tcp_subr.c revision 1.96
1 /* $NetBSD: tcp_subr.c,v 1.96 2000/10/17 03:06:43 itojun Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998 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 176 177 #ifndef TCBHASHSIZE 178 #define TCBHASHSIZE 128 179 #endif 180 int tcbhashsize = TCBHASHSIZE; 181 182 int tcp_freeq __P((struct tcpcb *)); 183 184 struct pool tcpcb_pool; 185 186 /* 187 * Tcp initialization 188 */ 189 void 190 tcp_init() 191 { 192 int hlen; 193 194 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 195 0, NULL, NULL, M_PCB); 196 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 197 #ifdef INET6 198 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 199 #endif 200 LIST_INIT(&tcp_delacks); 201 202 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 203 #ifdef INET6 204 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 205 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 206 #endif 207 if (max_protohdr < hlen) 208 max_protohdr = hlen; 209 if (max_linkhdr + hlen > MHLEN) 210 panic("tcp_init"); 211 212 /* Initialize the compressed state engine. */ 213 syn_cache_init(); 214 } 215 216 /* 217 * Create template to be used to send tcp packets on a connection. 218 * Call after host entry created, allocates an mbuf and fills 219 * in a skeletal tcp/ip header, minimizing the amount of work 220 * necessary when the connection is used. 221 */ 222 struct mbuf * 223 tcp_template(tp) 224 struct tcpcb *tp; 225 { 226 struct inpcb *inp = tp->t_inpcb; 227 #ifdef INET6 228 struct in6pcb *in6p = tp->t_in6pcb; 229 #endif 230 struct tcphdr *n; 231 struct mbuf *m; 232 int hlen; 233 234 switch (tp->t_family) { 235 case AF_INET: 236 hlen = sizeof(struct ip); 237 if (inp) 238 break; 239 #ifdef INET6 240 if (in6p) { 241 /* mapped addr case */ 242 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 243 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 244 break; 245 } 246 #endif 247 return NULL; /*EINVAL*/ 248 #ifdef INET6 249 case AF_INET6: 250 hlen = sizeof(struct ip6_hdr); 251 if (in6p) { 252 /* more sainty check? */ 253 break; 254 } 255 return NULL; /*EINVAL*/ 256 #endif 257 default: 258 hlen = 0; /*pacify gcc*/ 259 return NULL; /*EAFNOSUPPORT*/ 260 } 261 #ifdef DIAGNOSTIC 262 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 263 panic("mclbytes too small for t_template"); 264 #endif 265 m = tp->t_template; 266 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 267 ; 268 else { 269 if (m) 270 m_freem(m); 271 m = tp->t_template = NULL; 272 MGETHDR(m, M_DONTWAIT, MT_HEADER); 273 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 274 MCLGET(m, M_DONTWAIT); 275 if ((m->m_flags & M_EXT) == 0) { 276 m_free(m); 277 m = NULL; 278 } 279 } 280 if (m == NULL) 281 return NULL; 282 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 283 } 284 bzero(mtod(m, caddr_t), m->m_len); 285 switch (tp->t_family) { 286 case AF_INET: 287 { 288 struct ipovly *ipov; 289 mtod(m, struct ip *)->ip_v = 4; 290 ipov = mtod(m, struct ipovly *); 291 ipov->ih_pr = IPPROTO_TCP; 292 ipov->ih_len = htons(sizeof(struct tcphdr)); 293 if (inp) { 294 ipov->ih_src = inp->inp_laddr; 295 ipov->ih_dst = inp->inp_faddr; 296 } 297 #ifdef INET6 298 else if (in6p) { 299 /* mapped addr case */ 300 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 301 sizeof(ipov->ih_src)); 302 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 303 sizeof(ipov->ih_dst)); 304 } 305 #endif 306 break; 307 } 308 #ifdef INET6 309 case AF_INET6: 310 { 311 struct ip6_hdr *ip6; 312 mtod(m, struct ip *)->ip_v = 6; 313 ip6 = mtod(m, struct ip6_hdr *); 314 ip6->ip6_nxt = IPPROTO_TCP; 315 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 316 ip6->ip6_src = in6p->in6p_laddr; 317 ip6->ip6_dst = in6p->in6p_faddr; 318 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 319 if (ip6_auto_flowlabel) { 320 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 321 ip6->ip6_flow |= 322 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 323 } 324 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 325 ip6->ip6_vfc |= IPV6_VERSION; 326 break; 327 } 328 #endif 329 } 330 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 331 if (inp) { 332 n->th_sport = inp->inp_lport; 333 n->th_dport = inp->inp_fport; 334 } 335 #ifdef INET6 336 else if (in6p) { 337 n->th_sport = in6p->in6p_lport; 338 n->th_dport = in6p->in6p_fport; 339 } 340 #endif 341 n->th_seq = 0; 342 n->th_ack = 0; 343 n->th_x2 = 0; 344 n->th_off = 5; 345 n->th_flags = 0; 346 n->th_win = 0; 347 n->th_sum = 0; 348 n->th_urp = 0; 349 return (m); 350 } 351 352 /* 353 * Send a single message to the TCP at address specified by 354 * the given TCP/IP header. If m == 0, then we make a copy 355 * of the tcpiphdr at ti and send directly to the addressed host. 356 * This is used to force keep alive messages out using the TCP 357 * template for a connection tp->t_template. If flags are given 358 * then we send a message back to the TCP which originated the 359 * segment ti, and discard the mbuf containing it and any other 360 * attached mbufs. 361 * 362 * In any case the ack and sequence number of the transmitted 363 * segment are as specified by the parameters. 364 */ 365 int 366 tcp_respond(tp, template, m, th0, ack, seq, flags) 367 struct tcpcb *tp; 368 struct mbuf *template; 369 struct mbuf *m; 370 struct tcphdr *th0; 371 tcp_seq ack, seq; 372 int flags; 373 { 374 struct route *ro; 375 int error, tlen, win = 0; 376 int hlen; 377 struct ip *ip; 378 #ifdef INET6 379 struct ip6_hdr *ip6; 380 #endif 381 int family; /* family on packet, not inpcb/in6pcb! */ 382 struct tcphdr *th; 383 384 if (tp != NULL && (flags & TH_RST) == 0) { 385 #ifdef DIAGNOSTIC 386 if (tp->t_inpcb && tp->t_in6pcb) 387 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 388 #endif 389 #ifdef INET 390 if (tp->t_inpcb) 391 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 392 #endif 393 #ifdef INET6 394 if (tp->t_in6pcb) 395 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 396 #endif 397 } 398 399 ip = NULL; 400 #ifdef INET6 401 ip6 = NULL; 402 #endif 403 if (m == 0) { 404 if (!template) 405 return EINVAL; 406 407 /* get family information from template */ 408 switch (mtod(template, struct ip *)->ip_v) { 409 case 4: 410 family = AF_INET; 411 hlen = sizeof(struct ip); 412 break; 413 #ifdef INET6 414 case 6: 415 family = AF_INET6; 416 hlen = sizeof(struct ip6_hdr); 417 break; 418 #endif 419 default: 420 return EAFNOSUPPORT; 421 } 422 423 MGETHDR(m, M_DONTWAIT, MT_HEADER); 424 if (m) { 425 MCLGET(m, M_DONTWAIT); 426 if ((m->m_flags & M_EXT) == 0) { 427 m_free(m); 428 m = NULL; 429 } 430 } 431 if (m == NULL) 432 return (ENOBUFS); 433 434 if (tcp_compat_42) 435 tlen = 1; 436 else 437 tlen = 0; 438 439 m->m_data += max_linkhdr; 440 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 441 template->m_len); 442 switch (family) { 443 case AF_INET: 444 ip = mtod(m, struct ip *); 445 th = (struct tcphdr *)(ip + 1); 446 break; 447 #ifdef INET6 448 case AF_INET6: 449 ip6 = mtod(m, struct ip6_hdr *); 450 th = (struct tcphdr *)(ip6 + 1); 451 break; 452 #endif 453 #if 0 454 default: 455 /* noone will visit here */ 456 m_freem(m); 457 return EAFNOSUPPORT; 458 #endif 459 } 460 flags = TH_ACK; 461 } else { 462 463 if ((m->m_flags & M_PKTHDR) == 0) { 464 #if 0 465 printf("non PKTHDR to tcp_respond\n"); 466 #endif 467 m_freem(m); 468 return EINVAL; 469 } 470 #ifdef DIAGNOSTIC 471 if (!th0) 472 panic("th0 == NULL in tcp_respond"); 473 #endif 474 475 /* get family information from m */ 476 switch (mtod(m, struct ip *)->ip_v) { 477 case 4: 478 family = AF_INET; 479 hlen = sizeof(struct ip); 480 ip = mtod(m, struct ip *); 481 break; 482 #ifdef INET6 483 case 6: 484 family = AF_INET6; 485 hlen = sizeof(struct ip6_hdr); 486 ip6 = mtod(m, struct ip6_hdr *); 487 break; 488 #endif 489 default: 490 m_freem(m); 491 return EAFNOSUPPORT; 492 } 493 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 494 tlen = sizeof(*th0); 495 else 496 tlen = th0->th_off << 2; 497 498 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 499 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 500 m->m_len = hlen + tlen; 501 m_freem(m->m_next); 502 m->m_next = NULL; 503 } else { 504 struct mbuf *n; 505 506 #ifdef DIAGNOSTIC 507 if (max_linkhdr + hlen + tlen > MCLBYTES) { 508 m_freem(m); 509 return EMSGSIZE; 510 } 511 #endif 512 MGETHDR(n, M_DONTWAIT, MT_HEADER); 513 if (n && max_linkhdr + hlen + tlen > MHLEN) { 514 MCLGET(n, M_DONTWAIT); 515 if ((n->m_flags & M_EXT) == 0) { 516 m_freem(n); 517 n = NULL; 518 } 519 } 520 if (!n) { 521 m_freem(m); 522 return ENOBUFS; 523 } 524 525 n->m_data += max_linkhdr; 526 n->m_len = hlen + tlen; 527 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 528 m_copyback(n, hlen, tlen, (caddr_t)th0); 529 530 m_freem(m); 531 m = n; 532 n = NULL; 533 } 534 535 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 536 switch (family) { 537 case AF_INET: 538 ip = mtod(m, struct ip *); 539 th = (struct tcphdr *)(ip + 1); 540 ip->ip_p = IPPROTO_TCP; 541 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 542 ip->ip_p = IPPROTO_TCP; 543 break; 544 #ifdef INET6 545 case AF_INET6: 546 ip6 = mtod(m, struct ip6_hdr *); 547 th = (struct tcphdr *)(ip6 + 1); 548 ip6->ip6_nxt = IPPROTO_TCP; 549 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 550 ip6->ip6_nxt = IPPROTO_TCP; 551 break; 552 #endif 553 #if 0 554 default: 555 /* noone will visit here */ 556 m_freem(m); 557 return EAFNOSUPPORT; 558 #endif 559 } 560 xchg(th->th_dport, th->th_sport, u_int16_t); 561 #undef xchg 562 tlen = 0; /*be friendly with the following code*/ 563 } 564 th->th_seq = htonl(seq); 565 th->th_ack = htonl(ack); 566 th->th_x2 = 0; 567 if ((flags & TH_SYN) == 0) { 568 if (tp) 569 win >>= tp->rcv_scale; 570 if (win > TCP_MAXWIN) 571 win = TCP_MAXWIN; 572 th->th_win = htons((u_int16_t)win); 573 th->th_off = sizeof (struct tcphdr) >> 2; 574 tlen += sizeof(*th); 575 } else 576 tlen += th->th_off << 2; 577 m->m_len = hlen + tlen; 578 m->m_pkthdr.len = hlen + tlen; 579 m->m_pkthdr.rcvif = (struct ifnet *) 0; 580 th->th_flags = flags; 581 th->th_urp = 0; 582 583 switch (family) { 584 #ifdef INET 585 case AF_INET: 586 { 587 struct ipovly *ipov = (struct ipovly *)ip; 588 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 589 ipov->ih_len = htons((u_int16_t)tlen); 590 591 th->th_sum = 0; 592 th->th_sum = in_cksum(m, hlen + tlen); 593 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 594 ip->ip_ttl = ip_defttl; 595 break; 596 } 597 #endif 598 #ifdef INET6 599 case AF_INET6: 600 { 601 th->th_sum = 0; 602 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 603 tlen); 604 ip6->ip6_plen = ntohs(tlen); 605 if (tp && tp->t_in6pcb) { 606 struct ifnet *oifp; 607 ro = (struct route *)&tp->t_in6pcb->in6p_route; 608 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 609 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 610 } else 611 ip6->ip6_hlim = ip6_defhlim; 612 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 613 if (ip6_auto_flowlabel) { 614 ip6->ip6_flow |= 615 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 616 } 617 break; 618 } 619 #endif 620 } 621 622 #ifdef IPSEC 623 ipsec_setsocket(m, NULL); 624 #endif /*IPSEC*/ 625 626 if (tp != NULL && tp->t_inpcb != NULL) { 627 ro = &tp->t_inpcb->inp_route; 628 #ifdef IPSEC 629 ipsec_setsocket(m, tp->t_inpcb->inp_socket); 630 #endif 631 #ifdef DIAGNOSTIC 632 if (family != AF_INET) 633 panic("tcp_respond: address family mismatch"); 634 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 635 panic("tcp_respond: ip_dst %x != inp_faddr %x", 636 ntohl(ip->ip_dst.s_addr), 637 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 638 } 639 #endif 640 } 641 #ifdef INET6 642 else if (tp != NULL && tp->t_in6pcb != NULL) { 643 ro = (struct route *)&tp->t_in6pcb->in6p_route; 644 #ifdef IPSEC 645 ipsec_setsocket(m, tp->t_in6pcb->in6p_socket); 646 #endif 647 #ifdef DIAGNOSTIC 648 if (family == AF_INET) { 649 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 650 panic("tcp_respond: not mapped addr"); 651 if (bcmp(&ip->ip_dst, 652 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 653 sizeof(ip->ip_dst)) != 0) { 654 panic("tcp_respond: ip_dst != in6p_faddr"); 655 } 656 } else if (family == AF_INET6) { 657 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 658 panic("tcp_respond: ip6_dst != in6p_faddr"); 659 } else 660 panic("tcp_respond: address family mismatch"); 661 #endif 662 } 663 #endif 664 else 665 ro = NULL; 666 667 switch (family) { 668 #ifdef INET 669 case AF_INET: 670 error = ip_output(m, NULL, ro, 671 (ip_mtudisc ? IP_MTUDISC : 0), 672 NULL); 673 break; 674 #endif 675 #ifdef INET6 676 case AF_INET6: 677 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 678 NULL); 679 break; 680 #endif 681 default: 682 error = EAFNOSUPPORT; 683 break; 684 } 685 686 return (error); 687 } 688 689 /* 690 * Create a new TCP control block, making an 691 * empty reassembly queue and hooking it to the argument 692 * protocol control block. 693 */ 694 struct tcpcb * 695 tcp_newtcpcb(family, aux) 696 int family; /* selects inpcb, or in6pcb */ 697 void *aux; 698 { 699 struct tcpcb *tp; 700 701 switch (family) { 702 case PF_INET: 703 break; 704 #ifdef INET6 705 case PF_INET6: 706 break; 707 #endif 708 default: 709 return NULL; 710 } 711 712 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 713 if (tp == NULL) 714 return (NULL); 715 bzero((caddr_t)tp, sizeof(struct tcpcb)); 716 LIST_INIT(&tp->segq); 717 LIST_INIT(&tp->timeq); 718 tp->t_family = family; /* may be overridden later on */ 719 tp->t_peermss = tcp_mssdflt; 720 tp->t_ourmss = tcp_mssdflt; 721 tp->t_segsz = tcp_mssdflt; 722 LIST_INIT(&tp->t_sc); 723 724 tp->t_flags = 0; 725 if (tcp_do_rfc1323 && tcp_do_win_scale) 726 tp->t_flags |= TF_REQ_SCALE; 727 if (tcp_do_rfc1323 && tcp_do_timestamps) 728 tp->t_flags |= TF_REQ_TSTMP; 729 if (tcp_do_sack == 2) 730 tp->t_flags |= TF_WILL_SACK; 731 else if (tcp_do_sack == 1) 732 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 733 tp->t_flags |= TF_CANT_TXSACK; 734 switch (family) { 735 case PF_INET: 736 tp->t_inpcb = (struct inpcb *)aux; 737 break; 738 #ifdef INET6 739 case PF_INET6: 740 tp->t_in6pcb = (struct in6pcb *)aux; 741 break; 742 #endif 743 } 744 /* 745 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 746 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 747 * reasonable initial retransmit time. 748 */ 749 tp->t_srtt = TCPTV_SRTTBASE; 750 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 751 tp->t_rttmin = TCPTV_MIN; 752 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 753 TCPTV_MIN, TCPTV_REXMTMAX); 754 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 755 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 756 if (family == AF_INET) { 757 struct inpcb *inp = (struct inpcb *)aux; 758 inp->inp_ip.ip_ttl = ip_defttl; 759 inp->inp_ppcb = (caddr_t)tp; 760 } 761 #ifdef INET6 762 else if (family == AF_INET6) { 763 struct in6pcb *in6p = (struct in6pcb *)aux; 764 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 765 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 766 : NULL); 767 in6p->in6p_ppcb = (caddr_t)tp; 768 } 769 #endif 770 return (tp); 771 } 772 773 /* 774 * Drop a TCP connection, reporting 775 * the specified error. If connection is synchronized, 776 * then send a RST to peer. 777 */ 778 struct tcpcb * 779 tcp_drop(tp, errno) 780 struct tcpcb *tp; 781 int errno; 782 { 783 struct socket *so; 784 785 #ifdef DIAGNOSTIC 786 if (tp->t_inpcb && tp->t_in6pcb) 787 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 788 #endif 789 #ifdef INET 790 if (tp->t_inpcb) 791 so = tp->t_inpcb->inp_socket; 792 #endif 793 #ifdef INET6 794 if (tp->t_in6pcb) 795 so = tp->t_in6pcb->in6p_socket; 796 #endif 797 else 798 return NULL; 799 800 if (TCPS_HAVERCVDSYN(tp->t_state)) { 801 tp->t_state = TCPS_CLOSED; 802 (void) tcp_output(tp); 803 tcpstat.tcps_drops++; 804 } else 805 tcpstat.tcps_conndrops++; 806 if (errno == ETIMEDOUT && tp->t_softerror) 807 errno = tp->t_softerror; 808 so->so_error = errno; 809 return (tcp_close(tp)); 810 } 811 812 /* 813 * Close a TCP control block: 814 * discard all space held by the tcp 815 * discard internet protocol block 816 * wake up any sleepers 817 */ 818 struct tcpcb * 819 tcp_close(tp) 820 struct tcpcb *tp; 821 { 822 struct inpcb *inp; 823 #ifdef INET6 824 struct in6pcb *in6p; 825 #endif 826 struct socket *so; 827 #ifdef RTV_RTT 828 struct rtentry *rt; 829 #endif 830 struct route *ro; 831 832 inp = tp->t_inpcb; 833 #ifdef INET6 834 in6p = tp->t_in6pcb; 835 #endif 836 so = NULL; 837 ro = NULL; 838 if (inp) { 839 so = inp->inp_socket; 840 ro = &inp->inp_route; 841 } 842 #ifdef INET6 843 else if (in6p) { 844 so = in6p->in6p_socket; 845 ro = (struct route *)&in6p->in6p_route; 846 } 847 #endif 848 849 #ifdef RTV_RTT 850 /* 851 * If we sent enough data to get some meaningful characteristics, 852 * save them in the routing entry. 'Enough' is arbitrarily 853 * defined as the sendpipesize (default 4K) * 16. This would 854 * give us 16 rtt samples assuming we only get one sample per 855 * window (the usual case on a long haul net). 16 samples is 856 * enough for the srtt filter to converge to within 5% of the correct 857 * value; fewer samples and we could save a very bogus rtt. 858 * 859 * Don't update the default route's characteristics and don't 860 * update anything that the user "locked". 861 */ 862 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 863 ro && (rt = ro->ro_rt) && 864 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 865 u_long i = 0; 866 867 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 868 i = tp->t_srtt * 869 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 870 if (rt->rt_rmx.rmx_rtt && i) 871 /* 872 * filter this update to half the old & half 873 * the new values, converting scale. 874 * See route.h and tcp_var.h for a 875 * description of the scaling constants. 876 */ 877 rt->rt_rmx.rmx_rtt = 878 (rt->rt_rmx.rmx_rtt + i) / 2; 879 else 880 rt->rt_rmx.rmx_rtt = i; 881 } 882 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 883 i = tp->t_rttvar * 884 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 885 if (rt->rt_rmx.rmx_rttvar && i) 886 rt->rt_rmx.rmx_rttvar = 887 (rt->rt_rmx.rmx_rttvar + i) / 2; 888 else 889 rt->rt_rmx.rmx_rttvar = i; 890 } 891 /* 892 * update the pipelimit (ssthresh) if it has been updated 893 * already or if a pipesize was specified & the threshhold 894 * got below half the pipesize. I.e., wait for bad news 895 * before we start updating, then update on both good 896 * and bad news. 897 */ 898 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 899 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 900 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 901 /* 902 * convert the limit from user data bytes to 903 * packets then to packet data bytes. 904 */ 905 i = (i + tp->t_segsz / 2) / tp->t_segsz; 906 if (i < 2) 907 i = 2; 908 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 909 if (rt->rt_rmx.rmx_ssthresh) 910 rt->rt_rmx.rmx_ssthresh = 911 (rt->rt_rmx.rmx_ssthresh + i) / 2; 912 else 913 rt->rt_rmx.rmx_ssthresh = i; 914 } 915 } 916 #endif /* RTV_RTT */ 917 /* free the reassembly queue, if any */ 918 TCP_REASS_LOCK(tp); 919 (void) tcp_freeq(tp); 920 TCP_REASS_UNLOCK(tp); 921 922 TCP_CLEAR_DELACK(tp); 923 syn_cache_cleanup(tp); 924 925 if (tp->t_template) { 926 m_free(tp->t_template); 927 tp->t_template = NULL; 928 } 929 pool_put(&tcpcb_pool, tp); 930 if (inp) { 931 inp->inp_ppcb = 0; 932 soisdisconnected(so); 933 in_pcbdetach(inp); 934 } 935 #ifdef INET6 936 else if (in6p) { 937 in6p->in6p_ppcb = 0; 938 soisdisconnected(so); 939 in6_pcbdetach(in6p); 940 } 941 #endif 942 tcpstat.tcps_closed++; 943 return ((struct tcpcb *)0); 944 } 945 946 int 947 tcp_freeq(tp) 948 struct tcpcb *tp; 949 { 950 struct ipqent *qe; 951 int rv = 0; 952 #ifdef TCPREASS_DEBUG 953 int i = 0; 954 #endif 955 956 TCP_REASS_LOCK_CHECK(tp); 957 958 while ((qe = tp->segq.lh_first) != NULL) { 959 #ifdef TCPREASS_DEBUG 960 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 961 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 962 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 963 #endif 964 LIST_REMOVE(qe, ipqe_q); 965 LIST_REMOVE(qe, ipqe_timeq); 966 m_freem(qe->ipqe_m); 967 pool_put(&ipqent_pool, qe); 968 rv = 1; 969 } 970 return (rv); 971 } 972 973 /* 974 * Protocol drain routine. Called when memory is in short supply. 975 */ 976 void 977 tcp_drain() 978 { 979 struct inpcb *inp; 980 struct tcpcb *tp; 981 982 /* 983 * Free the sequence queue of all TCP connections. 984 */ 985 inp = tcbtable.inpt_queue.cqh_first; 986 if (inp) /* XXX */ 987 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 988 inp = inp->inp_queue.cqe_next) { 989 if ((tp = intotcpcb(inp)) != NULL) { 990 /* 991 * We may be called from a device's interrupt 992 * context. If the tcpcb is already busy, 993 * just bail out now. 994 */ 995 if (tcp_reass_lock_try(tp) == 0) 996 continue; 997 if (tcp_freeq(tp)) 998 tcpstat.tcps_connsdrained++; 999 TCP_REASS_UNLOCK(tp); 1000 } 1001 } 1002 } 1003 1004 /* 1005 * Notify a tcp user of an asynchronous error; 1006 * store error as soft error, but wake up user 1007 * (for now, won't do anything until can select for soft error). 1008 */ 1009 void 1010 tcp_notify(inp, error) 1011 struct inpcb *inp; 1012 int error; 1013 { 1014 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1015 struct socket *so = inp->inp_socket; 1016 1017 /* 1018 * Ignore some errors if we are hooked up. 1019 * If connection hasn't completed, has retransmitted several times, 1020 * and receives a second error, give up now. This is better 1021 * than waiting a long time to establish a connection that 1022 * can never complete. 1023 */ 1024 if (tp->t_state == TCPS_ESTABLISHED && 1025 (error == EHOSTUNREACH || error == ENETUNREACH || 1026 error == EHOSTDOWN)) { 1027 return; 1028 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1029 tp->t_rxtshift > 3 && tp->t_softerror) 1030 so->so_error = error; 1031 else 1032 tp->t_softerror = error; 1033 wakeup((caddr_t) &so->so_timeo); 1034 sorwakeup(so); 1035 sowwakeup(so); 1036 } 1037 1038 #if defined(INET6) && !defined(TCP6) 1039 void 1040 tcp6_notify(in6p, error) 1041 struct in6pcb *in6p; 1042 int error; 1043 { 1044 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1045 struct socket *so = in6p->in6p_socket; 1046 1047 /* 1048 * Ignore some errors if we are hooked up. 1049 * If connection hasn't completed, has retransmitted several times, 1050 * and receives a second error, give up now. This is better 1051 * than waiting a long time to establish a connection that 1052 * can never complete. 1053 */ 1054 if (tp->t_state == TCPS_ESTABLISHED && 1055 (error == EHOSTUNREACH || error == ENETUNREACH || 1056 error == EHOSTDOWN)) { 1057 return; 1058 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1059 tp->t_rxtshift > 3 && tp->t_softerror) 1060 so->so_error = error; 1061 else 1062 tp->t_softerror = error; 1063 wakeup((caddr_t) &so->so_timeo); 1064 sorwakeup(so); 1065 sowwakeup(so); 1066 } 1067 #endif 1068 1069 #if defined(INET6) && !defined(TCP6) 1070 void 1071 tcp6_ctlinput(cmd, sa, d) 1072 int cmd; 1073 struct sockaddr *sa; 1074 void *d; 1075 { 1076 struct tcphdr *thp; 1077 struct tcphdr th; 1078 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1079 int nmatch; 1080 struct sockaddr_in6 sa6; 1081 struct ip6_hdr *ip6; 1082 struct mbuf *m; 1083 int off; 1084 1085 if (sa->sa_family != AF_INET6 || 1086 sa->sa_len != sizeof(struct sockaddr_in6)) 1087 return; 1088 if ((unsigned)cmd >= PRC_NCMDS) 1089 return; 1090 else if (cmd == PRC_QUENCH) { 1091 /* XXX there's no PRC_QUENCH in IPv6 */ 1092 notify = tcp6_quench; 1093 } else if (PRC_IS_REDIRECT(cmd)) 1094 notify = in6_rtchange, d = NULL; 1095 else if (cmd == PRC_MSGSIZE) 1096 notify = tcp6_mtudisc, d = NULL; 1097 else if (cmd == PRC_HOSTDEAD) 1098 d = NULL; 1099 else if (inet6ctlerrmap[cmd] == 0) 1100 return; 1101 1102 /* if the parameter is from icmp6, decode it. */ 1103 if (d != NULL) { 1104 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1105 m = ip6cp->ip6c_m; 1106 ip6 = ip6cp->ip6c_ip6; 1107 off = ip6cp->ip6c_off; 1108 } else { 1109 m = NULL; 1110 ip6 = NULL; 1111 } 1112 1113 /* translate addresses into internal form */ 1114 sa6 = *(struct sockaddr_in6 *)sa; 1115 if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) && m && m->m_pkthdr.rcvif) 1116 sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index); 1117 1118 if (ip6) { 1119 /* 1120 * XXX: We assume that when ip6 is non NULL, 1121 * M and OFF are valid. 1122 */ 1123 struct in6_addr s; 1124 1125 /* translate addresses into internal form */ 1126 memcpy(&s, &ip6->ip6_src, sizeof(s)); 1127 if (IN6_IS_ADDR_LINKLOCAL(&s)) 1128 s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index); 1129 1130 /* check if we can safely examine src and dst ports */ 1131 if (m->m_pkthdr.len < off + sizeof(th)) 1132 return; 1133 1134 if (m->m_len < off + sizeof(th)) { 1135 /* 1136 * this should be rare case, 1137 * so we compromise on this copy... 1138 */ 1139 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1140 thp = &th; 1141 } else 1142 thp = (struct tcphdr *)(mtod(m, caddr_t) + off); 1143 nmatch = in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 1144 thp->th_dport, &s, thp->th_sport, cmd, notify); 1145 if (nmatch == 0 && syn_cache_count && 1146 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1147 inet6ctlerrmap[cmd] == ENETUNREACH || 1148 inet6ctlerrmap[cmd] == EHOSTDOWN)) { 1149 struct sockaddr_in6 sin6; 1150 bzero(&sin6, sizeof(sin6)); 1151 sin6.sin6_len = sizeof(sin6); 1152 sin6.sin6_family = AF_INET6; 1153 sin6.sin6_port = thp->th_sport; 1154 sin6.sin6_addr = s; 1155 syn_cache_unreach((struct sockaddr *)&sin6, sa, thp); 1156 } 1157 } else { 1158 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 0, 1159 &zeroin6_addr, 0, cmd, notify); 1160 } 1161 } 1162 #endif 1163 1164 #ifdef INET 1165 /* assumes that ip header and tcp header are contiguous on mbuf */ 1166 void * 1167 tcp_ctlinput(cmd, sa, v) 1168 int cmd; 1169 struct sockaddr *sa; 1170 void *v; 1171 { 1172 struct ip *ip = v; 1173 struct tcphdr *th; 1174 extern int inetctlerrmap[]; 1175 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1176 int errno; 1177 int nmatch; 1178 1179 if (sa->sa_family != AF_INET || 1180 sa->sa_len != sizeof(struct sockaddr_in)) 1181 return NULL; 1182 if ((unsigned)cmd >= PRC_NCMDS) 1183 return NULL; 1184 errno = inetctlerrmap[cmd]; 1185 if (cmd == PRC_QUENCH) 1186 notify = tcp_quench; 1187 else if (PRC_IS_REDIRECT(cmd)) 1188 notify = in_rtchange, ip = 0; 1189 else if (cmd == PRC_MSGSIZE && ip_mtudisc) 1190 notify = tcp_mtudisc, ip = 0; 1191 else if (cmd == PRC_HOSTDEAD) 1192 ip = 0; 1193 else if (errno == 0) 1194 return NULL; 1195 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1196 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1197 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1198 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1199 if (nmatch == 0 && syn_cache_count && 1200 (inetctlerrmap[cmd] == EHOSTUNREACH || 1201 inetctlerrmap[cmd] == ENETUNREACH || 1202 inetctlerrmap[cmd] == EHOSTDOWN)) { 1203 struct sockaddr_in sin; 1204 bzero(&sin, sizeof(sin)); 1205 sin.sin_len = sizeof(sin); 1206 sin.sin_family = AF_INET; 1207 sin.sin_port = th->th_sport; 1208 sin.sin_addr = ip->ip_src; 1209 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1210 } 1211 1212 /* XXX mapped address case */ 1213 } 1214 else { 1215 (void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1216 notify); 1217 } 1218 return NULL; 1219 } 1220 1221 /* 1222 * When a source quence is received, we are being notifed of congestion. 1223 * Close the congestion window down to the Loss Window (one segment). 1224 * We will gradually open it again as we proceed. 1225 */ 1226 void 1227 tcp_quench(inp, errno) 1228 struct inpcb *inp; 1229 int errno; 1230 { 1231 struct tcpcb *tp = intotcpcb(inp); 1232 1233 if (tp) 1234 tp->snd_cwnd = tp->t_segsz; 1235 } 1236 #endif 1237 1238 #if defined(INET6) && !defined(TCP6) 1239 void 1240 tcp6_quench(in6p, errno) 1241 struct in6pcb *in6p; 1242 int errno; 1243 { 1244 struct tcpcb *tp = in6totcpcb(in6p); 1245 1246 if (tp) 1247 tp->snd_cwnd = tp->t_segsz; 1248 } 1249 #endif 1250 1251 /* 1252 * On receipt of path MTU corrections, flush old route and replace it 1253 * with the new one. Retransmit all unacknowledged packets, to ensure 1254 * that all packets will be received. 1255 */ 1256 void 1257 tcp_mtudisc(inp, errno) 1258 struct inpcb *inp; 1259 int errno; 1260 { 1261 struct tcpcb *tp = intotcpcb(inp); 1262 struct rtentry *rt = in_pcbrtentry(inp); 1263 1264 if (tp != 0) { 1265 if (rt != 0) { 1266 /* 1267 * If this was not a host route, remove and realloc. 1268 */ 1269 if ((rt->rt_flags & RTF_HOST) == 0) { 1270 in_rtchange(inp, errno); 1271 if ((rt = in_pcbrtentry(inp)) == 0) 1272 return; 1273 } 1274 1275 /* 1276 * Slow start out of the error condition. We 1277 * use the MTU because we know it's smaller 1278 * than the previously transmitted segment. 1279 * 1280 * Note: This is more conservative than the 1281 * suggestion in draft-floyd-incr-init-win-03. 1282 */ 1283 if (rt->rt_rmx.rmx_mtu != 0) 1284 tp->snd_cwnd = 1285 TCP_INITIAL_WINDOW(tcp_init_win, 1286 rt->rt_rmx.rmx_mtu); 1287 } 1288 1289 /* 1290 * Resend unacknowledged packets. 1291 */ 1292 tp->snd_nxt = tp->snd_una; 1293 tcp_output(tp); 1294 } 1295 } 1296 1297 #if defined(INET6) && !defined(TCP6) 1298 void 1299 tcp6_mtudisc(in6p, errno) 1300 struct in6pcb *in6p; 1301 int errno; 1302 { 1303 struct tcpcb *tp = in6totcpcb(in6p); 1304 struct rtentry *rt = in6_pcbrtentry(in6p); 1305 1306 if (tp != 0) { 1307 if (rt != 0) { 1308 /* 1309 * If this was not a host route, remove and realloc. 1310 */ 1311 if ((rt->rt_flags & RTF_HOST) == 0) { 1312 in6_rtchange(in6p, errno); 1313 if ((rt = in6_pcbrtentry(in6p)) == 0) 1314 return; 1315 } 1316 1317 /* 1318 * Slow start out of the error condition. We 1319 * use the MTU because we know it's smaller 1320 * than the previously transmitted segment. 1321 * 1322 * Note: This is more conservative than the 1323 * suggestion in draft-floyd-incr-init-win-03. 1324 */ 1325 if (rt->rt_rmx.rmx_mtu != 0) 1326 tp->snd_cwnd = 1327 TCP_INITIAL_WINDOW(tcp_init_win, 1328 rt->rt_rmx.rmx_mtu); 1329 } 1330 1331 /* 1332 * Resend unacknowledged packets. 1333 */ 1334 tp->snd_nxt = tp->snd_una; 1335 tcp_output(tp); 1336 } 1337 } 1338 #endif 1339 1340 /* 1341 * Compute the MSS to advertise to the peer. Called only during 1342 * the 3-way handshake. If we are the server (peer initiated 1343 * connection), we are called with a pointer to the interface 1344 * on which the SYN packet arrived. If we are the client (we 1345 * initiated connection), we are called with a pointer to the 1346 * interface out which this connection should go. 1347 * 1348 * NOTE: Do not subtract IP option/extension header size nor IPsec 1349 * header size from MSS advertisement. MSS option must hold the maximum 1350 * segment size we can accept, so it must always be: 1351 * max(if mtu) - ip header - tcp header 1352 */ 1353 u_long 1354 tcp_mss_to_advertise(ifp, af) 1355 const struct ifnet *ifp; 1356 int af; 1357 { 1358 extern u_long in_maxmtu; 1359 u_long mss = 0; 1360 u_long hdrsiz; 1361 1362 /* 1363 * In order to avoid defeating path MTU discovery on the peer, 1364 * we advertise the max MTU of all attached networks as our MSS, 1365 * per RFC 1191, section 3.1. 1366 * 1367 * We provide the option to advertise just the MTU of 1368 * the interface on which we hope this connection will 1369 * be receiving. If we are responding to a SYN, we 1370 * will have a pretty good idea about this, but when 1371 * initiating a connection there is a bit more doubt. 1372 * 1373 * We also need to ensure that loopback has a large enough 1374 * MSS, as the loopback MTU is never included in in_maxmtu. 1375 */ 1376 1377 if (ifp != NULL) 1378 mss = ifp->if_mtu; 1379 1380 if (tcp_mss_ifmtu == 0) 1381 mss = max(in_maxmtu, mss); 1382 1383 switch (af) { 1384 case AF_INET: 1385 hdrsiz = sizeof(struct ip); 1386 break; 1387 #ifdef INET6 1388 case AF_INET6: 1389 hdrsiz = sizeof(struct ip6_hdr); 1390 break; 1391 #endif 1392 default: 1393 hdrsiz = 0; 1394 break; 1395 } 1396 hdrsiz += sizeof(struct tcphdr); 1397 if (mss > hdrsiz) 1398 mss -= hdrsiz; 1399 1400 mss = max(tcp_mssdflt, mss); 1401 return (mss); 1402 } 1403 1404 /* 1405 * Set connection variables based on the peer's advertised MSS. 1406 * We are passed the TCPCB for the actual connection. If we 1407 * are the server, we are called by the compressed state engine 1408 * when the 3-way handshake is complete. If we are the client, 1409 * we are called when we recieve the SYN,ACK from the server. 1410 * 1411 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1412 * before this routine is called! 1413 */ 1414 void 1415 tcp_mss_from_peer(tp, offer) 1416 struct tcpcb *tp; 1417 int offer; 1418 { 1419 struct socket *so; 1420 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1421 struct rtentry *rt; 1422 #endif 1423 u_long bufsize; 1424 int mss; 1425 1426 #ifdef DIAGNOSTIC 1427 if (tp->t_inpcb && tp->t_in6pcb) 1428 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1429 #endif 1430 so = NULL; 1431 rt = NULL; 1432 #ifdef INET 1433 if (tp->t_inpcb) { 1434 so = tp->t_inpcb->inp_socket; 1435 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1436 rt = in_pcbrtentry(tp->t_inpcb); 1437 #endif 1438 } 1439 #endif 1440 #ifdef INET6 1441 if (tp->t_in6pcb) { 1442 so = tp->t_in6pcb->in6p_socket; 1443 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1444 #ifdef TCP6 1445 rt = NULL; 1446 #else 1447 rt = in6_pcbrtentry(tp->t_in6pcb); 1448 #endif 1449 #endif 1450 } 1451 #endif 1452 1453 /* 1454 * As per RFC1122, use the default MSS value, unless they 1455 * sent us an offer. Do not accept offers less than 32 bytes. 1456 */ 1457 mss = tcp_mssdflt; 1458 if (offer) 1459 mss = offer; 1460 mss = max(mss, 32); /* sanity */ 1461 tp->t_peermss = mss; 1462 mss -= tcp_optlen(tp); 1463 #ifdef INET 1464 if (tp->t_inpcb) 1465 mss -= ip_optlen(tp->t_inpcb); 1466 #endif 1467 #ifdef INET6 1468 if (tp->t_in6pcb) 1469 mss -= ip6_optlen(tp->t_in6pcb); 1470 #endif 1471 1472 /* 1473 * If there's a pipesize, change the socket buffer to that size. 1474 * Make the socket buffer an integral number of MSS units. If 1475 * the MSS is larger than the socket buffer, artificially decrease 1476 * the MSS. 1477 */ 1478 #ifdef RTV_SPIPE 1479 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1480 bufsize = rt->rt_rmx.rmx_sendpipe; 1481 else 1482 #endif 1483 bufsize = so->so_snd.sb_hiwat; 1484 if (bufsize < mss) 1485 mss = bufsize; 1486 else { 1487 bufsize = roundup(bufsize, mss); 1488 if (bufsize > sb_max) 1489 bufsize = sb_max; 1490 (void) sbreserve(&so->so_snd, bufsize); 1491 } 1492 tp->t_segsz = mss; 1493 1494 #ifdef RTV_SSTHRESH 1495 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1496 /* 1497 * There's some sort of gateway or interface buffer 1498 * limit on the path. Use this to set the slow 1499 * start threshold, but set the threshold to no less 1500 * than 2 * MSS. 1501 */ 1502 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1503 } 1504 #endif 1505 } 1506 1507 /* 1508 * Processing necessary when a TCP connection is established. 1509 */ 1510 void 1511 tcp_established(tp) 1512 struct tcpcb *tp; 1513 { 1514 struct socket *so; 1515 #ifdef RTV_RPIPE 1516 struct rtentry *rt; 1517 #endif 1518 u_long bufsize; 1519 1520 #ifdef DIAGNOSTIC 1521 if (tp->t_inpcb && tp->t_in6pcb) 1522 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1523 #endif 1524 so = NULL; 1525 rt = NULL; 1526 #ifdef INET 1527 if (tp->t_inpcb) { 1528 so = tp->t_inpcb->inp_socket; 1529 #if defined(RTV_RPIPE) 1530 rt = in_pcbrtentry(tp->t_inpcb); 1531 #endif 1532 } 1533 #endif 1534 #ifdef INET6 1535 if (tp->t_in6pcb) { 1536 so = tp->t_in6pcb->in6p_socket; 1537 #if defined(RTV_RPIPE) 1538 #ifdef TCP6 1539 rt = NULL; 1540 #else 1541 rt = in6_pcbrtentry(tp->t_in6pcb); 1542 #endif 1543 #endif 1544 } 1545 #endif 1546 1547 tp->t_state = TCPS_ESTABLISHED; 1548 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1549 1550 #ifdef RTV_RPIPE 1551 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1552 bufsize = rt->rt_rmx.rmx_recvpipe; 1553 else 1554 #endif 1555 bufsize = so->so_rcv.sb_hiwat; 1556 if (bufsize > tp->t_ourmss) { 1557 bufsize = roundup(bufsize, tp->t_ourmss); 1558 if (bufsize > sb_max) 1559 bufsize = sb_max; 1560 (void) sbreserve(&so->so_rcv, bufsize); 1561 } 1562 } 1563 1564 /* 1565 * Check if there's an initial rtt or rttvar. Convert from the 1566 * route-table units to scaled multiples of the slow timeout timer. 1567 * Called only during the 3-way handshake. 1568 */ 1569 void 1570 tcp_rmx_rtt(tp) 1571 struct tcpcb *tp; 1572 { 1573 #ifdef RTV_RTT 1574 struct rtentry *rt = NULL; 1575 int rtt; 1576 1577 #ifdef DIAGNOSTIC 1578 if (tp->t_inpcb && tp->t_in6pcb) 1579 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1580 #endif 1581 #ifdef INET 1582 if (tp->t_inpcb) 1583 rt = in_pcbrtentry(tp->t_inpcb); 1584 #endif 1585 #ifdef INET6 1586 if (tp->t_in6pcb) { 1587 #ifdef TCP6 1588 rt = NULL; 1589 #else 1590 rt = in6_pcbrtentry(tp->t_in6pcb); 1591 #endif 1592 } 1593 #endif 1594 if (rt == NULL) 1595 return; 1596 1597 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1598 /* 1599 * XXX The lock bit for MTU indicates that the value 1600 * is also a minimum value; this is subject to time. 1601 */ 1602 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1603 TCPT_RANGESET(tp->t_rttmin, 1604 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1605 TCPTV_MIN, TCPTV_REXMTMAX); 1606 tp->t_srtt = rtt / 1607 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1608 if (rt->rt_rmx.rmx_rttvar) { 1609 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1610 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1611 (TCP_RTTVAR_SHIFT + 2)); 1612 } else { 1613 /* Default variation is +- 1 rtt */ 1614 tp->t_rttvar = 1615 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1616 } 1617 TCPT_RANGESET(tp->t_rxtcur, 1618 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1619 tp->t_rttmin, TCPTV_REXMTMAX); 1620 } 1621 #endif 1622 } 1623 1624 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1625 1626 /* 1627 * Get a new sequence value given a tcp control block 1628 */ 1629 tcp_seq 1630 tcp_new_iss(tp, len, addin) 1631 void *tp; 1632 u_long len; 1633 tcp_seq addin; 1634 { 1635 tcp_seq tcp_iss; 1636 1637 /* 1638 * Randomize. 1639 */ 1640 #if NRND > 0 1641 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 1642 #else 1643 tcp_iss = random(); 1644 #endif 1645 1646 /* 1647 * If we were asked to add some amount to a known value, 1648 * we will take a random value obtained above, mask off the upper 1649 * bits, and add in the known value. We also add in a constant to 1650 * ensure that we are at least a certain distance from the original 1651 * value. 1652 * 1653 * This is used when an old connection is in timed wait 1654 * and we have a new one coming in, for instance. 1655 */ 1656 if (addin != 0) { 1657 #ifdef TCPISS_DEBUG 1658 printf("Random %08x, ", tcp_iss); 1659 #endif 1660 tcp_iss &= TCP_ISS_RANDOM_MASK; 1661 tcp_iss += addin + TCP_ISSINCR; 1662 #ifdef TCPISS_DEBUG 1663 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 1664 #endif 1665 } else { 1666 tcp_iss &= TCP_ISS_RANDOM_MASK; 1667 tcp_iss += tcp_iss_seq; 1668 tcp_iss_seq += TCP_ISSINCR; 1669 #ifdef TCPISS_DEBUG 1670 printf("ISS %08x\n", tcp_iss); 1671 #endif 1672 } 1673 1674 if (tcp_compat_42) { 1675 /* 1676 * Limit it to the positive range for really old TCP 1677 * implementations. 1678 */ 1679 if (tcp_iss >= 0x80000000) 1680 tcp_iss &= 0x7fffffff; /* XXX */ 1681 } 1682 1683 return tcp_iss; 1684 } 1685 1686 #ifdef IPSEC 1687 /* compute ESP/AH header size for TCP, including outer IP header. */ 1688 size_t 1689 ipsec4_hdrsiz_tcp(tp) 1690 struct tcpcb *tp; 1691 { 1692 struct inpcb *inp; 1693 size_t hdrsiz; 1694 1695 /* XXX mapped addr case (tp->t_in6pcb) */ 1696 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 1697 return 0; 1698 switch (tp->t_family) { 1699 case AF_INET: 1700 /* XXX: should use currect direction. */ 1701 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 1702 break; 1703 default: 1704 hdrsiz = 0; 1705 break; 1706 } 1707 1708 return hdrsiz; 1709 } 1710 1711 #if defined(INET6) && !defined(TCP6) 1712 size_t 1713 ipsec6_hdrsiz_tcp(tp) 1714 struct tcpcb *tp; 1715 { 1716 struct in6pcb *in6p; 1717 size_t hdrsiz; 1718 1719 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 1720 return 0; 1721 switch (tp->t_family) { 1722 case AF_INET6: 1723 /* XXX: should use currect direction. */ 1724 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 1725 break; 1726 case AF_INET: 1727 /* mapped address case - tricky */ 1728 default: 1729 hdrsiz = 0; 1730 break; 1731 } 1732 1733 return hdrsiz; 1734 } 1735 #endif 1736 #endif /*IPSEC*/ 1737 1738 /* 1739 * Determine the length of the TCP options for this connection. 1740 * 1741 * XXX: What do we do for SACK, when we add that? Just reserve 1742 * all of the space? Otherwise we can't exactly be incrementing 1743 * cwnd by an amount that varies depending on the amount we last 1744 * had to SACK! 1745 */ 1746 1747 u_int 1748 tcp_optlen(tp) 1749 struct tcpcb *tp; 1750 { 1751 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1752 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1753 return TCPOLEN_TSTAMP_APPA; 1754 else 1755 return 0; 1756 } 1757
Indexes created Wed Sep 24 02:09:48 GMT 2025