tcp_timewait.c revision 12172
11541Srgrimes/* 211150Swollman * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * Redistribution and use in source and binary forms, with or without 61541Srgrimes * modification, are permitted provided that the following conditions 71541Srgrimes * are met: 81541Srgrimes * 1. Redistributions of source code must retain the above copyright 91541Srgrimes * notice, this list of conditions and the following disclaimer. 101541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 111541Srgrimes * notice, this list of conditions and the following disclaimer in the 121541Srgrimes * documentation and/or other materials provided with the distribution. 131541Srgrimes * 3. All advertising materials mentioning features or use of this software 141541Srgrimes * must display the following acknowledgement: 151541Srgrimes * This product includes software developed by the University of 161541Srgrimes * California, Berkeley and its contributors. 171541Srgrimes * 4. Neither the name of the University nor the names of its contributors 181541Srgrimes * may be used to endorse or promote products derived from this software 191541Srgrimes * without specific prior written permission. 201541Srgrimes * 211541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 221541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 231541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 241541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 251541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 261541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 271541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 281541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 291541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 301541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 311541Srgrimes * SUCH DAMAGE. 321541Srgrimes * 3311150Swollman * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 3412172Sphk * $Id: tcp_subr.c,v 1.20 1995/10/16 18:21:20 wollman Exp $ 351541Srgrimes */ 361541Srgrimes 371541Srgrimes#include <sys/param.h> 381541Srgrimes#include <sys/proc.h> 391541Srgrimes#include <sys/systm.h> 4012172Sphk#include <sys/kernel.h> 4112172Sphk#include <sys/sysctl.h> 421541Srgrimes#include <sys/malloc.h> 431541Srgrimes#include <sys/mbuf.h> 441541Srgrimes#include <sys/socket.h> 451541Srgrimes#include <sys/socketvar.h> 461541Srgrimes#include <sys/protosw.h> 471541Srgrimes#include <sys/errno.h> 487684Sdg#include <sys/queue.h> 491541Srgrimes 501541Srgrimes#include <net/route.h> 511541Srgrimes#include <net/if.h> 521541Srgrimes 531541Srgrimes#include <netinet/in.h> 541541Srgrimes#include <netinet/in_systm.h> 551541Srgrimes#include <netinet/ip.h> 561541Srgrimes#include <netinet/in_pcb.h> 577090Sbde#include <netinet/in_var.h> 581541Srgrimes#include <netinet/ip_var.h> 591541Srgrimes#include <netinet/ip_icmp.h> 601541Srgrimes#include <netinet/tcp.h> 611541Srgrimes#include <netinet/tcp_fsm.h> 621541Srgrimes#include <netinet/tcp_seq.h> 631541Srgrimes#include <netinet/tcp_timer.h> 641541Srgrimes#include <netinet/tcp_var.h> 651541Srgrimes#include <netinet/tcpip.h> 666283Swollman#ifdef TCPDEBUG 676283Swollman#include <netinet/tcp_debug.h> 686283Swollman#endif 691541Srgrimes 701541Srgrimes/* patchable/settable parameters for tcp */ 711541Srgrimesint tcp_mssdflt = TCP_MSS; 7212172SphkSYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 7312172Sphk CTLFLAG_RW, &tcp_mssdflt , 0, ""); 741541Srgrimesint tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 7512172SphkSYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, 7612172Sphk CTLFLAG_RW, &tcp_rttdflt , 0, ""); 771541Srgrimesint tcp_do_rfc1323 = 1; 7812172SphkSYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, 7912172Sphk CTLFLAG_RW, &tcp_do_rfc1323 , 0, ""); 806283Swollmanint tcp_do_rfc1644 = 1; 8112172SphkSYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, 8212172Sphk CTLFLAG_RW, &tcp_do_rfc1644 , 0, ""); 836283Swollmanstatic void tcp_cleartaocache(void); 841541Srgrimes 857684Sdg/* 867684Sdg * Target size of TCP PCB hash table. Will be rounded down to a prime 877684Sdg * number. 887684Sdg */ 897684Sdg#ifndef TCBHASHSIZE 907684Sdg#define TCBHASHSIZE 128 917684Sdg#endif 921541Srgrimes 931541Srgrimes/* 941541Srgrimes * Tcp initialization 951541Srgrimes */ 961541Srgrimesvoid 971541Srgrimestcp_init() 981541Srgrimes{ 991541Srgrimes 10011150Swollman tcp_iss = random(); /* wrong, but better than a constant */ 1016283Swollman tcp_ccgen = 1; 1026283Swollman tcp_cleartaocache(); 1037684Sdg LIST_INIT(&tcb); 1047684Sdg tcbinfo.listhead = &tcb; 1057684Sdg tcbinfo.hashbase = phashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashsize); 1061541Srgrimes if (max_protohdr < sizeof(struct tcpiphdr)) 1071541Srgrimes max_protohdr = sizeof(struct tcpiphdr); 1081541Srgrimes if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 1091541Srgrimes panic("tcp_init"); 1101541Srgrimes} 1111541Srgrimes 1121541Srgrimes/* 1131541Srgrimes * Create template to be used to send tcp packets on a connection. 1141541Srgrimes * Call after host entry created, allocates an mbuf and fills 1151541Srgrimes * in a skeletal tcp/ip header, minimizing the amount of work 1161541Srgrimes * necessary when the connection is used. 1171541Srgrimes */ 1181541Srgrimesstruct tcpiphdr * 1191541Srgrimestcp_template(tp) 1201541Srgrimes struct tcpcb *tp; 1211541Srgrimes{ 1221541Srgrimes register struct inpcb *inp = tp->t_inpcb; 1231541Srgrimes register struct mbuf *m; 1241541Srgrimes register struct tcpiphdr *n; 1251541Srgrimes 1261541Srgrimes if ((n = tp->t_template) == 0) { 1271541Srgrimes m = m_get(M_DONTWAIT, MT_HEADER); 1281541Srgrimes if (m == NULL) 1291541Srgrimes return (0); 1301541Srgrimes m->m_len = sizeof (struct tcpiphdr); 1311541Srgrimes n = mtod(m, struct tcpiphdr *); 1321541Srgrimes } 1331541Srgrimes n->ti_next = n->ti_prev = 0; 1341541Srgrimes n->ti_x1 = 0; 1351541Srgrimes n->ti_pr = IPPROTO_TCP; 1361541Srgrimes n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 1371541Srgrimes n->ti_src = inp->inp_laddr; 1381541Srgrimes n->ti_dst = inp->inp_faddr; 1391541Srgrimes n->ti_sport = inp->inp_lport; 1401541Srgrimes n->ti_dport = inp->inp_fport; 1411541Srgrimes n->ti_seq = 0; 1421541Srgrimes n->ti_ack = 0; 1431541Srgrimes n->ti_x2 = 0; 1441541Srgrimes n->ti_off = 5; 1451541Srgrimes n->ti_flags = 0; 1461541Srgrimes n->ti_win = 0; 1471541Srgrimes n->ti_sum = 0; 1481541Srgrimes n->ti_urp = 0; 1491541Srgrimes return (n); 1501541Srgrimes} 1511541Srgrimes 1521541Srgrimes/* 1531541Srgrimes * Send a single message to the TCP at address specified by 1541541Srgrimes * the given TCP/IP header. If m == 0, then we make a copy 1551541Srgrimes * of the tcpiphdr at ti and send directly to the addressed host. 1561541Srgrimes * This is used to force keep alive messages out using the TCP 1571541Srgrimes * template for a connection tp->t_template. If flags are given 1581541Srgrimes * then we send a message back to the TCP which originated the 1591541Srgrimes * segment ti, and discard the mbuf containing it and any other 1601541Srgrimes * attached mbufs. 1611541Srgrimes * 1621541Srgrimes * In any case the ack and sequence number of the transmitted 1631541Srgrimes * segment are as specified by the parameters. 1641541Srgrimes */ 1651541Srgrimesvoid 1661541Srgrimestcp_respond(tp, ti, m, ack, seq, flags) 1671541Srgrimes struct tcpcb *tp; 1681541Srgrimes register struct tcpiphdr *ti; 1691541Srgrimes register struct mbuf *m; 1701541Srgrimes tcp_seq ack, seq; 1711541Srgrimes int flags; 1721541Srgrimes{ 1731541Srgrimes register int tlen; 1741541Srgrimes int win = 0; 1751541Srgrimes struct route *ro = 0; 1761541Srgrimes 1771541Srgrimes if (tp) { 1781541Srgrimes win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 1791541Srgrimes ro = &tp->t_inpcb->inp_route; 1801541Srgrimes } 1811541Srgrimes if (m == 0) { 1821541Srgrimes m = m_gethdr(M_DONTWAIT, MT_HEADER); 1831541Srgrimes if (m == NULL) 1841541Srgrimes return; 1851541Srgrimes#ifdef TCP_COMPAT_42 1861541Srgrimes tlen = 1; 1871541Srgrimes#else 1881541Srgrimes tlen = 0; 1891541Srgrimes#endif 1901541Srgrimes m->m_data += max_linkhdr; 1911541Srgrimes *mtod(m, struct tcpiphdr *) = *ti; 1921541Srgrimes ti = mtod(m, struct tcpiphdr *); 1931541Srgrimes flags = TH_ACK; 1941541Srgrimes } else { 1951541Srgrimes m_freem(m->m_next); 1961541Srgrimes m->m_next = 0; 1971541Srgrimes m->m_data = (caddr_t)ti; 1981541Srgrimes m->m_len = sizeof (struct tcpiphdr); 1991541Srgrimes tlen = 0; 2001541Srgrimes#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 2011541Srgrimes xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long); 2021541Srgrimes xchg(ti->ti_dport, ti->ti_sport, u_short); 2031541Srgrimes#undef xchg 2041541Srgrimes } 2051541Srgrimes ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); 2061541Srgrimes tlen += sizeof (struct tcpiphdr); 2071541Srgrimes m->m_len = tlen; 2081541Srgrimes m->m_pkthdr.len = tlen; 2091541Srgrimes m->m_pkthdr.rcvif = (struct ifnet *) 0; 2101541Srgrimes ti->ti_next = ti->ti_prev = 0; 2111541Srgrimes ti->ti_x1 = 0; 2121541Srgrimes ti->ti_seq = htonl(seq); 2131541Srgrimes ti->ti_ack = htonl(ack); 2141541Srgrimes ti->ti_x2 = 0; 2151541Srgrimes ti->ti_off = sizeof (struct tcphdr) >> 2; 2161541Srgrimes ti->ti_flags = flags; 2171541Srgrimes if (tp) 2181541Srgrimes ti->ti_win = htons((u_short) (win >> tp->rcv_scale)); 2191541Srgrimes else 2201541Srgrimes ti->ti_win = htons((u_short)win); 2211541Srgrimes ti->ti_urp = 0; 2221541Srgrimes ti->ti_sum = 0; 2231541Srgrimes ti->ti_sum = in_cksum(m, tlen); 2241541Srgrimes ((struct ip *)ti)->ip_len = tlen; 2251541Srgrimes ((struct ip *)ti)->ip_ttl = ip_defttl; 2266283Swollman#ifdef TCPDEBUG 2276283Swollman if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2286283Swollman tcp_trace(TA_OUTPUT, 0, tp, ti, 0); 2296283Swollman#endif 2301541Srgrimes (void) ip_output(m, NULL, ro, 0, NULL); 2311541Srgrimes} 2321541Srgrimes 2331541Srgrimes/* 2341541Srgrimes * Create a new TCP control block, making an 2351541Srgrimes * empty reassembly queue and hooking it to the argument 2361541Srgrimes * protocol control block. 2371541Srgrimes */ 2381541Srgrimesstruct tcpcb * 2391541Srgrimestcp_newtcpcb(inp) 2401541Srgrimes struct inpcb *inp; 2411541Srgrimes{ 2421541Srgrimes register struct tcpcb *tp; 2431541Srgrimes 2441541Srgrimes tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT); 2451541Srgrimes if (tp == NULL) 2461541Srgrimes return ((struct tcpcb *)0); 2471541Srgrimes bzero((char *) tp, sizeof(struct tcpcb)); 2481541Srgrimes tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp; 2496283Swollman tp->t_maxseg = tp->t_maxopd = tcp_mssdflt; 2501541Srgrimes 2516283Swollman if (tcp_do_rfc1323) 2526283Swollman tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2536283Swollman if (tcp_do_rfc1644) 2546283Swollman tp->t_flags |= TF_REQ_CC; 2551541Srgrimes tp->t_inpcb = inp; 2561541Srgrimes /* 2571541Srgrimes * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2581541Srgrimes * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 2591541Srgrimes * reasonable initial retransmit time. 2601541Srgrimes */ 2611541Srgrimes tp->t_srtt = TCPTV_SRTTBASE; 2621541Srgrimes tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2; 2631541Srgrimes tp->t_rttmin = TCPTV_MIN; 2648876Srgrimes TCPT_RANGESET(tp->t_rxtcur, 2651541Srgrimes ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1, 2661541Srgrimes TCPTV_MIN, TCPTV_REXMTMAX); 2671541Srgrimes tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2681541Srgrimes tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2691541Srgrimes inp->inp_ip.ip_ttl = ip_defttl; 2701541Srgrimes inp->inp_ppcb = (caddr_t)tp; 2711541Srgrimes return (tp); 2721541Srgrimes} 2731541Srgrimes 2741541Srgrimes/* 2751541Srgrimes * Drop a TCP connection, reporting 2761541Srgrimes * the specified error. If connection is synchronized, 2771541Srgrimes * then send a RST to peer. 2781541Srgrimes */ 2791541Srgrimesstruct tcpcb * 2801541Srgrimestcp_drop(tp, errno) 2811541Srgrimes register struct tcpcb *tp; 2821541Srgrimes int errno; 2831541Srgrimes{ 2841541Srgrimes struct socket *so = tp->t_inpcb->inp_socket; 2851541Srgrimes 2861541Srgrimes if (TCPS_HAVERCVDSYN(tp->t_state)) { 2871541Srgrimes tp->t_state = TCPS_CLOSED; 2881541Srgrimes (void) tcp_output(tp); 2891541Srgrimes tcpstat.tcps_drops++; 2901541Srgrimes } else 2911541Srgrimes tcpstat.tcps_conndrops++; 2921541Srgrimes if (errno == ETIMEDOUT && tp->t_softerror) 2931541Srgrimes errno = tp->t_softerror; 2941541Srgrimes so->so_error = errno; 2951541Srgrimes return (tcp_close(tp)); 2961541Srgrimes} 2971541Srgrimes 2981541Srgrimes/* 2991541Srgrimes * Close a TCP control block: 3001541Srgrimes * discard all space held by the tcp 3011541Srgrimes * discard internet protocol block 3021541Srgrimes * wake up any sleepers 3031541Srgrimes */ 3041541Srgrimesstruct tcpcb * 3051541Srgrimestcp_close(tp) 3061541Srgrimes register struct tcpcb *tp; 3071541Srgrimes{ 3081541Srgrimes register struct tcpiphdr *t; 3091541Srgrimes struct inpcb *inp = tp->t_inpcb; 3101541Srgrimes struct socket *so = inp->inp_socket; 3111541Srgrimes register struct mbuf *m; 3121541Srgrimes#ifdef RTV_RTT 3131541Srgrimes register struct rtentry *rt; 3141541Srgrimes 3151541Srgrimes /* 3169373Swollman * If we got enough samples through the srtt filter, 3179373Swollman * save the rtt and rttvar in the routing entry. 3189373Swollman * 'Enough' is arbitrarily defined as the 16 samples. 3199373Swollman * 16 samples is enough for the srtt filter to converge 3209373Swollman * to within 5% of the correct value; fewer samples and 3219373Swollman * we could save a very bogus rtt. 3221541Srgrimes * 3231541Srgrimes * Don't update the default route's characteristics and don't 3241541Srgrimes * update anything that the user "locked". 3251541Srgrimes */ 3269373Swollman if (tp->t_rttupdated >= 16 && 3271541Srgrimes (rt = inp->inp_route.ro_rt) && 3281541Srgrimes ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) { 3291549Srgrimes register u_long i = 0; 3301541Srgrimes 3311541Srgrimes if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 3321541Srgrimes i = tp->t_srtt * 3331541Srgrimes (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 3341541Srgrimes if (rt->rt_rmx.rmx_rtt && i) 3351541Srgrimes /* 3361541Srgrimes * filter this update to half the old & half 3371541Srgrimes * the new values, converting scale. 3381541Srgrimes * See route.h and tcp_var.h for a 3391541Srgrimes * description of the scaling constants. 3401541Srgrimes */ 3411541Srgrimes rt->rt_rmx.rmx_rtt = 3421541Srgrimes (rt->rt_rmx.rmx_rtt + i) / 2; 3431541Srgrimes else 3441541Srgrimes rt->rt_rmx.rmx_rtt = i; 3459263Swollman tcpstat.tcps_cachedrtt++; 3461541Srgrimes } 3471541Srgrimes if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 3481541Srgrimes i = tp->t_rttvar * 3491541Srgrimes (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 3501541Srgrimes if (rt->rt_rmx.rmx_rttvar && i) 3511541Srgrimes rt->rt_rmx.rmx_rttvar = 3521541Srgrimes (rt->rt_rmx.rmx_rttvar + i) / 2; 3531541Srgrimes else 3541541Srgrimes rt->rt_rmx.rmx_rttvar = i; 3559263Swollman tcpstat.tcps_cachedrttvar++; 3561541Srgrimes } 3571541Srgrimes /* 3581541Srgrimes * update the pipelimit (ssthresh) if it has been updated 3591541Srgrimes * already or if a pipesize was specified & the threshhold 3601541Srgrimes * got below half the pipesize. I.e., wait for bad news 3611541Srgrimes * before we start updating, then update on both good 3621541Srgrimes * and bad news. 3631541Srgrimes */ 3643444Sphk if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 3653444Sphk ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) || 3661541Srgrimes i < (rt->rt_rmx.rmx_sendpipe / 2)) { 3671541Srgrimes /* 3681541Srgrimes * convert the limit from user data bytes to 3691541Srgrimes * packets then to packet data bytes. 3701541Srgrimes */ 3711541Srgrimes i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 3721541Srgrimes if (i < 2) 3731541Srgrimes i = 2; 3741541Srgrimes i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr)); 3751541Srgrimes if (rt->rt_rmx.rmx_ssthresh) 3761541Srgrimes rt->rt_rmx.rmx_ssthresh = 3771541Srgrimes (rt->rt_rmx.rmx_ssthresh + i) / 2; 3781541Srgrimes else 3791541Srgrimes rt->rt_rmx.rmx_ssthresh = i; 3809263Swollman tcpstat.tcps_cachedssthresh++; 3811541Srgrimes } 3821541Srgrimes } 3831541Srgrimes#endif /* RTV_RTT */ 3841541Srgrimes /* free the reassembly queue, if any */ 3851541Srgrimes t = tp->seg_next; 3861541Srgrimes while (t != (struct tcpiphdr *)tp) { 3871541Srgrimes t = (struct tcpiphdr *)t->ti_next; 3881541Srgrimes m = REASS_MBUF((struct tcpiphdr *)t->ti_prev); 3891541Srgrimes remque(t->ti_prev); 3901541Srgrimes m_freem(m); 3911541Srgrimes } 3921541Srgrimes if (tp->t_template) 3931541Srgrimes (void) m_free(dtom(tp->t_template)); 3941541Srgrimes free(tp, M_PCB); 3951541Srgrimes inp->inp_ppcb = 0; 3961541Srgrimes soisdisconnected(so); 3971541Srgrimes in_pcbdetach(inp); 3981541Srgrimes tcpstat.tcps_closed++; 3991541Srgrimes return ((struct tcpcb *)0); 4001541Srgrimes} 4011541Srgrimes 4021541Srgrimesvoid 4031541Srgrimestcp_drain() 4041541Srgrimes{ 4051541Srgrimes 4061541Srgrimes} 4071541Srgrimes 4081541Srgrimes/* 4091541Srgrimes * Notify a tcp user of an asynchronous error; 4101541Srgrimes * store error as soft error, but wake up user 4111541Srgrimes * (for now, won't do anything until can select for soft error). 4121541Srgrimes */ 4131541Srgrimesvoid 4141541Srgrimestcp_notify(inp, error) 4151541Srgrimes struct inpcb *inp; 4161541Srgrimes int error; 4171541Srgrimes{ 4181541Srgrimes register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 4191541Srgrimes register struct socket *so = inp->inp_socket; 4201541Srgrimes 4211541Srgrimes /* 4221541Srgrimes * Ignore some errors if we are hooked up. 4231541Srgrimes * If connection hasn't completed, has retransmitted several times, 4241541Srgrimes * and receives a second error, give up now. This is better 4251541Srgrimes * than waiting a long time to establish a connection that 4261541Srgrimes * can never complete. 4271541Srgrimes */ 4281541Srgrimes if (tp->t_state == TCPS_ESTABLISHED && 4291541Srgrimes (error == EHOSTUNREACH || error == ENETUNREACH || 4301541Srgrimes error == EHOSTDOWN)) { 4311541Srgrimes return; 4321541Srgrimes } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 4331541Srgrimes tp->t_softerror) 4341541Srgrimes so->so_error = error; 4358876Srgrimes else 4361541Srgrimes tp->t_softerror = error; 4371541Srgrimes wakeup((caddr_t) &so->so_timeo); 4381541Srgrimes sorwakeup(so); 4391541Srgrimes sowwakeup(so); 4401541Srgrimes} 4411541Srgrimes 4421541Srgrimesvoid 4431541Srgrimestcp_ctlinput(cmd, sa, ip) 4441541Srgrimes int cmd; 4451541Srgrimes struct sockaddr *sa; 4461541Srgrimes register struct ip *ip; 4471541Srgrimes{ 4481541Srgrimes register struct tcphdr *th; 4491541Srgrimes void (*notify) __P((struct inpcb *, int)) = tcp_notify; 4501541Srgrimes 4511541Srgrimes if (cmd == PRC_QUENCH) 4521541Srgrimes notify = tcp_quench; 45310881Swollman#ifdef MTUDISC 45410881Swollman else if (cmd == PRC_MSGSIZE) 45510881Swollman notify = tcp_mtudisc; 45610881Swollman#endif /* MTUDISC */ 4571541Srgrimes else if (!PRC_IS_REDIRECT(cmd) && 4581541Srgrimes ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)) 4591541Srgrimes return; 4601541Srgrimes if (ip) { 4611541Srgrimes th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 4621541Srgrimes in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport, 4631541Srgrimes cmd, notify); 4641541Srgrimes } else 4651541Srgrimes in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify); 4661541Srgrimes} 4671541Srgrimes 4681541Srgrimes/* 4691541Srgrimes * When a source quench is received, close congestion window 4701541Srgrimes * to one segment. We will gradually open it again as we proceed. 4711541Srgrimes */ 4721541Srgrimesvoid 4731541Srgrimestcp_quench(inp, errno) 4741541Srgrimes struct inpcb *inp; 4751541Srgrimes int errno; 4761541Srgrimes{ 4771541Srgrimes struct tcpcb *tp = intotcpcb(inp); 4781541Srgrimes 4791541Srgrimes if (tp) 4801541Srgrimes tp->snd_cwnd = tp->t_maxseg; 4811541Srgrimes} 4826283Swollman 48310881Swollman#ifdef MTUDISC 4846283Swollman/* 48510881Swollman * When `need fragmentation' ICMP is received, update our idea of the MSS 48610881Swollman * based on the new value in the route. Also nudge TCP to send something, 48710881Swollman * since we know the packet we just sent was dropped. 48810930Swollman * This duplicates some code in the tcp_mss() function in tcp_input.c. 48910881Swollman */ 49011537Swollmanvoid 49110881Swollmantcp_mtudisc(inp, errno) 49210881Swollman struct inpcb *inp; 49310881Swollman int errno; 49410881Swollman{ 49510881Swollman struct tcpcb *tp = intotcpcb(inp); 49610930Swollman struct rtentry *rt; 49710930Swollman struct rmxp_tao *taop; 49810930Swollman struct socket *so = inp->inp_socket; 49910930Swollman int offered; 50010930Swollman int mss; 50110881Swollman 50210930Swollman if (tp) { 50310930Swollman rt = tcp_rtlookup(inp); 50410930Swollman if (!rt || !rt->rt_rmx.rmx_mtu) { 50510930Swollman tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 50610930Swollman return; 50710930Swollman } 50810930Swollman taop = rmx_taop(rt->rt_rmx); 50910930Swollman offered = taop->tao_mssopt; 51010930Swollman mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 51110956Swollman mss = min(mss, offered); 51211415Swollman if (tp->t_maxopd <= mss) 51311415Swollman return; 51410930Swollman tp->t_maxopd = mss; 51510930Swollman 51610930Swollman if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 51710930Swollman (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 51810930Swollman mss -= TCPOLEN_TSTAMP_APPA; 51910930Swollman if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 52010930Swollman (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 52110930Swollman mss -= TCPOLEN_CC_APPA; 52210930Swollman#if (MCLBYTES & (MCLBYTES - 1)) == 0 52310930Swollman if (mss > MCLBYTES) 52410930Swollman mss &= ~(MCLBYTES-1); 52510930Swollman#else 52610930Swollman if (mss > MCLBYTES) 52710930Swollman mss = mss / MCLBYTES * MCLBYTES; 52810881Swollman#endif 52910930Swollman if (so->so_snd.sb_hiwat < mss) 53010930Swollman mss = so->so_snd.sb_hiwat; 53110930Swollman 53210930Swollman tp->t_maxseg = mss; 53310930Swollman 53411450Swollman tcpstat.tcps_mturesent++; 53511450Swollman tp->t_rtt = 0; 53611450Swollman tp->snd_nxt = tp->snd_una; 53711450Swollman tcp_output(tp); 53810930Swollman } 53910881Swollman} 54010881Swollman#endif /* MTUDISC */ 54110881Swollman 54210881Swollman/* 5436283Swollman * Look-up the routing entry to the peer of this inpcb. If no route 5446283Swollman * is found and it cannot be allocated the return NULL. This routine 5456283Swollman * is called by TCP routines that access the rmx structure and by tcp_mss 5466283Swollman * to get the interface MTU. 5476283Swollman */ 5486283Swollmanstruct rtentry * 5496283Swollmantcp_rtlookup(inp) 5506283Swollman struct inpcb *inp; 5516283Swollman{ 5526283Swollman struct route *ro; 5536283Swollman struct rtentry *rt; 5546283Swollman 5556283Swollman ro = &inp->inp_route; 5566283Swollman rt = ro->ro_rt; 5576283Swollman if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 5586283Swollman /* No route yet, so try to acquire one */ 5596283Swollman if (inp->inp_faddr.s_addr != INADDR_ANY) { 5606283Swollman ro->ro_dst.sa_family = AF_INET; 5616283Swollman ro->ro_dst.sa_len = sizeof(ro->ro_dst); 5626283Swollman ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 5636283Swollman inp->inp_faddr; 5646283Swollman rtalloc(ro); 5656283Swollman rt = ro->ro_rt; 5666283Swollman } 5676283Swollman } 5686283Swollman return rt; 5696283Swollman} 5706283Swollman 5716283Swollman/* 5726283Swollman * Return a pointer to the cached information about the remote host. 5736283Swollman * The cached information is stored in the protocol specific part of 5746283Swollman * the route metrics. 5756283Swollman */ 5766283Swollmanstruct rmxp_tao * 5776283Swollmantcp_gettaocache(inp) 5786283Swollman struct inpcb *inp; 5796283Swollman{ 5806283Swollman struct rtentry *rt = tcp_rtlookup(inp); 5816283Swollman 5826283Swollman /* Make sure this is a host route and is up. */ 5836283Swollman if (rt == NULL || 5846283Swollman (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 5856283Swollman return NULL; 5866283Swollman 5876283Swollman return rmx_taop(rt->rt_rmx); 5886283Swollman} 5896283Swollman 5906283Swollman/* 5916283Swollman * Clear all the TAO cache entries, called from tcp_init. 5926283Swollman * 5936283Swollman * XXX 5946283Swollman * This routine is just an empty one, because we assume that the routing 5956283Swollman * routing tables are initialized at the same time when TCP, so there is 5966283Swollman * nothing in the cache left over. 5976283Swollman */ 5986283Swollmanstatic void 5996283Swollmantcp_cleartaocache(void) 6006283Swollman{ } 601