xref: /csrg-svn/sys/netinet/tcp_input.c (revision 5440)
1*5440Swnj /*	tcp_input.c	1.49	82/01/17	*/
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;
28*5440Swnj extern	tcpnodelack;
294601Swnj 
305267Sroot struct	tcpcb *tcp_newtcpcb();
315065Swnj /*
325065Swnj  * TCP input routine, follows pages 65-76 of the
335065Swnj  * protocol specification dated September, 1981 very closely.
345065Swnj  */
354924Swnj tcp_input(m0)
364924Swnj 	struct mbuf *m0;
374601Swnj {
384924Swnj 	register struct tcpiphdr *ti;
394924Swnj 	struct inpcb *inp;
404924Swnj 	register struct mbuf *m;
41*5440Swnj 	struct mbuf *om = 0;
424924Swnj 	int len, tlen, off;
435391Swnj 	register struct tcpcb *tp = 0;
444924Swnj 	register int tiflags;
454803Swnj 	struct socket *so;
465109Swnj 	int todrop, acked;
475267Sroot 	short ostate;
484924Swnj 
494601Swnj COUNT(TCP_INPUT);
504924Swnj 	/*
515244Sroot 	 * Get IP and TCP header together in first mbuf.
525244Sroot 	 * Note: IP leaves IP header in first mbuf.
534924Swnj 	 */
544924Swnj 	m = m0;
555020Sroot 	ti = mtod(m, struct tcpiphdr *);
565244Sroot 	if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2))
575208Swnj 		ip_stripoptions((struct ip *)ti, (struct mbuf *)0);
585307Sroot 	if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) {
595307Sroot 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
605085Swnj 			tcpstat.tcps_hdrops++;
615307Sroot 			return;
625085Swnj 		}
635085Swnj 		ti = mtod(m, struct tcpiphdr *);
645085Swnj 	}
654601Swnj 
664601Swnj 	/*
675244Sroot 	 * Checksum extended TCP header and data.
684601Swnj 	 */
694924Swnj 	tlen = ((struct ip *)ti)->ip_len;
704924Swnj 	len = sizeof (struct ip) + tlen;
714679Swnj 	if (tcpcksum) {
724924Swnj 		ti->ti_next = ti->ti_prev = 0;
734924Swnj 		ti->ti_x1 = 0;
745223Swnj 		ti->ti_len = (u_short)tlen;
755223Swnj #if vax
765223Swnj 		ti->ti_len = htons(ti->ti_len);
775223Swnj #endif
785231Swnj 		if (ti->ti_sum = in_cksum(m, len)) {
794924Swnj 			tcpstat.tcps_badsum++;
805065Swnj 			printf("tcp cksum %x\n", ti->ti_sum);
815085Swnj 			goto drop;
824601Swnj 		}
834601Swnj 	}
844601Swnj 
854601Swnj 	/*
865244Sroot 	 * Check that TCP offset makes sense,
87*5440Swnj 	 * pull out TCP options and adjust length.
884601Swnj 	 */
894924Swnj 	off = ti->ti_off << 2;
905231Swnj 	if (off < sizeof (struct tcphdr) || off > tlen) {
914924Swnj 		tcpstat.tcps_badoff++;
925085Swnj 		goto drop;
934924Swnj 	}
944924Swnj 	ti->ti_len = tlen - off;
95*5440Swnj 	if (off > sizeof (struct tcphdr)) {
96*5440Swnj 		if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
97*5440Swnj 			tcpstat.tcps_hdrops++;
98*5440Swnj 			goto drop;
99*5440Swnj 		}
100*5440Swnj 		ti = mtod(m, struct tcpiphdr *);
101*5440Swnj 		om = m_get(M_DONTWAIT);
102*5440Swnj 		if (om == 0)
103*5440Swnj 			goto drop;
104*5440Swnj 		om->m_off = MMINOFF;
105*5440Swnj 		om->m_len = off - sizeof (struct tcphdr);
106*5440Swnj 		{ caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
107*5440Swnj 		  bcopy(op, mtod(om, caddr_t), om->m_len);
108*5440Swnj 		  m->m_len -= om->m_len;
109*5440Swnj 		  bcopy(op+om->m_len, op, m->m_len-sizeof (struct tcpiphdr));
110*5440Swnj 		}
111*5440Swnj 	}
1125065Swnj 	tiflags = ti->ti_flags;
1134924Swnj 
1145231Swnj #if vax
1154924Swnj 	/*
1165244Sroot 	 * Convert TCP protocol specific fields to host format.
1175085Swnj 	 */
1185085Swnj 	ti->ti_seq = ntohl(ti->ti_seq);
1195085Swnj 	ti->ti_ack = ntohl(ti->ti_ack);
1205085Swnj 	ti->ti_win = ntohs(ti->ti_win);
1215085Swnj 	ti->ti_urp = ntohs(ti->ti_urp);
1225231Swnj #endif
1235085Swnj 
1245085Swnj 	/*
1254924Swnj 	 * Locate pcb for segment.
1264924Swnj 	 */
1275065Swnj 	inp = in_pcblookup
1285065Swnj 		(&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport);
1295065Swnj 
1305065Swnj 	/*
1315065Swnj 	 * If the state is CLOSED (i.e., TCB does not exist) then
1325244Sroot 	 * all data in the incoming segment is discarded.
1335065Swnj 	 */
1345300Sroot 	if (inp == 0)
1355085Swnj 		goto dropwithreset;
1365065Swnj 	tp = intotcpcb(inp);
1375300Sroot 	if (tp == 0)
1385085Swnj 		goto dropwithreset;
1395109Swnj 	so = inp->inp_socket;
1405267Sroot 	if (so->so_options & SO_DEBUG) {
1415267Sroot 		ostate = tp->t_state;
1425267Sroot 		tcp_saveti = *ti;
1435267Sroot 	}
1444601Swnj 
1454601Swnj 	/*
1465162Swnj 	 * Segment received on connection.
1475162Swnj 	 * Reset idle time and keep-alive timer.
1485162Swnj 	 */
1495162Swnj 	tp->t_idle = 0;
1505162Swnj 	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
1515162Swnj 
1525162Swnj 	/*
153*5440Swnj 	 * Process options.
154*5440Swnj 	 */
155*5440Swnj 	if (om) {
156*5440Swnj 		tcp_dooptions(tp, om);
157*5440Swnj 		om = 0;
158*5440Swnj 	}
159*5440Swnj 
160*5440Swnj 	/*
1615085Swnj 	 * Calculate amount of space in receive window,
1625085Swnj 	 * and then do TCP input processing.
1634601Swnj 	 */
1645085Swnj 	tp->rcv_wnd = sbspace(&so->so_rcv);
1655231Swnj 	if (tp->rcv_wnd < 0)
1665231Swnj 		tp->rcv_wnd = 0;
1674601Swnj 
1684601Swnj 	switch (tp->t_state) {
1694601Swnj 
1705065Swnj 	/*
1715065Swnj 	 * If the state is LISTEN then ignore segment if it contains an RST.
1725065Swnj 	 * If the segment contains an ACK then it is bad and send a RST.
1735065Swnj 	 * If it does not contain a SYN then it is not interesting; drop it.
1745085Swnj 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
1755065Swnj 	 * tp->iss, and send a segment:
1765085Swnj 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1775065Swnj 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
1785065Swnj 	 * Fill in remote peer address fields if not previously specified.
1795065Swnj 	 * Enter SYN_RECEIVED state, and process any other fields of this
1805244Sroot 	 * segment in this state.
1815065Swnj 	 */
1825065Swnj 	case TCPS_LISTEN:
1835065Swnj 		if (tiflags & TH_RST)
1845065Swnj 			goto drop;
1855300Sroot 		if (tiflags & TH_ACK)
1865085Swnj 			goto dropwithreset;
1875300Sroot 		if ((tiflags & TH_SYN) == 0)
1885065Swnj 			goto drop;
1895244Sroot 		tcp_in.sin_addr = ti->ti_src;
1905244Sroot 		tcp_in.sin_port = ti->ti_sport;
1915300Sroot 		if (in_pcbconnect(inp, (struct sockaddr *)&tcp_in))
1925244Sroot 			goto drop;
1935244Sroot 		tp->t_template = tcp_template(tp);
1945244Sroot 		if (tp->t_template == 0) {
1955244Sroot 			in_pcbdisconnect(inp);
1965244Sroot 			goto drop;
1975244Sroot 		}
1985085Swnj 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
1995065Swnj 		tp->irs = ti->ti_seq;
2005085Swnj 		tcp_sendseqinit(tp);
2015085Swnj 		tcp_rcvseqinit(tp);
2025065Swnj 		tp->t_state = TCPS_SYN_RECEIVED;
2035244Sroot 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
2045085Swnj 		goto trimthenstep6;
2054601Swnj 
2065065Swnj 	/*
2075065Swnj 	 * If the state is SYN_SENT:
2085065Swnj 	 *	if seg contains an ACK, but not for our SYN, drop the input.
2095065Swnj 	 *	if seg contains a RST, then drop the connection.
2105065Swnj 	 *	if seg does not contain SYN, then drop it.
2115065Swnj 	 * Otherwise this is an acceptable SYN segment
2125065Swnj 	 *	initialize tp->rcv_nxt and tp->irs
2135065Swnj 	 *	if seg contains ack then advance tp->snd_una
2145065Swnj 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2155065Swnj 	 *	arrange for segment to be acked (eventually)
2165065Swnj 	 *	continue processing rest of data/controls, beginning with URG
2175065Swnj 	 */
2185065Swnj 	case TCPS_SYN_SENT:
2195065Swnj 		if ((tiflags & TH_ACK) &&
2205300Sroot /* this should be SEQ_LT; is SEQ_LEQ for BBN vax TCP only */
2215300Sroot 		    (SEQ_LT(ti->ti_ack, tp->iss) ||
2225231Swnj 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
2235085Swnj 			goto dropwithreset;
2245065Swnj 		if (tiflags & TH_RST) {
2255065Swnj 			if (tiflags & TH_ACK)
2265267Sroot 				tcp_drop(tp, ECONNREFUSED);
2275065Swnj 			goto drop;
2284601Swnj 		}
2295065Swnj 		if ((tiflags & TH_SYN) == 0)
2305065Swnj 			goto drop;
2315231Swnj 		tp->snd_una = ti->ti_ack;
2325357Sroot 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2335357Sroot 			tp->snd_nxt = tp->snd_una;
2345244Sroot 		tp->t_timer[TCPT_REXMT] = 0;
2355065Swnj 		tp->irs = ti->ti_seq;
2365085Swnj 		tcp_rcvseqinit(tp);
2375085Swnj 		tp->t_flags |= TF_ACKNOW;
2385162Swnj 		if (SEQ_GT(tp->snd_una, tp->iss)) {
2395391Swnj 			if (so->so_options & SO_ACCEPTCONN)
2405391Swnj 				so->so_state |= SS_CONNAWAITING;
2415244Sroot 			soisconnected(so);
2425065Swnj 			tp->t_state = TCPS_ESTABLISHED;
2435162Swnj 			(void) tcp_reass(tp, (struct tcpiphdr *)0);
2445162Swnj 		} else
2455085Swnj 			tp->t_state = TCPS_SYN_RECEIVED;
2465085Swnj 		goto trimthenstep6;
2475085Swnj 
2485085Swnj trimthenstep6:
2495085Swnj 		/*
2505231Swnj 		 * Advance ti->ti_seq to correspond to first data byte.
2515085Swnj 		 * If data, trim to stay within window,
2525085Swnj 		 * dropping FIN if necessary.
2535085Swnj 		 */
2545231Swnj 		ti->ti_seq++;
2555085Swnj 		if (ti->ti_len > tp->rcv_wnd) {
2565085Swnj 			todrop = ti->ti_len - tp->rcv_wnd;
2575085Swnj 			m_adj(m, -todrop);
2585085Swnj 			ti->ti_len = tp->rcv_wnd;
2595085Swnj 			ti->ti_flags &= ~TH_FIN;
2605065Swnj 		}
2615263Swnj 		tp->snd_wl1 = ti->ti_seq - 1;
2625085Swnj 		goto step6;
2635065Swnj 	}
2644601Swnj 
2655065Swnj 	/*
2665065Swnj 	 * States other than LISTEN or SYN_SENT.
2675065Swnj 	 * First check that at least some bytes of segment are within
2685065Swnj 	 * receive window.
2695065Swnj 	 */
2705065Swnj 	if (tp->rcv_wnd == 0) {
2715065Swnj 		/*
2725065Swnj 		 * If window is closed can only take segments at
2735231Swnj 		 * window edge, and have to drop data and PUSH from
2745065Swnj 		 * incoming segments.
2755065Swnj 		 */
2765300Sroot 		if (tp->rcv_nxt != ti->ti_seq)
2775065Swnj 			goto dropafterack;
2785085Swnj 		if (ti->ti_len > 0) {
2795085Swnj 			ti->ti_len = 0;
2805085Swnj 			ti->ti_flags &= ~(TH_PUSH|TH_FIN);
2815065Swnj 		}
2825065Swnj 	} else {
2835065Swnj 		/*
2845231Swnj 		 * If segment begins before rcv_nxt, drop leading
2855065Swnj 		 * data (and SYN); if nothing left, just ack.
2865065Swnj 		 */
2875065Swnj 		if (SEQ_GT(tp->rcv_nxt, ti->ti_seq)) {
2885085Swnj 			todrop = tp->rcv_nxt - ti->ti_seq;
2895085Swnj 			if (tiflags & TH_SYN) {
2905300Sroot 				tiflags &= ~TH_SYN;
2915085Swnj 				ti->ti_seq++;
2925085Swnj 				if (ti->ti_urp > 1)
2935085Swnj 					ti->ti_urp--;
2945085Swnj 				else
2955085Swnj 					tiflags &= ~TH_URG;
2965085Swnj 				todrop--;
2975085Swnj 			}
2985300Sroot 			if (todrop > ti->ti_len)
2995065Swnj 				goto dropafterack;
3005065Swnj 			m_adj(m, todrop);
3015065Swnj 			ti->ti_seq += todrop;
3025065Swnj 			ti->ti_len -= todrop;
3035085Swnj 			if (ti->ti_urp > todrop)
3045085Swnj 				ti->ti_urp -= todrop;
3055085Swnj 			else {
3065085Swnj 				tiflags &= ~TH_URG;
3075085Swnj 				/* ti->ti_flags &= ~TH_URG; */
3085085Swnj 				/* ti->ti_urp = 0; */
3095085Swnj 			}
3105085Swnj 			/* tiflags &= ~TH_SYN; */
3115085Swnj 			/* ti->ti_flags &= ~TH_SYN; */
3125065Swnj 		}
3135065Swnj 		/*
3145065Swnj 		 * If segment ends after window, drop trailing data
3155085Swnj 		 * (and PUSH and FIN); if nothing left, just ACK.
3165065Swnj 		 */
3175065Swnj 		if (SEQ_GT(ti->ti_seq+ti->ti_len, tp->rcv_nxt+tp->rcv_wnd)) {
3185085Swnj 			todrop =
3195065Swnj 			     ti->ti_seq+ti->ti_len - (tp->rcv_nxt+tp->rcv_wnd);
3205300Sroot 			if (todrop > ti->ti_len)
3215065Swnj 				goto dropafterack;
3225065Swnj 			m_adj(m, -todrop);
3235065Swnj 			ti->ti_len -= todrop;
3245085Swnj 			ti->ti_flags &= ~(TH_PUSH|TH_FIN);
3255065Swnj 		}
3265065Swnj 	}
3274601Swnj 
3285065Swnj 	/*
3295065Swnj 	 * If the RST bit is set examine the state:
3305065Swnj 	 *    SYN_RECEIVED STATE:
3315065Swnj 	 *	If passive open, return to LISTEN state.
3325065Swnj 	 *	If active open, inform user that connection was refused.
3335065Swnj 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
3345065Swnj 	 *	Inform user that connection was reset, and close tcb.
3355065Swnj 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
3365065Swnj 	 *	Close the tcb.
3375065Swnj 	 */
3385065Swnj 	if (tiflags&TH_RST) switch (tp->t_state) {
3395267Sroot 
3405065Swnj 	case TCPS_SYN_RECEIVED:
3415065Swnj 		if (inp->inp_socket->so_options & SO_ACCEPTCONN) {
3425267Sroot 			/* a miniature tcp_close, but invisible to user */
3435267Sroot 			(void) m_free(dtom(tp->t_template));
3445267Sroot 			(void) m_free(dtom(tp));
3455267Sroot 			inp->inp_ppcb = 0;
3465267Sroot 			tp = tcp_newtcpcb(inp);
3475085Swnj 			tp->t_state = TCPS_LISTEN;
3485065Swnj 			goto drop;
3494601Swnj 		}
3505085Swnj 		tcp_drop(tp, ECONNREFUSED);
3515065Swnj 		goto drop;
3524601Swnj 
3535065Swnj 	case TCPS_ESTABLISHED:
3545065Swnj 	case TCPS_FIN_WAIT_1:
3555065Swnj 	case TCPS_FIN_WAIT_2:
3565065Swnj 	case TCPS_CLOSE_WAIT:
3575065Swnj 		tcp_drop(tp, ECONNRESET);
3585065Swnj 		goto drop;
3595065Swnj 
3605065Swnj 	case TCPS_CLOSING:
3615065Swnj 	case TCPS_LAST_ACK:
3625065Swnj 	case TCPS_TIME_WAIT:
3635065Swnj 		tcp_close(tp);
3645065Swnj 		goto drop;
3654601Swnj 	}
3664601Swnj 
3674601Swnj 	/*
3685065Swnj 	 * If a SYN is in the window, then this is an
3695065Swnj 	 * error and we send an RST and drop the connection.
3704601Swnj 	 */
3715065Swnj 	if (tiflags & TH_SYN) {
3725231Swnj 		tcp_drop(tp, ECONNRESET);
3735085Swnj 		goto dropwithreset;
3744601Swnj 	}
3754601Swnj 
3764601Swnj 	/*
3775065Swnj 	 * If the ACK bit is off we drop the segment and return.
3784601Swnj 	 */
3795085Swnj 	if ((tiflags & TH_ACK) == 0)
3805065Swnj 		goto drop;
3815065Swnj 
3825065Swnj 	/*
3835065Swnj 	 * Ack processing.
3845065Swnj 	 */
3854601Swnj 	switch (tp->t_state) {
3864601Swnj 
3875065Swnj 	/*
3885065Swnj 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
3895065Swnj 	 * ESTABLISHED state and continue processing, othewise
3905065Swnj 	 * send an RST.
3915065Swnj 	 */
3925065Swnj 	case TCPS_SYN_RECEIVED:
3935085Swnj 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
3945231Swnj 		    SEQ_GT(ti->ti_ack, tp->snd_max))
3955085Swnj 			goto dropwithreset;
3965244Sroot 		tp->snd_una++;			/* SYN acked */
3975357Sroot 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
3985357Sroot 			tp->snd_nxt = tp->snd_una;
3995244Sroot 		tp->t_timer[TCPT_REXMT] = 0;
4005391Swnj 		if (so->so_options & SO_ACCEPTCONN)
4015391Swnj 			so->so_state |= SS_CONNAWAITING;
4025085Swnj 		soisconnected(so);
4035085Swnj 		tp->t_state = TCPS_ESTABLISHED;
4045162Swnj 		(void) tcp_reass(tp, (struct tcpiphdr *)0);
4055244Sroot 		tp->snd_wl1 = ti->ti_seq - 1;
4065085Swnj 		/* fall into ... */
4074601Swnj 
4085065Swnj 	/*
4095065Swnj 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
4105065Swnj 	 * ACKs.  If the ack is in the range
4115231Swnj 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
4125065Swnj 	 * then advance tp->snd_una to ti->ti_ack and drop
4135065Swnj 	 * data from the retransmission queue.  If this ACK reflects
4145065Swnj 	 * more up to date window information we update our window information.
4155065Swnj 	 */
4165065Swnj 	case TCPS_ESTABLISHED:
4175065Swnj 	case TCPS_FIN_WAIT_1:
4185065Swnj 	case TCPS_FIN_WAIT_2:
4195065Swnj 	case TCPS_CLOSE_WAIT:
4205065Swnj 	case TCPS_CLOSING:
4215244Sroot 	case TCPS_LAST_ACK:
4225244Sroot 	case TCPS_TIME_WAIT:
4235085Swnj #define	ourfinisacked	(acked > 0)
4245085Swnj 
4255244Sroot 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
4265065Swnj 			break;
4275300Sroot 		if (SEQ_GT(ti->ti_ack, tp->snd_max))
4285065Swnj 			goto dropafterack;
4295085Swnj 		acked = ti->ti_ack - tp->snd_una;
4305307Sroot 		if (ti->ti_ack == tp->snd_max)
4315244Sroot 			tp->t_timer[TCPT_REXMT] = 0;
4325307Sroot 		else {
4335244Sroot 			TCPT_RANGESET(tp->t_timer[TCPT_REXMT],
4345244Sroot 			    tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX);
4355300Sroot 			tp->t_rtt = 0;
4365300Sroot 			tp->t_rxtshift = 0;
4375085Swnj 		}
4385307Sroot 		if (acked > so->so_snd.sb_cc) {
4395307Sroot 			sbdrop(&so->so_snd, so->so_snd.sb_cc);
4405307Sroot 			tp->snd_wnd -= so->so_snd.sb_cc;
4415307Sroot 		} else {
4425307Sroot 			sbdrop(&so->so_snd.sb_cc, acked);
4435307Sroot 			tp->snd_wnd -= acked;
4445307Sroot 			acked = 0;
4455307Sroot 		}
4465300Sroot 		if (so->so_snd.sb_flags & SB_WAIT)
4475300Sroot 			sowwakeup(so);
4485231Swnj 		tp->snd_una = ti->ti_ack;
4495357Sroot 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
4505357Sroot 			tp->snd_nxt = tp->snd_una;
4515162Swnj 
4525162Swnj 		/*
4535162Swnj 		 * If transmit timer is running and timed sequence
4545162Swnj 		 * number was acked, update smoothed round trip time.
4555162Swnj 		 */
4565162Swnj 		if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) {
4575244Sroot 			if (tp->t_srtt == 0)
4585244Sroot 				tp->t_srtt = tp->t_rtt;
4595244Sroot 			else
4605244Sroot 				tp->t_srtt =
4615244Sroot 				    tcp_alpha * tp->t_srtt +
4625244Sroot 				    (1 - tcp_alpha) * tp->t_rtt;
4635162Swnj 			tp->t_rtt = 0;
4645162Swnj 		}
4655162Swnj 
4664601Swnj 		switch (tp->t_state) {
4674601Swnj 
4685065Swnj 		/*
4695065Swnj 		 * In FIN_WAIT_1 STATE in addition to the processing
4705065Swnj 		 * for the ESTABLISHED state if our FIN is now acknowledged
4715085Swnj 		 * then enter FIN_WAIT_2.
4725065Swnj 		 */
4735065Swnj 		case TCPS_FIN_WAIT_1:
4745085Swnj 			if (ourfinisacked)
4755085Swnj 				tp->t_state = TCPS_FIN_WAIT_2;
4764601Swnj 			break;
4774601Swnj 
4785065Swnj 	 	/*
4795065Swnj 		 * In CLOSING STATE in addition to the processing for
4805065Swnj 		 * the ESTABLISHED state if the ACK acknowledges our FIN
4815065Swnj 		 * then enter the TIME-WAIT state, otherwise ignore
4825065Swnj 		 * the segment.
4835065Swnj 		 */
4845065Swnj 		case TCPS_CLOSING:
4855244Sroot 			if (ourfinisacked) {
4865065Swnj 				tp->t_state = TCPS_TIME_WAIT;
4875244Sroot 				tcp_canceltimers(tp);
4885244Sroot 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
4895244Sroot 				soisdisconnected(so);
4905244Sroot 			}
4915244Sroot 			break;
4924601Swnj 
4935065Swnj 		/*
4945085Swnj 		 * The only thing that can arrive in  LAST_ACK state
4955085Swnj 		 * is an acknowledgment of our FIN.  If our FIN is now
4965085Swnj 		 * acknowledged, delete the TCB, enter the closed state
4975085Swnj 		 * and return.
4985065Swnj 		 */
4995065Swnj 		case TCPS_LAST_ACK:
5005251Sroot 			if (ourfinisacked)
5015065Swnj 				tcp_close(tp);
5025065Swnj 			goto drop;
5034601Swnj 
5045065Swnj 		/*
5055065Swnj 		 * In TIME_WAIT state the only thing that should arrive
5065065Swnj 		 * is a retransmission of the remote FIN.  Acknowledge
5075065Swnj 		 * it and restart the finack timer.
5085065Swnj 		 */
5095065Swnj 		case TCPS_TIME_WAIT:
5105162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
5115065Swnj 			goto dropafterack;
5124601Swnj 		}
5135085Swnj #undef ourfinisacked
5145085Swnj 	}
5154601Swnj 
5165065Swnj step6:
5175065Swnj 	/*
5185244Sroot 	 * Update window information.
5195244Sroot 	 */
5205300Sroot 	if (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq &&
5215391Swnj 	    (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
5225300Sroot 	     tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd)) {
5235391Swnj /*
5245391Swnj printf("wl1 %x seq %x wl2 %x ack %x win %x wnd %x\n", tp->snd_wl1, ti->ti_seq, tp->snd_wl2, ti->ti_ack, ti->ti_win, tp->snd_wnd);
5255391Swnj */
5265244Sroot 		tp->snd_wnd = ti->ti_win;
5275244Sroot 		tp->snd_wl1 = ti->ti_seq;
5285244Sroot 		tp->snd_wl2 = ti->ti_ack;
5295244Sroot 		if (tp->snd_wnd > 0)
5305244Sroot 			tp->t_timer[TCPT_PERSIST] = 0;
5315244Sroot 	}
5325244Sroot 
5335244Sroot 	/*
5345419Swnj 	 * If an URG bit is set and in the segment and is greater than the
535*5440Swnj 	 * current known urgent pointer, then mark the data stream.
536*5440Swnj 	 * If the TCP is not doing out-of-band data, then indicate
537*5440Swnj 	 * urgent to the user.  This should not happen
5385065Swnj 	 * in CLOSE_WAIT, CLOSING, LAST-ACK or TIME_WAIT STATES since
5395065Swnj 	 * a FIN has been received from the remote side.  In these states
5405065Swnj 	 * we ignore the URG.
5415065Swnj 	 */
5425419Swnj 	if ((tiflags & TH_URG) && TCPS_HAVERCVDFIN(tp->t_state) == 0 &&
5435419Swnj 	    SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
5445419Swnj 		tp->rcv_up = ti->ti_seq + ti->ti_urp;
5455419Swnj 		so->so_oobmark = so->so_rcv.sb_cc +
5465419Swnj 		    (tp->rcv_up - tp->rcv_nxt) - 1;
5475419Swnj 		if (so->so_oobmark == 0)
5485419Swnj 			so->so_state |= SS_RCVATMARK;
549*5440Swnj #ifdef TCPTRUEOOB
550*5440Swnj 		if ((tp->t_flags & TF_DOOOB) == 0)
551*5440Swnj #endif
552*5440Swnj 		{
553*5440Swnj 			sohasoutofband(so);
554*5440Swnj 			tp->t_oobflags |= TCPOOB_HAVEDATA;
555*5440Swnj 		}
556*5440Swnj 
5575419Swnj 	}
5584601Swnj 
5594601Swnj 	/*
5605065Swnj 	 * Process the segment text, merging it into the TCP sequencing queue,
5615065Swnj 	 * and arranging for acknowledgment of receipt if necessary.
5625065Swnj 	 * This process logically involves adjusting tp->rcv_wnd as data
5635065Swnj 	 * is presented to the user (this happens in tcp_usrreq.c,
5645065Swnj 	 * case PRU_RCVD).  If a FIN has already been received on this
5655065Swnj 	 * connection then we just ignore the text.
5664601Swnj 	 */
5675263Swnj 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
5685263Swnj 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
5695085Swnj 		off += sizeof (struct ip);		/* drop IP header */
5705085Swnj 		m->m_off += off;
5715085Swnj 		m->m_len -= off;
5725065Swnj 		tiflags = tcp_reass(tp, ti);
573*5440Swnj 		if (tcpnodelack == 0)
574*5440Swnj 			tp->t_flags |= TF_DELACK;
575*5440Swnj 		else
576*5440Swnj 			tp->t_flags |= TF_ACKNOW;
5775244Sroot 	} else {
5784924Swnj 		m_freem(m);
5795263Swnj 		tiflags &= ~TH_FIN;
5805244Sroot 	}
5814601Swnj 
5824601Swnj 	/*
5835263Swnj 	 * If FIN is received ACK the FIN and let the user know
5845263Swnj 	 * that the connection is closing.
5854601Swnj 	 */
5865263Swnj 	if (tiflags & TH_FIN) {
5875244Sroot 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
5885244Sroot 			socantrcvmore(so);
5895244Sroot 			tp->t_flags |= TF_ACKNOW;
5905244Sroot 			tp->rcv_nxt++;
5915244Sroot 		}
5925065Swnj 		switch (tp->t_state) {
5934601Swnj 
5945065Swnj 	 	/*
5955065Swnj 		 * In SYN_RECEIVED and ESTABLISHED STATES
5965065Swnj 		 * enter the CLOSE_WAIT state.
5974884Swnj 		 */
5985065Swnj 		case TCPS_SYN_RECEIVED:
5995065Swnj 		case TCPS_ESTABLISHED:
6005065Swnj 			tp->t_state = TCPS_CLOSE_WAIT;
6015065Swnj 			break;
6024884Swnj 
6035065Swnj 	 	/*
6045085Swnj 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
6055085Swnj 		 * enter the CLOSING state.
6064884Swnj 		 */
6075065Swnj 		case TCPS_FIN_WAIT_1:
6085085Swnj 			tp->t_state = TCPS_CLOSING;
6095065Swnj 			break;
6104601Swnj 
6115065Swnj 	 	/*
6125065Swnj 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
6135065Swnj 		 * starting the time-wait timer, turning off the other
6145065Swnj 		 * standard timers.
6155065Swnj 		 */
6165065Swnj 		case TCPS_FIN_WAIT_2:
6175244Sroot 			tp->t_state = TCPS_TIME_WAIT;
6185074Swnj 			tcp_canceltimers(tp);
6195162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
6205244Sroot 			soisdisconnected(so);
6215065Swnj 			break;
6225065Swnj 
6234884Swnj 		/*
6245065Swnj 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
6254884Swnj 		 */
6265065Swnj 		case TCPS_TIME_WAIT:
6275162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
6285065Swnj 			break;
6295085Swnj 		}
6304601Swnj 	}
6315267Sroot 	if (so->so_options & SO_DEBUG)
6325267Sroot 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
6335085Swnj 
6345085Swnj 	/*
6355085Swnj 	 * Return any desired output.
6365085Swnj 	 */
6375085Swnj 	tcp_output(tp);
6385065Swnj 	return;
6395085Swnj 
6405065Swnj dropafterack:
6415085Swnj 	/*
6425244Sroot 	 * Generate an ACK dropping incoming segment.
6435085Swnj 	 * Make ACK reflect our state.
6445085Swnj 	 */
6455085Swnj 	if (tiflags & TH_RST)
6465085Swnj 		goto drop;
6475391Swnj 	tcp_respond(tp, ti, tp->rcv_nxt, tp->snd_nxt, TH_ACK);
6485231Swnj 	return;
6495085Swnj 
6505085Swnj dropwithreset:
651*5440Swnj 	if (om)
652*5440Swnj 		m_free(om);
6535085Swnj 	/*
6545244Sroot 	 * Generate a RST, dropping incoming segment.
6555085Swnj 	 * Make ACK acceptable to originator of segment.
6565085Swnj 	 */
6575085Swnj 	if (tiflags & TH_RST)
6585085Swnj 		goto drop;
6595085Swnj 	if (tiflags & TH_ACK)
6605391Swnj 		tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST);
6615085Swnj 	else {
6625085Swnj 		if (tiflags & TH_SYN)
6635085Swnj 			ti->ti_len++;
6645391Swnj 		tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, TH_RST|TH_ACK);
6655085Swnj 	}
6665231Swnj 	return;
6675085Swnj 
6685065Swnj drop:
6695085Swnj 	/*
6705085Swnj 	 * Drop space held by incoming segment and return.
6715085Swnj 	 */
6725065Swnj 	m_freem(m);
6735267Sroot 	return;
6745065Swnj }
6755065Swnj 
676*5440Swnj tcp_dooptions(tp, om)
677*5440Swnj 	struct tcpcb *tp;
678*5440Swnj 	struct mbuf *om;
6795419Swnj {
680*5440Swnj 	register u_char *cp;
681*5440Swnj 	int opt, optlen, cnt;
6825419Swnj 
683*5440Swnj 	cp = mtod(om, u_char *);
684*5440Swnj 	cnt = om->m_len;
685*5440Swnj 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
686*5440Swnj 		opt = cp[0];
687*5440Swnj 		if (opt == TCPOPT_EOL)
688*5440Swnj 			break;
689*5440Swnj 		if (opt == TCPOPT_NOP)
690*5440Swnj 			optlen = 1;
691*5440Swnj 		else
692*5440Swnj 			optlen = cp[1];
693*5440Swnj 		switch (opt) {
694*5440Swnj 
695*5440Swnj 		default:
696*5440Swnj 			break;
697*5440Swnj 
698*5440Swnj 		case TCPOPT_MAXSEG:
699*5440Swnj 			if (optlen != 4)
700*5440Swnj 				continue;
701*5440Swnj 			tp->t_maxseg = *(u_short *)(cp + 2);
702*5440Swnj #if vax
703*5440Swnj 			tp->t_maxseg = ntohs(tp->t_maxseg);
704*5440Swnj #endif
705*5440Swnj 			break;
706*5440Swnj 
707*5440Swnj #ifdef TCPTRUEOOB
708*5440Swnj 		case TCPOPT_WILLOOB:
709*5440Swnj 			tp->t_flags |= TF_DOOOB;
710*5440Swnj printf("tp %x dooob\n", tp);
711*5440Swnj 			break;
712*5440Swnj 
713*5440Swnj 		case TCPOPT_OOBDATA: {
714*5440Swnj 			int seq;
715*5440Swnj 
716*5440Swnj 			if (optlen != 4)
717*5440Swnj 				continue;
718*5440Swnj 			seq = cp[2];
719*5440Swnj 			if (seq < tp->t_iobseq)
720*5440Swnj 				seq += 256;
721*5440Swnj printf("oobdata cp[2] %d iobseq %d seq %d\n", cp[2], tp->t_iobseq, seq);
722*5440Swnj 			if (seq - tp->t_iobseq > 128) {
723*5440Swnj printf("bad seq\n");
724*5440Swnj 				tp->t_oobflags |= TCPOOB_OWEACK;
725*5440Swnj 				break;
726*5440Swnj 			}
727*5440Swnj 			tp->t_iobseq = cp[2];
728*5440Swnj 			tp->t_iobc = cp[3];
729*5440Swnj printf("take oob data %x input iobseq now %x\n", tp->t_iobc, tp->t_iobseq);
730*5440Swnj 			sohasoutofband(tp->t_inpcb->inp_socket);
731*5440Swnj 			break;
7325419Swnj 		}
733*5440Swnj 
734*5440Swnj 		case TCPOPT_OOBACK: {
735*5440Swnj 			int seq;
736*5440Swnj 
737*5440Swnj 			if (optlen != 4)
738*5440Swnj 				continue;
739*5440Swnj 			if (tp->t_oobseq != cp[2]) {
740*5440Swnj printf("wrong ack\n");
741*5440Swnj 				break;
742*5440Swnj 			}
743*5440Swnj printf("take oob ack %x and cancel rexmt\n", cp[2]);
744*5440Swnj 			tp->t_oobflags &= ~TCPOOB_NEEDACK;
745*5440Swnj 			tp->t_timer[TCPT_OOBREXMT] = 0;
7465419Swnj 			break;
747*5440Swnj 		}
748*5440Swnj #endif TCPTRUEOOB
749*5440Swnj 		}
7505419Swnj 	}
751*5440Swnj 	m_free(om);
7525419Swnj }
7535419Swnj 
7545419Swnj /*
7555065Swnj  * Insert segment ti into reassembly queue of tcp with
7565065Swnj  * control block tp.  Return TH_FIN if reassembly now includes
7575065Swnj  * a segment with FIN.
7585065Swnj  */
7595109Swnj tcp_reass(tp, ti)
7605065Swnj 	register struct tcpcb *tp;
7615065Swnj 	register struct tcpiphdr *ti;
7625065Swnj {
7635065Swnj 	register struct tcpiphdr *q;
7645085Swnj 	struct socket *so = tp->t_inpcb->inp_socket;
7655263Swnj 	struct mbuf *m;
7665263Swnj 	int flags;
7675085Swnj COUNT(TCP_REASS);
7685065Swnj 
7695065Swnj 	/*
7705162Swnj 	 * Call with ti==0 after become established to
7715162Swnj 	 * force pre-ESTABLISHED data up to user socket.
7725065Swnj 	 */
7735162Swnj 	if (ti == 0)
7745065Swnj 		goto present;
7754601Swnj 
7765065Swnj 	/*
7775065Swnj 	 * Find a segment which begins after this one does.
7785065Swnj 	 */
7795065Swnj 	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
7805065Swnj 	    q = (struct tcpiphdr *)q->ti_next)
7815065Swnj 		if (SEQ_GT(q->ti_seq, ti->ti_seq))
7825065Swnj 			break;
7834601Swnj 
7845065Swnj 	/*
7855065Swnj 	 * If there is a preceding segment, it may provide some of
7865065Swnj 	 * our data already.  If so, drop the data from the incoming
7875065Swnj 	 * segment.  If it provides all of our data, drop us.
7885065Swnj 	 */
7895065Swnj 	if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
7905065Swnj 		register int i;
7915065Swnj 		q = (struct tcpiphdr *)(q->ti_prev);
7925065Swnj 		/* conversion to int (in i) handles seq wraparound */
7935065Swnj 		i = q->ti_seq + q->ti_len - ti->ti_seq;
7945065Swnj 		if (i > 0) {
7954924Swnj 			if (i >= ti->ti_len)
7965065Swnj 				goto drop;
7975065Swnj 			m_adj(dtom(tp), i);
7985065Swnj 			ti->ti_len -= i;
7994924Swnj 			ti->ti_seq += i;
8004601Swnj 		}
8015065Swnj 		q = (struct tcpiphdr *)(q->ti_next);
8025065Swnj 	}
8034601Swnj 
8045065Swnj 	/*
8055065Swnj 	 * While we overlap succeeding segments trim them or,
8065065Swnj 	 * if they are completely covered, dequeue them.
8075065Swnj 	 */
8085065Swnj 	while (q != (struct tcpiphdr *)tp &&
8095065Swnj 	    SEQ_GT(ti->ti_seq + ti->ti_len, q->ti_seq)) {
8105065Swnj 		register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
8115065Swnj 		if (i < q->ti_len) {
8125065Swnj 			q->ti_len -= i;
8135065Swnj 			m_adj(dtom(q), i);
8145065Swnj 			break;
8154601Swnj 		}
8165065Swnj 		q = (struct tcpiphdr *)q->ti_next;
8175065Swnj 		m_freem(dtom(q->ti_prev));
8185065Swnj 		remque(q->ti_prev);
8195065Swnj 	}
8204601Swnj 
8215065Swnj 	/*
8225065Swnj 	 * Stick new segment in its place.
8235065Swnj 	 */
8245065Swnj 	insque(ti, q->ti_prev);
8254601Swnj 
8265065Swnj present:
8275065Swnj 	/*
8285244Sroot 	 * Present data to user, advancing rcv_nxt through
8295244Sroot 	 * completed sequence space.
8305065Swnj 	 */
8315263Swnj 	if (TCPS_HAVERCVDSYN(tp->t_state) == 0)
8325244Sroot 		return (0);
8334924Swnj 	ti = tp->seg_next;
8345263Swnj 	if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
8355263Swnj 		return (0);
8365263Swnj 	if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
8375263Swnj 		return (0);
8385263Swnj 	do {
8395244Sroot 		tp->rcv_nxt += ti->ti_len;
8405244Sroot 		flags = ti->ti_flags & TH_FIN;
8414924Swnj 		remque(ti);
8425263Swnj 		m = dtom(ti);
8434924Swnj 		ti = (struct tcpiphdr *)ti->ti_next;
8445263Swnj 		if (so->so_state & SS_CANTRCVMORE)
8455263Swnj 			(void) m_freem(m);
8465263Swnj 		else
8475263Swnj 			sbappend(&so->so_rcv, m);
8485263Swnj 	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
8495263Swnj 	sorwakeup(so);
8505065Swnj 	return (flags);
8515065Swnj drop:
8525065Swnj 	m_freem(dtom(ti));
8535263Swnj 	return (0);
8544601Swnj }
855