tcp_timewait.c revision 9263
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 1.11 1995/05/30 08:09:58 rgrimes 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#include <sys/queue.h> 47 48#include <net/route.h> 49#include <net/if.h> 50 51#include <netinet/in.h> 52#include <netinet/in_systm.h> 53#include <netinet/ip.h> 54#include <netinet/in_pcb.h> 55#include <netinet/in_var.h> 56#include <netinet/ip_var.h> 57#include <netinet/ip_icmp.h> 58#include <netinet/tcp.h> 59#include <netinet/tcp_fsm.h> 60#include <netinet/tcp_seq.h> 61#include <netinet/tcp_timer.h> 62#include <netinet/tcp_var.h> 63#include <netinet/tcpip.h> 64#ifdef TCPDEBUG 65#include <netinet/tcp_debug.h> 66#endif 67 68/* patchable/settable parameters for tcp */ 69int tcp_mssdflt = TCP_MSS; 70int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 71int tcp_do_rfc1323 = 1; 72int tcp_do_rfc1644 = 1; 73static void tcp_cleartaocache(void); 74 75/* 76 * Target size of TCP PCB hash table. Will be rounded down to a prime 77 * number. 78 */ 79#ifndef TCBHASHSIZE 80#define TCBHASHSIZE 128 81#endif 82 83/* 84 * Tcp initialization 85 */ 86void 87tcp_init() 88{ 89 90 tcp_iss = 1; /* wrong */ 91 tcp_ccgen = 1; 92 tcp_cleartaocache(); 93 LIST_INIT(&tcb); 94 tcbinfo.listhead = &tcb; 95 tcbinfo.hashbase = phashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashsize); 96 if (max_protohdr < sizeof(struct tcpiphdr)) 97 max_protohdr = sizeof(struct tcpiphdr); 98 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 99 panic("tcp_init"); 100} 101 102/* 103 * Create template to be used to send tcp packets on a connection. 104 * Call after host entry created, allocates an mbuf and fills 105 * in a skeletal tcp/ip header, minimizing the amount of work 106 * necessary when the connection is used. 107 */ 108struct tcpiphdr * 109tcp_template(tp) 110 struct tcpcb *tp; 111{ 112 register struct inpcb *inp = tp->t_inpcb; 113 register struct mbuf *m; 114 register struct tcpiphdr *n; 115 116 if ((n = tp->t_template) == 0) { 117 m = m_get(M_DONTWAIT, MT_HEADER); 118 if (m == NULL) 119 return (0); 120 m->m_len = sizeof (struct tcpiphdr); 121 n = mtod(m, struct tcpiphdr *); 122 } 123 n->ti_next = n->ti_prev = 0; 124 n->ti_x1 = 0; 125 n->ti_pr = IPPROTO_TCP; 126 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 127 n->ti_src = inp->inp_laddr; 128 n->ti_dst = inp->inp_faddr; 129 n->ti_sport = inp->inp_lport; 130 n->ti_dport = inp->inp_fport; 131 n->ti_seq = 0; 132 n->ti_ack = 0; 133 n->ti_x2 = 0; 134 n->ti_off = 5; 135 n->ti_flags = 0; 136 n->ti_win = 0; 137 n->ti_sum = 0; 138 n->ti_urp = 0; 139 return (n); 140} 141 142/* 143 * Send a single message to the TCP at address specified by 144 * the given TCP/IP header. If m == 0, then we make a copy 145 * of the tcpiphdr at ti and send directly to the addressed host. 146 * This is used to force keep alive messages out using the TCP 147 * template for a connection tp->t_template. If flags are given 148 * then we send a message back to the TCP which originated the 149 * segment ti, and discard the mbuf containing it and any other 150 * attached mbufs. 151 * 152 * In any case the ack and sequence number of the transmitted 153 * segment are as specified by the parameters. 154 */ 155void 156tcp_respond(tp, ti, m, ack, seq, flags) 157 struct tcpcb *tp; 158 register struct tcpiphdr *ti; 159 register struct mbuf *m; 160 tcp_seq ack, seq; 161 int flags; 162{ 163 register int tlen; 164 int win = 0; 165 struct route *ro = 0; 166 167 if (tp) { 168 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 169 ro = &tp->t_inpcb->inp_route; 170 } 171 if (m == 0) { 172 m = m_gethdr(M_DONTWAIT, MT_HEADER); 173 if (m == NULL) 174 return; 175#ifdef TCP_COMPAT_42 176 tlen = 1; 177#else 178 tlen = 0; 179#endif 180 m->m_data += max_linkhdr; 181 *mtod(m, struct tcpiphdr *) = *ti; 182 ti = mtod(m, struct tcpiphdr *); 183 flags = TH_ACK; 184 } else { 185 m_freem(m->m_next); 186 m->m_next = 0; 187 m->m_data = (caddr_t)ti; 188 m->m_len = sizeof (struct tcpiphdr); 189 tlen = 0; 190#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 191 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long); 192 xchg(ti->ti_dport, ti->ti_sport, u_short); 193#undef xchg 194 } 195 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); 196 tlen += sizeof (struct tcpiphdr); 197 m->m_len = tlen; 198 m->m_pkthdr.len = tlen; 199 m->m_pkthdr.rcvif = (struct ifnet *) 0; 200 ti->ti_next = ti->ti_prev = 0; 201 ti->ti_x1 = 0; 202 ti->ti_seq = htonl(seq); 203 ti->ti_ack = htonl(ack); 204 ti->ti_x2 = 0; 205 ti->ti_off = sizeof (struct tcphdr) >> 2; 206 ti->ti_flags = flags; 207 if (tp) 208 ti->ti_win = htons((u_short) (win >> tp->rcv_scale)); 209 else 210 ti->ti_win = htons((u_short)win); 211 ti->ti_urp = 0; 212 ti->ti_sum = 0; 213 ti->ti_sum = in_cksum(m, tlen); 214 ((struct ip *)ti)->ip_len = tlen; 215 ((struct ip *)ti)->ip_ttl = ip_defttl; 216#ifdef TCPDEBUG 217 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 218 tcp_trace(TA_OUTPUT, 0, tp, ti, 0); 219#endif 220 (void) ip_output(m, NULL, ro, 0, NULL); 221} 222 223/* 224 * Create a new TCP control block, making an 225 * empty reassembly queue and hooking it to the argument 226 * protocol control block. 227 */ 228struct tcpcb * 229tcp_newtcpcb(inp) 230 struct inpcb *inp; 231{ 232 register struct tcpcb *tp; 233 234 tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT); 235 if (tp == NULL) 236 return ((struct tcpcb *)0); 237 bzero((char *) tp, sizeof(struct tcpcb)); 238 tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp; 239 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt; 240 241 if (tcp_do_rfc1323) 242 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 243 if (tcp_do_rfc1644) 244 tp->t_flags |= TF_REQ_CC; 245 tp->t_inpcb = inp; 246 /* 247 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 248 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 249 * reasonable initial retransmit time. 250 */ 251 tp->t_srtt = TCPTV_SRTTBASE; 252 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2; 253 tp->t_rttmin = TCPTV_MIN; 254 TCPT_RANGESET(tp->t_rxtcur, 255 ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1, 256 TCPTV_MIN, TCPTV_REXMTMAX); 257 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 258 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 259 inp->inp_ip.ip_ttl = ip_defttl; 260 inp->inp_ppcb = (caddr_t)tp; 261 return (tp); 262} 263 264/* 265 * Drop a TCP connection, reporting 266 * the specified error. If connection is synchronized, 267 * then send a RST to peer. 268 */ 269struct tcpcb * 270tcp_drop(tp, errno) 271 register struct tcpcb *tp; 272 int errno; 273{ 274 struct socket *so = tp->t_inpcb->inp_socket; 275 276 if (TCPS_HAVERCVDSYN(tp->t_state)) { 277 tp->t_state = TCPS_CLOSED; 278 (void) tcp_output(tp); 279 tcpstat.tcps_drops++; 280 } else 281 tcpstat.tcps_conndrops++; 282 if (errno == ETIMEDOUT && tp->t_softerror) 283 errno = tp->t_softerror; 284 so->so_error = errno; 285 return (tcp_close(tp)); 286} 287 288/* 289 * Close a TCP control block: 290 * discard all space held by the tcp 291 * discard internet protocol block 292 * wake up any sleepers 293 */ 294struct tcpcb * 295tcp_close(tp) 296 register struct tcpcb *tp; 297{ 298 register struct tcpiphdr *t; 299 struct inpcb *inp = tp->t_inpcb; 300 struct socket *so = inp->inp_socket; 301 register struct mbuf *m; 302#ifdef RTV_RTT 303 register struct rtentry *rt; 304 305 /* 306 * If we sent enough data to get some meaningful characteristics, 307 * save them in the routing entry. 'Enough' is arbitrarily 308 * defined as the sendpipesize (default 4K) * 16. This would 309 * give us 16 rtt samples assuming we only get one sample per 310 * window (the usual case on a long haul net). 16 samples is 311 * enough for the srtt filter to converge to within 5% of the correct 312 * value; fewer samples and we could save a very bogus rtt. 313 * 314 * Don't update the default route's characteristics and don't 315 * update anything that the user "locked". 316 */ 317 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 318 (rt = inp->inp_route.ro_rt) && 319 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) { 320 register u_long i = 0; 321 322 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 323 i = tp->t_srtt * 324 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 325 if (rt->rt_rmx.rmx_rtt && i) 326 /* 327 * filter this update to half the old & half 328 * the new values, converting scale. 329 * See route.h and tcp_var.h for a 330 * description of the scaling constants. 331 */ 332 rt->rt_rmx.rmx_rtt = 333 (rt->rt_rmx.rmx_rtt + i) / 2; 334 else 335 rt->rt_rmx.rmx_rtt = i; 336 tcpstat.tcps_cachedrtt++; 337 } 338 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 339 i = tp->t_rttvar * 340 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 341 if (rt->rt_rmx.rmx_rttvar && i) 342 rt->rt_rmx.rmx_rttvar = 343 (rt->rt_rmx.rmx_rttvar + i) / 2; 344 else 345 rt->rt_rmx.rmx_rttvar = i; 346 tcpstat.tcps_cachedrttvar++; 347 } 348 /* 349 * update the pipelimit (ssthresh) if it has been updated 350 * already or if a pipesize was specified & the threshhold 351 * got below half the pipesize. I.e., wait for bad news 352 * before we start updating, then update on both good 353 * and bad news. 354 */ 355 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 356 ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) || 357 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 358 /* 359 * convert the limit from user data bytes to 360 * packets then to packet data bytes. 361 */ 362 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 363 if (i < 2) 364 i = 2; 365 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr)); 366 if (rt->rt_rmx.rmx_ssthresh) 367 rt->rt_rmx.rmx_ssthresh = 368 (rt->rt_rmx.rmx_ssthresh + i) / 2; 369 else 370 rt->rt_rmx.rmx_ssthresh = i; 371 tcpstat.tcps_cachedssthresh++; 372 } 373 } 374#endif /* RTV_RTT */ 375 /* free the reassembly queue, if any */ 376 t = tp->seg_next; 377 while (t != (struct tcpiphdr *)tp) { 378 t = (struct tcpiphdr *)t->ti_next; 379 m = REASS_MBUF((struct tcpiphdr *)t->ti_prev); 380 remque(t->ti_prev); 381 m_freem(m); 382 } 383 if (tp->t_template) 384 (void) m_free(dtom(tp->t_template)); 385 free(tp, M_PCB); 386 inp->inp_ppcb = 0; 387 soisdisconnected(so); 388 in_pcbdetach(inp); 389 tcpstat.tcps_closed++; 390 return ((struct tcpcb *)0); 391} 392 393void 394tcp_drain() 395{ 396 397} 398 399/* 400 * Notify a tcp user of an asynchronous error; 401 * store error as soft error, but wake up user 402 * (for now, won't do anything until can select for soft error). 403 */ 404void 405tcp_notify(inp, error) 406 struct inpcb *inp; 407 int error; 408{ 409 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 410 register struct socket *so = inp->inp_socket; 411 412 /* 413 * Ignore some errors if we are hooked up. 414 * If connection hasn't completed, has retransmitted several times, 415 * and receives a second error, give up now. This is better 416 * than waiting a long time to establish a connection that 417 * can never complete. 418 */ 419 if (tp->t_state == TCPS_ESTABLISHED && 420 (error == EHOSTUNREACH || error == ENETUNREACH || 421 error == EHOSTDOWN)) { 422 return; 423 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 424 tp->t_softerror) 425 so->so_error = error; 426 else 427 tp->t_softerror = error; 428 wakeup((caddr_t) &so->so_timeo); 429 sorwakeup(so); 430 sowwakeup(so); 431} 432 433void 434tcp_ctlinput(cmd, sa, ip) 435 int cmd; 436 struct sockaddr *sa; 437 register struct ip *ip; 438{ 439 register struct tcphdr *th; 440 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 441 442 if (cmd == PRC_QUENCH) 443 notify = tcp_quench; 444 else if (!PRC_IS_REDIRECT(cmd) && 445 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)) 446 return; 447 if (ip) { 448 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 449 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport, 450 cmd, notify); 451 } else 452 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify); 453} 454 455/* 456 * When a source quench is received, close congestion window 457 * to one segment. We will gradually open it again as we proceed. 458 */ 459void 460tcp_quench(inp, errno) 461 struct inpcb *inp; 462 int errno; 463{ 464 struct tcpcb *tp = intotcpcb(inp); 465 466 if (tp) 467 tp->snd_cwnd = tp->t_maxseg; 468} 469 470/* 471 * Look-up the routing entry to the peer of this inpcb. If no route 472 * is found and it cannot be allocated the return NULL. This routine 473 * is called by TCP routines that access the rmx structure and by tcp_mss 474 * to get the interface MTU. 475 */ 476struct rtentry * 477tcp_rtlookup(inp) 478 struct inpcb *inp; 479{ 480 struct route *ro; 481 struct rtentry *rt; 482 483 ro = &inp->inp_route; 484 rt = ro->ro_rt; 485 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 486 /* No route yet, so try to acquire one */ 487 if (inp->inp_faddr.s_addr != INADDR_ANY) { 488 ro->ro_dst.sa_family = AF_INET; 489 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 490 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 491 inp->inp_faddr; 492 rtalloc(ro); 493 rt = ro->ro_rt; 494 } 495 } 496 return rt; 497} 498 499/* 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