1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_subr.c 7.19 (Berkeley) 07/25/90 8 */ 9 10 #include "param.h" 11 #include "systm.h" 12 #include "malloc.h" 13 #include "mbuf.h" 14 #include "socket.h" 15 #include "socketvar.h" 16 #include "protosw.h" 17 #include "errno.h" 18 19 #include "../net/route.h" 20 #include "../net/if.h" 21 22 #include "in.h" 23 #include "in_systm.h" 24 #include "ip.h" 25 #include "in_pcb.h" 26 #include "ip_var.h" 27 #include "ip_icmp.h" 28 #include "tcp.h" 29 #include "tcp_fsm.h" 30 #include "tcp_seq.h" 31 #include "tcp_timer.h" 32 #include "tcp_var.h" 33 #include "tcpip.h" 34 35 /* patchable/settable parameters for tcp */ 36 int tcp_ttl = TCP_TTL; 37 int tcp_mssdflt = TCP_MSS; 38 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 39 40 extern struct inpcb *tcp_last_inpcb; 41 42 /* 43 * Tcp initialization 44 */ 45 tcp_init() 46 { 47 48 tcp_iss = 1; /* wrong */ 49 tcb.inp_next = tcb.inp_prev = &tcb; 50 if (max_protohdr < sizeof(struct tcpiphdr)) 51 max_protohdr = sizeof(struct tcpiphdr); 52 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 53 panic("tcp_init"); 54 } 55 56 /* 57 * Create template to be used to send tcp packets on a connection. 58 * Call after host entry created, allocates an mbuf and fills 59 * in a skeletal tcp/ip header, minimizing the amount of work 60 * necessary when the connection is used. 61 */ 62 struct tcpiphdr * 63 tcp_template(tp) 64 struct tcpcb *tp; 65 { 66 register struct inpcb *inp = tp->t_inpcb; 67 register struct mbuf *m; 68 register struct tcpiphdr *n; 69 70 if ((n = tp->t_template) == 0) { 71 m = m_get(M_DONTWAIT, MT_HEADER); 72 if (m == NULL) 73 return (0); 74 m->m_len = sizeof (struct tcpiphdr); 75 n = mtod(m, struct tcpiphdr *); 76 } 77 n->ti_next = n->ti_prev = 0; 78 n->ti_x1 = 0; 79 n->ti_pr = IPPROTO_TCP; 80 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 81 n->ti_src = inp->inp_laddr; 82 n->ti_dst = inp->inp_faddr; 83 n->ti_sport = inp->inp_lport; 84 n->ti_dport = inp->inp_fport; 85 n->ti_seq = 0; 86 n->ti_ack = 0; 87 n->ti_x2 = 0; 88 n->ti_off = 5; 89 n->ti_flags = 0; 90 n->ti_win = 0; 91 n->ti_sum = 0; 92 n->ti_urp = 0; 93 return (n); 94 } 95 96 /* 97 * Send a single message to the TCP at address specified by 98 * the given TCP/IP header. If m == 0, then we make a copy 99 * of the tcpiphdr at ti and send directly to the addressed host. 100 * This is used to force keep alive messages out using the TCP 101 * template for a connection tp->t_template. If flags are given 102 * then we send a message back to the TCP which originated the 103 * segment ti, and discard the mbuf containing it and any other 104 * attached mbufs. 105 * 106 * In any case the ack and sequence number of the transmitted 107 * segment are as specified by the parameters. 108 */ 109 tcp_respond(tp, ti, m, ack, seq, flags) 110 struct tcpcb *tp; 111 register struct tcpiphdr *ti; 112 register struct mbuf *m; 113 tcp_seq ack, seq; 114 int flags; 115 { 116 register int tlen; 117 int win = 0; 118 struct route *ro = 0; 119 120 if (tp) { 121 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 122 ro = &tp->t_inpcb->inp_route; 123 } 124 if (m == 0) { 125 m = m_gethdr(M_DONTWAIT, MT_HEADER); 126 if (m == NULL) 127 return; 128 #ifdef TCP_COMPAT_42 129 tlen = 1; 130 #else 131 tlen = 0; 132 #endif 133 m->m_data += max_linkhdr; 134 *mtod(m, struct tcpiphdr *) = *ti; 135 ti = mtod(m, struct tcpiphdr *); 136 flags = TH_ACK; 137 } else { 138 m_freem(m->m_next); 139 m->m_next = 0; 140 m->m_data = (caddr_t)ti; 141 m->m_len = sizeof (struct tcpiphdr); 142 tlen = 0; 143 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 144 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long); 145 xchg(ti->ti_dport, ti->ti_sport, u_short); 146 #undef xchg 147 } 148 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); 149 tlen += sizeof (struct tcpiphdr); 150 m->m_len = tlen; 151 m->m_pkthdr.len = tlen; 152 m->m_pkthdr.rcvif = (struct ifnet *) 0; 153 ti->ti_next = ti->ti_prev = 0; 154 ti->ti_x1 = 0; 155 ti->ti_seq = htonl(seq); 156 ti->ti_ack = htonl(ack); 157 ti->ti_x2 = 0; 158 ti->ti_off = sizeof (struct tcphdr) >> 2; 159 ti->ti_flags = flags; 160 ti->ti_win = htons((u_short)win); 161 ti->ti_urp = 0; 162 ti->ti_sum = in_cksum(m, tlen); 163 ((struct ip *)ti)->ip_len = tlen; 164 ((struct ip *)ti)->ip_ttl = tcp_ttl; 165 (void) ip_output(m, (struct mbuf *)0, ro, 0); 166 } 167 168 /* 169 * Create a new TCP control block, making an 170 * empty reassembly queue and hooking it to the argument 171 * protocol control block. 172 */ 173 struct tcpcb * 174 tcp_newtcpcb(inp) 175 struct inpcb *inp; 176 { 177 struct mbuf *m = m_getclr(M_DONTWAIT, MT_PCB); 178 register struct tcpcb *tp; 179 180 if (m == NULL) 181 return ((struct tcpcb *)0); 182 tp = mtod(m, struct tcpcb *); 183 tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp; 184 tp->t_maxseg = tcp_mssdflt; 185 186 tp->t_flags = 0; /* sends options! */ 187 tp->t_inpcb = inp; 188 /* 189 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 190 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 191 * reasonable initial retransmit time. 192 */ 193 tp->t_srtt = TCPTV_SRTTBASE; 194 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2; 195 tp->t_rttmin = TCPTV_MIN; 196 TCPT_RANGESET(tp->t_rxtcur, 197 ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1, 198 TCPTV_MIN, TCPTV_REXMTMAX); 199 tp->snd_cwnd = TCP_MAXWIN; 200 tp->snd_ssthresh = TCP_MAXWIN; 201 inp->inp_ip.ip_ttl = tcp_ttl; 202 inp->inp_ppcb = (caddr_t)tp; 203 return (tp); 204 } 205 206 /* 207 * Drop a TCP connection, reporting 208 * the specified error. If connection is synchronized, 209 * then send a RST to peer. 210 */ 211 struct tcpcb * 212 tcp_drop(tp, errno) 213 register struct tcpcb *tp; 214 int errno; 215 { 216 struct socket *so = tp->t_inpcb->inp_socket; 217 218 if (TCPS_HAVERCVDSYN(tp->t_state)) { 219 tp->t_state = TCPS_CLOSED; 220 (void) tcp_output(tp); 221 tcpstat.tcps_drops++; 222 } else 223 tcpstat.tcps_conndrops++; 224 if (errno == ETIMEDOUT && tp->t_softerror) 225 errno = tp->t_softerror; 226 so->so_error = errno; 227 return (tcp_close(tp)); 228 } 229 230 /* 231 * Close a TCP control block: 232 * discard all space held by the tcp 233 * discard internet protocol block 234 * wake up any sleepers 235 */ 236 struct tcpcb * 237 tcp_close(tp) 238 register struct tcpcb *tp; 239 { 240 register struct tcpiphdr *t; 241 struct inpcb *inp = tp->t_inpcb; 242 struct socket *so = inp->inp_socket; 243 register struct mbuf *m; 244 #ifdef RTV_RTT 245 register struct rtentry *rt; 246 247 /* 248 * If we sent enough data to get some meaningful characteristics, 249 * save them in the routing entry. 'Enough' is arbitrarily 250 * defined as 4K (default tcp_sendspace) * 16. This would 251 * give us 16 rtt samples assuming we only get one sample per 252 * window (the usual case on a long haul net). 16 samples is 253 * enough for the srtt filter to converge to within 5% of the correct 254 * value; fewer samples and we could save a very bogus rtt. 255 * 256 * Don't update the default route's characteristics and don't 257 * update anything that the user "locked". 258 */ 259 if (SEQ_LT(tp->iss+(4096*16), tp->snd_max) && 260 (rt = inp->inp_route.ro_rt) && 261 ((struct sockaddr_in *) rt_key(rt))->sin_addr.s_addr != 262 INADDR_ANY) { 263 register u_long i; 264 265 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 266 i = tp->t_srtt * 267 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 268 if (rt->rt_rmx.rmx_rtt && i) 269 /* 270 * filter this update to half the old & half 271 * the new values, converting scale. 272 * See route.h and tcp_var.h for a 273 * description of the scaling constants. 274 */ 275 rt->rt_rmx.rmx_rtt = 276 (rt->rt_rmx.rmx_rtt + i) / 2; 277 else 278 rt->rt_rmx.rmx_rtt = i; 279 } 280 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 281 i = tp->t_rttvar * 282 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 283 if (rt->rt_rmx.rmx_rttvar && i) 284 rt->rt_rmx.rmx_rttvar = 285 (rt->rt_rmx.rmx_rttvar + i) / 2; 286 else 287 rt->rt_rmx.rmx_rttvar = i; 288 } 289 /* 290 * update the pipelimit (ssthresh) if it has been updated 291 * already or if a pipesize was specified & the threshhold 292 * got below half the pipesize. I.e., wait for bad news 293 * before we start updating, then update on both good 294 * and bad news. 295 */ 296 if ((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 297 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh || 298 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 299 /* 300 * convert the limit from user data bytes to 301 * packets then to packet data bytes. 302 */ 303 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 304 if (i < 2) 305 i = 2; 306 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr)); 307 if (rt->rt_rmx.rmx_ssthresh) 308 rt->rt_rmx.rmx_ssthresh = 309 (rt->rt_rmx.rmx_ssthresh + i) / 2; 310 else 311 rt->rt_rmx.rmx_ssthresh = i; 312 } 313 } 314 #endif RTV_RTT 315 /* free the reassembly queue, if any */ 316 t = tp->seg_next; 317 while (t != (struct tcpiphdr *)tp) { 318 t = (struct tcpiphdr *)t->ti_next; 319 m = REASS_MBUF((struct tcpiphdr *)t->ti_prev); 320 remque(t->ti_prev); 321 m_freem(m); 322 } 323 if (tp->t_template) 324 (void) m_free(dtom(tp->t_template)); 325 (void) m_free(dtom(tp)); 326 inp->inp_ppcb = 0; 327 soisdisconnected(so); 328 /* clobber input pcb cache if we're closing the cached connection */ 329 if (inp == tcp_last_inpcb) 330 tcp_last_inpcb = &tcb; 331 in_pcbdetach(inp); 332 tcpstat.tcps_closed++; 333 return ((struct tcpcb *)0); 334 } 335 336 tcp_drain() 337 { 338 339 } 340 341 /* 342 * Notify a tcp user of an asynchronous error; 343 * store error as soft error, but wake up user 344 * (for now, won't do anything until can select for soft error). 345 */ 346 tcp_notify(inp, error) 347 register struct inpcb *inp; 348 int error; 349 { 350 351 ((struct tcpcb *)inp->inp_ppcb)->t_softerror = error; 352 wakeup((caddr_t) &inp->inp_socket->so_timeo); 353 sorwakeup(inp->inp_socket); 354 sowwakeup(inp->inp_socket); 355 } 356 357 tcp_ctlinput(cmd, sa, ip) 358 int cmd; 359 struct sockaddr *sa; 360 register struct ip *ip; 361 { 362 register struct tcphdr *th; 363 extern struct in_addr zeroin_addr; 364 extern u_char inetctlerrmap[]; 365 int (*notify)() = tcp_notify, tcp_quench(); 366 367 if (cmd == PRC_QUENCH) 368 notify = tcp_quench; 369 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 370 return; 371 if (ip) { 372 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 373 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport, 374 cmd, notify); 375 } else 376 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify); 377 } 378 379 /* 380 * When a source quench is received, close congestion window 381 * to one segment. We will gradually open it again as we proceed. 382 */ 383 tcp_quench(inp) 384 struct inpcb *inp; 385 { 386 struct tcpcb *tp = intotcpcb(inp); 387 388 if (tp) 389 tp->snd_cwnd = tp->t_maxseg; 390 } 391