tcp_timewait.c revision 80428
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 * $FreeBSD: head/sys/netinet/tcp_timewait.c 80428 2001-07-27 00:03:49Z peter $ 35 */ 36 37#include "opt_compat.h" 38#include "opt_inet6.h" 39#include "opt_ipsec.h" 40#include "opt_tcpdebug.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/callout.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#ifdef INET6 50#include <sys/domain.h> 51#endif 52#include <sys/proc.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55#include <sys/protosw.h> 56#include <sys/random.h> 57 58#include <vm/vm_zone.h> 59 60#include <net/route.h> 61#include <net/if.h> 62 63#define _IP_VHL 64#include <netinet/in.h> 65#include <netinet/in_systm.h> 66#include <netinet/ip.h> 67#ifdef INET6 68#include <netinet/ip6.h> 69#endif 70#include <netinet/in_pcb.h> 71#ifdef INET6 72#include <netinet6/in6_pcb.h> 73#endif 74#include <netinet/in_var.h> 75#include <netinet/ip_var.h> 76#ifdef INET6 77#include <netinet6/ip6_var.h> 78#endif 79#include <netinet/tcp.h> 80#include <netinet/tcp_fsm.h> 81#include <netinet/tcp_seq.h> 82#include <netinet/tcp_timer.h> 83#include <netinet/tcp_var.h> 84#ifdef INET6 85#include <netinet6/tcp6_var.h> 86#endif 87#include <netinet/tcpip.h> 88#ifdef TCPDEBUG 89#include <netinet/tcp_debug.h> 90#endif 91#include <netinet6/ip6protosw.h> 92 93#ifdef IPSEC 94#include <netinet6/ipsec.h> 95#ifdef INET6 96#include <netinet6/ipsec6.h> 97#endif 98#endif /*IPSEC*/ 99 100#include <machine/in_cksum.h> 101 102int tcp_mssdflt = TCP_MSS; 103SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 104 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 105 106#ifdef INET6 107int tcp_v6mssdflt = TCP6_MSS; 108SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 109 CTLFLAG_RW, &tcp_v6mssdflt , 0, 110 "Default TCP Maximum Segment Size for IPv6"); 111#endif 112 113#if 0 114static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 115SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 116 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 117#endif 118 119static int tcp_do_rfc1323 = 1; 120SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 121 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 122 123static int tcp_do_rfc1644 = 0; 124SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 125 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 126 127static int tcp_tcbhashsize = 0; 128SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 129 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 130 131static int do_tcpdrain = 1; 132SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 133 "Enable tcp_drain routine for extra help when low on mbufs"); 134 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 136 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 137 138static int icmp_may_rst = 1; 139SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 140 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 141 142static int tcp_seq_genscheme = 1; 143SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_seq_genscheme, CTLFLAG_RW, 144 &tcp_seq_genscheme, 0, "TCP ISN generation scheme"); 145 146static void tcp_cleartaocache __P((void)); 147static void tcp_notify __P((struct inpcb *, int)); 148 149/* 150 * Target size of TCP PCB hash tables. Must be a power of two. 151 * 152 * Note that this can be overridden by the kernel environment 153 * variable net.inet.tcp.tcbhashsize 154 */ 155#ifndef TCBHASHSIZE 156#define TCBHASHSIZE 512 157#endif 158 159/* 160 * This is the actual shape of what we allocate using the zone 161 * allocator. Doing it this way allows us to protect both structures 162 * using the same generation count, and also eliminates the overhead 163 * of allocating tcpcbs separately. By hiding the structure here, 164 * we avoid changing most of the rest of the code (although it needs 165 * to be changed, eventually, for greater efficiency). 166 */ 167#define ALIGNMENT 32 168#define ALIGNM1 (ALIGNMENT - 1) 169struct inp_tp { 170 union { 171 struct inpcb inp; 172 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 173 } inp_tp_u; 174 struct tcpcb tcb; 175 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 176 struct callout inp_tp_delack; 177}; 178#undef ALIGNMENT 179#undef ALIGNM1 180 181/* 182 * Tcp initialization 183 */ 184void 185tcp_init() 186{ 187 int hashsize = TCBHASHSIZE; 188 189 tcp_iss = arc4random(); /* wrong, but better than a constant */ 190 tcp_ccgen = 1; 191 tcp_cleartaocache(); 192 193 tcp_delacktime = TCPTV_DELACK; 194 tcp_keepinit = TCPTV_KEEP_INIT; 195 tcp_keepidle = TCPTV_KEEP_IDLE; 196 tcp_keepintvl = TCPTV_KEEPINTVL; 197 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 198 tcp_msl = TCPTV_MSL; 199 200 LIST_INIT(&tcb); 201 tcbinfo.listhead = &tcb; 202 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 203 if (!powerof2(hashsize)) { 204 printf("WARNING: TCB hash size not a power of 2\n"); 205 hashsize = 512; /* safe default */ 206 } 207 tcp_tcbhashsize = hashsize; 208 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 209 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 210 &tcbinfo.porthashmask); 211 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets, 212 ZONE_INTERRUPT, 0); 213#ifdef INET6 214#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 215#else /* INET6 */ 216#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 217#endif /* INET6 */ 218 if (max_protohdr < TCP_MINPROTOHDR) 219 max_protohdr = TCP_MINPROTOHDR; 220 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 221 panic("tcp_init"); 222#undef TCP_MINPROTOHDR 223} 224 225/* 226 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 227 * tcp_template used to store this data in mbufs, but we now recopy it out 228 * of the tcpcb each time to conserve mbufs. 229 */ 230void 231tcp_fillheaders(tp, ip_ptr, tcp_ptr) 232 struct tcpcb *tp; 233 void *ip_ptr; 234 void *tcp_ptr; 235{ 236 struct inpcb *inp = tp->t_inpcb; 237 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 238 239#ifdef INET6 240 if ((inp->inp_vflag & INP_IPV6) != 0) { 241 struct ip6_hdr *ip6; 242 243 ip6 = (struct ip6_hdr *)ip_ptr; 244 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 245 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 246 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 247 (IPV6_VERSION & IPV6_VERSION_MASK); 248 ip6->ip6_nxt = IPPROTO_TCP; 249 ip6->ip6_plen = sizeof(struct tcphdr); 250 ip6->ip6_src = inp->in6p_laddr; 251 ip6->ip6_dst = inp->in6p_faddr; 252 tcp_hdr->th_sum = 0; 253 } else 254#endif 255 { 256 struct ip *ip = (struct ip *) ip_ptr; 257 258 ip->ip_vhl = IP_VHL_BORING; 259 ip->ip_tos = 0; 260 ip->ip_len = 0; 261 ip->ip_id = 0; 262 ip->ip_off = 0; 263 ip->ip_ttl = 0; 264 ip->ip_sum = 0; 265 ip->ip_p = IPPROTO_TCP; 266 ip->ip_src = inp->inp_laddr; 267 ip->ip_dst = inp->inp_faddr; 268 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 269 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 270 } 271 272 tcp_hdr->th_sport = inp->inp_lport; 273 tcp_hdr->th_dport = inp->inp_fport; 274 tcp_hdr->th_seq = 0; 275 tcp_hdr->th_ack = 0; 276 tcp_hdr->th_x2 = 0; 277 tcp_hdr->th_off = 5; 278 tcp_hdr->th_flags = 0; 279 tcp_hdr->th_win = 0; 280 tcp_hdr->th_urp = 0; 281} 282 283/* 284 * Create template to be used to send tcp packets on a connection. 285 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 286 * use for this function is in keepalives, which use tcp_respond. 287 */ 288struct tcptemp * 289tcp_maketemplate(tp) 290 struct tcpcb *tp; 291{ 292 struct mbuf *m; 293 struct tcptemp *n; 294 295 m = m_get(M_DONTWAIT, MT_HEADER); 296 if (m == NULL) 297 return (0); 298 m->m_len = sizeof(struct tcptemp); 299 n = mtod(m, struct tcptemp *); 300 301 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 302 return (n); 303} 304 305/* 306 * Send a single message to the TCP at address specified by 307 * the given TCP/IP header. If m == 0, then we make a copy 308 * of the tcpiphdr at ti and send directly to the addressed host. 309 * This is used to force keep alive messages out using the TCP 310 * template for a connection. If flags are given then we send 311 * a message back to the TCP which originated the * segment ti, 312 * and discard the mbuf containing it and any other attached mbufs. 313 * 314 * In any case the ack and sequence number of the transmitted 315 * segment are as specified by the parameters. 316 * 317 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 318 */ 319void 320tcp_respond(tp, ipgen, th, m, ack, seq, flags) 321 struct tcpcb *tp; 322 void *ipgen; 323 register struct tcphdr *th; 324 register struct mbuf *m; 325 tcp_seq ack, seq; 326 int flags; 327{ 328 register int tlen; 329 int win = 0; 330 struct route *ro = 0; 331 struct route sro; 332 struct ip *ip; 333 struct tcphdr *nth; 334#ifdef INET6 335 struct route_in6 *ro6 = 0; 336 struct route_in6 sro6; 337 struct ip6_hdr *ip6; 338 int isipv6; 339#endif /* INET6 */ 340 int ipflags = 0; 341 342#ifdef INET6 343 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; 344 ip6 = ipgen; 345#endif /* INET6 */ 346 ip = ipgen; 347 348 if (tp) { 349 if (!(flags & TH_RST)) { 350 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 351 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 352 win = (long)TCP_MAXWIN << tp->rcv_scale; 353 } 354#ifdef INET6 355 if (isipv6) 356 ro6 = &tp->t_inpcb->in6p_route; 357 else 358#endif /* INET6 */ 359 ro = &tp->t_inpcb->inp_route; 360 } else { 361#ifdef INET6 362 if (isipv6) { 363 ro6 = &sro6; 364 bzero(ro6, sizeof *ro6); 365 } else 366#endif /* INET6 */ 367 { 368 ro = &sro; 369 bzero(ro, sizeof *ro); 370 } 371 } 372 if (m == 0) { 373 m = m_gethdr(M_DONTWAIT, MT_HEADER); 374 if (m == NULL) 375 return; 376 tlen = 0; 377 m->m_data += max_linkhdr; 378#ifdef INET6 379 if (isipv6) { 380 bcopy((caddr_t)ip6, mtod(m, caddr_t), 381 sizeof(struct ip6_hdr)); 382 ip6 = mtod(m, struct ip6_hdr *); 383 nth = (struct tcphdr *)(ip6 + 1); 384 } else 385#endif /* INET6 */ 386 { 387 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 388 ip = mtod(m, struct ip *); 389 nth = (struct tcphdr *)(ip + 1); 390 } 391 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 392 flags = TH_ACK; 393 } else { 394 m_freem(m->m_next); 395 m->m_next = 0; 396 m->m_data = (caddr_t)ipgen; 397 /* m_len is set later */ 398 tlen = 0; 399#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 400#ifdef INET6 401 if (isipv6) { 402 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 403 nth = (struct tcphdr *)(ip6 + 1); 404 } else 405#endif /* INET6 */ 406 { 407 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 408 nth = (struct tcphdr *)(ip + 1); 409 } 410 if (th != nth) { 411 /* 412 * this is usually a case when an extension header 413 * exists between the IPv6 header and the 414 * TCP header. 415 */ 416 nth->th_sport = th->th_sport; 417 nth->th_dport = th->th_dport; 418 } 419 xchg(nth->th_dport, nth->th_sport, n_short); 420#undef xchg 421 } 422#ifdef INET6 423 if (isipv6) { 424 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 425 tlen)); 426 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 427 } else 428#endif 429 { 430 tlen += sizeof (struct tcpiphdr); 431 ip->ip_len = tlen; 432 ip->ip_ttl = ip_defttl; 433 } 434 m->m_len = tlen; 435 m->m_pkthdr.len = tlen; 436 m->m_pkthdr.rcvif = (struct ifnet *) 0; 437 nth->th_seq = htonl(seq); 438 nth->th_ack = htonl(ack); 439 nth->th_x2 = 0; 440 nth->th_off = sizeof (struct tcphdr) >> 2; 441 nth->th_flags = flags; 442 if (tp) 443 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 444 else 445 nth->th_win = htons((u_short)win); 446 nth->th_urp = 0; 447#ifdef INET6 448 if (isipv6) { 449 nth->th_sum = 0; 450 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 451 sizeof(struct ip6_hdr), 452 tlen - sizeof(struct ip6_hdr)); 453 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 454 ro6 && ro6->ro_rt ? 455 ro6->ro_rt->rt_ifp : 456 NULL); 457 } else 458#endif /* INET6 */ 459 { 460 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 461 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 462 m->m_pkthdr.csum_flags = CSUM_TCP; 463 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 464 } 465#ifdef TCPDEBUG 466 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 467 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 468#endif 469#ifdef IPSEC 470 if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { 471 m_freem(m); 472 return; 473 } 474#endif 475#ifdef INET6 476 if (isipv6) { 477 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL); 478 if (ro6 == &sro6 && ro6->ro_rt) { 479 RTFREE(ro6->ro_rt); 480 ro6->ro_rt = NULL; 481 } 482 } else 483#endif /* INET6 */ 484 { 485 (void) ip_output(m, NULL, ro, ipflags, NULL); 486 if (ro == &sro && ro->ro_rt) { 487 RTFREE(ro->ro_rt); 488 ro->ro_rt = NULL; 489 } 490 } 491} 492 493/* 494 * Create a new TCP control block, making an 495 * empty reassembly queue and hooking it to the argument 496 * protocol control block. The `inp' parameter must have 497 * come from the zone allocator set up in tcp_init(). 498 */ 499struct tcpcb * 500tcp_newtcpcb(inp) 501 struct inpcb *inp; 502{ 503 struct inp_tp *it; 504 register struct tcpcb *tp; 505#ifdef INET6 506 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 507#endif /* INET6 */ 508 509 it = (struct inp_tp *)inp; 510 tp = &it->tcb; 511 bzero((char *) tp, sizeof(struct tcpcb)); 512 LIST_INIT(&tp->t_segq); 513 tp->t_maxseg = tp->t_maxopd = 514#ifdef INET6 515 isipv6 ? tcp_v6mssdflt : 516#endif /* INET6 */ 517 tcp_mssdflt; 518 519 /* Set up our timeouts. */ 520 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0); 521 callout_init(tp->tt_persist = &it->inp_tp_persist, 0); 522 callout_init(tp->tt_keep = &it->inp_tp_keep, 0); 523 callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0); 524 callout_init(tp->tt_delack = &it->inp_tp_delack, 0); 525 526 if (tcp_do_rfc1323) 527 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 528 if (tcp_do_rfc1644) 529 tp->t_flags |= TF_REQ_CC; 530 tp->t_inpcb = inp; /* XXX */ 531 /* 532 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 533 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 534 * reasonable initial retransmit time. 535 */ 536 tp->t_srtt = TCPTV_SRTTBASE; 537 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 538 tp->t_rttmin = TCPTV_MIN; 539 tp->t_rxtcur = TCPTV_RTOBASE; 540 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 541 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 542 tp->t_rcvtime = ticks; 543 /* 544 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 545 * because the socket may be bound to an IPv6 wildcard address, 546 * which may match an IPv4-mapped IPv6 address. 547 */ 548 inp->inp_ip_ttl = ip_defttl; 549 inp->inp_ppcb = (caddr_t)tp; 550 return (tp); /* XXX */ 551} 552 553/* 554 * Drop a TCP connection, reporting 555 * the specified error. If connection is synchronized, 556 * then send a RST to peer. 557 */ 558struct tcpcb * 559tcp_drop(tp, errno) 560 register struct tcpcb *tp; 561 int errno; 562{ 563 struct socket *so = tp->t_inpcb->inp_socket; 564 565 if (TCPS_HAVERCVDSYN(tp->t_state)) { 566 tp->t_state = TCPS_CLOSED; 567 (void) tcp_output(tp); 568 tcpstat.tcps_drops++; 569 } else 570 tcpstat.tcps_conndrops++; 571 if (errno == ETIMEDOUT && tp->t_softerror) 572 errno = tp->t_softerror; 573 so->so_error = errno; 574 return (tcp_close(tp)); 575} 576 577/* 578 * Close a TCP control block: 579 * discard all space held by the tcp 580 * discard internet protocol block 581 * wake up any sleepers 582 */ 583struct tcpcb * 584tcp_close(tp) 585 register struct tcpcb *tp; 586{ 587 register struct tseg_qent *q; 588 struct inpcb *inp = tp->t_inpcb; 589 struct socket *so = inp->inp_socket; 590#ifdef INET6 591 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 592#endif /* INET6 */ 593 register struct rtentry *rt; 594 int dosavessthresh; 595 596 /* 597 * Make sure that all of our timers are stopped before we 598 * delete the PCB. 599 */ 600 callout_stop(tp->tt_rexmt); 601 callout_stop(tp->tt_persist); 602 callout_stop(tp->tt_keep); 603 callout_stop(tp->tt_2msl); 604 callout_stop(tp->tt_delack); 605 606 /* 607 * If we got enough samples through the srtt filter, 608 * save the rtt and rttvar in the routing entry. 609 * 'Enough' is arbitrarily defined as the 16 samples. 610 * 16 samples is enough for the srtt filter to converge 611 * to within 5% of the correct value; fewer samples and 612 * we could save a very bogus rtt. 613 * 614 * Don't update the default route's characteristics and don't 615 * update anything that the user "locked". 616 */ 617 if (tp->t_rttupdated >= 16) { 618 register u_long i = 0; 619#ifdef INET6 620 if (isipv6) { 621 struct sockaddr_in6 *sin6; 622 623 if ((rt = inp->in6p_route.ro_rt) == NULL) 624 goto no_valid_rt; 625 sin6 = (struct sockaddr_in6 *)rt_key(rt); 626 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 627 goto no_valid_rt; 628 } 629 else 630#endif /* INET6 */ 631 if ((rt = inp->inp_route.ro_rt) == NULL || 632 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr 633 == INADDR_ANY) 634 goto no_valid_rt; 635 636 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 637 i = tp->t_srtt * 638 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 639 if (rt->rt_rmx.rmx_rtt && i) 640 /* 641 * filter this update to half the old & half 642 * the new values, converting scale. 643 * See route.h and tcp_var.h for a 644 * description of the scaling constants. 645 */ 646 rt->rt_rmx.rmx_rtt = 647 (rt->rt_rmx.rmx_rtt + i) / 2; 648 else 649 rt->rt_rmx.rmx_rtt = i; 650 tcpstat.tcps_cachedrtt++; 651 } 652 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 653 i = tp->t_rttvar * 654 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 655 if (rt->rt_rmx.rmx_rttvar && i) 656 rt->rt_rmx.rmx_rttvar = 657 (rt->rt_rmx.rmx_rttvar + i) / 2; 658 else 659 rt->rt_rmx.rmx_rttvar = i; 660 tcpstat.tcps_cachedrttvar++; 661 } 662 /* 663 * The old comment here said: 664 * update the pipelimit (ssthresh) if it has been updated 665 * already or if a pipesize was specified & the threshhold 666 * got below half the pipesize. I.e., wait for bad news 667 * before we start updating, then update on both good 668 * and bad news. 669 * 670 * But we want to save the ssthresh even if no pipesize is 671 * specified explicitly in the route, because such 672 * connections still have an implicit pipesize specified 673 * by the global tcp_sendspace. In the absence of a reliable 674 * way to calculate the pipesize, it will have to do. 675 */ 676 i = tp->snd_ssthresh; 677 if (rt->rt_rmx.rmx_sendpipe != 0) 678 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 679 else 680 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 681 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 682 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 683 || dosavessthresh) { 684 /* 685 * convert the limit from user data bytes to 686 * packets then to packet data bytes. 687 */ 688 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 689 if (i < 2) 690 i = 2; 691 i *= (u_long)(tp->t_maxseg + 692#ifdef INET6 693 (isipv6 ? sizeof (struct ip6_hdr) + 694 sizeof (struct tcphdr) : 695#endif 696 sizeof (struct tcpiphdr) 697#ifdef INET6 698 ) 699#endif 700 ); 701 if (rt->rt_rmx.rmx_ssthresh) 702 rt->rt_rmx.rmx_ssthresh = 703 (rt->rt_rmx.rmx_ssthresh + i) / 2; 704 else 705 rt->rt_rmx.rmx_ssthresh = i; 706 tcpstat.tcps_cachedssthresh++; 707 } 708 } 709 rt = inp->inp_route.ro_rt; 710 if (rt) { 711 /* 712 * mark route for deletion if no information is 713 * cached. 714 */ 715 if ((tp->t_flags & TF_LQ_OVERFLOW) && 716 ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){ 717 if (rt->rt_rmx.rmx_rtt == 0) 718 rt->rt_flags |= RTF_DELCLONE; 719 } 720 } 721 no_valid_rt: 722 /* free the reassembly queue, if any */ 723 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 724 LIST_REMOVE(q, tqe_q); 725 m_freem(q->tqe_m); 726 FREE(q, M_TSEGQ); 727 } 728 inp->inp_ppcb = NULL; 729 soisdisconnected(so); 730#ifdef INET6 731 if (INP_CHECK_SOCKAF(so, AF_INET6)) 732 in6_pcbdetach(inp); 733 else 734#endif /* INET6 */ 735 in_pcbdetach(inp); 736 tcpstat.tcps_closed++; 737 return ((struct tcpcb *)0); 738} 739 740void 741tcp_drain() 742{ 743 if (do_tcpdrain) 744 { 745 struct inpcb *inpb; 746 struct tcpcb *tcpb; 747 struct tseg_qent *te; 748 749 /* 750 * Walk the tcpbs, if existing, and flush the reassembly queue, 751 * if there is one... 752 * XXX: The "Net/3" implementation doesn't imply that the TCP 753 * reassembly queue should be flushed, but in a situation 754 * where we're really low on mbufs, this is potentially 755 * usefull. 756 */ 757 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 758 if ((tcpb = intotcpcb(inpb))) { 759 while ((te = LIST_FIRST(&tcpb->t_segq)) 760 != NULL) { 761 LIST_REMOVE(te, tqe_q); 762 m_freem(te->tqe_m); 763 FREE(te, M_TSEGQ); 764 } 765 } 766 } 767 } 768} 769 770/* 771 * Notify a tcp user of an asynchronous error; 772 * store error as soft error, but wake up user 773 * (for now, won't do anything until can select for soft error). 774 * 775 * Do not wake up user since there currently is no mechanism for 776 * reporting soft errors (yet - a kqueue filter may be added). 777 */ 778static void 779tcp_notify(inp, error) 780 struct inpcb *inp; 781 int error; 782{ 783 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 784 785 /* 786 * Ignore some errors if we are hooked up. 787 * If connection hasn't completed, has retransmitted several times, 788 * and receives a second error, give up now. This is better 789 * than waiting a long time to establish a connection that 790 * can never complete. 791 */ 792 if (tp->t_state == TCPS_ESTABLISHED && 793 (error == EHOSTUNREACH || error == ENETUNREACH || 794 error == EHOSTDOWN)) { 795 return; 796 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 797 tp->t_softerror) 798 tcp_drop(tp, error); 799 else 800 tp->t_softerror = error; 801#if 0 802 wakeup((caddr_t) &so->so_timeo); 803 sorwakeup(so); 804 sowwakeup(so); 805#endif 806} 807 808static int 809tcp_pcblist(SYSCTL_HANDLER_ARGS) 810{ 811 int error, i, n, s; 812 struct inpcb *inp, **inp_list; 813 inp_gen_t gencnt; 814 struct xinpgen xig; 815 816 /* 817 * The process of preparing the TCB list is too time-consuming and 818 * resource-intensive to repeat twice on every request. 819 */ 820 if (req->oldptr == 0) { 821 n = tcbinfo.ipi_count; 822 req->oldidx = 2 * (sizeof xig) 823 + (n + n/8) * sizeof(struct xtcpcb); 824 return 0; 825 } 826 827 if (req->newptr != 0) 828 return EPERM; 829 830 /* 831 * OK, now we're committed to doing something. 832 */ 833 s = splnet(); 834 gencnt = tcbinfo.ipi_gencnt; 835 n = tcbinfo.ipi_count; 836 splx(s); 837 838 xig.xig_len = sizeof xig; 839 xig.xig_count = n; 840 xig.xig_gen = gencnt; 841 xig.xig_sogen = so_gencnt; 842 error = SYSCTL_OUT(req, &xig, sizeof xig); 843 if (error) 844 return error; 845 846 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 847 if (inp_list == 0) 848 return ENOMEM; 849 850 s = splnet(); 851 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; 852 inp = LIST_NEXT(inp, inp_list)) { 853 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp)) 854 inp_list[i++] = inp; 855 } 856 splx(s); 857 n = i; 858 859 error = 0; 860 for (i = 0; i < n; i++) { 861 inp = inp_list[i]; 862 if (inp->inp_gencnt <= gencnt) { 863 struct xtcpcb xt; 864 caddr_t inp_ppcb; 865 xt.xt_len = sizeof xt; 866 /* XXX should avoid extra copy */ 867 bcopy(inp, &xt.xt_inp, sizeof *inp); 868 inp_ppcb = inp->inp_ppcb; 869 if (inp_ppcb != NULL) 870 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 871 else 872 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 873 if (inp->inp_socket) 874 sotoxsocket(inp->inp_socket, &xt.xt_socket); 875 error = SYSCTL_OUT(req, &xt, sizeof xt); 876 } 877 } 878 if (!error) { 879 /* 880 * Give the user an updated idea of our state. 881 * If the generation differs from what we told 882 * her before, she knows that something happened 883 * while we were processing this request, and it 884 * might be necessary to retry. 885 */ 886 s = splnet(); 887 xig.xig_gen = tcbinfo.ipi_gencnt; 888 xig.xig_sogen = so_gencnt; 889 xig.xig_count = tcbinfo.ipi_count; 890 splx(s); 891 error = SYSCTL_OUT(req, &xig, sizeof xig); 892 } 893 free(inp_list, M_TEMP); 894 return error; 895} 896 897SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 898 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 899 900static int 901tcp_getcred(SYSCTL_HANDLER_ARGS) 902{ 903 struct xucred xuc; 904 struct sockaddr_in addrs[2]; 905 struct inpcb *inp; 906 int error, s; 907 908 error = suser_xxx(0, req->p, PRISON_ROOT); 909 if (error) 910 return (error); 911 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 912 if (error) 913 return (error); 914 s = splnet(); 915 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 916 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 917 if (inp == NULL || inp->inp_socket == NULL) { 918 error = ENOENT; 919 goto out; 920 } 921 error = u_cansee(req->p->p_ucred, inp->inp_socket->so_cred); 922 if (error) 923 goto out; 924 bzero(&xuc, sizeof(xuc)); 925 xuc.cr_uid = inp->inp_socket->so_cred->cr_uid; 926 xuc.cr_ngroups = inp->inp_socket->so_cred->cr_ngroups; 927 bcopy(inp->inp_socket->so_cred->cr_groups, xuc.cr_groups, 928 sizeof(xuc.cr_groups)); 929 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 930out: 931 splx(s); 932 return (error); 933} 934 935SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 936 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 937 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 938 939#ifdef INET6 940static int 941tcp6_getcred(SYSCTL_HANDLER_ARGS) 942{ 943 struct xucred xuc; 944 struct sockaddr_in6 addrs[2]; 945 struct inpcb *inp; 946 int error, s, mapped = 0; 947 948 error = suser_xxx(0, req->p, PRISON_ROOT); 949 if (error) 950 return (error); 951 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 952 if (error) 953 return (error); 954 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 955 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 956 mapped = 1; 957 else 958 return (EINVAL); 959 } 960 s = splnet(); 961 if (mapped == 1) 962 inp = in_pcblookup_hash(&tcbinfo, 963 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 964 addrs[1].sin6_port, 965 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 966 addrs[0].sin6_port, 967 0, NULL); 968 else 969 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 970 addrs[1].sin6_port, 971 &addrs[0].sin6_addr, addrs[0].sin6_port, 972 0, NULL); 973 if (inp == NULL || inp->inp_socket == NULL) { 974 error = ENOENT; 975 goto out; 976 } 977 error = u_cansee(req->p->p_ucred, inp->inp_socket->so_cred); 978 if (error) 979 goto out; 980 bzero(&xuc, sizeof(xuc)); 981 xuc.cr_uid = inp->inp_socket->so_cred->cr_uid; 982 xuc.cr_ngroups = inp->inp_socket->so_cred->cr_ngroups; 983 bcopy(inp->inp_socket->so_cred->cr_groups, xuc.cr_groups, 984 sizeof(xuc.cr_groups)); 985 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 986out: 987 splx(s); 988 return (error); 989} 990 991SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 992 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 993 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 994#endif 995 996 997void 998tcp_ctlinput(cmd, sa, vip) 999 int cmd; 1000 struct sockaddr *sa; 1001 void *vip; 1002{ 1003 struct ip *ip = vip; 1004 struct tcphdr *th; 1005 struct in_addr faddr; 1006 struct inpcb *inp; 1007 struct tcpcb *tp; 1008 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1009 tcp_seq icmp_seq; 1010 int s; 1011 1012 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1013 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1014 return; 1015 1016 if (cmd == PRC_QUENCH) 1017 notify = tcp_quench; 1018 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1019 cmd == PRC_UNREACH_PORT) && ip) 1020 notify = tcp_drop_syn_sent; 1021 else if (cmd == PRC_MSGSIZE) 1022 notify = tcp_mtudisc; 1023 else if (PRC_IS_REDIRECT(cmd)) { 1024 ip = 0; 1025 notify = in_rtchange; 1026 } else if (cmd == PRC_HOSTDEAD) 1027 ip = 0; 1028 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1029 return; 1030 if (ip) { 1031 s = splnet(); 1032 th = (struct tcphdr *)((caddr_t)ip 1033 + (IP_VHL_HL(ip->ip_vhl) << 2)); 1034 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1035 ip->ip_src, th->th_sport, 0, NULL); 1036 if (inp != NULL && inp->inp_socket != NULL) { 1037 icmp_seq = htonl(th->th_seq); 1038 tp = intotcpcb(inp); 1039 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1040 SEQ_LT(icmp_seq, tp->snd_max)) 1041 (*notify)(inp, inetctlerrmap[cmd]); 1042 } 1043 splx(s); 1044 } else 1045 in_pcbnotifyall(&tcb, faddr, inetctlerrmap[cmd], notify); 1046} 1047 1048#ifdef INET6 1049void 1050tcp6_ctlinput(cmd, sa, d) 1051 int cmd; 1052 struct sockaddr *sa; 1053 void *d; 1054{ 1055 struct tcphdr th; 1056 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1057 struct ip6_hdr *ip6; 1058 struct mbuf *m; 1059 struct ip6ctlparam *ip6cp = NULL; 1060 const struct sockaddr_in6 *sa6_src = NULL; 1061 int off; 1062 struct tcp_portonly { 1063 u_int16_t th_sport; 1064 u_int16_t th_dport; 1065 } *thp; 1066 1067 if (sa->sa_family != AF_INET6 || 1068 sa->sa_len != sizeof(struct sockaddr_in6)) 1069 return; 1070 1071 if (cmd == PRC_QUENCH) 1072 notify = tcp_quench; 1073 else if (cmd == PRC_MSGSIZE) 1074 notify = tcp_mtudisc; 1075 else if (!PRC_IS_REDIRECT(cmd) && 1076 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1077 return; 1078 1079 /* if the parameter is from icmp6, decode it. */ 1080 if (d != NULL) { 1081 ip6cp = (struct ip6ctlparam *)d; 1082 m = ip6cp->ip6c_m; 1083 ip6 = ip6cp->ip6c_ip6; 1084 off = ip6cp->ip6c_off; 1085 sa6_src = ip6cp->ip6c_src; 1086 } else { 1087 m = NULL; 1088 ip6 = NULL; 1089 off = 0; /* fool gcc */ 1090 sa6_src = &sa6_any; 1091 } 1092 1093 if (ip6) { 1094 /* 1095 * XXX: We assume that when IPV6 is non NULL, 1096 * M and OFF are valid. 1097 */ 1098 1099 /* check if we can safely examine src and dst ports */ 1100 if (m->m_pkthdr.len < off + sizeof(*thp)) 1101 return; 1102 1103 bzero(&th, sizeof(th)); 1104 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1105 1106 in6_pcbnotify(&tcb, sa, th.th_dport, 1107 (struct sockaddr *)ip6cp->ip6c_src, 1108 th.th_sport, cmd, notify); 1109 } else 1110 in6_pcbnotify(&tcb, sa, 0, (struct sockaddr *)sa6_src, 1111 0, cmd, notify); 1112} 1113#endif /* INET6 */ 1114 1115tcp_seq 1116tcp_new_isn() 1117{ 1118 1119 if (tcp_seq_genscheme > 1 || tcp_seq_genscheme < 0) 1120 tcp_seq_genscheme = 1; 1121 1122 switch (tcp_seq_genscheme) { 1123 case 0: /* Random positive increments */ 1124 tcp_iss += TCP_ISSINCR/2; 1125 return tcp_iss; 1126 case 1: /* OpenBSD randomized scheme */ 1127 return tcp_rndiss_next(); 1128 } 1129} 1130 1131#define TCP_RNDISS_ROUNDS 16 1132#define TCP_RNDISS_OUT 7200 1133#define TCP_RNDISS_MAX 30000 1134 1135u_int8_t tcp_rndiss_sbox[128]; 1136u_int16_t tcp_rndiss_msb; 1137u_int16_t tcp_rndiss_cnt; 1138long tcp_rndiss_reseed; 1139 1140u_int16_t 1141tcp_rndiss_encrypt(val) 1142 u_int16_t val; 1143{ 1144 u_int16_t sum = 0, i; 1145 1146 for (i = 0; i < TCP_RNDISS_ROUNDS; i++) { 1147 sum += 0x79b9; 1148 val ^= ((u_int16_t)tcp_rndiss_sbox[(val^sum) & 0x7f]) << 7; 1149 val = ((val & 0xff) << 7) | (val >> 8); 1150 } 1151 1152 return val; 1153} 1154 1155void 1156tcp_rndiss_init() 1157{ 1158 struct timeval time; 1159 1160 getmicrotime(&time); 1161 read_random(tcp_rndiss_sbox, sizeof(tcp_rndiss_sbox)); 1162 1163 tcp_rndiss_reseed = time.tv_sec + TCP_RNDISS_OUT; 1164 tcp_rndiss_msb = tcp_rndiss_msb == 0x8000 ? 0 : 0x8000; 1165 tcp_rndiss_cnt = 0; 1166} 1167 1168tcp_seq 1169tcp_rndiss_next() 1170{ 1171 u_int16_t tmp; 1172 struct timeval time; 1173 1174 getmicrotime(&time); 1175 1176 if (tcp_rndiss_cnt >= TCP_RNDISS_MAX || 1177 time.tv_sec > tcp_rndiss_reseed) 1178 tcp_rndiss_init(); 1179 1180 read_random(&tmp, sizeof(tmp)); 1181 1182 /* (tmp & 0x7fff) ensures a 32768 byte gap between ISS */ 1183 return ((tcp_rndiss_encrypt(tcp_rndiss_cnt++) | tcp_rndiss_msb) <<16) | 1184 (tmp & 0x7fff); 1185} 1186 1187 1188/* 1189 * When a source quench is received, close congestion window 1190 * to one segment. We will gradually open it again as we proceed. 1191 */ 1192void 1193tcp_quench(inp, errno) 1194 struct inpcb *inp; 1195 int errno; 1196{ 1197 struct tcpcb *tp = intotcpcb(inp); 1198 1199 if (tp) 1200 tp->snd_cwnd = tp->t_maxseg; 1201} 1202 1203/* 1204 * When a specific ICMP unreachable message is received and the 1205 * connection state is SYN-SENT, drop the connection. This behavior 1206 * is controlled by the icmp_may_rst sysctl. 1207 */ 1208void 1209tcp_drop_syn_sent(inp, errno) 1210 struct inpcb *inp; 1211 int errno; 1212{ 1213 struct tcpcb *tp = intotcpcb(inp); 1214 1215 if (tp && tp->t_state == TCPS_SYN_SENT) 1216 tcp_drop(tp, errno); 1217} 1218 1219/* 1220 * When `need fragmentation' ICMP is received, update our idea of the MSS 1221 * based on the new value in the route. Also nudge TCP to send something, 1222 * since we know the packet we just sent was dropped. 1223 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1224 */ 1225void 1226tcp_mtudisc(inp, errno) 1227 struct inpcb *inp; 1228 int errno; 1229{ 1230 struct tcpcb *tp = intotcpcb(inp); 1231 struct rtentry *rt; 1232 struct rmxp_tao *taop; 1233 struct socket *so = inp->inp_socket; 1234 int offered; 1235 int mss; 1236#ifdef INET6 1237 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1238#endif /* INET6 */ 1239 1240 if (tp) { 1241#ifdef INET6 1242 if (isipv6) 1243 rt = tcp_rtlookup6(inp); 1244 else 1245#endif /* INET6 */ 1246 rt = tcp_rtlookup(inp); 1247 if (!rt || !rt->rt_rmx.rmx_mtu) { 1248 tp->t_maxopd = tp->t_maxseg = 1249#ifdef INET6 1250 isipv6 ? tcp_v6mssdflt : 1251#endif /* INET6 */ 1252 tcp_mssdflt; 1253 return; 1254 } 1255 taop = rmx_taop(rt->rt_rmx); 1256 offered = taop->tao_mssopt; 1257 mss = rt->rt_rmx.rmx_mtu - 1258#ifdef INET6 1259 (isipv6 ? 1260 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1261#endif /* INET6 */ 1262 sizeof(struct tcpiphdr) 1263#ifdef INET6 1264 ) 1265#endif /* INET6 */ 1266 ; 1267 1268 if (offered) 1269 mss = min(mss, offered); 1270 /* 1271 * XXX - The above conditional probably violates the TCP 1272 * spec. The problem is that, since we don't know the 1273 * other end's MSS, we are supposed to use a conservative 1274 * default. But, if we do that, then MTU discovery will 1275 * never actually take place, because the conservative 1276 * default is much less than the MTUs typically seen 1277 * on the Internet today. For the moment, we'll sweep 1278 * this under the carpet. 1279 * 1280 * The conservative default might not actually be a problem 1281 * if the only case this occurs is when sending an initial 1282 * SYN with options and data to a host we've never talked 1283 * to before. Then, they will reply with an MSS value which 1284 * will get recorded and the new parameters should get 1285 * recomputed. For Further Study. 1286 */ 1287 if (tp->t_maxopd <= mss) 1288 return; 1289 tp->t_maxopd = mss; 1290 1291 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1292 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1293 mss -= TCPOLEN_TSTAMP_APPA; 1294 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1295 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1296 mss -= TCPOLEN_CC_APPA; 1297#if (MCLBYTES & (MCLBYTES - 1)) == 0 1298 if (mss > MCLBYTES) 1299 mss &= ~(MCLBYTES-1); 1300#else 1301 if (mss > MCLBYTES) 1302 mss = mss / MCLBYTES * MCLBYTES; 1303#endif 1304 if (so->so_snd.sb_hiwat < mss) 1305 mss = so->so_snd.sb_hiwat; 1306 1307 tp->t_maxseg = mss; 1308 1309 tcpstat.tcps_mturesent++; 1310 tp->t_rtttime = 0; 1311 tp->snd_nxt = tp->snd_una; 1312 tcp_output(tp); 1313 } 1314} 1315 1316/* 1317 * Look-up the routing entry to the peer of this inpcb. If no route 1318 * is found and it cannot be allocated the return NULL. This routine 1319 * is called by TCP routines that access the rmx structure and by tcp_mss 1320 * to get the interface MTU. 1321 */ 1322struct rtentry * 1323tcp_rtlookup(inp) 1324 struct inpcb *inp; 1325{ 1326 struct route *ro; 1327 struct rtentry *rt; 1328 1329 ro = &inp->inp_route; 1330 rt = ro->ro_rt; 1331 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1332 /* No route yet, so try to acquire one */ 1333 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1334 ro->ro_dst.sa_family = AF_INET; 1335 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1336 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1337 inp->inp_faddr; 1338 rtalloc(ro); 1339 rt = ro->ro_rt; 1340 } 1341 } 1342 return rt; 1343} 1344 1345#ifdef INET6 1346struct rtentry * 1347tcp_rtlookup6(inp) 1348 struct inpcb *inp; 1349{ 1350 struct route_in6 *ro6; 1351 struct rtentry *rt; 1352 1353 ro6 = &inp->in6p_route; 1354 rt = ro6->ro_rt; 1355 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1356 /* No route yet, so try to acquire one */ 1357 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) { 1358 struct sockaddr_in6 *dst6; 1359 1360 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst; 1361 dst6->sin6_family = AF_INET6; 1362 dst6->sin6_len = sizeof(*dst6); 1363 dst6->sin6_addr = inp->in6p_faddr; 1364 rtalloc((struct route *)ro6); 1365 rt = ro6->ro_rt; 1366 } 1367 } 1368 return rt; 1369} 1370#endif /* INET6 */ 1371 1372#ifdef IPSEC 1373/* compute ESP/AH header size for TCP, including outer IP header. */ 1374size_t 1375ipsec_hdrsiz_tcp(tp) 1376 struct tcpcb *tp; 1377{ 1378 struct inpcb *inp; 1379 struct mbuf *m; 1380 size_t hdrsiz; 1381 struct ip *ip; 1382#ifdef INET6 1383 struct ip6_hdr *ip6; 1384#endif /* INET6 */ 1385 struct tcphdr *th; 1386 1387 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1388 return 0; 1389 MGETHDR(m, M_DONTWAIT, MT_DATA); 1390 if (!m) 1391 return 0; 1392 1393#ifdef INET6 1394 if ((inp->inp_vflag & INP_IPV6) != 0) { 1395 ip6 = mtod(m, struct ip6_hdr *); 1396 th = (struct tcphdr *)(ip6 + 1); 1397 m->m_pkthdr.len = m->m_len = 1398 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1399 tcp_fillheaders(tp, ip6, th); 1400 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1401 } else 1402#endif /* INET6 */ 1403 { 1404 ip = mtod(m, struct ip *); 1405 th = (struct tcphdr *)(ip + 1); 1406 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1407 tcp_fillheaders(tp, ip, th); 1408 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1409 } 1410 1411 m_free(m); 1412 return hdrsiz; 1413} 1414#endif /*IPSEC*/ 1415 1416/* 1417 * Return a pointer to the cached information about the remote host. 1418 * The cached information is stored in the protocol specific part of 1419 * the route metrics. 1420 */ 1421struct rmxp_tao * 1422tcp_gettaocache(inp) 1423 struct inpcb *inp; 1424{ 1425 struct rtentry *rt; 1426 1427#ifdef INET6 1428 if ((inp->inp_vflag & INP_IPV6) != 0) 1429 rt = tcp_rtlookup6(inp); 1430 else 1431#endif /* INET6 */ 1432 rt = tcp_rtlookup(inp); 1433 1434 /* Make sure this is a host route and is up. */ 1435 if (rt == NULL || 1436 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 1437 return NULL; 1438 1439 return rmx_taop(rt->rt_rmx); 1440} 1441 1442/* 1443 * Clear all the TAO cache entries, called from tcp_init. 1444 * 1445 * XXX 1446 * This routine is just an empty one, because we assume that the routing 1447 * routing tables are initialized at the same time when TCP, so there is 1448 * nothing in the cache left over. 1449 */ 1450static void 1451tcp_cleartaocache() 1452{ 1453} 1454