tcp_subr.c revision 1.98
1 /* $NetBSD: tcp_subr.c,v 1.98 2000/10/18 17:09:15 thorpej Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998, 2000 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 #endif 144 145 #include <netinet/tcp.h> 146 #include <netinet/tcp_fsm.h> 147 #include <netinet/tcp_seq.h> 148 #include <netinet/tcp_timer.h> 149 #include <netinet/tcp_var.h> 150 #include <netinet/tcpip.h> 151 152 #ifdef IPSEC 153 #include <netinet6/ipsec.h> 154 #endif /*IPSEC*/ 155 156 #ifdef INET6 157 struct in6pcb tcb6; 158 #endif 159 160 /* patchable/settable parameters for tcp */ 161 int tcp_mssdflt = TCP_MSS; 162 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 163 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 164 int tcp_do_sack = 1; /* selective acknowledgement */ 165 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 166 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 167 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 168 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 169 int tcp_init_win = 1; 170 int tcp_mss_ifmtu = 0; 171 #ifdef TCP_COMPAT_42 172 int tcp_compat_42 = 1; 173 #else 174 int tcp_compat_42 = 0; 175 #endif 176 int tcp_rst_ppslim = 100; /* 100pps */ 177 178 /* tcb hash */ 179 #ifndef TCBHASHSIZE 180 #define TCBHASHSIZE 128 181 #endif 182 int tcbhashsize = TCBHASHSIZE; 183 184 /* syn hash parameters */ 185 #define TCP_SYN_HASH_SIZE 293 186 #define TCP_SYN_BUCKET_SIZE 35 187 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 188 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 189 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 190 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 191 int tcp_syn_cache_interval = 1; /* runs timer twice a second */ 192 193 int tcp_freeq __P((struct tcpcb *)); 194 195 void tcp_mtudisc_callback __P((struct in_addr)); 196 197 void tcp_mtudisc __P((struct inpcb *, int)); 198 #if defined(INET6) && !defined(TCP6) 199 void tcp6_mtudisc __P((struct in6pcb *, int)); 200 #endif 201 202 struct pool tcpcb_pool; 203 204 /* 205 * Tcp initialization 206 */ 207 void 208 tcp_init() 209 { 210 int hlen; 211 212 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 213 0, NULL, NULL, M_PCB); 214 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 215 #ifdef INET6 216 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 217 #endif 218 LIST_INIT(&tcp_delacks); 219 220 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 221 #ifdef INET6 222 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 223 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 224 #endif 225 if (max_protohdr < hlen) 226 max_protohdr = hlen; 227 if (max_linkhdr + hlen > MHLEN) 228 panic("tcp_init"); 229 230 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 231 232 /* Initialize the compressed state engine. */ 233 syn_cache_init(); 234 } 235 236 /* 237 * Create template to be used to send tcp packets on a connection. 238 * Call after host entry created, allocates an mbuf and fills 239 * in a skeletal tcp/ip header, minimizing the amount of work 240 * necessary when the connection is used. 241 */ 242 struct mbuf * 243 tcp_template(tp) 244 struct tcpcb *tp; 245 { 246 struct inpcb *inp = tp->t_inpcb; 247 #ifdef INET6 248 struct in6pcb *in6p = tp->t_in6pcb; 249 #endif 250 struct tcphdr *n; 251 struct mbuf *m; 252 int hlen; 253 254 switch (tp->t_family) { 255 case AF_INET: 256 hlen = sizeof(struct ip); 257 if (inp) 258 break; 259 #ifdef INET6 260 if (in6p) { 261 /* mapped addr case */ 262 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 263 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 264 break; 265 } 266 #endif 267 return NULL; /*EINVAL*/ 268 #ifdef INET6 269 case AF_INET6: 270 hlen = sizeof(struct ip6_hdr); 271 if (in6p) { 272 /* more sainty check? */ 273 break; 274 } 275 return NULL; /*EINVAL*/ 276 #endif 277 default: 278 hlen = 0; /*pacify gcc*/ 279 return NULL; /*EAFNOSUPPORT*/ 280 } 281 #ifdef DIAGNOSTIC 282 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 283 panic("mclbytes too small for t_template"); 284 #endif 285 m = tp->t_template; 286 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 287 ; 288 else { 289 if (m) 290 m_freem(m); 291 m = tp->t_template = NULL; 292 MGETHDR(m, M_DONTWAIT, MT_HEADER); 293 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 294 MCLGET(m, M_DONTWAIT); 295 if ((m->m_flags & M_EXT) == 0) { 296 m_free(m); 297 m = NULL; 298 } 299 } 300 if (m == NULL) 301 return NULL; 302 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 303 } 304 bzero(mtod(m, caddr_t), m->m_len); 305 switch (tp->t_family) { 306 case AF_INET: 307 { 308 struct ipovly *ipov; 309 mtod(m, struct ip *)->ip_v = 4; 310 ipov = mtod(m, struct ipovly *); 311 ipov->ih_pr = IPPROTO_TCP; 312 ipov->ih_len = htons(sizeof(struct tcphdr)); 313 if (inp) { 314 ipov->ih_src = inp->inp_laddr; 315 ipov->ih_dst = inp->inp_faddr; 316 } 317 #ifdef INET6 318 else if (in6p) { 319 /* mapped addr case */ 320 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 321 sizeof(ipov->ih_src)); 322 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 323 sizeof(ipov->ih_dst)); 324 } 325 #endif 326 break; 327 } 328 #ifdef INET6 329 case AF_INET6: 330 { 331 struct ip6_hdr *ip6; 332 mtod(m, struct ip *)->ip_v = 6; 333 ip6 = mtod(m, struct ip6_hdr *); 334 ip6->ip6_nxt = IPPROTO_TCP; 335 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 336 ip6->ip6_src = in6p->in6p_laddr; 337 ip6->ip6_dst = in6p->in6p_faddr; 338 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 339 if (ip6_auto_flowlabel) { 340 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 341 ip6->ip6_flow |= 342 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 343 } 344 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 345 ip6->ip6_vfc |= IPV6_VERSION; 346 break; 347 } 348 #endif 349 } 350 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 351 if (inp) { 352 n->th_sport = inp->inp_lport; 353 n->th_dport = inp->inp_fport; 354 } 355 #ifdef INET6 356 else if (in6p) { 357 n->th_sport = in6p->in6p_lport; 358 n->th_dport = in6p->in6p_fport; 359 } 360 #endif 361 n->th_seq = 0; 362 n->th_ack = 0; 363 n->th_x2 = 0; 364 n->th_off = 5; 365 n->th_flags = 0; 366 n->th_win = 0; 367 n->th_sum = 0; 368 n->th_urp = 0; 369 return (m); 370 } 371 372 /* 373 * Send a single message to the TCP at address specified by 374 * the given TCP/IP header. If m == 0, then we make a copy 375 * of the tcpiphdr at ti and send directly to the addressed host. 376 * This is used to force keep alive messages out using the TCP 377 * template for a connection tp->t_template. If flags are given 378 * then we send a message back to the TCP which originated the 379 * segment ti, and discard the mbuf containing it and any other 380 * attached mbufs. 381 * 382 * In any case the ack and sequence number of the transmitted 383 * segment are as specified by the parameters. 384 */ 385 int 386 tcp_respond(tp, template, m, th0, ack, seq, flags) 387 struct tcpcb *tp; 388 struct mbuf *template; 389 struct mbuf *m; 390 struct tcphdr *th0; 391 tcp_seq ack, seq; 392 int flags; 393 { 394 struct route *ro; 395 int error, tlen, win = 0; 396 int hlen; 397 struct ip *ip; 398 #ifdef INET6 399 struct ip6_hdr *ip6; 400 #endif 401 int family; /* family on packet, not inpcb/in6pcb! */ 402 struct tcphdr *th; 403 404 if (tp != NULL && (flags & TH_RST) == 0) { 405 #ifdef DIAGNOSTIC 406 if (tp->t_inpcb && tp->t_in6pcb) 407 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 408 #endif 409 #ifdef INET 410 if (tp->t_inpcb) 411 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 412 #endif 413 #ifdef INET6 414 if (tp->t_in6pcb) 415 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 416 #endif 417 } 418 419 ip = NULL; 420 #ifdef INET6 421 ip6 = NULL; 422 #endif 423 if (m == 0) { 424 if (!template) 425 return EINVAL; 426 427 /* get family information from template */ 428 switch (mtod(template, struct ip *)->ip_v) { 429 case 4: 430 family = AF_INET; 431 hlen = sizeof(struct ip); 432 break; 433 #ifdef INET6 434 case 6: 435 family = AF_INET6; 436 hlen = sizeof(struct ip6_hdr); 437 break; 438 #endif 439 default: 440 return EAFNOSUPPORT; 441 } 442 443 MGETHDR(m, M_DONTWAIT, MT_HEADER); 444 if (m) { 445 MCLGET(m, M_DONTWAIT); 446 if ((m->m_flags & M_EXT) == 0) { 447 m_free(m); 448 m = NULL; 449 } 450 } 451 if (m == NULL) 452 return (ENOBUFS); 453 454 if (tcp_compat_42) 455 tlen = 1; 456 else 457 tlen = 0; 458 459 m->m_data += max_linkhdr; 460 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 461 template->m_len); 462 switch (family) { 463 case AF_INET: 464 ip = mtod(m, struct ip *); 465 th = (struct tcphdr *)(ip + 1); 466 break; 467 #ifdef INET6 468 case AF_INET6: 469 ip6 = mtod(m, struct ip6_hdr *); 470 th = (struct tcphdr *)(ip6 + 1); 471 break; 472 #endif 473 #if 0 474 default: 475 /* noone will visit here */ 476 m_freem(m); 477 return EAFNOSUPPORT; 478 #endif 479 } 480 flags = TH_ACK; 481 } else { 482 483 if ((m->m_flags & M_PKTHDR) == 0) { 484 #if 0 485 printf("non PKTHDR to tcp_respond\n"); 486 #endif 487 m_freem(m); 488 return EINVAL; 489 } 490 #ifdef DIAGNOSTIC 491 if (!th0) 492 panic("th0 == NULL in tcp_respond"); 493 #endif 494 495 /* get family information from m */ 496 switch (mtod(m, struct ip *)->ip_v) { 497 case 4: 498 family = AF_INET; 499 hlen = sizeof(struct ip); 500 ip = mtod(m, struct ip *); 501 break; 502 #ifdef INET6 503 case 6: 504 family = AF_INET6; 505 hlen = sizeof(struct ip6_hdr); 506 ip6 = mtod(m, struct ip6_hdr *); 507 break; 508 #endif 509 default: 510 m_freem(m); 511 return EAFNOSUPPORT; 512 } 513 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 514 tlen = sizeof(*th0); 515 else 516 tlen = th0->th_off << 2; 517 518 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 519 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 520 m->m_len = hlen + tlen; 521 m_freem(m->m_next); 522 m->m_next = NULL; 523 } else { 524 struct mbuf *n; 525 526 #ifdef DIAGNOSTIC 527 if (max_linkhdr + hlen + tlen > MCLBYTES) { 528 m_freem(m); 529 return EMSGSIZE; 530 } 531 #endif 532 MGETHDR(n, M_DONTWAIT, MT_HEADER); 533 if (n && max_linkhdr + hlen + tlen > MHLEN) { 534 MCLGET(n, M_DONTWAIT); 535 if ((n->m_flags & M_EXT) == 0) { 536 m_freem(n); 537 n = NULL; 538 } 539 } 540 if (!n) { 541 m_freem(m); 542 return ENOBUFS; 543 } 544 545 n->m_data += max_linkhdr; 546 n->m_len = hlen + tlen; 547 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 548 m_copyback(n, hlen, tlen, (caddr_t)th0); 549 550 m_freem(m); 551 m = n; 552 n = NULL; 553 } 554 555 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 556 switch (family) { 557 case AF_INET: 558 ip = mtod(m, struct ip *); 559 th = (struct tcphdr *)(ip + 1); 560 ip->ip_p = IPPROTO_TCP; 561 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 562 ip->ip_p = IPPROTO_TCP; 563 break; 564 #ifdef INET6 565 case AF_INET6: 566 ip6 = mtod(m, struct ip6_hdr *); 567 th = (struct tcphdr *)(ip6 + 1); 568 ip6->ip6_nxt = IPPROTO_TCP; 569 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 570 ip6->ip6_nxt = IPPROTO_TCP; 571 break; 572 #endif 573 #if 0 574 default: 575 /* noone will visit here */ 576 m_freem(m); 577 return EAFNOSUPPORT; 578 #endif 579 } 580 xchg(th->th_dport, th->th_sport, u_int16_t); 581 #undef xchg 582 tlen = 0; /*be friendly with the following code*/ 583 } 584 th->th_seq = htonl(seq); 585 th->th_ack = htonl(ack); 586 th->th_x2 = 0; 587 if ((flags & TH_SYN) == 0) { 588 if (tp) 589 win >>= tp->rcv_scale; 590 if (win > TCP_MAXWIN) 591 win = TCP_MAXWIN; 592 th->th_win = htons((u_int16_t)win); 593 th->th_off = sizeof (struct tcphdr) >> 2; 594 tlen += sizeof(*th); 595 } else 596 tlen += th->th_off << 2; 597 m->m_len = hlen + tlen; 598 m->m_pkthdr.len = hlen + tlen; 599 m->m_pkthdr.rcvif = (struct ifnet *) 0; 600 th->th_flags = flags; 601 th->th_urp = 0; 602 603 switch (family) { 604 #ifdef INET 605 case AF_INET: 606 { 607 struct ipovly *ipov = (struct ipovly *)ip; 608 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 609 ipov->ih_len = htons((u_int16_t)tlen); 610 611 th->th_sum = 0; 612 th->th_sum = in_cksum(m, hlen + tlen); 613 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 614 ip->ip_ttl = ip_defttl; 615 break; 616 } 617 #endif 618 #ifdef INET6 619 case AF_INET6: 620 { 621 th->th_sum = 0; 622 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 623 tlen); 624 ip6->ip6_plen = ntohs(tlen); 625 if (tp && tp->t_in6pcb) { 626 struct ifnet *oifp; 627 ro = (struct route *)&tp->t_in6pcb->in6p_route; 628 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 629 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 630 } else 631 ip6->ip6_hlim = ip6_defhlim; 632 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 633 if (ip6_auto_flowlabel) { 634 ip6->ip6_flow |= 635 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 636 } 637 break; 638 } 639 #endif 640 } 641 642 #ifdef IPSEC 643 ipsec_setsocket(m, NULL); 644 #endif /*IPSEC*/ 645 646 if (tp != NULL && tp->t_inpcb != NULL) { 647 ro = &tp->t_inpcb->inp_route; 648 #ifdef IPSEC 649 ipsec_setsocket(m, tp->t_inpcb->inp_socket); 650 #endif 651 #ifdef DIAGNOSTIC 652 if (family != AF_INET) 653 panic("tcp_respond: address family mismatch"); 654 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 655 panic("tcp_respond: ip_dst %x != inp_faddr %x", 656 ntohl(ip->ip_dst.s_addr), 657 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 658 } 659 #endif 660 } 661 #ifdef INET6 662 else if (tp != NULL && tp->t_in6pcb != NULL) { 663 ro = (struct route *)&tp->t_in6pcb->in6p_route; 664 #ifdef IPSEC 665 ipsec_setsocket(m, tp->t_in6pcb->in6p_socket); 666 #endif 667 #ifdef DIAGNOSTIC 668 if (family == AF_INET) { 669 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 670 panic("tcp_respond: not mapped addr"); 671 if (bcmp(&ip->ip_dst, 672 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 673 sizeof(ip->ip_dst)) != 0) { 674 panic("tcp_respond: ip_dst != in6p_faddr"); 675 } 676 } else if (family == AF_INET6) { 677 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 678 panic("tcp_respond: ip6_dst != in6p_faddr"); 679 } else 680 panic("tcp_respond: address family mismatch"); 681 #endif 682 } 683 #endif 684 else 685 ro = NULL; 686 687 switch (family) { 688 #ifdef INET 689 case AF_INET: 690 error = ip_output(m, NULL, ro, 691 (ip_mtudisc ? IP_MTUDISC : 0), 692 NULL); 693 break; 694 #endif 695 #ifdef INET6 696 case AF_INET6: 697 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 698 NULL); 699 break; 700 #endif 701 default: 702 error = EAFNOSUPPORT; 703 break; 704 } 705 706 return (error); 707 } 708 709 /* 710 * Create a new TCP control block, making an 711 * empty reassembly queue and hooking it to the argument 712 * protocol control block. 713 */ 714 struct tcpcb * 715 tcp_newtcpcb(family, aux) 716 int family; /* selects inpcb, or in6pcb */ 717 void *aux; 718 { 719 struct tcpcb *tp; 720 721 switch (family) { 722 case PF_INET: 723 break; 724 #ifdef INET6 725 case PF_INET6: 726 break; 727 #endif 728 default: 729 return NULL; 730 } 731 732 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 733 if (tp == NULL) 734 return (NULL); 735 bzero((caddr_t)tp, sizeof(struct tcpcb)); 736 LIST_INIT(&tp->segq); 737 LIST_INIT(&tp->timeq); 738 tp->t_family = family; /* may be overridden later on */ 739 tp->t_peermss = tcp_mssdflt; 740 tp->t_ourmss = tcp_mssdflt; 741 tp->t_segsz = tcp_mssdflt; 742 LIST_INIT(&tp->t_sc); 743 744 tp->t_flags = 0; 745 if (tcp_do_rfc1323 && tcp_do_win_scale) 746 tp->t_flags |= TF_REQ_SCALE; 747 if (tcp_do_rfc1323 && tcp_do_timestamps) 748 tp->t_flags |= TF_REQ_TSTMP; 749 if (tcp_do_sack == 2) 750 tp->t_flags |= TF_WILL_SACK; 751 else if (tcp_do_sack == 1) 752 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 753 tp->t_flags |= TF_CANT_TXSACK; 754 switch (family) { 755 case PF_INET: 756 tp->t_inpcb = (struct inpcb *)aux; 757 break; 758 #ifdef INET6 759 case PF_INET6: 760 tp->t_in6pcb = (struct in6pcb *)aux; 761 break; 762 #endif 763 } 764 /* 765 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 766 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 767 * reasonable initial retransmit time. 768 */ 769 tp->t_srtt = TCPTV_SRTTBASE; 770 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 771 tp->t_rttmin = TCPTV_MIN; 772 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 773 TCPTV_MIN, TCPTV_REXMTMAX); 774 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 775 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 776 if (family == AF_INET) { 777 struct inpcb *inp = (struct inpcb *)aux; 778 inp->inp_ip.ip_ttl = ip_defttl; 779 inp->inp_ppcb = (caddr_t)tp; 780 } 781 #ifdef INET6 782 else if (family == AF_INET6) { 783 struct in6pcb *in6p = (struct in6pcb *)aux; 784 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 785 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 786 : NULL); 787 in6p->in6p_ppcb = (caddr_t)tp; 788 } 789 #endif 790 return (tp); 791 } 792 793 /* 794 * Drop a TCP connection, reporting 795 * the specified error. If connection is synchronized, 796 * then send a RST to peer. 797 */ 798 struct tcpcb * 799 tcp_drop(tp, errno) 800 struct tcpcb *tp; 801 int errno; 802 { 803 struct socket *so; 804 805 #ifdef DIAGNOSTIC 806 if (tp->t_inpcb && tp->t_in6pcb) 807 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 808 #endif 809 #ifdef INET 810 if (tp->t_inpcb) 811 so = tp->t_inpcb->inp_socket; 812 #endif 813 #ifdef INET6 814 if (tp->t_in6pcb) 815 so = tp->t_in6pcb->in6p_socket; 816 #endif 817 else 818 return NULL; 819 820 if (TCPS_HAVERCVDSYN(tp->t_state)) { 821 tp->t_state = TCPS_CLOSED; 822 (void) tcp_output(tp); 823 tcpstat.tcps_drops++; 824 } else 825 tcpstat.tcps_conndrops++; 826 if (errno == ETIMEDOUT && tp->t_softerror) 827 errno = tp->t_softerror; 828 so->so_error = errno; 829 return (tcp_close(tp)); 830 } 831 832 /* 833 * Close a TCP control block: 834 * discard all space held by the tcp 835 * discard internet protocol block 836 * wake up any sleepers 837 */ 838 struct tcpcb * 839 tcp_close(tp) 840 struct tcpcb *tp; 841 { 842 struct inpcb *inp; 843 #ifdef INET6 844 struct in6pcb *in6p; 845 #endif 846 struct socket *so; 847 #ifdef RTV_RTT 848 struct rtentry *rt; 849 #endif 850 struct route *ro; 851 852 inp = tp->t_inpcb; 853 #ifdef INET6 854 in6p = tp->t_in6pcb; 855 #endif 856 so = NULL; 857 ro = NULL; 858 if (inp) { 859 so = inp->inp_socket; 860 ro = &inp->inp_route; 861 } 862 #ifdef INET6 863 else if (in6p) { 864 so = in6p->in6p_socket; 865 ro = (struct route *)&in6p->in6p_route; 866 } 867 #endif 868 869 #ifdef RTV_RTT 870 /* 871 * If we sent enough data to get some meaningful characteristics, 872 * save them in the routing entry. 'Enough' is arbitrarily 873 * defined as the sendpipesize (default 4K) * 16. This would 874 * give us 16 rtt samples assuming we only get one sample per 875 * window (the usual case on a long haul net). 16 samples is 876 * enough for the srtt filter to converge to within 5% of the correct 877 * value; fewer samples and we could save a very bogus rtt. 878 * 879 * Don't update the default route's characteristics and don't 880 * update anything that the user "locked". 881 */ 882 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 883 ro && (rt = ro->ro_rt) && 884 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 885 u_long i = 0; 886 887 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 888 i = tp->t_srtt * 889 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 890 if (rt->rt_rmx.rmx_rtt && i) 891 /* 892 * filter this update to half the old & half 893 * the new values, converting scale. 894 * See route.h and tcp_var.h for a 895 * description of the scaling constants. 896 */ 897 rt->rt_rmx.rmx_rtt = 898 (rt->rt_rmx.rmx_rtt + i) / 2; 899 else 900 rt->rt_rmx.rmx_rtt = i; 901 } 902 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 903 i = tp->t_rttvar * 904 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 905 if (rt->rt_rmx.rmx_rttvar && i) 906 rt->rt_rmx.rmx_rttvar = 907 (rt->rt_rmx.rmx_rttvar + i) / 2; 908 else 909 rt->rt_rmx.rmx_rttvar = i; 910 } 911 /* 912 * update the pipelimit (ssthresh) if it has been updated 913 * already or if a pipesize was specified & the threshhold 914 * got below half the pipesize. I.e., wait for bad news 915 * before we start updating, then update on both good 916 * and bad news. 917 */ 918 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 919 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 920 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 921 /* 922 * convert the limit from user data bytes to 923 * packets then to packet data bytes. 924 */ 925 i = (i + tp->t_segsz / 2) / tp->t_segsz; 926 if (i < 2) 927 i = 2; 928 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 929 if (rt->rt_rmx.rmx_ssthresh) 930 rt->rt_rmx.rmx_ssthresh = 931 (rt->rt_rmx.rmx_ssthresh + i) / 2; 932 else 933 rt->rt_rmx.rmx_ssthresh = i; 934 } 935 } 936 #endif /* RTV_RTT */ 937 /* free the reassembly queue, if any */ 938 TCP_REASS_LOCK(tp); 939 (void) tcp_freeq(tp); 940 TCP_REASS_UNLOCK(tp); 941 942 TCP_CLEAR_DELACK(tp); 943 syn_cache_cleanup(tp); 944 945 if (tp->t_template) { 946 m_free(tp->t_template); 947 tp->t_template = NULL; 948 } 949 pool_put(&tcpcb_pool, tp); 950 if (inp) { 951 inp->inp_ppcb = 0; 952 soisdisconnected(so); 953 in_pcbdetach(inp); 954 } 955 #ifdef INET6 956 else if (in6p) { 957 in6p->in6p_ppcb = 0; 958 soisdisconnected(so); 959 in6_pcbdetach(in6p); 960 } 961 #endif 962 tcpstat.tcps_closed++; 963 return ((struct tcpcb *)0); 964 } 965 966 int 967 tcp_freeq(tp) 968 struct tcpcb *tp; 969 { 970 struct ipqent *qe; 971 int rv = 0; 972 #ifdef TCPREASS_DEBUG 973 int i = 0; 974 #endif 975 976 TCP_REASS_LOCK_CHECK(tp); 977 978 while ((qe = tp->segq.lh_first) != NULL) { 979 #ifdef TCPREASS_DEBUG 980 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 981 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 982 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 983 #endif 984 LIST_REMOVE(qe, ipqe_q); 985 LIST_REMOVE(qe, ipqe_timeq); 986 m_freem(qe->ipqe_m); 987 pool_put(&ipqent_pool, qe); 988 rv = 1; 989 } 990 return (rv); 991 } 992 993 /* 994 * Protocol drain routine. Called when memory is in short supply. 995 */ 996 void 997 tcp_drain() 998 { 999 struct inpcb *inp; 1000 struct tcpcb *tp; 1001 1002 /* 1003 * Free the sequence queue of all TCP connections. 1004 */ 1005 inp = tcbtable.inpt_queue.cqh_first; 1006 if (inp) /* XXX */ 1007 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 1008 inp = inp->inp_queue.cqe_next) { 1009 if ((tp = intotcpcb(inp)) != NULL) { 1010 /* 1011 * We may be called from a device's interrupt 1012 * context. If the tcpcb is already busy, 1013 * just bail out now. 1014 */ 1015 if (tcp_reass_lock_try(tp) == 0) 1016 continue; 1017 if (tcp_freeq(tp)) 1018 tcpstat.tcps_connsdrained++; 1019 TCP_REASS_UNLOCK(tp); 1020 } 1021 } 1022 } 1023 1024 /* 1025 * Notify a tcp user of an asynchronous error; 1026 * store error as soft error, but wake up user 1027 * (for now, won't do anything until can select for soft error). 1028 */ 1029 void 1030 tcp_notify(inp, error) 1031 struct inpcb *inp; 1032 int error; 1033 { 1034 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1035 struct socket *so = inp->inp_socket; 1036 1037 /* 1038 * Ignore some errors if we are hooked up. 1039 * If connection hasn't completed, has retransmitted several times, 1040 * and receives a second error, give up now. This is better 1041 * than waiting a long time to establish a connection that 1042 * can never complete. 1043 */ 1044 if (tp->t_state == TCPS_ESTABLISHED && 1045 (error == EHOSTUNREACH || error == ENETUNREACH || 1046 error == EHOSTDOWN)) { 1047 return; 1048 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1049 tp->t_rxtshift > 3 && tp->t_softerror) 1050 so->so_error = error; 1051 else 1052 tp->t_softerror = error; 1053 wakeup((caddr_t) &so->so_timeo); 1054 sorwakeup(so); 1055 sowwakeup(so); 1056 } 1057 1058 #if defined(INET6) && !defined(TCP6) 1059 void 1060 tcp6_notify(in6p, error) 1061 struct in6pcb *in6p; 1062 int error; 1063 { 1064 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1065 struct socket *so = in6p->in6p_socket; 1066 1067 /* 1068 * Ignore some errors if we are hooked up. 1069 * If connection hasn't completed, has retransmitted several times, 1070 * and receives a second error, give up now. This is better 1071 * than waiting a long time to establish a connection that 1072 * can never complete. 1073 */ 1074 if (tp->t_state == TCPS_ESTABLISHED && 1075 (error == EHOSTUNREACH || error == ENETUNREACH || 1076 error == EHOSTDOWN)) { 1077 return; 1078 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1079 tp->t_rxtshift > 3 && tp->t_softerror) 1080 so->so_error = error; 1081 else 1082 tp->t_softerror = error; 1083 wakeup((caddr_t) &so->so_timeo); 1084 sorwakeup(so); 1085 sowwakeup(so); 1086 } 1087 #endif 1088 1089 #if defined(INET6) && !defined(TCP6) 1090 void 1091 tcp6_ctlinput(cmd, sa, d) 1092 int cmd; 1093 struct sockaddr *sa; 1094 void *d; 1095 { 1096 struct tcphdr *thp; 1097 struct tcphdr th; 1098 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1099 int nmatch; 1100 struct sockaddr_in6 sa6; 1101 struct ip6_hdr *ip6; 1102 struct mbuf *m; 1103 int off; 1104 1105 if (sa->sa_family != AF_INET6 || 1106 sa->sa_len != sizeof(struct sockaddr_in6)) 1107 return; 1108 if ((unsigned)cmd >= PRC_NCMDS) 1109 return; 1110 else if (cmd == PRC_QUENCH) { 1111 /* XXX there's no PRC_QUENCH in IPv6 */ 1112 notify = tcp6_quench; 1113 } else if (PRC_IS_REDIRECT(cmd)) 1114 notify = in6_rtchange, d = NULL; 1115 else if (cmd == PRC_MSGSIZE) 1116 notify = tcp6_mtudisc, d = NULL; 1117 else if (cmd == PRC_HOSTDEAD) 1118 d = NULL; 1119 else if (inet6ctlerrmap[cmd] == 0) 1120 return; 1121 1122 /* if the parameter is from icmp6, decode it. */ 1123 if (d != NULL) { 1124 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1125 m = ip6cp->ip6c_m; 1126 ip6 = ip6cp->ip6c_ip6; 1127 off = ip6cp->ip6c_off; 1128 } else { 1129 m = NULL; 1130 ip6 = NULL; 1131 } 1132 1133 /* translate addresses into internal form */ 1134 sa6 = *(struct sockaddr_in6 *)sa; 1135 if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) && m && m->m_pkthdr.rcvif) 1136 sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index); 1137 1138 if (ip6) { 1139 /* 1140 * XXX: We assume that when ip6 is non NULL, 1141 * M and OFF are valid. 1142 */ 1143 struct in6_addr s; 1144 1145 /* translate addresses into internal form */ 1146 memcpy(&s, &ip6->ip6_src, sizeof(s)); 1147 if (IN6_IS_ADDR_LINKLOCAL(&s)) 1148 s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index); 1149 1150 /* check if we can safely examine src and dst ports */ 1151 if (m->m_pkthdr.len < off + sizeof(th)) 1152 return; 1153 1154 if (m->m_len < off + sizeof(th)) { 1155 /* 1156 * this should be rare case, 1157 * so we compromise on this copy... 1158 */ 1159 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1160 thp = &th; 1161 } else 1162 thp = (struct tcphdr *)(mtod(m, caddr_t) + off); 1163 nmatch = in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 1164 thp->th_dport, &s, thp->th_sport, cmd, notify); 1165 if (nmatch == 0 && syn_cache_count && 1166 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1167 inet6ctlerrmap[cmd] == ENETUNREACH || 1168 inet6ctlerrmap[cmd] == EHOSTDOWN)) { 1169 struct sockaddr_in6 sin6; 1170 bzero(&sin6, sizeof(sin6)); 1171 sin6.sin6_len = sizeof(sin6); 1172 sin6.sin6_family = AF_INET6; 1173 sin6.sin6_port = thp->th_sport; 1174 sin6.sin6_addr = s; 1175 syn_cache_unreach((struct sockaddr *)&sin6, sa, thp); 1176 } 1177 } else { 1178 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 0, 1179 &zeroin6_addr, 0, cmd, notify); 1180 } 1181 } 1182 #endif 1183 1184 #ifdef INET 1185 /* assumes that ip header and tcp header are contiguous on mbuf */ 1186 void * 1187 tcp_ctlinput(cmd, sa, v) 1188 int cmd; 1189 struct sockaddr *sa; 1190 void *v; 1191 { 1192 struct ip *ip = v; 1193 struct tcphdr *th; 1194 struct icmp *icp; 1195 extern int inetctlerrmap[]; 1196 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1197 int errno; 1198 int nmatch; 1199 1200 if (sa->sa_family != AF_INET || 1201 sa->sa_len != sizeof(struct sockaddr_in)) 1202 return NULL; 1203 if ((unsigned)cmd >= PRC_NCMDS) 1204 return NULL; 1205 errno = inetctlerrmap[cmd]; 1206 if (cmd == PRC_QUENCH) 1207 notify = tcp_quench; 1208 else if (PRC_IS_REDIRECT(cmd)) 1209 notify = in_rtchange, ip = 0; 1210 else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) { 1211 /* 1212 * Check to see if we have a valid TCP connection 1213 * corresponding to the address in the ICMP message 1214 * payload. 1215 */ 1216 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1217 if (in_pcblookup_connect(&tcbtable, 1218 ip->ip_dst, th->th_dport, 1219 ip->ip_src, th->th_sport) == NULL) 1220 return NULL; 1221 1222 /* 1223 * Now that we've validated that we are actually communicating 1224 * with the host indicated in the ICMP message, locate the 1225 * ICMP header, recalculate the new MTU, and create the 1226 * corresponding routing entry. 1227 */ 1228 icp = (struct icmp *)((caddr_t)ip - 1229 offsetof(struct icmp, icmp_ip)); 1230 icmp_mtudisc(icp, ip->ip_dst); 1231 1232 return NULL; 1233 } else if (cmd == PRC_HOSTDEAD) 1234 ip = 0; 1235 else if (errno == 0) 1236 return NULL; 1237 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1238 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1239 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1240 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1241 if (nmatch == 0 && syn_cache_count && 1242 (inetctlerrmap[cmd] == EHOSTUNREACH || 1243 inetctlerrmap[cmd] == ENETUNREACH || 1244 inetctlerrmap[cmd] == EHOSTDOWN)) { 1245 struct sockaddr_in sin; 1246 bzero(&sin, sizeof(sin)); 1247 sin.sin_len = sizeof(sin); 1248 sin.sin_family = AF_INET; 1249 sin.sin_port = th->th_sport; 1250 sin.sin_addr = ip->ip_src; 1251 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1252 } 1253 1254 /* XXX mapped address case */ 1255 } else 1256 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1257 notify); 1258 return NULL; 1259 } 1260 1261 /* 1262 * When a source quence is received, we are being notifed of congestion. 1263 * Close the congestion window down to the Loss Window (one segment). 1264 * We will gradually open it again as we proceed. 1265 */ 1266 void 1267 tcp_quench(inp, errno) 1268 struct inpcb *inp; 1269 int errno; 1270 { 1271 struct tcpcb *tp = intotcpcb(inp); 1272 1273 if (tp) 1274 tp->snd_cwnd = tp->t_segsz; 1275 } 1276 #endif 1277 1278 #if defined(INET6) && !defined(TCP6) 1279 void 1280 tcp6_quench(in6p, errno) 1281 struct in6pcb *in6p; 1282 int errno; 1283 { 1284 struct tcpcb *tp = in6totcpcb(in6p); 1285 1286 if (tp) 1287 tp->snd_cwnd = tp->t_segsz; 1288 } 1289 #endif 1290 1291 /* 1292 * Path MTU Discovery handlers. 1293 */ 1294 void 1295 tcp_mtudisc_callback(faddr) 1296 struct in_addr faddr; 1297 { 1298 1299 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1300 } 1301 1302 /* 1303 * On receipt of path MTU corrections, flush old route and replace it 1304 * with the new one. Retransmit all unacknowledged packets, to ensure 1305 * that all packets will be received. 1306 */ 1307 void 1308 tcp_mtudisc(inp, errno) 1309 struct inpcb *inp; 1310 int errno; 1311 { 1312 struct tcpcb *tp = intotcpcb(inp); 1313 struct rtentry *rt = in_pcbrtentry(inp); 1314 1315 if (tp != 0) { 1316 if (rt != 0) { 1317 /* 1318 * If this was not a host route, remove and realloc. 1319 */ 1320 if ((rt->rt_flags & RTF_HOST) == 0) { 1321 in_rtchange(inp, errno); 1322 if ((rt = in_pcbrtentry(inp)) == 0) 1323 return; 1324 } 1325 1326 /* 1327 * Slow start out of the error condition. We 1328 * use the MTU because we know it's smaller 1329 * than the previously transmitted segment. 1330 * 1331 * Note: This is more conservative than the 1332 * suggestion in draft-floyd-incr-init-win-03. 1333 */ 1334 if (rt->rt_rmx.rmx_mtu != 0) 1335 tp->snd_cwnd = 1336 TCP_INITIAL_WINDOW(tcp_init_win, 1337 rt->rt_rmx.rmx_mtu); 1338 } 1339 1340 /* 1341 * Resend unacknowledged packets. 1342 */ 1343 tp->snd_nxt = tp->snd_una; 1344 tcp_output(tp); 1345 } 1346 } 1347 1348 #if defined(INET6) && !defined(TCP6) 1349 void 1350 tcp6_mtudisc(in6p, errno) 1351 struct in6pcb *in6p; 1352 int errno; 1353 { 1354 struct tcpcb *tp = in6totcpcb(in6p); 1355 struct rtentry *rt = in6_pcbrtentry(in6p); 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 in6_rtchange(in6p, errno); 1364 if ((rt = in6_pcbrtentry(in6p)) == 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 /* INET6 && !TCP6 */ 1390 1391 /* 1392 * Compute the MSS to advertise to the peer. Called only during 1393 * the 3-way handshake. If we are the server (peer initiated 1394 * connection), we are called with a pointer to the interface 1395 * on which the SYN packet arrived. If we are the client (we 1396 * initiated connection), we are called with a pointer to the 1397 * interface out which this connection should go. 1398 * 1399 * NOTE: Do not subtract IP option/extension header size nor IPsec 1400 * header size from MSS advertisement. MSS option must hold the maximum 1401 * segment size we can accept, so it must always be: 1402 * max(if mtu) - ip header - tcp header 1403 */ 1404 u_long 1405 tcp_mss_to_advertise(ifp, af) 1406 const struct ifnet *ifp; 1407 int af; 1408 { 1409 extern u_long in_maxmtu; 1410 u_long mss = 0; 1411 u_long hdrsiz; 1412 1413 /* 1414 * In order to avoid defeating path MTU discovery on the peer, 1415 * we advertise the max MTU of all attached networks as our MSS, 1416 * per RFC 1191, section 3.1. 1417 * 1418 * We provide the option to advertise just the MTU of 1419 * the interface on which we hope this connection will 1420 * be receiving. If we are responding to a SYN, we 1421 * will have a pretty good idea about this, but when 1422 * initiating a connection there is a bit more doubt. 1423 * 1424 * We also need to ensure that loopback has a large enough 1425 * MSS, as the loopback MTU is never included in in_maxmtu. 1426 */ 1427 1428 if (ifp != NULL) 1429 mss = ifp->if_mtu; 1430 1431 if (tcp_mss_ifmtu == 0) 1432 mss = max(in_maxmtu, mss); 1433 1434 switch (af) { 1435 case AF_INET: 1436 hdrsiz = sizeof(struct ip); 1437 break; 1438 #ifdef INET6 1439 case AF_INET6: 1440 hdrsiz = sizeof(struct ip6_hdr); 1441 break; 1442 #endif 1443 default: 1444 hdrsiz = 0; 1445 break; 1446 } 1447 hdrsiz += sizeof(struct tcphdr); 1448 if (mss > hdrsiz) 1449 mss -= hdrsiz; 1450 1451 mss = max(tcp_mssdflt, mss); 1452 return (mss); 1453 } 1454 1455 /* 1456 * Set connection variables based on the peer's advertised MSS. 1457 * We are passed the TCPCB for the actual connection. If we 1458 * are the server, we are called by the compressed state engine 1459 * when the 3-way handshake is complete. If we are the client, 1460 * we are called when we recieve the SYN,ACK from the server. 1461 * 1462 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1463 * before this routine is called! 1464 */ 1465 void 1466 tcp_mss_from_peer(tp, offer) 1467 struct tcpcb *tp; 1468 int offer; 1469 { 1470 struct socket *so; 1471 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1472 struct rtentry *rt; 1473 #endif 1474 u_long bufsize; 1475 int mss; 1476 1477 #ifdef DIAGNOSTIC 1478 if (tp->t_inpcb && tp->t_in6pcb) 1479 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1480 #endif 1481 so = NULL; 1482 rt = NULL; 1483 #ifdef INET 1484 if (tp->t_inpcb) { 1485 so = tp->t_inpcb->inp_socket; 1486 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1487 rt = in_pcbrtentry(tp->t_inpcb); 1488 #endif 1489 } 1490 #endif 1491 #ifdef INET6 1492 if (tp->t_in6pcb) { 1493 so = tp->t_in6pcb->in6p_socket; 1494 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1495 #ifdef TCP6 1496 rt = NULL; 1497 #else 1498 rt = in6_pcbrtentry(tp->t_in6pcb); 1499 #endif 1500 #endif 1501 } 1502 #endif 1503 1504 /* 1505 * As per RFC1122, use the default MSS value, unless they 1506 * sent us an offer. Do not accept offers less than 32 bytes. 1507 */ 1508 mss = tcp_mssdflt; 1509 if (offer) 1510 mss = offer; 1511 mss = max(mss, 32); /* sanity */ 1512 tp->t_peermss = mss; 1513 mss -= tcp_optlen(tp); 1514 #ifdef INET 1515 if (tp->t_inpcb) 1516 mss -= ip_optlen(tp->t_inpcb); 1517 #endif 1518 #ifdef INET6 1519 if (tp->t_in6pcb) 1520 mss -= ip6_optlen(tp->t_in6pcb); 1521 #endif 1522 1523 /* 1524 * If there's a pipesize, change the socket buffer to that size. 1525 * Make the socket buffer an integral number of MSS units. If 1526 * the MSS is larger than the socket buffer, artificially decrease 1527 * the MSS. 1528 */ 1529 #ifdef RTV_SPIPE 1530 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1531 bufsize = rt->rt_rmx.rmx_sendpipe; 1532 else 1533 #endif 1534 bufsize = so->so_snd.sb_hiwat; 1535 if (bufsize < mss) 1536 mss = bufsize; 1537 else { 1538 bufsize = roundup(bufsize, mss); 1539 if (bufsize > sb_max) 1540 bufsize = sb_max; 1541 (void) sbreserve(&so->so_snd, bufsize); 1542 } 1543 tp->t_segsz = mss; 1544 1545 #ifdef RTV_SSTHRESH 1546 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1547 /* 1548 * There's some sort of gateway or interface buffer 1549 * limit on the path. Use this to set the slow 1550 * start threshold, but set the threshold to no less 1551 * than 2 * MSS. 1552 */ 1553 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1554 } 1555 #endif 1556 } 1557 1558 /* 1559 * Processing necessary when a TCP connection is established. 1560 */ 1561 void 1562 tcp_established(tp) 1563 struct tcpcb *tp; 1564 { 1565 struct socket *so; 1566 #ifdef RTV_RPIPE 1567 struct rtentry *rt; 1568 #endif 1569 u_long bufsize; 1570 1571 #ifdef DIAGNOSTIC 1572 if (tp->t_inpcb && tp->t_in6pcb) 1573 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1574 #endif 1575 so = NULL; 1576 rt = NULL; 1577 #ifdef INET 1578 if (tp->t_inpcb) { 1579 so = tp->t_inpcb->inp_socket; 1580 #if defined(RTV_RPIPE) 1581 rt = in_pcbrtentry(tp->t_inpcb); 1582 #endif 1583 } 1584 #endif 1585 #ifdef INET6 1586 if (tp->t_in6pcb) { 1587 so = tp->t_in6pcb->in6p_socket; 1588 #if defined(RTV_RPIPE) 1589 #ifdef TCP6 1590 rt = NULL; 1591 #else 1592 rt = in6_pcbrtentry(tp->t_in6pcb); 1593 #endif 1594 #endif 1595 } 1596 #endif 1597 1598 tp->t_state = TCPS_ESTABLISHED; 1599 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1600 1601 #ifdef RTV_RPIPE 1602 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1603 bufsize = rt->rt_rmx.rmx_recvpipe; 1604 else 1605 #endif 1606 bufsize = so->so_rcv.sb_hiwat; 1607 if (bufsize > tp->t_ourmss) { 1608 bufsize = roundup(bufsize, tp->t_ourmss); 1609 if (bufsize > sb_max) 1610 bufsize = sb_max; 1611 (void) sbreserve(&so->so_rcv, bufsize); 1612 } 1613 } 1614 1615 /* 1616 * Check if there's an initial rtt or rttvar. Convert from the 1617 * route-table units to scaled multiples of the slow timeout timer. 1618 * Called only during the 3-way handshake. 1619 */ 1620 void 1621 tcp_rmx_rtt(tp) 1622 struct tcpcb *tp; 1623 { 1624 #ifdef RTV_RTT 1625 struct rtentry *rt = NULL; 1626 int rtt; 1627 1628 #ifdef DIAGNOSTIC 1629 if (tp->t_inpcb && tp->t_in6pcb) 1630 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1631 #endif 1632 #ifdef INET 1633 if (tp->t_inpcb) 1634 rt = in_pcbrtentry(tp->t_inpcb); 1635 #endif 1636 #ifdef INET6 1637 if (tp->t_in6pcb) { 1638 #ifdef TCP6 1639 rt = NULL; 1640 #else 1641 rt = in6_pcbrtentry(tp->t_in6pcb); 1642 #endif 1643 } 1644 #endif 1645 if (rt == NULL) 1646 return; 1647 1648 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1649 /* 1650 * XXX The lock bit for MTU indicates that the value 1651 * is also a minimum value; this is subject to time. 1652 */ 1653 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1654 TCPT_RANGESET(tp->t_rttmin, 1655 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1656 TCPTV_MIN, TCPTV_REXMTMAX); 1657 tp->t_srtt = rtt / 1658 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1659 if (rt->rt_rmx.rmx_rttvar) { 1660 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1661 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1662 (TCP_RTTVAR_SHIFT + 2)); 1663 } else { 1664 /* Default variation is +- 1 rtt */ 1665 tp->t_rttvar = 1666 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1667 } 1668 TCPT_RANGESET(tp->t_rxtcur, 1669 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1670 tp->t_rttmin, TCPTV_REXMTMAX); 1671 } 1672 #endif 1673 } 1674 1675 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1676 1677 /* 1678 * Get a new sequence value given a tcp control block 1679 */ 1680 tcp_seq 1681 tcp_new_iss(tp, len, addin) 1682 void *tp; 1683 u_long len; 1684 tcp_seq addin; 1685 { 1686 tcp_seq tcp_iss; 1687 1688 /* 1689 * Randomize. 1690 */ 1691 #if NRND > 0 1692 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 1693 #else 1694 tcp_iss = random(); 1695 #endif 1696 1697 /* 1698 * If we were asked to add some amount to a known value, 1699 * we will take a random value obtained above, mask off the upper 1700 * bits, and add in the known value. We also add in a constant to 1701 * ensure that we are at least a certain distance from the original 1702 * value. 1703 * 1704 * This is used when an old connection is in timed wait 1705 * and we have a new one coming in, for instance. 1706 */ 1707 if (addin != 0) { 1708 #ifdef TCPISS_DEBUG 1709 printf("Random %08x, ", tcp_iss); 1710 #endif 1711 tcp_iss &= TCP_ISS_RANDOM_MASK; 1712 tcp_iss += addin + TCP_ISSINCR; 1713 #ifdef TCPISS_DEBUG 1714 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 1715 #endif 1716 } else { 1717 tcp_iss &= TCP_ISS_RANDOM_MASK; 1718 tcp_iss += tcp_iss_seq; 1719 tcp_iss_seq += TCP_ISSINCR; 1720 #ifdef TCPISS_DEBUG 1721 printf("ISS %08x\n", tcp_iss); 1722 #endif 1723 } 1724 1725 if (tcp_compat_42) { 1726 /* 1727 * Limit it to the positive range for really old TCP 1728 * implementations. 1729 */ 1730 if (tcp_iss >= 0x80000000) 1731 tcp_iss &= 0x7fffffff; /* XXX */ 1732 } 1733 1734 return tcp_iss; 1735 } 1736 1737 #ifdef IPSEC 1738 /* compute ESP/AH header size for TCP, including outer IP header. */ 1739 size_t 1740 ipsec4_hdrsiz_tcp(tp) 1741 struct tcpcb *tp; 1742 { 1743 struct inpcb *inp; 1744 size_t hdrsiz; 1745 1746 /* XXX mapped addr case (tp->t_in6pcb) */ 1747 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 1748 return 0; 1749 switch (tp->t_family) { 1750 case AF_INET: 1751 /* XXX: should use currect direction. */ 1752 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 1753 break; 1754 default: 1755 hdrsiz = 0; 1756 break; 1757 } 1758 1759 return hdrsiz; 1760 } 1761 1762 #if defined(INET6) && !defined(TCP6) 1763 size_t 1764 ipsec6_hdrsiz_tcp(tp) 1765 struct tcpcb *tp; 1766 { 1767 struct in6pcb *in6p; 1768 size_t hdrsiz; 1769 1770 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 1771 return 0; 1772 switch (tp->t_family) { 1773 case AF_INET6: 1774 /* XXX: should use currect direction. */ 1775 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 1776 break; 1777 case AF_INET: 1778 /* mapped address case - tricky */ 1779 default: 1780 hdrsiz = 0; 1781 break; 1782 } 1783 1784 return hdrsiz; 1785 } 1786 #endif 1787 #endif /*IPSEC*/ 1788 1789 /* 1790 * Determine the length of the TCP options for this connection. 1791 * 1792 * XXX: What do we do for SACK, when we add that? Just reserve 1793 * all of the space? Otherwise we can't exactly be incrementing 1794 * cwnd by an amount that varies depending on the amount we last 1795 * had to SACK! 1796 */ 1797 1798 u_int 1799 tcp_optlen(tp) 1800 struct tcpcb *tp; 1801 { 1802 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1803 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1804 return TCPOLEN_TSTAMP_APPA; 1805 else 1806 return 0; 1807 } 1808
Indexes created Mon Sep 22 05:09:51 GMT 2025