xref: /csrg-svn/sys/netinet/tcp_input.c (revision 11731)
1*11731Ssam /*	tcp_input.c	1.91	83/03/27	*/
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"
910894Ssam #include "../h/errno.h"
1010894Ssam 
1110894Ssam #include "../net/if.h"
1210894Ssam #include "../net/route.h"
1310894Ssam 
148401Swnj #include "../netinet/in.h"
158401Swnj #include "../netinet/in_pcb.h"
168401Swnj #include "../netinet/in_systm.h"
178401Swnj #include "../netinet/ip.h"
188401Swnj #include "../netinet/ip_var.h"
198401Swnj #include "../netinet/tcp.h"
208401Swnj #include "../netinet/tcp_fsm.h"
218401Swnj #include "../netinet/tcp_seq.h"
228401Swnj #include "../netinet/tcp_timer.h"
238401Swnj #include "../netinet/tcp_var.h"
248401Swnj #include "../netinet/tcpip.h"
258401Swnj #include "../netinet/tcp_debug.h"
264601Swnj 
275300Sroot int	tcpprintfs = 0;
284679Swnj int	tcpcksum = 1;
295267Sroot struct	tcpiphdr tcp_saveti;
305440Swnj extern	tcpnodelack;
314601Swnj 
325267Sroot struct	tcpcb *tcp_newtcpcb();
335065Swnj /*
345065Swnj  * TCP input routine, follows pages 65-76 of the
355065Swnj  * protocol specification dated September, 1981 very closely.
365065Swnj  */
374924Swnj tcp_input(m0)
384924Swnj 	struct mbuf *m0;
394601Swnj {
404924Swnj 	register struct tcpiphdr *ti;
414924Swnj 	struct inpcb *inp;
424924Swnj 	register struct mbuf *m;
435440Swnj 	struct mbuf *om = 0;
444924Swnj 	int len, tlen, off;
455391Swnj 	register struct tcpcb *tp = 0;
464924Swnj 	register int tiflags;
474803Swnj 	struct socket *so;
485109Swnj 	int todrop, acked;
495267Sroot 	short ostate;
506028Sroot 	struct in_addr laddr;
5110769Ssam 	int dropsocket = 0;
524924Swnj 
534924Swnj 	/*
545244Sroot 	 * Get IP and TCP header together in first mbuf.
555244Sroot 	 * Note: IP leaves IP header in first mbuf.
564924Swnj 	 */
574924Swnj 	m = m0;
585020Sroot 	ti = mtod(m, struct tcpiphdr *);
595244Sroot 	if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2))
605208Swnj 		ip_stripoptions((struct ip *)ti, (struct mbuf *)0);
615307Sroot 	if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) {
625307Sroot 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
635085Swnj 			tcpstat.tcps_hdrops++;
645307Sroot 			return;
655085Swnj 		}
665085Swnj 		ti = mtod(m, struct tcpiphdr *);
675085Swnj 	}
684601Swnj 
694601Swnj 	/*
705244Sroot 	 * Checksum extended TCP header and data.
714601Swnj 	 */
724924Swnj 	tlen = ((struct ip *)ti)->ip_len;
734924Swnj 	len = sizeof (struct ip) + tlen;
744679Swnj 	if (tcpcksum) {
754924Swnj 		ti->ti_next = ti->ti_prev = 0;
764924Swnj 		ti->ti_x1 = 0;
775223Swnj 		ti->ti_len = (u_short)tlen;
786161Ssam 		ti->ti_len = htons((u_short)ti->ti_len);
795231Swnj 		if (ti->ti_sum = in_cksum(m, len)) {
804924Swnj 			tcpstat.tcps_badsum++;
815085Swnj 			goto drop;
824601Swnj 		}
834601Swnj 	}
844601Swnj 
854601Swnj 	/*
865244Sroot 	 * Check that TCP offset makes sense,
875440Swnj 	 * 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 	}
946211Swnj 	tlen -= off;
956211Swnj 	ti->ti_len = tlen;
965440Swnj 	if (off > sizeof (struct tcphdr)) {
975440Swnj 		if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
985440Swnj 			tcpstat.tcps_hdrops++;
9911730Ssam 			return;
1005440Swnj 		}
1015440Swnj 		ti = mtod(m, struct tcpiphdr *);
1029642Ssam 		om = m_get(M_DONTWAIT, MT_DATA);
1035440Swnj 		if (om == 0)
1045440Swnj 			goto drop;
1055440Swnj 		om->m_len = off - sizeof (struct tcphdr);
1065440Swnj 		{ caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
1076161Ssam 		  bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len);
1085440Swnj 		  m->m_len -= om->m_len;
1096161Ssam 		  bcopy(op+om->m_len, op,
1106161Ssam 		   (unsigned)(m->m_len-sizeof (struct tcpiphdr)));
1115440Swnj 		}
1125440Swnj 	}
1135065Swnj 	tiflags = ti->ti_flags;
1144924Swnj 
1156093Sroot 	/*
1166211Swnj 	 * Drop TCP and IP headers.
1176093Sroot 	 */
1186093Sroot 	off += sizeof (struct ip);
1196093Sroot 	m->m_off += off;
1206093Sroot 	m->m_len -= off;
1216093Sroot 
1224924Swnj 	/*
1235244Sroot 	 * Convert TCP protocol specific fields to host format.
1245085Swnj 	 */
1255085Swnj 	ti->ti_seq = ntohl(ti->ti_seq);
1265085Swnj 	ti->ti_ack = ntohl(ti->ti_ack);
1275085Swnj 	ti->ti_win = ntohs(ti->ti_win);
1285085Swnj 	ti->ti_urp = ntohs(ti->ti_urp);
1295085Swnj 
1305085Swnj 	/*
1318271Sroot 	 * Locate pcb for segment.
1324924Swnj 	 */
1335065Swnj 	inp = in_pcblookup
1346028Sroot 		(&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport,
1356028Sroot 		INPLOOKUP_WILDCARD);
1365065Swnj 
1375065Swnj 	/*
1385065Swnj 	 * If the state is CLOSED (i.e., TCB does not exist) then
1395244Sroot 	 * all data in the incoming segment is discarded.
1405065Swnj 	 */
1415300Sroot 	if (inp == 0)
1425085Swnj 		goto dropwithreset;
1435065Swnj 	tp = intotcpcb(inp);
1445300Sroot 	if (tp == 0)
1455085Swnj 		goto dropwithreset;
1465109Swnj 	so = inp->inp_socket;
1475267Sroot 	if (so->so_options & SO_DEBUG) {
1485267Sroot 		ostate = tp->t_state;
1495267Sroot 		tcp_saveti = *ti;
1505267Sroot 	}
1517510Sroot 	if (so->so_options & SO_ACCEPTCONN) {
1527510Sroot 		so = sonewconn(so);
1537510Sroot 		if (so == 0)
1547510Sroot 			goto drop;
15510769Ssam 		/*
15610769Ssam 		 * This is ugly, but ....
15710769Ssam 		 *
15810769Ssam 		 * Mark socket as temporary until we're
15910769Ssam 		 * committed to keeping it.  The code at
16010769Ssam 		 * ``drop'' and ``dropwithreset'' check the
16110769Ssam 		 * flag dropsocket to see if the temporary
16210769Ssam 		 * socket created here should be discarded.
16310769Ssam 		 * We mark the socket as discardable until
16410769Ssam 		 * we're committed to it below in TCPS_LISTEN.
16510769Ssam 		 */
16610769Ssam 		dropsocket++;
1677510Sroot 		inp = (struct inpcb *)so->so_pcb;
1687510Sroot 		inp->inp_laddr = ti->ti_dst;
1697510Sroot 		inp->inp_lport = ti->ti_dport;
1707510Sroot 		tp = intotcpcb(inp);
1717510Sroot 		tp->t_state = TCPS_LISTEN;
1727510Sroot 	}
1734601Swnj 
1744601Swnj 	/*
1755162Swnj 	 * Segment received on connection.
1765162Swnj 	 * Reset idle time and keep-alive timer.
1775162Swnj 	 */
1785162Swnj 	tp->t_idle = 0;
1795162Swnj 	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
1805162Swnj 
1815162Swnj 	/*
1825440Swnj 	 * Process options.
1835440Swnj 	 */
1845440Swnj 	if (om) {
1855440Swnj 		tcp_dooptions(tp, om);
1865440Swnj 		om = 0;
1875440Swnj 	}
1885440Swnj 
1895440Swnj 	/*
1905085Swnj 	 * Calculate amount of space in receive window,
1915085Swnj 	 * and then do TCP input processing.
1924601Swnj 	 */
1935085Swnj 	tp->rcv_wnd = sbspace(&so->so_rcv);
1945231Swnj 	if (tp->rcv_wnd < 0)
1955231Swnj 		tp->rcv_wnd = 0;
1964601Swnj 
1974601Swnj 	switch (tp->t_state) {
1984601Swnj 
1995065Swnj 	/*
2005065Swnj 	 * If the state is LISTEN then ignore segment if it contains an RST.
2015065Swnj 	 * If the segment contains an ACK then it is bad and send a RST.
2025065Swnj 	 * If it does not contain a SYN then it is not interesting; drop it.
2035085Swnj 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
2045065Swnj 	 * tp->iss, and send a segment:
2055085Swnj 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
2065065Swnj 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
2075065Swnj 	 * Fill in remote peer address fields if not previously specified.
2085065Swnj 	 * Enter SYN_RECEIVED state, and process any other fields of this
2095244Sroot 	 * segment in this state.
2105065Swnj 	 */
2118271Sroot 	case TCPS_LISTEN: {
21210145Ssam 		struct mbuf *am;
2138271Sroot 		register struct sockaddr_in *sin;
2148271Sroot 
2155065Swnj 		if (tiflags & TH_RST)
2165065Swnj 			goto drop;
2175300Sroot 		if (tiflags & TH_ACK)
2185085Swnj 			goto dropwithreset;
2195300Sroot 		if ((tiflags & TH_SYN) == 0)
2205065Swnj 			goto drop;
22110145Ssam 		am = m_get(M_DONTWAIT, MT_SONAME);
22210145Ssam 		if (am == NULL)
22310145Ssam 			goto drop;
22410145Ssam 		am->m_len = sizeof (struct sockaddr_in);
2258599Sroot 		sin = mtod(am, struct sockaddr_in *);
2268271Sroot 		sin->sin_family = AF_INET;
2278271Sroot 		sin->sin_addr = ti->ti_src;
2288271Sroot 		sin->sin_port = ti->ti_sport;
2296028Sroot 		laddr = inp->inp_laddr;
23010145Ssam 		if (inp->inp_laddr.s_addr == INADDR_ANY)
2316028Sroot 			inp->inp_laddr = ti->ti_dst;
2328599Sroot 		if (in_pcbconnect(inp, am)) {
2336028Sroot 			inp->inp_laddr = laddr;
2348716Sroot 			(void) m_free(am);
2355244Sroot 			goto drop;
2366028Sroot 		}
2378716Sroot 		(void) m_free(am);
2385244Sroot 		tp->t_template = tcp_template(tp);
2395244Sroot 		if (tp->t_template == 0) {
2405244Sroot 			in_pcbdisconnect(inp);
2416028Sroot 			inp->inp_laddr = laddr;
2426320Swnj 			tp = 0;
2435244Sroot 			goto drop;
2445244Sroot 		}
2455085Swnj 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
2465065Swnj 		tp->irs = ti->ti_seq;
2475085Swnj 		tcp_sendseqinit(tp);
2485085Swnj 		tcp_rcvseqinit(tp);
2495065Swnj 		tp->t_state = TCPS_SYN_RECEIVED;
2505244Sroot 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
25110769Ssam 		dropsocket = 0;		/* committed to socket */
2525085Swnj 		goto trimthenstep6;
2538271Sroot 		}
2544601Swnj 
2555065Swnj 	/*
2565065Swnj 	 * If the state is SYN_SENT:
2575065Swnj 	 *	if seg contains an ACK, but not for our SYN, drop the input.
2585065Swnj 	 *	if seg contains a RST, then drop the connection.
2595065Swnj 	 *	if seg does not contain SYN, then drop it.
2605065Swnj 	 * Otherwise this is an acceptable SYN segment
2615065Swnj 	 *	initialize tp->rcv_nxt and tp->irs
2625065Swnj 	 *	if seg contains ack then advance tp->snd_una
2635065Swnj 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2645065Swnj 	 *	arrange for segment to be acked (eventually)
2655065Swnj 	 *	continue processing rest of data/controls, beginning with URG
2665065Swnj 	 */
2675065Swnj 	case TCPS_SYN_SENT:
2685065Swnj 		if ((tiflags & TH_ACK) &&
2695300Sroot /* this should be SEQ_LT; is SEQ_LEQ for BBN vax TCP only */
2705300Sroot 		    (SEQ_LT(ti->ti_ack, tp->iss) ||
2715231Swnj 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
2725085Swnj 			goto dropwithreset;
2735065Swnj 		if (tiflags & TH_RST) {
27410394Ssam 			if (tiflags & TH_ACK)
27510394Ssam 				tp = tcp_drop(tp, ECONNREFUSED);
2765065Swnj 			goto drop;
2774601Swnj 		}
2785065Swnj 		if ((tiflags & TH_SYN) == 0)
2795065Swnj 			goto drop;
2805231Swnj 		tp->snd_una = ti->ti_ack;
2815357Sroot 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2825357Sroot 			tp->snd_nxt = tp->snd_una;
2835244Sroot 		tp->t_timer[TCPT_REXMT] = 0;
2845065Swnj 		tp->irs = ti->ti_seq;
2855085Swnj 		tcp_rcvseqinit(tp);
2865085Swnj 		tp->t_flags |= TF_ACKNOW;
2875162Swnj 		if (SEQ_GT(tp->snd_una, tp->iss)) {
2885244Sroot 			soisconnected(so);
2895065Swnj 			tp->t_state = TCPS_ESTABLISHED;
2905162Swnj 			(void) tcp_reass(tp, (struct tcpiphdr *)0);
2915162Swnj 		} else
2925085Swnj 			tp->t_state = TCPS_SYN_RECEIVED;
2935085Swnj 		goto trimthenstep6;
2945085Swnj 
2955085Swnj trimthenstep6:
2965085Swnj 		/*
2975231Swnj 		 * Advance ti->ti_seq to correspond to first data byte.
2985085Swnj 		 * If data, trim to stay within window,
2995085Swnj 		 * dropping FIN if necessary.
3005085Swnj 		 */
3015231Swnj 		ti->ti_seq++;
3025085Swnj 		if (ti->ti_len > tp->rcv_wnd) {
3035085Swnj 			todrop = ti->ti_len - tp->rcv_wnd;
3045085Swnj 			m_adj(m, -todrop);
3055085Swnj 			ti->ti_len = tp->rcv_wnd;
3065085Swnj 			ti->ti_flags &= ~TH_FIN;
3075065Swnj 		}
3085263Swnj 		tp->snd_wl1 = ti->ti_seq - 1;
3095085Swnj 		goto step6;
3105065Swnj 	}
3114601Swnj 
3125065Swnj 	/*
3135065Swnj 	 * States other than LISTEN or SYN_SENT.
3145065Swnj 	 * First check that at least some bytes of segment are within
3155065Swnj 	 * receive window.
3165065Swnj 	 */
3175065Swnj 	if (tp->rcv_wnd == 0) {
3185065Swnj 		/*
3195065Swnj 		 * If window is closed can only take segments at
3205231Swnj 		 * window edge, and have to drop data and PUSH from
3215065Swnj 		 * incoming segments.
3225065Swnj 		 */
3235300Sroot 		if (tp->rcv_nxt != ti->ti_seq)
3245065Swnj 			goto dropafterack;
3255085Swnj 		if (ti->ti_len > 0) {
3265690Swnj 			m_adj(m, ti->ti_len);
3275085Swnj 			ti->ti_len = 0;
3285085Swnj 			ti->ti_flags &= ~(TH_PUSH|TH_FIN);
3295065Swnj 		}
3305065Swnj 	} else {
3315065Swnj 		/*
3325231Swnj 		 * If segment begins before rcv_nxt, drop leading
3335065Swnj 		 * data (and SYN); if nothing left, just ack.
3345065Swnj 		 */
3355690Swnj 		todrop = tp->rcv_nxt - ti->ti_seq;
3365690Swnj 		if (todrop > 0) {
3375085Swnj 			if (tiflags & TH_SYN) {
3385300Sroot 				tiflags &= ~TH_SYN;
3395690Swnj 				ti->ti_flags &= ~TH_SYN;
3405085Swnj 				ti->ti_seq++;
3415085Swnj 				if (ti->ti_urp > 1)
3425085Swnj 					ti->ti_urp--;
3435085Swnj 				else
3445085Swnj 					tiflags &= ~TH_URG;
3455085Swnj 				todrop--;
3465085Swnj 			}
3476211Swnj 			if (todrop > ti->ti_len ||
3486211Swnj 			    todrop == ti->ti_len && (tiflags&TH_FIN) == 0)
3495065Swnj 				goto dropafterack;
3505065Swnj 			m_adj(m, todrop);
3515065Swnj 			ti->ti_seq += todrop;
3525065Swnj 			ti->ti_len -= todrop;
3535085Swnj 			if (ti->ti_urp > todrop)
3545085Swnj 				ti->ti_urp -= todrop;
3555085Swnj 			else {
3565085Swnj 				tiflags &= ~TH_URG;
3575690Swnj 				ti->ti_flags &= ~TH_URG;
3585690Swnj 				ti->ti_urp = 0;
3595085Swnj 			}
3605065Swnj 		}
3615065Swnj 		/*
3625065Swnj 		 * If segment ends after window, drop trailing data
3635085Swnj 		 * (and PUSH and FIN); if nothing left, just ACK.
3645065Swnj 		 */
3655690Swnj 		todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
3665690Swnj 		if (todrop > 0) {
3676211Swnj 			if (todrop >= ti->ti_len)
3685065Swnj 				goto dropafterack;
3695065Swnj 			m_adj(m, -todrop);
3705065Swnj 			ti->ti_len -= todrop;
3715085Swnj 			ti->ti_flags &= ~(TH_PUSH|TH_FIN);
3725065Swnj 		}
3735065Swnj 	}
3744601Swnj 
3755065Swnj 	/*
37610013Ssam 	 * If data is received on a connection after the
3775951Swnj 	 * user processes are gone, then RST the other end.
3785951Swnj 	 */
37910394Ssam 	if ((so->so_state & SS_NOFDREF) && tp->t_state > TCPS_CLOSE_WAIT &&
38010203Ssam 	    ti->ti_len) {
38110394Ssam 		tp = tcp_close(tp);
3825951Swnj 		goto dropwithreset;
3835951Swnj 	}
3845951Swnj 
3855951Swnj 	/*
3865065Swnj 	 * If the RST bit is set examine the state:
3875065Swnj 	 *    SYN_RECEIVED STATE:
3885065Swnj 	 *	If passive open, return to LISTEN state.
3895065Swnj 	 *	If active open, inform user that connection was refused.
3905065Swnj 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
3915065Swnj 	 *	Inform user that connection was reset, and close tcb.
3925065Swnj 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
3935065Swnj 	 *	Close the tcb.
3945065Swnj 	 */
3955065Swnj 	if (tiflags&TH_RST) switch (tp->t_state) {
3965267Sroot 
3975065Swnj 	case TCPS_SYN_RECEIVED:
39810394Ssam 		tp = tcp_drop(tp, ECONNREFUSED);
3995065Swnj 		goto drop;
4004601Swnj 
4015065Swnj 	case TCPS_ESTABLISHED:
4025065Swnj 	case TCPS_FIN_WAIT_1:
4035065Swnj 	case TCPS_FIN_WAIT_2:
4045065Swnj 	case TCPS_CLOSE_WAIT:
40510394Ssam 		tp = tcp_drop(tp, ECONNRESET);
4065065Swnj 		goto drop;
4075065Swnj 
4085065Swnj 	case TCPS_CLOSING:
4095065Swnj 	case TCPS_LAST_ACK:
4105065Swnj 	case TCPS_TIME_WAIT:
41110394Ssam 		tp = tcp_close(tp);
4125065Swnj 		goto drop;
4134601Swnj 	}
4144601Swnj 
4154601Swnj 	/*
4165065Swnj 	 * If a SYN is in the window, then this is an
4175065Swnj 	 * error and we send an RST and drop the connection.
4184601Swnj 	 */
4195065Swnj 	if (tiflags & TH_SYN) {
42010394Ssam 		tp = tcp_drop(tp, ECONNRESET);
4215085Swnj 		goto dropwithreset;
4224601Swnj 	}
4234601Swnj 
4244601Swnj 	/*
4255065Swnj 	 * If the ACK bit is off we drop the segment and return.
4264601Swnj 	 */
4275085Swnj 	if ((tiflags & TH_ACK) == 0)
4285065Swnj 		goto drop;
4295065Swnj 
4305065Swnj 	/*
4315065Swnj 	 * Ack processing.
4325065Swnj 	 */
4334601Swnj 	switch (tp->t_state) {
4344601Swnj 
4355065Swnj 	/*
4365065Swnj 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
4375065Swnj 	 * ESTABLISHED state and continue processing, othewise
4385065Swnj 	 * send an RST.
4395065Swnj 	 */
4405065Swnj 	case TCPS_SYN_RECEIVED:
4415085Swnj 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
4425231Swnj 		    SEQ_GT(ti->ti_ack, tp->snd_max))
4435085Swnj 			goto dropwithreset;
4445244Sroot 		tp->snd_una++;			/* SYN acked */
4455357Sroot 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
4465357Sroot 			tp->snd_nxt = tp->snd_una;
4475244Sroot 		tp->t_timer[TCPT_REXMT] = 0;
4485085Swnj 		soisconnected(so);
4495085Swnj 		tp->t_state = TCPS_ESTABLISHED;
4505162Swnj 		(void) tcp_reass(tp, (struct tcpiphdr *)0);
4515244Sroot 		tp->snd_wl1 = ti->ti_seq - 1;
4525085Swnj 		/* fall into ... */
4534601Swnj 
4545065Swnj 	/*
4555065Swnj 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
4565065Swnj 	 * ACKs.  If the ack is in the range
4575231Swnj 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
4585065Swnj 	 * then advance tp->snd_una to ti->ti_ack and drop
4595065Swnj 	 * data from the retransmission queue.  If this ACK reflects
4605065Swnj 	 * more up to date window information we update our window information.
4615065Swnj 	 */
4625065Swnj 	case TCPS_ESTABLISHED:
4635065Swnj 	case TCPS_FIN_WAIT_1:
4645065Swnj 	case TCPS_FIN_WAIT_2:
4655065Swnj 	case TCPS_CLOSE_WAIT:
4665065Swnj 	case TCPS_CLOSING:
4675244Sroot 	case TCPS_LAST_ACK:
4685244Sroot 	case TCPS_TIME_WAIT:
4695085Swnj #define	ourfinisacked	(acked > 0)
4705085Swnj 
4715244Sroot 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
4725065Swnj 			break;
4735300Sroot 		if (SEQ_GT(ti->ti_ack, tp->snd_max))
4745065Swnj 			goto dropafterack;
4755085Swnj 		acked = ti->ti_ack - tp->snd_una;
4765951Swnj 
4775951Swnj 		/*
4785951Swnj 		 * If transmit timer is running and timed sequence
4795951Swnj 		 * number was acked, update smoothed round trip time.
4805951Swnj 		 */
4815951Swnj 		if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) {
4825951Swnj 			if (tp->t_srtt == 0)
4835951Swnj 				tp->t_srtt = tp->t_rtt;
4845951Swnj 			else
4855951Swnj 				tp->t_srtt =
4865951Swnj 				    tcp_alpha * tp->t_srtt +
4875951Swnj 				    (1 - tcp_alpha) * tp->t_rtt;
4885951Swnj 			tp->t_rtt = 0;
4895951Swnj 		}
4905951Swnj 
4915307Sroot 		if (ti->ti_ack == tp->snd_max)
4925244Sroot 			tp->t_timer[TCPT_REXMT] = 0;
4935307Sroot 		else {
4945244Sroot 			TCPT_RANGESET(tp->t_timer[TCPT_REXMT],
4955244Sroot 			    tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX);
4965951Swnj 			tp->t_rtt = 1;
4975300Sroot 			tp->t_rxtshift = 0;
4985085Swnj 		}
4995307Sroot 		if (acked > so->so_snd.sb_cc) {
5005307Sroot 			sbdrop(&so->so_snd, so->so_snd.sb_cc);
5015307Sroot 			tp->snd_wnd -= so->so_snd.sb_cc;
5025307Sroot 		} else {
5036161Ssam 			sbdrop(&so->so_snd, acked);
5045307Sroot 			tp->snd_wnd -= acked;
5055307Sroot 			acked = 0;
5065307Sroot 		}
5076434Swnj 		if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel)
5085300Sroot 			sowwakeup(so);
5095231Swnj 		tp->snd_una = ti->ti_ack;
5105357Sroot 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
5115357Sroot 			tp->snd_nxt = tp->snd_una;
5125162Swnj 
5134601Swnj 		switch (tp->t_state) {
5144601Swnj 
5155065Swnj 		/*
5165065Swnj 		 * In FIN_WAIT_1 STATE in addition to the processing
5175065Swnj 		 * for the ESTABLISHED state if our FIN is now acknowledged
5185085Swnj 		 * then enter FIN_WAIT_2.
5195065Swnj 		 */
5205065Swnj 		case TCPS_FIN_WAIT_1:
5215896Swnj 			if (ourfinisacked) {
5225896Swnj 				/*
5235896Swnj 				 * If we can't receive any more
5245896Swnj 				 * data, then closing user can proceed.
5255896Swnj 				 */
5265896Swnj 				if (so->so_state & SS_CANTRCVMORE)
5275896Swnj 					soisdisconnected(so);
5285085Swnj 				tp->t_state = TCPS_FIN_WAIT_2;
5295896Swnj 			}
5304601Swnj 			break;
5314601Swnj 
5325065Swnj 	 	/*
5335065Swnj 		 * In CLOSING STATE in addition to the processing for
5345065Swnj 		 * the ESTABLISHED state if the ACK acknowledges our FIN
5355065Swnj 		 * then enter the TIME-WAIT state, otherwise ignore
5365065Swnj 		 * the segment.
5375065Swnj 		 */
5385065Swnj 		case TCPS_CLOSING:
5395244Sroot 			if (ourfinisacked) {
5405065Swnj 				tp->t_state = TCPS_TIME_WAIT;
5415244Sroot 				tcp_canceltimers(tp);
5425244Sroot 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
5435244Sroot 				soisdisconnected(so);
5445244Sroot 			}
5455244Sroot 			break;
5464601Swnj 
5475065Swnj 		/*
5485085Swnj 		 * The only thing that can arrive in  LAST_ACK state
5495085Swnj 		 * is an acknowledgment of our FIN.  If our FIN is now
5505085Swnj 		 * acknowledged, delete the TCB, enter the closed state
5515085Swnj 		 * and return.
5525065Swnj 		 */
5535065Swnj 		case TCPS_LAST_ACK:
55410394Ssam 			if (ourfinisacked)
55510394Ssam 				tp = tcp_close(tp);
5565065Swnj 			goto drop;
5574601Swnj 
5585065Swnj 		/*
5595065Swnj 		 * In TIME_WAIT state the only thing that should arrive
5605065Swnj 		 * is a retransmission of the remote FIN.  Acknowledge
5615065Swnj 		 * it and restart the finack timer.
5625065Swnj 		 */
5635065Swnj 		case TCPS_TIME_WAIT:
5645162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
5655065Swnj 			goto dropafterack;
5664601Swnj 		}
5675085Swnj #undef ourfinisacked
5685085Swnj 	}
5694601Swnj 
5705065Swnj step6:
5715065Swnj 	/*
5725244Sroot 	 * Update window information.
5735244Sroot 	 */
5745300Sroot 	if (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq &&
5755391Swnj 	    (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
5765300Sroot 	     tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd)) {
5775244Sroot 		tp->snd_wnd = ti->ti_win;
5785244Sroot 		tp->snd_wl1 = ti->ti_seq;
5795244Sroot 		tp->snd_wl2 = ti->ti_ack;
5808599Sroot 		if (tp->snd_wnd != 0)
5815244Sroot 			tp->t_timer[TCPT_PERSIST] = 0;
5825244Sroot 	}
5835244Sroot 
5845244Sroot 	/*
5855547Swnj 	 * Process segments with URG.
5865065Swnj 	 */
5877267Swnj 	if ((tiflags & TH_URG) && ti->ti_urp &&
5887267Swnj 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
5895547Swnj 		/*
5905547Swnj 		 * If this segment advances the known urgent pointer,
5915547Swnj 		 * then mark the data stream.  This should not happen
5925547Swnj 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
5935547Swnj 		 * a FIN has been received from the remote side.
5945547Swnj 		 * In these states we ignore the URG.
5955547Swnj 		 */
5965547Swnj 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
5975547Swnj 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
5985547Swnj 			so->so_oobmark = so->so_rcv.sb_cc +
5995547Swnj 			    (tp->rcv_up - tp->rcv_nxt) - 1;
6005547Swnj 			if (so->so_oobmark == 0)
6015547Swnj 				so->so_state |= SS_RCVATMARK;
6028313Sroot 			sohasoutofband(so);
6035547Swnj 			tp->t_oobflags &= ~TCPOOB_HAVEDATA;
6045440Swnj 		}
6055547Swnj 		/*
6065547Swnj 		 * Remove out of band data so doesn't get presented to user.
6075547Swnj 		 * This can happen independent of advancing the URG pointer,
6085547Swnj 		 * but if two URG's are pending at once, some out-of-band
6095547Swnj 		 * data may creep in... ick.
6105547Swnj 		 */
6117510Sroot 		if (ti->ti_urp <= ti->ti_len)
6125547Swnj 			tcp_pulloutofband(so, ti);
6135419Swnj 	}
6144601Swnj 
6154601Swnj 	/*
6165065Swnj 	 * Process the segment text, merging it into the TCP sequencing queue,
6175065Swnj 	 * and arranging for acknowledgment of receipt if necessary.
6185065Swnj 	 * This process logically involves adjusting tp->rcv_wnd as data
6195065Swnj 	 * is presented to the user (this happens in tcp_usrreq.c,
6205065Swnj 	 * case PRU_RCVD).  If a FIN has already been received on this
6215065Swnj 	 * connection then we just ignore the text.
6224601Swnj 	 */
6235263Swnj 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
6245263Swnj 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
6255065Swnj 		tiflags = tcp_reass(tp, ti);
6265440Swnj 		if (tcpnodelack == 0)
6275440Swnj 			tp->t_flags |= TF_DELACK;
6285440Swnj 		else
6295440Swnj 			tp->t_flags |= TF_ACKNOW;
6305244Sroot 	} else {
6314924Swnj 		m_freem(m);
6325263Swnj 		tiflags &= ~TH_FIN;
6335244Sroot 	}
6344601Swnj 
6354601Swnj 	/*
6365263Swnj 	 * If FIN is received ACK the FIN and let the user know
6375263Swnj 	 * that the connection is closing.
6384601Swnj 	 */
6395263Swnj 	if (tiflags & TH_FIN) {
6405244Sroot 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
6415244Sroot 			socantrcvmore(so);
6425244Sroot 			tp->t_flags |= TF_ACKNOW;
6435244Sroot 			tp->rcv_nxt++;
6445244Sroot 		}
6455065Swnj 		switch (tp->t_state) {
6464601Swnj 
6475065Swnj 	 	/*
6485065Swnj 		 * In SYN_RECEIVED and ESTABLISHED STATES
6495065Swnj 		 * enter the CLOSE_WAIT state.
6504884Swnj 		 */
6515065Swnj 		case TCPS_SYN_RECEIVED:
6525065Swnj 		case TCPS_ESTABLISHED:
6535065Swnj 			tp->t_state = TCPS_CLOSE_WAIT;
6545065Swnj 			break;
6554884Swnj 
6565065Swnj 	 	/*
6575085Swnj 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
6585085Swnj 		 * enter the CLOSING state.
6594884Swnj 		 */
6605065Swnj 		case TCPS_FIN_WAIT_1:
6615085Swnj 			tp->t_state = TCPS_CLOSING;
6625065Swnj 			break;
6634601Swnj 
6645065Swnj 	 	/*
6655065Swnj 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
6665065Swnj 		 * starting the time-wait timer, turning off the other
6675065Swnj 		 * standard timers.
6685065Swnj 		 */
6695065Swnj 		case TCPS_FIN_WAIT_2:
6705244Sroot 			tp->t_state = TCPS_TIME_WAIT;
6715074Swnj 			tcp_canceltimers(tp);
6725162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
6735244Sroot 			soisdisconnected(so);
6745065Swnj 			break;
6755065Swnj 
6764884Swnj 		/*
6775065Swnj 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
6784884Swnj 		 */
6795065Swnj 		case TCPS_TIME_WAIT:
6805162Swnj 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
6815065Swnj 			break;
6825085Swnj 		}
6834601Swnj 	}
6845267Sroot 	if (so->so_options & SO_DEBUG)
6855267Sroot 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
6865085Swnj 
6875085Swnj 	/*
6885085Swnj 	 * Return any desired output.
6895085Swnj 	 */
6906161Ssam 	(void) tcp_output(tp);
6915065Swnj 	return;
6925085Swnj 
6935065Swnj dropafterack:
6945085Swnj 	/*
6956211Swnj 	 * Generate an ACK dropping incoming segment if it occupies
6966211Swnj 	 * sequence space, where the ACK reflects our state.
6975085Swnj 	 */
6986211Swnj 	if ((tiflags&TH_RST) ||
6996211Swnj 	    tlen == 0 && (tiflags&(TH_SYN|TH_FIN)) == 0)
7005085Swnj 		goto drop;
7016303Sroot 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
7026303Sroot 		tcp_trace(TA_RESPOND, ostate, tp, &tcp_saveti, 0);
7035391Swnj 	tcp_respond(tp, ti, tp->rcv_nxt, tp->snd_nxt, TH_ACK);
7045231Swnj 	return;
7055085Swnj 
7065085Swnj dropwithreset:
707*11731Ssam 	if (om) {
7086161Ssam 		(void) m_free(om);
709*11731Ssam 		om = 0;
710*11731Ssam 	}
7115085Swnj 	/*
7125244Sroot 	 * Generate a RST, dropping incoming segment.
7135085Swnj 	 * Make ACK acceptable to originator of segment.
7145085Swnj 	 */
7155085Swnj 	if (tiflags & TH_RST)
7165085Swnj 		goto drop;
7175085Swnj 	if (tiflags & TH_ACK)
7185391Swnj 		tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST);
7195085Swnj 	else {
7205085Swnj 		if (tiflags & TH_SYN)
7215085Swnj 			ti->ti_len++;
7226211Swnj 		tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0,
7236211Swnj 		    TH_RST|TH_ACK);
7245085Swnj 	}
72510769Ssam 	/* destroy temporarily created socket */
72610769Ssam 	if (dropsocket)
72710769Ssam 		(void) soabort(so);
7285231Swnj 	return;
7295085Swnj 
7305065Swnj drop:
73111730Ssam 	if (om)
73211730Ssam 		(void) m_free(om);
7335085Swnj 	/*
7345085Swnj 	 * Drop space held by incoming segment and return.
7355085Swnj 	 */
7366303Sroot 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
7376303Sroot 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
7385065Swnj 	m_freem(m);
73910769Ssam 	/* destroy temporarily created socket */
74010769Ssam 	if (dropsocket)
74110769Ssam 		(void) soabort(so);
7425267Sroot 	return;
7435065Swnj }
7445065Swnj 
7455440Swnj tcp_dooptions(tp, om)
7465440Swnj 	struct tcpcb *tp;
7475440Swnj 	struct mbuf *om;
7485419Swnj {
7495440Swnj 	register u_char *cp;
7505440Swnj 	int opt, optlen, cnt;
7515419Swnj 
7525440Swnj 	cp = mtod(om, u_char *);
7535440Swnj 	cnt = om->m_len;
7545440Swnj 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
7555440Swnj 		opt = cp[0];
7565440Swnj 		if (opt == TCPOPT_EOL)
7575440Swnj 			break;
7585440Swnj 		if (opt == TCPOPT_NOP)
7595440Swnj 			optlen = 1;
7605440Swnj 		else
7615440Swnj 			optlen = cp[1];
7625440Swnj 		switch (opt) {
7635440Swnj 
7645440Swnj 		default:
7655440Swnj 			break;
7665440Swnj 
7675440Swnj 		case TCPOPT_MAXSEG:
7685440Swnj 			if (optlen != 4)
7695440Swnj 				continue;
7705440Swnj 			tp->t_maxseg = *(u_short *)(cp + 2);
7716161Ssam 			tp->t_maxseg = ntohs((u_short)tp->t_maxseg);
7725440Swnj 			break;
7735419Swnj 		}
7745419Swnj 	}
7756161Ssam 	(void) m_free(om);
7765419Swnj }
7775419Swnj 
7785419Swnj /*
7795547Swnj  * Pull out of band byte out of a segment so
7805547Swnj  * it doesn't appear in the user's data queue.
7815547Swnj  * It is still reflected in the segment length for
7825547Swnj  * sequencing purposes.
7835547Swnj  */
7845547Swnj tcp_pulloutofband(so, ti)
7855547Swnj 	struct socket *so;
7865547Swnj 	struct tcpiphdr *ti;
7875547Swnj {
7885547Swnj 	register struct mbuf *m;
7896116Swnj 	int cnt = ti->ti_urp - 1;
7905547Swnj 
7915547Swnj 	m = dtom(ti);
7925547Swnj 	while (cnt >= 0) {
7935547Swnj 		if (m->m_len > cnt) {
7945547Swnj 			char *cp = mtod(m, caddr_t) + cnt;
7955547Swnj 			struct tcpcb *tp = sototcpcb(so);
7965547Swnj 
7975547Swnj 			tp->t_iobc = *cp;
7985547Swnj 			tp->t_oobflags |= TCPOOB_HAVEDATA;
7996161Ssam 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
8005547Swnj 			m->m_len--;
8015547Swnj 			return;
8025547Swnj 		}
8035547Swnj 		cnt -= m->m_len;
8045547Swnj 		m = m->m_next;
8055547Swnj 		if (m == 0)
8065547Swnj 			break;
8075547Swnj 	}
8085547Swnj 	panic("tcp_pulloutofband");
8095547Swnj }
8105547Swnj 
8115547Swnj /*
8125065Swnj  * Insert segment ti into reassembly queue of tcp with
8135065Swnj  * control block tp.  Return TH_FIN if reassembly now includes
8145065Swnj  * a segment with FIN.
8155065Swnj  */
8165109Swnj tcp_reass(tp, ti)
8175065Swnj 	register struct tcpcb *tp;
8185065Swnj 	register struct tcpiphdr *ti;
8195065Swnj {
8205065Swnj 	register struct tcpiphdr *q;
8215085Swnj 	struct socket *so = tp->t_inpcb->inp_socket;
8225263Swnj 	struct mbuf *m;
8235263Swnj 	int flags;
8245065Swnj 
8255065Swnj 	/*
8265162Swnj 	 * Call with ti==0 after become established to
8275162Swnj 	 * force pre-ESTABLISHED data up to user socket.
8285065Swnj 	 */
8295162Swnj 	if (ti == 0)
8305065Swnj 		goto present;
8314601Swnj 
8325065Swnj 	/*
8335065Swnj 	 * Find a segment which begins after this one does.
8345065Swnj 	 */
8355065Swnj 	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
8365065Swnj 	    q = (struct tcpiphdr *)q->ti_next)
8375065Swnj 		if (SEQ_GT(q->ti_seq, ti->ti_seq))
8385065Swnj 			break;
8394601Swnj 
8405065Swnj 	/*
8415065Swnj 	 * If there is a preceding segment, it may provide some of
8425065Swnj 	 * our data already.  If so, drop the data from the incoming
8435065Swnj 	 * segment.  If it provides all of our data, drop us.
8445065Swnj 	 */
8455065Swnj 	if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
8465065Swnj 		register int i;
8475690Swnj 		q = (struct tcpiphdr *)q->ti_prev;
8485065Swnj 		/* conversion to int (in i) handles seq wraparound */
8495065Swnj 		i = q->ti_seq + q->ti_len - ti->ti_seq;
8505065Swnj 		if (i > 0) {
8514924Swnj 			if (i >= ti->ti_len)
8525065Swnj 				goto drop;
8537338Swnj 			m_adj(dtom(ti), i);
8545065Swnj 			ti->ti_len -= i;
8554924Swnj 			ti->ti_seq += i;
8564601Swnj 		}
8575065Swnj 		q = (struct tcpiphdr *)(q->ti_next);
8585065Swnj 	}
8594601Swnj 
8605065Swnj 	/*
8615065Swnj 	 * While we overlap succeeding segments trim them or,
8625065Swnj 	 * if they are completely covered, dequeue them.
8635065Swnj 	 */
8645690Swnj 	while (q != (struct tcpiphdr *)tp) {
8655065Swnj 		register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
8665690Swnj 		if (i <= 0)
8675690Swnj 			break;
8685065Swnj 		if (i < q->ti_len) {
8695690Swnj 			q->ti_seq += i;
8705065Swnj 			q->ti_len -= i;
8715065Swnj 			m_adj(dtom(q), i);
8725065Swnj 			break;
8734601Swnj 		}
8745065Swnj 		q = (struct tcpiphdr *)q->ti_next;
8755623Swnj 		m = dtom(q->ti_prev);
8765065Swnj 		remque(q->ti_prev);
8775623Swnj 		m_freem(m);
8785065Swnj 	}
8794601Swnj 
8805065Swnj 	/*
8815065Swnj 	 * Stick new segment in its place.
8825065Swnj 	 */
8835065Swnj 	insque(ti, q->ti_prev);
8844601Swnj 
8855065Swnj present:
8865065Swnj 	/*
8875244Sroot 	 * Present data to user, advancing rcv_nxt through
8885244Sroot 	 * completed sequence space.
8895065Swnj 	 */
8905263Swnj 	if (TCPS_HAVERCVDSYN(tp->t_state) == 0)
8915244Sroot 		return (0);
8924924Swnj 	ti = tp->seg_next;
8935263Swnj 	if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
8945263Swnj 		return (0);
8955263Swnj 	if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
8965263Swnj 		return (0);
8975263Swnj 	do {
8985244Sroot 		tp->rcv_nxt += ti->ti_len;
8995244Sroot 		flags = ti->ti_flags & TH_FIN;
9004924Swnj 		remque(ti);
9015263Swnj 		m = dtom(ti);
9024924Swnj 		ti = (struct tcpiphdr *)ti->ti_next;
9035263Swnj 		if (so->so_state & SS_CANTRCVMORE)
9046161Ssam 			m_freem(m);
90510145Ssam 		else
9065263Swnj 			sbappend(&so->so_rcv, m);
9075263Swnj 	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
9085263Swnj 	sorwakeup(so);
9095065Swnj 	return (flags);
9105065Swnj drop:
9115065Swnj 	m_freem(dtom(ti));
9125263Swnj 	return (0);
9134601Swnj }
914