1 /* $NetBSD: tcp_subr.c,v 1.109 2001/03/21 03:35:11 chs 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 "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/md5.h> 122 #include <sys/rnd.h> 123 #endif 124 125 #include <net/route.h> 126 #include <net/if.h> 127 128 #include <netinet/in.h> 129 #include <netinet/in_systm.h> 130 #include <netinet/ip.h> 131 #include <netinet/in_pcb.h> 132 #include <netinet/ip_var.h> 133 #include <netinet/ip_icmp.h> 134 135 #ifdef INET6 136 #ifndef INET 137 #include <netinet/in.h> 138 #endif 139 #include <netinet/ip6.h> 140 #include <netinet6/in6_pcb.h> 141 #include <netinet6/ip6_var.h> 142 #include <netinet6/in6_var.h> 143 #include <netinet6/ip6protosw.h> 144 #include <netinet/icmp6.h> 145 #endif 146 147 #include <netinet/tcp.h> 148 #include <netinet/tcp_fsm.h> 149 #include <netinet/tcp_seq.h> 150 #include <netinet/tcp_timer.h> 151 #include <netinet/tcp_var.h> 152 #include <netinet/tcpip.h> 153 154 #ifdef IPSEC 155 #include <netinet6/ipsec.h> 156 #endif /*IPSEC*/ 157 158 #ifdef INET6 159 struct in6pcb tcb6; 160 #endif 161 162 /* patchable/settable parameters for tcp */ 163 int tcp_mssdflt = TCP_MSS; 164 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 165 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 166 #if NRND > 0 167 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 168 #endif 169 int tcp_do_sack = 1; /* selective acknowledgement */ 170 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 171 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 172 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 173 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 174 int tcp_init_win = 1; 175 int tcp_mss_ifmtu = 0; 176 #ifdef TCP_COMPAT_42 177 int tcp_compat_42 = 1; 178 #else 179 int tcp_compat_42 = 0; 180 #endif 181 int tcp_rst_ppslim = 100; /* 100pps */ 182 183 /* tcb hash */ 184 #ifndef TCBHASHSIZE 185 #define TCBHASHSIZE 128 186 #endif 187 int tcbhashsize = TCBHASHSIZE; 188 189 /* syn hash parameters */ 190 #define TCP_SYN_HASH_SIZE 293 191 #define TCP_SYN_BUCKET_SIZE 35 192 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 193 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 194 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 195 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 196 int tcp_syn_cache_interval = 1; /* runs timer twice a second */ 197 198 int tcp_freeq __P((struct tcpcb *)); 199 200 #ifdef INET 201 void tcp_mtudisc_callback __P((struct in_addr)); 202 #endif 203 #ifdef INET6 204 void tcp6_mtudisc_callback __P((struct in6_addr *)); 205 #endif 206 207 void tcp_mtudisc __P((struct inpcb *, int)); 208 #ifdef INET6 209 void tcp6_mtudisc __P((struct in6pcb *, int)); 210 #endif 211 212 struct pool tcpcb_pool; 213 214 /* 215 * Tcp initialization 216 */ 217 void 218 tcp_init() 219 { 220 int hlen; 221 222 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 223 0, NULL, NULL, M_PCB); 224 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 225 #ifdef INET6 226 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 227 #endif 228 LIST_INIT(&tcp_delacks); 229 230 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 231 #ifdef INET6 232 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 233 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 234 #endif 235 if (max_protohdr < hlen) 236 max_protohdr = hlen; 237 if (max_linkhdr + hlen > MHLEN) 238 panic("tcp_init"); 239 240 #ifdef INET 241 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 242 #endif 243 #ifdef INET6 244 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 245 #endif 246 247 /* Initialize the compressed state engine. */ 248 syn_cache_init(); 249 } 250 251 /* 252 * Create template to be used to send tcp packets on a connection. 253 * Call after host entry created, allocates an mbuf and fills 254 * in a skeletal tcp/ip header, minimizing the amount of work 255 * necessary when the connection is used. 256 */ 257 struct mbuf * 258 tcp_template(tp) 259 struct tcpcb *tp; 260 { 261 struct inpcb *inp = tp->t_inpcb; 262 #ifdef INET6 263 struct in6pcb *in6p = tp->t_in6pcb; 264 #endif 265 struct tcphdr *n; 266 struct mbuf *m; 267 int hlen; 268 269 switch (tp->t_family) { 270 case AF_INET: 271 hlen = sizeof(struct ip); 272 if (inp) 273 break; 274 #ifdef INET6 275 if (in6p) { 276 /* mapped addr case */ 277 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 278 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 279 break; 280 } 281 #endif 282 return NULL; /*EINVAL*/ 283 #ifdef INET6 284 case AF_INET6: 285 hlen = sizeof(struct ip6_hdr); 286 if (in6p) { 287 /* more sainty check? */ 288 break; 289 } 290 return NULL; /*EINVAL*/ 291 #endif 292 default: 293 hlen = 0; /*pacify gcc*/ 294 return NULL; /*EAFNOSUPPORT*/ 295 } 296 #ifdef DIAGNOSTIC 297 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 298 panic("mclbytes too small for t_template"); 299 #endif 300 m = tp->t_template; 301 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 302 ; 303 else { 304 if (m) 305 m_freem(m); 306 m = tp->t_template = NULL; 307 MGETHDR(m, M_DONTWAIT, MT_HEADER); 308 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 309 MCLGET(m, M_DONTWAIT); 310 if ((m->m_flags & M_EXT) == 0) { 311 m_free(m); 312 m = NULL; 313 } 314 } 315 if (m == NULL) 316 return NULL; 317 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 318 } 319 bzero(mtod(m, caddr_t), m->m_len); 320 switch (tp->t_family) { 321 case AF_INET: 322 { 323 struct ipovly *ipov; 324 mtod(m, struct ip *)->ip_v = 4; 325 ipov = mtod(m, struct ipovly *); 326 ipov->ih_pr = IPPROTO_TCP; 327 ipov->ih_len = htons(sizeof(struct tcphdr)); 328 if (inp) { 329 ipov->ih_src = inp->inp_laddr; 330 ipov->ih_dst = inp->inp_faddr; 331 } 332 #ifdef INET6 333 else if (in6p) { 334 /* mapped addr case */ 335 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 336 sizeof(ipov->ih_src)); 337 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 338 sizeof(ipov->ih_dst)); 339 } 340 #endif 341 break; 342 } 343 #ifdef INET6 344 case AF_INET6: 345 { 346 struct ip6_hdr *ip6; 347 mtod(m, struct ip *)->ip_v = 6; 348 ip6 = mtod(m, struct ip6_hdr *); 349 ip6->ip6_nxt = IPPROTO_TCP; 350 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 351 ip6->ip6_src = in6p->in6p_laddr; 352 ip6->ip6_dst = in6p->in6p_faddr; 353 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 354 if (ip6_auto_flowlabel) { 355 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 356 ip6->ip6_flow |= 357 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 358 } 359 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 360 ip6->ip6_vfc |= IPV6_VERSION; 361 break; 362 } 363 #endif 364 } 365 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 366 if (inp) { 367 n->th_sport = inp->inp_lport; 368 n->th_dport = inp->inp_fport; 369 } 370 #ifdef INET6 371 else if (in6p) { 372 n->th_sport = in6p->in6p_lport; 373 n->th_dport = in6p->in6p_fport; 374 } 375 #endif 376 n->th_seq = 0; 377 n->th_ack = 0; 378 n->th_x2 = 0; 379 n->th_off = 5; 380 n->th_flags = 0; 381 n->th_win = 0; 382 n->th_sum = 0; 383 n->th_urp = 0; 384 return (m); 385 } 386 387 /* 388 * Send a single message to the TCP at address specified by 389 * the given TCP/IP header. If m == 0, then we make a copy 390 * of the tcpiphdr at ti and send directly to the addressed host. 391 * This is used to force keep alive messages out using the TCP 392 * template for a connection tp->t_template. If flags are given 393 * then we send a message back to the TCP which originated the 394 * segment ti, and discard the mbuf containing it and any other 395 * attached mbufs. 396 * 397 * In any case the ack and sequence number of the transmitted 398 * segment are as specified by the parameters. 399 */ 400 int 401 tcp_respond(tp, template, m, th0, ack, seq, flags) 402 struct tcpcb *tp; 403 struct mbuf *template; 404 struct mbuf *m; 405 struct tcphdr *th0; 406 tcp_seq ack, seq; 407 int flags; 408 { 409 struct route *ro; 410 int error, tlen, win = 0; 411 int hlen; 412 struct ip *ip; 413 #ifdef INET6 414 struct ip6_hdr *ip6; 415 #endif 416 int family; /* family on packet, not inpcb/in6pcb! */ 417 struct tcphdr *th; 418 419 if (tp != NULL && (flags & TH_RST) == 0) { 420 #ifdef DIAGNOSTIC 421 if (tp->t_inpcb && tp->t_in6pcb) 422 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 423 #endif 424 #ifdef INET 425 if (tp->t_inpcb) 426 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 427 #endif 428 #ifdef INET6 429 if (tp->t_in6pcb) 430 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 431 #endif 432 } 433 434 ip = NULL; 435 #ifdef INET6 436 ip6 = NULL; 437 #endif 438 if (m == 0) { 439 if (!template) 440 return EINVAL; 441 442 /* get family information from template */ 443 switch (mtod(template, struct ip *)->ip_v) { 444 case 4: 445 family = AF_INET; 446 hlen = sizeof(struct ip); 447 break; 448 #ifdef INET6 449 case 6: 450 family = AF_INET6; 451 hlen = sizeof(struct ip6_hdr); 452 break; 453 #endif 454 default: 455 return EAFNOSUPPORT; 456 } 457 458 MGETHDR(m, M_DONTWAIT, MT_HEADER); 459 if (m) { 460 MCLGET(m, M_DONTWAIT); 461 if ((m->m_flags & M_EXT) == 0) { 462 m_free(m); 463 m = NULL; 464 } 465 } 466 if (m == NULL) 467 return (ENOBUFS); 468 469 if (tcp_compat_42) 470 tlen = 1; 471 else 472 tlen = 0; 473 474 m->m_data += max_linkhdr; 475 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 476 template->m_len); 477 switch (family) { 478 case AF_INET: 479 ip = mtod(m, struct ip *); 480 th = (struct tcphdr *)(ip + 1); 481 break; 482 #ifdef INET6 483 case AF_INET6: 484 ip6 = mtod(m, struct ip6_hdr *); 485 th = (struct tcphdr *)(ip6 + 1); 486 break; 487 #endif 488 #if 0 489 default: 490 /* noone will visit here */ 491 m_freem(m); 492 return EAFNOSUPPORT; 493 #endif 494 } 495 flags = TH_ACK; 496 } else { 497 498 if ((m->m_flags & M_PKTHDR) == 0) { 499 #if 0 500 printf("non PKTHDR to tcp_respond\n"); 501 #endif 502 m_freem(m); 503 return EINVAL; 504 } 505 #ifdef DIAGNOSTIC 506 if (!th0) 507 panic("th0 == NULL in tcp_respond"); 508 #endif 509 510 /* get family information from m */ 511 switch (mtod(m, struct ip *)->ip_v) { 512 case 4: 513 family = AF_INET; 514 hlen = sizeof(struct ip); 515 ip = mtod(m, struct ip *); 516 break; 517 #ifdef INET6 518 case 6: 519 family = AF_INET6; 520 hlen = sizeof(struct ip6_hdr); 521 ip6 = mtod(m, struct ip6_hdr *); 522 break; 523 #endif 524 default: 525 m_freem(m); 526 return EAFNOSUPPORT; 527 } 528 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 529 tlen = sizeof(*th0); 530 else 531 tlen = th0->th_off << 2; 532 533 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 534 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 535 m->m_len = hlen + tlen; 536 m_freem(m->m_next); 537 m->m_next = NULL; 538 } else { 539 struct mbuf *n; 540 541 #ifdef DIAGNOSTIC 542 if (max_linkhdr + hlen + tlen > MCLBYTES) { 543 m_freem(m); 544 return EMSGSIZE; 545 } 546 #endif 547 MGETHDR(n, M_DONTWAIT, MT_HEADER); 548 if (n && max_linkhdr + hlen + tlen > MHLEN) { 549 MCLGET(n, M_DONTWAIT); 550 if ((n->m_flags & M_EXT) == 0) { 551 m_freem(n); 552 n = NULL; 553 } 554 } 555 if (!n) { 556 m_freem(m); 557 return ENOBUFS; 558 } 559 560 n->m_data += max_linkhdr; 561 n->m_len = hlen + tlen; 562 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 563 m_copyback(n, hlen, tlen, (caddr_t)th0); 564 565 m_freem(m); 566 m = n; 567 n = NULL; 568 } 569 570 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 571 switch (family) { 572 case AF_INET: 573 ip = mtod(m, struct ip *); 574 th = (struct tcphdr *)(ip + 1); 575 ip->ip_p = IPPROTO_TCP; 576 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 577 ip->ip_p = IPPROTO_TCP; 578 break; 579 #ifdef INET6 580 case AF_INET6: 581 ip6 = mtod(m, struct ip6_hdr *); 582 th = (struct tcphdr *)(ip6 + 1); 583 ip6->ip6_nxt = IPPROTO_TCP; 584 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 585 ip6->ip6_nxt = IPPROTO_TCP; 586 break; 587 #endif 588 #if 0 589 default: 590 /* noone will visit here */ 591 m_freem(m); 592 return EAFNOSUPPORT; 593 #endif 594 } 595 xchg(th->th_dport, th->th_sport, u_int16_t); 596 #undef xchg 597 tlen = 0; /*be friendly with the following code*/ 598 } 599 th->th_seq = htonl(seq); 600 th->th_ack = htonl(ack); 601 th->th_x2 = 0; 602 if ((flags & TH_SYN) == 0) { 603 if (tp) 604 win >>= tp->rcv_scale; 605 if (win > TCP_MAXWIN) 606 win = TCP_MAXWIN; 607 th->th_win = htons((u_int16_t)win); 608 th->th_off = sizeof (struct tcphdr) >> 2; 609 tlen += sizeof(*th); 610 } else 611 tlen += th->th_off << 2; 612 m->m_len = hlen + tlen; 613 m->m_pkthdr.len = hlen + tlen; 614 m->m_pkthdr.rcvif = (struct ifnet *) 0; 615 th->th_flags = flags; 616 th->th_urp = 0; 617 618 switch (family) { 619 #ifdef INET 620 case AF_INET: 621 { 622 struct ipovly *ipov = (struct ipovly *)ip; 623 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 624 ipov->ih_len = htons((u_int16_t)tlen); 625 626 th->th_sum = 0; 627 th->th_sum = in_cksum(m, hlen + tlen); 628 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 629 ip->ip_ttl = ip_defttl; 630 break; 631 } 632 #endif 633 #ifdef INET6 634 case AF_INET6: 635 { 636 th->th_sum = 0; 637 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 638 tlen); 639 ip6->ip6_plen = ntohs(tlen); 640 if (tp && tp->t_in6pcb) { 641 struct ifnet *oifp; 642 ro = (struct route *)&tp->t_in6pcb->in6p_route; 643 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 644 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 645 } else 646 ip6->ip6_hlim = ip6_defhlim; 647 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 648 if (ip6_auto_flowlabel) { 649 ip6->ip6_flow |= 650 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 651 } 652 break; 653 } 654 #endif 655 } 656 657 #ifdef IPSEC 658 (void)ipsec_setsocket(m, NULL); 659 #endif /*IPSEC*/ 660 661 if (tp != NULL && tp->t_inpcb != NULL) { 662 ro = &tp->t_inpcb->inp_route; 663 #ifdef IPSEC 664 if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) { 665 m_freem(m); 666 return ENOBUFS; 667 } 668 #endif 669 #ifdef DIAGNOSTIC 670 if (family != AF_INET) 671 panic("tcp_respond: address family mismatch"); 672 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 673 panic("tcp_respond: ip_dst %x != inp_faddr %x", 674 ntohl(ip->ip_dst.s_addr), 675 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 676 } 677 #endif 678 } 679 #ifdef INET6 680 else if (tp != NULL && tp->t_in6pcb != NULL) { 681 ro = (struct route *)&tp->t_in6pcb->in6p_route; 682 #ifdef IPSEC 683 if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) { 684 m_freem(m); 685 return ENOBUFS; 686 } 687 #endif 688 #ifdef DIAGNOSTIC 689 if (family == AF_INET) { 690 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 691 panic("tcp_respond: not mapped addr"); 692 if (bcmp(&ip->ip_dst, 693 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 694 sizeof(ip->ip_dst)) != 0) { 695 panic("tcp_respond: ip_dst != in6p_faddr"); 696 } 697 } else if (family == AF_INET6) { 698 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 699 panic("tcp_respond: ip6_dst != in6p_faddr"); 700 } else 701 panic("tcp_respond: address family mismatch"); 702 #endif 703 } 704 #endif 705 else 706 ro = NULL; 707 708 switch (family) { 709 #ifdef INET 710 case AF_INET: 711 error = ip_output(m, NULL, ro, 712 (ip_mtudisc ? IP_MTUDISC : 0), 713 NULL); 714 break; 715 #endif 716 #ifdef INET6 717 case AF_INET6: 718 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 719 NULL); 720 break; 721 #endif 722 default: 723 error = EAFNOSUPPORT; 724 break; 725 } 726 727 return (error); 728 } 729 730 /* 731 * Create a new TCP control block, making an 732 * empty reassembly queue and hooking it to the argument 733 * protocol control block. 734 */ 735 struct tcpcb * 736 tcp_newtcpcb(family, aux) 737 int family; /* selects inpcb, or in6pcb */ 738 void *aux; 739 { 740 struct tcpcb *tp; 741 742 switch (family) { 743 case PF_INET: 744 break; 745 #ifdef INET6 746 case PF_INET6: 747 break; 748 #endif 749 default: 750 return NULL; 751 } 752 753 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 754 if (tp == NULL) 755 return (NULL); 756 bzero((caddr_t)tp, sizeof(struct tcpcb)); 757 LIST_INIT(&tp->segq); 758 LIST_INIT(&tp->timeq); 759 tp->t_family = family; /* may be overridden later on */ 760 tp->t_peermss = tcp_mssdflt; 761 tp->t_ourmss = tcp_mssdflt; 762 tp->t_segsz = tcp_mssdflt; 763 LIST_INIT(&tp->t_sc); 764 765 tp->t_flags = 0; 766 if (tcp_do_rfc1323 && tcp_do_win_scale) 767 tp->t_flags |= TF_REQ_SCALE; 768 if (tcp_do_rfc1323 && tcp_do_timestamps) 769 tp->t_flags |= TF_REQ_TSTMP; 770 if (tcp_do_sack == 2) 771 tp->t_flags |= TF_WILL_SACK; 772 else if (tcp_do_sack == 1) 773 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 774 tp->t_flags |= TF_CANT_TXSACK; 775 switch (family) { 776 case PF_INET: 777 tp->t_inpcb = (struct inpcb *)aux; 778 break; 779 #ifdef INET6 780 case PF_INET6: 781 tp->t_in6pcb = (struct in6pcb *)aux; 782 break; 783 #endif 784 } 785 /* 786 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 787 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 788 * reasonable initial retransmit time. 789 */ 790 tp->t_srtt = TCPTV_SRTTBASE; 791 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 792 tp->t_rttmin = TCPTV_MIN; 793 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 794 TCPTV_MIN, TCPTV_REXMTMAX); 795 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 796 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 797 if (family == AF_INET) { 798 struct inpcb *inp = (struct inpcb *)aux; 799 inp->inp_ip.ip_ttl = ip_defttl; 800 inp->inp_ppcb = (caddr_t)tp; 801 } 802 #ifdef INET6 803 else if (family == AF_INET6) { 804 struct in6pcb *in6p = (struct in6pcb *)aux; 805 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 806 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 807 : NULL); 808 in6p->in6p_ppcb = (caddr_t)tp; 809 } 810 #endif 811 812 /* 813 * Initialize our timebase. When we send timestamps, we take 814 * the delta from tcp_now -- this means each connection always 815 * gets a timebase of 0, which makes it, among other things, 816 * more difficult to determine how long a system has been up, 817 * and thus how many TCP sequence increments have occurred. 818 */ 819 tp->ts_timebase = tcp_now; 820 821 return (tp); 822 } 823 824 /* 825 * Drop a TCP connection, reporting 826 * the specified error. If connection is synchronized, 827 * then send a RST to peer. 828 */ 829 struct tcpcb * 830 tcp_drop(tp, errno) 831 struct tcpcb *tp; 832 int errno; 833 { 834 struct socket *so = NULL; 835 836 #ifdef DIAGNOSTIC 837 if (tp->t_inpcb && tp->t_in6pcb) 838 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 839 #endif 840 #ifdef INET 841 if (tp->t_inpcb) 842 so = tp->t_inpcb->inp_socket; 843 #endif 844 #ifdef INET6 845 if (tp->t_in6pcb) 846 so = tp->t_in6pcb->in6p_socket; 847 #endif 848 if (!so) 849 return NULL; 850 851 if (TCPS_HAVERCVDSYN(tp->t_state)) { 852 tp->t_state = TCPS_CLOSED; 853 (void) tcp_output(tp); 854 tcpstat.tcps_drops++; 855 } else 856 tcpstat.tcps_conndrops++; 857 if (errno == ETIMEDOUT && tp->t_softerror) 858 errno = tp->t_softerror; 859 so->so_error = errno; 860 return (tcp_close(tp)); 861 } 862 863 /* 864 * Close a TCP control block: 865 * discard all space held by the tcp 866 * discard internet protocol block 867 * wake up any sleepers 868 */ 869 struct tcpcb * 870 tcp_close(tp) 871 struct tcpcb *tp; 872 { 873 struct inpcb *inp; 874 #ifdef INET6 875 struct in6pcb *in6p; 876 #endif 877 struct socket *so; 878 #ifdef RTV_RTT 879 struct rtentry *rt; 880 #endif 881 struct route *ro; 882 883 inp = tp->t_inpcb; 884 #ifdef INET6 885 in6p = tp->t_in6pcb; 886 #endif 887 so = NULL; 888 ro = NULL; 889 if (inp) { 890 so = inp->inp_socket; 891 ro = &inp->inp_route; 892 } 893 #ifdef INET6 894 else if (in6p) { 895 so = in6p->in6p_socket; 896 ro = (struct route *)&in6p->in6p_route; 897 } 898 #endif 899 900 #ifdef RTV_RTT 901 /* 902 * If we sent enough data to get some meaningful characteristics, 903 * save them in the routing entry. 'Enough' is arbitrarily 904 * defined as the sendpipesize (default 4K) * 16. This would 905 * give us 16 rtt samples assuming we only get one sample per 906 * window (the usual case on a long haul net). 16 samples is 907 * enough for the srtt filter to converge to within 5% of the correct 908 * value; fewer samples and we could save a very bogus rtt. 909 * 910 * Don't update the default route's characteristics and don't 911 * update anything that the user "locked". 912 */ 913 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 914 ro && (rt = ro->ro_rt) && 915 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 916 u_long i = 0; 917 918 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 919 i = tp->t_srtt * 920 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 921 if (rt->rt_rmx.rmx_rtt && i) 922 /* 923 * filter this update to half the old & half 924 * the new values, converting scale. 925 * See route.h and tcp_var.h for a 926 * description of the scaling constants. 927 */ 928 rt->rt_rmx.rmx_rtt = 929 (rt->rt_rmx.rmx_rtt + i) / 2; 930 else 931 rt->rt_rmx.rmx_rtt = i; 932 } 933 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 934 i = tp->t_rttvar * 935 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 936 if (rt->rt_rmx.rmx_rttvar && i) 937 rt->rt_rmx.rmx_rttvar = 938 (rt->rt_rmx.rmx_rttvar + i) / 2; 939 else 940 rt->rt_rmx.rmx_rttvar = i; 941 } 942 /* 943 * update the pipelimit (ssthresh) if it has been updated 944 * already or if a pipesize was specified & the threshhold 945 * got below half the pipesize. I.e., wait for bad news 946 * before we start updating, then update on both good 947 * and bad news. 948 */ 949 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 950 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 951 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 952 /* 953 * convert the limit from user data bytes to 954 * packets then to packet data bytes. 955 */ 956 i = (i + tp->t_segsz / 2) / tp->t_segsz; 957 if (i < 2) 958 i = 2; 959 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 960 if (rt->rt_rmx.rmx_ssthresh) 961 rt->rt_rmx.rmx_ssthresh = 962 (rt->rt_rmx.rmx_ssthresh + i) / 2; 963 else 964 rt->rt_rmx.rmx_ssthresh = i; 965 } 966 } 967 #endif /* RTV_RTT */ 968 /* free the reassembly queue, if any */ 969 TCP_REASS_LOCK(tp); 970 (void) tcp_freeq(tp); 971 TCP_REASS_UNLOCK(tp); 972 973 TCP_CLEAR_DELACK(tp); 974 syn_cache_cleanup(tp); 975 976 if (tp->t_template) { 977 m_free(tp->t_template); 978 tp->t_template = NULL; 979 } 980 pool_put(&tcpcb_pool, tp); 981 if (inp) { 982 inp->inp_ppcb = 0; 983 soisdisconnected(so); 984 in_pcbdetach(inp); 985 } 986 #ifdef INET6 987 else if (in6p) { 988 in6p->in6p_ppcb = 0; 989 soisdisconnected(so); 990 in6_pcbdetach(in6p); 991 } 992 #endif 993 tcpstat.tcps_closed++; 994 return ((struct tcpcb *)0); 995 } 996 997 int 998 tcp_freeq(tp) 999 struct tcpcb *tp; 1000 { 1001 struct ipqent *qe; 1002 int rv = 0; 1003 #ifdef TCPREASS_DEBUG 1004 int i = 0; 1005 #endif 1006 1007 TCP_REASS_LOCK_CHECK(tp); 1008 1009 while ((qe = tp->segq.lh_first) != NULL) { 1010 #ifdef TCPREASS_DEBUG 1011 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1012 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1013 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1014 #endif 1015 LIST_REMOVE(qe, ipqe_q); 1016 LIST_REMOVE(qe, ipqe_timeq); 1017 m_freem(qe->ipqe_m); 1018 pool_put(&ipqent_pool, qe); 1019 rv = 1; 1020 } 1021 return (rv); 1022 } 1023 1024 /* 1025 * Protocol drain routine. Called when memory is in short supply. 1026 */ 1027 void 1028 tcp_drain() 1029 { 1030 struct inpcb *inp; 1031 struct tcpcb *tp; 1032 1033 /* 1034 * Free the sequence queue of all TCP connections. 1035 */ 1036 inp = tcbtable.inpt_queue.cqh_first; 1037 if (inp) /* XXX */ 1038 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 1039 inp = inp->inp_queue.cqe_next) { 1040 if ((tp = intotcpcb(inp)) != NULL) { 1041 /* 1042 * We may be called from a device's interrupt 1043 * context. If the tcpcb is already busy, 1044 * just bail out now. 1045 */ 1046 if (tcp_reass_lock_try(tp) == 0) 1047 continue; 1048 if (tcp_freeq(tp)) 1049 tcpstat.tcps_connsdrained++; 1050 TCP_REASS_UNLOCK(tp); 1051 } 1052 } 1053 } 1054 1055 /* 1056 * Notify a tcp user of an asynchronous error; 1057 * store error as soft error, but wake up user 1058 * (for now, won't do anything until can select for soft error). 1059 */ 1060 void 1061 tcp_notify(inp, error) 1062 struct inpcb *inp; 1063 int error; 1064 { 1065 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1066 struct socket *so = inp->inp_socket; 1067 1068 /* 1069 * Ignore some errors if we are hooked up. 1070 * If connection hasn't completed, has retransmitted several times, 1071 * and receives a second error, give up now. This is better 1072 * than waiting a long time to establish a connection that 1073 * can never complete. 1074 */ 1075 if (tp->t_state == TCPS_ESTABLISHED && 1076 (error == EHOSTUNREACH || error == ENETUNREACH || 1077 error == EHOSTDOWN)) { 1078 return; 1079 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1080 tp->t_rxtshift > 3 && tp->t_softerror) 1081 so->so_error = error; 1082 else 1083 tp->t_softerror = error; 1084 wakeup((caddr_t) &so->so_timeo); 1085 sorwakeup(so); 1086 sowwakeup(so); 1087 } 1088 1089 #ifdef INET6 1090 void 1091 tcp6_notify(in6p, error) 1092 struct in6pcb *in6p; 1093 int error; 1094 { 1095 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1096 struct socket *so = in6p->in6p_socket; 1097 1098 /* 1099 * Ignore some errors if we are hooked up. 1100 * If connection hasn't completed, has retransmitted several times, 1101 * and receives a second error, give up now. This is better 1102 * than waiting a long time to establish a connection that 1103 * can never complete. 1104 */ 1105 if (tp->t_state == TCPS_ESTABLISHED && 1106 (error == EHOSTUNREACH || error == ENETUNREACH || 1107 error == EHOSTDOWN)) { 1108 return; 1109 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1110 tp->t_rxtshift > 3 && tp->t_softerror) 1111 so->so_error = error; 1112 else 1113 tp->t_softerror = error; 1114 wakeup((caddr_t) &so->so_timeo); 1115 sorwakeup(so); 1116 sowwakeup(so); 1117 } 1118 #endif 1119 1120 #ifdef INET6 1121 void 1122 tcp6_ctlinput(cmd, sa, d) 1123 int cmd; 1124 struct sockaddr *sa; 1125 void *d; 1126 { 1127 struct tcphdr th; 1128 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1129 int nmatch; 1130 struct ip6_hdr *ip6; 1131 const struct sockaddr_in6 *sa6_src = NULL; 1132 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1133 struct mbuf *m; 1134 int off; 1135 1136 if (sa->sa_family != AF_INET6 || 1137 sa->sa_len != sizeof(struct sockaddr_in6)) 1138 return; 1139 if ((unsigned)cmd >= PRC_NCMDS) 1140 return; 1141 else if (cmd == PRC_QUENCH) { 1142 /* XXX there's no PRC_QUENCH in IPv6 */ 1143 notify = tcp6_quench; 1144 } else if (PRC_IS_REDIRECT(cmd)) 1145 notify = in6_rtchange, d = NULL; 1146 else if (cmd == PRC_MSGSIZE) 1147 ; /* special code is present, see below */ 1148 else if (cmd == PRC_HOSTDEAD) 1149 d = NULL; 1150 else if (inet6ctlerrmap[cmd] == 0) 1151 return; 1152 1153 /* if the parameter is from icmp6, decode it. */ 1154 if (d != NULL) { 1155 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1156 m = ip6cp->ip6c_m; 1157 ip6 = ip6cp->ip6c_ip6; 1158 off = ip6cp->ip6c_off; 1159 sa6_src = ip6cp->ip6c_src; 1160 } else { 1161 m = NULL; 1162 ip6 = NULL; 1163 sa6_src = &sa6_any; 1164 } 1165 1166 if (ip6) { 1167 /* 1168 * XXX: We assume that when ip6 is non NULL, 1169 * M and OFF are valid. 1170 */ 1171 1172 /* check if we can safely examine src and dst ports */ 1173 if (m->m_pkthdr.len < off + sizeof(th)) 1174 return; 1175 1176 bzero(&th, sizeof(th)); 1177 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1178 1179 if (cmd == PRC_MSGSIZE) { 1180 int valid = 0; 1181 1182 /* 1183 * Check to see if we have a valid TCP connection 1184 * corresponding to the address in the ICMPv6 message 1185 * payload. 1186 */ 1187 if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr, 1188 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1189 th.th_sport, 0)) 1190 valid++; 1191 1192 /* 1193 * Depending on the value of "valid" and routing table 1194 * size (mtudisc_{hi,lo}wat), we will: 1195 * - recalcurate the new MTU and create the 1196 * corresponding routing entry, or 1197 * - ignore the MTU change notification. 1198 */ 1199 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1200 1201 /* 1202 * no need to call in6_pcbnotify, it should have been 1203 * called via callback if necessary 1204 */ 1205 return; 1206 } 1207 1208 nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport, 1209 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1210 if (nmatch == 0 && syn_cache_count && 1211 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1212 inet6ctlerrmap[cmd] == ENETUNREACH || 1213 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1214 syn_cache_unreach((struct sockaddr *)sa6_src, 1215 sa, &th); 1216 } else { 1217 (void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src, 1218 0, cmd, NULL, notify); 1219 } 1220 } 1221 #endif 1222 1223 #ifdef INET 1224 /* assumes that ip header and tcp header are contiguous on mbuf */ 1225 void * 1226 tcp_ctlinput(cmd, sa, v) 1227 int cmd; 1228 struct sockaddr *sa; 1229 void *v; 1230 { 1231 struct ip *ip = v; 1232 struct tcphdr *th; 1233 struct icmp *icp; 1234 extern int inetctlerrmap[]; 1235 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1236 int errno; 1237 int nmatch; 1238 1239 if (sa->sa_family != AF_INET || 1240 sa->sa_len != sizeof(struct sockaddr_in)) 1241 return NULL; 1242 if ((unsigned)cmd >= PRC_NCMDS) 1243 return NULL; 1244 errno = inetctlerrmap[cmd]; 1245 if (cmd == PRC_QUENCH) 1246 notify = tcp_quench; 1247 else if (PRC_IS_REDIRECT(cmd)) 1248 notify = in_rtchange, ip = 0; 1249 else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) { 1250 /* 1251 * Check to see if we have a valid TCP connection 1252 * corresponding to the address in the ICMP message 1253 * payload. 1254 */ 1255 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1256 if (in_pcblookup_connect(&tcbtable, 1257 ip->ip_dst, th->th_dport, 1258 ip->ip_src, th->th_sport) == NULL) 1259 return NULL; 1260 1261 /* 1262 * Now that we've validated that we are actually communicating 1263 * with the host indicated in the ICMP message, locate the 1264 * ICMP header, recalculate the new MTU, and create the 1265 * corresponding routing entry. 1266 */ 1267 icp = (struct icmp *)((caddr_t)ip - 1268 offsetof(struct icmp, icmp_ip)); 1269 icmp_mtudisc(icp, ip->ip_dst); 1270 1271 return NULL; 1272 } else if (cmd == PRC_HOSTDEAD) 1273 ip = 0; 1274 else if (errno == 0) 1275 return NULL; 1276 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1277 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1278 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1279 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1280 if (nmatch == 0 && syn_cache_count && 1281 (inetctlerrmap[cmd] == EHOSTUNREACH || 1282 inetctlerrmap[cmd] == ENETUNREACH || 1283 inetctlerrmap[cmd] == EHOSTDOWN)) { 1284 struct sockaddr_in sin; 1285 bzero(&sin, sizeof(sin)); 1286 sin.sin_len = sizeof(sin); 1287 sin.sin_family = AF_INET; 1288 sin.sin_port = th->th_sport; 1289 sin.sin_addr = ip->ip_src; 1290 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1291 } 1292 1293 /* XXX mapped address case */ 1294 } else 1295 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1296 notify); 1297 return NULL; 1298 } 1299 1300 /* 1301 * When a source quence is received, we are being notifed of congestion. 1302 * Close the congestion window down to the Loss Window (one segment). 1303 * We will gradually open it again as we proceed. 1304 */ 1305 void 1306 tcp_quench(inp, errno) 1307 struct inpcb *inp; 1308 int errno; 1309 { 1310 struct tcpcb *tp = intotcpcb(inp); 1311 1312 if (tp) 1313 tp->snd_cwnd = tp->t_segsz; 1314 } 1315 #endif 1316 1317 #ifdef INET6 1318 void 1319 tcp6_quench(in6p, errno) 1320 struct in6pcb *in6p; 1321 int errno; 1322 { 1323 struct tcpcb *tp = in6totcpcb(in6p); 1324 1325 if (tp) 1326 tp->snd_cwnd = tp->t_segsz; 1327 } 1328 #endif 1329 1330 #ifdef INET 1331 /* 1332 * Path MTU Discovery handlers. 1333 */ 1334 void 1335 tcp_mtudisc_callback(faddr) 1336 struct in_addr faddr; 1337 { 1338 1339 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1340 } 1341 1342 /* 1343 * On receipt of path MTU corrections, flush old route and replace it 1344 * with the new one. Retransmit all unacknowledged packets, to ensure 1345 * that all packets will be received. 1346 */ 1347 void 1348 tcp_mtudisc(inp, errno) 1349 struct inpcb *inp; 1350 int errno; 1351 { 1352 struct tcpcb *tp = intotcpcb(inp); 1353 struct rtentry *rt = in_pcbrtentry(inp); 1354 1355 if (tp != 0) { 1356 if (rt != 0) { 1357 /* 1358 * If this was not a host route, remove and realloc. 1359 */ 1360 if ((rt->rt_flags & RTF_HOST) == 0) { 1361 in_rtchange(inp, errno); 1362 if ((rt = in_pcbrtentry(inp)) == 0) 1363 return; 1364 } 1365 1366 /* 1367 * Slow start out of the error condition. We 1368 * use the MTU because we know it's smaller 1369 * than the previously transmitted segment. 1370 * 1371 * Note: This is more conservative than the 1372 * suggestion in draft-floyd-incr-init-win-03. 1373 */ 1374 if (rt->rt_rmx.rmx_mtu != 0) 1375 tp->snd_cwnd = 1376 TCP_INITIAL_WINDOW(tcp_init_win, 1377 rt->rt_rmx.rmx_mtu); 1378 } 1379 1380 /* 1381 * Resend unacknowledged packets. 1382 */ 1383 tp->snd_nxt = tp->snd_una; 1384 tcp_output(tp); 1385 } 1386 } 1387 #endif 1388 1389 #ifdef INET6 1390 /* 1391 * Path MTU Discovery handlers. 1392 */ 1393 void 1394 tcp6_mtudisc_callback(faddr) 1395 struct in6_addr *faddr; 1396 { 1397 struct sockaddr_in6 sin6; 1398 1399 bzero(&sin6, sizeof(sin6)); 1400 sin6.sin6_family = AF_INET6; 1401 sin6.sin6_len = sizeof(struct sockaddr_in6); 1402 sin6.sin6_addr = *faddr; 1403 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1404 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1405 } 1406 1407 void 1408 tcp6_mtudisc(in6p, errno) 1409 struct in6pcb *in6p; 1410 int errno; 1411 { 1412 struct tcpcb *tp = in6totcpcb(in6p); 1413 struct rtentry *rt = in6_pcbrtentry(in6p); 1414 1415 if (tp != 0) { 1416 if (rt != 0) { 1417 /* 1418 * If this was not a host route, remove and realloc. 1419 */ 1420 if ((rt->rt_flags & RTF_HOST) == 0) { 1421 in6_rtchange(in6p, errno); 1422 if ((rt = in6_pcbrtentry(in6p)) == 0) 1423 return; 1424 } 1425 1426 /* 1427 * Slow start out of the error condition. We 1428 * use the MTU because we know it's smaller 1429 * than the previously transmitted segment. 1430 * 1431 * Note: This is more conservative than the 1432 * suggestion in draft-floyd-incr-init-win-03. 1433 */ 1434 if (rt->rt_rmx.rmx_mtu != 0) 1435 tp->snd_cwnd = 1436 TCP_INITIAL_WINDOW(tcp_init_win, 1437 rt->rt_rmx.rmx_mtu); 1438 } 1439 1440 /* 1441 * Resend unacknowledged packets. 1442 */ 1443 tp->snd_nxt = tp->snd_una; 1444 tcp_output(tp); 1445 } 1446 } 1447 #endif /* INET6 */ 1448 1449 /* 1450 * Compute the MSS to advertise to the peer. Called only during 1451 * the 3-way handshake. If we are the server (peer initiated 1452 * connection), we are called with a pointer to the interface 1453 * on which the SYN packet arrived. If we are the client (we 1454 * initiated connection), we are called with a pointer to the 1455 * interface out which this connection should go. 1456 * 1457 * NOTE: Do not subtract IP option/extension header size nor IPsec 1458 * header size from MSS advertisement. MSS option must hold the maximum 1459 * segment size we can accept, so it must always be: 1460 * max(if mtu) - ip header - tcp header 1461 */ 1462 u_long 1463 tcp_mss_to_advertise(ifp, af) 1464 const struct ifnet *ifp; 1465 int af; 1466 { 1467 extern u_long in_maxmtu; 1468 u_long mss = 0; 1469 u_long hdrsiz; 1470 1471 /* 1472 * In order to avoid defeating path MTU discovery on the peer, 1473 * we advertise the max MTU of all attached networks as our MSS, 1474 * per RFC 1191, section 3.1. 1475 * 1476 * We provide the option to advertise just the MTU of 1477 * the interface on which we hope this connection will 1478 * be receiving. If we are responding to a SYN, we 1479 * will have a pretty good idea about this, but when 1480 * initiating a connection there is a bit more doubt. 1481 * 1482 * We also need to ensure that loopback has a large enough 1483 * MSS, as the loopback MTU is never included in in_maxmtu. 1484 */ 1485 1486 if (ifp != NULL) 1487 mss = ifp->if_mtu; 1488 1489 if (tcp_mss_ifmtu == 0) 1490 mss = max(in_maxmtu, mss); 1491 1492 switch (af) { 1493 case AF_INET: 1494 hdrsiz = sizeof(struct ip); 1495 break; 1496 #ifdef INET6 1497 case AF_INET6: 1498 hdrsiz = sizeof(struct ip6_hdr); 1499 break; 1500 #endif 1501 default: 1502 hdrsiz = 0; 1503 break; 1504 } 1505 hdrsiz += sizeof(struct tcphdr); 1506 if (mss > hdrsiz) 1507 mss -= hdrsiz; 1508 1509 mss = max(tcp_mssdflt, mss); 1510 return (mss); 1511 } 1512 1513 /* 1514 * Set connection variables based on the peer's advertised MSS. 1515 * We are passed the TCPCB for the actual connection. If we 1516 * are the server, we are called by the compressed state engine 1517 * when the 3-way handshake is complete. If we are the client, 1518 * we are called when we recieve the SYN,ACK from the server. 1519 * 1520 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1521 * before this routine is called! 1522 */ 1523 void 1524 tcp_mss_from_peer(tp, offer) 1525 struct tcpcb *tp; 1526 int offer; 1527 { 1528 struct socket *so; 1529 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1530 struct rtentry *rt; 1531 #endif 1532 u_long bufsize; 1533 int mss; 1534 1535 #ifdef DIAGNOSTIC 1536 if (tp->t_inpcb && tp->t_in6pcb) 1537 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1538 #endif 1539 so = NULL; 1540 rt = NULL; 1541 #ifdef INET 1542 if (tp->t_inpcb) { 1543 so = tp->t_inpcb->inp_socket; 1544 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1545 rt = in_pcbrtentry(tp->t_inpcb); 1546 #endif 1547 } 1548 #endif 1549 #ifdef INET6 1550 if (tp->t_in6pcb) { 1551 so = tp->t_in6pcb->in6p_socket; 1552 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1553 rt = in6_pcbrtentry(tp->t_in6pcb); 1554 #endif 1555 } 1556 #endif 1557 1558 /* 1559 * As per RFC1122, use the default MSS value, unless they 1560 * sent us an offer. Do not accept offers less than 32 bytes. 1561 */ 1562 mss = tcp_mssdflt; 1563 if (offer) 1564 mss = offer; 1565 mss = max(mss, 32); /* sanity */ 1566 tp->t_peermss = mss; 1567 mss -= tcp_optlen(tp); 1568 #ifdef INET 1569 if (tp->t_inpcb) 1570 mss -= ip_optlen(tp->t_inpcb); 1571 #endif 1572 #ifdef INET6 1573 if (tp->t_in6pcb) 1574 mss -= ip6_optlen(tp->t_in6pcb); 1575 #endif 1576 1577 /* 1578 * If there's a pipesize, change the socket buffer to that size. 1579 * Make the socket buffer an integral number of MSS units. If 1580 * the MSS is larger than the socket buffer, artificially decrease 1581 * the MSS. 1582 */ 1583 #ifdef RTV_SPIPE 1584 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1585 bufsize = rt->rt_rmx.rmx_sendpipe; 1586 else 1587 #endif 1588 bufsize = so->so_snd.sb_hiwat; 1589 if (bufsize < mss) 1590 mss = bufsize; 1591 else { 1592 bufsize = roundup(bufsize, mss); 1593 if (bufsize > sb_max) 1594 bufsize = sb_max; 1595 (void) sbreserve(&so->so_snd, bufsize); 1596 } 1597 tp->t_segsz = mss; 1598 1599 #ifdef RTV_SSTHRESH 1600 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1601 /* 1602 * There's some sort of gateway or interface buffer 1603 * limit on the path. Use this to set the slow 1604 * start threshold, but set the threshold to no less 1605 * than 2 * MSS. 1606 */ 1607 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1608 } 1609 #endif 1610 } 1611 1612 /* 1613 * Processing necessary when a TCP connection is established. 1614 */ 1615 void 1616 tcp_established(tp) 1617 struct tcpcb *tp; 1618 { 1619 struct socket *so; 1620 #ifdef RTV_RPIPE 1621 struct rtentry *rt; 1622 #endif 1623 u_long bufsize; 1624 1625 #ifdef DIAGNOSTIC 1626 if (tp->t_inpcb && tp->t_in6pcb) 1627 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1628 #endif 1629 so = NULL; 1630 rt = NULL; 1631 #ifdef INET 1632 if (tp->t_inpcb) { 1633 so = tp->t_inpcb->inp_socket; 1634 #if defined(RTV_RPIPE) 1635 rt = in_pcbrtentry(tp->t_inpcb); 1636 #endif 1637 } 1638 #endif 1639 #ifdef INET6 1640 if (tp->t_in6pcb) { 1641 so = tp->t_in6pcb->in6p_socket; 1642 #if defined(RTV_RPIPE) 1643 rt = in6_pcbrtentry(tp->t_in6pcb); 1644 #endif 1645 } 1646 #endif 1647 1648 tp->t_state = TCPS_ESTABLISHED; 1649 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1650 1651 #ifdef RTV_RPIPE 1652 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1653 bufsize = rt->rt_rmx.rmx_recvpipe; 1654 else 1655 #endif 1656 bufsize = so->so_rcv.sb_hiwat; 1657 if (bufsize > tp->t_ourmss) { 1658 bufsize = roundup(bufsize, tp->t_ourmss); 1659 if (bufsize > sb_max) 1660 bufsize = sb_max; 1661 (void) sbreserve(&so->so_rcv, bufsize); 1662 } 1663 } 1664 1665 /* 1666 * Check if there's an initial rtt or rttvar. Convert from the 1667 * route-table units to scaled multiples of the slow timeout timer. 1668 * Called only during the 3-way handshake. 1669 */ 1670 void 1671 tcp_rmx_rtt(tp) 1672 struct tcpcb *tp; 1673 { 1674 #ifdef RTV_RTT 1675 struct rtentry *rt = NULL; 1676 int rtt; 1677 1678 #ifdef DIAGNOSTIC 1679 if (tp->t_inpcb && tp->t_in6pcb) 1680 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1681 #endif 1682 #ifdef INET 1683 if (tp->t_inpcb) 1684 rt = in_pcbrtentry(tp->t_inpcb); 1685 #endif 1686 #ifdef INET6 1687 if (tp->t_in6pcb) 1688 rt = in6_pcbrtentry(tp->t_in6pcb); 1689 #endif 1690 if (rt == NULL) 1691 return; 1692 1693 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1694 /* 1695 * XXX The lock bit for MTU indicates that the value 1696 * is also a minimum value; this is subject to time. 1697 */ 1698 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1699 TCPT_RANGESET(tp->t_rttmin, 1700 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1701 TCPTV_MIN, TCPTV_REXMTMAX); 1702 tp->t_srtt = rtt / 1703 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1704 if (rt->rt_rmx.rmx_rttvar) { 1705 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1706 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1707 (TCP_RTTVAR_SHIFT + 2)); 1708 } else { 1709 /* Default variation is +- 1 rtt */ 1710 tp->t_rttvar = 1711 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1712 } 1713 TCPT_RANGESET(tp->t_rxtcur, 1714 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1715 tp->t_rttmin, TCPTV_REXMTMAX); 1716 } 1717 #endif 1718 } 1719 1720 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1721 #if NRND > 0 1722 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1723 #endif 1724 1725 /* 1726 * Get a new sequence value given a tcp control block 1727 */ 1728 tcp_seq 1729 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1730 { 1731 1732 #ifdef INET 1733 if (tp->t_inpcb != NULL) { 1734 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1735 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1736 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1737 addin)); 1738 } 1739 #endif 1740 #ifdef INET6 1741 if (tp->t_in6pcb != NULL) { 1742 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 1743 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 1744 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 1745 addin)); 1746 } 1747 #endif 1748 /* Not possible. */ 1749 panic("tcp_new_iss"); 1750 } 1751 1752 /* 1753 * This routine actually generates a new TCP initial sequence number. 1754 */ 1755 tcp_seq 1756 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 1757 size_t addrsz, tcp_seq addin) 1758 { 1759 tcp_seq tcp_iss; 1760 1761 #if NRND > 0 1762 static int beenhere; 1763 1764 /* 1765 * If we haven't been here before, initialize our cryptographic 1766 * hash secret. 1767 */ 1768 if (beenhere == 0) { 1769 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 1770 RND_EXTRACT_ANY); 1771 beenhere = 1; 1772 } 1773 1774 if (tcp_do_rfc1948) { 1775 MD5_CTX ctx; 1776 u_int8_t hash[16]; /* XXX MD5 knowledge */ 1777 1778 /* 1779 * Compute the base value of the ISS. It is a hash 1780 * of (saddr, sport, daddr, dport, secret). 1781 */ 1782 MD5Init(&ctx); 1783 1784 MD5Update(&ctx, (u_char *) laddr, addrsz); 1785 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 1786 1787 MD5Update(&ctx, (u_char *) faddr, addrsz); 1788 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 1789 1790 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 1791 1792 MD5Final(hash, &ctx); 1793 1794 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 1795 1796 /* 1797 * Now increment our "timer", and add it in to 1798 * the computed value. 1799 * 1800 * XXX Use `addin'? 1801 * XXX TCP_ISSINCR too large to use? 1802 */ 1803 tcp_iss_seq += TCP_ISSINCR; 1804 #ifdef TCPISS_DEBUG 1805 printf("ISS hash 0x%08x, ", tcp_iss); 1806 #endif 1807 tcp_iss += tcp_iss_seq + addin; 1808 #ifdef TCPISS_DEBUG 1809 printf("new ISS 0x%08x\n", tcp_iss); 1810 #endif 1811 } else 1812 #endif /* NRND > 0 */ 1813 { 1814 /* 1815 * Randomize. 1816 */ 1817 #if NRND > 0 1818 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 1819 #else 1820 tcp_iss = random(); 1821 #endif 1822 1823 /* 1824 * If we were asked to add some amount to a known value, 1825 * we will take a random value obtained above, mask off 1826 * the upper bits, and add in the known value. We also 1827 * add in a constant to ensure that we are at least a 1828 * certain distance from the original value. 1829 * 1830 * This is used when an old connection is in timed wait 1831 * and we have a new one coming in, for instance. 1832 */ 1833 if (addin != 0) { 1834 #ifdef TCPISS_DEBUG 1835 printf("Random %08x, ", tcp_iss); 1836 #endif 1837 tcp_iss &= TCP_ISS_RANDOM_MASK; 1838 tcp_iss += addin + TCP_ISSINCR; 1839 #ifdef TCPISS_DEBUG 1840 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 1841 #endif 1842 } else { 1843 tcp_iss &= TCP_ISS_RANDOM_MASK; 1844 tcp_iss += tcp_iss_seq; 1845 tcp_iss_seq += TCP_ISSINCR; 1846 #ifdef TCPISS_DEBUG 1847 printf("ISS %08x\n", tcp_iss); 1848 #endif 1849 } 1850 } 1851 1852 if (tcp_compat_42) { 1853 /* 1854 * Limit it to the positive range for really old TCP 1855 * implementations. 1856 */ 1857 if (tcp_iss >= 0x80000000) 1858 tcp_iss &= 0x7fffffff; /* XXX */ 1859 } 1860 1861 return (tcp_iss); 1862 } 1863 1864 #ifdef IPSEC 1865 /* compute ESP/AH header size for TCP, including outer IP header. */ 1866 size_t 1867 ipsec4_hdrsiz_tcp(tp) 1868 struct tcpcb *tp; 1869 { 1870 struct inpcb *inp; 1871 size_t hdrsiz; 1872 1873 /* XXX mapped addr case (tp->t_in6pcb) */ 1874 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 1875 return 0; 1876 switch (tp->t_family) { 1877 case AF_INET: 1878 /* XXX: should use currect direction. */ 1879 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 1880 break; 1881 default: 1882 hdrsiz = 0; 1883 break; 1884 } 1885 1886 return hdrsiz; 1887 } 1888 1889 #ifdef INET6 1890 size_t 1891 ipsec6_hdrsiz_tcp(tp) 1892 struct tcpcb *tp; 1893 { 1894 struct in6pcb *in6p; 1895 size_t hdrsiz; 1896 1897 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 1898 return 0; 1899 switch (tp->t_family) { 1900 case AF_INET6: 1901 /* XXX: should use currect direction. */ 1902 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 1903 break; 1904 case AF_INET: 1905 /* mapped address case - tricky */ 1906 default: 1907 hdrsiz = 0; 1908 break; 1909 } 1910 1911 return hdrsiz; 1912 } 1913 #endif 1914 #endif /*IPSEC*/ 1915 1916 /* 1917 * Determine the length of the TCP options for this connection. 1918 * 1919 * XXX: What do we do for SACK, when we add that? Just reserve 1920 * all of the space? Otherwise we can't exactly be incrementing 1921 * cwnd by an amount that varies depending on the amount we last 1922 * had to SACK! 1923 */ 1924 1925 u_int 1926 tcp_optlen(tp) 1927 struct tcpcb *tp; 1928 { 1929 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1930 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1931 return TCPOLEN_TSTAMP_APPA; 1932 else 1933 return 0; 1934 } 1935