tcp_timewait.c revision 6283
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_subr.c 8.1 (Berkeley) 6/10/93 34 * $Id: tcp_subr.c,v 1.5 1994/10/08 22:39:58 phk Exp $ 35 */ 36 37#include <sys/param.h> 38#include <sys/proc.h> 39#include <sys/systm.h> 40#include <sys/malloc.h> 41#include <sys/mbuf.h> 42#include <sys/socket.h> 43#include <sys/socketvar.h> 44#include <sys/protosw.h> 45#include <sys/errno.h> 46 47#include <net/route.h> 48#include <net/if.h> 49 50#include <netinet/in.h> 51#include <netinet/in_systm.h> 52#include <netinet/ip.h> 53#include <netinet/in_pcb.h> 54#include <netinet/ip_var.h> 55#include <netinet/ip_icmp.h> 56#include <netinet/tcp.h> 57#define TCPOUTFLAGS 58#include <netinet/tcp_fsm.h> 59#include <netinet/tcp_seq.h> 60#include <netinet/tcp_timer.h> 61#include <netinet/tcp_var.h> 62#include <netinet/tcpip.h> 63#ifdef TCPDEBUG 64#include <netinet/tcp_debug.h> 65#endif 66 67/* patchable/settable parameters for tcp */ 68int tcp_mssdflt = TCP_MSS; 69int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 70int tcp_do_rfc1323 = 1; 71#ifdef TTCP 72int tcp_do_rfc1644 = 1; 73static void tcp_cleartaocache(void); 74#endif 75 76extern struct inpcb *tcp_last_inpcb; 77 78/* 79 * Tcp initialization 80 */ 81void 82tcp_init() 83{ 84 85 tcp_iss = 1; /* wrong */ 86#ifdef TTCP 87 tcp_ccgen = 1; 88 tcp_cleartaocache(); 89#endif 90 tcb.inp_next = tcb.inp_prev = &tcb; 91 if (max_protohdr < sizeof(struct tcpiphdr)) 92 max_protohdr = sizeof(struct tcpiphdr); 93 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 94 panic("tcp_init"); 95} 96 97/* 98 * Create template to be used to send tcp packets on a connection. 99 * Call after host entry created, allocates an mbuf and fills 100 * in a skeletal tcp/ip header, minimizing the amount of work 101 * necessary when the connection is used. 102 */ 103struct tcpiphdr * 104tcp_template(tp) 105 struct tcpcb *tp; 106{ 107 register struct inpcb *inp = tp->t_inpcb; 108 register struct mbuf *m; 109 register struct tcpiphdr *n; 110 111 if ((n = tp->t_template) == 0) { 112 m = m_get(M_DONTWAIT, MT_HEADER); 113 if (m == NULL) 114 return (0); 115 m->m_len = sizeof (struct tcpiphdr); 116 n = mtod(m, struct tcpiphdr *); 117 } 118 n->ti_next = n->ti_prev = 0; 119 n->ti_x1 = 0; 120 n->ti_pr = IPPROTO_TCP; 121 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 122 n->ti_src = inp->inp_laddr; 123 n->ti_dst = inp->inp_faddr; 124 n->ti_sport = inp->inp_lport; 125 n->ti_dport = inp->inp_fport; 126 n->ti_seq = 0; 127 n->ti_ack = 0; 128 n->ti_x2 = 0; 129 n->ti_off = 5; 130 n->ti_flags = 0; 131 n->ti_win = 0; 132 n->ti_sum = 0; 133 n->ti_urp = 0; 134 return (n); 135} 136 137/* 138 * Send a single message to the TCP at address specified by 139 * the given TCP/IP header. If m == 0, then we make a copy 140 * of the tcpiphdr at ti and send directly to the addressed host. 141 * This is used to force keep alive messages out using the TCP 142 * template for a connection tp->t_template. If flags are given 143 * then we send a message back to the TCP which originated the 144 * segment ti, and discard the mbuf containing it and any other 145 * attached mbufs. 146 * 147 * In any case the ack and sequence number of the transmitted 148 * segment are as specified by the parameters. 149 */ 150void 151tcp_respond(tp, ti, m, ack, seq, flags) 152 struct tcpcb *tp; 153 register struct tcpiphdr *ti; 154 register struct mbuf *m; 155 tcp_seq ack, seq; 156 int flags; 157{ 158 register int tlen; 159 int win = 0; 160 struct route *ro = 0; 161 162 if (tp) { 163 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 164 ro = &tp->t_inpcb->inp_route; 165 } 166 if (m == 0) { 167 m = m_gethdr(M_DONTWAIT, MT_HEADER); 168 if (m == NULL) 169 return; 170#ifdef TCP_COMPAT_42 171 tlen = 1; 172#else 173 tlen = 0; 174#endif 175 m->m_data += max_linkhdr; 176 *mtod(m, struct tcpiphdr *) = *ti; 177 ti = mtod(m, struct tcpiphdr *); 178 flags = TH_ACK; 179 } else { 180 m_freem(m->m_next); 181 m->m_next = 0; 182 m->m_data = (caddr_t)ti; 183 m->m_len = sizeof (struct tcpiphdr); 184 tlen = 0; 185#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 186 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long); 187 xchg(ti->ti_dport, ti->ti_sport, u_short); 188#undef xchg 189 } 190 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); 191 tlen += sizeof (struct tcpiphdr); 192 m->m_len = tlen; 193 m->m_pkthdr.len = tlen; 194 m->m_pkthdr.rcvif = (struct ifnet *) 0; 195 ti->ti_next = ti->ti_prev = 0; 196 ti->ti_x1 = 0; 197 ti->ti_seq = htonl(seq); 198 ti->ti_ack = htonl(ack); 199 ti->ti_x2 = 0; 200 ti->ti_off = sizeof (struct tcphdr) >> 2; 201 ti->ti_flags = flags; 202 if (tp) 203 ti->ti_win = htons((u_short) (win >> tp->rcv_scale)); 204 else 205 ti->ti_win = htons((u_short)win); 206 ti->ti_urp = 0; 207 ti->ti_sum = 0; 208 ti->ti_sum = in_cksum(m, tlen); 209 ((struct ip *)ti)->ip_len = tlen; 210 ((struct ip *)ti)->ip_ttl = ip_defttl; 211#ifdef TCPDEBUG 212 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 213 tcp_trace(TA_OUTPUT, 0, tp, ti, 0); 214#endif 215 (void) ip_output(m, NULL, ro, 0, NULL); 216} 217 218/* 219 * Create a new TCP control block, making an 220 * empty reassembly queue and hooking it to the argument 221 * protocol control block. 222 */ 223struct tcpcb * 224tcp_newtcpcb(inp) 225 struct inpcb *inp; 226{ 227 register struct tcpcb *tp; 228 229 tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT); 230 if (tp == NULL) 231 return ((struct tcpcb *)0); 232 bzero((char *) tp, sizeof(struct tcpcb)); 233 tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp; 234 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt; 235 236 if (tcp_do_rfc1323) 237 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 238#ifdef TTCP 239 if (tcp_do_rfc1644) 240 tp->t_flags |= TF_REQ_CC; 241#endif 242 tp->t_inpcb = inp; 243 /* 244 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 245 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 246 * reasonable initial retransmit time. 247 */ 248 tp->t_srtt = TCPTV_SRTTBASE; 249 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2; 250 tp->t_rttmin = TCPTV_MIN; 251 TCPT_RANGESET(tp->t_rxtcur, 252 ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1, 253 TCPTV_MIN, TCPTV_REXMTMAX); 254 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 255 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 256 inp->inp_ip.ip_ttl = ip_defttl; 257 inp->inp_ppcb = (caddr_t)tp; 258 return (tp); 259} 260 261/* 262 * Drop a TCP connection, reporting 263 * the specified error. If connection is synchronized, 264 * then send a RST to peer. 265 */ 266struct tcpcb * 267tcp_drop(tp, errno) 268 register struct tcpcb *tp; 269 int errno; 270{ 271 struct socket *so = tp->t_inpcb->inp_socket; 272 273 if (TCPS_HAVERCVDSYN(tp->t_state)) { 274 tp->t_state = TCPS_CLOSED; 275 (void) tcp_output(tp); 276 tcpstat.tcps_drops++; 277 } else 278 tcpstat.tcps_conndrops++; 279 if (errno == ETIMEDOUT && tp->t_softerror) 280 errno = tp->t_softerror; 281 so->so_error = errno; 282 return (tcp_close(tp)); 283} 284 285/* 286 * Close a TCP control block: 287 * discard all space held by the tcp 288 * discard internet protocol block 289 * wake up any sleepers 290 */ 291struct tcpcb * 292tcp_close(tp) 293 register struct tcpcb *tp; 294{ 295 register struct tcpiphdr *t; 296 struct inpcb *inp = tp->t_inpcb; 297 struct socket *so = inp->inp_socket; 298 register struct mbuf *m; 299#ifdef RTV_RTT 300 register struct rtentry *rt; 301 302 /* 303 * If we sent enough data to get some meaningful characteristics, 304 * save them in the routing entry. 'Enough' is arbitrarily 305 * defined as the sendpipesize (default 4K) * 16. This would 306 * give us 16 rtt samples assuming we only get one sample per 307 * window (the usual case on a long haul net). 16 samples is 308 * enough for the srtt filter to converge to within 5% of the correct 309 * value; fewer samples and we could save a very bogus rtt. 310 * 311 * Don't update the default route's characteristics and don't 312 * update anything that the user "locked". 313 */ 314 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 315 (rt = inp->inp_route.ro_rt) && 316 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) { 317 register u_long i = 0; 318 319 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 320 i = tp->t_srtt * 321 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 322 if (rt->rt_rmx.rmx_rtt && i) 323 /* 324 * filter this update to half the old & half 325 * the new values, converting scale. 326 * See route.h and tcp_var.h for a 327 * description of the scaling constants. 328 */ 329 rt->rt_rmx.rmx_rtt = 330 (rt->rt_rmx.rmx_rtt + i) / 2; 331 else 332 rt->rt_rmx.rmx_rtt = i; 333 } 334 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 335 i = tp->t_rttvar * 336 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 337 if (rt->rt_rmx.rmx_rttvar && i) 338 rt->rt_rmx.rmx_rttvar = 339 (rt->rt_rmx.rmx_rttvar + i) / 2; 340 else 341 rt->rt_rmx.rmx_rttvar = i; 342 } 343 /* 344 * update the pipelimit (ssthresh) if it has been updated 345 * already or if a pipesize was specified & the threshhold 346 * got below half the pipesize. I.e., wait for bad news 347 * before we start updating, then update on both good 348 * and bad news. 349 */ 350 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 351 ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) || 352 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 353 /* 354 * convert the limit from user data bytes to 355 * packets then to packet data bytes. 356 */ 357 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 358 if (i < 2) 359 i = 2; 360 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr)); 361 if (rt->rt_rmx.rmx_ssthresh) 362 rt->rt_rmx.rmx_ssthresh = 363 (rt->rt_rmx.rmx_ssthresh + i) / 2; 364 else 365 rt->rt_rmx.rmx_ssthresh = i; 366 } 367 } 368#endif /* RTV_RTT */ 369 /* free the reassembly queue, if any */ 370 t = tp->seg_next; 371 while (t != (struct tcpiphdr *)tp) { 372 t = (struct tcpiphdr *)t->ti_next; 373 m = REASS_MBUF((struct tcpiphdr *)t->ti_prev); 374 remque(t->ti_prev); 375 m_freem(m); 376 } 377 if (tp->t_template) 378 (void) m_free(dtom(tp->t_template)); 379 free(tp, M_PCB); 380 inp->inp_ppcb = 0; 381 soisdisconnected(so); 382 /* clobber input pcb cache if we're closing the cached connection */ 383 if (inp == tcp_last_inpcb) 384 tcp_last_inpcb = &tcb; 385 in_pcbdetach(inp); 386 tcpstat.tcps_closed++; 387 return ((struct tcpcb *)0); 388} 389 390void 391tcp_drain() 392{ 393 394} 395 396/* 397 * Notify a tcp user of an asynchronous error; 398 * store error as soft error, but wake up user 399 * (for now, won't do anything until can select for soft error). 400 */ 401void 402tcp_notify(inp, error) 403 struct inpcb *inp; 404 int error; 405{ 406 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 407 register struct socket *so = inp->inp_socket; 408 409 /* 410 * Ignore some errors if we are hooked up. 411 * If connection hasn't completed, has retransmitted several times, 412 * and receives a second error, give up now. This is better 413 * than waiting a long time to establish a connection that 414 * can never complete. 415 */ 416 if (tp->t_state == TCPS_ESTABLISHED && 417 (error == EHOSTUNREACH || error == ENETUNREACH || 418 error == EHOSTDOWN)) { 419 return; 420 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 421 tp->t_softerror) 422 so->so_error = error; 423 else 424 tp->t_softerror = error; 425 wakeup((caddr_t) &so->so_timeo); 426 sorwakeup(so); 427 sowwakeup(so); 428} 429 430void 431tcp_ctlinput(cmd, sa, ip) 432 int cmd; 433 struct sockaddr *sa; 434 register struct ip *ip; 435{ 436 register struct tcphdr *th; 437 extern struct in_addr zeroin_addr; 438 extern u_char inetctlerrmap[]; 439 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 440 441 if (cmd == PRC_QUENCH) 442 notify = tcp_quench; 443 else if (!PRC_IS_REDIRECT(cmd) && 444 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)) 445 return; 446 if (ip) { 447 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 448 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport, 449 cmd, notify); 450 } else 451 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify); 452} 453 454/* 455 * When a source quench is received, close congestion window 456 * to one segment. We will gradually open it again as we proceed. 457 */ 458void 459tcp_quench(inp, errno) 460 struct inpcb *inp; 461 int errno; 462{ 463 struct tcpcb *tp = intotcpcb(inp); 464 465 if (tp) 466 tp->snd_cwnd = tp->t_maxseg; 467} 468 469/* 470 * Look-up the routing entry to the peer of this inpcb. If no route 471 * is found and it cannot be allocated the return NULL. This routine 472 * is called by TCP routines that access the rmx structure and by tcp_mss 473 * to get the interface MTU. 474 */ 475struct rtentry * 476tcp_rtlookup(inp) 477 struct inpcb *inp; 478{ 479 struct route *ro; 480 struct rtentry *rt; 481 482 ro = &inp->inp_route; 483 rt = ro->ro_rt; 484 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 485 /* No route yet, so try to acquire one */ 486 if (inp->inp_faddr.s_addr != INADDR_ANY) { 487 ro->ro_dst.sa_family = AF_INET; 488 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 489 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 490 inp->inp_faddr; 491 rtalloc(ro); 492 rt = ro->ro_rt; 493 } 494 } 495 return rt; 496} 497 498#ifdef TTCP 499/* 500 * Return a pointer to the cached information about the remote host. 501 * The cached information is stored in the protocol specific part of 502 * the route metrics. 503 */ 504struct rmxp_tao * 505tcp_gettaocache(inp) 506 struct inpcb *inp; 507{ 508 struct rtentry *rt = tcp_rtlookup(inp); 509 510 /* Make sure this is a host route and is up. */ 511 if (rt == NULL || 512 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 513 return NULL; 514 515 return rmx_taop(rt->rt_rmx); 516} 517 518/* 519 * Clear all the TAO cache entries, called from tcp_init. 520 * 521 * XXX 522 * This routine is just an empty one, because we assume that the routing 523 * routing tables are initialized at the same time when TCP, so there is 524 * nothing in the cache left over. 525 */ 526static void 527tcp_cleartaocache(void) 528{ } 529#endif /* TTCP */ 530