1*7267Swnj /* tcp_input.c 1.69 82/06/24 */ 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" 106351Ssam #include "../net/route.h" 115085Swnj #include "../net/in_pcb.h" 125085Swnj #include "../net/in_systm.h" 135085Swnj #include "../net/if.h" 144803Swnj #include "../net/ip.h" 154899Swnj #include "../net/ip_var.h" 164803Swnj #include "../net/tcp.h" 174803Swnj #include "../net/tcp_fsm.h" 185085Swnj #include "../net/tcp_seq.h" 195085Swnj #include "../net/tcp_timer.h" 204803Swnj #include "../net/tcp_var.h" 215085Swnj #include "../net/tcpip.h" 225267Sroot #include "../net/tcp_debug.h" 235109Swnj #include "../errno.h" 244601Swnj 255300Sroot int tcpprintfs = 0; 264679Swnj int tcpcksum = 1; 275244Sroot struct sockaddr_in tcp_in = { AF_INET }; 285267Sroot struct tcpiphdr tcp_saveti; 295440Swnj extern tcpnodelack; 304601Swnj 315267Sroot struct tcpcb *tcp_newtcpcb(); 325065Swnj /* 335065Swnj * TCP input routine, follows pages 65-76 of the 345065Swnj * protocol specification dated September, 1981 very closely. 355065Swnj */ 364924Swnj tcp_input(m0) 374924Swnj struct mbuf *m0; 384601Swnj { 394924Swnj register struct tcpiphdr *ti; 404924Swnj struct inpcb *inp; 414924Swnj register struct mbuf *m; 425440Swnj struct mbuf *om = 0; 434924Swnj int len, tlen, off; 445391Swnj register struct tcpcb *tp = 0; 454924Swnj register int tiflags; 464803Swnj struct socket *so; 475109Swnj int todrop, acked; 485267Sroot short ostate; 496028Sroot struct in_addr laddr; 504924Swnj 514924Swnj /* 525244Sroot * Get IP and TCP header together in first mbuf. 535244Sroot * Note: IP leaves IP header in first mbuf. 544924Swnj */ 554924Swnj m = m0; 565020Sroot ti = mtod(m, struct tcpiphdr *); 575244Sroot if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2)) 585208Swnj ip_stripoptions((struct ip *)ti, (struct mbuf *)0); 595307Sroot if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) { 605307Sroot if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 615085Swnj tcpstat.tcps_hdrops++; 625307Sroot return; 635085Swnj } 645085Swnj ti = mtod(m, struct tcpiphdr *); 655085Swnj } 664601Swnj 674601Swnj /* 685244Sroot * Checksum extended TCP header and data. 694601Swnj */ 704924Swnj tlen = ((struct ip *)ti)->ip_len; 714924Swnj len = sizeof (struct ip) + tlen; 724679Swnj if (tcpcksum) { 734924Swnj ti->ti_next = ti->ti_prev = 0; 744924Swnj ti->ti_x1 = 0; 755223Swnj ti->ti_len = (u_short)tlen; 766320Swnj #if vax || pdp11 776161Ssam ti->ti_len = htons((u_short)ti->ti_len); 785223Swnj #endif 795231Swnj if (ti->ti_sum = in_cksum(m, len)) { 804924Swnj tcpstat.tcps_badsum++; 816211Swnj if (tcpprintfs) 826211Swnj printf("tcp cksum %x\n", ti->ti_sum); 835085Swnj goto drop; 844601Swnj } 854601Swnj } 864601Swnj 874601Swnj /* 885244Sroot * Check that TCP offset makes sense, 895440Swnj * pull out TCP options and adjust length. 904601Swnj */ 914924Swnj off = ti->ti_off << 2; 925231Swnj if (off < sizeof (struct tcphdr) || off > tlen) { 934924Swnj tcpstat.tcps_badoff++; 945085Swnj goto drop; 954924Swnj } 966211Swnj tlen -= off; 976211Swnj ti->ti_len = tlen; 985440Swnj if (off > sizeof (struct tcphdr)) { 995440Swnj if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 1005440Swnj tcpstat.tcps_hdrops++; 1015440Swnj goto drop; 1025440Swnj } 1035440Swnj ti = mtod(m, struct tcpiphdr *); 1045440Swnj om = m_get(M_DONTWAIT); 1055440Swnj if (om == 0) 1065440Swnj goto drop; 1075440Swnj om->m_off = MMINOFF; 1085440Swnj om->m_len = off - sizeof (struct tcphdr); 1095440Swnj { caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 1106161Ssam bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len); 1115440Swnj m->m_len -= om->m_len; 1126161Ssam bcopy(op+om->m_len, op, 1136161Ssam (unsigned)(m->m_len-sizeof (struct tcpiphdr))); 1145440Swnj } 1155440Swnj } 1165065Swnj tiflags = ti->ti_flags; 1174924Swnj 1186093Sroot /* 1196211Swnj * Drop TCP and IP headers. 1206093Sroot */ 1216093Sroot off += sizeof (struct ip); 1226093Sroot m->m_off += off; 1236093Sroot m->m_len -= off; 1246093Sroot 1256320Swnj #if vax || pdp11 1264924Swnj /* 1275244Sroot * Convert TCP protocol specific fields to host format. 1285085Swnj */ 1295085Swnj ti->ti_seq = ntohl(ti->ti_seq); 1305085Swnj ti->ti_ack = ntohl(ti->ti_ack); 1315085Swnj ti->ti_win = ntohs(ti->ti_win); 1325085Swnj ti->ti_urp = ntohs(ti->ti_urp); 1335231Swnj #endif 1345085Swnj 1355085Swnj /* 1365994Swnj * Locate pcb for segment. On match, update the local 1375994Swnj * address stored in the block to reflect anchoring. 1384924Swnj */ 1395065Swnj inp = in_pcblookup 1406028Sroot (&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport, 1416028Sroot INPLOOKUP_WILDCARD); 1425065Swnj 1435065Swnj /* 1445065Swnj * If the state is CLOSED (i.e., TCB does not exist) then 1455244Sroot * all data in the incoming segment is discarded. 1465065Swnj */ 1475300Sroot if (inp == 0) 1485085Swnj goto dropwithreset; 1495065Swnj tp = intotcpcb(inp); 1505300Sroot if (tp == 0) 1515085Swnj goto dropwithreset; 1525109Swnj so = inp->inp_socket; 1535267Sroot if (so->so_options & SO_DEBUG) { 1545267Sroot ostate = tp->t_state; 1555267Sroot tcp_saveti = *ti; 1565267Sroot } 1574601Swnj 1584601Swnj /* 1595162Swnj * Segment received on connection. 1605162Swnj * Reset idle time and keep-alive timer. 1615162Swnj */ 1625162Swnj tp->t_idle = 0; 1635162Swnj tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 1645162Swnj 1655162Swnj /* 1665440Swnj * Process options. 1675440Swnj */ 1685440Swnj if (om) { 1695440Swnj tcp_dooptions(tp, om); 1705440Swnj om = 0; 1715440Swnj } 1725440Swnj 1735440Swnj /* 1745085Swnj * Calculate amount of space in receive window, 1755085Swnj * and then do TCP input processing. 1764601Swnj */ 1775085Swnj tp->rcv_wnd = sbspace(&so->so_rcv); 1785231Swnj if (tp->rcv_wnd < 0) 1795231Swnj tp->rcv_wnd = 0; 1804601Swnj 1814601Swnj switch (tp->t_state) { 1824601Swnj 1835065Swnj /* 1845065Swnj * If the state is LISTEN then ignore segment if it contains an RST. 1855065Swnj * If the segment contains an ACK then it is bad and send a RST. 1865065Swnj * If it does not contain a SYN then it is not interesting; drop it. 1875085Swnj * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 1885065Swnj * tp->iss, and send a segment: 1895085Swnj * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 1905065Swnj * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 1915065Swnj * Fill in remote peer address fields if not previously specified. 1925065Swnj * Enter SYN_RECEIVED state, and process any other fields of this 1935244Sroot * segment in this state. 1945065Swnj */ 1955065Swnj case TCPS_LISTEN: 1965065Swnj if (tiflags & TH_RST) 1975065Swnj goto drop; 1985300Sroot if (tiflags & TH_ACK) 1995085Swnj goto dropwithreset; 2005300Sroot if ((tiflags & TH_SYN) == 0) 2015065Swnj goto drop; 2025244Sroot tcp_in.sin_addr = ti->ti_src; 2035244Sroot tcp_in.sin_port = ti->ti_sport; 2046028Sroot laddr = inp->inp_laddr; 2056028Sroot if (inp->inp_laddr.s_addr == 0) 2066028Sroot inp->inp_laddr = ti->ti_dst; 2076161Ssam if (in_pcbconnect(inp, (struct sockaddr_in *)&tcp_in)) { 2086028Sroot inp->inp_laddr = laddr; 2095244Sroot goto drop; 2106028Sroot } 2115244Sroot tp->t_template = tcp_template(tp); 2125244Sroot if (tp->t_template == 0) { 2135244Sroot in_pcbdisconnect(inp); 2146028Sroot inp->inp_laddr = laddr; 2156320Swnj tp = 0; 2165244Sroot goto drop; 2175244Sroot } 2185085Swnj tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2; 2195065Swnj tp->irs = ti->ti_seq; 2205085Swnj tcp_sendseqinit(tp); 2215085Swnj tcp_rcvseqinit(tp); 2225065Swnj tp->t_state = TCPS_SYN_RECEIVED; 2235244Sroot tp->t_timer[TCPT_KEEP] = TCPTV_KEEP; 2245085Swnj goto trimthenstep6; 2254601Swnj 2265065Swnj /* 2275065Swnj * If the state is SYN_SENT: 2285065Swnj * if seg contains an ACK, but not for our SYN, drop the input. 2295065Swnj * if seg contains a RST, then drop the connection. 2305065Swnj * if seg does not contain SYN, then drop it. 2315065Swnj * Otherwise this is an acceptable SYN segment 2325065Swnj * initialize tp->rcv_nxt and tp->irs 2335065Swnj * if seg contains ack then advance tp->snd_una 2345065Swnj * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2355065Swnj * arrange for segment to be acked (eventually) 2365065Swnj * continue processing rest of data/controls, beginning with URG 2375065Swnj */ 2385065Swnj case TCPS_SYN_SENT: 2395065Swnj if ((tiflags & TH_ACK) && 2405300Sroot /* this should be SEQ_LT; is SEQ_LEQ for BBN vax TCP only */ 2415300Sroot (SEQ_LT(ti->ti_ack, tp->iss) || 2425231Swnj SEQ_GT(ti->ti_ack, tp->snd_max))) 2435085Swnj goto dropwithreset; 2445065Swnj if (tiflags & TH_RST) { 2456320Swnj if (tiflags & TH_ACK) { 2465267Sroot tcp_drop(tp, ECONNREFUSED); 2476320Swnj tp = 0; 2486320Swnj } 2495065Swnj goto drop; 2504601Swnj } 2515065Swnj if ((tiflags & TH_SYN) == 0) 2525065Swnj goto drop; 2535231Swnj tp->snd_una = ti->ti_ack; 2545357Sroot if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2555357Sroot tp->snd_nxt = tp->snd_una; 2565244Sroot tp->t_timer[TCPT_REXMT] = 0; 2575065Swnj tp->irs = ti->ti_seq; 2585085Swnj tcp_rcvseqinit(tp); 2595085Swnj tp->t_flags |= TF_ACKNOW; 2605162Swnj if (SEQ_GT(tp->snd_una, tp->iss)) { 2615391Swnj if (so->so_options & SO_ACCEPTCONN) 2625391Swnj so->so_state |= SS_CONNAWAITING; 2635244Sroot soisconnected(so); 2645065Swnj tp->t_state = TCPS_ESTABLISHED; 2655162Swnj (void) tcp_reass(tp, (struct tcpiphdr *)0); 2665162Swnj } else 2675085Swnj tp->t_state = TCPS_SYN_RECEIVED; 2685085Swnj goto trimthenstep6; 2695085Swnj 2705085Swnj trimthenstep6: 2715085Swnj /* 2725231Swnj * Advance ti->ti_seq to correspond to first data byte. 2735085Swnj * If data, trim to stay within window, 2745085Swnj * dropping FIN if necessary. 2755085Swnj */ 2765231Swnj ti->ti_seq++; 2775085Swnj if (ti->ti_len > tp->rcv_wnd) { 2785085Swnj todrop = ti->ti_len - tp->rcv_wnd; 2795085Swnj m_adj(m, -todrop); 2805085Swnj ti->ti_len = tp->rcv_wnd; 2815085Swnj ti->ti_flags &= ~TH_FIN; 2825065Swnj } 2835263Swnj tp->snd_wl1 = ti->ti_seq - 1; 2845085Swnj goto step6; 2855065Swnj } 2864601Swnj 2875065Swnj /* 2885065Swnj * States other than LISTEN or SYN_SENT. 2895065Swnj * First check that at least some bytes of segment are within 2905065Swnj * receive window. 2915065Swnj */ 2925065Swnj if (tp->rcv_wnd == 0) { 2935065Swnj /* 2945065Swnj * If window is closed can only take segments at 2955231Swnj * window edge, and have to drop data and PUSH from 2965065Swnj * incoming segments. 2975065Swnj */ 2985300Sroot if (tp->rcv_nxt != ti->ti_seq) 2995065Swnj goto dropafterack; 3005085Swnj if (ti->ti_len > 0) { 3015690Swnj m_adj(m, ti->ti_len); 3025085Swnj ti->ti_len = 0; 3035085Swnj ti->ti_flags &= ~(TH_PUSH|TH_FIN); 3045065Swnj } 3055065Swnj } else { 3065065Swnj /* 3075231Swnj * If segment begins before rcv_nxt, drop leading 3085065Swnj * data (and SYN); if nothing left, just ack. 3095065Swnj */ 3105690Swnj todrop = tp->rcv_nxt - ti->ti_seq; 3115690Swnj if (todrop > 0) { 3125085Swnj if (tiflags & TH_SYN) { 3135300Sroot tiflags &= ~TH_SYN; 3145690Swnj ti->ti_flags &= ~TH_SYN; 3155085Swnj ti->ti_seq++; 3165085Swnj if (ti->ti_urp > 1) 3175085Swnj ti->ti_urp--; 3185085Swnj else 3195085Swnj tiflags &= ~TH_URG; 3205085Swnj todrop--; 3215085Swnj } 3226211Swnj if (todrop > ti->ti_len || 3236211Swnj todrop == ti->ti_len && (tiflags&TH_FIN) == 0) 3245065Swnj goto dropafterack; 3255065Swnj m_adj(m, todrop); 3265065Swnj ti->ti_seq += todrop; 3275065Swnj ti->ti_len -= todrop; 3285085Swnj if (ti->ti_urp > todrop) 3295085Swnj ti->ti_urp -= todrop; 3305085Swnj else { 3315085Swnj tiflags &= ~TH_URG; 3325690Swnj ti->ti_flags &= ~TH_URG; 3335690Swnj ti->ti_urp = 0; 3345085Swnj } 3355065Swnj } 3365065Swnj /* 3375065Swnj * If segment ends after window, drop trailing data 3385085Swnj * (and PUSH and FIN); if nothing left, just ACK. 3395065Swnj */ 3405690Swnj todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 3415690Swnj if (todrop > 0) { 3426211Swnj if (todrop >= ti->ti_len) 3435065Swnj goto dropafterack; 3445065Swnj m_adj(m, -todrop); 3455065Swnj ti->ti_len -= todrop; 3465085Swnj ti->ti_flags &= ~(TH_PUSH|TH_FIN); 3475065Swnj } 3485065Swnj } 3494601Swnj 3505065Swnj /* 3515951Swnj * If a segment is received on a connection after the 3525951Swnj * user processes are gone, then RST the other end. 3535951Swnj */ 3545951Swnj if (so->so_state & SS_USERGONE) { 3555951Swnj tcp_close(tp); 3566266Swnj tp = 0; 3575951Swnj goto dropwithreset; 3585951Swnj } 3595951Swnj 3605951Swnj /* 3615065Swnj * If the RST bit is set examine the state: 3625065Swnj * SYN_RECEIVED STATE: 3635065Swnj * If passive open, return to LISTEN state. 3645065Swnj * If active open, inform user that connection was refused. 3655065Swnj * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 3665065Swnj * Inform user that connection was reset, and close tcb. 3675065Swnj * CLOSING, LAST_ACK, TIME_WAIT STATES 3685065Swnj * Close the tcb. 3695065Swnj */ 3705065Swnj if (tiflags&TH_RST) switch (tp->t_state) { 3715267Sroot 3725065Swnj case TCPS_SYN_RECEIVED: 3735065Swnj if (inp->inp_socket->so_options & SO_ACCEPTCONN) { 3745267Sroot /* a miniature tcp_close, but invisible to user */ 3755267Sroot (void) m_free(dtom(tp->t_template)); 3765267Sroot (void) m_free(dtom(tp)); 3775267Sroot inp->inp_ppcb = 0; 3785267Sroot tp = tcp_newtcpcb(inp); 3795085Swnj tp->t_state = TCPS_LISTEN; 3806028Sroot inp->inp_faddr.s_addr = 0; 3816028Sroot inp->inp_fport = 0; 3826028Sroot inp->inp_laddr.s_addr = 0; /* not quite right */ 3836320Swnj tp = 0; 3845065Swnj goto drop; 3854601Swnj } 3865085Swnj tcp_drop(tp, ECONNREFUSED); 3876320Swnj tp = 0; 3885065Swnj goto drop; 3894601Swnj 3905065Swnj case TCPS_ESTABLISHED: 3915065Swnj case TCPS_FIN_WAIT_1: 3925065Swnj case TCPS_FIN_WAIT_2: 3935065Swnj case TCPS_CLOSE_WAIT: 3945065Swnj tcp_drop(tp, ECONNRESET); 3956320Swnj tp = 0; 3965065Swnj goto drop; 3975065Swnj 3985065Swnj case TCPS_CLOSING: 3995065Swnj case TCPS_LAST_ACK: 4005065Swnj case TCPS_TIME_WAIT: 4015065Swnj tcp_close(tp); 4026320Swnj tp = 0; 4035065Swnj goto drop; 4044601Swnj } 4054601Swnj 4064601Swnj /* 4075065Swnj * If a SYN is in the window, then this is an 4085065Swnj * error and we send an RST and drop the connection. 4094601Swnj */ 4105065Swnj if (tiflags & TH_SYN) { 4115231Swnj tcp_drop(tp, ECONNRESET); 4126266Swnj tp = 0; 4135085Swnj goto dropwithreset; 4144601Swnj } 4154601Swnj 4164601Swnj /* 4175065Swnj * If the ACK bit is off we drop the segment and return. 4184601Swnj */ 4195085Swnj if ((tiflags & TH_ACK) == 0) 4205065Swnj goto drop; 4215065Swnj 4225065Swnj /* 4235065Swnj * Ack processing. 4245065Swnj */ 4254601Swnj switch (tp->t_state) { 4264601Swnj 4275065Swnj /* 4285065Swnj * In SYN_RECEIVED state if the ack ACKs our SYN then enter 4295065Swnj * ESTABLISHED state and continue processing, othewise 4305065Swnj * send an RST. 4315065Swnj */ 4325065Swnj case TCPS_SYN_RECEIVED: 4335085Swnj if (SEQ_GT(tp->snd_una, ti->ti_ack) || 4345231Swnj SEQ_GT(ti->ti_ack, tp->snd_max)) 4355085Swnj goto dropwithreset; 4365244Sroot tp->snd_una++; /* SYN acked */ 4375357Sroot if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 4385357Sroot tp->snd_nxt = tp->snd_una; 4395244Sroot tp->t_timer[TCPT_REXMT] = 0; 4405391Swnj if (so->so_options & SO_ACCEPTCONN) 4415391Swnj so->so_state |= SS_CONNAWAITING; 4425085Swnj soisconnected(so); 4435085Swnj tp->t_state = TCPS_ESTABLISHED; 4445162Swnj (void) tcp_reass(tp, (struct tcpiphdr *)0); 4455244Sroot tp->snd_wl1 = ti->ti_seq - 1; 4465085Swnj /* fall into ... */ 4474601Swnj 4485065Swnj /* 4495065Swnj * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 4505065Swnj * ACKs. If the ack is in the range 4515231Swnj * tp->snd_una < ti->ti_ack <= tp->snd_max 4525065Swnj * then advance tp->snd_una to ti->ti_ack and drop 4535065Swnj * data from the retransmission queue. If this ACK reflects 4545065Swnj * more up to date window information we update our window information. 4555065Swnj */ 4565065Swnj case TCPS_ESTABLISHED: 4575065Swnj case TCPS_FIN_WAIT_1: 4585065Swnj case TCPS_FIN_WAIT_2: 4595065Swnj case TCPS_CLOSE_WAIT: 4605065Swnj case TCPS_CLOSING: 4615244Sroot case TCPS_LAST_ACK: 4625244Sroot case TCPS_TIME_WAIT: 4635085Swnj #define ourfinisacked (acked > 0) 4645085Swnj 4655244Sroot if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) 4665065Swnj break; 4675300Sroot if (SEQ_GT(ti->ti_ack, tp->snd_max)) 4685065Swnj goto dropafterack; 4695085Swnj acked = ti->ti_ack - tp->snd_una; 4705951Swnj 4715951Swnj /* 4725951Swnj * If transmit timer is running and timed sequence 4735951Swnj * number was acked, update smoothed round trip time. 4745951Swnj */ 4755951Swnj if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) { 4765951Swnj if (tp->t_srtt == 0) 4775951Swnj tp->t_srtt = tp->t_rtt; 4785951Swnj else 4795951Swnj tp->t_srtt = 4805951Swnj tcp_alpha * tp->t_srtt + 4815951Swnj (1 - tcp_alpha) * tp->t_rtt; 4825951Swnj /* printf("rtt %d srtt*100 now %d\n", tp->t_rtt, (int)(tp->t_srtt*100)); */ 4835951Swnj tp->t_rtt = 0; 4845951Swnj } 4855951Swnj 4865307Sroot if (ti->ti_ack == tp->snd_max) 4875244Sroot tp->t_timer[TCPT_REXMT] = 0; 4885307Sroot else { 4895244Sroot TCPT_RANGESET(tp->t_timer[TCPT_REXMT], 4905244Sroot tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX); 4915951Swnj tp->t_rtt = 1; 4925300Sroot tp->t_rxtshift = 0; 4935085Swnj } 4945307Sroot if (acked > so->so_snd.sb_cc) { 4955307Sroot sbdrop(&so->so_snd, so->so_snd.sb_cc); 4965307Sroot tp->snd_wnd -= so->so_snd.sb_cc; 4975307Sroot } else { 4986161Ssam sbdrop(&so->so_snd, acked); 4995307Sroot tp->snd_wnd -= acked; 5005307Sroot acked = 0; 5015307Sroot } 5026434Swnj if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel) 5035300Sroot sowwakeup(so); 5045231Swnj tp->snd_una = ti->ti_ack; 5055357Sroot if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 5065357Sroot tp->snd_nxt = tp->snd_una; 5075162Swnj 5084601Swnj switch (tp->t_state) { 5094601Swnj 5105065Swnj /* 5115065Swnj * In FIN_WAIT_1 STATE in addition to the processing 5125065Swnj * for the ESTABLISHED state if our FIN is now acknowledged 5135085Swnj * then enter FIN_WAIT_2. 5145065Swnj */ 5155065Swnj case TCPS_FIN_WAIT_1: 5165896Swnj if (ourfinisacked) { 5175896Swnj /* 5185896Swnj * If we can't receive any more 5195896Swnj * data, then closing user can proceed. 5205896Swnj */ 5215896Swnj if (so->so_state & SS_CANTRCVMORE) 5225896Swnj soisdisconnected(so); 5235085Swnj tp->t_state = TCPS_FIN_WAIT_2; 5245896Swnj } 5254601Swnj break; 5264601Swnj 5275065Swnj /* 5285065Swnj * In CLOSING STATE in addition to the processing for 5295065Swnj * the ESTABLISHED state if the ACK acknowledges our FIN 5305065Swnj * then enter the TIME-WAIT state, otherwise ignore 5315065Swnj * the segment. 5325065Swnj */ 5335065Swnj case TCPS_CLOSING: 5345244Sroot if (ourfinisacked) { 5355065Swnj tp->t_state = TCPS_TIME_WAIT; 5365244Sroot tcp_canceltimers(tp); 5375244Sroot tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 5385244Sroot soisdisconnected(so); 5395244Sroot } 5405244Sroot break; 5414601Swnj 5425065Swnj /* 5435085Swnj * The only thing that can arrive in LAST_ACK state 5445085Swnj * is an acknowledgment of our FIN. If our FIN is now 5455085Swnj * acknowledged, delete the TCB, enter the closed state 5465085Swnj * and return. 5475065Swnj */ 5485065Swnj case TCPS_LAST_ACK: 5496320Swnj if (ourfinisacked) { 5505065Swnj tcp_close(tp); 5516320Swnj tp = 0; 5526320Swnj } 5535065Swnj goto drop; 5544601Swnj 5555065Swnj /* 5565065Swnj * In TIME_WAIT state the only thing that should arrive 5575065Swnj * is a retransmission of the remote FIN. Acknowledge 5585065Swnj * it and restart the finack timer. 5595065Swnj */ 5605065Swnj case TCPS_TIME_WAIT: 5615162Swnj tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 5625065Swnj goto dropafterack; 5634601Swnj } 5645085Swnj #undef ourfinisacked 5655085Swnj } 5664601Swnj 5675065Swnj step6: 5685065Swnj /* 5695244Sroot * Update window information. 5705244Sroot */ 5715300Sroot if (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 5725391Swnj (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 5735300Sroot tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd)) { 5745244Sroot tp->snd_wnd = ti->ti_win; 5755244Sroot tp->snd_wl1 = ti->ti_seq; 5765244Sroot tp->snd_wl2 = ti->ti_ack; 5775244Sroot if (tp->snd_wnd > 0) 5785244Sroot tp->t_timer[TCPT_PERSIST] = 0; 5795244Sroot } 5805244Sroot 5815244Sroot /* 5825547Swnj * Process segments with URG. 5835065Swnj */ 584*7267Swnj if ((tiflags & TH_URG) && ti->ti_urp && 585*7267Swnj TCPS_HAVERCVDFIN(tp->t_state) == 0) { 5865547Swnj /* 5875547Swnj * If this segment advances the known urgent pointer, 5885547Swnj * then mark the data stream. This should not happen 5895547Swnj * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 5905547Swnj * a FIN has been received from the remote side. 5915547Swnj * In these states we ignore the URG. 5925547Swnj */ 5935547Swnj if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 5945547Swnj tp->rcv_up = ti->ti_seq + ti->ti_urp; 5955547Swnj so->so_oobmark = so->so_rcv.sb_cc + 5965547Swnj (tp->rcv_up - tp->rcv_nxt) - 1; 5975547Swnj if (so->so_oobmark == 0) 5985547Swnj so->so_state |= SS_RCVATMARK; 5995440Swnj #ifdef TCPTRUEOOB 6005547Swnj if ((tp->t_flags & TF_DOOOB) == 0) 6015440Swnj #endif 6025547Swnj 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 */ 6115547Swnj if (ti->ti_urp <= ti->ti_len) { 6125547Swnj tcp_pulloutofband(so, ti); 6135547Swnj } 6145419Swnj } 6154601Swnj 6164601Swnj /* 6175065Swnj * Process the segment text, merging it into the TCP sequencing queue, 6185065Swnj * and arranging for acknowledgment of receipt if necessary. 6195065Swnj * This process logically involves adjusting tp->rcv_wnd as data 6205065Swnj * is presented to the user (this happens in tcp_usrreq.c, 6215065Swnj * case PRU_RCVD). If a FIN has already been received on this 6225065Swnj * connection then we just ignore the text. 6234601Swnj */ 6245263Swnj if ((ti->ti_len || (tiflags&TH_FIN)) && 6255263Swnj TCPS_HAVERCVDFIN(tp->t_state) == 0) { 6265065Swnj tiflags = tcp_reass(tp, ti); 6275440Swnj if (tcpnodelack == 0) 6285440Swnj tp->t_flags |= TF_DELACK; 6295440Swnj else 6305440Swnj tp->t_flags |= TF_ACKNOW; 6315244Sroot } else { 6324924Swnj m_freem(m); 6335263Swnj tiflags &= ~TH_FIN; 6345244Sroot } 6354601Swnj 6364601Swnj /* 6375263Swnj * If FIN is received ACK the FIN and let the user know 6385263Swnj * that the connection is closing. 6394601Swnj */ 6405263Swnj if (tiflags & TH_FIN) { 6415244Sroot if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 6425244Sroot socantrcvmore(so); 6435244Sroot tp->t_flags |= TF_ACKNOW; 6445244Sroot tp->rcv_nxt++; 6455244Sroot } 6465065Swnj switch (tp->t_state) { 6474601Swnj 6485065Swnj /* 6495065Swnj * In SYN_RECEIVED and ESTABLISHED STATES 6505065Swnj * enter the CLOSE_WAIT state. 6514884Swnj */ 6525065Swnj case TCPS_SYN_RECEIVED: 6535065Swnj case TCPS_ESTABLISHED: 6545065Swnj tp->t_state = TCPS_CLOSE_WAIT; 6555065Swnj break; 6564884Swnj 6575065Swnj /* 6585085Swnj * If still in FIN_WAIT_1 STATE FIN has not been acked so 6595085Swnj * enter the CLOSING state. 6604884Swnj */ 6615065Swnj case TCPS_FIN_WAIT_1: 6625085Swnj tp->t_state = TCPS_CLOSING; 6635065Swnj break; 6644601Swnj 6655065Swnj /* 6665065Swnj * In FIN_WAIT_2 state enter the TIME_WAIT state, 6675065Swnj * starting the time-wait timer, turning off the other 6685065Swnj * standard timers. 6695065Swnj */ 6705065Swnj case TCPS_FIN_WAIT_2: 6715244Sroot tp->t_state = TCPS_TIME_WAIT; 6725074Swnj tcp_canceltimers(tp); 6735162Swnj tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 6745244Sroot soisdisconnected(so); 6755065Swnj break; 6765065Swnj 6774884Swnj /* 6785065Swnj * In TIME_WAIT state restart the 2 MSL time_wait timer. 6794884Swnj */ 6805065Swnj case TCPS_TIME_WAIT: 6815162Swnj tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 6825065Swnj break; 6835085Swnj } 6844601Swnj } 6855267Sroot if (so->so_options & SO_DEBUG) 6865267Sroot tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 6875085Swnj 6885085Swnj /* 6895085Swnj * Return any desired output. 6905085Swnj */ 6916161Ssam (void) tcp_output(tp); 6925065Swnj return; 6935085Swnj 6945065Swnj dropafterack: 6955085Swnj /* 6966211Swnj * Generate an ACK dropping incoming segment if it occupies 6976211Swnj * sequence space, where the ACK reflects our state. 6985085Swnj */ 6996211Swnj if ((tiflags&TH_RST) || 7006211Swnj tlen == 0 && (tiflags&(TH_SYN|TH_FIN)) == 0) 7015085Swnj goto drop; 7026303Sroot if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 7036303Sroot tcp_trace(TA_RESPOND, ostate, tp, &tcp_saveti, 0); 7045391Swnj tcp_respond(tp, ti, tp->rcv_nxt, tp->snd_nxt, TH_ACK); 7055231Swnj return; 7065085Swnj 7075085Swnj dropwithreset: 7085440Swnj if (om) 7096161Ssam (void) m_free(om); 7105085Swnj /* 7115244Sroot * Generate a RST, dropping incoming segment. 7125085Swnj * Make ACK acceptable to originator of segment. 7135085Swnj */ 7145085Swnj if (tiflags & TH_RST) 7155085Swnj goto drop; 7165085Swnj if (tiflags & TH_ACK) 7175391Swnj tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST); 7185085Swnj else { 7195085Swnj if (tiflags & TH_SYN) 7205085Swnj ti->ti_len++; 7216211Swnj tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0, 7226211Swnj TH_RST|TH_ACK); 7235085Swnj } 7245231Swnj return; 7255085Swnj 7265065Swnj drop: 7275085Swnj /* 7285085Swnj * Drop space held by incoming segment and return. 7295085Swnj */ 7306303Sroot if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 7316303Sroot tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 7325065Swnj m_freem(m); 7335267Sroot return; 7345065Swnj } 7355065Swnj 7365440Swnj tcp_dooptions(tp, om) 7375440Swnj struct tcpcb *tp; 7385440Swnj struct mbuf *om; 7395419Swnj { 7405440Swnj register u_char *cp; 7415440Swnj int opt, optlen, cnt; 7425419Swnj 7435440Swnj cp = mtod(om, u_char *); 7445440Swnj cnt = om->m_len; 7455440Swnj for (; cnt > 0; cnt -= optlen, cp += optlen) { 7465440Swnj opt = cp[0]; 7475440Swnj if (opt == TCPOPT_EOL) 7485440Swnj break; 7495440Swnj if (opt == TCPOPT_NOP) 7505440Swnj optlen = 1; 7515440Swnj else 7525440Swnj optlen = cp[1]; 7535440Swnj switch (opt) { 7545440Swnj 7555440Swnj default: 7565440Swnj break; 7575440Swnj 7585440Swnj case TCPOPT_MAXSEG: 7595440Swnj if (optlen != 4) 7605440Swnj continue; 7615440Swnj tp->t_maxseg = *(u_short *)(cp + 2); 7626320Swnj #if vax || pdp11 7636161Ssam tp->t_maxseg = ntohs((u_short)tp->t_maxseg); 7645440Swnj #endif 7655440Swnj break; 7665440Swnj 7675440Swnj #ifdef TCPTRUEOOB 7685440Swnj case TCPOPT_WILLOOB: 7695440Swnj tp->t_flags |= TF_DOOOB; 7705440Swnj printf("tp %x dooob\n", tp); 7715440Swnj break; 7725440Swnj 7735440Swnj case TCPOPT_OOBDATA: { 7745440Swnj int seq; 7755547Swnj register struct socket *so = tp->t_inpcb->inp_socket; 7765547Swnj tcp_seq mark; 7775440Swnj 7785547Swnj if (optlen != 8) 7795440Swnj continue; 7805440Swnj seq = cp[2]; 7815440Swnj if (seq < tp->t_iobseq) 7825440Swnj seq += 256; 7835440Swnj printf("oobdata cp[2] %d iobseq %d seq %d\n", cp[2], tp->t_iobseq, seq); 7845440Swnj if (seq - tp->t_iobseq > 128) { 7855440Swnj printf("bad seq\n"); 7865440Swnj tp->t_oobflags |= TCPOOB_OWEACK; 7875440Swnj break; 7885440Swnj } 7895440Swnj tp->t_iobseq = cp[2]; 7905440Swnj tp->t_iobc = cp[3]; 7915547Swnj mark = *(tcp_seq *)(cp + 4); 7926320Swnj #if vax || pdp11 7935547Swnj mark = ntohl(mark); 7945547Swnj #endif 7955547Swnj so->so_oobmark = so->so_rcv.sb_cc + (mark-tp->rcv_nxt); 7965547Swnj if (so->so_oobmark == 0) 7975547Swnj so->so_state |= SS_RCVATMARK; 7985440Swnj printf("take oob data %x input iobseq now %x\n", tp->t_iobc, tp->t_iobseq); 7995547Swnj sohasoutofband(so); 8005440Swnj break; 8015419Swnj } 8025440Swnj 8035440Swnj case TCPOPT_OOBACK: { 8045440Swnj int seq; 8055440Swnj 8065440Swnj if (optlen != 4) 8075440Swnj continue; 8085440Swnj if (tp->t_oobseq != cp[2]) { 8095440Swnj printf("wrong ack\n"); 8105440Swnj break; 8115440Swnj } 8125440Swnj printf("take oob ack %x and cancel rexmt\n", cp[2]); 8135440Swnj tp->t_oobflags &= ~TCPOOB_NEEDACK; 8145440Swnj tp->t_timer[TCPT_OOBREXMT] = 0; 8155419Swnj break; 8165440Swnj } 8175440Swnj #endif TCPTRUEOOB 8185440Swnj } 8195419Swnj } 8206161Ssam (void) m_free(om); 8215419Swnj } 8225419Swnj 8235419Swnj /* 8245547Swnj * Pull out of band byte out of a segment so 8255547Swnj * it doesn't appear in the user's data queue. 8265547Swnj * It is still reflected in the segment length for 8275547Swnj * sequencing purposes. 8285547Swnj */ 8295547Swnj tcp_pulloutofband(so, ti) 8305547Swnj struct socket *so; 8315547Swnj struct tcpiphdr *ti; 8325547Swnj { 8335547Swnj register struct mbuf *m; 8346116Swnj int cnt = ti->ti_urp - 1; 8355547Swnj 8365547Swnj m = dtom(ti); 8375547Swnj while (cnt >= 0) { 8385547Swnj if (m->m_len > cnt) { 8395547Swnj char *cp = mtod(m, caddr_t) + cnt; 8405547Swnj struct tcpcb *tp = sototcpcb(so); 8415547Swnj 8425547Swnj tp->t_iobc = *cp; 8435547Swnj tp->t_oobflags |= TCPOOB_HAVEDATA; 8446161Ssam bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 8455547Swnj m->m_len--; 8465547Swnj return; 8475547Swnj } 8485547Swnj cnt -= m->m_len; 8495547Swnj m = m->m_next; 8505547Swnj if (m == 0) 8515547Swnj break; 8525547Swnj } 8535547Swnj panic("tcp_pulloutofband"); 8545547Swnj } 8555547Swnj 8565547Swnj /* 8575065Swnj * Insert segment ti into reassembly queue of tcp with 8585065Swnj * control block tp. Return TH_FIN if reassembly now includes 8595065Swnj * a segment with FIN. 8605065Swnj */ 8615109Swnj tcp_reass(tp, ti) 8625065Swnj register struct tcpcb *tp; 8635065Swnj register struct tcpiphdr *ti; 8645065Swnj { 8655065Swnj register struct tcpiphdr *q; 8665085Swnj struct socket *so = tp->t_inpcb->inp_socket; 8675263Swnj struct mbuf *m; 8685263Swnj int flags; 8695065Swnj 8705065Swnj /* 8715162Swnj * Call with ti==0 after become established to 8725162Swnj * force pre-ESTABLISHED data up to user socket. 8735065Swnj */ 8745162Swnj if (ti == 0) 8755065Swnj goto present; 8764601Swnj 8775065Swnj /* 8785065Swnj * Find a segment which begins after this one does. 8795065Swnj */ 8805065Swnj for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 8815065Swnj q = (struct tcpiphdr *)q->ti_next) 8825065Swnj if (SEQ_GT(q->ti_seq, ti->ti_seq)) 8835065Swnj break; 8844601Swnj 8855065Swnj /* 8865065Swnj * If there is a preceding segment, it may provide some of 8875065Swnj * our data already. If so, drop the data from the incoming 8885065Swnj * segment. If it provides all of our data, drop us. 8895065Swnj */ 8905065Swnj if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 8915065Swnj register int i; 8925690Swnj q = (struct tcpiphdr *)q->ti_prev; 8935065Swnj /* conversion to int (in i) handles seq wraparound */ 8945065Swnj i = q->ti_seq + q->ti_len - ti->ti_seq; 8955065Swnj if (i > 0) { 8964924Swnj if (i >= ti->ti_len) 8975065Swnj goto drop; 8985065Swnj m_adj(dtom(tp), i); 8995065Swnj ti->ti_len -= i; 9004924Swnj ti->ti_seq += i; 9014601Swnj } 9025065Swnj q = (struct tcpiphdr *)(q->ti_next); 9035065Swnj } 9044601Swnj 9055065Swnj /* 9065065Swnj * While we overlap succeeding segments trim them or, 9075065Swnj * if they are completely covered, dequeue them. 9085065Swnj */ 9095690Swnj while (q != (struct tcpiphdr *)tp) { 9105065Swnj register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 9115690Swnj if (i <= 0) 9125690Swnj break; 9135065Swnj if (i < q->ti_len) { 9145690Swnj q->ti_seq += i; 9155065Swnj q->ti_len -= i; 9165065Swnj m_adj(dtom(q), i); 9175065Swnj break; 9184601Swnj } 9195065Swnj q = (struct tcpiphdr *)q->ti_next; 9205623Swnj m = dtom(q->ti_prev); 9215065Swnj remque(q->ti_prev); 9225623Swnj m_freem(m); 9235065Swnj } 9244601Swnj 9255065Swnj /* 9265065Swnj * Stick new segment in its place. 9275065Swnj */ 9285065Swnj insque(ti, q->ti_prev); 9294601Swnj 9305065Swnj present: 9315065Swnj /* 9325244Sroot * Present data to user, advancing rcv_nxt through 9335244Sroot * completed sequence space. 9345065Swnj */ 9355263Swnj if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 9365244Sroot return (0); 9374924Swnj ti = tp->seg_next; 9385263Swnj if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 9395263Swnj return (0); 9405263Swnj if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 9415263Swnj return (0); 9425263Swnj do { 9435244Sroot tp->rcv_nxt += ti->ti_len; 9445244Sroot flags = ti->ti_flags & TH_FIN; 9454924Swnj remque(ti); 9465263Swnj m = dtom(ti); 9474924Swnj ti = (struct tcpiphdr *)ti->ti_next; 9485263Swnj if (so->so_state & SS_CANTRCVMORE) 9496161Ssam m_freem(m); 9505263Swnj else 9515263Swnj sbappend(&so->so_rcv, m); 9525263Swnj } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 9535263Swnj sorwakeup(so); 9545065Swnj return (flags); 9555065Swnj drop: 9565065Swnj m_freem(dtom(ti)); 9575263Swnj return (0); 9584601Swnj } 959