| 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 7.21 (Berkeley) %G% |
| 8 | */ |
| 9 | |
| 10 | #include "param.h" |
| 11 | #include "proc.h" |
| 12 | #include "systm.h" |
| 13 | #include "malloc.h" |
| 14 | #include "mbuf.h" |
| 15 | #include "socket.h" |
| 16 | #include "socketvar.h" |
| 17 | #include "protosw.h" |
| 18 | #include "errno.h" |
| 19 | |
| 20 | #include "../net/route.h" |
| 21 | #include "../net/if.h" |
| 22 | |
| 23 | #include "in.h" |
| 24 | #include "in_systm.h" |
| 25 | #include "ip.h" |
| 26 | #include "in_pcb.h" |
| 27 | #include "ip_var.h" |
| 28 | #include "ip_icmp.h" |
| 29 | #include "tcp.h" |
| 30 | #include "tcp_fsm.h" |
| 31 | #include "tcp_seq.h" |
| 32 | #include "tcp_timer.h" |
| 33 | #include "tcp_var.h" |
| 34 | #include "tcpip.h" |
| 35 | |
| 36 | /* patchable/settable parameters for tcp */ |
| 37 | int tcp_ttl = TCP_TTL; |
| 38 | int tcp_mssdflt = TCP_MSS; |
| 39 | int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; |
| 40 | |
| 41 | extern struct inpcb *tcp_last_inpcb; |
| 42 | |
| 43 | /* |
| 44 | * Tcp initialization |
| 45 | */ |
| 46 | tcp_init() |
| 47 | { |
| 48 | |
| 49 | tcp_iss = 1; /* wrong */ |
| 50 | tcb.inp_next = tcb.inp_prev = &tcb; |
| 51 | if (max_protohdr < sizeof(struct tcpiphdr)) |
| 52 | max_protohdr = sizeof(struct tcpiphdr); |
| 53 | if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) |
| 54 | panic("tcp_init"); |
| 55 | } |
| 56 | |
| 57 | /* |
| 58 | * Create template to be used to send tcp packets on a connection. |
| 59 | * Call after host entry created, allocates an mbuf and fills |
| 60 | * in a skeletal tcp/ip header, minimizing the amount of work |
| 61 | * necessary when the connection is used. |
| 62 | */ |
| 63 | struct tcpiphdr * |
| 64 | tcp_template(tp) |
| 65 | struct tcpcb *tp; |
| 66 | { |
| 67 | register struct inpcb *inp = tp->t_inpcb; |
| 68 | register struct mbuf *m; |
| 69 | register struct tcpiphdr *n; |
| 70 | |
| 71 | if ((n = tp->t_template) == 0) { |
| 72 | m = m_get(M_DONTWAIT, MT_HEADER); |
| 73 | if (m == NULL) |
| 74 | return (0); |
| 75 | m->m_len = sizeof (struct tcpiphdr); |
| 76 | n = mtod(m, struct tcpiphdr *); |
| 77 | } |
| 78 | n->ti_next = n->ti_prev = 0; |
| 79 | n->ti_x1 = 0; |
| 80 | n->ti_pr = IPPROTO_TCP; |
| 81 | n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); |
| 82 | n->ti_src = inp->inp_laddr; |
| 83 | n->ti_dst = inp->inp_faddr; |
| 84 | n->ti_sport = inp->inp_lport; |
| 85 | n->ti_dport = inp->inp_fport; |
| 86 | n->ti_seq = 0; |
| 87 | n->ti_ack = 0; |
| 88 | n->ti_x2 = 0; |
| 89 | n->ti_off = 5; |
| 90 | n->ti_flags = 0; |
| 91 | n->ti_win = 0; |
| 92 | n->ti_sum = 0; |
| 93 | n->ti_urp = 0; |
| 94 | return (n); |
| 95 | } |
| 96 | |
| 97 | /* |
| 98 | * Send a single message to the TCP at address specified by |
| 99 | * the given TCP/IP header. If m == 0, then we make a copy |
| 100 | * of the tcpiphdr at ti and send directly to the addressed host. |
| 101 | * This is used to force keep alive messages out using the TCP |
| 102 | * template for a connection tp->t_template. If flags are given |
| 103 | * then we send a message back to the TCP which originated the |
| 104 | * segment ti, and discard the mbuf containing it and any other |
| 105 | * attached mbufs. |
| 106 | * |
| 107 | * In any case the ack and sequence number of the transmitted |
| 108 | * segment are as specified by the parameters. |
| 109 | */ |
| 110 | tcp_respond(tp, ti, m, ack, seq, flags) |
| 111 | struct tcpcb *tp; |
| 112 | register struct tcpiphdr *ti; |
| 113 | register struct mbuf *m; |
| 114 | tcp_seq ack, seq; |
| 115 | int flags; |
| 116 | { |
| 117 | register int tlen; |
| 118 | int win = 0; |
| 119 | struct route *ro = 0; |
| 120 | |
| 121 | if (tp) { |
| 122 | win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); |
| 123 | ro = &tp->t_inpcb->inp_route; |
| 124 | } |
| 125 | if (m == 0) { |
| 126 | m = m_gethdr(M_DONTWAIT, MT_HEADER); |
| 127 | if (m == NULL) |
| 128 | return; |
| 129 | #ifdef TCP_COMPAT_42 |
| 130 | tlen = 1; |
| 131 | #else |
| 132 | tlen = 0; |
| 133 | #endif |
| 134 | m->m_data += max_linkhdr; |
| 135 | *mtod(m, struct tcpiphdr *) = *ti; |
| 136 | ti = mtod(m, struct tcpiphdr *); |
| 137 | flags = TH_ACK; |
| 138 | } else { |
| 139 | m_freem(m->m_next); |
| 140 | m->m_next = 0; |
| 141 | m->m_data = (caddr_t)ti; |
| 142 | m->m_len = sizeof (struct tcpiphdr); |
| 143 | tlen = 0; |
| 144 | #define xchg(a,b,type) { type t; t=a; a=b; b=t; } |
| 145 | xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long); |
| 146 | xchg(ti->ti_dport, ti->ti_sport, u_short); |
| 147 | #undef xchg |
| 148 | } |
| 149 | ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); |
| 150 | tlen += sizeof (struct tcpiphdr); |
| 151 | m->m_len = tlen; |
| 152 | m->m_pkthdr.len = tlen; |
| 153 | m->m_pkthdr.rcvif = (struct ifnet *) 0; |
| 154 | ti->ti_next = ti->ti_prev = 0; |
| 155 | ti->ti_x1 = 0; |
| 156 | ti->ti_seq = htonl(seq); |
| 157 | ti->ti_ack = htonl(ack); |
| 158 | ti->ti_x2 = 0; |
| 159 | ti->ti_off = sizeof (struct tcphdr) >> 2; |
| 160 | ti->ti_flags = flags; |
| 161 | ti->ti_win = htons((u_short)win); |
| 162 | ti->ti_urp = 0; |
| 163 | ti->ti_sum = in_cksum(m, tlen); |
| 164 | ((struct ip *)ti)->ip_len = tlen; |
| 165 | ((struct ip *)ti)->ip_ttl = tcp_ttl; |
| 166 | (void) ip_output(m, (struct mbuf *)0, ro, 0); |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | * Create a new TCP control block, making an |
| 171 | * empty reassembly queue and hooking it to the argument |
| 172 | * protocol control block. |
| 173 | */ |
| 174 | struct tcpcb * |
| 175 | tcp_newtcpcb(inp) |
| 176 | struct inpcb *inp; |
| 177 | { |
| 178 | struct mbuf *m = m_getclr(M_DONTWAIT, MT_PCB); |
| 179 | register struct tcpcb *tp; |
| 180 | |
| 181 | if (m == NULL) |
| 182 | return ((struct tcpcb *)0); |
| 183 | tp = mtod(m, struct tcpcb *); |
| 184 | tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp; |
| 185 | tp->t_maxseg = tcp_mssdflt; |
| 186 | |
| 187 | tp->t_flags = 0; /* sends options! */ |
| 188 | tp->t_inpcb = inp; |
| 189 | /* |
| 190 | * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no |
| 191 | * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives |
| 192 | * reasonable initial retransmit time. |
| 193 | */ |
| 194 | tp->t_srtt = TCPTV_SRTTBASE; |
| 195 | tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2; |
| 196 | tp->t_rttmin = TCPTV_MIN; |
| 197 | TCPT_RANGESET(tp->t_rxtcur, |
| 198 | ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1, |
| 199 | TCPTV_MIN, TCPTV_REXMTMAX); |
| 200 | tp->snd_cwnd = TCP_MAXWIN; |
| 201 | tp->snd_ssthresh = TCP_MAXWIN; |
| 202 | inp->inp_ip.ip_ttl = tcp_ttl; |
| 203 | inp->inp_ppcb = (caddr_t)tp; |
| 204 | return (tp); |
| 205 | } |
| 206 | |
| 207 | /* |
| 208 | * Drop a TCP connection, reporting |
| 209 | * the specified error. If connection is synchronized, |
| 210 | * then send a RST to peer. |
| 211 | */ |
| 212 | struct tcpcb * |
| 213 | tcp_drop(tp, errno) |
| 214 | register struct tcpcb *tp; |
| 215 | int errno; |
| 216 | { |
| 217 | struct socket *so = tp->t_inpcb->inp_socket; |
| 218 | |
| 219 | if (TCPS_HAVERCVDSYN(tp->t_state)) { |
| 220 | tp->t_state = TCPS_CLOSED; |
| 221 | (void) tcp_output(tp); |
| 222 | tcpstat.tcps_drops++; |
| 223 | } else |
| 224 | tcpstat.tcps_conndrops++; |
| 225 | if (errno == ETIMEDOUT && tp->t_softerror) |
| 226 | errno = tp->t_softerror; |
| 227 | so->so_error = errno; |
| 228 | return (tcp_close(tp)); |
| 229 | } |
| 230 | |
| 231 | /* |
| 232 | * Close a TCP control block: |
| 233 | * discard all space held by the tcp |
| 234 | * discard internet protocol block |
| 235 | * wake up any sleepers |
| 236 | */ |
| 237 | struct tcpcb * |
| 238 | tcp_close(tp) |
| 239 | register struct tcpcb *tp; |
| 240 | { |
| 241 | register struct tcpiphdr *t; |
| 242 | struct inpcb *inp = tp->t_inpcb; |
| 243 | struct socket *so = inp->inp_socket; |
| 244 | register struct mbuf *m; |
| 245 | #ifdef RTV_RTT |
| 246 | register struct rtentry *rt; |
| 247 | |
| 248 | /* |
| 249 | * If we sent enough data to get some meaningful characteristics, |
| 250 | * save them in the routing entry. 'Enough' is arbitrarily |
| 251 | * defined as the sendpipesize (default 4K) * 16. This would |
| 252 | * give us 16 rtt samples assuming we only get one sample per |
| 253 | * window (the usual case on a long haul net). 16 samples is |
| 254 | * enough for the srtt filter to converge to within 5% of the correct |
| 255 | * value; fewer samples and we could save a very bogus rtt. |
| 256 | * |
| 257 | * Don't update the default route's characteristics and don't |
| 258 | * update anything that the user "locked". |
| 259 | */ |
| 260 | if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && |
| 261 | (rt = inp->inp_route.ro_rt) && |
| 262 | ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != 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 | #if BSD<43 |
| 380 | /* XXX fake routine */ |
| 381 | tcp_abort(inp) |
| 382 | struct inpcb *inp; |
| 383 | { |
| 384 | return; |
| 385 | } |
| 386 | #endif |
| 387 | |
| 388 | /* |
| 389 | * When a source quench is received, close congestion window |
| 390 | * to one segment. We will gradually open it again as we proceed. |
| 391 | */ |
| 392 | tcp_quench(inp) |
| 393 | struct inpcb *inp; |
| 394 | { |
| 395 | struct tcpcb *tp = intotcpcb(inp); |
| 396 | |
| 397 | if (tp) |
| 398 | tp->snd_cwnd = tp->t_maxseg; |
| 399 | } |