tcp_timewait.c revision 34881
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 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 8.2 (Berkeley) 5/24/95 34 * $Id: tcp_subr.c,v 1.42 1998/01/27 09:15:10 davidg Exp $ 35 */ 36 37#include "opt_compat.h" 38#include "opt_tcpdebug.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/kernel.h> 43#include <sys/sysctl.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/socket.h> 47#include <sys/socketvar.h> 48#include <sys/protosw.h> 49#include <vm/vm_zone.h> 50 51#include <net/route.h> 52#include <net/if.h> 53 54#define _IP_VHL 55#include <netinet/in.h> 56#include <netinet/in_systm.h> 57#include <netinet/ip.h> 58#include <netinet/in_pcb.h> 59#include <netinet/in_var.h> 60#include <netinet/ip_var.h> 61#include <netinet/tcp.h> 62#include <netinet/tcp_fsm.h> 63#include <netinet/tcp_seq.h> 64#include <netinet/tcp_timer.h> 65#include <netinet/tcp_var.h> 66#include <netinet/tcpip.h> 67#ifdef TCPDEBUG 68#include <netinet/tcp_debug.h> 69#endif 70 71int tcp_mssdflt = TCP_MSS; 72SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 73 CTLFLAG_RW, &tcp_mssdflt , 0, ""); 74 75static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 76SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, 77 CTLFLAG_RW, &tcp_rttdflt , 0, ""); 78 79static int tcp_do_rfc1323 = 1; 80SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, 81 CTLFLAG_RW, &tcp_do_rfc1323 , 0, ""); 82 83static int tcp_do_rfc1644 = 1; 84SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, 85 CTLFLAG_RW, &tcp_do_rfc1644 , 0, ""); 86 87static void tcp_cleartaocache __P((void)); 88static void tcp_notify __P((struct inpcb *, int)); 89 90/* 91 * Target size of TCP PCB hash tables. Must be a power of two. 92 */ 93#ifndef TCBHASHSIZE 94#define TCBHASHSIZE 512 95#endif 96 97/* 98 * This is the actual shape of what we allocate using the zone 99 * allocator. Doing it this way allows us to protect both structures 100 * using the same generation count, and also eliminates the overhead 101 * of allocating tcpcbs separately. By hiding the structure here, 102 * we avoid changing most of the rest of the code (although it needs 103 * to be changed, eventually, for greater efficiency). 104 */ 105#define ALIGNMENT 32 106#define ALIGNM1 (ALIGNMENT-1) 107struct inp_tp { 108 union { 109 struct inpcb inp; 110 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 111 } inp_tp_u; 112 struct tcpcb tcb; 113}; 114#undef ALIGNMENT 115#undef ALIGNM1 116 117/* 118 * Tcp initialization 119 */ 120void 121tcp_init() 122{ 123 124 tcp_iss = random(); /* wrong, but better than a constant */ 125 tcp_ccgen = 1; 126 tcp_cleartaocache(); 127 LIST_INIT(&tcb); 128 tcbinfo.listhead = &tcb; 129 tcbinfo.hashbase = hashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashmask); 130 tcbinfo.porthashbase = hashinit(TCBHASHSIZE, M_PCB, &tcbinfo.porthashmask); 131 /* For the moment, we just worry about putting inpcbs here. */ 132 /* 133 * Rationale for a maximum of `nmbclusters': 134 * 1) It's a convenient value, sized by config, based on 135 * parameters already known to be tweakable as needed 136 * for network-intensive systems. 137 * 2) Under the Old World Order, when pcbs were stored in 138 * mbufs, it was of course impossible to have more 139 * pcbs than mbufs. 140 * 3) The zone allocator doesn't allocate physical memory 141 * for this many pcbs; it just sizes the virtual 142 * address space appropriately. Thus, even for very large 143 * values of nmbclusters, we don't actually take up much 144 * memory unless required. 145 */ 146 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), nmbclusters, 147 ZONE_INTERRUPT, 0); 148 if (max_protohdr < sizeof(struct tcpiphdr)) 149 max_protohdr = sizeof(struct tcpiphdr); 150 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 151 panic("tcp_init"); 152} 153 154/* 155 * Create template to be used to send tcp packets on a connection. 156 * Call after host entry created, allocates an mbuf and fills 157 * in a skeletal tcp/ip header, minimizing the amount of work 158 * necessary when the connection is used. 159 */ 160struct tcpiphdr * 161tcp_template(tp) 162 struct tcpcb *tp; 163{ 164 register struct inpcb *inp = tp->t_inpcb; 165 register struct mbuf *m; 166 register struct tcpiphdr *n; 167 168 if ((n = tp->t_template) == 0) { 169 m = m_get(M_DONTWAIT, MT_HEADER); 170 if (m == NULL) 171 return (0); 172 m->m_len = sizeof (struct tcpiphdr); 173 n = mtod(m, struct tcpiphdr *); 174 } 175 n->ti_next = n->ti_prev = 0; 176 n->ti_x1 = 0; 177 n->ti_pr = IPPROTO_TCP; 178 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 179 n->ti_src = inp->inp_laddr; 180 n->ti_dst = inp->inp_faddr; 181 n->ti_sport = inp->inp_lport; 182 n->ti_dport = inp->inp_fport; 183 n->ti_seq = 0; 184 n->ti_ack = 0; 185 n->ti_x2 = 0; 186 n->ti_off = 5; 187 n->ti_flags = 0; 188 n->ti_win = 0; 189 n->ti_sum = 0; 190 n->ti_urp = 0; 191 return (n); 192} 193 194/* 195 * Send a single message to the TCP at address specified by 196 * the given TCP/IP header. If m == 0, then we make a copy 197 * of the tcpiphdr at ti and send directly to the addressed host. 198 * This is used to force keep alive messages out using the TCP 199 * template for a connection tp->t_template. If flags are given 200 * then we send a message back to the TCP which originated the 201 * segment ti, and discard the mbuf containing it and any other 202 * attached mbufs. 203 * 204 * In any case the ack and sequence number of the transmitted 205 * segment are as specified by the parameters. 206 * 207 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 208 */ 209void 210tcp_respond(tp, ti, m, ack, seq, flags) 211 struct tcpcb *tp; 212 register struct tcpiphdr *ti; 213 register struct mbuf *m; 214 tcp_seq ack, seq; 215 int flags; 216{ 217 register int tlen; 218 int win = 0; 219 struct route *ro = 0; 220 struct route sro; 221 222 if (tp) { 223 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 224 ro = &tp->t_inpcb->inp_route; 225 } else { 226 ro = &sro; 227 bzero(ro, sizeof *ro); 228 } 229 if (m == 0) { 230 m = m_gethdr(M_DONTWAIT, MT_HEADER); 231 if (m == NULL) 232 return; 233#ifdef TCP_COMPAT_42 234 tlen = 1; 235#else 236 tlen = 0; 237#endif 238 m->m_data += max_linkhdr; 239 *mtod(m, struct tcpiphdr *) = *ti; 240 ti = mtod(m, struct tcpiphdr *); 241 flags = TH_ACK; 242 } else { 243 m_freem(m->m_next); 244 m->m_next = 0; 245 m->m_data = (caddr_t)ti; 246 m->m_len = sizeof (struct tcpiphdr); 247 tlen = 0; 248#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 249 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long); 250 xchg(ti->ti_dport, ti->ti_sport, u_short); 251#undef xchg 252 } 253 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); 254 tlen += sizeof (struct tcpiphdr); 255 m->m_len = tlen; 256 m->m_pkthdr.len = tlen; 257 m->m_pkthdr.rcvif = (struct ifnet *) 0; 258 ti->ti_next = ti->ti_prev = 0; 259 ti->ti_x1 = 0; 260 ti->ti_seq = htonl(seq); 261 ti->ti_ack = htonl(ack); 262 ti->ti_x2 = 0; 263 ti->ti_off = sizeof (struct tcphdr) >> 2; 264 ti->ti_flags = flags; 265 if (tp) 266 ti->ti_win = htons((u_short) (win >> tp->rcv_scale)); 267 else 268 ti->ti_win = htons((u_short)win); 269 ti->ti_urp = 0; 270 ti->ti_sum = 0; 271 ti->ti_sum = in_cksum(m, tlen); 272 ((struct ip *)ti)->ip_len = tlen; 273 ((struct ip *)ti)->ip_ttl = ip_defttl; 274#ifdef TCPDEBUG 275 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 276 tcp_trace(TA_OUTPUT, 0, tp, ti, 0); 277#endif 278 (void) ip_output(m, NULL, ro, 0, NULL); 279 if (ro == &sro && ro->ro_rt) { 280 RTFREE(ro->ro_rt); 281 } 282} 283 284/* 285 * Create a new TCP control block, making an 286 * empty reassembly queue and hooking it to the argument 287 * protocol control block. The `inp' parameter must have 288 * come from the zone allocator set up in tcp_init(). 289 */ 290struct tcpcb * 291tcp_newtcpcb(inp) 292 struct inpcb *inp; 293{ 294 struct inp_tp *it; 295 register struct tcpcb *tp; 296 297 it = (struct inp_tp *)inp; 298 tp = &it->tcb; 299 bzero((char *) tp, sizeof(struct tcpcb)); 300 tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp; 301 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt; 302 303 if (tcp_do_rfc1323) 304 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 305 if (tcp_do_rfc1644) 306 tp->t_flags |= TF_REQ_CC; 307 tp->t_inpcb = inp; /* XXX */ 308 /* 309 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 310 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 311 * reasonable initial retransmit time. 312 */ 313 tp->t_srtt = TCPTV_SRTTBASE; 314 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 315 tp->t_rttmin = TCPTV_MIN; 316 tp->t_rxtcur = TCPTV_RTOBASE; 317 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 318 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 319 inp->inp_ip_ttl = ip_defttl; 320 inp->inp_ppcb = (caddr_t)tp; 321 return (tp); /* XXX */ 322} 323 324/* 325 * Drop a TCP connection, reporting 326 * the specified error. If connection is synchronized, 327 * then send a RST to peer. 328 */ 329struct tcpcb * 330tcp_drop(tp, errno) 331 register struct tcpcb *tp; 332 int errno; 333{ 334 struct socket *so = tp->t_inpcb->inp_socket; 335 336 if (TCPS_HAVERCVDSYN(tp->t_state)) { 337 tp->t_state = TCPS_CLOSED; 338 (void) tcp_output(tp); 339 tcpstat.tcps_drops++; 340 } else 341 tcpstat.tcps_conndrops++; 342 if (errno == ETIMEDOUT && tp->t_softerror) 343 errno = tp->t_softerror; 344 so->so_error = errno; 345 return (tcp_close(tp)); 346} 347 348/* 349 * Close a TCP control block: 350 * discard all space held by the tcp 351 * discard internet protocol block 352 * wake up any sleepers 353 */ 354struct tcpcb * 355tcp_close(tp) 356 register struct tcpcb *tp; 357{ 358 register struct tcpiphdr *t; 359 struct inpcb *inp = tp->t_inpcb; 360 struct socket *so = inp->inp_socket; 361 register struct mbuf *m; 362 register struct rtentry *rt; 363 int dosavessthresh; 364 365 /* 366 * If we got enough samples through the srtt filter, 367 * save the rtt and rttvar in the routing entry. 368 * 'Enough' is arbitrarily defined as the 16 samples. 369 * 16 samples is enough for the srtt filter to converge 370 * to within 5% of the correct value; fewer samples and 371 * we could save a very bogus rtt. 372 * 373 * Don't update the default route's characteristics and don't 374 * update anything that the user "locked". 375 */ 376 if (tp->t_rttupdated >= 16 && 377 (rt = inp->inp_route.ro_rt) && 378 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) { 379 register u_long i = 0; 380 381 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 382 i = tp->t_srtt * 383 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 384 if (rt->rt_rmx.rmx_rtt && i) 385 /* 386 * filter this update to half the old & half 387 * the new values, converting scale. 388 * See route.h and tcp_var.h for a 389 * description of the scaling constants. 390 */ 391 rt->rt_rmx.rmx_rtt = 392 (rt->rt_rmx.rmx_rtt + i) / 2; 393 else 394 rt->rt_rmx.rmx_rtt = i; 395 tcpstat.tcps_cachedrtt++; 396 } 397 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 398 i = tp->t_rttvar * 399 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 400 if (rt->rt_rmx.rmx_rttvar && i) 401 rt->rt_rmx.rmx_rttvar = 402 (rt->rt_rmx.rmx_rttvar + i) / 2; 403 else 404 rt->rt_rmx.rmx_rttvar = i; 405 tcpstat.tcps_cachedrttvar++; 406 } 407 /* 408 * The old comment here said: 409 * update the pipelimit (ssthresh) if it has been updated 410 * already or if a pipesize was specified & the threshhold 411 * got below half the pipesize. I.e., wait for bad news 412 * before we start updating, then update on both good 413 * and bad news. 414 * 415 * But we want to save the ssthresh even if no pipesize is 416 * specified explicitly in the route, because such 417 * connections still have an implicit pipesize specified 418 * by the global tcp_sendspace. In the absence of a reliable 419 * way to calculate the pipesize, it will have to do. 420 */ 421 i = tp->snd_ssthresh; 422#if 1 423 if (rt->rt_rmx.rmx_sendpipe != 0) 424 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 425 else 426 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 427#else 428 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 429#endif 430 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 431 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 432 || dosavessthresh) { 433 /* 434 * convert the limit from user data bytes to 435 * packets then to packet data bytes. 436 */ 437 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 438 if (i < 2) 439 i = 2; 440 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr)); 441 if (rt->rt_rmx.rmx_ssthresh) 442 rt->rt_rmx.rmx_ssthresh = 443 (rt->rt_rmx.rmx_ssthresh + i) / 2; 444 else 445 rt->rt_rmx.rmx_ssthresh = i; 446 tcpstat.tcps_cachedssthresh++; 447 } 448 } 449 /* free the reassembly queue, if any */ 450 t = tp->seg_next; 451 while (t != (struct tcpiphdr *)tp) { 452 t = (struct tcpiphdr *)t->ti_next; 453 m = REASS_MBUF((struct tcpiphdr *)t->ti_prev); 454 remque(t->ti_prev); 455 m_freem(m); 456 } 457 if (tp->t_template) 458 (void) m_free(dtom(tp->t_template)); 459 inp->inp_ppcb = NULL; 460 soisdisconnected(so); 461 in_pcbdetach(inp); 462 tcpstat.tcps_closed++; 463 return ((struct tcpcb *)0); 464} 465 466void 467tcp_drain() 468{ 469 470} 471 472/* 473 * Notify a tcp user of an asynchronous error; 474 * store error as soft error, but wake up user 475 * (for now, won't do anything until can select for soft error). 476 */ 477static void 478tcp_notify(inp, error) 479 struct inpcb *inp; 480 int error; 481{ 482 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 483 register struct socket *so = inp->inp_socket; 484 485 /* 486 * Ignore some errors if we are hooked up. 487 * If connection hasn't completed, has retransmitted several times, 488 * and receives a second error, give up now. This is better 489 * than waiting a long time to establish a connection that 490 * can never complete. 491 */ 492 if (tp->t_state == TCPS_ESTABLISHED && 493 (error == EHOSTUNREACH || error == ENETUNREACH || 494 error == EHOSTDOWN)) { 495 return; 496 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 497 tp->t_softerror) 498 so->so_error = error; 499 else 500 tp->t_softerror = error; 501 wakeup((caddr_t) &so->so_timeo); 502 sorwakeup(so); 503 sowwakeup(so); 504} 505 506void 507tcp_ctlinput(cmd, sa, vip) 508 int cmd; 509 struct sockaddr *sa; 510 void *vip; 511{ 512 register struct ip *ip = vip; 513 register struct tcphdr *th; 514 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 515 516 if (cmd == PRC_QUENCH) 517 notify = tcp_quench; 518#if 1 519 else if (cmd == PRC_MSGSIZE) 520 notify = tcp_mtudisc; 521#endif 522 else if (!PRC_IS_REDIRECT(cmd) && 523 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)) 524 return; 525 if (ip) { 526 th = (struct tcphdr *)((caddr_t)ip 527 + (IP_VHL_HL(ip->ip_vhl) << 2)); 528 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport, 529 cmd, notify); 530 } else 531 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify); 532} 533 534/* 535 * When a source quench is received, close congestion window 536 * to one segment. We will gradually open it again as we proceed. 537 */ 538void 539tcp_quench(inp, errno) 540 struct inpcb *inp; 541 int errno; 542{ 543 struct tcpcb *tp = intotcpcb(inp); 544 545 if (tp) 546 tp->snd_cwnd = tp->t_maxseg; 547} 548 549#if 1 550/* 551 * When `need fragmentation' ICMP is received, update our idea of the MSS 552 * based on the new value in the route. Also nudge TCP to send something, 553 * since we know the packet we just sent was dropped. 554 * This duplicates some code in the tcp_mss() function in tcp_input.c. 555 */ 556void 557tcp_mtudisc(inp, errno) 558 struct inpcb *inp; 559 int errno; 560{ 561 struct tcpcb *tp = intotcpcb(inp); 562 struct rtentry *rt; 563 struct rmxp_tao *taop; 564 struct socket *so = inp->inp_socket; 565 int offered; 566 int mss; 567 568 if (tp) { 569 rt = tcp_rtlookup(inp); 570 if (!rt || !rt->rt_rmx.rmx_mtu) { 571 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 572 return; 573 } 574 taop = rmx_taop(rt->rt_rmx); 575 offered = taop->tao_mssopt; 576 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 577 if (offered) 578 mss = min(mss, offered); 579 /* 580 * XXX - The above conditional probably violates the TCP 581 * spec. The problem is that, since we don't know the 582 * other end's MSS, we are supposed to use a conservative 583 * default. But, if we do that, then MTU discovery will 584 * never actually take place, because the conservative 585 * default is much less than the MTUs typically seen 586 * on the Internet today. For the moment, we'll sweep 587 * this under the carpet. 588 * 589 * The conservative default might not actually be a problem 590 * if the only case this occurs is when sending an initial 591 * SYN with options and data to a host we've never talked 592 * to before. Then, they will reply with an MSS value which 593 * will get recorded and the new parameters should get 594 * recomputed. For Further Study. 595 */ 596 if (tp->t_maxopd <= mss) 597 return; 598 tp->t_maxopd = mss; 599 600 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 601 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 602 mss -= TCPOLEN_TSTAMP_APPA; 603 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 604 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 605 mss -= TCPOLEN_CC_APPA; 606#if (MCLBYTES & (MCLBYTES - 1)) == 0 607 if (mss > MCLBYTES) 608 mss &= ~(MCLBYTES-1); 609#else 610 if (mss > MCLBYTES) 611 mss = mss / MCLBYTES * MCLBYTES; 612#endif 613 if (so->so_snd.sb_hiwat < mss) 614 mss = so->so_snd.sb_hiwat; 615 616 tp->t_maxseg = mss; 617 618 tcpstat.tcps_mturesent++; 619 tp->t_rtt = 0; 620 tp->snd_nxt = tp->snd_una; 621 tcp_output(tp); 622 } 623} 624#endif 625 626/* 627 * Look-up the routing entry to the peer of this inpcb. If no route 628 * is found and it cannot be allocated the return NULL. This routine 629 * is called by TCP routines that access the rmx structure and by tcp_mss 630 * to get the interface MTU. 631 */ 632struct rtentry * 633tcp_rtlookup(inp) 634 struct inpcb *inp; 635{ 636 struct route *ro; 637 struct rtentry *rt; 638 639 ro = &inp->inp_route; 640 rt = ro->ro_rt; 641 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 642 /* No route yet, so try to acquire one */ 643 if (inp->inp_faddr.s_addr != INADDR_ANY) { 644 ro->ro_dst.sa_family = AF_INET; 645 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 646 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 647 inp->inp_faddr; 648 rtalloc(ro); 649 rt = ro->ro_rt; 650 } 651 } 652 return rt; 653} 654 655/* 656 * Return a pointer to the cached information about the remote host. 657 * The cached information is stored in the protocol specific part of 658 * the route metrics. 659 */ 660struct rmxp_tao * 661tcp_gettaocache(inp) 662 struct inpcb *inp; 663{ 664 struct rtentry *rt = tcp_rtlookup(inp); 665 666 /* Make sure this is a host route and is up. */ 667 if (rt == NULL || 668 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 669 return NULL; 670 671 return rmx_taop(rt->rt_rmx); 672} 673 674/* 675 * Clear all the TAO cache entries, called from tcp_init. 676 * 677 * XXX 678 * This routine is just an empty one, because we assume that the routing 679 * routing tables are initialized at the same time when TCP, so there is 680 * nothing in the cache left over. 681 */ 682static void 683tcp_cleartaocache() 684{ 685} 686