tcp_timewait.c revision 41187
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 1.47 1998/09/06 08:17:35 phk 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 50#include <vm/vm_zone.h> 51 52#include <net/route.h> 53#include <net/if.h> 54 55#define _IP_VHL 56#include <netinet/in.h> 57#include <netinet/in_systm.h> 58#include <netinet/ip.h> 59#include <netinet/in_pcb.h> 60#include <netinet/in_var.h> 61#include <netinet/ip_var.h> 62#include <netinet/tcp.h> 63#include <netinet/tcp_fsm.h> 64#include <netinet/tcp_seq.h> 65#include <netinet/tcp_timer.h> 66#include <netinet/tcp_var.h> 67#include <netinet/tcpip.h> 68#ifdef TCPDEBUG 69#include <netinet/tcp_debug.h> 70#endif 71 72int tcp_mssdflt = TCP_MSS; 73SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 74 CTLFLAG_RW, &tcp_mssdflt , 0, ""); 75 76static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 77SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, 78 CTLFLAG_RW, &tcp_rttdflt , 0, ""); 79 80static int tcp_do_rfc1323 = 1; 81SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, 82 CTLFLAG_RW, &tcp_do_rfc1323 , 0, ""); 83 84static int tcp_do_rfc1644 = 0; 85SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, 86 CTLFLAG_RW, &tcp_do_rfc1644 , 0, ""); 87 88SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, &tcbinfo.ipi_count, 89 0, "Number of active PCBs"); 90 91static void tcp_cleartaocache __P((void)); 92static void tcp_notify __P((struct inpcb *, int)); 93 94/* 95 * Target size of TCP PCB hash tables. Must be a power of two. 96 */ 97#ifndef TCBHASHSIZE 98#define TCBHASHSIZE 512 99#endif 100 101/* 102 * This is the actual shape of what we allocate using the zone 103 * allocator. Doing it this way allows us to protect both structures 104 * using the same generation count, and also eliminates the overhead 105 * of allocating tcpcbs separately. By hiding the structure here, 106 * we avoid changing most of the rest of the code (although it needs 107 * to be changed, eventually, for greater efficiency). 108 */ 109#define ALIGNMENT 32 110#define ALIGNM1 (ALIGNMENT - 1) 111struct inp_tp { 112 union { 113 struct inpcb inp; 114 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 115 } inp_tp_u; 116 struct tcpcb tcb; 117}; 118#undef ALIGNMENT 119#undef ALIGNM1 120 121/* 122 * Tcp initialization 123 */ 124void 125tcp_init() 126{ 127 128 tcp_iss = random(); /* wrong, but better than a constant */ 129 tcp_ccgen = 1; 130 tcp_cleartaocache(); 131 LIST_INIT(&tcb); 132 tcbinfo.listhead = &tcb; 133 tcbinfo.hashbase = hashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashmask); 134 tcbinfo.porthashbase = hashinit(TCBHASHSIZE, M_PCB, 135 &tcbinfo.porthashmask); 136 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets, 137 ZONE_INTERRUPT, 0); 138 if (max_protohdr < sizeof(struct tcpiphdr)) 139 max_protohdr = sizeof(struct tcpiphdr); 140 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN) 141 panic("tcp_init"); 142} 143 144/* 145 * Create template to be used to send tcp packets on a connection. 146 * Call after host entry created, allocates an mbuf and fills 147 * in a skeletal tcp/ip header, minimizing the amount of work 148 * necessary when the connection is used. 149 */ 150struct tcpiphdr * 151tcp_template(tp) 152 struct tcpcb *tp; 153{ 154 register struct inpcb *inp = tp->t_inpcb; 155 register struct mbuf *m; 156 register struct tcpiphdr *n; 157 158 if ((n = tp->t_template) == 0) { 159 m = m_get(M_DONTWAIT, MT_HEADER); 160 if (m == NULL) 161 return (0); 162 m->m_len = sizeof (struct tcpiphdr); 163 n = mtod(m, struct tcpiphdr *); 164 } 165 bzero(n->ti_x1, sizeof(n->ti_x1)); 166 n->ti_pr = IPPROTO_TCP; 167 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip)); 168 n->ti_src = inp->inp_laddr; 169 n->ti_dst = inp->inp_faddr; 170 n->ti_sport = inp->inp_lport; 171 n->ti_dport = inp->inp_fport; 172 n->ti_seq = 0; 173 n->ti_ack = 0; 174 n->ti_x2 = 0; 175 n->ti_off = 5; 176 n->ti_flags = 0; 177 n->ti_win = 0; 178 n->ti_sum = 0; 179 n->ti_urp = 0; 180 return (n); 181} 182 183/* 184 * Send a single message to the TCP at address specified by 185 * the given TCP/IP header. If m == 0, then we make a copy 186 * of the tcpiphdr at ti and send directly to the addressed host. 187 * This is used to force keep alive messages out using the TCP 188 * template for a connection tp->t_template. If flags are given 189 * then we send a message back to the TCP which originated the 190 * segment ti, and discard the mbuf containing it and any other 191 * attached mbufs. 192 * 193 * In any case the ack and sequence number of the transmitted 194 * segment are as specified by the parameters. 195 * 196 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 197 */ 198void 199tcp_respond(tp, ti, m, ack, seq, flags) 200 struct tcpcb *tp; 201 register struct tcpiphdr *ti; 202 register struct mbuf *m; 203 tcp_seq ack, seq; 204 int flags; 205{ 206 register int tlen; 207 int win = 0; 208 struct route *ro = 0; 209 struct route sro; 210 211 if (tp) { 212 if (!(flags & TH_RST)) 213 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 214 ro = &tp->t_inpcb->inp_route; 215 } else { 216 ro = &sro; 217 bzero(ro, sizeof *ro); 218 } 219 if (m == 0) { 220 m = m_gethdr(M_DONTWAIT, MT_HEADER); 221 if (m == NULL) 222 return; 223#ifdef TCP_COMPAT_42 224 tlen = 1; 225#else 226 tlen = 0; 227#endif 228 m->m_data += max_linkhdr; 229 *mtod(m, struct tcpiphdr *) = *ti; 230 ti = mtod(m, struct tcpiphdr *); 231 flags = TH_ACK; 232 } else { 233 m_freem(m->m_next); 234 m->m_next = 0; 235 m->m_data = (caddr_t)ti; 236 m->m_len = sizeof (struct tcpiphdr); 237 tlen = 0; 238#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 239 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, n_long); 240 xchg(ti->ti_dport, ti->ti_sport, n_short); 241#undef xchg 242 } 243 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen)); 244 tlen += sizeof (struct tcpiphdr); 245 m->m_len = tlen; 246 m->m_pkthdr.len = tlen; 247 m->m_pkthdr.rcvif = (struct ifnet *) 0; 248 bzero(ti->ti_x1, sizeof(ti->ti_x1)); 249 ti->ti_seq = htonl(seq); 250 ti->ti_ack = htonl(ack); 251 ti->ti_x2 = 0; 252 ti->ti_off = sizeof (struct tcphdr) >> 2; 253 ti->ti_flags = flags; 254 if (tp) 255 ti->ti_win = htons((u_short) (win >> tp->rcv_scale)); 256 else 257 ti->ti_win = htons((u_short)win); 258 ti->ti_urp = 0; 259 ti->ti_sum = 0; 260 ti->ti_sum = in_cksum(m, tlen); 261 ((struct ip *)ti)->ip_len = tlen; 262 ((struct ip *)ti)->ip_ttl = ip_defttl; 263#ifdef TCPDEBUG 264 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 265 tcp_trace(TA_OUTPUT, 0, tp, ti, 0); 266#endif 267 (void) ip_output(m, NULL, ro, 0, NULL); 268 if (ro == &sro && ro->ro_rt) { 269 RTFREE(ro->ro_rt); 270 } 271} 272 273/* 274 * Create a new TCP control block, making an 275 * empty reassembly queue and hooking it to the argument 276 * protocol control block. The `inp' parameter must have 277 * come from the zone allocator set up in tcp_init(). 278 */ 279struct tcpcb * 280tcp_newtcpcb(inp) 281 struct inpcb *inp; 282{ 283 struct inp_tp *it; 284 register struct tcpcb *tp; 285 286 it = (struct inp_tp *)inp; 287 tp = &it->tcb; 288 bzero((char *) tp, sizeof(struct tcpcb)); 289 tp->t_segq = NULL; 290 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt; 291 292 if (tcp_do_rfc1323) 293 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 294 if (tcp_do_rfc1644) 295 tp->t_flags |= TF_REQ_CC; 296 tp->t_inpcb = inp; /* XXX */ 297 /* 298 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 299 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 300 * reasonable initial retransmit time. 301 */ 302 tp->t_srtt = TCPTV_SRTTBASE; 303 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 304 tp->t_rttmin = TCPTV_MIN; 305 tp->t_rxtcur = TCPTV_RTOBASE; 306 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 307 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 308 inp->inp_ip_ttl = ip_defttl; 309 inp->inp_ppcb = (caddr_t)tp; 310 return (tp); /* XXX */ 311} 312 313/* 314 * Drop a TCP connection, reporting 315 * the specified error. If connection is synchronized, 316 * then send a RST to peer. 317 */ 318struct tcpcb * 319tcp_drop(tp, errno) 320 register struct tcpcb *tp; 321 int errno; 322{ 323 struct socket *so = tp->t_inpcb->inp_socket; 324 325 if (TCPS_HAVERCVDSYN(tp->t_state)) { 326 tp->t_state = TCPS_CLOSED; 327 (void) tcp_output(tp); 328 tcpstat.tcps_drops++; 329 } else 330 tcpstat.tcps_conndrops++; 331 if (errno == ETIMEDOUT && tp->t_softerror) 332 errno = tp->t_softerror; 333 so->so_error = errno; 334 return (tcp_close(tp)); 335} 336 337/* 338 * Close a TCP control block: 339 * discard all space held by the tcp 340 * discard internet protocol block 341 * wake up any sleepers 342 */ 343struct tcpcb * 344tcp_close(tp) 345 register struct tcpcb *tp; 346{ 347 register struct mbuf *q; 348 register struct mbuf *nq; 349 struct inpcb *inp = tp->t_inpcb; 350 struct socket *so = inp->inp_socket; 351 register struct mbuf *m; 352 register struct rtentry *rt; 353 int dosavessthresh; 354 355 /* 356 * If we got enough samples through the srtt filter, 357 * save the rtt and rttvar in the routing entry. 358 * 'Enough' is arbitrarily defined as the 16 samples. 359 * 16 samples is enough for the srtt filter to converge 360 * to within 5% of the correct value; fewer samples and 361 * we could save a very bogus rtt. 362 * 363 * Don't update the default route's characteristics and don't 364 * update anything that the user "locked". 365 */ 366 if (tp->t_rttupdated >= 16 && 367 (rt = inp->inp_route.ro_rt) && 368 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) { 369 register u_long i = 0; 370 371 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 372 i = tp->t_srtt * 373 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 374 if (rt->rt_rmx.rmx_rtt && i) 375 /* 376 * filter this update to half the old & half 377 * the new values, converting scale. 378 * See route.h and tcp_var.h for a 379 * description of the scaling constants. 380 */ 381 rt->rt_rmx.rmx_rtt = 382 (rt->rt_rmx.rmx_rtt + i) / 2; 383 else 384 rt->rt_rmx.rmx_rtt = i; 385 tcpstat.tcps_cachedrtt++; 386 } 387 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 388 i = tp->t_rttvar * 389 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 390 if (rt->rt_rmx.rmx_rttvar && i) 391 rt->rt_rmx.rmx_rttvar = 392 (rt->rt_rmx.rmx_rttvar + i) / 2; 393 else 394 rt->rt_rmx.rmx_rttvar = i; 395 tcpstat.tcps_cachedrttvar++; 396 } 397 /* 398 * The old comment here said: 399 * update the pipelimit (ssthresh) if it has been updated 400 * already or if a pipesize was specified & the threshhold 401 * got below half the pipesize. I.e., wait for bad news 402 * before we start updating, then update on both good 403 * and bad news. 404 * 405 * But we want to save the ssthresh even if no pipesize is 406 * specified explicitly in the route, because such 407 * connections still have an implicit pipesize specified 408 * by the global tcp_sendspace. In the absence of a reliable 409 * way to calculate the pipesize, it will have to do. 410 */ 411 i = tp->snd_ssthresh; 412 if (rt->rt_rmx.rmx_sendpipe != 0) 413 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 414 else 415 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 416 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 417 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 418 || dosavessthresh) { 419 /* 420 * convert the limit from user data bytes to 421 * packets then to packet data bytes. 422 */ 423 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 424 if (i < 2) 425 i = 2; 426 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr)); 427 if (rt->rt_rmx.rmx_ssthresh) 428 rt->rt_rmx.rmx_ssthresh = 429 (rt->rt_rmx.rmx_ssthresh + i) / 2; 430 else 431 rt->rt_rmx.rmx_ssthresh = i; 432 tcpstat.tcps_cachedssthresh++; 433 } 434 } 435 /* free the reassembly queue, if any */ 436 for (q = tp->t_segq; q; q = nq) { 437 nq = q->m_nextpkt; 438 tp->t_segq = nq; 439 m_freem(q); 440 } 441 if (tp->t_template) 442 (void) m_free(dtom(tp->t_template)); 443 inp->inp_ppcb = NULL; 444 soisdisconnected(so); 445 in_pcbdetach(inp); 446 tcpstat.tcps_closed++; 447 return ((struct tcpcb *)0); 448} 449 450void 451tcp_drain() 452{ 453 454} 455 456/* 457 * Notify a tcp user of an asynchronous error; 458 * store error as soft error, but wake up user 459 * (for now, won't do anything until can select for soft error). 460 */ 461static void 462tcp_notify(inp, error) 463 struct inpcb *inp; 464 int error; 465{ 466 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 467 register struct socket *so = inp->inp_socket; 468 469 /* 470 * Ignore some errors if we are hooked up. 471 * If connection hasn't completed, has retransmitted several times, 472 * and receives a second error, give up now. This is better 473 * than waiting a long time to establish a connection that 474 * can never complete. 475 */ 476 if (tp->t_state == TCPS_ESTABLISHED && 477 (error == EHOSTUNREACH || error == ENETUNREACH || 478 error == EHOSTDOWN)) { 479 return; 480 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 481 tp->t_softerror) 482 so->so_error = error; 483 else 484 tp->t_softerror = error; 485 wakeup((caddr_t) &so->so_timeo); 486 sorwakeup(so); 487 sowwakeup(so); 488} 489 490static int 491tcp_pcblist SYSCTL_HANDLER_ARGS 492{ 493 int error, i, n, s; 494 struct inpcb *inp, **inp_list; 495 inp_gen_t gencnt; 496 struct xinpgen xig; 497 498 /* 499 * The process of preparing the TCB list is too time-consuming and 500 * resource-intensive to repeat twice on every request. 501 */ 502 if (req->oldptr == 0) { 503 n = tcbinfo.ipi_count; 504 req->oldidx = 2 * (sizeof xig) 505 + (n + n/8) * sizeof(struct xtcpcb); 506 return 0; 507 } 508 509 if (req->newptr != 0) 510 return EPERM; 511 512 /* 513 * OK, now we're committed to doing something. 514 */ 515 s = splnet(); 516 gencnt = tcbinfo.ipi_gencnt; 517 n = tcbinfo.ipi_count; 518 splx(s); 519 520 xig.xig_len = sizeof xig; 521 xig.xig_count = n; 522 xig.xig_gen = gencnt; 523 xig.xig_sogen = so_gencnt; 524 error = SYSCTL_OUT(req, &xig, sizeof xig); 525 if (error) 526 return error; 527 528 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 529 if (inp_list == 0) 530 return ENOMEM; 531 532 s = splnet(); 533 for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n; 534 inp = inp->inp_list.le_next) { 535 if (inp->inp_gencnt <= gencnt) 536 inp_list[i++] = inp; 537 } 538 splx(s); 539 n = i; 540 541 error = 0; 542 for (i = 0; i < n; i++) { 543 inp = inp_list[i]; 544 if (inp->inp_gencnt <= gencnt) { 545 struct xtcpcb xt; 546 xt.xt_len = sizeof xt; 547 /* XXX should avoid extra copy */ 548 bcopy(inp, &xt.xt_inp, sizeof *inp); 549 bcopy(inp->inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 550 if (inp->inp_socket) 551 sotoxsocket(inp->inp_socket, &xt.xt_socket); 552 error = SYSCTL_OUT(req, &xt, sizeof xt); 553 } 554 } 555 if (!error) { 556 /* 557 * Give the user an updated idea of our state. 558 * If the generation differs from what we told 559 * her before, she knows that something happened 560 * while we were processing this request, and it 561 * might be necessary to retry. 562 */ 563 s = splnet(); 564 xig.xig_gen = tcbinfo.ipi_gencnt; 565 xig.xig_sogen = so_gencnt; 566 xig.xig_count = tcbinfo.ipi_count; 567 splx(s); 568 error = SYSCTL_OUT(req, &xig, sizeof xig); 569 } 570 free(inp_list, M_TEMP); 571 return error; 572} 573 574SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 575 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 576 577void 578tcp_ctlinput(cmd, sa, vip) 579 int cmd; 580 struct sockaddr *sa; 581 void *vip; 582{ 583 register struct ip *ip = vip; 584 register struct tcphdr *th; 585 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 586 587 if (cmd == PRC_QUENCH) 588 notify = tcp_quench; 589 else if (cmd == PRC_MSGSIZE) 590 notify = tcp_mtudisc; 591 else if (!PRC_IS_REDIRECT(cmd) && 592 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)) 593 return; 594 if (ip) { 595 th = (struct tcphdr *)((caddr_t)ip 596 + (IP_VHL_HL(ip->ip_vhl) << 2)); 597 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport, 598 cmd, notify); 599 } else 600 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify); 601} 602 603/* 604 * When a source quench is received, close congestion window 605 * to one segment. We will gradually open it again as we proceed. 606 */ 607void 608tcp_quench(inp, errno) 609 struct inpcb *inp; 610 int errno; 611{ 612 struct tcpcb *tp = intotcpcb(inp); 613 614 if (tp) 615 tp->snd_cwnd = tp->t_maxseg; 616} 617 618/* 619 * When `need fragmentation' ICMP is received, update our idea of the MSS 620 * based on the new value in the route. Also nudge TCP to send something, 621 * since we know the packet we just sent was dropped. 622 * This duplicates some code in the tcp_mss() function in tcp_input.c. 623 */ 624void 625tcp_mtudisc(inp, errno) 626 struct inpcb *inp; 627 int errno; 628{ 629 struct tcpcb *tp = intotcpcb(inp); 630 struct rtentry *rt; 631 struct rmxp_tao *taop; 632 struct socket *so = inp->inp_socket; 633 int offered; 634 int mss; 635 636 if (tp) { 637 rt = tcp_rtlookup(inp); 638 if (!rt || !rt->rt_rmx.rmx_mtu) { 639 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 640 return; 641 } 642 taop = rmx_taop(rt->rt_rmx); 643 offered = taop->tao_mssopt; 644 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 645 if (offered) 646 mss = min(mss, offered); 647 /* 648 * XXX - The above conditional probably violates the TCP 649 * spec. The problem is that, since we don't know the 650 * other end's MSS, we are supposed to use a conservative 651 * default. But, if we do that, then MTU discovery will 652 * never actually take place, because the conservative 653 * default is much less than the MTUs typically seen 654 * on the Internet today. For the moment, we'll sweep 655 * this under the carpet. 656 * 657 * The conservative default might not actually be a problem 658 * if the only case this occurs is when sending an initial 659 * SYN with options and data to a host we've never talked 660 * to before. Then, they will reply with an MSS value which 661 * will get recorded and the new parameters should get 662 * recomputed. For Further Study. 663 */ 664 if (tp->t_maxopd <= mss) 665 return; 666 tp->t_maxopd = mss; 667 668 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 669 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 670 mss -= TCPOLEN_TSTAMP_APPA; 671 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 672 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 673 mss -= TCPOLEN_CC_APPA; 674#if (MCLBYTES & (MCLBYTES - 1)) == 0 675 if (mss > MCLBYTES) 676 mss &= ~(MCLBYTES-1); 677#else 678 if (mss > MCLBYTES) 679 mss = mss / MCLBYTES * MCLBYTES; 680#endif 681 if (so->so_snd.sb_hiwat < mss) 682 mss = so->so_snd.sb_hiwat; 683 684 tp->t_maxseg = mss; 685 686 tcpstat.tcps_mturesent++; 687 tp->t_rtt = 0; 688 tp->snd_nxt = tp->snd_una; 689 tcp_output(tp); 690 } 691} 692 693/* 694 * Look-up the routing entry to the peer of this inpcb. If no route 695 * is found and it cannot be allocated the return NULL. This routine 696 * is called by TCP routines that access the rmx structure and by tcp_mss 697 * to get the interface MTU. 698 */ 699struct rtentry * 700tcp_rtlookup(inp) 701 struct inpcb *inp; 702{ 703 struct route *ro; 704 struct rtentry *rt; 705 706 ro = &inp->inp_route; 707 rt = ro->ro_rt; 708 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 709 /* No route yet, so try to acquire one */ 710 if (inp->inp_faddr.s_addr != INADDR_ANY) { 711 ro->ro_dst.sa_family = AF_INET; 712 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 713 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 714 inp->inp_faddr; 715 rtalloc(ro); 716 rt = ro->ro_rt; 717 } 718 } 719 return rt; 720} 721 722/* 723 * Return a pointer to the cached information about the remote host. 724 * The cached information is stored in the protocol specific part of 725 * the route metrics. 726 */ 727struct rmxp_tao * 728tcp_gettaocache(inp) 729 struct inpcb *inp; 730{ 731 struct rtentry *rt = tcp_rtlookup(inp); 732 733 /* Make sure this is a host route and is up. */ 734 if (rt == NULL || 735 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 736 return NULL; 737 738 return rmx_taop(rt->rt_rmx); 739} 740 741/* 742 * Clear all the TAO cache entries, called from tcp_init. 743 * 744 * XXX 745 * This routine is just an empty one, because we assume that the routing 746 * routing tables are initialized at the same time when TCP, so there is 747 * nothing in the cache left over. 748 */ 749static void 750tcp_cleartaocache() 751{ 752} 753