1 /* $NetBSD: tcp_subr.c,v 1.105 2000/12/21 00:45:17 itojun 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 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 "rnd.h" 108 109 #include <sys/param.h> 110 #include <sys/proc.h> 111 #include <sys/systm.h> 112 #include <sys/malloc.h> 113 #include <sys/mbuf.h> 114 #include <sys/socket.h> 115 #include <sys/socketvar.h> 116 #include <sys/protosw.h> 117 #include <sys/errno.h> 118 #include <sys/kernel.h> 119 #include <sys/pool.h> 120 #if NRND > 0 121 #include <sys/rnd.h> 122 #endif 123 124 #include <net/route.h> 125 #include <net/if.h> 126 127 #include <netinet/in.h> 128 #include <netinet/in_systm.h> 129 #include <netinet/ip.h> 130 #include <netinet/in_pcb.h> 131 #include <netinet/ip_var.h> 132 #include <netinet/ip_icmp.h> 133 134 #ifdef INET6 135 #ifndef INET 136 #include <netinet/in.h> 137 #endif 138 #include <netinet/ip6.h> 139 #include <netinet6/in6_pcb.h> 140 #include <netinet6/ip6_var.h> 141 #include <netinet6/in6_var.h> 142 #include <netinet6/ip6protosw.h> 143 #include <netinet/icmp6.h> 144 #endif 145 146 #include <netinet/tcp.h> 147 #include <netinet/tcp_fsm.h> 148 #include <netinet/tcp_seq.h> 149 #include <netinet/tcp_timer.h> 150 #include <netinet/tcp_var.h> 151 #include <netinet/tcpip.h> 152 153 #ifdef IPSEC 154 #include <netinet6/ipsec.h> 155 #endif /*IPSEC*/ 156 157 #ifdef INET6 158 struct in6pcb tcb6; 159 #endif 160 161 /* patchable/settable parameters for tcp */ 162 int tcp_mssdflt = TCP_MSS; 163 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 164 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 165 int tcp_do_sack = 1; /* selective acknowledgement */ 166 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 167 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 168 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 169 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 170 int tcp_init_win = 1; 171 int tcp_mss_ifmtu = 0; 172 #ifdef TCP_COMPAT_42 173 int tcp_compat_42 = 1; 174 #else 175 int tcp_compat_42 = 0; 176 #endif 177 int tcp_rst_ppslim = 100; /* 100pps */ 178 179 /* tcb hash */ 180 #ifndef TCBHASHSIZE 181 #define TCBHASHSIZE 128 182 #endif 183 int tcbhashsize = TCBHASHSIZE; 184 185 /* syn hash parameters */ 186 #define TCP_SYN_HASH_SIZE 293 187 #define TCP_SYN_BUCKET_SIZE 35 188 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 189 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 190 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 191 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 192 int tcp_syn_cache_interval = 1; /* runs timer twice a second */ 193 194 int tcp_freeq __P((struct tcpcb *)); 195 196 #ifdef INET 197 void tcp_mtudisc_callback __P((struct in_addr)); 198 #endif 199 #ifdef INET6 200 void tcp6_mtudisc_callback __P((struct in6_addr *)); 201 #endif 202 203 void tcp_mtudisc __P((struct inpcb *, int)); 204 #ifdef INET6 205 void tcp6_mtudisc __P((struct in6pcb *, int)); 206 #endif 207 208 struct pool tcpcb_pool; 209 210 /* 211 * Tcp initialization 212 */ 213 void 214 tcp_init() 215 { 216 int hlen; 217 218 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 219 0, NULL, NULL, M_PCB); 220 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 221 #ifdef INET6 222 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 223 #endif 224 LIST_INIT(&tcp_delacks); 225 226 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 227 #ifdef INET6 228 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 229 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 230 #endif 231 if (max_protohdr < hlen) 232 max_protohdr = hlen; 233 if (max_linkhdr + hlen > MHLEN) 234 panic("tcp_init"); 235 236 #ifdef INET 237 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 238 #endif 239 #ifdef INET6 240 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 241 #endif 242 243 /* Initialize the compressed state engine. */ 244 syn_cache_init(); 245 } 246 247 /* 248 * Create template to be used to send tcp packets on a connection. 249 * Call after host entry created, allocates an mbuf and fills 250 * in a skeletal tcp/ip header, minimizing the amount of work 251 * necessary when the connection is used. 252 */ 253 struct mbuf * 254 tcp_template(tp) 255 struct tcpcb *tp; 256 { 257 struct inpcb *inp = tp->t_inpcb; 258 #ifdef INET6 259 struct in6pcb *in6p = tp->t_in6pcb; 260 #endif 261 struct tcphdr *n; 262 struct mbuf *m; 263 int hlen; 264 265 switch (tp->t_family) { 266 case AF_INET: 267 hlen = sizeof(struct ip); 268 if (inp) 269 break; 270 #ifdef INET6 271 if (in6p) { 272 /* mapped addr case */ 273 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 274 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 275 break; 276 } 277 #endif 278 return NULL; /*EINVAL*/ 279 #ifdef INET6 280 case AF_INET6: 281 hlen = sizeof(struct ip6_hdr); 282 if (in6p) { 283 /* more sainty check? */ 284 break; 285 } 286 return NULL; /*EINVAL*/ 287 #endif 288 default: 289 hlen = 0; /*pacify gcc*/ 290 return NULL; /*EAFNOSUPPORT*/ 291 } 292 #ifdef DIAGNOSTIC 293 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 294 panic("mclbytes too small for t_template"); 295 #endif 296 m = tp->t_template; 297 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 298 ; 299 else { 300 if (m) 301 m_freem(m); 302 m = tp->t_template = NULL; 303 MGETHDR(m, M_DONTWAIT, MT_HEADER); 304 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 305 MCLGET(m, M_DONTWAIT); 306 if ((m->m_flags & M_EXT) == 0) { 307 m_free(m); 308 m = NULL; 309 } 310 } 311 if (m == NULL) 312 return NULL; 313 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 314 } 315 bzero(mtod(m, caddr_t), m->m_len); 316 switch (tp->t_family) { 317 case AF_INET: 318 { 319 struct ipovly *ipov; 320 mtod(m, struct ip *)->ip_v = 4; 321 ipov = mtod(m, struct ipovly *); 322 ipov->ih_pr = IPPROTO_TCP; 323 ipov->ih_len = htons(sizeof(struct tcphdr)); 324 if (inp) { 325 ipov->ih_src = inp->inp_laddr; 326 ipov->ih_dst = inp->inp_faddr; 327 } 328 #ifdef INET6 329 else if (in6p) { 330 /* mapped addr case */ 331 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 332 sizeof(ipov->ih_src)); 333 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 334 sizeof(ipov->ih_dst)); 335 } 336 #endif 337 break; 338 } 339 #ifdef INET6 340 case AF_INET6: 341 { 342 struct ip6_hdr *ip6; 343 mtod(m, struct ip *)->ip_v = 6; 344 ip6 = mtod(m, struct ip6_hdr *); 345 ip6->ip6_nxt = IPPROTO_TCP; 346 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 347 ip6->ip6_src = in6p->in6p_laddr; 348 ip6->ip6_dst = in6p->in6p_faddr; 349 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 350 if (ip6_auto_flowlabel) { 351 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 352 ip6->ip6_flow |= 353 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 354 } 355 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 356 ip6->ip6_vfc |= IPV6_VERSION; 357 break; 358 } 359 #endif 360 } 361 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 362 if (inp) { 363 n->th_sport = inp->inp_lport; 364 n->th_dport = inp->inp_fport; 365 } 366 #ifdef INET6 367 else if (in6p) { 368 n->th_sport = in6p->in6p_lport; 369 n->th_dport = in6p->in6p_fport; 370 } 371 #endif 372 n->th_seq = 0; 373 n->th_ack = 0; 374 n->th_x2 = 0; 375 n->th_off = 5; 376 n->th_flags = 0; 377 n->th_win = 0; 378 n->th_sum = 0; 379 n->th_urp = 0; 380 return (m); 381 } 382 383 /* 384 * Send a single message to the TCP at address specified by 385 * the given TCP/IP header. If m == 0, then we make a copy 386 * of the tcpiphdr at ti and send directly to the addressed host. 387 * This is used to force keep alive messages out using the TCP 388 * template for a connection tp->t_template. If flags are given 389 * then we send a message back to the TCP which originated the 390 * segment ti, and discard the mbuf containing it and any other 391 * attached mbufs. 392 * 393 * In any case the ack and sequence number of the transmitted 394 * segment are as specified by the parameters. 395 */ 396 int 397 tcp_respond(tp, template, m, th0, ack, seq, flags) 398 struct tcpcb *tp; 399 struct mbuf *template; 400 struct mbuf *m; 401 struct tcphdr *th0; 402 tcp_seq ack, seq; 403 int flags; 404 { 405 struct route *ro; 406 int error, tlen, win = 0; 407 int hlen; 408 struct ip *ip; 409 #ifdef INET6 410 struct ip6_hdr *ip6; 411 #endif 412 int family; /* family on packet, not inpcb/in6pcb! */ 413 struct tcphdr *th; 414 415 if (tp != NULL && (flags & TH_RST) == 0) { 416 #ifdef DIAGNOSTIC 417 if (tp->t_inpcb && tp->t_in6pcb) 418 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 419 #endif 420 #ifdef INET 421 if (tp->t_inpcb) 422 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 423 #endif 424 #ifdef INET6 425 if (tp->t_in6pcb) 426 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 427 #endif 428 } 429 430 ip = NULL; 431 #ifdef INET6 432 ip6 = NULL; 433 #endif 434 if (m == 0) { 435 if (!template) 436 return EINVAL; 437 438 /* get family information from template */ 439 switch (mtod(template, struct ip *)->ip_v) { 440 case 4: 441 family = AF_INET; 442 hlen = sizeof(struct ip); 443 break; 444 #ifdef INET6 445 case 6: 446 family = AF_INET6; 447 hlen = sizeof(struct ip6_hdr); 448 break; 449 #endif 450 default: 451 return EAFNOSUPPORT; 452 } 453 454 MGETHDR(m, M_DONTWAIT, MT_HEADER); 455 if (m) { 456 MCLGET(m, M_DONTWAIT); 457 if ((m->m_flags & M_EXT) == 0) { 458 m_free(m); 459 m = NULL; 460 } 461 } 462 if (m == NULL) 463 return (ENOBUFS); 464 465 if (tcp_compat_42) 466 tlen = 1; 467 else 468 tlen = 0; 469 470 m->m_data += max_linkhdr; 471 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 472 template->m_len); 473 switch (family) { 474 case AF_INET: 475 ip = mtod(m, struct ip *); 476 th = (struct tcphdr *)(ip + 1); 477 break; 478 #ifdef INET6 479 case AF_INET6: 480 ip6 = mtod(m, struct ip6_hdr *); 481 th = (struct tcphdr *)(ip6 + 1); 482 break; 483 #endif 484 #if 0 485 default: 486 /* noone will visit here */ 487 m_freem(m); 488 return EAFNOSUPPORT; 489 #endif 490 } 491 flags = TH_ACK; 492 } else { 493 494 if ((m->m_flags & M_PKTHDR) == 0) { 495 #if 0 496 printf("non PKTHDR to tcp_respond\n"); 497 #endif 498 m_freem(m); 499 return EINVAL; 500 } 501 #ifdef DIAGNOSTIC 502 if (!th0) 503 panic("th0 == NULL in tcp_respond"); 504 #endif 505 506 /* get family information from m */ 507 switch (mtod(m, struct ip *)->ip_v) { 508 case 4: 509 family = AF_INET; 510 hlen = sizeof(struct ip); 511 ip = mtod(m, struct ip *); 512 break; 513 #ifdef INET6 514 case 6: 515 family = AF_INET6; 516 hlen = sizeof(struct ip6_hdr); 517 ip6 = mtod(m, struct ip6_hdr *); 518 break; 519 #endif 520 default: 521 m_freem(m); 522 return EAFNOSUPPORT; 523 } 524 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 525 tlen = sizeof(*th0); 526 else 527 tlen = th0->th_off << 2; 528 529 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 530 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 531 m->m_len = hlen + tlen; 532 m_freem(m->m_next); 533 m->m_next = NULL; 534 } else { 535 struct mbuf *n; 536 537 #ifdef DIAGNOSTIC 538 if (max_linkhdr + hlen + tlen > MCLBYTES) { 539 m_freem(m); 540 return EMSGSIZE; 541 } 542 #endif 543 MGETHDR(n, M_DONTWAIT, MT_HEADER); 544 if (n && max_linkhdr + hlen + tlen > MHLEN) { 545 MCLGET(n, M_DONTWAIT); 546 if ((n->m_flags & M_EXT) == 0) { 547 m_freem(n); 548 n = NULL; 549 } 550 } 551 if (!n) { 552 m_freem(m); 553 return ENOBUFS; 554 } 555 556 n->m_data += max_linkhdr; 557 n->m_len = hlen + tlen; 558 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 559 m_copyback(n, hlen, tlen, (caddr_t)th0); 560 561 m_freem(m); 562 m = n; 563 n = NULL; 564 } 565 566 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 567 switch (family) { 568 case AF_INET: 569 ip = mtod(m, struct ip *); 570 th = (struct tcphdr *)(ip + 1); 571 ip->ip_p = IPPROTO_TCP; 572 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 573 ip->ip_p = IPPROTO_TCP; 574 break; 575 #ifdef INET6 576 case AF_INET6: 577 ip6 = mtod(m, struct ip6_hdr *); 578 th = (struct tcphdr *)(ip6 + 1); 579 ip6->ip6_nxt = IPPROTO_TCP; 580 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 581 ip6->ip6_nxt = IPPROTO_TCP; 582 break; 583 #endif 584 #if 0 585 default: 586 /* noone will visit here */ 587 m_freem(m); 588 return EAFNOSUPPORT; 589 #endif 590 } 591 xchg(th->th_dport, th->th_sport, u_int16_t); 592 #undef xchg 593 tlen = 0; /*be friendly with the following code*/ 594 } 595 th->th_seq = htonl(seq); 596 th->th_ack = htonl(ack); 597 th->th_x2 = 0; 598 if ((flags & TH_SYN) == 0) { 599 if (tp) 600 win >>= tp->rcv_scale; 601 if (win > TCP_MAXWIN) 602 win = TCP_MAXWIN; 603 th->th_win = htons((u_int16_t)win); 604 th->th_off = sizeof (struct tcphdr) >> 2; 605 tlen += sizeof(*th); 606 } else 607 tlen += th->th_off << 2; 608 m->m_len = hlen + tlen; 609 m->m_pkthdr.len = hlen + tlen; 610 m->m_pkthdr.rcvif = (struct ifnet *) 0; 611 th->th_flags = flags; 612 th->th_urp = 0; 613 614 switch (family) { 615 #ifdef INET 616 case AF_INET: 617 { 618 struct ipovly *ipov = (struct ipovly *)ip; 619 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 620 ipov->ih_len = htons((u_int16_t)tlen); 621 622 th->th_sum = 0; 623 th->th_sum = in_cksum(m, hlen + tlen); 624 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 625 ip->ip_ttl = ip_defttl; 626 break; 627 } 628 #endif 629 #ifdef INET6 630 case AF_INET6: 631 { 632 th->th_sum = 0; 633 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 634 tlen); 635 ip6->ip6_plen = ntohs(tlen); 636 if (tp && tp->t_in6pcb) { 637 struct ifnet *oifp; 638 ro = (struct route *)&tp->t_in6pcb->in6p_route; 639 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 640 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 641 } else 642 ip6->ip6_hlim = ip6_defhlim; 643 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 644 if (ip6_auto_flowlabel) { 645 ip6->ip6_flow |= 646 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 647 } 648 break; 649 } 650 #endif 651 } 652 653 #ifdef IPSEC 654 ipsec_setsocket(m, NULL); 655 #endif /*IPSEC*/ 656 657 if (tp != NULL && tp->t_inpcb != NULL) { 658 ro = &tp->t_inpcb->inp_route; 659 #ifdef IPSEC 660 ipsec_setsocket(m, tp->t_inpcb->inp_socket); 661 #endif 662 #ifdef DIAGNOSTIC 663 if (family != AF_INET) 664 panic("tcp_respond: address family mismatch"); 665 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 666 panic("tcp_respond: ip_dst %x != inp_faddr %x", 667 ntohl(ip->ip_dst.s_addr), 668 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 669 } 670 #endif 671 } 672 #ifdef INET6 673 else if (tp != NULL && tp->t_in6pcb != NULL) { 674 ro = (struct route *)&tp->t_in6pcb->in6p_route; 675 #ifdef IPSEC 676 ipsec_setsocket(m, tp->t_in6pcb->in6p_socket); 677 #endif 678 #ifdef DIAGNOSTIC 679 if (family == AF_INET) { 680 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 681 panic("tcp_respond: not mapped addr"); 682 if (bcmp(&ip->ip_dst, 683 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 684 sizeof(ip->ip_dst)) != 0) { 685 panic("tcp_respond: ip_dst != in6p_faddr"); 686 } 687 } else if (family == AF_INET6) { 688 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 689 panic("tcp_respond: ip6_dst != in6p_faddr"); 690 } else 691 panic("tcp_respond: address family mismatch"); 692 #endif 693 } 694 #endif 695 else 696 ro = NULL; 697 698 switch (family) { 699 #ifdef INET 700 case AF_INET: 701 error = ip_output(m, NULL, ro, 702 (ip_mtudisc ? IP_MTUDISC : 0), 703 NULL); 704 break; 705 #endif 706 #ifdef INET6 707 case AF_INET6: 708 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 709 NULL); 710 break; 711 #endif 712 default: 713 error = EAFNOSUPPORT; 714 break; 715 } 716 717 return (error); 718 } 719 720 /* 721 * Create a new TCP control block, making an 722 * empty reassembly queue and hooking it to the argument 723 * protocol control block. 724 */ 725 struct tcpcb * 726 tcp_newtcpcb(family, aux) 727 int family; /* selects inpcb, or in6pcb */ 728 void *aux; 729 { 730 struct tcpcb *tp; 731 732 switch (family) { 733 case PF_INET: 734 break; 735 #ifdef INET6 736 case PF_INET6: 737 break; 738 #endif 739 default: 740 return NULL; 741 } 742 743 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 744 if (tp == NULL) 745 return (NULL); 746 bzero((caddr_t)tp, sizeof(struct tcpcb)); 747 LIST_INIT(&tp->segq); 748 LIST_INIT(&tp->timeq); 749 tp->t_family = family; /* may be overridden later on */ 750 tp->t_peermss = tcp_mssdflt; 751 tp->t_ourmss = tcp_mssdflt; 752 tp->t_segsz = tcp_mssdflt; 753 LIST_INIT(&tp->t_sc); 754 755 tp->t_flags = 0; 756 if (tcp_do_rfc1323 && tcp_do_win_scale) 757 tp->t_flags |= TF_REQ_SCALE; 758 if (tcp_do_rfc1323 && tcp_do_timestamps) 759 tp->t_flags |= TF_REQ_TSTMP; 760 if (tcp_do_sack == 2) 761 tp->t_flags |= TF_WILL_SACK; 762 else if (tcp_do_sack == 1) 763 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 764 tp->t_flags |= TF_CANT_TXSACK; 765 switch (family) { 766 case PF_INET: 767 tp->t_inpcb = (struct inpcb *)aux; 768 break; 769 #ifdef INET6 770 case PF_INET6: 771 tp->t_in6pcb = (struct in6pcb *)aux; 772 break; 773 #endif 774 } 775 /* 776 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 777 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 778 * reasonable initial retransmit time. 779 */ 780 tp->t_srtt = TCPTV_SRTTBASE; 781 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 782 tp->t_rttmin = TCPTV_MIN; 783 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 784 TCPTV_MIN, TCPTV_REXMTMAX); 785 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 786 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 787 if (family == AF_INET) { 788 struct inpcb *inp = (struct inpcb *)aux; 789 inp->inp_ip.ip_ttl = ip_defttl; 790 inp->inp_ppcb = (caddr_t)tp; 791 } 792 #ifdef INET6 793 else if (family == AF_INET6) { 794 struct in6pcb *in6p = (struct in6pcb *)aux; 795 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 796 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 797 : NULL); 798 in6p->in6p_ppcb = (caddr_t)tp; 799 } 800 #endif 801 return (tp); 802 } 803 804 /* 805 * Drop a TCP connection, reporting 806 * the specified error. If connection is synchronized, 807 * then send a RST to peer. 808 */ 809 struct tcpcb * 810 tcp_drop(tp, errno) 811 struct tcpcb *tp; 812 int errno; 813 { 814 struct socket *so = NULL; 815 816 #ifdef DIAGNOSTIC 817 if (tp->t_inpcb && tp->t_in6pcb) 818 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 819 #endif 820 #ifdef INET 821 if (tp->t_inpcb) 822 so = tp->t_inpcb->inp_socket; 823 #endif 824 #ifdef INET6 825 if (tp->t_in6pcb) 826 so = tp->t_in6pcb->in6p_socket; 827 #endif 828 if (!so) 829 return NULL; 830 831 if (TCPS_HAVERCVDSYN(tp->t_state)) { 832 tp->t_state = TCPS_CLOSED; 833 (void) tcp_output(tp); 834 tcpstat.tcps_drops++; 835 } else 836 tcpstat.tcps_conndrops++; 837 if (errno == ETIMEDOUT && tp->t_softerror) 838 errno = tp->t_softerror; 839 so->so_error = errno; 840 return (tcp_close(tp)); 841 } 842 843 /* 844 * Close a TCP control block: 845 * discard all space held by the tcp 846 * discard internet protocol block 847 * wake up any sleepers 848 */ 849 struct tcpcb * 850 tcp_close(tp) 851 struct tcpcb *tp; 852 { 853 struct inpcb *inp; 854 #ifdef INET6 855 struct in6pcb *in6p; 856 #endif 857 struct socket *so; 858 #ifdef RTV_RTT 859 struct rtentry *rt; 860 #endif 861 struct route *ro; 862 863 inp = tp->t_inpcb; 864 #ifdef INET6 865 in6p = tp->t_in6pcb; 866 #endif 867 so = NULL; 868 ro = NULL; 869 if (inp) { 870 so = inp->inp_socket; 871 ro = &inp->inp_route; 872 } 873 #ifdef INET6 874 else if (in6p) { 875 so = in6p->in6p_socket; 876 ro = (struct route *)&in6p->in6p_route; 877 } 878 #endif 879 880 #ifdef RTV_RTT 881 /* 882 * If we sent enough data to get some meaningful characteristics, 883 * save them in the routing entry. 'Enough' is arbitrarily 884 * defined as the sendpipesize (default 4K) * 16. This would 885 * give us 16 rtt samples assuming we only get one sample per 886 * window (the usual case on a long haul net). 16 samples is 887 * enough for the srtt filter to converge to within 5% of the correct 888 * value; fewer samples and we could save a very bogus rtt. 889 * 890 * Don't update the default route's characteristics and don't 891 * update anything that the user "locked". 892 */ 893 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 894 ro && (rt = ro->ro_rt) && 895 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 896 u_long i = 0; 897 898 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 899 i = tp->t_srtt * 900 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 901 if (rt->rt_rmx.rmx_rtt && i) 902 /* 903 * filter this update to half the old & half 904 * the new values, converting scale. 905 * See route.h and tcp_var.h for a 906 * description of the scaling constants. 907 */ 908 rt->rt_rmx.rmx_rtt = 909 (rt->rt_rmx.rmx_rtt + i) / 2; 910 else 911 rt->rt_rmx.rmx_rtt = i; 912 } 913 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 914 i = tp->t_rttvar * 915 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 916 if (rt->rt_rmx.rmx_rttvar && i) 917 rt->rt_rmx.rmx_rttvar = 918 (rt->rt_rmx.rmx_rttvar + i) / 2; 919 else 920 rt->rt_rmx.rmx_rttvar = i; 921 } 922 /* 923 * update the pipelimit (ssthresh) if it has been updated 924 * already or if a pipesize was specified & the threshhold 925 * got below half the pipesize. I.e., wait for bad news 926 * before we start updating, then update on both good 927 * and bad news. 928 */ 929 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 930 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 931 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 932 /* 933 * convert the limit from user data bytes to 934 * packets then to packet data bytes. 935 */ 936 i = (i + tp->t_segsz / 2) / tp->t_segsz; 937 if (i < 2) 938 i = 2; 939 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 940 if (rt->rt_rmx.rmx_ssthresh) 941 rt->rt_rmx.rmx_ssthresh = 942 (rt->rt_rmx.rmx_ssthresh + i) / 2; 943 else 944 rt->rt_rmx.rmx_ssthresh = i; 945 } 946 } 947 #endif /* RTV_RTT */ 948 /* free the reassembly queue, if any */ 949 TCP_REASS_LOCK(tp); 950 (void) tcp_freeq(tp); 951 TCP_REASS_UNLOCK(tp); 952 953 TCP_CLEAR_DELACK(tp); 954 syn_cache_cleanup(tp); 955 956 if (tp->t_template) { 957 m_free(tp->t_template); 958 tp->t_template = NULL; 959 } 960 pool_put(&tcpcb_pool, tp); 961 if (inp) { 962 inp->inp_ppcb = 0; 963 soisdisconnected(so); 964 in_pcbdetach(inp); 965 } 966 #ifdef INET6 967 else if (in6p) { 968 in6p->in6p_ppcb = 0; 969 soisdisconnected(so); 970 in6_pcbdetach(in6p); 971 } 972 #endif 973 tcpstat.tcps_closed++; 974 return ((struct tcpcb *)0); 975 } 976 977 int 978 tcp_freeq(tp) 979 struct tcpcb *tp; 980 { 981 struct ipqent *qe; 982 int rv = 0; 983 #ifdef TCPREASS_DEBUG 984 int i = 0; 985 #endif 986 987 TCP_REASS_LOCK_CHECK(tp); 988 989 while ((qe = tp->segq.lh_first) != NULL) { 990 #ifdef TCPREASS_DEBUG 991 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 992 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 993 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 994 #endif 995 LIST_REMOVE(qe, ipqe_q); 996 LIST_REMOVE(qe, ipqe_timeq); 997 m_freem(qe->ipqe_m); 998 pool_put(&ipqent_pool, qe); 999 rv = 1; 1000 } 1001 return (rv); 1002 } 1003 1004 /* 1005 * Protocol drain routine. Called when memory is in short supply. 1006 */ 1007 void 1008 tcp_drain() 1009 { 1010 struct inpcb *inp; 1011 struct tcpcb *tp; 1012 1013 /* 1014 * Free the sequence queue of all TCP connections. 1015 */ 1016 inp = tcbtable.inpt_queue.cqh_first; 1017 if (inp) /* XXX */ 1018 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 1019 inp = inp->inp_queue.cqe_next) { 1020 if ((tp = intotcpcb(inp)) != NULL) { 1021 /* 1022 * We may be called from a device's interrupt 1023 * context. If the tcpcb is already busy, 1024 * just bail out now. 1025 */ 1026 if (tcp_reass_lock_try(tp) == 0) 1027 continue; 1028 if (tcp_freeq(tp)) 1029 tcpstat.tcps_connsdrained++; 1030 TCP_REASS_UNLOCK(tp); 1031 } 1032 } 1033 } 1034 1035 /* 1036 * Notify a tcp user of an asynchronous error; 1037 * store error as soft error, but wake up user 1038 * (for now, won't do anything until can select for soft error). 1039 */ 1040 void 1041 tcp_notify(inp, error) 1042 struct inpcb *inp; 1043 int error; 1044 { 1045 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1046 struct socket *so = inp->inp_socket; 1047 1048 /* 1049 * Ignore some errors if we are hooked up. 1050 * If connection hasn't completed, has retransmitted several times, 1051 * and receives a second error, give up now. This is better 1052 * than waiting a long time to establish a connection that 1053 * can never complete. 1054 */ 1055 if (tp->t_state == TCPS_ESTABLISHED && 1056 (error == EHOSTUNREACH || error == ENETUNREACH || 1057 error == EHOSTDOWN)) { 1058 return; 1059 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1060 tp->t_rxtshift > 3 && tp->t_softerror) 1061 so->so_error = error; 1062 else 1063 tp->t_softerror = error; 1064 wakeup((caddr_t) &so->so_timeo); 1065 sorwakeup(so); 1066 sowwakeup(so); 1067 } 1068 1069 #ifdef INET6 1070 void 1071 tcp6_notify(in6p, error) 1072 struct in6pcb *in6p; 1073 int error; 1074 { 1075 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1076 struct socket *so = in6p->in6p_socket; 1077 1078 /* 1079 * Ignore some errors if we are hooked up. 1080 * If connection hasn't completed, has retransmitted several times, 1081 * and receives a second error, give up now. This is better 1082 * than waiting a long time to establish a connection that 1083 * can never complete. 1084 */ 1085 if (tp->t_state == TCPS_ESTABLISHED && 1086 (error == EHOSTUNREACH || error == ENETUNREACH || 1087 error == EHOSTDOWN)) { 1088 return; 1089 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1090 tp->t_rxtshift > 3 && tp->t_softerror) 1091 so->so_error = error; 1092 else 1093 tp->t_softerror = error; 1094 wakeup((caddr_t) &so->so_timeo); 1095 sorwakeup(so); 1096 sowwakeup(so); 1097 } 1098 #endif 1099 1100 #ifdef INET6 1101 void 1102 tcp6_ctlinput(cmd, sa, d) 1103 int cmd; 1104 struct sockaddr *sa; 1105 void *d; 1106 { 1107 struct tcphdr *thp; 1108 struct tcphdr th; 1109 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1110 int nmatch; 1111 struct sockaddr_in6 sa6; 1112 struct ip6_hdr *ip6; 1113 struct mbuf *m; 1114 int off; 1115 struct in6_addr finaldst; 1116 struct in6_addr s; 1117 1118 if (sa->sa_family != AF_INET6 || 1119 sa->sa_len != sizeof(struct sockaddr_in6)) 1120 return; 1121 if ((unsigned)cmd >= PRC_NCMDS) 1122 return; 1123 else if (cmd == PRC_QUENCH) { 1124 /* XXX there's no PRC_QUENCH in IPv6 */ 1125 notify = tcp6_quench; 1126 } else if (PRC_IS_REDIRECT(cmd)) 1127 notify = in6_rtchange, d = NULL; 1128 else if (cmd == PRC_MSGSIZE) 1129 ; /* special code is present, see below */ 1130 else if (cmd == PRC_HOSTDEAD) 1131 d = NULL; 1132 else if (inet6ctlerrmap[cmd] == 0) 1133 return; 1134 1135 /* if the parameter is from icmp6, decode it. */ 1136 if (d != NULL) { 1137 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1138 m = ip6cp->ip6c_m; 1139 ip6 = ip6cp->ip6c_ip6; 1140 off = ip6cp->ip6c_off; 1141 1142 /* translate addresses into internal form */ 1143 bcopy(ip6cp->ip6c_finaldst, &finaldst, sizeof(finaldst)); 1144 if (IN6_IS_ADDR_LINKLOCAL(&finaldst)) { 1145 finaldst.s6_addr16[1] = 1146 htons(m->m_pkthdr.rcvif->if_index); 1147 } 1148 bcopy(&ip6->ip6_src, &s, sizeof(s)); 1149 if (IN6_IS_ADDR_LINKLOCAL(&s)) 1150 s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index); 1151 } else { 1152 m = NULL; 1153 ip6 = NULL; 1154 } 1155 1156 /* translate addresses into internal form */ 1157 sa6 = *(struct sockaddr_in6 *)sa; 1158 if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) && m && m->m_pkthdr.rcvif) 1159 sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index); 1160 1161 if (ip6) { 1162 /* 1163 * XXX: We assume that when ip6 is non NULL, 1164 * M and OFF are valid. 1165 */ 1166 1167 /* check if we can safely examine src and dst ports */ 1168 if (m->m_pkthdr.len < off + sizeof(th)) 1169 return; 1170 1171 if (m->m_len < off + sizeof(th)) { 1172 /* 1173 * this should be rare case, 1174 * so we compromise on this copy... 1175 */ 1176 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1177 thp = &th; 1178 } else 1179 thp = (struct tcphdr *)(mtod(m, caddr_t) + off); 1180 1181 if (cmd == PRC_MSGSIZE) { 1182 int valid = 0; 1183 1184 /* 1185 * Check to see if we have a valid TCP connection 1186 * corresponding to the address in the ICMPv6 message 1187 * payload. 1188 */ 1189 if (in6_pcblookup_connect(&tcb6, &finaldst, 1190 thp->th_dport, &s, thp->th_sport, 0)) 1191 valid++; 1192 1193 /* 1194 * Now that we've validated that we are actually 1195 * communicating with the host indicated in the ICMPv6 1196 * message, recalculate the new MTU, and create the 1197 * corresponding routing entry. 1198 */ 1199 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1200 1201 return; 1202 } 1203 1204 nmatch = in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 1205 thp->th_dport, &s, thp->th_sport, cmd, notify); 1206 if (nmatch == 0 && syn_cache_count && 1207 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1208 inet6ctlerrmap[cmd] == ENETUNREACH || 1209 inet6ctlerrmap[cmd] == EHOSTDOWN)) { 1210 struct sockaddr_in6 sin6; 1211 bzero(&sin6, sizeof(sin6)); 1212 sin6.sin6_len = sizeof(sin6); 1213 sin6.sin6_family = AF_INET6; 1214 sin6.sin6_port = thp->th_sport; 1215 sin6.sin6_addr = s; 1216 syn_cache_unreach((struct sockaddr *)&sin6, sa, thp); 1217 } 1218 } else { 1219 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 0, 1220 &zeroin6_addr, 0, cmd, notify); 1221 } 1222 } 1223 #endif 1224 1225 #ifdef INET 1226 /* assumes that ip header and tcp header are contiguous on mbuf */ 1227 void * 1228 tcp_ctlinput(cmd, sa, v) 1229 int cmd; 1230 struct sockaddr *sa; 1231 void *v; 1232 { 1233 struct ip *ip = v; 1234 struct tcphdr *th; 1235 struct icmp *icp; 1236 extern int inetctlerrmap[]; 1237 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1238 int errno; 1239 int nmatch; 1240 1241 if (sa->sa_family != AF_INET || 1242 sa->sa_len != sizeof(struct sockaddr_in)) 1243 return NULL; 1244 if ((unsigned)cmd >= PRC_NCMDS) 1245 return NULL; 1246 errno = inetctlerrmap[cmd]; 1247 if (cmd == PRC_QUENCH) 1248 notify = tcp_quench; 1249 else if (PRC_IS_REDIRECT(cmd)) 1250 notify = in_rtchange, ip = 0; 1251 else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) { 1252 /* 1253 * Check to see if we have a valid TCP connection 1254 * corresponding to the address in the ICMP message 1255 * payload. 1256 */ 1257 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1258 if (in_pcblookup_connect(&tcbtable, 1259 ip->ip_dst, th->th_dport, 1260 ip->ip_src, th->th_sport) == NULL) 1261 return NULL; 1262 1263 /* 1264 * Now that we've validated that we are actually communicating 1265 * with the host indicated in the ICMP message, locate the 1266 * ICMP header, recalculate the new MTU, and create the 1267 * corresponding routing entry. 1268 */ 1269 icp = (struct icmp *)((caddr_t)ip - 1270 offsetof(struct icmp, icmp_ip)); 1271 icmp_mtudisc(icp, ip->ip_dst); 1272 1273 return NULL; 1274 } else if (cmd == PRC_HOSTDEAD) 1275 ip = 0; 1276 else if (errno == 0) 1277 return NULL; 1278 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1279 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1280 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1281 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1282 if (nmatch == 0 && syn_cache_count && 1283 (inetctlerrmap[cmd] == EHOSTUNREACH || 1284 inetctlerrmap[cmd] == ENETUNREACH || 1285 inetctlerrmap[cmd] == EHOSTDOWN)) { 1286 struct sockaddr_in sin; 1287 bzero(&sin, sizeof(sin)); 1288 sin.sin_len = sizeof(sin); 1289 sin.sin_family = AF_INET; 1290 sin.sin_port = th->th_sport; 1291 sin.sin_addr = ip->ip_src; 1292 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1293 } 1294 1295 /* XXX mapped address case */ 1296 } else 1297 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1298 notify); 1299 return NULL; 1300 } 1301 1302 /* 1303 * When a source quence is received, we are being notifed of congestion. 1304 * Close the congestion window down to the Loss Window (one segment). 1305 * We will gradually open it again as we proceed. 1306 */ 1307 void 1308 tcp_quench(inp, errno) 1309 struct inpcb *inp; 1310 int errno; 1311 { 1312 struct tcpcb *tp = intotcpcb(inp); 1313 1314 if (tp) 1315 tp->snd_cwnd = tp->t_segsz; 1316 } 1317 #endif 1318 1319 #ifdef INET6 1320 void 1321 tcp6_quench(in6p, errno) 1322 struct in6pcb *in6p; 1323 int errno; 1324 { 1325 struct tcpcb *tp = in6totcpcb(in6p); 1326 1327 if (tp) 1328 tp->snd_cwnd = tp->t_segsz; 1329 } 1330 #endif 1331 1332 #ifdef INET 1333 /* 1334 * Path MTU Discovery handlers. 1335 */ 1336 void 1337 tcp_mtudisc_callback(faddr) 1338 struct in_addr faddr; 1339 { 1340 1341 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1342 } 1343 1344 /* 1345 * On receipt of path MTU corrections, flush old route and replace it 1346 * with the new one. Retransmit all unacknowledged packets, to ensure 1347 * that all packets will be received. 1348 */ 1349 void 1350 tcp_mtudisc(inp, errno) 1351 struct inpcb *inp; 1352 int errno; 1353 { 1354 struct tcpcb *tp = intotcpcb(inp); 1355 struct rtentry *rt = in_pcbrtentry(inp); 1356 1357 if (tp != 0) { 1358 if (rt != 0) { 1359 /* 1360 * If this was not a host route, remove and realloc. 1361 */ 1362 if ((rt->rt_flags & RTF_HOST) == 0) { 1363 in_rtchange(inp, errno); 1364 if ((rt = in_pcbrtentry(inp)) == 0) 1365 return; 1366 } 1367 1368 /* 1369 * Slow start out of the error condition. We 1370 * use the MTU because we know it's smaller 1371 * than the previously transmitted segment. 1372 * 1373 * Note: This is more conservative than the 1374 * suggestion in draft-floyd-incr-init-win-03. 1375 */ 1376 if (rt->rt_rmx.rmx_mtu != 0) 1377 tp->snd_cwnd = 1378 TCP_INITIAL_WINDOW(tcp_init_win, 1379 rt->rt_rmx.rmx_mtu); 1380 } 1381 1382 /* 1383 * Resend unacknowledged packets. 1384 */ 1385 tp->snd_nxt = tp->snd_una; 1386 tcp_output(tp); 1387 } 1388 } 1389 #endif 1390 1391 #ifdef INET6 1392 /* 1393 * Path MTU Discovery handlers. 1394 */ 1395 void 1396 tcp6_mtudisc_callback(faddr) 1397 struct in6_addr *faddr; 1398 { 1399 struct sockaddr_in6 sin6; 1400 1401 bzero(&sin6, sizeof(sin6)); 1402 sin6.sin6_family = AF_INET6; 1403 sin6.sin6_len = sizeof(struct sockaddr_in6); 1404 sin6.sin6_addr = *faddr; 1405 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1406 &zeroin6_addr, 0, PRC_MSGSIZE, tcp6_mtudisc); 1407 } 1408 1409 void 1410 tcp6_mtudisc(in6p, errno) 1411 struct in6pcb *in6p; 1412 int errno; 1413 { 1414 struct tcpcb *tp = in6totcpcb(in6p); 1415 struct rtentry *rt = in6_pcbrtentry(in6p); 1416 1417 if (tp != 0) { 1418 if (rt != 0) { 1419 /* 1420 * If this was not a host route, remove and realloc. 1421 */ 1422 if ((rt->rt_flags & RTF_HOST) == 0) { 1423 in6_rtchange(in6p, errno); 1424 if ((rt = in6_pcbrtentry(in6p)) == 0) 1425 return; 1426 } 1427 1428 /* 1429 * Slow start out of the error condition. We 1430 * use the MTU because we know it's smaller 1431 * than the previously transmitted segment. 1432 * 1433 * Note: This is more conservative than the 1434 * suggestion in draft-floyd-incr-init-win-03. 1435 */ 1436 if (rt->rt_rmx.rmx_mtu != 0) 1437 tp->snd_cwnd = 1438 TCP_INITIAL_WINDOW(tcp_init_win, 1439 rt->rt_rmx.rmx_mtu); 1440 } 1441 1442 /* 1443 * Resend unacknowledged packets. 1444 */ 1445 tp->snd_nxt = tp->snd_una; 1446 tcp_output(tp); 1447 } 1448 } 1449 #endif /* INET6 */ 1450 1451 /* 1452 * Compute the MSS to advertise to the peer. Called only during 1453 * the 3-way handshake. If we are the server (peer initiated 1454 * connection), we are called with a pointer to the interface 1455 * on which the SYN packet arrived. If we are the client (we 1456 * initiated connection), we are called with a pointer to the 1457 * interface out which this connection should go. 1458 * 1459 * NOTE: Do not subtract IP option/extension header size nor IPsec 1460 * header size from MSS advertisement. MSS option must hold the maximum 1461 * segment size we can accept, so it must always be: 1462 * max(if mtu) - ip header - tcp header 1463 */ 1464 u_long 1465 tcp_mss_to_advertise(ifp, af) 1466 const struct ifnet *ifp; 1467 int af; 1468 { 1469 extern u_long in_maxmtu; 1470 u_long mss = 0; 1471 u_long hdrsiz; 1472 1473 /* 1474 * In order to avoid defeating path MTU discovery on the peer, 1475 * we advertise the max MTU of all attached networks as our MSS, 1476 * per RFC 1191, section 3.1. 1477 * 1478 * We provide the option to advertise just the MTU of 1479 * the interface on which we hope this connection will 1480 * be receiving. If we are responding to a SYN, we 1481 * will have a pretty good idea about this, but when 1482 * initiating a connection there is a bit more doubt. 1483 * 1484 * We also need to ensure that loopback has a large enough 1485 * MSS, as the loopback MTU is never included in in_maxmtu. 1486 */ 1487 1488 if (ifp != NULL) 1489 mss = ifp->if_mtu; 1490 1491 if (tcp_mss_ifmtu == 0) 1492 mss = max(in_maxmtu, mss); 1493 1494 switch (af) { 1495 case AF_INET: 1496 hdrsiz = sizeof(struct ip); 1497 break; 1498 #ifdef INET6 1499 case AF_INET6: 1500 hdrsiz = sizeof(struct ip6_hdr); 1501 break; 1502 #endif 1503 default: 1504 hdrsiz = 0; 1505 break; 1506 } 1507 hdrsiz += sizeof(struct tcphdr); 1508 if (mss > hdrsiz) 1509 mss -= hdrsiz; 1510 1511 mss = max(tcp_mssdflt, mss); 1512 return (mss); 1513 } 1514 1515 /* 1516 * Set connection variables based on the peer's advertised MSS. 1517 * We are passed the TCPCB for the actual connection. If we 1518 * are the server, we are called by the compressed state engine 1519 * when the 3-way handshake is complete. If we are the client, 1520 * we are called when we recieve the SYN,ACK from the server. 1521 * 1522 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1523 * before this routine is called! 1524 */ 1525 void 1526 tcp_mss_from_peer(tp, offer) 1527 struct tcpcb *tp; 1528 int offer; 1529 { 1530 struct socket *so; 1531 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1532 struct rtentry *rt; 1533 #endif 1534 u_long bufsize; 1535 int mss; 1536 1537 #ifdef DIAGNOSTIC 1538 if (tp->t_inpcb && tp->t_in6pcb) 1539 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1540 #endif 1541 so = NULL; 1542 rt = NULL; 1543 #ifdef INET 1544 if (tp->t_inpcb) { 1545 so = tp->t_inpcb->inp_socket; 1546 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1547 rt = in_pcbrtentry(tp->t_inpcb); 1548 #endif 1549 } 1550 #endif 1551 #ifdef INET6 1552 if (tp->t_in6pcb) { 1553 so = tp->t_in6pcb->in6p_socket; 1554 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1555 rt = in6_pcbrtentry(tp->t_in6pcb); 1556 #endif 1557 } 1558 #endif 1559 1560 /* 1561 * As per RFC1122, use the default MSS value, unless they 1562 * sent us an offer. Do not accept offers less than 32 bytes. 1563 */ 1564 mss = tcp_mssdflt; 1565 if (offer) 1566 mss = offer; 1567 mss = max(mss, 32); /* sanity */ 1568 tp->t_peermss = mss; 1569 mss -= tcp_optlen(tp); 1570 #ifdef INET 1571 if (tp->t_inpcb) 1572 mss -= ip_optlen(tp->t_inpcb); 1573 #endif 1574 #ifdef INET6 1575 if (tp->t_in6pcb) 1576 mss -= ip6_optlen(tp->t_in6pcb); 1577 #endif 1578 1579 /* 1580 * If there's a pipesize, change the socket buffer to that size. 1581 * Make the socket buffer an integral number of MSS units. If 1582 * the MSS is larger than the socket buffer, artificially decrease 1583 * the MSS. 1584 */ 1585 #ifdef RTV_SPIPE 1586 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1587 bufsize = rt->rt_rmx.rmx_sendpipe; 1588 else 1589 #endif 1590 bufsize = so->so_snd.sb_hiwat; 1591 if (bufsize < mss) 1592 mss = bufsize; 1593 else { 1594 bufsize = roundup(bufsize, mss); 1595 if (bufsize > sb_max) 1596 bufsize = sb_max; 1597 (void) sbreserve(&so->so_snd, bufsize); 1598 } 1599 tp->t_segsz = mss; 1600 1601 #ifdef RTV_SSTHRESH 1602 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1603 /* 1604 * There's some sort of gateway or interface buffer 1605 * limit on the path. Use this to set the slow 1606 * start threshold, but set the threshold to no less 1607 * than 2 * MSS. 1608 */ 1609 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1610 } 1611 #endif 1612 } 1613 1614 /* 1615 * Processing necessary when a TCP connection is established. 1616 */ 1617 void 1618 tcp_established(tp) 1619 struct tcpcb *tp; 1620 { 1621 struct socket *so; 1622 #ifdef RTV_RPIPE 1623 struct rtentry *rt; 1624 #endif 1625 u_long bufsize; 1626 1627 #ifdef DIAGNOSTIC 1628 if (tp->t_inpcb && tp->t_in6pcb) 1629 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1630 #endif 1631 so = NULL; 1632 rt = NULL; 1633 #ifdef INET 1634 if (tp->t_inpcb) { 1635 so = tp->t_inpcb->inp_socket; 1636 #if defined(RTV_RPIPE) 1637 rt = in_pcbrtentry(tp->t_inpcb); 1638 #endif 1639 } 1640 #endif 1641 #ifdef INET6 1642 if (tp->t_in6pcb) { 1643 so = tp->t_in6pcb->in6p_socket; 1644 #if defined(RTV_RPIPE) 1645 rt = in6_pcbrtentry(tp->t_in6pcb); 1646 #endif 1647 } 1648 #endif 1649 1650 tp->t_state = TCPS_ESTABLISHED; 1651 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1652 1653 #ifdef RTV_RPIPE 1654 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1655 bufsize = rt->rt_rmx.rmx_recvpipe; 1656 else 1657 #endif 1658 bufsize = so->so_rcv.sb_hiwat; 1659 if (bufsize > tp->t_ourmss) { 1660 bufsize = roundup(bufsize, tp->t_ourmss); 1661 if (bufsize > sb_max) 1662 bufsize = sb_max; 1663 (void) sbreserve(&so->so_rcv, bufsize); 1664 } 1665 } 1666 1667 /* 1668 * Check if there's an initial rtt or rttvar. Convert from the 1669 * route-table units to scaled multiples of the slow timeout timer. 1670 * Called only during the 3-way handshake. 1671 */ 1672 void 1673 tcp_rmx_rtt(tp) 1674 struct tcpcb *tp; 1675 { 1676 #ifdef RTV_RTT 1677 struct rtentry *rt = NULL; 1678 int rtt; 1679 1680 #ifdef DIAGNOSTIC 1681 if (tp->t_inpcb && tp->t_in6pcb) 1682 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1683 #endif 1684 #ifdef INET 1685 if (tp->t_inpcb) 1686 rt = in_pcbrtentry(tp->t_inpcb); 1687 #endif 1688 #ifdef INET6 1689 if (tp->t_in6pcb) 1690 rt = in6_pcbrtentry(tp->t_in6pcb); 1691 #endif 1692 if (rt == NULL) 1693 return; 1694 1695 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1696 /* 1697 * XXX The lock bit for MTU indicates that the value 1698 * is also a minimum value; this is subject to time. 1699 */ 1700 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1701 TCPT_RANGESET(tp->t_rttmin, 1702 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1703 TCPTV_MIN, TCPTV_REXMTMAX); 1704 tp->t_srtt = rtt / 1705 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1706 if (rt->rt_rmx.rmx_rttvar) { 1707 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1708 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1709 (TCP_RTTVAR_SHIFT + 2)); 1710 } else { 1711 /* Default variation is +- 1 rtt */ 1712 tp->t_rttvar = 1713 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1714 } 1715 TCPT_RANGESET(tp->t_rxtcur, 1716 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1717 tp->t_rttmin, TCPTV_REXMTMAX); 1718 } 1719 #endif 1720 } 1721 1722 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1723 1724 /* 1725 * Get a new sequence value given a tcp control block 1726 */ 1727 tcp_seq 1728 tcp_new_iss(tp, len, addin) 1729 void *tp; 1730 u_long len; 1731 tcp_seq addin; 1732 { 1733 tcp_seq tcp_iss; 1734 1735 /* 1736 * Randomize. 1737 */ 1738 #if NRND > 0 1739 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 1740 #else 1741 tcp_iss = random(); 1742 #endif 1743 1744 /* 1745 * If we were asked to add some amount to a known value, 1746 * we will take a random value obtained above, mask off the upper 1747 * bits, and add in the known value. We also add in a constant to 1748 * ensure that we are at least a certain distance from the original 1749 * value. 1750 * 1751 * This is used when an old connection is in timed wait 1752 * and we have a new one coming in, for instance. 1753 */ 1754 if (addin != 0) { 1755 #ifdef TCPISS_DEBUG 1756 printf("Random %08x, ", tcp_iss); 1757 #endif 1758 tcp_iss &= TCP_ISS_RANDOM_MASK; 1759 tcp_iss += addin + TCP_ISSINCR; 1760 #ifdef TCPISS_DEBUG 1761 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 1762 #endif 1763 } else { 1764 tcp_iss &= TCP_ISS_RANDOM_MASK; 1765 tcp_iss += tcp_iss_seq; 1766 tcp_iss_seq += TCP_ISSINCR; 1767 #ifdef TCPISS_DEBUG 1768 printf("ISS %08x\n", tcp_iss); 1769 #endif 1770 } 1771 1772 if (tcp_compat_42) { 1773 /* 1774 * Limit it to the positive range for really old TCP 1775 * implementations. 1776 */ 1777 if (tcp_iss >= 0x80000000) 1778 tcp_iss &= 0x7fffffff; /* XXX */ 1779 } 1780 1781 return tcp_iss; 1782 } 1783 1784 #ifdef IPSEC 1785 /* compute ESP/AH header size for TCP, including outer IP header. */ 1786 size_t 1787 ipsec4_hdrsiz_tcp(tp) 1788 struct tcpcb *tp; 1789 { 1790 struct inpcb *inp; 1791 size_t hdrsiz; 1792 1793 /* XXX mapped addr case (tp->t_in6pcb) */ 1794 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 1795 return 0; 1796 switch (tp->t_family) { 1797 case AF_INET: 1798 /* XXX: should use currect direction. */ 1799 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 1800 break; 1801 default: 1802 hdrsiz = 0; 1803 break; 1804 } 1805 1806 return hdrsiz; 1807 } 1808 1809 #ifdef INET6 1810 size_t 1811 ipsec6_hdrsiz_tcp(tp) 1812 struct tcpcb *tp; 1813 { 1814 struct in6pcb *in6p; 1815 size_t hdrsiz; 1816 1817 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 1818 return 0; 1819 switch (tp->t_family) { 1820 case AF_INET6: 1821 /* XXX: should use currect direction. */ 1822 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 1823 break; 1824 case AF_INET: 1825 /* mapped address case - tricky */ 1826 default: 1827 hdrsiz = 0; 1828 break; 1829 } 1830 1831 return hdrsiz; 1832 } 1833 #endif 1834 #endif /*IPSEC*/ 1835 1836 /* 1837 * Determine the length of the TCP options for this connection. 1838 * 1839 * XXX: What do we do for SACK, when we add that? Just reserve 1840 * all of the space? Otherwise we can't exactly be incrementing 1841 * cwnd by an amount that varies depending on the amount we last 1842 * had to SACK! 1843 */ 1844 1845 u_int 1846 tcp_optlen(tp) 1847 struct tcpcb *tp; 1848 { 1849 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1850 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1851 return TCPOLEN_TSTAMP_APPA; 1852 else 1853 return 0; 1854 } 1855