tcp_timewait.c revision 96972
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 96972 2002-05-20 05:41:09Z tanimura $ 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/uma.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#include <sys/md5.h> 102 103int tcp_mssdflt = TCP_MSS; 104SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 105 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 106 107#ifdef INET6 108int tcp_v6mssdflt = TCP6_MSS; 109SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 110 CTLFLAG_RW, &tcp_v6mssdflt , 0, 111 "Default TCP Maximum Segment Size for IPv6"); 112#endif 113 114#if 0 115static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 116SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 117 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 118#endif 119 120int tcp_do_rfc1323 = 1; 121SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 122 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 123 124int tcp_do_rfc1644 = 0; 125SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 126 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 127 128static int tcp_tcbhashsize = 0; 129SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 130 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 131 132static int do_tcpdrain = 1; 133SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 134 "Enable tcp_drain routine for extra help when low on mbufs"); 135 136SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 137 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 138 139static int icmp_may_rst = 1; 140SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 141 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 142 143static int tcp_isn_reseed_interval = 0; 144SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 145 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 146 147static void tcp_cleartaocache(void); 148static void tcp_notify(struct inpcb *, int); 149 150/* 151 * Target size of TCP PCB hash tables. Must be a power of two. 152 * 153 * Note that this can be overridden by the kernel environment 154 * variable net.inet.tcp.tcbhashsize 155 */ 156#ifndef TCBHASHSIZE 157#define TCBHASHSIZE 512 158#endif 159 160/* 161 * This is the actual shape of what we allocate using the zone 162 * allocator. Doing it this way allows us to protect both structures 163 * using the same generation count, and also eliminates the overhead 164 * of allocating tcpcbs separately. By hiding the structure here, 165 * we avoid changing most of the rest of the code (although it needs 166 * to be changed, eventually, for greater efficiency). 167 */ 168#define ALIGNMENT 32 169#define ALIGNM1 (ALIGNMENT - 1) 170struct inp_tp { 171 union { 172 struct inpcb inp; 173 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 174 } inp_tp_u; 175 struct tcpcb tcb; 176 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 177 struct callout inp_tp_delack; 178}; 179#undef ALIGNMENT 180#undef ALIGNM1 181 182/* 183 * Tcp initialization 184 */ 185void 186tcp_init() 187{ 188 int hashsize = TCBHASHSIZE; 189 190 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 = uma_zcreate("tcpcb", sizeof(struct inp_tp), 212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 213 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 214#ifdef INET6 215#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 216#else /* INET6 */ 217#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 218#endif /* INET6 */ 219 if (max_protohdr < TCP_MINPROTOHDR) 220 max_protohdr = TCP_MINPROTOHDR; 221 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 222 panic("tcp_init"); 223#undef TCP_MINPROTOHDR 224 225 syncache_init(); 226} 227 228/* 229 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 230 * tcp_template used to store this data in mbufs, but we now recopy it out 231 * of the tcpcb each time to conserve mbufs. 232 */ 233void 234tcp_fillheaders(tp, ip_ptr, tcp_ptr) 235 struct tcpcb *tp; 236 void *ip_ptr; 237 void *tcp_ptr; 238{ 239 struct inpcb *inp = tp->t_inpcb; 240 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 241 242#ifdef INET6 243 if ((inp->inp_vflag & INP_IPV6) != 0) { 244 struct ip6_hdr *ip6; 245 246 ip6 = (struct ip6_hdr *)ip_ptr; 247 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 248 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 249 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 250 (IPV6_VERSION & IPV6_VERSION_MASK); 251 ip6->ip6_nxt = IPPROTO_TCP; 252 ip6->ip6_plen = sizeof(struct tcphdr); 253 ip6->ip6_src = inp->in6p_laddr; 254 ip6->ip6_dst = inp->in6p_faddr; 255 tcp_hdr->th_sum = 0; 256 } else 257#endif 258 { 259 struct ip *ip = (struct ip *) ip_ptr; 260 261 ip->ip_vhl = IP_VHL_BORING; 262 ip->ip_tos = 0; 263 ip->ip_len = 0; 264 ip->ip_id = 0; 265 ip->ip_off = 0; 266 ip->ip_ttl = 0; 267 ip->ip_sum = 0; 268 ip->ip_p = IPPROTO_TCP; 269 ip->ip_src = inp->inp_laddr; 270 ip->ip_dst = inp->inp_faddr; 271 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 272 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 273 } 274 275 tcp_hdr->th_sport = inp->inp_lport; 276 tcp_hdr->th_dport = inp->inp_fport; 277 tcp_hdr->th_seq = 0; 278 tcp_hdr->th_ack = 0; 279 tcp_hdr->th_x2 = 0; 280 tcp_hdr->th_off = 5; 281 tcp_hdr->th_flags = 0; 282 tcp_hdr->th_win = 0; 283 tcp_hdr->th_urp = 0; 284} 285 286/* 287 * Create template to be used to send tcp packets on a connection. 288 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 289 * use for this function is in keepalives, which use tcp_respond. 290 */ 291struct tcptemp * 292tcp_maketemplate(tp) 293 struct tcpcb *tp; 294{ 295 struct mbuf *m; 296 struct tcptemp *n; 297 298 m = m_get(M_DONTWAIT, MT_HEADER); 299 if (m == NULL) 300 return (0); 301 m->m_len = sizeof(struct tcptemp); 302 n = mtod(m, struct tcptemp *); 303 304 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 305 return (n); 306} 307 308/* 309 * Send a single message to the TCP at address specified by 310 * the given TCP/IP header. If m == 0, then we make a copy 311 * of the tcpiphdr at ti and send directly to the addressed host. 312 * This is used to force keep alive messages out using the TCP 313 * template for a connection. If flags are given then we send 314 * a message back to the TCP which originated the * segment ti, 315 * and discard the mbuf containing it and any other attached mbufs. 316 * 317 * In any case the ack and sequence number of the transmitted 318 * segment are as specified by the parameters. 319 * 320 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 321 */ 322void 323tcp_respond(tp, ipgen, th, m, ack, seq, flags) 324 struct tcpcb *tp; 325 void *ipgen; 326 register struct tcphdr *th; 327 register struct mbuf *m; 328 tcp_seq ack, seq; 329 int flags; 330{ 331 register int tlen; 332 int win = 0; 333 struct route *ro = 0; 334 struct route sro; 335 struct ip *ip; 336 struct tcphdr *nth; 337#ifdef INET6 338 struct route_in6 *ro6 = 0; 339 struct route_in6 sro6; 340 struct ip6_hdr *ip6; 341 int isipv6; 342#endif /* INET6 */ 343 int ipflags = 0; 344 345#ifdef INET6 346 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; 347 ip6 = ipgen; 348#endif /* INET6 */ 349 ip = ipgen; 350 351 if (tp) { 352 if (!(flags & TH_RST)) { 353 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 354 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 355 win = (long)TCP_MAXWIN << tp->rcv_scale; 356 } 357#ifdef INET6 358 if (isipv6) 359 ro6 = &tp->t_inpcb->in6p_route; 360 else 361#endif /* INET6 */ 362 ro = &tp->t_inpcb->inp_route; 363 } else { 364#ifdef INET6 365 if (isipv6) { 366 ro6 = &sro6; 367 bzero(ro6, sizeof *ro6); 368 } else 369#endif /* INET6 */ 370 { 371 ro = &sro; 372 bzero(ro, sizeof *ro); 373 } 374 } 375 if (m == 0) { 376 m = m_gethdr(M_DONTWAIT, MT_HEADER); 377 if (m == NULL) 378 return; 379 tlen = 0; 380 m->m_data += max_linkhdr; 381#ifdef INET6 382 if (isipv6) { 383 bcopy((caddr_t)ip6, mtod(m, caddr_t), 384 sizeof(struct ip6_hdr)); 385 ip6 = mtod(m, struct ip6_hdr *); 386 nth = (struct tcphdr *)(ip6 + 1); 387 } else 388#endif /* INET6 */ 389 { 390 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 391 ip = mtod(m, struct ip *); 392 nth = (struct tcphdr *)(ip + 1); 393 } 394 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 395 flags = TH_ACK; 396 } else { 397 m_freem(m->m_next); 398 m->m_next = 0; 399 m->m_data = (caddr_t)ipgen; 400 /* m_len is set later */ 401 tlen = 0; 402#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 403#ifdef INET6 404 if (isipv6) { 405 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 406 nth = (struct tcphdr *)(ip6 + 1); 407 } else 408#endif /* INET6 */ 409 { 410 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 411 nth = (struct tcphdr *)(ip + 1); 412 } 413 if (th != nth) { 414 /* 415 * this is usually a case when an extension header 416 * exists between the IPv6 header and the 417 * TCP header. 418 */ 419 nth->th_sport = th->th_sport; 420 nth->th_dport = th->th_dport; 421 } 422 xchg(nth->th_dport, nth->th_sport, n_short); 423#undef xchg 424 } 425#ifdef INET6 426 if (isipv6) { 427 ip6->ip6_flow = 0; 428 ip6->ip6_vfc = IPV6_VERSION; 429 ip6->ip6_nxt = IPPROTO_TCP; 430 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 431 tlen)); 432 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 433 } else 434#endif 435 { 436 tlen += sizeof (struct tcpiphdr); 437 ip->ip_len = tlen; 438 ip->ip_ttl = ip_defttl; 439 } 440 m->m_len = tlen; 441 m->m_pkthdr.len = tlen; 442 m->m_pkthdr.rcvif = (struct ifnet *) 0; 443 nth->th_seq = htonl(seq); 444 nth->th_ack = htonl(ack); 445 nth->th_x2 = 0; 446 nth->th_off = sizeof (struct tcphdr) >> 2; 447 nth->th_flags = flags; 448 if (tp) 449 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 450 else 451 nth->th_win = htons((u_short)win); 452 nth->th_urp = 0; 453#ifdef INET6 454 if (isipv6) { 455 nth->th_sum = 0; 456 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 457 sizeof(struct ip6_hdr), 458 tlen - sizeof(struct ip6_hdr)); 459 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 460 ro6 && ro6->ro_rt ? 461 ro6->ro_rt->rt_ifp : 462 NULL); 463 } else 464#endif /* INET6 */ 465 { 466 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 467 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 468 m->m_pkthdr.csum_flags = CSUM_TCP; 469 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 470 } 471#ifdef TCPDEBUG 472 if (tp == NULL) 473 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 474 else { 475 SOCK_LOCK(tp->t_inpcb->inp_socket); 476 if ((tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { 477 SOCK_UNLOCK(tp->t_inpcb->inp_socket); 478 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 479 } else 480 SOCK_UNLOCK(tp->t_inpcb->inp_socket); 481 } 482#endif 483#ifdef IPSEC 484 if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { 485 m_freem(m); 486 return; 487 } 488#endif 489#ifdef INET6 490 if (isipv6) { 491 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL); 492 if (ro6 == &sro6 && ro6->ro_rt) { 493 RTFREE(ro6->ro_rt); 494 ro6->ro_rt = NULL; 495 } 496 } else 497#endif /* INET6 */ 498 { 499 (void) ip_output(m, NULL, ro, ipflags, NULL); 500 if (ro == &sro && ro->ro_rt) { 501 RTFREE(ro->ro_rt); 502 ro->ro_rt = NULL; 503 } 504 } 505} 506 507/* 508 * Create a new TCP control block, making an 509 * empty reassembly queue and hooking it to the argument 510 * protocol control block. The `inp' parameter must have 511 * come from the zone allocator set up in tcp_init(). 512 */ 513struct tcpcb * 514tcp_newtcpcb(inp) 515 struct inpcb *inp; 516{ 517 struct inp_tp *it; 518 register struct tcpcb *tp; 519#ifdef INET6 520 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 521#endif /* INET6 */ 522 523 it = (struct inp_tp *)inp; 524 tp = &it->tcb; 525 bzero((char *) tp, sizeof(struct tcpcb)); 526 LIST_INIT(&tp->t_segq); 527 tp->t_maxseg = tp->t_maxopd = 528#ifdef INET6 529 isipv6 ? tcp_v6mssdflt : 530#endif /* INET6 */ 531 tcp_mssdflt; 532 533 /* Set up our timeouts. */ 534 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0); 535 callout_init(tp->tt_persist = &it->inp_tp_persist, 0); 536 callout_init(tp->tt_keep = &it->inp_tp_keep, 0); 537 callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0); 538 callout_init(tp->tt_delack = &it->inp_tp_delack, 0); 539 540 if (tcp_do_rfc1323) 541 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 542 if (tcp_do_rfc1644) 543 tp->t_flags |= TF_REQ_CC; 544 tp->t_inpcb = inp; /* XXX */ 545 /* 546 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 547 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 548 * reasonable initial retransmit time. 549 */ 550 tp->t_srtt = TCPTV_SRTTBASE; 551 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 552 tp->t_rttmin = TCPTV_MIN; 553 tp->t_rxtcur = TCPTV_RTOBASE; 554 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 555 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 556 tp->t_rcvtime = ticks; 557 /* 558 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 559 * because the socket may be bound to an IPv6 wildcard address, 560 * which may match an IPv4-mapped IPv6 address. 561 */ 562 inp->inp_ip_ttl = ip_defttl; 563 inp->inp_ppcb = (caddr_t)tp; 564 return (tp); /* XXX */ 565} 566 567/* 568 * Drop a TCP connection, reporting 569 * the specified error. If connection is synchronized, 570 * then send a RST to peer. 571 */ 572struct tcpcb * 573tcp_drop(tp, errno) 574 register struct tcpcb *tp; 575 int errno; 576{ 577 struct socket *so = tp->t_inpcb->inp_socket; 578 579 if (TCPS_HAVERCVDSYN(tp->t_state)) { 580 tp->t_state = TCPS_CLOSED; 581 (void) tcp_output(tp); 582 tcpstat.tcps_drops++; 583 } else 584 tcpstat.tcps_conndrops++; 585 if (errno == ETIMEDOUT && tp->t_softerror) 586 errno = tp->t_softerror; 587 so->so_error = errno; 588 return (tcp_close(tp)); 589} 590 591/* 592 * Close a TCP control block: 593 * discard all space held by the tcp 594 * discard internet protocol block 595 * wake up any sleepers 596 */ 597struct tcpcb * 598tcp_close(tp) 599 register struct tcpcb *tp; 600{ 601 register struct tseg_qent *q; 602 struct inpcb *inp = tp->t_inpcb; 603 struct socket *so = inp->inp_socket; 604#ifdef INET6 605 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 606#endif /* INET6 */ 607 register struct rtentry *rt; 608 int dosavessthresh; 609 610 /* 611 * Make sure that all of our timers are stopped before we 612 * delete the PCB. 613 */ 614 callout_stop(tp->tt_rexmt); 615 callout_stop(tp->tt_persist); 616 callout_stop(tp->tt_keep); 617 callout_stop(tp->tt_2msl); 618 callout_stop(tp->tt_delack); 619 620 /* 621 * If we got enough samples through the srtt filter, 622 * save the rtt and rttvar in the routing entry. 623 * 'Enough' is arbitrarily defined as the 16 samples. 624 * 16 samples is enough for the srtt filter to converge 625 * to within 5% of the correct value; fewer samples and 626 * we could save a very bogus rtt. 627 * 628 * Don't update the default route's characteristics and don't 629 * update anything that the user "locked". 630 */ 631 if (tp->t_rttupdated >= 16) { 632 register u_long i = 0; 633#ifdef INET6 634 if (isipv6) { 635 struct sockaddr_in6 *sin6; 636 637 if ((rt = inp->in6p_route.ro_rt) == NULL) 638 goto no_valid_rt; 639 sin6 = (struct sockaddr_in6 *)rt_key(rt); 640 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 641 goto no_valid_rt; 642 } 643 else 644#endif /* INET6 */ 645 if ((rt = inp->inp_route.ro_rt) == NULL || 646 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr 647 == INADDR_ANY) 648 goto no_valid_rt; 649 650 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 651 i = tp->t_srtt * 652 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 653 if (rt->rt_rmx.rmx_rtt && i) 654 /* 655 * filter this update to half the old & half 656 * the new values, converting scale. 657 * See route.h and tcp_var.h for a 658 * description of the scaling constants. 659 */ 660 rt->rt_rmx.rmx_rtt = 661 (rt->rt_rmx.rmx_rtt + i) / 2; 662 else 663 rt->rt_rmx.rmx_rtt = i; 664 tcpstat.tcps_cachedrtt++; 665 } 666 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 667 i = tp->t_rttvar * 668 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 669 if (rt->rt_rmx.rmx_rttvar && i) 670 rt->rt_rmx.rmx_rttvar = 671 (rt->rt_rmx.rmx_rttvar + i) / 2; 672 else 673 rt->rt_rmx.rmx_rttvar = i; 674 tcpstat.tcps_cachedrttvar++; 675 } 676 /* 677 * The old comment here said: 678 * update the pipelimit (ssthresh) if it has been updated 679 * already or if a pipesize was specified & the threshhold 680 * got below half the pipesize. I.e., wait for bad news 681 * before we start updating, then update on both good 682 * and bad news. 683 * 684 * But we want to save the ssthresh even if no pipesize is 685 * specified explicitly in the route, because such 686 * connections still have an implicit pipesize specified 687 * by the global tcp_sendspace. In the absence of a reliable 688 * way to calculate the pipesize, it will have to do. 689 */ 690 i = tp->snd_ssthresh; 691 if (rt->rt_rmx.rmx_sendpipe != 0) 692 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 693 else 694 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 695 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 696 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 697 || dosavessthresh) { 698 /* 699 * convert the limit from user data bytes to 700 * packets then to packet data bytes. 701 */ 702 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 703 if (i < 2) 704 i = 2; 705 i *= (u_long)(tp->t_maxseg + 706#ifdef INET6 707 (isipv6 ? sizeof (struct ip6_hdr) + 708 sizeof (struct tcphdr) : 709#endif 710 sizeof (struct tcpiphdr) 711#ifdef INET6 712 ) 713#endif 714 ); 715 if (rt->rt_rmx.rmx_ssthresh) 716 rt->rt_rmx.rmx_ssthresh = 717 (rt->rt_rmx.rmx_ssthresh + i) / 2; 718 else 719 rt->rt_rmx.rmx_ssthresh = i; 720 tcpstat.tcps_cachedssthresh++; 721 } 722 } 723 no_valid_rt: 724 /* free the reassembly queue, if any */ 725 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 726 LIST_REMOVE(q, tqe_q); 727 m_freem(q->tqe_m); 728 FREE(q, M_TSEGQ); 729 } 730 inp->inp_ppcb = NULL; 731 SOCK_LOCK(so); 732 soisdisconnected(so); 733 SOCK_UNLOCK(so); 734#ifdef INET6 735 if (INP_CHECK_SOCKAF(so, AF_INET6)) 736 in6_pcbdetach(inp); 737 else 738#endif /* INET6 */ 739 in_pcbdetach(inp); 740 tcpstat.tcps_closed++; 741 return ((struct tcpcb *)0); 742} 743 744void 745tcp_drain() 746{ 747 if (do_tcpdrain) 748 { 749 struct inpcb *inpb; 750 struct tcpcb *tcpb; 751 struct tseg_qent *te; 752 753 /* 754 * Walk the tcpbs, if existing, and flush the reassembly queue, 755 * if there is one... 756 * XXX: The "Net/3" implementation doesn't imply that the TCP 757 * reassembly queue should be flushed, but in a situation 758 * where we're really low on mbufs, this is potentially 759 * usefull. 760 */ 761 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 762 if ((tcpb = intotcpcb(inpb))) { 763 while ((te = LIST_FIRST(&tcpb->t_segq)) 764 != NULL) { 765 LIST_REMOVE(te, tqe_q); 766 m_freem(te->tqe_m); 767 FREE(te, M_TSEGQ); 768 } 769 } 770 } 771 } 772} 773 774/* 775 * Notify a tcp user of an asynchronous error; 776 * store error as soft error, but wake up user 777 * (for now, won't do anything until can select for soft error). 778 * 779 * Do not wake up user since there currently is no mechanism for 780 * reporting soft errors (yet - a kqueue filter may be added). 781 */ 782static void 783tcp_notify(inp, error) 784 struct inpcb *inp; 785 int error; 786{ 787 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 788 789 /* 790 * Ignore some errors if we are hooked up. 791 * If connection hasn't completed, has retransmitted several times, 792 * and receives a second error, give up now. This is better 793 * than waiting a long time to establish a connection that 794 * can never complete. 795 */ 796 if (tp->t_state == TCPS_ESTABLISHED && 797 (error == EHOSTUNREACH || error == ENETUNREACH || 798 error == EHOSTDOWN)) { 799 return; 800 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 801 tp->t_softerror) 802 tcp_drop(tp, error); 803 else 804 tp->t_softerror = error; 805#if 0 806 SOCK_LOCK(so); 807 wakeup((caddr_t) &so->so_timeo); 808 sorwakeup(so); 809 sowwakeup(so); 810 SOCK_UNLOCK(so); 811#endif 812} 813 814static int 815tcp_pcblist(SYSCTL_HANDLER_ARGS) 816{ 817 int error, i, n, s; 818 struct inpcb *inp, **inp_list; 819 inp_gen_t gencnt; 820 struct xinpgen xig; 821 822 /* 823 * The process of preparing the TCB list is too time-consuming and 824 * resource-intensive to repeat twice on every request. 825 */ 826 if (req->oldptr == 0) { 827 n = tcbinfo.ipi_count; 828 req->oldidx = 2 * (sizeof xig) 829 + (n + n/8) * sizeof(struct xtcpcb); 830 return 0; 831 } 832 833 if (req->newptr != 0) 834 return EPERM; 835 836 /* 837 * OK, now we're committed to doing something. 838 */ 839 s = splnet(); 840 gencnt = tcbinfo.ipi_gencnt; 841 n = tcbinfo.ipi_count; 842 splx(s); 843 844 xig.xig_len = sizeof xig; 845 xig.xig_count = n; 846 xig.xig_gen = gencnt; 847 xig.xig_sogen = so_gencnt; 848 error = SYSCTL_OUT(req, &xig, sizeof xig); 849 if (error) 850 return error; 851 852 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 853 if (inp_list == 0) 854 return ENOMEM; 855 856 s = splnet(); 857 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; 858 inp = LIST_NEXT(inp, inp_list)) { 859 if (inp->inp_gencnt <= gencnt) { 860 if (cr_canseesocket(req->td->td_ucred, 861 inp->inp_socket)) 862 continue; 863 inp_list[i++] = inp; 864 } 865 } 866 splx(s); 867 n = i; 868 869 error = 0; 870 for (i = 0; i < n; i++) { 871 inp = inp_list[i]; 872 if (inp->inp_gencnt <= gencnt) { 873 struct xtcpcb xt; 874 caddr_t inp_ppcb; 875 xt.xt_len = sizeof xt; 876 /* XXX should avoid extra copy */ 877 bcopy(inp, &xt.xt_inp, sizeof *inp); 878 inp_ppcb = inp->inp_ppcb; 879 if (inp_ppcb != NULL) 880 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 881 else 882 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 883 if (inp->inp_socket) 884 sotoxsocket(inp->inp_socket, &xt.xt_socket); 885 error = SYSCTL_OUT(req, &xt, sizeof xt); 886 } 887 } 888 if (!error) { 889 /* 890 * Give the user an updated idea of our state. 891 * If the generation differs from what we told 892 * her before, she knows that something happened 893 * while we were processing this request, and it 894 * might be necessary to retry. 895 */ 896 s = splnet(); 897 xig.xig_gen = tcbinfo.ipi_gencnt; 898 xig.xig_sogen = so_gencnt; 899 xig.xig_count = tcbinfo.ipi_count; 900 splx(s); 901 error = SYSCTL_OUT(req, &xig, sizeof xig); 902 } 903 free(inp_list, M_TEMP); 904 return error; 905} 906 907SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 908 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 909 910static int 911tcp_getcred(SYSCTL_HANDLER_ARGS) 912{ 913 struct xucred xuc; 914 struct sockaddr_in addrs[2]; 915 struct inpcb *inp; 916 int error, s; 917 918 error = suser_cred(req->td->td_ucred, PRISON_ROOT); 919 if (error) 920 return (error); 921 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 922 if (error) 923 return (error); 924 s = splnet(); 925 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 926 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 927 if (inp == NULL || inp->inp_socket == NULL) { 928 error = ENOENT; 929 goto out; 930 } 931 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 932 if (error) 933 goto out; 934 cru2x(inp->inp_socket->so_cred, &xuc); 935 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 936out: 937 splx(s); 938 return (error); 939} 940 941SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 942 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 943 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 944 945#ifdef INET6 946static int 947tcp6_getcred(SYSCTL_HANDLER_ARGS) 948{ 949 struct xucred xuc; 950 struct sockaddr_in6 addrs[2]; 951 struct inpcb *inp; 952 int error, s, mapped = 0; 953 954 error = suser_cred(req->td->td_ucred, PRISON_ROOT); 955 if (error) 956 return (error); 957 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 958 if (error) 959 return (error); 960 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 961 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 962 mapped = 1; 963 else 964 return (EINVAL); 965 } 966 s = splnet(); 967 if (mapped == 1) 968 inp = in_pcblookup_hash(&tcbinfo, 969 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 970 addrs[1].sin6_port, 971 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 972 addrs[0].sin6_port, 973 0, NULL); 974 else 975 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 976 addrs[1].sin6_port, 977 &addrs[0].sin6_addr, addrs[0].sin6_port, 978 0, NULL); 979 if (inp == NULL || inp->inp_socket == NULL) { 980 error = ENOENT; 981 goto out; 982 } 983 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 984 if (error) 985 goto out; 986 cru2x(inp->inp_socket->so_cred, &xuc); 987 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 988out: 989 splx(s); 990 return (error); 991} 992 993SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 994 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 995 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 996#endif 997 998 999void 1000tcp_ctlinput(cmd, sa, vip) 1001 int cmd; 1002 struct sockaddr *sa; 1003 void *vip; 1004{ 1005 struct ip *ip = vip; 1006 struct tcphdr *th; 1007 struct in_addr faddr; 1008 struct inpcb *inp; 1009 struct tcpcb *tp; 1010 void (*notify)(struct inpcb *, int) = tcp_notify; 1011 tcp_seq icmp_seq; 1012 int s; 1013 1014 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1015 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1016 return; 1017 1018 if (cmd == PRC_QUENCH) 1019 notify = tcp_quench; 1020 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1021 cmd == PRC_UNREACH_PORT) && ip) 1022 notify = tcp_drop_syn_sent; 1023 else if (cmd == PRC_MSGSIZE) 1024 notify = tcp_mtudisc; 1025 else if (PRC_IS_REDIRECT(cmd)) { 1026 ip = 0; 1027 notify = in_rtchange; 1028 } else if (cmd == PRC_HOSTDEAD) 1029 ip = 0; 1030 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1031 return; 1032 if (ip) { 1033 s = splnet(); 1034 th = (struct tcphdr *)((caddr_t)ip 1035 + (IP_VHL_HL(ip->ip_vhl) << 2)); 1036 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1037 ip->ip_src, th->th_sport, 0, NULL); 1038 if (inp != NULL && inp->inp_socket != NULL) { 1039 icmp_seq = htonl(th->th_seq); 1040 tp = intotcpcb(inp); 1041 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1042 SEQ_LT(icmp_seq, tp->snd_max)) 1043 (*notify)(inp, inetctlerrmap[cmd]); 1044 } else { 1045 struct in_conninfo inc; 1046 1047 inc.inc_fport = th->th_dport; 1048 inc.inc_lport = th->th_sport; 1049 inc.inc_faddr = faddr; 1050 inc.inc_laddr = ip->ip_src; 1051#ifdef INET6 1052 inc.inc_isipv6 = 0; 1053#endif 1054 syncache_unreach(&inc, th); 1055 } 1056 splx(s); 1057 } else 1058 in_pcbnotifyall(&tcb, faddr, inetctlerrmap[cmd], notify); 1059} 1060 1061#ifdef INET6 1062void 1063tcp6_ctlinput(cmd, sa, d) 1064 int cmd; 1065 struct sockaddr *sa; 1066 void *d; 1067{ 1068 struct tcphdr th; 1069 void (*notify)(struct inpcb *, int) = tcp_notify; 1070 struct ip6_hdr *ip6; 1071 struct mbuf *m; 1072 struct ip6ctlparam *ip6cp = NULL; 1073 const struct sockaddr_in6 *sa6_src = NULL; 1074 int off; 1075 struct tcp_portonly { 1076 u_int16_t th_sport; 1077 u_int16_t th_dport; 1078 } *thp; 1079 1080 if (sa->sa_family != AF_INET6 || 1081 sa->sa_len != sizeof(struct sockaddr_in6)) 1082 return; 1083 1084 if (cmd == PRC_QUENCH) 1085 notify = tcp_quench; 1086 else if (cmd == PRC_MSGSIZE) 1087 notify = tcp_mtudisc; 1088 else if (!PRC_IS_REDIRECT(cmd) && 1089 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1090 return; 1091 1092 /* if the parameter is from icmp6, decode it. */ 1093 if (d != NULL) { 1094 ip6cp = (struct ip6ctlparam *)d; 1095 m = ip6cp->ip6c_m; 1096 ip6 = ip6cp->ip6c_ip6; 1097 off = ip6cp->ip6c_off; 1098 sa6_src = ip6cp->ip6c_src; 1099 } else { 1100 m = NULL; 1101 ip6 = NULL; 1102 off = 0; /* fool gcc */ 1103 sa6_src = &sa6_any; 1104 } 1105 1106 if (ip6) { 1107 struct in_conninfo inc; 1108 /* 1109 * XXX: We assume that when IPV6 is non NULL, 1110 * M and OFF are valid. 1111 */ 1112 1113 /* check if we can safely examine src and dst ports */ 1114 if (m->m_pkthdr.len < off + sizeof(*thp)) 1115 return; 1116 1117 bzero(&th, sizeof(th)); 1118 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1119 1120 in6_pcbnotify(&tcb, sa, th.th_dport, 1121 (struct sockaddr *)ip6cp->ip6c_src, 1122 th.th_sport, cmd, notify); 1123 1124 inc.inc_fport = th.th_dport; 1125 inc.inc_lport = th.th_sport; 1126 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1127 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1128 inc.inc_isipv6 = 1; 1129 syncache_unreach(&inc, &th); 1130 } else 1131 in6_pcbnotify(&tcb, sa, 0, (const struct sockaddr *)sa6_src, 1132 0, cmd, notify); 1133} 1134#endif /* INET6 */ 1135 1136 1137/* 1138 * Following is where TCP initial sequence number generation occurs. 1139 * 1140 * There are two places where we must use initial sequence numbers: 1141 * 1. In SYN-ACK packets. 1142 * 2. In SYN packets. 1143 * 1144 * All ISNs for SYN-ACK packets are generated by the syncache. See 1145 * tcp_syncache.c for details. 1146 * 1147 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1148 * depends on this property. In addition, these ISNs should be 1149 * unguessable so as to prevent connection hijacking. To satisfy 1150 * the requirements of this situation, the algorithm outlined in 1151 * RFC 1948 is used to generate sequence numbers. 1152 * 1153 * Implementation details: 1154 * 1155 * Time is based off the system timer, and is corrected so that it 1156 * increases by one megabyte per second. This allows for proper 1157 * recycling on high speed LANs while still leaving over an hour 1158 * before rollover. 1159 * 1160 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1161 * between seeding of isn_secret. This is normally set to zero, 1162 * as reseeding should not be necessary. 1163 * 1164 */ 1165 1166#define ISN_BYTES_PER_SECOND 1048576 1167 1168u_char isn_secret[32]; 1169int isn_last_reseed; 1170MD5_CTX isn_ctx; 1171 1172tcp_seq 1173tcp_new_isn(tp) 1174 struct tcpcb *tp; 1175{ 1176 u_int32_t md5_buffer[4]; 1177 tcp_seq new_isn; 1178 1179 /* Seed if this is the first use, reseed if requested. */ 1180 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1181 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1182 < (u_int)ticks))) { 1183 read_random(&isn_secret, sizeof(isn_secret)); 1184 isn_last_reseed = ticks; 1185 } 1186 1187 /* Compute the md5 hash and return the ISN. */ 1188 MD5Init(&isn_ctx); 1189 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1190 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1191#ifdef INET6 1192 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1193 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1194 sizeof(struct in6_addr)); 1195 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1196 sizeof(struct in6_addr)); 1197 } else 1198#endif 1199 { 1200 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1201 sizeof(struct in_addr)); 1202 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1203 sizeof(struct in_addr)); 1204 } 1205 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1206 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1207 new_isn = (tcp_seq) md5_buffer[0]; 1208 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1209 return new_isn; 1210} 1211 1212/* 1213 * When a source quench is received, close congestion window 1214 * to one segment. We will gradually open it again as we proceed. 1215 */ 1216void 1217tcp_quench(inp, errno) 1218 struct inpcb *inp; 1219 int errno; 1220{ 1221 struct tcpcb *tp = intotcpcb(inp); 1222 1223 if (tp) 1224 tp->snd_cwnd = tp->t_maxseg; 1225} 1226 1227/* 1228 * When a specific ICMP unreachable message is received and the 1229 * connection state is SYN-SENT, drop the connection. This behavior 1230 * is controlled by the icmp_may_rst sysctl. 1231 */ 1232void 1233tcp_drop_syn_sent(inp, errno) 1234 struct inpcb *inp; 1235 int errno; 1236{ 1237 struct tcpcb *tp = intotcpcb(inp); 1238 1239 if (tp && tp->t_state == TCPS_SYN_SENT) 1240 tcp_drop(tp, errno); 1241} 1242 1243/* 1244 * When `need fragmentation' ICMP is received, update our idea of the MSS 1245 * based on the new value in the route. Also nudge TCP to send something, 1246 * since we know the packet we just sent was dropped. 1247 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1248 */ 1249void 1250tcp_mtudisc(inp, errno) 1251 struct inpcb *inp; 1252 int errno; 1253{ 1254 struct tcpcb *tp = intotcpcb(inp); 1255 struct rtentry *rt; 1256 struct rmxp_tao *taop; 1257 struct socket *so = inp->inp_socket; 1258 int offered; 1259 int mss; 1260#ifdef INET6 1261 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1262#endif /* INET6 */ 1263 1264 if (tp) { 1265#ifdef INET6 1266 if (isipv6) 1267 rt = tcp_rtlookup6(&inp->inp_inc); 1268 else 1269#endif /* INET6 */ 1270 rt = tcp_rtlookup(&inp->inp_inc); 1271 if (!rt || !rt->rt_rmx.rmx_mtu) { 1272 tp->t_maxopd = tp->t_maxseg = 1273#ifdef INET6 1274 isipv6 ? tcp_v6mssdflt : 1275#endif /* INET6 */ 1276 tcp_mssdflt; 1277 return; 1278 } 1279 taop = rmx_taop(rt->rt_rmx); 1280 offered = taop->tao_mssopt; 1281 mss = rt->rt_rmx.rmx_mtu - 1282#ifdef INET6 1283 (isipv6 ? 1284 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1285#endif /* INET6 */ 1286 sizeof(struct tcpiphdr) 1287#ifdef INET6 1288 ) 1289#endif /* INET6 */ 1290 ; 1291 1292 if (offered) 1293 mss = min(mss, offered); 1294 /* 1295 * XXX - The above conditional probably violates the TCP 1296 * spec. The problem is that, since we don't know the 1297 * other end's MSS, we are supposed to use a conservative 1298 * default. But, if we do that, then MTU discovery will 1299 * never actually take place, because the conservative 1300 * default is much less than the MTUs typically seen 1301 * on the Internet today. For the moment, we'll sweep 1302 * this under the carpet. 1303 * 1304 * The conservative default might not actually be a problem 1305 * if the only case this occurs is when sending an initial 1306 * SYN with options and data to a host we've never talked 1307 * to before. Then, they will reply with an MSS value which 1308 * will get recorded and the new parameters should get 1309 * recomputed. For Further Study. 1310 */ 1311 if (tp->t_maxopd <= mss) 1312 return; 1313 tp->t_maxopd = mss; 1314 1315 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1316 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1317 mss -= TCPOLEN_TSTAMP_APPA; 1318 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1319 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1320 mss -= TCPOLEN_CC_APPA; 1321#if (MCLBYTES & (MCLBYTES - 1)) == 0 1322 if (mss > MCLBYTES) 1323 mss &= ~(MCLBYTES-1); 1324#else 1325 if (mss > MCLBYTES) 1326 mss = mss / MCLBYTES * MCLBYTES; 1327#endif 1328 if (so->so_snd.sb_hiwat < mss) 1329 mss = so->so_snd.sb_hiwat; 1330 1331 tp->t_maxseg = mss; 1332 1333 tcpstat.tcps_mturesent++; 1334 tp->t_rtttime = 0; 1335 tp->snd_nxt = tp->snd_una; 1336 tcp_output(tp); 1337 } 1338} 1339 1340/* 1341 * Look-up the routing entry to the peer of this inpcb. If no route 1342 * is found and it cannot be allocated the return NULL. This routine 1343 * is called by TCP routines that access the rmx structure and by tcp_mss 1344 * to get the interface MTU. 1345 */ 1346struct rtentry * 1347tcp_rtlookup(inc) 1348 struct in_conninfo *inc; 1349{ 1350 struct route *ro; 1351 struct rtentry *rt; 1352 1353 ro = &inc->inc_route; 1354 rt = ro->ro_rt; 1355 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1356 /* No route yet, so try to acquire one */ 1357 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1358 ro->ro_dst.sa_family = AF_INET; 1359 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1360 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1361 inc->inc_faddr; 1362 rtalloc(ro); 1363 rt = ro->ro_rt; 1364 } 1365 } 1366 return rt; 1367} 1368 1369#ifdef INET6 1370struct rtentry * 1371tcp_rtlookup6(inc) 1372 struct in_conninfo *inc; 1373{ 1374 struct route_in6 *ro6; 1375 struct rtentry *rt; 1376 1377 ro6 = &inc->inc6_route; 1378 rt = ro6->ro_rt; 1379 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1380 /* No route yet, so try to acquire one */ 1381 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1382 ro6->ro_dst.sin6_family = AF_INET6; 1383 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1384 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1385 rtalloc((struct route *)ro6); 1386 rt = ro6->ro_rt; 1387 } 1388 } 1389 return rt; 1390} 1391#endif /* INET6 */ 1392 1393#ifdef IPSEC 1394/* compute ESP/AH header size for TCP, including outer IP header. */ 1395size_t 1396ipsec_hdrsiz_tcp(tp) 1397 struct tcpcb *tp; 1398{ 1399 struct inpcb *inp; 1400 struct mbuf *m; 1401 size_t hdrsiz; 1402 struct ip *ip; 1403#ifdef INET6 1404 struct ip6_hdr *ip6; 1405#endif /* INET6 */ 1406 struct tcphdr *th; 1407 1408 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1409 return 0; 1410 MGETHDR(m, M_DONTWAIT, MT_DATA); 1411 if (!m) 1412 return 0; 1413 1414#ifdef INET6 1415 if ((inp->inp_vflag & INP_IPV6) != 0) { 1416 ip6 = mtod(m, struct ip6_hdr *); 1417 th = (struct tcphdr *)(ip6 + 1); 1418 m->m_pkthdr.len = m->m_len = 1419 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1420 tcp_fillheaders(tp, ip6, th); 1421 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1422 } else 1423#endif /* INET6 */ 1424 { 1425 ip = mtod(m, struct ip *); 1426 th = (struct tcphdr *)(ip + 1); 1427 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1428 tcp_fillheaders(tp, ip, th); 1429 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1430 } 1431 1432 m_free(m); 1433 return hdrsiz; 1434} 1435#endif /*IPSEC*/ 1436 1437/* 1438 * Return a pointer to the cached information about the remote host. 1439 * The cached information is stored in the protocol specific part of 1440 * the route metrics. 1441 */ 1442struct rmxp_tao * 1443tcp_gettaocache(inc) 1444 struct in_conninfo *inc; 1445{ 1446 struct rtentry *rt; 1447 1448#ifdef INET6 1449 if (inc->inc_isipv6) 1450 rt = tcp_rtlookup6(inc); 1451 else 1452#endif /* INET6 */ 1453 rt = tcp_rtlookup(inc); 1454 1455 /* Make sure this is a host route and is up. */ 1456 if (rt == NULL || 1457 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 1458 return NULL; 1459 1460 return rmx_taop(rt->rt_rmx); 1461} 1462 1463/* 1464 * Clear all the TAO cache entries, called from tcp_init. 1465 * 1466 * XXX 1467 * This routine is just an empty one, because we assume that the routing 1468 * routing tables are initialized at the same time when TCP, so there is 1469 * nothing in the cache left over. 1470 */ 1471static void 1472tcp_cleartaocache() 1473{ 1474} 1475