
5085085Swnj 		if (SEQ_GT(ti->ti_urp, tp->rcv_up)) {
5095065Swnj 			tp->rcv_up = ti->ti_urp;
5105085Swnj #if 0
5115244Sroot 			sohasoutofband(so);		/* XXX */
5125085Swnj #endif
5134601Swnj 		}
5144601Swnj 
5154601Swnj 	/*
5165065Swnj 	 * Process the segment text, merging it into the TCP sequencing queue,
5175065Swnj 	 * and arranging for acknowledgment of receipt if necessary.
5185065Swnj 	 * This process logically involves adjusting tp->rcv_wnd as data
5195065Swnj 	 * is presented to the user (this happens in tcp_usrreq.c,
5205065Swnj 	 * case PRU_RCVD).  If a FIN has already been received on this
5215065Swnj 	 * connection then we just ignore the text.
5224601Swnj 	 */
5235263Swnj 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
5245263Swnj 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
5255085Swnj 		off += sizeof (struct ip);		/* drop IP header */
5265085Swnj 		m->m_off += off;
5275085Swnj 		m->m_len -= off;
5285065Swnj 		tiflags = tcp_reass(tp, ti);
5295287Sroot { extern tcpdelack;
5305287Sroot if (tcpdelack) tp->t_flags |= TF_DELACK; else
5315085Swnj 		tp->t_flags |= TF_ACKNOW;		/* XXX TF_DELACK */
5325287Sroot }
5335244Sroot 	} else {
5344924Swnj 		m_freem(m);
5355263Swnj 		tiflags &= ~TH_FIN;
5365244Sroot 	}
5374601Swnj 
5384601Swnj 	/*
5395263Swnj 	 * If FIN is received ACK the FIN and let the user know
5405263Swnj 	 * that the connection is closing.
5414601Swnj 	 */
5425263Swnj 	if (tiflags & TH_FIN) {
5435244Sroot 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
5445244Sroot 			socantrcvmore(so);
5455244Sroot 			tp->t_flags |= TF_ACKNOW;
5465244Sroot 			tp->rcv_nxt++;
5475244Sroot 		}
5485065Swnj 		switch (tp->t_state) {
5494601Swnj 
5505065Swnj 	 	/*
5515065Swnj 		 * In SYN_RECEIVED and ESTABLISHED STATES
5525065Swnj 		 * enter the CLOSE_WAIT state.
5534884Swnj 		 */
5545065Swnj 		case TCPS_SYN_RECEIVED:
5555065Swnj 		case TCPS_ESTABLISHED:
5565065Swnj 			tp->t_state = TCPS_CLOSE_WAIT;
5575065Swnj 			break;
5584884Swnj 
5595065Swnj 	 	/*
5605085Swnj 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
5615085Swnj 		 * enter the CLOSING state.
5624884Swnj 		 */
5635065Swnj 		case TCPS_FIN_WAIT_1:
5645085Swnj 			tp->t_state = TCPS_CLOSING;
5655065Swnj 			break;
5664601Swnj 
5675065Swnj 	 	/*
5685065Swnj 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
5695065Swnj 		 * starting the time-wait timer, turning off the other
5705065Swnj 		 * standard timers.
5715065Swnj 		 */
5725065Swnj 		case TCPS_FIN_WAIT_2:
5735244Sroot 			tp->t_state = TCPS_TIME_WAIT;
5745074Swnj 			tcp_canceltimers(tp);
5755162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
5765244Sroot 			soisdisconnected(so);
5775065Swnj 			break;
5785065Swnj 
5794884Swnj 		/*
5805065Swnj 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
5814884Swnj 		 */
5825065Swnj 		case TCPS_TIME_WAIT:
5835162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
5845065Swnj 			break;
5855085Swnj 		}
5864601Swnj 	}
5875267Sroot 	if (so->so_options & SO_DEBUG)
5885267Sroot 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
5895085Swnj 
5905085Swnj 	/*
5915085Swnj 	 * Return any desired output.
5925085Swnj 	 */
5935085Swnj 	tcp_output(tp);
5945065Swnj 	return;
5955085Swnj 
5965065Swnj dropafterack:
5975085Swnj 	/*
5985244Sroot 	 * Generate an ACK dropping incoming segment.
5995085Swnj 	 * Make ACK reflect our state.
6005085Swnj 	 */
6015085Swnj 	if (tiflags & TH_RST)
6025085Swnj 		goto drop;
6035085Swnj 	tcp_respond(ti, tp->rcv_nxt, tp->snd_nxt, TH_ACK);
6045231Swnj 	return;
6055085Swnj 
6065085Swnj dropwithreset:
6075085Swnj 	/*
6085244Sroot 	 * Generate a RST, dropping incoming segment.
6095085Swnj 	 * Make ACK acceptable to originator of segment.
6105085Swnj 	 */
6115085Swnj 	if (tiflags & TH_RST)
6125085Swnj 		goto drop;
6135085Swnj 	if (tiflags & TH_ACK)
6145109Swnj 		tcp_respond(ti, (tcp_seq)0, ti->ti_ack, TH_RST);
6155085Swnj 	else {
6165085Swnj 		if (tiflags & TH_SYN)
6175085Swnj 			ti->ti_len++;
6185109Swnj 		tcp_respond(ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, TH_RST|TH_ACK);
6195085Swnj 	}
6205231Swnj 	return;
6215085Swnj 
6225065Swnj drop:
6235085Swnj 	/*
6245085Swnj 	 * Drop space held by incoming segment and return.
6255085Swnj 	 */
6265065Swnj 	m_freem(m);
6275267Sroot 	return;
6285065Swnj }
6295065Swnj 
6305065Swnj /*
6315065Swnj  * Insert segment ti into reassembly queue of tcp with
6325065Swnj  * control block tp.  Return TH_FIN if reassembly now includes
6335065Swnj  * a segment with FIN.
6345065Swnj  */
6355109Swnj tcp_reass(tp, ti)
6365065Swnj 	register struct tcpcb *tp;
6375065Swnj 	register struct tcpiphdr *ti;
6385065Swnj {
6395065Swnj 	register struct tcpiphdr *q;
6405085Swnj 	struct socket *so = tp->t_inpcb->inp_socket;
6415263Swnj 	struct mbuf *m;
6425263Swnj 	int flags;
6435085Swnj COUNT(TCP_REASS);
6445065Swnj 
6455065Swnj 	/*
6465162Swnj 	 * Call with ti==0 after become established to
6475162Swnj 	 * force pre-ESTABLISHED data up to user socket.
6485065Swnj 	 */
6495162Swnj 	if (ti == 0)
6505065Swnj 		goto present;
6514601Swnj 
6525065Swnj 	/*
6535065Swnj 	 * Find a segment which begins after this one does.
6545065Swnj 	 */
6555065Swnj 	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
6565065Swnj 	    q = (struct tcpiphdr *)q->ti_next)
6575065Swnj 		if (SEQ_GT(q->ti_seq, ti->ti_seq))
6585065Swnj 			break;
6594601Swnj 
6605065Swnj 	/*
6615065Swnj 	 * If there is a preceding segment, it may provide some of
6625065Swnj 	 * our data already.  If so, drop the data from the incoming
6635065Swnj 	 * segment.  If it provides all of our data, drop us.
6645065Swnj 	 */
6655065Swnj 	if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
6665065Swnj 		register int i;
6675065Swnj 		q = (struct tcpiphdr *)(q->ti_prev);
6685065Swnj 		/* conversion to int (in i) handles seq wraparound */
6695065Swnj 		i = q->ti_seq + q->ti_len - ti->ti_seq;
6705065Swnj 		if (i > 0) {
6714924Swnj 			if (i >= ti->ti_len)
6725065Swnj 				goto drop;
6735065Swnj 			m_adj(dtom(tp), i);
6745065Swnj 			ti->ti_len -= i;
6754924Swnj 			ti->ti_seq += i;
6764601Swnj 		}
6775065Swnj 		q = (struct tcpiphdr *)(q->ti_next);
6785065Swnj 	}
6794601Swnj 
6805065Swnj 	/*
6815065Swnj 	 * While we overlap succeeding segments trim them or,
6825065Swnj 	 * if they are completely covered, dequeue them.
6835065Swnj 	 */
6845065Swnj 	while (q != (struct tcpiphdr *)tp &&
6855065Swnj 	    SEQ_GT(ti->ti_seq + ti->ti_len, q->ti_seq)) {
6865065Swnj 		register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
6875065Swnj 		if (i < q->ti_len) {
6885065Swnj 			q->ti_len -= i;
6895065Swnj 			m_adj(dtom(q), i);
6905065Swnj 			break;
6914601Swnj 		}
6925065Swnj 		q = (struct tcpiphdr *)q->ti_next;
6935065Swnj 		m_freem(dtom(q->ti_prev));
6945065Swnj 		remque(q->ti_prev);
6955065Swnj 	}
6964601Swnj 
6975065Swnj 	/*
6985065Swnj 	 * Stick new segment in its place.
6995065Swnj 	 */
7005065Swnj 	insque(ti, q->ti_prev);
7014601Swnj 
7025065Swnj present:
7035065Swnj 	/*
7045244Sroot 	 * Present data to user, advancing rcv_nxt through
7055244Sroot 	 * completed sequence space.
7065065Swnj 	 */
7075263Swnj 	if (TCPS_HAVERCVDSYN(tp->t_state) == 0)
7085244Sroot 		return (0);
7094924Swnj 	ti = tp->seg_next;
7105263Swnj 	if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
7115263Swnj 		return (0);
7125263Swnj 	if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
7135263Swnj 		return (0);
7145263Swnj 	do {
7155244Sroot 		tp->rcv_nxt += ti->ti_len;
7165244Sroot 		flags = ti->ti_flags & TH_FIN;
7174924Swnj 		remque(ti);
7185263Swnj 		m = dtom(ti);
7194924Swnj 		ti = (struct tcpiphdr *)ti->ti_next;
7205263Swnj 		if (so->so_state & SS_CANTRCVMORE)
7215263Swnj 			(void) m_freem(m);
7225263Swnj 		else
7235263Swnj 			sbappend(&so->so_rcv, m);
7245263Swnj 	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
7255263Swnj 	sorwakeup(so);
7265065Swnj 	return (flags);
7275065Swnj drop:
7285065Swnj 	m_freem(dtom(ti));
7295263Swnj 	return (0);
7304601Swnj }
731