tcp_subr.c revision 1.118
1 /* $NetBSD: tcp_subr.c,v 1.118 2001/09/10 22:14:27 thorpej Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998, 2000, 2001 The NetBSD Foundation, Inc. 34 * All rights reserved. 35 * 36 * This code is derived from software contributed to The NetBSD Foundation 37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 38 * Facility, NASA Ames Research Center. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. All advertising materials mentioning features or use of this software 49 * must display the following acknowledgement: 50 * This product includes software developed by the NetBSD 51 * Foundation, Inc. and its contributors. 52 * 4. Neither the name of The NetBSD Foundation nor the names of its 53 * contributors may be used to endorse or promote products derived 54 * from this software without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 57 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 66 * POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 /* 70 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 71 * The Regents of the University of California. All rights reserved. 72 * 73 * Redistribution and use in source and binary forms, with or without 74 * modification, are permitted provided that the following conditions 75 * are met: 76 * 1. Redistributions of source code must retain the above copyright 77 * notice, this list of conditions and the following disclaimer. 78 * 2. Redistributions in binary form must reproduce the above copyright 79 * notice, this list of conditions and the following disclaimer in the 80 * documentation and/or other materials provided with the distribution. 81 * 3. All advertising materials mentioning features or use of this software 82 * must display the following acknowledgement: 83 * This product includes software developed by the University of 84 * California, Berkeley and its contributors. 85 * 4. Neither the name of the University nor the names of its contributors 86 * may be used to endorse or promote products derived from this software 87 * without specific prior written permission. 88 * 89 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 90 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 91 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 92 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 93 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 94 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 95 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 96 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 97 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 98 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 99 * SUCH DAMAGE. 100 * 101 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 102 */ 103 104 #include "opt_inet.h" 105 #include "opt_ipsec.h" 106 #include "opt_tcp_compat_42.h" 107 #include "opt_inet_csum.h" 108 #include "rnd.h" 109 110 #include <sys/param.h> 111 #include <sys/proc.h> 112 #include <sys/systm.h> 113 #include <sys/malloc.h> 114 #include <sys/mbuf.h> 115 #include <sys/socket.h> 116 #include <sys/socketvar.h> 117 #include <sys/protosw.h> 118 #include <sys/errno.h> 119 #include <sys/kernel.h> 120 #include <sys/pool.h> 121 #if NRND > 0 122 #include <sys/md5.h> 123 #include <sys/rnd.h> 124 #endif 125 126 #include <net/route.h> 127 #include <net/if.h> 128 129 #include <netinet/in.h> 130 #include <netinet/in_systm.h> 131 #include <netinet/ip.h> 132 #include <netinet/in_pcb.h> 133 #include <netinet/ip_var.h> 134 #include <netinet/ip_icmp.h> 135 136 #ifdef INET6 137 #ifndef INET 138 #include <netinet/in.h> 139 #endif 140 #include <netinet/ip6.h> 141 #include <netinet6/in6_pcb.h> 142 #include <netinet6/ip6_var.h> 143 #include <netinet6/in6_var.h> 144 #include <netinet6/ip6protosw.h> 145 #include <netinet/icmp6.h> 146 #endif 147 148 #include <netinet/tcp.h> 149 #include <netinet/tcp_fsm.h> 150 #include <netinet/tcp_seq.h> 151 #include <netinet/tcp_timer.h> 152 #include <netinet/tcp_var.h> 153 #include <netinet/tcpip.h> 154 155 #ifdef IPSEC 156 #include <netinet6/ipsec.h> 157 #endif /*IPSEC*/ 158 159 #ifdef INET6 160 struct in6pcb tcb6; 161 #endif 162 163 /* patchable/settable parameters for tcp */ 164 int tcp_mssdflt = TCP_MSS; 165 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 166 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 167 #if NRND > 0 168 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 169 #endif 170 int tcp_do_sack = 1; /* selective acknowledgement */ 171 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 172 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 173 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 174 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 175 int tcp_init_win = 1; 176 int tcp_mss_ifmtu = 0; 177 #ifdef TCP_COMPAT_42 178 int tcp_compat_42 = 1; 179 #else 180 int tcp_compat_42 = 0; 181 #endif 182 int tcp_rst_ppslim = 100; /* 100pps */ 183 184 /* tcb hash */ 185 #ifndef TCBHASHSIZE 186 #define TCBHASHSIZE 128 187 #endif 188 int tcbhashsize = TCBHASHSIZE; 189 190 /* syn hash parameters */ 191 #define TCP_SYN_HASH_SIZE 293 192 #define TCP_SYN_BUCKET_SIZE 35 193 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 194 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 195 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 196 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 197 int tcp_syn_cache_interval = 1; /* runs timer twice a second */ 198 199 int tcp_freeq __P((struct tcpcb *)); 200 201 #ifdef INET 202 void tcp_mtudisc_callback __P((struct in_addr)); 203 #endif 204 #ifdef INET6 205 void tcp6_mtudisc_callback __P((struct in6_addr *)); 206 #endif 207 208 void tcp_mtudisc __P((struct inpcb *, int)); 209 #ifdef INET6 210 void tcp6_mtudisc __P((struct in6pcb *, int)); 211 #endif 212 213 struct pool tcpcb_pool; 214 215 #ifdef TCP_CSUM_COUNTERS 216 #include <sys/device.h> 217 218 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 219 NULL, "tcp", "hwcsum bad"); 220 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 221 NULL, "tcp", "hwcsum ok"); 222 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 223 NULL, "tcp", "hwcsum data"); 224 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 225 NULL, "tcp", "swcsum"); 226 #endif /* TCP_CSUM_COUNTERS */ 227 228 /* 229 * Tcp initialization 230 */ 231 void 232 tcp_init() 233 { 234 int hlen; 235 236 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 237 0, NULL, NULL, M_PCB); 238 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 239 #ifdef INET6 240 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 241 #endif 242 243 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 244 #ifdef INET6 245 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 246 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 247 #endif 248 if (max_protohdr < hlen) 249 max_protohdr = hlen; 250 if (max_linkhdr + hlen > MHLEN) 251 panic("tcp_init"); 252 253 #ifdef INET 254 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 255 #endif 256 #ifdef INET6 257 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 258 #endif 259 260 /* Initialize timer state. */ 261 tcp_timer_init(); 262 263 /* Initialize the compressed state engine. */ 264 syn_cache_init(); 265 266 #ifdef TCP_CSUM_COUNTERS 267 evcnt_attach_static(&tcp_hwcsum_bad); 268 evcnt_attach_static(&tcp_hwcsum_ok); 269 evcnt_attach_static(&tcp_hwcsum_data); 270 evcnt_attach_static(&tcp_swcsum); 271 #endif /* TCP_CSUM_COUNTERS */ 272 } 273 274 /* 275 * Create template to be used to send tcp packets on a connection. 276 * Call after host entry created, allocates an mbuf and fills 277 * in a skeletal tcp/ip header, minimizing the amount of work 278 * necessary when the connection is used. 279 */ 280 struct mbuf * 281 tcp_template(tp) 282 struct tcpcb *tp; 283 { 284 struct inpcb *inp = tp->t_inpcb; 285 #ifdef INET6 286 struct in6pcb *in6p = tp->t_in6pcb; 287 #endif 288 struct tcphdr *n; 289 struct mbuf *m; 290 int hlen; 291 292 switch (tp->t_family) { 293 case AF_INET: 294 hlen = sizeof(struct ip); 295 if (inp) 296 break; 297 #ifdef INET6 298 if (in6p) { 299 /* mapped addr case */ 300 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 301 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 302 break; 303 } 304 #endif 305 return NULL; /*EINVAL*/ 306 #ifdef INET6 307 case AF_INET6: 308 hlen = sizeof(struct ip6_hdr); 309 if (in6p) { 310 /* more sainty check? */ 311 break; 312 } 313 return NULL; /*EINVAL*/ 314 #endif 315 default: 316 hlen = 0; /*pacify gcc*/ 317 return NULL; /*EAFNOSUPPORT*/ 318 } 319 #ifdef DIAGNOSTIC 320 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 321 panic("mclbytes too small for t_template"); 322 #endif 323 m = tp->t_template; 324 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 325 ; 326 else { 327 if (m) 328 m_freem(m); 329 m = tp->t_template = NULL; 330 MGETHDR(m, M_DONTWAIT, MT_HEADER); 331 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 332 MCLGET(m, M_DONTWAIT); 333 if ((m->m_flags & M_EXT) == 0) { 334 m_free(m); 335 m = NULL; 336 } 337 } 338 if (m == NULL) 339 return NULL; 340 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 341 } 342 343 bzero(mtod(m, caddr_t), m->m_len); 344 345 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 346 347 switch (tp->t_family) { 348 case AF_INET: 349 { 350 struct ipovly *ipov; 351 mtod(m, struct ip *)->ip_v = 4; 352 ipov = mtod(m, struct ipovly *); 353 ipov->ih_pr = IPPROTO_TCP; 354 ipov->ih_len = htons(sizeof(struct tcphdr)); 355 if (inp) { 356 ipov->ih_src = inp->inp_laddr; 357 ipov->ih_dst = inp->inp_faddr; 358 } 359 #ifdef INET6 360 else if (in6p) { 361 /* mapped addr case */ 362 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 363 sizeof(ipov->ih_src)); 364 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 365 sizeof(ipov->ih_dst)); 366 } 367 #endif 368 /* 369 * Compute the pseudo-header portion of the checksum 370 * now. We incrementally add in the TCP option and 371 * payload lengths later, and then compute the TCP 372 * checksum right before the packet is sent off onto 373 * the wire. 374 */ 375 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 376 ipov->ih_dst.s_addr, 377 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 378 break; 379 } 380 #ifdef INET6 381 case AF_INET6: 382 { 383 struct ip6_hdr *ip6; 384 mtod(m, struct ip *)->ip_v = 6; 385 ip6 = mtod(m, struct ip6_hdr *); 386 ip6->ip6_nxt = IPPROTO_TCP; 387 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 388 ip6->ip6_src = in6p->in6p_laddr; 389 ip6->ip6_dst = in6p->in6p_faddr; 390 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 391 if (ip6_auto_flowlabel) { 392 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 393 ip6->ip6_flow |= 394 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 395 } 396 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 397 ip6->ip6_vfc |= IPV6_VERSION; 398 399 /* 400 * Compute the pseudo-header portion of the checksum 401 * now. We incrementally add in the TCP option and 402 * payload lengths later, and then compute the TCP 403 * checksum right before the packet is sent off onto 404 * the wire. 405 */ 406 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 407 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 408 htonl(IPPROTO_TCP)); 409 break; 410 } 411 #endif 412 } 413 if (inp) { 414 n->th_sport = inp->inp_lport; 415 n->th_dport = inp->inp_fport; 416 } 417 #ifdef INET6 418 else if (in6p) { 419 n->th_sport = in6p->in6p_lport; 420 n->th_dport = in6p->in6p_fport; 421 } 422 #endif 423 n->th_seq = 0; 424 n->th_ack = 0; 425 n->th_x2 = 0; 426 n->th_off = 5; 427 n->th_flags = 0; 428 n->th_win = 0; 429 n->th_urp = 0; 430 return (m); 431 } 432 433 /* 434 * Send a single message to the TCP at address specified by 435 * the given TCP/IP header. If m == 0, then we make a copy 436 * of the tcpiphdr at ti and send directly to the addressed host. 437 * This is used to force keep alive messages out using the TCP 438 * template for a connection tp->t_template. If flags are given 439 * then we send a message back to the TCP which originated the 440 * segment ti, and discard the mbuf containing it and any other 441 * attached mbufs. 442 * 443 * In any case the ack and sequence number of the transmitted 444 * segment are as specified by the parameters. 445 */ 446 int 447 tcp_respond(tp, template, m, th0, ack, seq, flags) 448 struct tcpcb *tp; 449 struct mbuf *template; 450 struct mbuf *m; 451 struct tcphdr *th0; 452 tcp_seq ack, seq; 453 int flags; 454 { 455 struct route *ro; 456 int error, tlen, win = 0; 457 int hlen; 458 struct ip *ip; 459 #ifdef INET6 460 struct ip6_hdr *ip6; 461 #endif 462 int family; /* family on packet, not inpcb/in6pcb! */ 463 struct tcphdr *th; 464 465 if (tp != NULL && (flags & TH_RST) == 0) { 466 #ifdef DIAGNOSTIC 467 if (tp->t_inpcb && tp->t_in6pcb) 468 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 469 #endif 470 #ifdef INET 471 if (tp->t_inpcb) 472 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 473 #endif 474 #ifdef INET6 475 if (tp->t_in6pcb) 476 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 477 #endif 478 } 479 480 ip = NULL; 481 #ifdef INET6 482 ip6 = NULL; 483 #endif 484 if (m == 0) { 485 if (!template) 486 return EINVAL; 487 488 /* get family information from template */ 489 switch (mtod(template, struct ip *)->ip_v) { 490 case 4: 491 family = AF_INET; 492 hlen = sizeof(struct ip); 493 break; 494 #ifdef INET6 495 case 6: 496 family = AF_INET6; 497 hlen = sizeof(struct ip6_hdr); 498 break; 499 #endif 500 default: 501 return EAFNOSUPPORT; 502 } 503 504 MGETHDR(m, M_DONTWAIT, MT_HEADER); 505 if (m) { 506 MCLGET(m, M_DONTWAIT); 507 if ((m->m_flags & M_EXT) == 0) { 508 m_free(m); 509 m = NULL; 510 } 511 } 512 if (m == NULL) 513 return (ENOBUFS); 514 515 if (tcp_compat_42) 516 tlen = 1; 517 else 518 tlen = 0; 519 520 m->m_data += max_linkhdr; 521 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 522 template->m_len); 523 switch (family) { 524 case AF_INET: 525 ip = mtod(m, struct ip *); 526 th = (struct tcphdr *)(ip + 1); 527 break; 528 #ifdef INET6 529 case AF_INET6: 530 ip6 = mtod(m, struct ip6_hdr *); 531 th = (struct tcphdr *)(ip6 + 1); 532 break; 533 #endif 534 #if 0 535 default: 536 /* noone will visit here */ 537 m_freem(m); 538 return EAFNOSUPPORT; 539 #endif 540 } 541 flags = TH_ACK; 542 } else { 543 544 if ((m->m_flags & M_PKTHDR) == 0) { 545 #if 0 546 printf("non PKTHDR to tcp_respond\n"); 547 #endif 548 m_freem(m); 549 return EINVAL; 550 } 551 #ifdef DIAGNOSTIC 552 if (!th0) 553 panic("th0 == NULL in tcp_respond"); 554 #endif 555 556 /* get family information from m */ 557 switch (mtod(m, struct ip *)->ip_v) { 558 case 4: 559 family = AF_INET; 560 hlen = sizeof(struct ip); 561 ip = mtod(m, struct ip *); 562 break; 563 #ifdef INET6 564 case 6: 565 family = AF_INET6; 566 hlen = sizeof(struct ip6_hdr); 567 ip6 = mtod(m, struct ip6_hdr *); 568 break; 569 #endif 570 default: 571 m_freem(m); 572 return EAFNOSUPPORT; 573 } 574 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 575 tlen = sizeof(*th0); 576 else 577 tlen = th0->th_off << 2; 578 579 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 580 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 581 m->m_len = hlen + tlen; 582 m_freem(m->m_next); 583 m->m_next = NULL; 584 } else { 585 struct mbuf *n; 586 587 #ifdef DIAGNOSTIC 588 if (max_linkhdr + hlen + tlen > MCLBYTES) { 589 m_freem(m); 590 return EMSGSIZE; 591 } 592 #endif 593 MGETHDR(n, M_DONTWAIT, MT_HEADER); 594 if (n && max_linkhdr + hlen + tlen > MHLEN) { 595 MCLGET(n, M_DONTWAIT); 596 if ((n->m_flags & M_EXT) == 0) { 597 m_freem(n); 598 n = NULL; 599 } 600 } 601 if (!n) { 602 m_freem(m); 603 return ENOBUFS; 604 } 605 606 n->m_data += max_linkhdr; 607 n->m_len = hlen + tlen; 608 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 609 m_copyback(n, hlen, tlen, (caddr_t)th0); 610 611 m_freem(m); 612 m = n; 613 n = NULL; 614 } 615 616 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 617 switch (family) { 618 case AF_INET: 619 ip = mtod(m, struct ip *); 620 th = (struct tcphdr *)(ip + 1); 621 ip->ip_p = IPPROTO_TCP; 622 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 623 ip->ip_p = IPPROTO_TCP; 624 break; 625 #ifdef INET6 626 case AF_INET6: 627 ip6 = mtod(m, struct ip6_hdr *); 628 th = (struct tcphdr *)(ip6 + 1); 629 ip6->ip6_nxt = IPPROTO_TCP; 630 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 631 ip6->ip6_nxt = IPPROTO_TCP; 632 break; 633 #endif 634 #if 0 635 default: 636 /* noone will visit here */ 637 m_freem(m); 638 return EAFNOSUPPORT; 639 #endif 640 } 641 xchg(th->th_dport, th->th_sport, u_int16_t); 642 #undef xchg 643 tlen = 0; /*be friendly with the following code*/ 644 } 645 th->th_seq = htonl(seq); 646 th->th_ack = htonl(ack); 647 th->th_x2 = 0; 648 if ((flags & TH_SYN) == 0) { 649 if (tp) 650 win >>= tp->rcv_scale; 651 if (win > TCP_MAXWIN) 652 win = TCP_MAXWIN; 653 th->th_win = htons((u_int16_t)win); 654 th->th_off = sizeof (struct tcphdr) >> 2; 655 tlen += sizeof(*th); 656 } else 657 tlen += th->th_off << 2; 658 m->m_len = hlen + tlen; 659 m->m_pkthdr.len = hlen + tlen; 660 m->m_pkthdr.rcvif = (struct ifnet *) 0; 661 th->th_flags = flags; 662 th->th_urp = 0; 663 664 switch (family) { 665 #ifdef INET 666 case AF_INET: 667 { 668 struct ipovly *ipov = (struct ipovly *)ip; 669 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 670 ipov->ih_len = htons((u_int16_t)tlen); 671 672 th->th_sum = 0; 673 th->th_sum = in_cksum(m, hlen + tlen); 674 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 675 ip->ip_ttl = ip_defttl; 676 break; 677 } 678 #endif 679 #ifdef INET6 680 case AF_INET6: 681 { 682 th->th_sum = 0; 683 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 684 tlen); 685 ip6->ip6_plen = ntohs(tlen); 686 if (tp && tp->t_in6pcb) { 687 struct ifnet *oifp; 688 ro = (struct route *)&tp->t_in6pcb->in6p_route; 689 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 690 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 691 } else 692 ip6->ip6_hlim = ip6_defhlim; 693 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 694 if (ip6_auto_flowlabel) { 695 ip6->ip6_flow |= 696 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 697 } 698 break; 699 } 700 #endif 701 } 702 703 #ifdef IPSEC 704 (void)ipsec_setsocket(m, NULL); 705 #endif /*IPSEC*/ 706 707 if (tp != NULL && tp->t_inpcb != NULL) { 708 ro = &tp->t_inpcb->inp_route; 709 #ifdef IPSEC 710 if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) { 711 m_freem(m); 712 return ENOBUFS; 713 } 714 #endif 715 #ifdef DIAGNOSTIC 716 if (family != AF_INET) 717 panic("tcp_respond: address family mismatch"); 718 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 719 panic("tcp_respond: ip_dst %x != inp_faddr %x", 720 ntohl(ip->ip_dst.s_addr), 721 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 722 } 723 #endif 724 } 725 #ifdef INET6 726 else if (tp != NULL && tp->t_in6pcb != NULL) { 727 ro = (struct route *)&tp->t_in6pcb->in6p_route; 728 #ifdef IPSEC 729 if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) { 730 m_freem(m); 731 return ENOBUFS; 732 } 733 #endif 734 #ifdef DIAGNOSTIC 735 if (family == AF_INET) { 736 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 737 panic("tcp_respond: not mapped addr"); 738 if (bcmp(&ip->ip_dst, 739 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 740 sizeof(ip->ip_dst)) != 0) { 741 panic("tcp_respond: ip_dst != in6p_faddr"); 742 } 743 } else if (family == AF_INET6) { 744 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 745 panic("tcp_respond: ip6_dst != in6p_faddr"); 746 } else 747 panic("tcp_respond: address family mismatch"); 748 #endif 749 } 750 #endif 751 else 752 ro = NULL; 753 754 switch (family) { 755 #ifdef INET 756 case AF_INET: 757 error = ip_output(m, NULL, ro, 758 (ip_mtudisc ? IP_MTUDISC : 0), 759 NULL); 760 break; 761 #endif 762 #ifdef INET6 763 case AF_INET6: 764 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 765 NULL); 766 break; 767 #endif 768 default: 769 error = EAFNOSUPPORT; 770 break; 771 } 772 773 return (error); 774 } 775 776 /* 777 * Create a new TCP control block, making an 778 * empty reassembly queue and hooking it to the argument 779 * protocol control block. 780 */ 781 struct tcpcb * 782 tcp_newtcpcb(family, aux) 783 int family; /* selects inpcb, or in6pcb */ 784 void *aux; 785 { 786 struct tcpcb *tp; 787 int i; 788 789 switch (family) { 790 case PF_INET: 791 break; 792 #ifdef INET6 793 case PF_INET6: 794 break; 795 #endif 796 default: 797 return NULL; 798 } 799 800 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 801 if (tp == NULL) 802 return (NULL); 803 bzero((caddr_t)tp, sizeof(struct tcpcb)); 804 LIST_INIT(&tp->segq); 805 LIST_INIT(&tp->timeq); 806 tp->t_family = family; /* may be overridden later on */ 807 tp->t_peermss = tcp_mssdflt; 808 tp->t_ourmss = tcp_mssdflt; 809 tp->t_segsz = tcp_mssdflt; 810 LIST_INIT(&tp->t_sc); 811 812 callout_init(&tp->t_delack_ch); 813 for (i = 0; i < TCPT_NTIMERS; i++) 814 TCP_TIMER_INIT(tp, i); 815 816 tp->t_flags = 0; 817 if (tcp_do_rfc1323 && tcp_do_win_scale) 818 tp->t_flags |= TF_REQ_SCALE; 819 if (tcp_do_rfc1323 && tcp_do_timestamps) 820 tp->t_flags |= TF_REQ_TSTMP; 821 if (tcp_do_sack == 2) 822 tp->t_flags |= TF_WILL_SACK; 823 else if (tcp_do_sack == 1) 824 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 825 tp->t_flags |= TF_CANT_TXSACK; 826 switch (family) { 827 case PF_INET: 828 tp->t_inpcb = (struct inpcb *)aux; 829 break; 830 #ifdef INET6 831 case PF_INET6: 832 tp->t_in6pcb = (struct in6pcb *)aux; 833 break; 834 #endif 835 } 836 /* 837 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 838 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 839 * reasonable initial retransmit time. 840 */ 841 tp->t_srtt = TCPTV_SRTTBASE; 842 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 843 tp->t_rttmin = TCPTV_MIN; 844 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 845 TCPTV_MIN, TCPTV_REXMTMAX); 846 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 847 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 848 if (family == AF_INET) { 849 struct inpcb *inp = (struct inpcb *)aux; 850 inp->inp_ip.ip_ttl = ip_defttl; 851 inp->inp_ppcb = (caddr_t)tp; 852 } 853 #ifdef INET6 854 else if (family == AF_INET6) { 855 struct in6pcb *in6p = (struct in6pcb *)aux; 856 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 857 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 858 : NULL); 859 in6p->in6p_ppcb = (caddr_t)tp; 860 } 861 #endif 862 863 /* 864 * Initialize our timebase. When we send timestamps, we take 865 * the delta from tcp_now -- this means each connection always 866 * gets a timebase of 0, which makes it, among other things, 867 * more difficult to determine how long a system has been up, 868 * and thus how many TCP sequence increments have occurred. 869 */ 870 tp->ts_timebase = tcp_now; 871 872 return (tp); 873 } 874 875 /* 876 * Drop a TCP connection, reporting 877 * the specified error. If connection is synchronized, 878 * then send a RST to peer. 879 */ 880 struct tcpcb * 881 tcp_drop(tp, errno) 882 struct tcpcb *tp; 883 int errno; 884 { 885 struct socket *so = NULL; 886 887 #ifdef DIAGNOSTIC 888 if (tp->t_inpcb && tp->t_in6pcb) 889 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 890 #endif 891 #ifdef INET 892 if (tp->t_inpcb) 893 so = tp->t_inpcb->inp_socket; 894 #endif 895 #ifdef INET6 896 if (tp->t_in6pcb) 897 so = tp->t_in6pcb->in6p_socket; 898 #endif 899 if (!so) 900 return NULL; 901 902 if (TCPS_HAVERCVDSYN(tp->t_state)) { 903 tp->t_state = TCPS_CLOSED; 904 (void) tcp_output(tp); 905 tcpstat.tcps_drops++; 906 } else 907 tcpstat.tcps_conndrops++; 908 if (errno == ETIMEDOUT && tp->t_softerror) 909 errno = tp->t_softerror; 910 so->so_error = errno; 911 return (tcp_close(tp)); 912 } 913 914 /* 915 * Close a TCP control block: 916 * discard all space held by the tcp 917 * discard internet protocol block 918 * wake up any sleepers 919 */ 920 struct tcpcb * 921 tcp_close(tp) 922 struct tcpcb *tp; 923 { 924 struct inpcb *inp; 925 #ifdef INET6 926 struct in6pcb *in6p; 927 #endif 928 struct socket *so; 929 #ifdef RTV_RTT 930 struct rtentry *rt; 931 #endif 932 struct route *ro; 933 934 inp = tp->t_inpcb; 935 #ifdef INET6 936 in6p = tp->t_in6pcb; 937 #endif 938 so = NULL; 939 ro = NULL; 940 if (inp) { 941 so = inp->inp_socket; 942 ro = &inp->inp_route; 943 } 944 #ifdef INET6 945 else if (in6p) { 946 so = in6p->in6p_socket; 947 ro = (struct route *)&in6p->in6p_route; 948 } 949 #endif 950 951 #ifdef RTV_RTT 952 /* 953 * If we sent enough data to get some meaningful characteristics, 954 * save them in the routing entry. 'Enough' is arbitrarily 955 * defined as the sendpipesize (default 4K) * 16. This would 956 * give us 16 rtt samples assuming we only get one sample per 957 * window (the usual case on a long haul net). 16 samples is 958 * enough for the srtt filter to converge to within 5% of the correct 959 * value; fewer samples and we could save a very bogus rtt. 960 * 961 * Don't update the default route's characteristics and don't 962 * update anything that the user "locked". 963 */ 964 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 965 ro && (rt = ro->ro_rt) && 966 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 967 u_long i = 0; 968 969 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 970 i = tp->t_srtt * 971 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 972 if (rt->rt_rmx.rmx_rtt && i) 973 /* 974 * filter this update to half the old & half 975 * the new values, converting scale. 976 * See route.h and tcp_var.h for a 977 * description of the scaling constants. 978 */ 979 rt->rt_rmx.rmx_rtt = 980 (rt->rt_rmx.rmx_rtt + i) / 2; 981 else 982 rt->rt_rmx.rmx_rtt = i; 983 } 984 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 985 i = tp->t_rttvar * 986 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 987 if (rt->rt_rmx.rmx_rttvar && i) 988 rt->rt_rmx.rmx_rttvar = 989 (rt->rt_rmx.rmx_rttvar + i) / 2; 990 else 991 rt->rt_rmx.rmx_rttvar = i; 992 } 993 /* 994 * update the pipelimit (ssthresh) if it has been updated 995 * already or if a pipesize was specified & the threshhold 996 * got below half the pipesize. I.e., wait for bad news 997 * before we start updating, then update on both good 998 * and bad news. 999 */ 1000 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1001 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1002 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1003 /* 1004 * convert the limit from user data bytes to 1005 * packets then to packet data bytes. 1006 */ 1007 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1008 if (i < 2) 1009 i = 2; 1010 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1011 if (rt->rt_rmx.rmx_ssthresh) 1012 rt->rt_rmx.rmx_ssthresh = 1013 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1014 else 1015 rt->rt_rmx.rmx_ssthresh = i; 1016 } 1017 } 1018 #endif /* RTV_RTT */ 1019 /* free the reassembly queue, if any */ 1020 TCP_REASS_LOCK(tp); 1021 (void) tcp_freeq(tp); 1022 TCP_REASS_UNLOCK(tp); 1023 1024 tcp_canceltimers(tp); 1025 TCP_CLEAR_DELACK(tp); 1026 syn_cache_cleanup(tp); 1027 1028 if (tp->t_template) { 1029 m_free(tp->t_template); 1030 tp->t_template = NULL; 1031 } 1032 pool_put(&tcpcb_pool, tp); 1033 if (inp) { 1034 inp->inp_ppcb = 0; 1035 soisdisconnected(so); 1036 in_pcbdetach(inp); 1037 } 1038 #ifdef INET6 1039 else if (in6p) { 1040 in6p->in6p_ppcb = 0; 1041 soisdisconnected(so); 1042 in6_pcbdetach(in6p); 1043 } 1044 #endif 1045 tcpstat.tcps_closed++; 1046 return ((struct tcpcb *)0); 1047 } 1048 1049 int 1050 tcp_freeq(tp) 1051 struct tcpcb *tp; 1052 { 1053 struct ipqent *qe; 1054 int rv = 0; 1055 #ifdef TCPREASS_DEBUG 1056 int i = 0; 1057 #endif 1058 1059 TCP_REASS_LOCK_CHECK(tp); 1060 1061 while ((qe = tp->segq.lh_first) != NULL) { 1062 #ifdef TCPREASS_DEBUG 1063 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1064 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1065 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1066 #endif 1067 LIST_REMOVE(qe, ipqe_q); 1068 LIST_REMOVE(qe, ipqe_timeq); 1069 m_freem(qe->ipqe_m); 1070 pool_put(&ipqent_pool, qe); 1071 rv = 1; 1072 } 1073 return (rv); 1074 } 1075 1076 /* 1077 * Protocol drain routine. Called when memory is in short supply. 1078 */ 1079 void 1080 tcp_drain() 1081 { 1082 struct inpcb *inp; 1083 struct tcpcb *tp; 1084 1085 /* 1086 * Free the sequence queue of all TCP connections. 1087 */ 1088 inp = tcbtable.inpt_queue.cqh_first; 1089 if (inp) /* XXX */ 1090 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 1091 inp = inp->inp_queue.cqe_next) { 1092 if ((tp = intotcpcb(inp)) != NULL) { 1093 /* 1094 * We may be called from a device's interrupt 1095 * context. If the tcpcb is already busy, 1096 * just bail out now. 1097 */ 1098 if (tcp_reass_lock_try(tp) == 0) 1099 continue; 1100 if (tcp_freeq(tp)) 1101 tcpstat.tcps_connsdrained++; 1102 TCP_REASS_UNLOCK(tp); 1103 } 1104 } 1105 } 1106 1107 /* 1108 * Notify a tcp user of an asynchronous error; 1109 * store error as soft error, but wake up user 1110 * (for now, won't do anything until can select for soft error). 1111 */ 1112 void 1113 tcp_notify(inp, error) 1114 struct inpcb *inp; 1115 int error; 1116 { 1117 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1118 struct socket *so = inp->inp_socket; 1119 1120 /* 1121 * Ignore some errors if we are hooked up. 1122 * If connection hasn't completed, has retransmitted several times, 1123 * and receives a second error, give up now. This is better 1124 * than waiting a long time to establish a connection that 1125 * can never complete. 1126 */ 1127 if (tp->t_state == TCPS_ESTABLISHED && 1128 (error == EHOSTUNREACH || error == ENETUNREACH || 1129 error == EHOSTDOWN)) { 1130 return; 1131 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1132 tp->t_rxtshift > 3 && tp->t_softerror) 1133 so->so_error = error; 1134 else 1135 tp->t_softerror = error; 1136 wakeup((caddr_t) &so->so_timeo); 1137 sorwakeup(so); 1138 sowwakeup(so); 1139 } 1140 1141 #ifdef INET6 1142 void 1143 tcp6_notify(in6p, error) 1144 struct in6pcb *in6p; 1145 int error; 1146 { 1147 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1148 struct socket *so = in6p->in6p_socket; 1149 1150 /* 1151 * Ignore some errors if we are hooked up. 1152 * If connection hasn't completed, has retransmitted several times, 1153 * and receives a second error, give up now. This is better 1154 * than waiting a long time to establish a connection that 1155 * can never complete. 1156 */ 1157 if (tp->t_state == TCPS_ESTABLISHED && 1158 (error == EHOSTUNREACH || error == ENETUNREACH || 1159 error == EHOSTDOWN)) { 1160 return; 1161 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1162 tp->t_rxtshift > 3 && tp->t_softerror) 1163 so->so_error = error; 1164 else 1165 tp->t_softerror = error; 1166 wakeup((caddr_t) &so->so_timeo); 1167 sorwakeup(so); 1168 sowwakeup(so); 1169 } 1170 #endif 1171 1172 #ifdef INET6 1173 void 1174 tcp6_ctlinput(cmd, sa, d) 1175 int cmd; 1176 struct sockaddr *sa; 1177 void *d; 1178 { 1179 struct tcphdr th; 1180 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1181 int nmatch; 1182 struct ip6_hdr *ip6; 1183 const struct sockaddr_in6 *sa6_src = NULL; 1184 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1185 struct mbuf *m; 1186 int off; 1187 1188 if (sa->sa_family != AF_INET6 || 1189 sa->sa_len != sizeof(struct sockaddr_in6)) 1190 return; 1191 if ((unsigned)cmd >= PRC_NCMDS) 1192 return; 1193 else if (cmd == PRC_QUENCH) { 1194 /* XXX there's no PRC_QUENCH in IPv6 */ 1195 notify = tcp6_quench; 1196 } else if (PRC_IS_REDIRECT(cmd)) 1197 notify = in6_rtchange, d = NULL; 1198 else if (cmd == PRC_MSGSIZE) 1199 ; /* special code is present, see below */ 1200 else if (cmd == PRC_HOSTDEAD) 1201 d = NULL; 1202 else if (inet6ctlerrmap[cmd] == 0) 1203 return; 1204 1205 /* if the parameter is from icmp6, decode it. */ 1206 if (d != NULL) { 1207 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1208 m = ip6cp->ip6c_m; 1209 ip6 = ip6cp->ip6c_ip6; 1210 off = ip6cp->ip6c_off; 1211 sa6_src = ip6cp->ip6c_src; 1212 } else { 1213 m = NULL; 1214 ip6 = NULL; 1215 sa6_src = &sa6_any; 1216 } 1217 1218 if (ip6) { 1219 /* 1220 * XXX: We assume that when ip6 is non NULL, 1221 * M and OFF are valid. 1222 */ 1223 1224 /* check if we can safely examine src and dst ports */ 1225 if (m->m_pkthdr.len < off + sizeof(th)) { 1226 if (cmd == PRC_MSGSIZE) 1227 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1228 return; 1229 } 1230 1231 bzero(&th, sizeof(th)); 1232 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1233 1234 if (cmd == PRC_MSGSIZE) { 1235 int valid = 0; 1236 1237 /* 1238 * Check to see if we have a valid TCP connection 1239 * corresponding to the address in the ICMPv6 message 1240 * payload. 1241 */ 1242 if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr, 1243 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1244 th.th_sport, 0)) 1245 valid++; 1246 1247 /* 1248 * Depending on the value of "valid" and routing table 1249 * size (mtudisc_{hi,lo}wat), we will: 1250 * - recalcurate the new MTU and create the 1251 * corresponding routing entry, or 1252 * - ignore the MTU change notification. 1253 */ 1254 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1255 1256 /* 1257 * no need to call in6_pcbnotify, it should have been 1258 * called via callback if necessary 1259 */ 1260 return; 1261 } 1262 1263 nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport, 1264 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1265 if (nmatch == 0 && syn_cache_count && 1266 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1267 inet6ctlerrmap[cmd] == ENETUNREACH || 1268 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1269 syn_cache_unreach((struct sockaddr *)sa6_src, 1270 sa, &th); 1271 } else { 1272 (void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src, 1273 0, cmd, NULL, notify); 1274 } 1275 } 1276 #endif 1277 1278 #ifdef INET 1279 /* assumes that ip header and tcp header are contiguous on mbuf */ 1280 void * 1281 tcp_ctlinput(cmd, sa, v) 1282 int cmd; 1283 struct sockaddr *sa; 1284 void *v; 1285 { 1286 struct ip *ip = v; 1287 struct tcphdr *th; 1288 struct icmp *icp; 1289 extern int inetctlerrmap[]; 1290 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1291 int errno; 1292 int nmatch; 1293 1294 if (sa->sa_family != AF_INET || 1295 sa->sa_len != sizeof(struct sockaddr_in)) 1296 return NULL; 1297 if ((unsigned)cmd >= PRC_NCMDS) 1298 return NULL; 1299 errno = inetctlerrmap[cmd]; 1300 if (cmd == PRC_QUENCH) 1301 notify = tcp_quench; 1302 else if (PRC_IS_REDIRECT(cmd)) 1303 notify = in_rtchange, ip = 0; 1304 else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) { 1305 /* 1306 * Check to see if we have a valid TCP connection 1307 * corresponding to the address in the ICMP message 1308 * payload. 1309 * 1310 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1311 */ 1312 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1313 if (in_pcblookup_connect(&tcbtable, 1314 ip->ip_dst, th->th_dport, 1315 ip->ip_src, th->th_sport) == NULL) 1316 return NULL; 1317 1318 /* 1319 * Now that we've validated that we are actually communicating 1320 * with the host indicated in the ICMP message, locate the 1321 * ICMP header, recalculate the new MTU, and create the 1322 * corresponding routing entry. 1323 */ 1324 icp = (struct icmp *)((caddr_t)ip - 1325 offsetof(struct icmp, icmp_ip)); 1326 icmp_mtudisc(icp, ip->ip_dst); 1327 1328 return NULL; 1329 } else if (cmd == PRC_HOSTDEAD) 1330 ip = 0; 1331 else if (errno == 0) 1332 return NULL; 1333 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1334 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1335 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1336 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1337 if (nmatch == 0 && syn_cache_count && 1338 (inetctlerrmap[cmd] == EHOSTUNREACH || 1339 inetctlerrmap[cmd] == ENETUNREACH || 1340 inetctlerrmap[cmd] == EHOSTDOWN)) { 1341 struct sockaddr_in sin; 1342 bzero(&sin, sizeof(sin)); 1343 sin.sin_len = sizeof(sin); 1344 sin.sin_family = AF_INET; 1345 sin.sin_port = th->th_sport; 1346 sin.sin_addr = ip->ip_src; 1347 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1348 } 1349 1350 /* XXX mapped address case */ 1351 } else 1352 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1353 notify); 1354 return NULL; 1355 } 1356 1357 /* 1358 * When a source quence is received, we are being notifed of congestion. 1359 * Close the congestion window down to the Loss Window (one segment). 1360 * We will gradually open it again as we proceed. 1361 */ 1362 void 1363 tcp_quench(inp, errno) 1364 struct inpcb *inp; 1365 int errno; 1366 { 1367 struct tcpcb *tp = intotcpcb(inp); 1368 1369 if (tp) 1370 tp->snd_cwnd = tp->t_segsz; 1371 } 1372 #endif 1373 1374 #ifdef INET6 1375 void 1376 tcp6_quench(in6p, errno) 1377 struct in6pcb *in6p; 1378 int errno; 1379 { 1380 struct tcpcb *tp = in6totcpcb(in6p); 1381 1382 if (tp) 1383 tp->snd_cwnd = tp->t_segsz; 1384 } 1385 #endif 1386 1387 #ifdef INET 1388 /* 1389 * Path MTU Discovery handlers. 1390 */ 1391 void 1392 tcp_mtudisc_callback(faddr) 1393 struct in_addr faddr; 1394 { 1395 1396 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1397 } 1398 1399 /* 1400 * On receipt of path MTU corrections, flush old route and replace it 1401 * with the new one. Retransmit all unacknowledged packets, to ensure 1402 * that all packets will be received. 1403 */ 1404 void 1405 tcp_mtudisc(inp, errno) 1406 struct inpcb *inp; 1407 int errno; 1408 { 1409 struct tcpcb *tp = intotcpcb(inp); 1410 struct rtentry *rt = in_pcbrtentry(inp); 1411 1412 if (tp != 0) { 1413 if (rt != 0) { 1414 /* 1415 * If this was not a host route, remove and realloc. 1416 */ 1417 if ((rt->rt_flags & RTF_HOST) == 0) { 1418 in_rtchange(inp, errno); 1419 if ((rt = in_pcbrtentry(inp)) == 0) 1420 return; 1421 } 1422 1423 /* 1424 * Slow start out of the error condition. We 1425 * use the MTU because we know it's smaller 1426 * than the previously transmitted segment. 1427 * 1428 * Note: This is more conservative than the 1429 * suggestion in draft-floyd-incr-init-win-03. 1430 */ 1431 if (rt->rt_rmx.rmx_mtu != 0) 1432 tp->snd_cwnd = 1433 TCP_INITIAL_WINDOW(tcp_init_win, 1434 rt->rt_rmx.rmx_mtu); 1435 } 1436 1437 /* 1438 * Resend unacknowledged packets. 1439 */ 1440 tp->snd_nxt = tp->snd_una; 1441 tcp_output(tp); 1442 } 1443 } 1444 #endif 1445 1446 #ifdef INET6 1447 /* 1448 * Path MTU Discovery handlers. 1449 */ 1450 void 1451 tcp6_mtudisc_callback(faddr) 1452 struct in6_addr *faddr; 1453 { 1454 struct sockaddr_in6 sin6; 1455 1456 bzero(&sin6, sizeof(sin6)); 1457 sin6.sin6_family = AF_INET6; 1458 sin6.sin6_len = sizeof(struct sockaddr_in6); 1459 sin6.sin6_addr = *faddr; 1460 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1461 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1462 } 1463 1464 void 1465 tcp6_mtudisc(in6p, errno) 1466 struct in6pcb *in6p; 1467 int errno; 1468 { 1469 struct tcpcb *tp = in6totcpcb(in6p); 1470 struct rtentry *rt = in6_pcbrtentry(in6p); 1471 1472 if (tp != 0) { 1473 if (rt != 0) { 1474 /* 1475 * If this was not a host route, remove and realloc. 1476 */ 1477 if ((rt->rt_flags & RTF_HOST) == 0) { 1478 in6_rtchange(in6p, errno); 1479 if ((rt = in6_pcbrtentry(in6p)) == 0) 1480 return; 1481 } 1482 1483 /* 1484 * Slow start out of the error condition. We 1485 * use the MTU because we know it's smaller 1486 * than the previously transmitted segment. 1487 * 1488 * Note: This is more conservative than the 1489 * suggestion in draft-floyd-incr-init-win-03. 1490 */ 1491 if (rt->rt_rmx.rmx_mtu != 0) 1492 tp->snd_cwnd = 1493 TCP_INITIAL_WINDOW(tcp_init_win, 1494 rt->rt_rmx.rmx_mtu); 1495 } 1496 1497 /* 1498 * Resend unacknowledged packets. 1499 */ 1500 tp->snd_nxt = tp->snd_una; 1501 tcp_output(tp); 1502 } 1503 } 1504 #endif /* INET6 */ 1505 1506 /* 1507 * Compute the MSS to advertise to the peer. Called only during 1508 * the 3-way handshake. If we are the server (peer initiated 1509 * connection), we are called with a pointer to the interface 1510 * on which the SYN packet arrived. If we are the client (we 1511 * initiated connection), we are called with a pointer to the 1512 * interface out which this connection should go. 1513 * 1514 * NOTE: Do not subtract IP option/extension header size nor IPsec 1515 * header size from MSS advertisement. MSS option must hold the maximum 1516 * segment size we can accept, so it must always be: 1517 * max(if mtu) - ip header - tcp header 1518 */ 1519 u_long 1520 tcp_mss_to_advertise(ifp, af) 1521 const struct ifnet *ifp; 1522 int af; 1523 { 1524 extern u_long in_maxmtu; 1525 u_long mss = 0; 1526 u_long hdrsiz; 1527 1528 /* 1529 * In order to avoid defeating path MTU discovery on the peer, 1530 * we advertise the max MTU of all attached networks as our MSS, 1531 * per RFC 1191, section 3.1. 1532 * 1533 * We provide the option to advertise just the MTU of 1534 * the interface on which we hope this connection will 1535 * be receiving. If we are responding to a SYN, we 1536 * will have a pretty good idea about this, but when 1537 * initiating a connection there is a bit more doubt. 1538 * 1539 * We also need to ensure that loopback has a large enough 1540 * MSS, as the loopback MTU is never included in in_maxmtu. 1541 */ 1542 1543 if (ifp != NULL) 1544 mss = ifp->if_mtu; 1545 1546 if (tcp_mss_ifmtu == 0) 1547 switch (af) { 1548 case AF_INET: 1549 mss = max(in_maxmtu, mss); 1550 break; 1551 #ifdef INET6 1552 case AF_INET6: 1553 mss = max(in6_maxmtu, mss); 1554 break; 1555 #endif 1556 } 1557 1558 switch (af) { 1559 case AF_INET: 1560 hdrsiz = sizeof(struct ip); 1561 break; 1562 #ifdef INET6 1563 case AF_INET6: 1564 hdrsiz = sizeof(struct ip6_hdr); 1565 break; 1566 #endif 1567 default: 1568 hdrsiz = 0; 1569 break; 1570 } 1571 hdrsiz += sizeof(struct tcphdr); 1572 if (mss > hdrsiz) 1573 mss -= hdrsiz; 1574 1575 mss = max(tcp_mssdflt, mss); 1576 return (mss); 1577 } 1578 1579 /* 1580 * Set connection variables based on the peer's advertised MSS. 1581 * We are passed the TCPCB for the actual connection. If we 1582 * are the server, we are called by the compressed state engine 1583 * when the 3-way handshake is complete. If we are the client, 1584 * we are called when we receive the SYN,ACK from the server. 1585 * 1586 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1587 * before this routine is called! 1588 */ 1589 void 1590 tcp_mss_from_peer(tp, offer) 1591 struct tcpcb *tp; 1592 int offer; 1593 { 1594 struct socket *so; 1595 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1596 struct rtentry *rt; 1597 #endif 1598 u_long bufsize; 1599 int mss; 1600 1601 #ifdef DIAGNOSTIC 1602 if (tp->t_inpcb && tp->t_in6pcb) 1603 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1604 #endif 1605 so = NULL; 1606 rt = NULL; 1607 #ifdef INET 1608 if (tp->t_inpcb) { 1609 so = tp->t_inpcb->inp_socket; 1610 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1611 rt = in_pcbrtentry(tp->t_inpcb); 1612 #endif 1613 } 1614 #endif 1615 #ifdef INET6 1616 if (tp->t_in6pcb) { 1617 so = tp->t_in6pcb->in6p_socket; 1618 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1619 rt = in6_pcbrtentry(tp->t_in6pcb); 1620 #endif 1621 } 1622 #endif 1623 1624 /* 1625 * As per RFC1122, use the default MSS value, unless they 1626 * sent us an offer. Do not accept offers less than 32 bytes. 1627 */ 1628 mss = tcp_mssdflt; 1629 if (offer) 1630 mss = offer; 1631 mss = max(mss, 32); /* sanity */ 1632 tp->t_peermss = mss; 1633 mss -= tcp_optlen(tp); 1634 #ifdef INET 1635 if (tp->t_inpcb) 1636 mss -= ip_optlen(tp->t_inpcb); 1637 #endif 1638 #ifdef INET6 1639 if (tp->t_in6pcb) 1640 mss -= ip6_optlen(tp->t_in6pcb); 1641 #endif 1642 1643 /* 1644 * If there's a pipesize, change the socket buffer to that size. 1645 * Make the socket buffer an integral number of MSS units. If 1646 * the MSS is larger than the socket buffer, artificially decrease 1647 * the MSS. 1648 */ 1649 #ifdef RTV_SPIPE 1650 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1651 bufsize = rt->rt_rmx.rmx_sendpipe; 1652 else 1653 #endif 1654 bufsize = so->so_snd.sb_hiwat; 1655 if (bufsize < mss) 1656 mss = bufsize; 1657 else { 1658 bufsize = roundup(bufsize, mss); 1659 if (bufsize > sb_max) 1660 bufsize = sb_max; 1661 (void) sbreserve(&so->so_snd, bufsize); 1662 } 1663 tp->t_segsz = mss; 1664 1665 #ifdef RTV_SSTHRESH 1666 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1667 /* 1668 * There's some sort of gateway or interface buffer 1669 * limit on the path. Use this to set the slow 1670 * start threshold, but set the threshold to no less 1671 * than 2 * MSS. 1672 */ 1673 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1674 } 1675 #endif 1676 } 1677 1678 /* 1679 * Processing necessary when a TCP connection is established. 1680 */ 1681 void 1682 tcp_established(tp) 1683 struct tcpcb *tp; 1684 { 1685 struct socket *so; 1686 #ifdef RTV_RPIPE 1687 struct rtentry *rt; 1688 #endif 1689 u_long bufsize; 1690 1691 #ifdef DIAGNOSTIC 1692 if (tp->t_inpcb && tp->t_in6pcb) 1693 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1694 #endif 1695 so = NULL; 1696 rt = NULL; 1697 #ifdef INET 1698 if (tp->t_inpcb) { 1699 so = tp->t_inpcb->inp_socket; 1700 #if defined(RTV_RPIPE) 1701 rt = in_pcbrtentry(tp->t_inpcb); 1702 #endif 1703 } 1704 #endif 1705 #ifdef INET6 1706 if (tp->t_in6pcb) { 1707 so = tp->t_in6pcb->in6p_socket; 1708 #if defined(RTV_RPIPE) 1709 rt = in6_pcbrtentry(tp->t_in6pcb); 1710 #endif 1711 } 1712 #endif 1713 1714 tp->t_state = TCPS_ESTABLISHED; 1715 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1716 1717 #ifdef RTV_RPIPE 1718 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1719 bufsize = rt->rt_rmx.rmx_recvpipe; 1720 else 1721 #endif 1722 bufsize = so->so_rcv.sb_hiwat; 1723 if (bufsize > tp->t_ourmss) { 1724 bufsize = roundup(bufsize, tp->t_ourmss); 1725 if (bufsize > sb_max) 1726 bufsize = sb_max; 1727 (void) sbreserve(&so->so_rcv, bufsize); 1728 } 1729 } 1730 1731 /* 1732 * Check if there's an initial rtt or rttvar. Convert from the 1733 * route-table units to scaled multiples of the slow timeout timer. 1734 * Called only during the 3-way handshake. 1735 */ 1736 void 1737 tcp_rmx_rtt(tp) 1738 struct tcpcb *tp; 1739 { 1740 #ifdef RTV_RTT 1741 struct rtentry *rt = NULL; 1742 int rtt; 1743 1744 #ifdef DIAGNOSTIC 1745 if (tp->t_inpcb && tp->t_in6pcb) 1746 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1747 #endif 1748 #ifdef INET 1749 if (tp->t_inpcb) 1750 rt = in_pcbrtentry(tp->t_inpcb); 1751 #endif 1752 #ifdef INET6 1753 if (tp->t_in6pcb) 1754 rt = in6_pcbrtentry(tp->t_in6pcb); 1755 #endif 1756 if (rt == NULL) 1757 return; 1758 1759 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1760 /* 1761 * XXX The lock bit for MTU indicates that the value 1762 * is also a minimum value; this is subject to time. 1763 */ 1764 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1765 TCPT_RANGESET(tp->t_rttmin, 1766 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1767 TCPTV_MIN, TCPTV_REXMTMAX); 1768 tp->t_srtt = rtt / 1769 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1770 if (rt->rt_rmx.rmx_rttvar) { 1771 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1772 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1773 (TCP_RTTVAR_SHIFT + 2)); 1774 } else { 1775 /* Default variation is +- 1 rtt */ 1776 tp->t_rttvar = 1777 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1778 } 1779 TCPT_RANGESET(tp->t_rxtcur, 1780 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1781 tp->t_rttmin, TCPTV_REXMTMAX); 1782 } 1783 #endif 1784 } 1785 1786 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1787 #if NRND > 0 1788 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1789 #endif 1790 1791 /* 1792 * Get a new sequence value given a tcp control block 1793 */ 1794 tcp_seq 1795 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1796 { 1797 1798 #ifdef INET 1799 if (tp->t_inpcb != NULL) { 1800 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1801 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1802 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1803 addin)); 1804 } 1805 #endif 1806 #ifdef INET6 1807 if (tp->t_in6pcb != NULL) { 1808 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 1809 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 1810 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 1811 addin)); 1812 } 1813 #endif 1814 /* Not possible. */ 1815 panic("tcp_new_iss"); 1816 } 1817 1818 /* 1819 * This routine actually generates a new TCP initial sequence number. 1820 */ 1821 tcp_seq 1822 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 1823 size_t addrsz, tcp_seq addin) 1824 { 1825 tcp_seq tcp_iss; 1826 1827 #if NRND > 0 1828 static int beenhere; 1829 1830 /* 1831 * If we haven't been here before, initialize our cryptographic 1832 * hash secret. 1833 */ 1834 if (beenhere == 0) { 1835 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 1836 RND_EXTRACT_ANY); 1837 beenhere = 1; 1838 } 1839 1840 if (tcp_do_rfc1948) { 1841 MD5_CTX ctx; 1842 u_int8_t hash[16]; /* XXX MD5 knowledge */ 1843 1844 /* 1845 * Compute the base value of the ISS. It is a hash 1846 * of (saddr, sport, daddr, dport, secret). 1847 */ 1848 MD5Init(&ctx); 1849 1850 MD5Update(&ctx, (u_char *) laddr, addrsz); 1851 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 1852 1853 MD5Update(&ctx, (u_char *) faddr, addrsz); 1854 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 1855 1856 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 1857 1858 MD5Final(hash, &ctx); 1859 1860 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 1861 1862 /* 1863 * Now increment our "timer", and add it in to 1864 * the computed value. 1865 * 1866 * XXX Use `addin'? 1867 * XXX TCP_ISSINCR too large to use? 1868 */ 1869 tcp_iss_seq += TCP_ISSINCR; 1870 #ifdef TCPISS_DEBUG 1871 printf("ISS hash 0x%08x, ", tcp_iss); 1872 #endif 1873 tcp_iss += tcp_iss_seq + addin; 1874 #ifdef TCPISS_DEBUG 1875 printf("new ISS 0x%08x\n", tcp_iss); 1876 #endif 1877 } else 1878 #endif /* NRND > 0 */ 1879 { 1880 /* 1881 * Randomize. 1882 */ 1883 #if NRND > 0 1884 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 1885 #else 1886 tcp_iss = random(); 1887 #endif 1888 1889 /* 1890 * If we were asked to add some amount to a known value, 1891 * we will take a random value obtained above, mask off 1892 * the upper bits, and add in the known value. We also 1893 * add in a constant to ensure that we are at least a 1894 * certain distance from the original value. 1895 * 1896 * This is used when an old connection is in timed wait 1897 * and we have a new one coming in, for instance. 1898 */ 1899 if (addin != 0) { 1900 #ifdef TCPISS_DEBUG 1901 printf("Random %08x, ", tcp_iss); 1902 #endif 1903 tcp_iss &= TCP_ISS_RANDOM_MASK; 1904 tcp_iss += addin + TCP_ISSINCR; 1905 #ifdef TCPISS_DEBUG 1906 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 1907 #endif 1908 } else { 1909 tcp_iss &= TCP_ISS_RANDOM_MASK; 1910 tcp_iss += tcp_iss_seq; 1911 tcp_iss_seq += TCP_ISSINCR; 1912 #ifdef TCPISS_DEBUG 1913 printf("ISS %08x\n", tcp_iss); 1914 #endif 1915 } 1916 } 1917 1918 if (tcp_compat_42) { 1919 /* 1920 * Limit it to the positive range for really old TCP 1921 * implementations. 1922 */ 1923 if (tcp_iss >= 0x80000000) 1924 tcp_iss &= 0x7fffffff; /* XXX */ 1925 } 1926 1927 return (tcp_iss); 1928 } 1929 1930 #ifdef IPSEC 1931 /* compute ESP/AH header size for TCP, including outer IP header. */ 1932 size_t 1933 ipsec4_hdrsiz_tcp(tp) 1934 struct tcpcb *tp; 1935 { 1936 struct inpcb *inp; 1937 size_t hdrsiz; 1938 1939 /* XXX mapped addr case (tp->t_in6pcb) */ 1940 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 1941 return 0; 1942 switch (tp->t_family) { 1943 case AF_INET: 1944 /* XXX: should use currect direction. */ 1945 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 1946 break; 1947 default: 1948 hdrsiz = 0; 1949 break; 1950 } 1951 1952 return hdrsiz; 1953 } 1954 1955 #ifdef INET6 1956 size_t 1957 ipsec6_hdrsiz_tcp(tp) 1958 struct tcpcb *tp; 1959 { 1960 struct in6pcb *in6p; 1961 size_t hdrsiz; 1962 1963 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 1964 return 0; 1965 switch (tp->t_family) { 1966 case AF_INET6: 1967 /* XXX: should use currect direction. */ 1968 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 1969 break; 1970 case AF_INET: 1971 /* mapped address case - tricky */ 1972 default: 1973 hdrsiz = 0; 1974 break; 1975 } 1976 1977 return hdrsiz; 1978 } 1979 #endif 1980 #endif /*IPSEC*/ 1981 1982 /* 1983 * Determine the length of the TCP options for this connection. 1984 * 1985 * XXX: What do we do for SACK, when we add that? Just reserve 1986 * all of the space? Otherwise we can't exactly be incrementing 1987 * cwnd by an amount that varies depending on the amount we last 1988 * had to SACK! 1989 */ 1990 1991 u_int 1992 tcp_optlen(tp) 1993 struct tcpcb *tp; 1994 { 1995 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1996 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1997 return TCPOLEN_TSTAMP_APPA; 1998 else 1999 return 0; 2000 } 2001
Indexes created Mon Sep 22 13:09:51 GMT 2025