tcp_subr.c revision 1.127
1 /* $NetBSD: tcp_subr.c,v 1.127 2002/05/12 20:33:50 matt Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998, 2000, 2001 The NetBSD Foundation, Inc. 34 * All rights reserved. 35 * 36 * This code is derived from software contributed to The NetBSD Foundation 37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 38 * Facility, NASA Ames Research Center. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. All advertising materials mentioning features or use of this software 49 * must display the following acknowledgement: 50 * This product includes software developed by the NetBSD 51 * Foundation, Inc. and its contributors. 52 * 4. Neither the name of The NetBSD Foundation nor the names of its 53 * contributors may be used to endorse or promote products derived 54 * from this software without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 57 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 66 * POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 /* 70 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 71 * The Regents of the University of California. All rights reserved. 72 * 73 * Redistribution and use in source and binary forms, with or without 74 * modification, are permitted provided that the following conditions 75 * are met: 76 * 1. Redistributions of source code must retain the above copyright 77 * notice, this list of conditions and the following disclaimer. 78 * 2. Redistributions in binary form must reproduce the above copyright 79 * notice, this list of conditions and the following disclaimer in the 80 * documentation and/or other materials provided with the distribution. 81 * 3. All advertising materials mentioning features or use of this software 82 * must display the following acknowledgement: 83 * This product includes software developed by the University of 84 * California, Berkeley and its contributors. 85 * 4. Neither the name of the University nor the names of its contributors 86 * may be used to endorse or promote products derived from this software 87 * without specific prior written permission. 88 * 89 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 90 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 91 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 92 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 93 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 94 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 95 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 96 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 97 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 98 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 99 * SUCH DAMAGE. 100 * 101 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 102 */ 103 104 #include <sys/cdefs.h> 105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.127 2002/05/12 20:33:50 matt Exp $"); 106 107 #include "opt_inet.h" 108 #include "opt_ipsec.h" 109 #include "opt_tcp_compat_42.h" 110 #include "opt_inet_csum.h" 111 #include "rnd.h" 112 113 #include <sys/param.h> 114 #include <sys/proc.h> 115 #include <sys/systm.h> 116 #include <sys/malloc.h> 117 #include <sys/mbuf.h> 118 #include <sys/socket.h> 119 #include <sys/socketvar.h> 120 #include <sys/protosw.h> 121 #include <sys/errno.h> 122 #include <sys/kernel.h> 123 #include <sys/pool.h> 124 #if NRND > 0 125 #include <sys/md5.h> 126 #include <sys/rnd.h> 127 #endif 128 129 #include <net/route.h> 130 #include <net/if.h> 131 132 #include <netinet/in.h> 133 #include <netinet/in_systm.h> 134 #include <netinet/ip.h> 135 #include <netinet/in_pcb.h> 136 #include <netinet/ip_var.h> 137 #include <netinet/ip_icmp.h> 138 139 #ifdef INET6 140 #ifndef INET 141 #include <netinet/in.h> 142 #endif 143 #include <netinet/ip6.h> 144 #include <netinet6/in6_pcb.h> 145 #include <netinet6/ip6_var.h> 146 #include <netinet6/in6_var.h> 147 #include <netinet6/ip6protosw.h> 148 #include <netinet/icmp6.h> 149 #endif 150 151 #include <netinet/tcp.h> 152 #include <netinet/tcp_fsm.h> 153 #include <netinet/tcp_seq.h> 154 #include <netinet/tcp_timer.h> 155 #include <netinet/tcp_var.h> 156 #include <netinet/tcpip.h> 157 158 #ifdef IPSEC 159 #include <netinet6/ipsec.h> 160 #endif /*IPSEC*/ 161 162 #ifdef INET6 163 struct in6pcb tcb6; 164 #endif 165 166 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 167 struct tcpstat tcpstat; /* tcp statistics */ 168 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 169 170 /* patchable/settable parameters for tcp */ 171 int tcp_mssdflt = TCP_MSS; 172 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 173 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 174 #if NRND > 0 175 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 176 #endif 177 int tcp_do_sack = 1; /* selective acknowledgement */ 178 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 179 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 180 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 181 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 182 int tcp_init_win = 1; 183 int tcp_mss_ifmtu = 0; 184 #ifdef TCP_COMPAT_42 185 int tcp_compat_42 = 1; 186 #else 187 int tcp_compat_42 = 0; 188 #endif 189 int tcp_rst_ppslim = 100; /* 100pps */ 190 191 /* tcb hash */ 192 #ifndef TCBHASHSIZE 193 #define TCBHASHSIZE 128 194 #endif 195 int tcbhashsize = TCBHASHSIZE; 196 197 /* syn hash parameters */ 198 #define TCP_SYN_HASH_SIZE 293 199 #define TCP_SYN_BUCKET_SIZE 35 200 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 201 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 202 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 203 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 204 205 int tcp_freeq __P((struct tcpcb *)); 206 207 #ifdef INET 208 void tcp_mtudisc_callback __P((struct in_addr)); 209 #endif 210 #ifdef INET6 211 void tcp6_mtudisc_callback __P((struct in6_addr *)); 212 #endif 213 214 void tcp_mtudisc __P((struct inpcb *, int)); 215 #ifdef INET6 216 void tcp6_mtudisc __P((struct in6pcb *, int)); 217 #endif 218 219 struct pool tcpcb_pool; 220 221 #ifdef TCP_CSUM_COUNTERS 222 #include <sys/device.h> 223 224 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 225 NULL, "tcp", "hwcsum bad"); 226 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 227 NULL, "tcp", "hwcsum ok"); 228 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 229 NULL, "tcp", "hwcsum data"); 230 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 231 NULL, "tcp", "swcsum"); 232 #endif /* TCP_CSUM_COUNTERS */ 233 234 #ifdef TCP_OUTPUT_COUNTERS 235 #include <sys/device.h> 236 237 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 238 NULL, "tcp", "output big header"); 239 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 240 NULL, "tcp", "output copy small"); 241 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "output copy big"); 243 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "output reference big"); 245 #endif /* TCP_OUTPUT_COUNTERS */ 246 247 #ifdef TCP_REASS_COUNTERS 248 #include <sys/device.h> 249 250 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp_reass", "calls"); 252 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 &tcp_reass_, "tcp_reass", "insert into empty queue"); 254 struct evcnt tcp_reass_iteration[8] = { 255 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 256 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 257 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 258 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 259 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 260 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 261 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 262 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 263 }; 264 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 265 &tcp_reass_, "tcp_reass", "prepend to first"); 266 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 &tcp_reass_, "tcp_reass", "prepend"); 268 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 269 &tcp_reass_, "tcp_reass", "insert"); 270 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 271 &tcp_reass_, "tcp_reass", "insert at tail"); 272 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 273 &tcp_reass_, "tcp_reass", "append"); 274 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 275 &tcp_reass_, "tcp_reass", "append to tail fragment"); 276 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 277 &tcp_reass_, "tcp_reass", "overlap at end"); 278 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 279 &tcp_reass_, "tcp_reass", "overlap at start"); 280 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 281 &tcp_reass_, "tcp_reass", "duplicate segment"); 282 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 &tcp_reass_, "tcp_reass", "duplicate fragment"); 284 285 #endif /* TCP_REASS_COUNTERS */ 286 287 /* 288 * Tcp initialization 289 */ 290 void 291 tcp_init() 292 { 293 int hlen; 294 295 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 296 NULL); 297 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 298 #ifdef INET6 299 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 300 #endif 301 302 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 303 #ifdef INET6 304 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 305 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 306 #endif 307 if (max_protohdr < hlen) 308 max_protohdr = hlen; 309 if (max_linkhdr + hlen > MHLEN) 310 panic("tcp_init"); 311 312 #ifdef INET 313 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 314 #endif 315 #ifdef INET6 316 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 317 #endif 318 319 /* Initialize timer state. */ 320 tcp_timer_init(); 321 322 /* Initialize the compressed state engine. */ 323 syn_cache_init(); 324 325 #ifdef TCP_CSUM_COUNTERS 326 evcnt_attach_static(&tcp_hwcsum_bad); 327 evcnt_attach_static(&tcp_hwcsum_ok); 328 evcnt_attach_static(&tcp_hwcsum_data); 329 evcnt_attach_static(&tcp_swcsum); 330 #endif /* TCP_CSUM_COUNTERS */ 331 332 #ifdef TCP_OUTPUT_COUNTERS 333 evcnt_attach_static(&tcp_output_bigheader); 334 evcnt_attach_static(&tcp_output_copysmall); 335 evcnt_attach_static(&tcp_output_copybig); 336 evcnt_attach_static(&tcp_output_refbig); 337 #endif /* TCP_OUTPUT_COUNTERS */ 338 339 #ifdef TCP_REASS_COUNTERS 340 evcnt_attach_static(&tcp_reass_); 341 evcnt_attach_static(&tcp_reass_empty); 342 evcnt_attach_static(&tcp_reass_iteration[0]); 343 evcnt_attach_static(&tcp_reass_iteration[1]); 344 evcnt_attach_static(&tcp_reass_iteration[2]); 345 evcnt_attach_static(&tcp_reass_iteration[3]); 346 evcnt_attach_static(&tcp_reass_iteration[4]); 347 evcnt_attach_static(&tcp_reass_iteration[5]); 348 evcnt_attach_static(&tcp_reass_iteration[6]); 349 evcnt_attach_static(&tcp_reass_iteration[7]); 350 evcnt_attach_static(&tcp_reass_prependfirst); 351 evcnt_attach_static(&tcp_reass_prepend); 352 evcnt_attach_static(&tcp_reass_insert); 353 evcnt_attach_static(&tcp_reass_inserttail); 354 evcnt_attach_static(&tcp_reass_append); 355 evcnt_attach_static(&tcp_reass_appendtail); 356 evcnt_attach_static(&tcp_reass_overlaptail); 357 evcnt_attach_static(&tcp_reass_overlapfront); 358 evcnt_attach_static(&tcp_reass_segdup); 359 evcnt_attach_static(&tcp_reass_fragdup); 360 #endif /* TCP_REASS_COUNTERS */ 361 } 362 363 /* 364 * Create template to be used to send tcp packets on a connection. 365 * Call after host entry created, allocates an mbuf and fills 366 * in a skeletal tcp/ip header, minimizing the amount of work 367 * necessary when the connection is used. 368 */ 369 struct mbuf * 370 tcp_template(tp) 371 struct tcpcb *tp; 372 { 373 struct inpcb *inp = tp->t_inpcb; 374 #ifdef INET6 375 struct in6pcb *in6p = tp->t_in6pcb; 376 #endif 377 struct tcphdr *n; 378 struct mbuf *m; 379 int hlen; 380 381 switch (tp->t_family) { 382 case AF_INET: 383 hlen = sizeof(struct ip); 384 if (inp) 385 break; 386 #ifdef INET6 387 if (in6p) { 388 /* mapped addr case */ 389 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 390 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 391 break; 392 } 393 #endif 394 return NULL; /*EINVAL*/ 395 #ifdef INET6 396 case AF_INET6: 397 hlen = sizeof(struct ip6_hdr); 398 if (in6p) { 399 /* more sainty check? */ 400 break; 401 } 402 return NULL; /*EINVAL*/ 403 #endif 404 default: 405 hlen = 0; /*pacify gcc*/ 406 return NULL; /*EAFNOSUPPORT*/ 407 } 408 #ifdef DIAGNOSTIC 409 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 410 panic("mclbytes too small for t_template"); 411 #endif 412 m = tp->t_template; 413 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 414 ; 415 else { 416 if (m) 417 m_freem(m); 418 m = tp->t_template = NULL; 419 MGETHDR(m, M_DONTWAIT, MT_HEADER); 420 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 421 MCLGET(m, M_DONTWAIT); 422 if ((m->m_flags & M_EXT) == 0) { 423 m_free(m); 424 m = NULL; 425 } 426 } 427 if (m == NULL) 428 return NULL; 429 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 430 } 431 432 bzero(mtod(m, caddr_t), m->m_len); 433 434 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 435 436 switch (tp->t_family) { 437 case AF_INET: 438 { 439 struct ipovly *ipov; 440 mtod(m, struct ip *)->ip_v = 4; 441 ipov = mtod(m, struct ipovly *); 442 ipov->ih_pr = IPPROTO_TCP; 443 ipov->ih_len = htons(sizeof(struct tcphdr)); 444 if (inp) { 445 ipov->ih_src = inp->inp_laddr; 446 ipov->ih_dst = inp->inp_faddr; 447 } 448 #ifdef INET6 449 else if (in6p) { 450 /* mapped addr case */ 451 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 452 sizeof(ipov->ih_src)); 453 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 454 sizeof(ipov->ih_dst)); 455 } 456 #endif 457 /* 458 * Compute the pseudo-header portion of the checksum 459 * now. We incrementally add in the TCP option and 460 * payload lengths later, and then compute the TCP 461 * checksum right before the packet is sent off onto 462 * the wire. 463 */ 464 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 465 ipov->ih_dst.s_addr, 466 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 467 break; 468 } 469 #ifdef INET6 470 case AF_INET6: 471 { 472 struct ip6_hdr *ip6; 473 mtod(m, struct ip *)->ip_v = 6; 474 ip6 = mtod(m, struct ip6_hdr *); 475 ip6->ip6_nxt = IPPROTO_TCP; 476 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 477 ip6->ip6_src = in6p->in6p_laddr; 478 ip6->ip6_dst = in6p->in6p_faddr; 479 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 480 if (ip6_auto_flowlabel) { 481 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 482 ip6->ip6_flow |= 483 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 484 } 485 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 486 ip6->ip6_vfc |= IPV6_VERSION; 487 488 /* 489 * Compute the pseudo-header portion of the checksum 490 * now. We incrementally add in the TCP option and 491 * payload lengths later, and then compute the TCP 492 * checksum right before the packet is sent off onto 493 * the wire. 494 */ 495 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 496 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 497 htonl(IPPROTO_TCP)); 498 break; 499 } 500 #endif 501 } 502 if (inp) { 503 n->th_sport = inp->inp_lport; 504 n->th_dport = inp->inp_fport; 505 } 506 #ifdef INET6 507 else if (in6p) { 508 n->th_sport = in6p->in6p_lport; 509 n->th_dport = in6p->in6p_fport; 510 } 511 #endif 512 n->th_seq = 0; 513 n->th_ack = 0; 514 n->th_x2 = 0; 515 n->th_off = 5; 516 n->th_flags = 0; 517 n->th_win = 0; 518 n->th_urp = 0; 519 return (m); 520 } 521 522 /* 523 * Send a single message to the TCP at address specified by 524 * the given TCP/IP header. If m == 0, then we make a copy 525 * of the tcpiphdr at ti and send directly to the addressed host. 526 * This is used to force keep alive messages out using the TCP 527 * template for a connection tp->t_template. If flags are given 528 * then we send a message back to the TCP which originated the 529 * segment ti, and discard the mbuf containing it and any other 530 * attached mbufs. 531 * 532 * In any case the ack and sequence number of the transmitted 533 * segment are as specified by the parameters. 534 */ 535 int 536 tcp_respond(tp, template, m, th0, ack, seq, flags) 537 struct tcpcb *tp; 538 struct mbuf *template; 539 struct mbuf *m; 540 struct tcphdr *th0; 541 tcp_seq ack, seq; 542 int flags; 543 { 544 struct route *ro; 545 int error, tlen, win = 0; 546 int hlen; 547 struct ip *ip; 548 #ifdef INET6 549 struct ip6_hdr *ip6; 550 #endif 551 int family; /* family on packet, not inpcb/in6pcb! */ 552 struct tcphdr *th; 553 554 if (tp != NULL && (flags & TH_RST) == 0) { 555 #ifdef DIAGNOSTIC 556 if (tp->t_inpcb && tp->t_in6pcb) 557 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 558 #endif 559 #ifdef INET 560 if (tp->t_inpcb) 561 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 562 #endif 563 #ifdef INET6 564 if (tp->t_in6pcb) 565 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 566 #endif 567 } 568 569 ip = NULL; 570 #ifdef INET6 571 ip6 = NULL; 572 #endif 573 if (m == 0) { 574 if (!template) 575 return EINVAL; 576 577 /* get family information from template */ 578 switch (mtod(template, struct ip *)->ip_v) { 579 case 4: 580 family = AF_INET; 581 hlen = sizeof(struct ip); 582 break; 583 #ifdef INET6 584 case 6: 585 family = AF_INET6; 586 hlen = sizeof(struct ip6_hdr); 587 break; 588 #endif 589 default: 590 return EAFNOSUPPORT; 591 } 592 593 MGETHDR(m, M_DONTWAIT, MT_HEADER); 594 if (m) { 595 MCLGET(m, M_DONTWAIT); 596 if ((m->m_flags & M_EXT) == 0) { 597 m_free(m); 598 m = NULL; 599 } 600 } 601 if (m == NULL) 602 return (ENOBUFS); 603 604 if (tcp_compat_42) 605 tlen = 1; 606 else 607 tlen = 0; 608 609 m->m_data += max_linkhdr; 610 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 611 template->m_len); 612 switch (family) { 613 case AF_INET: 614 ip = mtod(m, struct ip *); 615 th = (struct tcphdr *)(ip + 1); 616 break; 617 #ifdef INET6 618 case AF_INET6: 619 ip6 = mtod(m, struct ip6_hdr *); 620 th = (struct tcphdr *)(ip6 + 1); 621 break; 622 #endif 623 #if 0 624 default: 625 /* noone will visit here */ 626 m_freem(m); 627 return EAFNOSUPPORT; 628 #endif 629 } 630 flags = TH_ACK; 631 } else { 632 633 if ((m->m_flags & M_PKTHDR) == 0) { 634 #if 0 635 printf("non PKTHDR to tcp_respond\n"); 636 #endif 637 m_freem(m); 638 return EINVAL; 639 } 640 #ifdef DIAGNOSTIC 641 if (!th0) 642 panic("th0 == NULL in tcp_respond"); 643 #endif 644 645 /* get family information from m */ 646 switch (mtod(m, struct ip *)->ip_v) { 647 case 4: 648 family = AF_INET; 649 hlen = sizeof(struct ip); 650 ip = mtod(m, struct ip *); 651 break; 652 #ifdef INET6 653 case 6: 654 family = AF_INET6; 655 hlen = sizeof(struct ip6_hdr); 656 ip6 = mtod(m, struct ip6_hdr *); 657 break; 658 #endif 659 default: 660 m_freem(m); 661 return EAFNOSUPPORT; 662 } 663 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 664 tlen = sizeof(*th0); 665 else 666 tlen = th0->th_off << 2; 667 668 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 669 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 670 m->m_len = hlen + tlen; 671 m_freem(m->m_next); 672 m->m_next = NULL; 673 } else { 674 struct mbuf *n; 675 676 #ifdef DIAGNOSTIC 677 if (max_linkhdr + hlen + tlen > MCLBYTES) { 678 m_freem(m); 679 return EMSGSIZE; 680 } 681 #endif 682 MGETHDR(n, M_DONTWAIT, MT_HEADER); 683 if (n && max_linkhdr + hlen + tlen > MHLEN) { 684 MCLGET(n, M_DONTWAIT); 685 if ((n->m_flags & M_EXT) == 0) { 686 m_freem(n); 687 n = NULL; 688 } 689 } 690 if (!n) { 691 m_freem(m); 692 return ENOBUFS; 693 } 694 695 n->m_data += max_linkhdr; 696 n->m_len = hlen + tlen; 697 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 698 m_copyback(n, hlen, tlen, (caddr_t)th0); 699 700 m_freem(m); 701 m = n; 702 n = NULL; 703 } 704 705 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 706 switch (family) { 707 case AF_INET: 708 ip = mtod(m, struct ip *); 709 th = (struct tcphdr *)(ip + 1); 710 ip->ip_p = IPPROTO_TCP; 711 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 712 ip->ip_p = IPPROTO_TCP; 713 break; 714 #ifdef INET6 715 case AF_INET6: 716 ip6 = mtod(m, struct ip6_hdr *); 717 th = (struct tcphdr *)(ip6 + 1); 718 ip6->ip6_nxt = IPPROTO_TCP; 719 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 720 ip6->ip6_nxt = IPPROTO_TCP; 721 break; 722 #endif 723 #if 0 724 default: 725 /* noone will visit here */ 726 m_freem(m); 727 return EAFNOSUPPORT; 728 #endif 729 } 730 xchg(th->th_dport, th->th_sport, u_int16_t); 731 #undef xchg 732 tlen = 0; /*be friendly with the following code*/ 733 } 734 th->th_seq = htonl(seq); 735 th->th_ack = htonl(ack); 736 th->th_x2 = 0; 737 if ((flags & TH_SYN) == 0) { 738 if (tp) 739 win >>= tp->rcv_scale; 740 if (win > TCP_MAXWIN) 741 win = TCP_MAXWIN; 742 th->th_win = htons((u_int16_t)win); 743 th->th_off = sizeof (struct tcphdr) >> 2; 744 tlen += sizeof(*th); 745 } else 746 tlen += th->th_off << 2; 747 m->m_len = hlen + tlen; 748 m->m_pkthdr.len = hlen + tlen; 749 m->m_pkthdr.rcvif = (struct ifnet *) 0; 750 th->th_flags = flags; 751 th->th_urp = 0; 752 753 switch (family) { 754 #ifdef INET 755 case AF_INET: 756 { 757 struct ipovly *ipov = (struct ipovly *)ip; 758 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 759 ipov->ih_len = htons((u_int16_t)tlen); 760 761 th->th_sum = 0; 762 th->th_sum = in_cksum(m, hlen + tlen); 763 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 764 ip->ip_ttl = ip_defttl; 765 break; 766 } 767 #endif 768 #ifdef INET6 769 case AF_INET6: 770 { 771 th->th_sum = 0; 772 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 773 tlen); 774 ip6->ip6_plen = ntohs(tlen); 775 if (tp && tp->t_in6pcb) { 776 struct ifnet *oifp; 777 ro = (struct route *)&tp->t_in6pcb->in6p_route; 778 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 779 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 780 } else 781 ip6->ip6_hlim = ip6_defhlim; 782 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 783 if (ip6_auto_flowlabel) { 784 ip6->ip6_flow |= 785 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 786 } 787 break; 788 } 789 #endif 790 } 791 792 #ifdef IPSEC 793 (void)ipsec_setsocket(m, NULL); 794 #endif /*IPSEC*/ 795 796 if (tp != NULL && tp->t_inpcb != NULL) { 797 ro = &tp->t_inpcb->inp_route; 798 #ifdef IPSEC 799 if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) { 800 m_freem(m); 801 return ENOBUFS; 802 } 803 #endif 804 #ifdef DIAGNOSTIC 805 if (family != AF_INET) 806 panic("tcp_respond: address family mismatch"); 807 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 808 panic("tcp_respond: ip_dst %x != inp_faddr %x", 809 ntohl(ip->ip_dst.s_addr), 810 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 811 } 812 #endif 813 } 814 #ifdef INET6 815 else if (tp != NULL && tp->t_in6pcb != NULL) { 816 ro = (struct route *)&tp->t_in6pcb->in6p_route; 817 #ifdef IPSEC 818 if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) { 819 m_freem(m); 820 return ENOBUFS; 821 } 822 #endif 823 #ifdef DIAGNOSTIC 824 if (family == AF_INET) { 825 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 826 panic("tcp_respond: not mapped addr"); 827 if (bcmp(&ip->ip_dst, 828 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 829 sizeof(ip->ip_dst)) != 0) { 830 panic("tcp_respond: ip_dst != in6p_faddr"); 831 } 832 } else if (family == AF_INET6) { 833 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 834 panic("tcp_respond: ip6_dst != in6p_faddr"); 835 } else 836 panic("tcp_respond: address family mismatch"); 837 #endif 838 } 839 #endif 840 else 841 ro = NULL; 842 843 switch (family) { 844 #ifdef INET 845 case AF_INET: 846 error = ip_output(m, NULL, ro, 847 (ip_mtudisc ? IP_MTUDISC : 0), 848 NULL); 849 break; 850 #endif 851 #ifdef INET6 852 case AF_INET6: 853 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 854 NULL); 855 break; 856 #endif 857 default: 858 error = EAFNOSUPPORT; 859 break; 860 } 861 862 return (error); 863 } 864 865 /* 866 * Create a new TCP control block, making an 867 * empty reassembly queue and hooking it to the argument 868 * protocol control block. 869 */ 870 struct tcpcb * 871 tcp_newtcpcb(family, aux) 872 int family; /* selects inpcb, or in6pcb */ 873 void *aux; 874 { 875 struct tcpcb *tp; 876 int i; 877 878 switch (family) { 879 case PF_INET: 880 break; 881 #ifdef INET6 882 case PF_INET6: 883 break; 884 #endif 885 default: 886 return NULL; 887 } 888 889 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 890 if (tp == NULL) 891 return (NULL); 892 bzero((caddr_t)tp, sizeof(struct tcpcb)); 893 TAILQ_INIT(&tp->segq); 894 TAILQ_INIT(&tp->timeq); 895 tp->t_family = family; /* may be overridden later on */ 896 tp->t_peermss = tcp_mssdflt; 897 tp->t_ourmss = tcp_mssdflt; 898 tp->t_segsz = tcp_mssdflt; 899 LIST_INIT(&tp->t_sc); 900 901 callout_init(&tp->t_delack_ch); 902 for (i = 0; i < TCPT_NTIMERS; i++) 903 TCP_TIMER_INIT(tp, i); 904 905 tp->t_flags = 0; 906 if (tcp_do_rfc1323 && tcp_do_win_scale) 907 tp->t_flags |= TF_REQ_SCALE; 908 if (tcp_do_rfc1323 && tcp_do_timestamps) 909 tp->t_flags |= TF_REQ_TSTMP; 910 if (tcp_do_sack == 2) 911 tp->t_flags |= TF_WILL_SACK; 912 else if (tcp_do_sack == 1) 913 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 914 tp->t_flags |= TF_CANT_TXSACK; 915 switch (family) { 916 case PF_INET: 917 tp->t_inpcb = (struct inpcb *)aux; 918 break; 919 #ifdef INET6 920 case PF_INET6: 921 tp->t_in6pcb = (struct in6pcb *)aux; 922 break; 923 #endif 924 } 925 /* 926 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 927 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 928 * reasonable initial retransmit time. 929 */ 930 tp->t_srtt = TCPTV_SRTTBASE; 931 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 932 tp->t_rttmin = TCPTV_MIN; 933 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 934 TCPTV_MIN, TCPTV_REXMTMAX); 935 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 936 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 937 if (family == AF_INET) { 938 struct inpcb *inp = (struct inpcb *)aux; 939 inp->inp_ip.ip_ttl = ip_defttl; 940 inp->inp_ppcb = (caddr_t)tp; 941 } 942 #ifdef INET6 943 else if (family == AF_INET6) { 944 struct in6pcb *in6p = (struct in6pcb *)aux; 945 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 946 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 947 : NULL); 948 in6p->in6p_ppcb = (caddr_t)tp; 949 } 950 #endif 951 952 /* 953 * Initialize our timebase. When we send timestamps, we take 954 * the delta from tcp_now -- this means each connection always 955 * gets a timebase of 0, which makes it, among other things, 956 * more difficult to determine how long a system has been up, 957 * and thus how many TCP sequence increments have occurred. 958 */ 959 tp->ts_timebase = tcp_now; 960 961 return (tp); 962 } 963 964 /* 965 * Drop a TCP connection, reporting 966 * the specified error. If connection is synchronized, 967 * then send a RST to peer. 968 */ 969 struct tcpcb * 970 tcp_drop(tp, errno) 971 struct tcpcb *tp; 972 int errno; 973 { 974 struct socket *so = NULL; 975 976 #ifdef DIAGNOSTIC 977 if (tp->t_inpcb && tp->t_in6pcb) 978 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 979 #endif 980 #ifdef INET 981 if (tp->t_inpcb) 982 so = tp->t_inpcb->inp_socket; 983 #endif 984 #ifdef INET6 985 if (tp->t_in6pcb) 986 so = tp->t_in6pcb->in6p_socket; 987 #endif 988 if (!so) 989 return NULL; 990 991 if (TCPS_HAVERCVDSYN(tp->t_state)) { 992 tp->t_state = TCPS_CLOSED; 993 (void) tcp_output(tp); 994 tcpstat.tcps_drops++; 995 } else 996 tcpstat.tcps_conndrops++; 997 if (errno == ETIMEDOUT && tp->t_softerror) 998 errno = tp->t_softerror; 999 so->so_error = errno; 1000 return (tcp_close(tp)); 1001 } 1002 1003 /* 1004 * Close a TCP control block: 1005 * discard all space held by the tcp 1006 * discard internet protocol block 1007 * wake up any sleepers 1008 */ 1009 struct tcpcb * 1010 tcp_close(tp) 1011 struct tcpcb *tp; 1012 { 1013 struct inpcb *inp; 1014 #ifdef INET6 1015 struct in6pcb *in6p; 1016 #endif 1017 struct socket *so; 1018 #ifdef RTV_RTT 1019 struct rtentry *rt; 1020 #endif 1021 struct route *ro; 1022 1023 inp = tp->t_inpcb; 1024 #ifdef INET6 1025 in6p = tp->t_in6pcb; 1026 #endif 1027 so = NULL; 1028 ro = NULL; 1029 if (inp) { 1030 so = inp->inp_socket; 1031 ro = &inp->inp_route; 1032 } 1033 #ifdef INET6 1034 else if (in6p) { 1035 so = in6p->in6p_socket; 1036 ro = (struct route *)&in6p->in6p_route; 1037 } 1038 #endif 1039 1040 #ifdef RTV_RTT 1041 /* 1042 * If we sent enough data to get some meaningful characteristics, 1043 * save them in the routing entry. 'Enough' is arbitrarily 1044 * defined as the sendpipesize (default 4K) * 16. This would 1045 * give us 16 rtt samples assuming we only get one sample per 1046 * window (the usual case on a long haul net). 16 samples is 1047 * enough for the srtt filter to converge to within 5% of the correct 1048 * value; fewer samples and we could save a very bogus rtt. 1049 * 1050 * Don't update the default route's characteristics and don't 1051 * update anything that the user "locked". 1052 */ 1053 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1054 ro && (rt = ro->ro_rt) && 1055 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1056 u_long i = 0; 1057 1058 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1059 i = tp->t_srtt * 1060 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1061 if (rt->rt_rmx.rmx_rtt && i) 1062 /* 1063 * filter this update to half the old & half 1064 * the new values, converting scale. 1065 * See route.h and tcp_var.h for a 1066 * description of the scaling constants. 1067 */ 1068 rt->rt_rmx.rmx_rtt = 1069 (rt->rt_rmx.rmx_rtt + i) / 2; 1070 else 1071 rt->rt_rmx.rmx_rtt = i; 1072 } 1073 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1074 i = tp->t_rttvar * 1075 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1076 if (rt->rt_rmx.rmx_rttvar && i) 1077 rt->rt_rmx.rmx_rttvar = 1078 (rt->rt_rmx.rmx_rttvar + i) / 2; 1079 else 1080 rt->rt_rmx.rmx_rttvar = i; 1081 } 1082 /* 1083 * update the pipelimit (ssthresh) if it has been updated 1084 * already or if a pipesize was specified & the threshhold 1085 * got below half the pipesize. I.e., wait for bad news 1086 * before we start updating, then update on both good 1087 * and bad news. 1088 */ 1089 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1090 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1091 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1092 /* 1093 * convert the limit from user data bytes to 1094 * packets then to packet data bytes. 1095 */ 1096 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1097 if (i < 2) 1098 i = 2; 1099 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1100 if (rt->rt_rmx.rmx_ssthresh) 1101 rt->rt_rmx.rmx_ssthresh = 1102 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1103 else 1104 rt->rt_rmx.rmx_ssthresh = i; 1105 } 1106 } 1107 #endif /* RTV_RTT */ 1108 /* free the reassembly queue, if any */ 1109 TCP_REASS_LOCK(tp); 1110 (void) tcp_freeq(tp); 1111 TCP_REASS_UNLOCK(tp); 1112 1113 tcp_canceltimers(tp); 1114 TCP_CLEAR_DELACK(tp); 1115 syn_cache_cleanup(tp); 1116 1117 if (tp->t_template) { 1118 m_free(tp->t_template); 1119 tp->t_template = NULL; 1120 } 1121 pool_put(&tcpcb_pool, tp); 1122 if (inp) { 1123 inp->inp_ppcb = 0; 1124 soisdisconnected(so); 1125 in_pcbdetach(inp); 1126 } 1127 #ifdef INET6 1128 else if (in6p) { 1129 in6p->in6p_ppcb = 0; 1130 soisdisconnected(so); 1131 in6_pcbdetach(in6p); 1132 } 1133 #endif 1134 tcpstat.tcps_closed++; 1135 return ((struct tcpcb *)0); 1136 } 1137 1138 int 1139 tcp_freeq(tp) 1140 struct tcpcb *tp; 1141 { 1142 struct ipqent *qe; 1143 int rv = 0; 1144 #ifdef TCPREASS_DEBUG 1145 int i = 0; 1146 #endif 1147 1148 TCP_REASS_LOCK_CHECK(tp); 1149 1150 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1151 #ifdef TCPREASS_DEBUG 1152 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1153 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1154 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1155 #endif 1156 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1157 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1158 m_freem(qe->ipqe_m); 1159 pool_put(&ipqent_pool, qe); 1160 rv = 1; 1161 } 1162 return (rv); 1163 } 1164 1165 /* 1166 * Protocol drain routine. Called when memory is in short supply. 1167 */ 1168 void 1169 tcp_drain() 1170 { 1171 struct inpcb *inp; 1172 struct tcpcb *tp; 1173 1174 /* 1175 * Free the sequence queue of all TCP connections. 1176 */ 1177 inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue); 1178 if (inp) /* XXX */ 1179 CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) { 1180 if ((tp = intotcpcb(inp)) != NULL) { 1181 /* 1182 * We may be called from a device's interrupt 1183 * context. If the tcpcb is already busy, 1184 * just bail out now. 1185 */ 1186 if (tcp_reass_lock_try(tp) == 0) 1187 continue; 1188 if (tcp_freeq(tp)) 1189 tcpstat.tcps_connsdrained++; 1190 TCP_REASS_UNLOCK(tp); 1191 } 1192 } 1193 } 1194 1195 #ifdef INET6 1196 void 1197 tcp6_drain() 1198 { 1199 struct in6pcb *in6p; 1200 struct tcpcb *tp; 1201 struct in6pcb *head = &tcb6; 1202 1203 /* 1204 * Free the sequence queue of all TCP connections. 1205 */ 1206 for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) { 1207 if ((tp = in6totcpcb(in6p)) != NULL) { 1208 /* 1209 * We may be called from a device's interrupt 1210 * context. If the tcpcb is already busy, 1211 * just bail out now. 1212 */ 1213 if (tcp_reass_lock_try(tp) == 0) 1214 continue; 1215 if (tcp_freeq(tp)) 1216 tcpstat.tcps_connsdrained++; 1217 TCP_REASS_UNLOCK(tp); 1218 } 1219 } 1220 } 1221 #endif 1222 1223 /* 1224 * Notify a tcp user of an asynchronous error; 1225 * store error as soft error, but wake up user 1226 * (for now, won't do anything until can select for soft error). 1227 */ 1228 void 1229 tcp_notify(inp, error) 1230 struct inpcb *inp; 1231 int error; 1232 { 1233 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1234 struct socket *so = inp->inp_socket; 1235 1236 /* 1237 * Ignore some errors if we are hooked up. 1238 * If connection hasn't completed, has retransmitted several times, 1239 * and receives a second error, give up now. This is better 1240 * than waiting a long time to establish a connection that 1241 * can never complete. 1242 */ 1243 if (tp->t_state == TCPS_ESTABLISHED && 1244 (error == EHOSTUNREACH || error == ENETUNREACH || 1245 error == EHOSTDOWN)) { 1246 return; 1247 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1248 tp->t_rxtshift > 3 && tp->t_softerror) 1249 so->so_error = error; 1250 else 1251 tp->t_softerror = error; 1252 wakeup((caddr_t) &so->so_timeo); 1253 sorwakeup(so); 1254 sowwakeup(so); 1255 } 1256 1257 #ifdef INET6 1258 void 1259 tcp6_notify(in6p, error) 1260 struct in6pcb *in6p; 1261 int error; 1262 { 1263 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1264 struct socket *so = in6p->in6p_socket; 1265 1266 /* 1267 * Ignore some errors if we are hooked up. 1268 * If connection hasn't completed, has retransmitted several times, 1269 * and receives a second error, give up now. This is better 1270 * than waiting a long time to establish a connection that 1271 * can never complete. 1272 */ 1273 if (tp->t_state == TCPS_ESTABLISHED && 1274 (error == EHOSTUNREACH || error == ENETUNREACH || 1275 error == EHOSTDOWN)) { 1276 return; 1277 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1278 tp->t_rxtshift > 3 && tp->t_softerror) 1279 so->so_error = error; 1280 else 1281 tp->t_softerror = error; 1282 wakeup((caddr_t) &so->so_timeo); 1283 sorwakeup(so); 1284 sowwakeup(so); 1285 } 1286 #endif 1287 1288 #ifdef INET6 1289 void 1290 tcp6_ctlinput(cmd, sa, d) 1291 int cmd; 1292 struct sockaddr *sa; 1293 void *d; 1294 { 1295 struct tcphdr th; 1296 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1297 int nmatch; 1298 struct ip6_hdr *ip6; 1299 const struct sockaddr_in6 *sa6_src = NULL; 1300 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1301 struct mbuf *m; 1302 int off; 1303 1304 if (sa->sa_family != AF_INET6 || 1305 sa->sa_len != sizeof(struct sockaddr_in6)) 1306 return; 1307 if ((unsigned)cmd >= PRC_NCMDS) 1308 return; 1309 else if (cmd == PRC_QUENCH) { 1310 /* XXX there's no PRC_QUENCH in IPv6 */ 1311 notify = tcp6_quench; 1312 } else if (PRC_IS_REDIRECT(cmd)) 1313 notify = in6_rtchange, d = NULL; 1314 else if (cmd == PRC_MSGSIZE) 1315 ; /* special code is present, see below */ 1316 else if (cmd == PRC_HOSTDEAD) 1317 d = NULL; 1318 else if (inet6ctlerrmap[cmd] == 0) 1319 return; 1320 1321 /* if the parameter is from icmp6, decode it. */ 1322 if (d != NULL) { 1323 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1324 m = ip6cp->ip6c_m; 1325 ip6 = ip6cp->ip6c_ip6; 1326 off = ip6cp->ip6c_off; 1327 sa6_src = ip6cp->ip6c_src; 1328 } else { 1329 m = NULL; 1330 ip6 = NULL; 1331 sa6_src = &sa6_any; 1332 } 1333 1334 if (ip6) { 1335 /* 1336 * XXX: We assume that when ip6 is non NULL, 1337 * M and OFF are valid. 1338 */ 1339 1340 /* check if we can safely examine src and dst ports */ 1341 if (m->m_pkthdr.len < off + sizeof(th)) { 1342 if (cmd == PRC_MSGSIZE) 1343 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1344 return; 1345 } 1346 1347 bzero(&th, sizeof(th)); 1348 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1349 1350 if (cmd == PRC_MSGSIZE) { 1351 int valid = 0; 1352 1353 /* 1354 * Check to see if we have a valid TCP connection 1355 * corresponding to the address in the ICMPv6 message 1356 * payload. 1357 */ 1358 if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr, 1359 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1360 th.th_sport, 0)) 1361 valid++; 1362 1363 /* 1364 * Depending on the value of "valid" and routing table 1365 * size (mtudisc_{hi,lo}wat), we will: 1366 * - recalcurate the new MTU and create the 1367 * corresponding routing entry, or 1368 * - ignore the MTU change notification. 1369 */ 1370 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1371 1372 /* 1373 * no need to call in6_pcbnotify, it should have been 1374 * called via callback if necessary 1375 */ 1376 return; 1377 } 1378 1379 nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport, 1380 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1381 if (nmatch == 0 && syn_cache_count && 1382 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1383 inet6ctlerrmap[cmd] == ENETUNREACH || 1384 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1385 syn_cache_unreach((struct sockaddr *)sa6_src, 1386 sa, &th); 1387 } else { 1388 (void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src, 1389 0, cmd, NULL, notify); 1390 } 1391 } 1392 #endif 1393 1394 #ifdef INET 1395 /* assumes that ip header and tcp header are contiguous on mbuf */ 1396 void * 1397 tcp_ctlinput(cmd, sa, v) 1398 int cmd; 1399 struct sockaddr *sa; 1400 void *v; 1401 { 1402 struct ip *ip = v; 1403 struct tcphdr *th; 1404 struct icmp *icp; 1405 extern const int inetctlerrmap[]; 1406 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1407 int errno; 1408 int nmatch; 1409 1410 if (sa->sa_family != AF_INET || 1411 sa->sa_len != sizeof(struct sockaddr_in)) 1412 return NULL; 1413 if ((unsigned)cmd >= PRC_NCMDS) 1414 return NULL; 1415 errno = inetctlerrmap[cmd]; 1416 if (cmd == PRC_QUENCH) 1417 notify = tcp_quench; 1418 else if (PRC_IS_REDIRECT(cmd)) 1419 notify = in_rtchange, ip = 0; 1420 else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) { 1421 /* 1422 * Check to see if we have a valid TCP connection 1423 * corresponding to the address in the ICMP message 1424 * payload. 1425 * 1426 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1427 */ 1428 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1429 if (in_pcblookup_connect(&tcbtable, 1430 ip->ip_dst, th->th_dport, 1431 ip->ip_src, th->th_sport) == NULL) 1432 return NULL; 1433 1434 /* 1435 * Now that we've validated that we are actually communicating 1436 * with the host indicated in the ICMP message, locate the 1437 * ICMP header, recalculate the new MTU, and create the 1438 * corresponding routing entry. 1439 */ 1440 icp = (struct icmp *)((caddr_t)ip - 1441 offsetof(struct icmp, icmp_ip)); 1442 icmp_mtudisc(icp, ip->ip_dst); 1443 1444 return NULL; 1445 } else if (cmd == PRC_HOSTDEAD) 1446 ip = 0; 1447 else if (errno == 0) 1448 return NULL; 1449 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1450 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1451 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1452 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1453 if (nmatch == 0 && syn_cache_count && 1454 (inetctlerrmap[cmd] == EHOSTUNREACH || 1455 inetctlerrmap[cmd] == ENETUNREACH || 1456 inetctlerrmap[cmd] == EHOSTDOWN)) { 1457 struct sockaddr_in sin; 1458 bzero(&sin, sizeof(sin)); 1459 sin.sin_len = sizeof(sin); 1460 sin.sin_family = AF_INET; 1461 sin.sin_port = th->th_sport; 1462 sin.sin_addr = ip->ip_src; 1463 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1464 } 1465 1466 /* XXX mapped address case */ 1467 } else 1468 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1469 notify); 1470 return NULL; 1471 } 1472 1473 /* 1474 * When a source quence is received, we are being notifed of congestion. 1475 * Close the congestion window down to the Loss Window (one segment). 1476 * We will gradually open it again as we proceed. 1477 */ 1478 void 1479 tcp_quench(inp, errno) 1480 struct inpcb *inp; 1481 int errno; 1482 { 1483 struct tcpcb *tp = intotcpcb(inp); 1484 1485 if (tp) 1486 tp->snd_cwnd = tp->t_segsz; 1487 } 1488 #endif 1489 1490 #ifdef INET6 1491 void 1492 tcp6_quench(in6p, errno) 1493 struct in6pcb *in6p; 1494 int errno; 1495 { 1496 struct tcpcb *tp = in6totcpcb(in6p); 1497 1498 if (tp) 1499 tp->snd_cwnd = tp->t_segsz; 1500 } 1501 #endif 1502 1503 #ifdef INET 1504 /* 1505 * Path MTU Discovery handlers. 1506 */ 1507 void 1508 tcp_mtudisc_callback(faddr) 1509 struct in_addr faddr; 1510 { 1511 1512 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1513 } 1514 1515 /* 1516 * On receipt of path MTU corrections, flush old route and replace it 1517 * with the new one. Retransmit all unacknowledged packets, to ensure 1518 * that all packets will be received. 1519 */ 1520 void 1521 tcp_mtudisc(inp, errno) 1522 struct inpcb *inp; 1523 int errno; 1524 { 1525 struct tcpcb *tp = intotcpcb(inp); 1526 struct rtentry *rt = in_pcbrtentry(inp); 1527 1528 if (tp != 0) { 1529 if (rt != 0) { 1530 /* 1531 * If this was not a host route, remove and realloc. 1532 */ 1533 if ((rt->rt_flags & RTF_HOST) == 0) { 1534 in_rtchange(inp, errno); 1535 if ((rt = in_pcbrtentry(inp)) == 0) 1536 return; 1537 } 1538 1539 /* 1540 * Slow start out of the error condition. We 1541 * use the MTU because we know it's smaller 1542 * than the previously transmitted segment. 1543 * 1544 * Note: This is more conservative than the 1545 * suggestion in draft-floyd-incr-init-win-03. 1546 */ 1547 if (rt->rt_rmx.rmx_mtu != 0) 1548 tp->snd_cwnd = 1549 TCP_INITIAL_WINDOW(tcp_init_win, 1550 rt->rt_rmx.rmx_mtu); 1551 } 1552 1553 /* 1554 * Resend unacknowledged packets. 1555 */ 1556 tp->snd_nxt = tp->snd_una; 1557 tcp_output(tp); 1558 } 1559 } 1560 #endif 1561 1562 #ifdef INET6 1563 /* 1564 * Path MTU Discovery handlers. 1565 */ 1566 void 1567 tcp6_mtudisc_callback(faddr) 1568 struct in6_addr *faddr; 1569 { 1570 struct sockaddr_in6 sin6; 1571 1572 bzero(&sin6, sizeof(sin6)); 1573 sin6.sin6_family = AF_INET6; 1574 sin6.sin6_len = sizeof(struct sockaddr_in6); 1575 sin6.sin6_addr = *faddr; 1576 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1577 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1578 } 1579 1580 void 1581 tcp6_mtudisc(in6p, errno) 1582 struct in6pcb *in6p; 1583 int errno; 1584 { 1585 struct tcpcb *tp = in6totcpcb(in6p); 1586 struct rtentry *rt = in6_pcbrtentry(in6p); 1587 1588 if (tp != 0) { 1589 if (rt != 0) { 1590 /* 1591 * If this was not a host route, remove and realloc. 1592 */ 1593 if ((rt->rt_flags & RTF_HOST) == 0) { 1594 in6_rtchange(in6p, errno); 1595 if ((rt = in6_pcbrtentry(in6p)) == 0) 1596 return; 1597 } 1598 1599 /* 1600 * Slow start out of the error condition. We 1601 * use the MTU because we know it's smaller 1602 * than the previously transmitted segment. 1603 * 1604 * Note: This is more conservative than the 1605 * suggestion in draft-floyd-incr-init-win-03. 1606 */ 1607 if (rt->rt_rmx.rmx_mtu != 0) 1608 tp->snd_cwnd = 1609 TCP_INITIAL_WINDOW(tcp_init_win, 1610 rt->rt_rmx.rmx_mtu); 1611 } 1612 1613 /* 1614 * Resend unacknowledged packets. 1615 */ 1616 tp->snd_nxt = tp->snd_una; 1617 tcp_output(tp); 1618 } 1619 } 1620 #endif /* INET6 */ 1621 1622 /* 1623 * Compute the MSS to advertise to the peer. Called only during 1624 * the 3-way handshake. If we are the server (peer initiated 1625 * connection), we are called with a pointer to the interface 1626 * on which the SYN packet arrived. If we are the client (we 1627 * initiated connection), we are called with a pointer to the 1628 * interface out which this connection should go. 1629 * 1630 * NOTE: Do not subtract IP option/extension header size nor IPsec 1631 * header size from MSS advertisement. MSS option must hold the maximum 1632 * segment size we can accept, so it must always be: 1633 * max(if mtu) - ip header - tcp header 1634 */ 1635 u_long 1636 tcp_mss_to_advertise(ifp, af) 1637 const struct ifnet *ifp; 1638 int af; 1639 { 1640 extern u_long in_maxmtu; 1641 u_long mss = 0; 1642 u_long hdrsiz; 1643 1644 /* 1645 * In order to avoid defeating path MTU discovery on the peer, 1646 * we advertise the max MTU of all attached networks as our MSS, 1647 * per RFC 1191, section 3.1. 1648 * 1649 * We provide the option to advertise just the MTU of 1650 * the interface on which we hope this connection will 1651 * be receiving. If we are responding to a SYN, we 1652 * will have a pretty good idea about this, but when 1653 * initiating a connection there is a bit more doubt. 1654 * 1655 * We also need to ensure that loopback has a large enough 1656 * MSS, as the loopback MTU is never included in in_maxmtu. 1657 */ 1658 1659 if (ifp != NULL) 1660 mss = ifp->if_mtu; 1661 1662 if (tcp_mss_ifmtu == 0) 1663 switch (af) { 1664 case AF_INET: 1665 mss = max(in_maxmtu, mss); 1666 break; 1667 #ifdef INET6 1668 case AF_INET6: 1669 mss = max(in6_maxmtu, mss); 1670 break; 1671 #endif 1672 } 1673 1674 switch (af) { 1675 case AF_INET: 1676 hdrsiz = sizeof(struct ip); 1677 break; 1678 #ifdef INET6 1679 case AF_INET6: 1680 hdrsiz = sizeof(struct ip6_hdr); 1681 break; 1682 #endif 1683 default: 1684 hdrsiz = 0; 1685 break; 1686 } 1687 hdrsiz += sizeof(struct tcphdr); 1688 if (mss > hdrsiz) 1689 mss -= hdrsiz; 1690 1691 mss = max(tcp_mssdflt, mss); 1692 return (mss); 1693 } 1694 1695 /* 1696 * Set connection variables based on the peer's advertised MSS. 1697 * We are passed the TCPCB for the actual connection. If we 1698 * are the server, we are called by the compressed state engine 1699 * when the 3-way handshake is complete. If we are the client, 1700 * we are called when we receive the SYN,ACK from the server. 1701 * 1702 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1703 * before this routine is called! 1704 */ 1705 void 1706 tcp_mss_from_peer(tp, offer) 1707 struct tcpcb *tp; 1708 int offer; 1709 { 1710 struct socket *so; 1711 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1712 struct rtentry *rt; 1713 #endif 1714 u_long bufsize; 1715 int mss; 1716 1717 #ifdef DIAGNOSTIC 1718 if (tp->t_inpcb && tp->t_in6pcb) 1719 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1720 #endif 1721 so = NULL; 1722 rt = NULL; 1723 #ifdef INET 1724 if (tp->t_inpcb) { 1725 so = tp->t_inpcb->inp_socket; 1726 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1727 rt = in_pcbrtentry(tp->t_inpcb); 1728 #endif 1729 } 1730 #endif 1731 #ifdef INET6 1732 if (tp->t_in6pcb) { 1733 so = tp->t_in6pcb->in6p_socket; 1734 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1735 rt = in6_pcbrtentry(tp->t_in6pcb); 1736 #endif 1737 } 1738 #endif 1739 1740 /* 1741 * As per RFC1122, use the default MSS value, unless they 1742 * sent us an offer. Do not accept offers less than 32 bytes. 1743 */ 1744 mss = tcp_mssdflt; 1745 if (offer) 1746 mss = offer; 1747 mss = max(mss, 32); /* sanity */ 1748 tp->t_peermss = mss; 1749 mss -= tcp_optlen(tp); 1750 #ifdef INET 1751 if (tp->t_inpcb) 1752 mss -= ip_optlen(tp->t_inpcb); 1753 #endif 1754 #ifdef INET6 1755 if (tp->t_in6pcb) 1756 mss -= ip6_optlen(tp->t_in6pcb); 1757 #endif 1758 1759 /* 1760 * If there's a pipesize, change the socket buffer to that size. 1761 * Make the socket buffer an integral number of MSS units. If 1762 * the MSS is larger than the socket buffer, artificially decrease 1763 * the MSS. 1764 */ 1765 #ifdef RTV_SPIPE 1766 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1767 bufsize = rt->rt_rmx.rmx_sendpipe; 1768 else 1769 #endif 1770 bufsize = so->so_snd.sb_hiwat; 1771 if (bufsize < mss) 1772 mss = bufsize; 1773 else { 1774 bufsize = roundup(bufsize, mss); 1775 if (bufsize > sb_max) 1776 bufsize = sb_max; 1777 (void) sbreserve(&so->so_snd, bufsize); 1778 } 1779 tp->t_segsz = mss; 1780 1781 #ifdef RTV_SSTHRESH 1782 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1783 /* 1784 * There's some sort of gateway or interface buffer 1785 * limit on the path. Use this to set the slow 1786 * start threshold, but set the threshold to no less 1787 * than 2 * MSS. 1788 */ 1789 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1790 } 1791 #endif 1792 } 1793 1794 /* 1795 * Processing necessary when a TCP connection is established. 1796 */ 1797 void 1798 tcp_established(tp) 1799 struct tcpcb *tp; 1800 { 1801 struct socket *so; 1802 #ifdef RTV_RPIPE 1803 struct rtentry *rt; 1804 #endif 1805 u_long bufsize; 1806 1807 #ifdef DIAGNOSTIC 1808 if (tp->t_inpcb && tp->t_in6pcb) 1809 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1810 #endif 1811 so = NULL; 1812 rt = NULL; 1813 #ifdef INET 1814 if (tp->t_inpcb) { 1815 so = tp->t_inpcb->inp_socket; 1816 #if defined(RTV_RPIPE) 1817 rt = in_pcbrtentry(tp->t_inpcb); 1818 #endif 1819 } 1820 #endif 1821 #ifdef INET6 1822 if (tp->t_in6pcb) { 1823 so = tp->t_in6pcb->in6p_socket; 1824 #if defined(RTV_RPIPE) 1825 rt = in6_pcbrtentry(tp->t_in6pcb); 1826 #endif 1827 } 1828 #endif 1829 1830 tp->t_state = TCPS_ESTABLISHED; 1831 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1832 1833 #ifdef RTV_RPIPE 1834 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1835 bufsize = rt->rt_rmx.rmx_recvpipe; 1836 else 1837 #endif 1838 bufsize = so->so_rcv.sb_hiwat; 1839 if (bufsize > tp->t_ourmss) { 1840 bufsize = roundup(bufsize, tp->t_ourmss); 1841 if (bufsize > sb_max) 1842 bufsize = sb_max; 1843 (void) sbreserve(&so->so_rcv, bufsize); 1844 } 1845 } 1846 1847 /* 1848 * Check if there's an initial rtt or rttvar. Convert from the 1849 * route-table units to scaled multiples of the slow timeout timer. 1850 * Called only during the 3-way handshake. 1851 */ 1852 void 1853 tcp_rmx_rtt(tp) 1854 struct tcpcb *tp; 1855 { 1856 #ifdef RTV_RTT 1857 struct rtentry *rt = NULL; 1858 int rtt; 1859 1860 #ifdef DIAGNOSTIC 1861 if (tp->t_inpcb && tp->t_in6pcb) 1862 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1863 #endif 1864 #ifdef INET 1865 if (tp->t_inpcb) 1866 rt = in_pcbrtentry(tp->t_inpcb); 1867 #endif 1868 #ifdef INET6 1869 if (tp->t_in6pcb) 1870 rt = in6_pcbrtentry(tp->t_in6pcb); 1871 #endif 1872 if (rt == NULL) 1873 return; 1874 1875 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1876 /* 1877 * XXX The lock bit for MTU indicates that the value 1878 * is also a minimum value; this is subject to time. 1879 */ 1880 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1881 TCPT_RANGESET(tp->t_rttmin, 1882 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1883 TCPTV_MIN, TCPTV_REXMTMAX); 1884 tp->t_srtt = rtt / 1885 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1886 if (rt->rt_rmx.rmx_rttvar) { 1887 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1888 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1889 (TCP_RTTVAR_SHIFT + 2)); 1890 } else { 1891 /* Default variation is +- 1 rtt */ 1892 tp->t_rttvar = 1893 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1894 } 1895 TCPT_RANGESET(tp->t_rxtcur, 1896 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1897 tp->t_rttmin, TCPTV_REXMTMAX); 1898 } 1899 #endif 1900 } 1901 1902 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1903 #if NRND > 0 1904 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1905 #endif 1906 1907 /* 1908 * Get a new sequence value given a tcp control block 1909 */ 1910 tcp_seq 1911 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1912 { 1913 1914 #ifdef INET 1915 if (tp->t_inpcb != NULL) { 1916 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1917 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1918 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1919 addin)); 1920 } 1921 #endif 1922 #ifdef INET6 1923 if (tp->t_in6pcb != NULL) { 1924 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 1925 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 1926 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 1927 addin)); 1928 } 1929 #endif 1930 /* Not possible. */ 1931 panic("tcp_new_iss"); 1932 } 1933 1934 /* 1935 * This routine actually generates a new TCP initial sequence number. 1936 */ 1937 tcp_seq 1938 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 1939 size_t addrsz, tcp_seq addin) 1940 { 1941 tcp_seq tcp_iss; 1942 1943 #if NRND > 0 1944 static int beenhere; 1945 1946 /* 1947 * If we haven't been here before, initialize our cryptographic 1948 * hash secret. 1949 */ 1950 if (beenhere == 0) { 1951 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 1952 RND_EXTRACT_ANY); 1953 beenhere = 1; 1954 } 1955 1956 if (tcp_do_rfc1948) { 1957 MD5_CTX ctx; 1958 u_int8_t hash[16]; /* XXX MD5 knowledge */ 1959 1960 /* 1961 * Compute the base value of the ISS. It is a hash 1962 * of (saddr, sport, daddr, dport, secret). 1963 */ 1964 MD5Init(&ctx); 1965 1966 MD5Update(&ctx, (u_char *) laddr, addrsz); 1967 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 1968 1969 MD5Update(&ctx, (u_char *) faddr, addrsz); 1970 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 1971 1972 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 1973 1974 MD5Final(hash, &ctx); 1975 1976 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 1977 1978 /* 1979 * Now increment our "timer", and add it in to 1980 * the computed value. 1981 * 1982 * XXX Use `addin'? 1983 * XXX TCP_ISSINCR too large to use? 1984 */ 1985 tcp_iss_seq += TCP_ISSINCR; 1986 #ifdef TCPISS_DEBUG 1987 printf("ISS hash 0x%08x, ", tcp_iss); 1988 #endif 1989 tcp_iss += tcp_iss_seq + addin; 1990 #ifdef TCPISS_DEBUG 1991 printf("new ISS 0x%08x\n", tcp_iss); 1992 #endif 1993 } else 1994 #endif /* NRND > 0 */ 1995 { 1996 /* 1997 * Randomize. 1998 */ 1999 #if NRND > 0 2000 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2001 #else 2002 tcp_iss = random(); 2003 #endif 2004 2005 /* 2006 * If we were asked to add some amount to a known value, 2007 * we will take a random value obtained above, mask off 2008 * the upper bits, and add in the known value. We also 2009 * add in a constant to ensure that we are at least a 2010 * certain distance from the original value. 2011 * 2012 * This is used when an old connection is in timed wait 2013 * and we have a new one coming in, for instance. 2014 */ 2015 if (addin != 0) { 2016 #ifdef TCPISS_DEBUG 2017 printf("Random %08x, ", tcp_iss); 2018 #endif 2019 tcp_iss &= TCP_ISS_RANDOM_MASK; 2020 tcp_iss += addin + TCP_ISSINCR; 2021 #ifdef TCPISS_DEBUG 2022 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2023 #endif 2024 } else { 2025 tcp_iss &= TCP_ISS_RANDOM_MASK; 2026 tcp_iss += tcp_iss_seq; 2027 tcp_iss_seq += TCP_ISSINCR; 2028 #ifdef TCPISS_DEBUG 2029 printf("ISS %08x\n", tcp_iss); 2030 #endif 2031 } 2032 } 2033 2034 if (tcp_compat_42) { 2035 /* 2036 * Limit it to the positive range for really old TCP 2037 * implementations. 2038 */ 2039 if (tcp_iss >= 0x80000000) 2040 tcp_iss &= 0x7fffffff; /* XXX */ 2041 } 2042 2043 return (tcp_iss); 2044 } 2045 2046 #ifdef IPSEC 2047 /* compute ESP/AH header size for TCP, including outer IP header. */ 2048 size_t 2049 ipsec4_hdrsiz_tcp(tp) 2050 struct tcpcb *tp; 2051 { 2052 struct inpcb *inp; 2053 size_t hdrsiz; 2054 2055 /* XXX mapped addr case (tp->t_in6pcb) */ 2056 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2057 return 0; 2058 switch (tp->t_family) { 2059 case AF_INET: 2060 /* XXX: should use currect direction. */ 2061 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2062 break; 2063 default: 2064 hdrsiz = 0; 2065 break; 2066 } 2067 2068 return hdrsiz; 2069 } 2070 2071 #ifdef INET6 2072 size_t 2073 ipsec6_hdrsiz_tcp(tp) 2074 struct tcpcb *tp; 2075 { 2076 struct in6pcb *in6p; 2077 size_t hdrsiz; 2078 2079 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2080 return 0; 2081 switch (tp->t_family) { 2082 case AF_INET6: 2083 /* XXX: should use currect direction. */ 2084 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2085 break; 2086 case AF_INET: 2087 /* mapped address case - tricky */ 2088 default: 2089 hdrsiz = 0; 2090 break; 2091 } 2092 2093 return hdrsiz; 2094 } 2095 #endif 2096 #endif /*IPSEC*/ 2097 2098 /* 2099 * Determine the length of the TCP options for this connection. 2100 * 2101 * XXX: What do we do for SACK, when we add that? Just reserve 2102 * all of the space? Otherwise we can't exactly be incrementing 2103 * cwnd by an amount that varies depending on the amount we last 2104 * had to SACK! 2105 */ 2106 2107 u_int 2108 tcp_optlen(tp) 2109 struct tcpcb *tp; 2110 { 2111 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2112 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2113 return TCPOLEN_TSTAMP_APPA; 2114 else 2115 return 0; 2116 } 2117
Indexes created Sun Sep 21 20:09:37 GMT 2025