tcp_subr.c revision 1.147
1 /* $NetBSD: tcp_subr.c,v 1.147 2003/08/22 20:20:11 jonathan 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. Neither the name of the University nor the names of its contributors 82 * may be used to endorse or promote products derived from this software 83 * without specific prior written permission. 84 * 85 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 86 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 87 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 88 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 89 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 90 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 91 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 92 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 93 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 94 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 95 * SUCH DAMAGE. 96 * 97 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 98 */ 99 100 #include <sys/cdefs.h> 101 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.147 2003/08/22 20:20:11 jonathan Exp $"); 102 103 #include "opt_inet.h" 104 #include "opt_ipsec.h" 105 #include "opt_tcp_compat_42.h" 106 #include "opt_inet_csum.h" 107 #include "opt_mbuftrace.h" 108 #include "rnd.h" 109 110 #include <sys/param.h> 111 #include <sys/proc.h> 112 #include <sys/systm.h> 113 #include <sys/malloc.h> 114 #include <sys/mbuf.h> 115 #include <sys/socket.h> 116 #include <sys/socketvar.h> 117 #include <sys/protosw.h> 118 #include <sys/errno.h> 119 #include <sys/kernel.h> 120 #include <sys/pool.h> 121 #if NRND > 0 122 #include <sys/md5.h> 123 #include <sys/rnd.h> 124 #endif 125 126 #include <net/route.h> 127 #include <net/if.h> 128 129 #include <netinet/in.h> 130 #include <netinet/in_systm.h> 131 #include <netinet/ip.h> 132 #include <netinet/in_pcb.h> 133 #include <netinet/ip_var.h> 134 #include <netinet/ip_icmp.h> 135 136 #ifdef INET6 137 #ifndef INET 138 #include <netinet/in.h> 139 #endif 140 #include <netinet/ip6.h> 141 #include <netinet6/in6_pcb.h> 142 #include <netinet6/ip6_var.h> 143 #include <netinet6/in6_var.h> 144 #include <netinet6/ip6protosw.h> 145 #include <netinet/icmp6.h> 146 #include <netinet6/nd6.h> 147 #endif 148 149 #include <netinet/tcp.h> 150 #include <netinet/tcp_fsm.h> 151 #include <netinet/tcp_seq.h> 152 #include <netinet/tcp_timer.h> 153 #include <netinet/tcp_var.h> 154 #include <netinet/tcpip.h> 155 156 #ifdef IPSEC 157 #include <netinet6/ipsec.h> 158 #endif /*IPSEC*/ 159 160 #ifdef FAST_IPSEC 161 #include <netipsec/ipsec.h> 162 #ifdef INET6 163 #include <netipsec/ipsec6.h> 164 #endif 165 #endif /* FAST_IPSEC*/ 166 167 168 #ifdef INET6 169 struct in6pcb tcb6; 170 #endif 171 172 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 173 struct tcpstat tcpstat; /* tcp statistics */ 174 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 175 176 /* patchable/settable parameters for tcp */ 177 int tcp_mssdflt = TCP_MSS; 178 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 179 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 180 #if NRND > 0 181 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 182 #endif 183 int tcp_do_sack = 1; /* selective acknowledgement */ 184 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 185 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 186 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 187 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 188 #ifndef TCP_INIT_WIN 189 #define TCP_INIT_WIN 1 /* initial slow start window */ 190 #endif 191 #ifndef TCP_INIT_WIN_LOCAL 192 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 193 #endif 194 int tcp_init_win = TCP_INIT_WIN; 195 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 196 int tcp_mss_ifmtu = 0; 197 #ifdef TCP_COMPAT_42 198 int tcp_compat_42 = 1; 199 #else 200 int tcp_compat_42 = 0; 201 #endif 202 int tcp_rst_ppslim = 100; /* 100pps */ 203 204 /* tcb hash */ 205 #ifndef TCBHASHSIZE 206 #define TCBHASHSIZE 128 207 #endif 208 int tcbhashsize = TCBHASHSIZE; 209 210 /* syn hash parameters */ 211 #define TCP_SYN_HASH_SIZE 293 212 #define TCP_SYN_BUCKET_SIZE 35 213 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 214 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 215 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 216 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 217 218 int tcp_freeq __P((struct tcpcb *)); 219 220 #ifdef INET 221 void tcp_mtudisc_callback __P((struct in_addr)); 222 #endif 223 #ifdef INET6 224 void tcp6_mtudisc_callback __P((struct in6_addr *)); 225 #endif 226 227 void tcp_mtudisc __P((struct inpcb *, int)); 228 #ifdef INET6 229 void tcp6_mtudisc __P((struct in6pcb *, int)); 230 #endif 231 232 struct pool tcpcb_pool; 233 234 #ifdef TCP_CSUM_COUNTERS 235 #include <sys/device.h> 236 237 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 238 NULL, "tcp", "hwcsum bad"); 239 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 240 NULL, "tcp", "hwcsum ok"); 241 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "hwcsum data"); 243 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "swcsum"); 245 #endif /* TCP_CSUM_COUNTERS */ 246 247 #ifdef TCP_OUTPUT_COUNTERS 248 #include <sys/device.h> 249 250 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp", "output big header"); 252 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 NULL, "tcp", "output copy small"); 254 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 255 NULL, "tcp", "output copy big"); 256 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 257 NULL, "tcp", "output reference big"); 258 #endif /* TCP_OUTPUT_COUNTERS */ 259 260 #ifdef TCP_REASS_COUNTERS 261 #include <sys/device.h> 262 263 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp_reass", "calls"); 265 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 &tcp_reass_, "tcp_reass", "insert into empty queue"); 267 struct evcnt tcp_reass_iteration[8] = { 268 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 269 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 270 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 271 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 272 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 273 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 274 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 275 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 276 }; 277 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 278 &tcp_reass_, "tcp_reass", "prepend to first"); 279 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 &tcp_reass_, "tcp_reass", "prepend"); 281 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 &tcp_reass_, "tcp_reass", "insert"); 283 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 &tcp_reass_, "tcp_reass", "insert at tail"); 285 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 &tcp_reass_, "tcp_reass", "append"); 287 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 &tcp_reass_, "tcp_reass", "append to tail fragment"); 289 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 &tcp_reass_, "tcp_reass", "overlap at end"); 291 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 &tcp_reass_, "tcp_reass", "overlap at start"); 293 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 294 &tcp_reass_, "tcp_reass", "duplicate segment"); 295 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 296 &tcp_reass_, "tcp_reass", "duplicate fragment"); 297 298 #endif /* TCP_REASS_COUNTERS */ 299 300 #ifdef MBUFTRACE 301 struct mowner tcp_mowner = { "tcp" }; 302 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 303 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 304 #endif 305 306 /* 307 * Tcp initialization 308 */ 309 void 310 tcp_init() 311 { 312 int hlen; 313 314 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 315 NULL); 316 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 317 #ifdef INET6 318 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 319 #endif 320 321 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 322 #ifdef INET6 323 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 324 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 325 #endif 326 if (max_protohdr < hlen) 327 max_protohdr = hlen; 328 if (max_linkhdr + hlen > MHLEN) 329 panic("tcp_init"); 330 331 #ifdef INET 332 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 333 #endif 334 #ifdef INET6 335 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 336 #endif 337 338 /* Initialize timer state. */ 339 tcp_timer_init(); 340 341 /* Initialize the compressed state engine. */ 342 syn_cache_init(); 343 344 #ifdef TCP_CSUM_COUNTERS 345 evcnt_attach_static(&tcp_hwcsum_bad); 346 evcnt_attach_static(&tcp_hwcsum_ok); 347 evcnt_attach_static(&tcp_hwcsum_data); 348 evcnt_attach_static(&tcp_swcsum); 349 #endif /* TCP_CSUM_COUNTERS */ 350 351 #ifdef TCP_OUTPUT_COUNTERS 352 evcnt_attach_static(&tcp_output_bigheader); 353 evcnt_attach_static(&tcp_output_copysmall); 354 evcnt_attach_static(&tcp_output_copybig); 355 evcnt_attach_static(&tcp_output_refbig); 356 #endif /* TCP_OUTPUT_COUNTERS */ 357 358 #ifdef TCP_REASS_COUNTERS 359 evcnt_attach_static(&tcp_reass_); 360 evcnt_attach_static(&tcp_reass_empty); 361 evcnt_attach_static(&tcp_reass_iteration[0]); 362 evcnt_attach_static(&tcp_reass_iteration[1]); 363 evcnt_attach_static(&tcp_reass_iteration[2]); 364 evcnt_attach_static(&tcp_reass_iteration[3]); 365 evcnt_attach_static(&tcp_reass_iteration[4]); 366 evcnt_attach_static(&tcp_reass_iteration[5]); 367 evcnt_attach_static(&tcp_reass_iteration[6]); 368 evcnt_attach_static(&tcp_reass_iteration[7]); 369 evcnt_attach_static(&tcp_reass_prependfirst); 370 evcnt_attach_static(&tcp_reass_prepend); 371 evcnt_attach_static(&tcp_reass_insert); 372 evcnt_attach_static(&tcp_reass_inserttail); 373 evcnt_attach_static(&tcp_reass_append); 374 evcnt_attach_static(&tcp_reass_appendtail); 375 evcnt_attach_static(&tcp_reass_overlaptail); 376 evcnt_attach_static(&tcp_reass_overlapfront); 377 evcnt_attach_static(&tcp_reass_segdup); 378 evcnt_attach_static(&tcp_reass_fragdup); 379 #endif /* TCP_REASS_COUNTERS */ 380 381 MOWNER_ATTACH(&tcp_tx_mowner); 382 MOWNER_ATTACH(&tcp_rx_mowner); 383 MOWNER_ATTACH(&tcp_mowner); 384 } 385 386 /* 387 * Create template to be used to send tcp packets on a connection. 388 * Call after host entry created, allocates an mbuf and fills 389 * in a skeletal tcp/ip header, minimizing the amount of work 390 * necessary when the connection is used. 391 */ 392 struct mbuf * 393 tcp_template(tp) 394 struct tcpcb *tp; 395 { 396 struct inpcb *inp = tp->t_inpcb; 397 #ifdef INET6 398 struct in6pcb *in6p = tp->t_in6pcb; 399 #endif 400 struct tcphdr *n; 401 struct mbuf *m; 402 int hlen; 403 404 switch (tp->t_family) { 405 case AF_INET: 406 hlen = sizeof(struct ip); 407 if (inp) 408 break; 409 #ifdef INET6 410 if (in6p) { 411 /* mapped addr case */ 412 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 413 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 414 break; 415 } 416 #endif 417 return NULL; /*EINVAL*/ 418 #ifdef INET6 419 case AF_INET6: 420 hlen = sizeof(struct ip6_hdr); 421 if (in6p) { 422 /* more sainty check? */ 423 break; 424 } 425 return NULL; /*EINVAL*/ 426 #endif 427 default: 428 hlen = 0; /*pacify gcc*/ 429 return NULL; /*EAFNOSUPPORT*/ 430 } 431 #ifdef DIAGNOSTIC 432 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 433 panic("mclbytes too small for t_template"); 434 #endif 435 m = tp->t_template; 436 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 437 ; 438 else { 439 if (m) 440 m_freem(m); 441 m = tp->t_template = NULL; 442 MGETHDR(m, M_DONTWAIT, MT_HEADER); 443 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 444 MCLGET(m, M_DONTWAIT); 445 if ((m->m_flags & M_EXT) == 0) { 446 m_free(m); 447 m = NULL; 448 } 449 } 450 if (m == NULL) 451 return NULL; 452 MCLAIM(m, &tcp_mowner); 453 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 454 } 455 456 bzero(mtod(m, caddr_t), m->m_len); 457 458 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 459 460 switch (tp->t_family) { 461 case AF_INET: 462 { 463 struct ipovly *ipov; 464 mtod(m, struct ip *)->ip_v = 4; 465 ipov = mtod(m, struct ipovly *); 466 ipov->ih_pr = IPPROTO_TCP; 467 ipov->ih_len = htons(sizeof(struct tcphdr)); 468 if (inp) { 469 ipov->ih_src = inp->inp_laddr; 470 ipov->ih_dst = inp->inp_faddr; 471 } 472 #ifdef INET6 473 else if (in6p) { 474 /* mapped addr case */ 475 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 476 sizeof(ipov->ih_src)); 477 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 478 sizeof(ipov->ih_dst)); 479 } 480 #endif 481 /* 482 * Compute the pseudo-header portion of the checksum 483 * now. We incrementally add in the TCP option and 484 * payload lengths later, and then compute the TCP 485 * checksum right before the packet is sent off onto 486 * the wire. 487 */ 488 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 489 ipov->ih_dst.s_addr, 490 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 491 break; 492 } 493 #ifdef INET6 494 case AF_INET6: 495 { 496 struct ip6_hdr *ip6; 497 mtod(m, struct ip *)->ip_v = 6; 498 ip6 = mtod(m, struct ip6_hdr *); 499 ip6->ip6_nxt = IPPROTO_TCP; 500 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 501 ip6->ip6_src = in6p->in6p_laddr; 502 ip6->ip6_dst = in6p->in6p_faddr; 503 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 504 if (ip6_auto_flowlabel) { 505 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 506 ip6->ip6_flow |= 507 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 508 } 509 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 510 ip6->ip6_vfc |= IPV6_VERSION; 511 512 /* 513 * Compute the pseudo-header portion of the checksum 514 * now. We incrementally add in the TCP option and 515 * payload lengths later, and then compute the TCP 516 * checksum right before the packet is sent off onto 517 * the wire. 518 */ 519 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 520 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 521 htonl(IPPROTO_TCP)); 522 break; 523 } 524 #endif 525 } 526 if (inp) { 527 n->th_sport = inp->inp_lport; 528 n->th_dport = inp->inp_fport; 529 } 530 #ifdef INET6 531 else if (in6p) { 532 n->th_sport = in6p->in6p_lport; 533 n->th_dport = in6p->in6p_fport; 534 } 535 #endif 536 n->th_seq = 0; 537 n->th_ack = 0; 538 n->th_x2 = 0; 539 n->th_off = 5; 540 n->th_flags = 0; 541 n->th_win = 0; 542 n->th_urp = 0; 543 return (m); 544 } 545 546 /* 547 * Send a single message to the TCP at address specified by 548 * the given TCP/IP header. If m == 0, then we make a copy 549 * of the tcpiphdr at ti and send directly to the addressed host. 550 * This is used to force keep alive messages out using the TCP 551 * template for a connection tp->t_template. If flags are given 552 * then we send a message back to the TCP which originated the 553 * segment ti, and discard the mbuf containing it and any other 554 * attached mbufs. 555 * 556 * In any case the ack and sequence number of the transmitted 557 * segment are as specified by the parameters. 558 */ 559 int 560 tcp_respond(tp, template, m, th0, ack, seq, flags) 561 struct tcpcb *tp; 562 struct mbuf *template; 563 struct mbuf *m; 564 struct tcphdr *th0; 565 tcp_seq ack, seq; 566 int flags; 567 { 568 struct route *ro; 569 int error, tlen, win = 0; 570 int hlen; 571 struct ip *ip; 572 #ifdef INET6 573 struct ip6_hdr *ip6; 574 #endif 575 int family; /* family on packet, not inpcb/in6pcb! */ 576 struct tcphdr *th; 577 578 if (tp != NULL && (flags & TH_RST) == 0) { 579 #ifdef DIAGNOSTIC 580 if (tp->t_inpcb && tp->t_in6pcb) 581 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 582 #endif 583 #ifdef INET 584 if (tp->t_inpcb) 585 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 586 #endif 587 #ifdef INET6 588 if (tp->t_in6pcb) 589 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 590 #endif 591 } 592 593 th = NULL; /* Quell uninitialized warning */ 594 ip = NULL; 595 #ifdef INET6 596 ip6 = NULL; 597 #endif 598 if (m == 0) { 599 if (!template) 600 return EINVAL; 601 602 /* get family information from template */ 603 switch (mtod(template, struct ip *)->ip_v) { 604 case 4: 605 family = AF_INET; 606 hlen = sizeof(struct ip); 607 break; 608 #ifdef INET6 609 case 6: 610 family = AF_INET6; 611 hlen = sizeof(struct ip6_hdr); 612 break; 613 #endif 614 default: 615 return EAFNOSUPPORT; 616 } 617 618 MGETHDR(m, M_DONTWAIT, MT_HEADER); 619 if (m) { 620 MCLAIM(m, &tcp_tx_mowner); 621 MCLGET(m, M_DONTWAIT); 622 if ((m->m_flags & M_EXT) == 0) { 623 m_free(m); 624 m = NULL; 625 } 626 } 627 if (m == NULL) 628 return (ENOBUFS); 629 630 if (tcp_compat_42) 631 tlen = 1; 632 else 633 tlen = 0; 634 635 m->m_data += max_linkhdr; 636 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 637 template->m_len); 638 switch (family) { 639 case AF_INET: 640 ip = mtod(m, struct ip *); 641 th = (struct tcphdr *)(ip + 1); 642 break; 643 #ifdef INET6 644 case AF_INET6: 645 ip6 = mtod(m, struct ip6_hdr *); 646 th = (struct tcphdr *)(ip6 + 1); 647 break; 648 #endif 649 #if 0 650 default: 651 /* noone will visit here */ 652 m_freem(m); 653 return EAFNOSUPPORT; 654 #endif 655 } 656 flags = TH_ACK; 657 } else { 658 659 if ((m->m_flags & M_PKTHDR) == 0) { 660 #if 0 661 printf("non PKTHDR to tcp_respond\n"); 662 #endif 663 m_freem(m); 664 return EINVAL; 665 } 666 #ifdef DIAGNOSTIC 667 if (!th0) 668 panic("th0 == NULL in tcp_respond"); 669 #endif 670 671 /* get family information from m */ 672 switch (mtod(m, struct ip *)->ip_v) { 673 case 4: 674 family = AF_INET; 675 hlen = sizeof(struct ip); 676 ip = mtod(m, struct ip *); 677 break; 678 #ifdef INET6 679 case 6: 680 family = AF_INET6; 681 hlen = sizeof(struct ip6_hdr); 682 ip6 = mtod(m, struct ip6_hdr *); 683 break; 684 #endif 685 default: 686 m_freem(m); 687 return EAFNOSUPPORT; 688 } 689 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 690 tlen = sizeof(*th0); 691 else 692 tlen = th0->th_off << 2; 693 694 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 695 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 696 m->m_len = hlen + tlen; 697 m_freem(m->m_next); 698 m->m_next = NULL; 699 } else { 700 struct mbuf *n; 701 702 #ifdef DIAGNOSTIC 703 if (max_linkhdr + hlen + tlen > MCLBYTES) { 704 m_freem(m); 705 return EMSGSIZE; 706 } 707 #endif 708 MGETHDR(n, M_DONTWAIT, MT_HEADER); 709 if (n && max_linkhdr + hlen + tlen > MHLEN) { 710 MCLGET(n, M_DONTWAIT); 711 if ((n->m_flags & M_EXT) == 0) { 712 m_freem(n); 713 n = NULL; 714 } 715 } 716 if (!n) { 717 m_freem(m); 718 return ENOBUFS; 719 } 720 721 MCLAIM(n, &tcp_tx_mowner); 722 n->m_data += max_linkhdr; 723 n->m_len = hlen + tlen; 724 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 725 m_copyback(n, hlen, tlen, (caddr_t)th0); 726 727 m_freem(m); 728 m = n; 729 n = NULL; 730 } 731 732 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 733 switch (family) { 734 case AF_INET: 735 ip = mtod(m, struct ip *); 736 th = (struct tcphdr *)(ip + 1); 737 ip->ip_p = IPPROTO_TCP; 738 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 739 ip->ip_p = IPPROTO_TCP; 740 break; 741 #ifdef INET6 742 case AF_INET6: 743 ip6 = mtod(m, struct ip6_hdr *); 744 th = (struct tcphdr *)(ip6 + 1); 745 ip6->ip6_nxt = IPPROTO_TCP; 746 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 747 ip6->ip6_nxt = IPPROTO_TCP; 748 break; 749 #endif 750 #if 0 751 default: 752 /* noone will visit here */ 753 m_freem(m); 754 return EAFNOSUPPORT; 755 #endif 756 } 757 xchg(th->th_dport, th->th_sport, u_int16_t); 758 #undef xchg 759 tlen = 0; /*be friendly with the following code*/ 760 } 761 th->th_seq = htonl(seq); 762 th->th_ack = htonl(ack); 763 th->th_x2 = 0; 764 if ((flags & TH_SYN) == 0) { 765 if (tp) 766 win >>= tp->rcv_scale; 767 if (win > TCP_MAXWIN) 768 win = TCP_MAXWIN; 769 th->th_win = htons((u_int16_t)win); 770 th->th_off = sizeof (struct tcphdr) >> 2; 771 tlen += sizeof(*th); 772 } else 773 tlen += th->th_off << 2; 774 m->m_len = hlen + tlen; 775 m->m_pkthdr.len = hlen + tlen; 776 m->m_pkthdr.rcvif = (struct ifnet *) 0; 777 th->th_flags = flags; 778 th->th_urp = 0; 779 780 switch (family) { 781 #ifdef INET 782 case AF_INET: 783 { 784 struct ipovly *ipov = (struct ipovly *)ip; 785 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 786 ipov->ih_len = htons((u_int16_t)tlen); 787 788 th->th_sum = 0; 789 th->th_sum = in_cksum(m, hlen + tlen); 790 ip->ip_len = htons(hlen + tlen); 791 ip->ip_ttl = ip_defttl; 792 break; 793 } 794 #endif 795 #ifdef INET6 796 case AF_INET6: 797 { 798 th->th_sum = 0; 799 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 800 tlen); 801 ip6->ip6_plen = ntohs(tlen); 802 if (tp && tp->t_in6pcb) { 803 struct ifnet *oifp; 804 ro = (struct route *)&tp->t_in6pcb->in6p_route; 805 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 806 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 807 } else 808 ip6->ip6_hlim = ip6_defhlim; 809 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 810 if (ip6_auto_flowlabel) { 811 ip6->ip6_flow |= 812 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 813 } 814 break; 815 } 816 #endif 817 } 818 819 #ifdef IPSEC 820 (void)ipsec_setsocket(m, NULL); 821 #endif /*IPSEC*/ 822 823 if (tp != NULL && tp->t_inpcb != NULL) { 824 ro = &tp->t_inpcb->inp_route; 825 #ifdef IPSEC 826 if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) { 827 m_freem(m); 828 return ENOBUFS; 829 } 830 #endif 831 #ifdef DIAGNOSTIC 832 if (family != AF_INET) 833 panic("tcp_respond: address family mismatch"); 834 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 835 panic("tcp_respond: ip_dst %x != inp_faddr %x", 836 ntohl(ip->ip_dst.s_addr), 837 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 838 } 839 #endif 840 } 841 #ifdef INET6 842 else if (tp != NULL && tp->t_in6pcb != NULL) { 843 ro = (struct route *)&tp->t_in6pcb->in6p_route; 844 #ifdef IPSEC 845 if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) { 846 m_freem(m); 847 return ENOBUFS; 848 } 849 #endif 850 #ifdef DIAGNOSTIC 851 if (family == AF_INET) { 852 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 853 panic("tcp_respond: not mapped addr"); 854 if (bcmp(&ip->ip_dst, 855 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 856 sizeof(ip->ip_dst)) != 0) { 857 panic("tcp_respond: ip_dst != in6p_faddr"); 858 } 859 } else if (family == AF_INET6) { 860 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 861 &tp->t_in6pcb->in6p_faddr)) 862 panic("tcp_respond: ip6_dst != in6p_faddr"); 863 } else 864 panic("tcp_respond: address family mismatch"); 865 #endif 866 } 867 #endif 868 else 869 ro = NULL; 870 871 switch (family) { 872 #ifdef INET 873 case AF_INET: 874 error = ip_output(m, NULL, ro, 875 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 876 (struct ip_moptions *)0, 877 (tp == NULL ? (struct inpcb *)0 : tp->t_inpcb)); 878 break; 879 #endif 880 #ifdef INET6 881 case AF_INET6: 882 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 883 (struct ip6_moptions *)0, (struct in6pcb *)0, NULL); 884 break; 885 #endif 886 default: 887 error = EAFNOSUPPORT; 888 break; 889 } 890 891 return (error); 892 } 893 894 /* 895 * Create a new TCP control block, making an 896 * empty reassembly queue and hooking it to the argument 897 * protocol control block. 898 */ 899 struct tcpcb * 900 tcp_newtcpcb(family, aux) 901 int family; /* selects inpcb, or in6pcb */ 902 void *aux; 903 { 904 struct tcpcb *tp; 905 int i; 906 907 switch (family) { 908 case PF_INET: 909 break; 910 #ifdef INET6 911 case PF_INET6: 912 break; 913 #endif 914 default: 915 return NULL; 916 } 917 918 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 919 if (tp == NULL) 920 return (NULL); 921 bzero((caddr_t)tp, sizeof(struct tcpcb)); 922 TAILQ_INIT(&tp->segq); 923 TAILQ_INIT(&tp->timeq); 924 tp->t_family = family; /* may be overridden later on */ 925 tp->t_peermss = tcp_mssdflt; 926 tp->t_ourmss = tcp_mssdflt; 927 tp->t_segsz = tcp_mssdflt; 928 LIST_INIT(&tp->t_sc); 929 930 tp->t_lastm = NULL; 931 tp->t_lastoff = 0; 932 933 callout_init(&tp->t_delack_ch); 934 for (i = 0; i < TCPT_NTIMERS; i++) 935 TCP_TIMER_INIT(tp, i); 936 937 tp->t_flags = 0; 938 if (tcp_do_rfc1323 && tcp_do_win_scale) 939 tp->t_flags |= TF_REQ_SCALE; 940 if (tcp_do_rfc1323 && tcp_do_timestamps) 941 tp->t_flags |= TF_REQ_TSTMP; 942 if (tcp_do_sack == 2) 943 tp->t_flags |= TF_WILL_SACK; 944 else if (tcp_do_sack == 1) 945 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 946 tp->t_flags |= TF_CANT_TXSACK; 947 switch (family) { 948 case PF_INET: 949 tp->t_inpcb = (struct inpcb *)aux; 950 tp->t_mtudisc = ip_mtudisc; 951 break; 952 #ifdef INET6 953 case PF_INET6: 954 tp->t_in6pcb = (struct in6pcb *)aux; 955 /* for IPv6, always try to run path MTU discovery */ 956 tp->t_mtudisc = 1; 957 break; 958 #endif 959 } 960 /* 961 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 962 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 963 * reasonable initial retransmit time. 964 */ 965 tp->t_srtt = TCPTV_SRTTBASE; 966 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 967 tp->t_rttmin = TCPTV_MIN; 968 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 969 TCPTV_MIN, TCPTV_REXMTMAX); 970 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 971 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 972 if (family == AF_INET) { 973 struct inpcb *inp = (struct inpcb *)aux; 974 inp->inp_ip.ip_ttl = ip_defttl; 975 inp->inp_ppcb = (caddr_t)tp; 976 } 977 #ifdef INET6 978 else if (family == AF_INET6) { 979 struct in6pcb *in6p = (struct in6pcb *)aux; 980 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 981 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 982 : NULL); 983 in6p->in6p_ppcb = (caddr_t)tp; 984 } 985 #endif 986 987 /* 988 * Initialize our timebase. When we send timestamps, we take 989 * the delta from tcp_now -- this means each connection always 990 * gets a timebase of 0, which makes it, among other things, 991 * more difficult to determine how long a system has been up, 992 * and thus how many TCP sequence increments have occurred. 993 */ 994 tp->ts_timebase = tcp_now; 995 996 return (tp); 997 } 998 999 /* 1000 * Drop a TCP connection, reporting 1001 * the specified error. If connection is synchronized, 1002 * then send a RST to peer. 1003 */ 1004 struct tcpcb * 1005 tcp_drop(tp, errno) 1006 struct tcpcb *tp; 1007 int errno; 1008 { 1009 struct socket *so = NULL; 1010 1011 #ifdef DIAGNOSTIC 1012 if (tp->t_inpcb && tp->t_in6pcb) 1013 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1014 #endif 1015 #ifdef INET 1016 if (tp->t_inpcb) 1017 so = tp->t_inpcb->inp_socket; 1018 #endif 1019 #ifdef INET6 1020 if (tp->t_in6pcb) 1021 so = tp->t_in6pcb->in6p_socket; 1022 #endif 1023 if (!so) 1024 return NULL; 1025 1026 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1027 tp->t_state = TCPS_CLOSED; 1028 (void) tcp_output(tp); 1029 tcpstat.tcps_drops++; 1030 } else 1031 tcpstat.tcps_conndrops++; 1032 if (errno == ETIMEDOUT && tp->t_softerror) 1033 errno = tp->t_softerror; 1034 so->so_error = errno; 1035 return (tcp_close(tp)); 1036 } 1037 1038 /* 1039 * Return whether this tcpcb is marked as dead, indicating 1040 * to the calling timer function that no further action should 1041 * be taken, as we are about to release this tcpcb. The release 1042 * of the storage will be done if this is the last timer running. 1043 * 1044 * This is typically called from the callout handler function before 1045 * callout_ack() is done, therefore we need to test the number of 1046 * running timer functions against 1 below, not 0. 1047 */ 1048 int 1049 tcp_isdead(tp) 1050 struct tcpcb *tp; 1051 { 1052 int dead = (tp->t_flags & TF_DEAD); 1053 1054 if (__predict_false(dead)) { 1055 if (tcp_timers_invoking(tp) > 1) 1056 /* not quite there yet -- count separately? */ 1057 return dead; 1058 tcpstat.tcps_delayed_free++; 1059 pool_put(&tcpcb_pool, tp); 1060 } 1061 return dead; 1062 } 1063 1064 /* 1065 * Close a TCP control block: 1066 * discard all space held by the tcp 1067 * discard internet protocol block 1068 * wake up any sleepers 1069 */ 1070 struct tcpcb * 1071 tcp_close(tp) 1072 struct tcpcb *tp; 1073 { 1074 struct inpcb *inp; 1075 #ifdef INET6 1076 struct in6pcb *in6p; 1077 #endif 1078 struct socket *so; 1079 #ifdef RTV_RTT 1080 struct rtentry *rt; 1081 #endif 1082 struct route *ro; 1083 1084 inp = tp->t_inpcb; 1085 #ifdef INET6 1086 in6p = tp->t_in6pcb; 1087 #endif 1088 so = NULL; 1089 ro = NULL; 1090 if (inp) { 1091 so = inp->inp_socket; 1092 ro = &inp->inp_route; 1093 } 1094 #ifdef INET6 1095 else if (in6p) { 1096 so = in6p->in6p_socket; 1097 ro = (struct route *)&in6p->in6p_route; 1098 } 1099 #endif 1100 1101 #ifdef RTV_RTT 1102 /* 1103 * If we sent enough data to get some meaningful characteristics, 1104 * save them in the routing entry. 'Enough' is arbitrarily 1105 * defined as the sendpipesize (default 4K) * 16. This would 1106 * give us 16 rtt samples assuming we only get one sample per 1107 * window (the usual case on a long haul net). 16 samples is 1108 * enough for the srtt filter to converge to within 5% of the correct 1109 * value; fewer samples and we could save a very bogus rtt. 1110 * 1111 * Don't update the default route's characteristics and don't 1112 * update anything that the user "locked". 1113 */ 1114 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1115 ro && (rt = ro->ro_rt) && 1116 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1117 u_long i = 0; 1118 1119 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1120 i = tp->t_srtt * 1121 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1122 if (rt->rt_rmx.rmx_rtt && i) 1123 /* 1124 * filter this update to half the old & half 1125 * the new values, converting scale. 1126 * See route.h and tcp_var.h for a 1127 * description of the scaling constants. 1128 */ 1129 rt->rt_rmx.rmx_rtt = 1130 (rt->rt_rmx.rmx_rtt + i) / 2; 1131 else 1132 rt->rt_rmx.rmx_rtt = i; 1133 } 1134 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1135 i = tp->t_rttvar * 1136 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1137 if (rt->rt_rmx.rmx_rttvar && i) 1138 rt->rt_rmx.rmx_rttvar = 1139 (rt->rt_rmx.rmx_rttvar + i) / 2; 1140 else 1141 rt->rt_rmx.rmx_rttvar = i; 1142 } 1143 /* 1144 * update the pipelimit (ssthresh) if it has been updated 1145 * already or if a pipesize was specified & the threshhold 1146 * got below half the pipesize. I.e., wait for bad news 1147 * before we start updating, then update on both good 1148 * and bad news. 1149 */ 1150 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1151 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1152 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1153 /* 1154 * convert the limit from user data bytes to 1155 * packets then to packet data bytes. 1156 */ 1157 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1158 if (i < 2) 1159 i = 2; 1160 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1161 if (rt->rt_rmx.rmx_ssthresh) 1162 rt->rt_rmx.rmx_ssthresh = 1163 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1164 else 1165 rt->rt_rmx.rmx_ssthresh = i; 1166 } 1167 } 1168 #endif /* RTV_RTT */ 1169 /* free the reassembly queue, if any */ 1170 TCP_REASS_LOCK(tp); 1171 (void) tcp_freeq(tp); 1172 TCP_REASS_UNLOCK(tp); 1173 1174 tcp_canceltimers(tp); 1175 TCP_CLEAR_DELACK(tp); 1176 syn_cache_cleanup(tp); 1177 1178 if (tp->t_template) { 1179 m_free(tp->t_template); 1180 tp->t_template = NULL; 1181 } 1182 if (tcp_timers_invoking(tp)) 1183 tp->t_flags |= TF_DEAD; 1184 else 1185 pool_put(&tcpcb_pool, tp); 1186 1187 if (inp) { 1188 inp->inp_ppcb = 0; 1189 soisdisconnected(so); 1190 in_pcbdetach(inp); 1191 } 1192 #ifdef INET6 1193 else if (in6p) { 1194 in6p->in6p_ppcb = 0; 1195 soisdisconnected(so); 1196 in6_pcbdetach(in6p); 1197 } 1198 #endif 1199 tcpstat.tcps_closed++; 1200 return ((struct tcpcb *)0); 1201 } 1202 1203 int 1204 tcp_freeq(tp) 1205 struct tcpcb *tp; 1206 { 1207 struct ipqent *qe; 1208 int rv = 0; 1209 #ifdef TCPREASS_DEBUG 1210 int i = 0; 1211 #endif 1212 1213 TCP_REASS_LOCK_CHECK(tp); 1214 1215 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1216 #ifdef TCPREASS_DEBUG 1217 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1218 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1219 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1220 #endif 1221 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1222 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1223 m_freem(qe->ipqe_m); 1224 pool_put(&ipqent_pool, qe); 1225 rv = 1; 1226 } 1227 return (rv); 1228 } 1229 1230 /* 1231 * Protocol drain routine. Called when memory is in short supply. 1232 */ 1233 void 1234 tcp_drain() 1235 { 1236 struct inpcb *inp; 1237 struct tcpcb *tp; 1238 1239 /* 1240 * Free the sequence queue of all TCP connections. 1241 */ 1242 inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue); 1243 if (inp) /* XXX */ 1244 CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) { 1245 if ((tp = intotcpcb(inp)) != NULL) { 1246 /* 1247 * We may be called from a device's interrupt 1248 * context. If the tcpcb is already busy, 1249 * just bail out now. 1250 */ 1251 if (tcp_reass_lock_try(tp) == 0) 1252 continue; 1253 if (tcp_freeq(tp)) 1254 tcpstat.tcps_connsdrained++; 1255 TCP_REASS_UNLOCK(tp); 1256 } 1257 } 1258 } 1259 1260 #ifdef INET6 1261 void 1262 tcp6_drain() 1263 { 1264 struct in6pcb *in6p; 1265 struct tcpcb *tp; 1266 struct in6pcb *head = &tcb6; 1267 1268 /* 1269 * Free the sequence queue of all TCP connections. 1270 */ 1271 for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) { 1272 if ((tp = in6totcpcb(in6p)) != NULL) { 1273 /* 1274 * We may be called from a device's interrupt 1275 * context. If the tcpcb is already busy, 1276 * just bail out now. 1277 */ 1278 if (tcp_reass_lock_try(tp) == 0) 1279 continue; 1280 if (tcp_freeq(tp)) 1281 tcpstat.tcps_connsdrained++; 1282 TCP_REASS_UNLOCK(tp); 1283 } 1284 } 1285 } 1286 #endif 1287 1288 /* 1289 * Notify a tcp user of an asynchronous error; 1290 * store error as soft error, but wake up user 1291 * (for now, won't do anything until can select for soft error). 1292 */ 1293 void 1294 tcp_notify(inp, error) 1295 struct inpcb *inp; 1296 int error; 1297 { 1298 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1299 struct socket *so = inp->inp_socket; 1300 1301 /* 1302 * Ignore some errors if we are hooked up. 1303 * If connection hasn't completed, has retransmitted several times, 1304 * and receives a second error, give up now. This is better 1305 * than waiting a long time to establish a connection that 1306 * can never complete. 1307 */ 1308 if (tp->t_state == TCPS_ESTABLISHED && 1309 (error == EHOSTUNREACH || error == ENETUNREACH || 1310 error == EHOSTDOWN)) { 1311 return; 1312 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1313 tp->t_rxtshift > 3 && tp->t_softerror) 1314 so->so_error = error; 1315 else 1316 tp->t_softerror = error; 1317 wakeup((caddr_t) &so->so_timeo); 1318 sorwakeup(so); 1319 sowwakeup(so); 1320 } 1321 1322 #ifdef INET6 1323 void 1324 tcp6_notify(in6p, error) 1325 struct in6pcb *in6p; 1326 int error; 1327 { 1328 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1329 struct socket *so = in6p->in6p_socket; 1330 1331 /* 1332 * Ignore some errors if we are hooked up. 1333 * If connection hasn't completed, has retransmitted several times, 1334 * and receives a second error, give up now. This is better 1335 * than waiting a long time to establish a connection that 1336 * can never complete. 1337 */ 1338 if (tp->t_state == TCPS_ESTABLISHED && 1339 (error == EHOSTUNREACH || error == ENETUNREACH || 1340 error == EHOSTDOWN)) { 1341 return; 1342 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1343 tp->t_rxtshift > 3 && tp->t_softerror) 1344 so->so_error = error; 1345 else 1346 tp->t_softerror = error; 1347 wakeup((caddr_t) &so->so_timeo); 1348 sorwakeup(so); 1349 sowwakeup(so); 1350 } 1351 #endif 1352 1353 #ifdef INET6 1354 void 1355 tcp6_ctlinput(cmd, sa, d) 1356 int cmd; 1357 struct sockaddr *sa; 1358 void *d; 1359 { 1360 struct tcphdr th; 1361 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1362 int nmatch; 1363 struct ip6_hdr *ip6; 1364 const struct sockaddr_in6 *sa6_src = NULL; 1365 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1366 struct mbuf *m; 1367 int off; 1368 1369 if (sa->sa_family != AF_INET6 || 1370 sa->sa_len != sizeof(struct sockaddr_in6)) 1371 return; 1372 if ((unsigned)cmd >= PRC_NCMDS) 1373 return; 1374 else if (cmd == PRC_QUENCH) { 1375 /* XXX there's no PRC_QUENCH in IPv6 */ 1376 notify = tcp6_quench; 1377 } else if (PRC_IS_REDIRECT(cmd)) 1378 notify = in6_rtchange, d = NULL; 1379 else if (cmd == PRC_MSGSIZE) 1380 ; /* special code is present, see below */ 1381 else if (cmd == PRC_HOSTDEAD) 1382 d = NULL; 1383 else if (inet6ctlerrmap[cmd] == 0) 1384 return; 1385 1386 /* if the parameter is from icmp6, decode it. */ 1387 if (d != NULL) { 1388 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1389 m = ip6cp->ip6c_m; 1390 ip6 = ip6cp->ip6c_ip6; 1391 off = ip6cp->ip6c_off; 1392 sa6_src = ip6cp->ip6c_src; 1393 } else { 1394 m = NULL; 1395 ip6 = NULL; 1396 sa6_src = &sa6_any; 1397 } 1398 1399 if (ip6) { 1400 /* 1401 * XXX: We assume that when ip6 is non NULL, 1402 * M and OFF are valid. 1403 */ 1404 1405 /* check if we can safely examine src and dst ports */ 1406 if (m->m_pkthdr.len < off + sizeof(th)) { 1407 if (cmd == PRC_MSGSIZE) 1408 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1409 return; 1410 } 1411 1412 bzero(&th, sizeof(th)); 1413 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1414 1415 if (cmd == PRC_MSGSIZE) { 1416 int valid = 0; 1417 1418 /* 1419 * Check to see if we have a valid TCP connection 1420 * corresponding to the address in the ICMPv6 message 1421 * payload. 1422 */ 1423 if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr, 1424 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1425 th.th_sport, 0)) 1426 valid++; 1427 1428 /* 1429 * Depending on the value of "valid" and routing table 1430 * size (mtudisc_{hi,lo}wat), we will: 1431 * - recalcurate the new MTU and create the 1432 * corresponding routing entry, or 1433 * - ignore the MTU change notification. 1434 */ 1435 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1436 1437 /* 1438 * no need to call in6_pcbnotify, it should have been 1439 * called via callback if necessary 1440 */ 1441 return; 1442 } 1443 1444 nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport, 1445 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1446 if (nmatch == 0 && syn_cache_count && 1447 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1448 inet6ctlerrmap[cmd] == ENETUNREACH || 1449 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1450 syn_cache_unreach((struct sockaddr *)sa6_src, 1451 sa, &th); 1452 } else { 1453 (void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src, 1454 0, cmd, NULL, notify); 1455 } 1456 } 1457 #endif 1458 1459 #ifdef INET 1460 /* assumes that ip header and tcp header are contiguous on mbuf */ 1461 void * 1462 tcp_ctlinput(cmd, sa, v) 1463 int cmd; 1464 struct sockaddr *sa; 1465 void *v; 1466 { 1467 struct ip *ip = v; 1468 struct tcphdr *th; 1469 struct icmp *icp; 1470 extern const int inetctlerrmap[]; 1471 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1472 int errno; 1473 int nmatch; 1474 #ifdef INET6 1475 struct in6_addr src6, dst6; 1476 #endif 1477 1478 if (sa->sa_family != AF_INET || 1479 sa->sa_len != sizeof(struct sockaddr_in)) 1480 return NULL; 1481 if ((unsigned)cmd >= PRC_NCMDS) 1482 return NULL; 1483 errno = inetctlerrmap[cmd]; 1484 if (cmd == PRC_QUENCH) 1485 notify = tcp_quench; 1486 else if (PRC_IS_REDIRECT(cmd)) 1487 notify = in_rtchange, ip = 0; 1488 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1489 /* 1490 * Check to see if we have a valid TCP connection 1491 * corresponding to the address in the ICMP message 1492 * payload. 1493 * 1494 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1495 */ 1496 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1497 #ifdef INET6 1498 memset(&src6, 0, sizeof(src6)); 1499 memset(&dst6, 0, sizeof(dst6)); 1500 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1501 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1502 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1503 #endif 1504 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1505 ip->ip_src, th->th_sport) != NULL) 1506 ; 1507 #ifdef INET6 1508 else if (in6_pcblookup_connect(&tcb6, &dst6, 1509 th->th_dport, &src6, th->th_sport, 0) != NULL) 1510 ; 1511 #endif 1512 else 1513 return NULL; 1514 1515 /* 1516 * Now that we've validated that we are actually communicating 1517 * with the host indicated in the ICMP message, locate the 1518 * ICMP header, recalculate the new MTU, and create the 1519 * corresponding routing entry. 1520 */ 1521 icp = (struct icmp *)((caddr_t)ip - 1522 offsetof(struct icmp, icmp_ip)); 1523 icmp_mtudisc(icp, ip->ip_dst); 1524 1525 return NULL; 1526 } else if (cmd == PRC_HOSTDEAD) 1527 ip = 0; 1528 else if (errno == 0) 1529 return NULL; 1530 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1531 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1532 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1533 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1534 if (nmatch == 0 && syn_cache_count && 1535 (inetctlerrmap[cmd] == EHOSTUNREACH || 1536 inetctlerrmap[cmd] == ENETUNREACH || 1537 inetctlerrmap[cmd] == EHOSTDOWN)) { 1538 struct sockaddr_in sin; 1539 bzero(&sin, sizeof(sin)); 1540 sin.sin_len = sizeof(sin); 1541 sin.sin_family = AF_INET; 1542 sin.sin_port = th->th_sport; 1543 sin.sin_addr = ip->ip_src; 1544 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1545 } 1546 1547 /* XXX mapped address case */ 1548 } else 1549 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1550 notify); 1551 return NULL; 1552 } 1553 1554 /* 1555 * When a source quence is received, we are being notifed of congestion. 1556 * Close the congestion window down to the Loss Window (one segment). 1557 * We will gradually open it again as we proceed. 1558 */ 1559 void 1560 tcp_quench(inp, errno) 1561 struct inpcb *inp; 1562 int errno; 1563 { 1564 struct tcpcb *tp = intotcpcb(inp); 1565 1566 if (tp) 1567 tp->snd_cwnd = tp->t_segsz; 1568 } 1569 #endif 1570 1571 #ifdef INET6 1572 void 1573 tcp6_quench(in6p, errno) 1574 struct in6pcb *in6p; 1575 int errno; 1576 { 1577 struct tcpcb *tp = in6totcpcb(in6p); 1578 1579 if (tp) 1580 tp->snd_cwnd = tp->t_segsz; 1581 } 1582 #endif 1583 1584 #ifdef INET 1585 /* 1586 * Path MTU Discovery handlers. 1587 */ 1588 void 1589 tcp_mtudisc_callback(faddr) 1590 struct in_addr faddr; 1591 { 1592 #ifdef INET6 1593 struct in6_addr in6; 1594 #endif 1595 1596 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1597 #ifdef INET6 1598 memset(&in6, 0, sizeof(in6)); 1599 in6.s6_addr16[5] = 0xffff; 1600 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1601 tcp6_mtudisc_callback(&in6); 1602 #endif 1603 } 1604 1605 /* 1606 * On receipt of path MTU corrections, flush old route and replace it 1607 * with the new one. Retransmit all unacknowledged packets, to ensure 1608 * that all packets will be received. 1609 */ 1610 void 1611 tcp_mtudisc(inp, errno) 1612 struct inpcb *inp; 1613 int errno; 1614 { 1615 struct tcpcb *tp = intotcpcb(inp); 1616 struct rtentry *rt = in_pcbrtentry(inp); 1617 1618 if (tp != 0) { 1619 if (rt != 0) { 1620 /* 1621 * If this was not a host route, remove and realloc. 1622 */ 1623 if ((rt->rt_flags & RTF_HOST) == 0) { 1624 in_rtchange(inp, errno); 1625 if ((rt = in_pcbrtentry(inp)) == 0) 1626 return; 1627 } 1628 1629 /* 1630 * Slow start out of the error condition. We 1631 * use the MTU because we know it's smaller 1632 * than the previously transmitted segment. 1633 * 1634 * Note: This is more conservative than the 1635 * suggestion in draft-floyd-incr-init-win-03. 1636 */ 1637 if (rt->rt_rmx.rmx_mtu != 0) 1638 tp->snd_cwnd = 1639 TCP_INITIAL_WINDOW(tcp_init_win, 1640 rt->rt_rmx.rmx_mtu); 1641 } 1642 1643 /* 1644 * Resend unacknowledged packets. 1645 */ 1646 tp->snd_nxt = tp->snd_una; 1647 tcp_output(tp); 1648 } 1649 } 1650 #endif 1651 1652 #ifdef INET6 1653 /* 1654 * Path MTU Discovery handlers. 1655 */ 1656 void 1657 tcp6_mtudisc_callback(faddr) 1658 struct in6_addr *faddr; 1659 { 1660 struct sockaddr_in6 sin6; 1661 1662 bzero(&sin6, sizeof(sin6)); 1663 sin6.sin6_family = AF_INET6; 1664 sin6.sin6_len = sizeof(struct sockaddr_in6); 1665 sin6.sin6_addr = *faddr; 1666 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1667 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1668 } 1669 1670 void 1671 tcp6_mtudisc(in6p, errno) 1672 struct in6pcb *in6p; 1673 int errno; 1674 { 1675 struct tcpcb *tp = in6totcpcb(in6p); 1676 struct rtentry *rt = in6_pcbrtentry(in6p); 1677 1678 if (tp != 0) { 1679 if (rt != 0) { 1680 /* 1681 * If this was not a host route, remove and realloc. 1682 */ 1683 if ((rt->rt_flags & RTF_HOST) == 0) { 1684 in6_rtchange(in6p, errno); 1685 if ((rt = in6_pcbrtentry(in6p)) == 0) 1686 return; 1687 } 1688 1689 /* 1690 * Slow start out of the error condition. We 1691 * use the MTU because we know it's smaller 1692 * than the previously transmitted segment. 1693 * 1694 * Note: This is more conservative than the 1695 * suggestion in draft-floyd-incr-init-win-03. 1696 */ 1697 if (rt->rt_rmx.rmx_mtu != 0) 1698 tp->snd_cwnd = 1699 TCP_INITIAL_WINDOW(tcp_init_win, 1700 rt->rt_rmx.rmx_mtu); 1701 } 1702 1703 /* 1704 * Resend unacknowledged packets. 1705 */ 1706 tp->snd_nxt = tp->snd_una; 1707 tcp_output(tp); 1708 } 1709 } 1710 #endif /* INET6 */ 1711 1712 /* 1713 * Compute the MSS to advertise to the peer. Called only during 1714 * the 3-way handshake. If we are the server (peer initiated 1715 * connection), we are called with a pointer to the interface 1716 * on which the SYN packet arrived. If we are the client (we 1717 * initiated connection), we are called with a pointer to the 1718 * interface out which this connection should go. 1719 * 1720 * NOTE: Do not subtract IP option/extension header size nor IPsec 1721 * header size from MSS advertisement. MSS option must hold the maximum 1722 * segment size we can accept, so it must always be: 1723 * max(if mtu) - ip header - tcp header 1724 */ 1725 u_long 1726 tcp_mss_to_advertise(ifp, af) 1727 const struct ifnet *ifp; 1728 int af; 1729 { 1730 extern u_long in_maxmtu; 1731 u_long mss = 0; 1732 u_long hdrsiz; 1733 1734 /* 1735 * In order to avoid defeating path MTU discovery on the peer, 1736 * we advertise the max MTU of all attached networks as our MSS, 1737 * per RFC 1191, section 3.1. 1738 * 1739 * We provide the option to advertise just the MTU of 1740 * the interface on which we hope this connection will 1741 * be receiving. If we are responding to a SYN, we 1742 * will have a pretty good idea about this, but when 1743 * initiating a connection there is a bit more doubt. 1744 * 1745 * We also need to ensure that loopback has a large enough 1746 * MSS, as the loopback MTU is never included in in_maxmtu. 1747 */ 1748 1749 if (ifp != NULL) 1750 switch (af) { 1751 case AF_INET: 1752 mss = ifp->if_mtu; 1753 break; 1754 #ifdef INET6 1755 case AF_INET6: 1756 mss = IN6_LINKMTU(ifp); 1757 break; 1758 #endif 1759 } 1760 1761 if (tcp_mss_ifmtu == 0) 1762 switch (af) { 1763 case AF_INET: 1764 mss = max(in_maxmtu, mss); 1765 break; 1766 #ifdef INET6 1767 case AF_INET6: 1768 mss = max(in6_maxmtu, mss); 1769 break; 1770 #endif 1771 } 1772 1773 switch (af) { 1774 case AF_INET: 1775 hdrsiz = sizeof(struct ip); 1776 break; 1777 #ifdef INET6 1778 case AF_INET6: 1779 hdrsiz = sizeof(struct ip6_hdr); 1780 break; 1781 #endif 1782 default: 1783 hdrsiz = 0; 1784 break; 1785 } 1786 hdrsiz += sizeof(struct tcphdr); 1787 if (mss > hdrsiz) 1788 mss -= hdrsiz; 1789 1790 mss = max(tcp_mssdflt, mss); 1791 return (mss); 1792 } 1793 1794 /* 1795 * Set connection variables based on the peer's advertised MSS. 1796 * We are passed the TCPCB for the actual connection. If we 1797 * are the server, we are called by the compressed state engine 1798 * when the 3-way handshake is complete. If we are the client, 1799 * we are called when we receive the SYN,ACK from the server. 1800 * 1801 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1802 * before this routine is called! 1803 */ 1804 void 1805 tcp_mss_from_peer(tp, offer) 1806 struct tcpcb *tp; 1807 int offer; 1808 { 1809 struct socket *so; 1810 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1811 struct rtentry *rt; 1812 #endif 1813 u_long bufsize; 1814 int mss; 1815 1816 #ifdef DIAGNOSTIC 1817 if (tp->t_inpcb && tp->t_in6pcb) 1818 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1819 #endif 1820 so = NULL; 1821 rt = NULL; 1822 #ifdef INET 1823 if (tp->t_inpcb) { 1824 so = tp->t_inpcb->inp_socket; 1825 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1826 rt = in_pcbrtentry(tp->t_inpcb); 1827 #endif 1828 } 1829 #endif 1830 #ifdef INET6 1831 if (tp->t_in6pcb) { 1832 so = tp->t_in6pcb->in6p_socket; 1833 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1834 rt = in6_pcbrtentry(tp->t_in6pcb); 1835 #endif 1836 } 1837 #endif 1838 1839 /* 1840 * As per RFC1122, use the default MSS value, unless they 1841 * sent us an offer. Do not accept offers less than 32 bytes. 1842 */ 1843 mss = tcp_mssdflt; 1844 if (offer) 1845 mss = offer; 1846 mss = max(mss, 32); /* sanity */ 1847 tp->t_peermss = mss; 1848 mss -= tcp_optlen(tp); 1849 #ifdef INET 1850 if (tp->t_inpcb) 1851 mss -= ip_optlen(tp->t_inpcb); 1852 #endif 1853 #ifdef INET6 1854 if (tp->t_in6pcb) 1855 mss -= ip6_optlen(tp->t_in6pcb); 1856 #endif 1857 1858 /* 1859 * If there's a pipesize, change the socket buffer to that size. 1860 * Make the socket buffer an integral number of MSS units. If 1861 * the MSS is larger than the socket buffer, artificially decrease 1862 * the MSS. 1863 */ 1864 #ifdef RTV_SPIPE 1865 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1866 bufsize = rt->rt_rmx.rmx_sendpipe; 1867 else 1868 #endif 1869 bufsize = so->so_snd.sb_hiwat; 1870 if (bufsize < mss) 1871 mss = bufsize; 1872 else { 1873 bufsize = roundup(bufsize, mss); 1874 if (bufsize > sb_max) 1875 bufsize = sb_max; 1876 (void) sbreserve(&so->so_snd, bufsize); 1877 } 1878 tp->t_segsz = mss; 1879 1880 #ifdef RTV_SSTHRESH 1881 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1882 /* 1883 * There's some sort of gateway or interface buffer 1884 * limit on the path. Use this to set the slow 1885 * start threshold, but set the threshold to no less 1886 * than 2 * MSS. 1887 */ 1888 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1889 } 1890 #endif 1891 } 1892 1893 /* 1894 * Processing necessary when a TCP connection is established. 1895 */ 1896 void 1897 tcp_established(tp) 1898 struct tcpcb *tp; 1899 { 1900 struct socket *so; 1901 #ifdef RTV_RPIPE 1902 struct rtentry *rt; 1903 #endif 1904 u_long bufsize; 1905 1906 #ifdef DIAGNOSTIC 1907 if (tp->t_inpcb && tp->t_in6pcb) 1908 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1909 #endif 1910 so = NULL; 1911 rt = NULL; 1912 #ifdef INET 1913 if (tp->t_inpcb) { 1914 so = tp->t_inpcb->inp_socket; 1915 #if defined(RTV_RPIPE) 1916 rt = in_pcbrtentry(tp->t_inpcb); 1917 #endif 1918 } 1919 #endif 1920 #ifdef INET6 1921 if (tp->t_in6pcb) { 1922 so = tp->t_in6pcb->in6p_socket; 1923 #if defined(RTV_RPIPE) 1924 rt = in6_pcbrtentry(tp->t_in6pcb); 1925 #endif 1926 } 1927 #endif 1928 1929 tp->t_state = TCPS_ESTABLISHED; 1930 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1931 1932 #ifdef RTV_RPIPE 1933 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1934 bufsize = rt->rt_rmx.rmx_recvpipe; 1935 else 1936 #endif 1937 bufsize = so->so_rcv.sb_hiwat; 1938 if (bufsize > tp->t_ourmss) { 1939 bufsize = roundup(bufsize, tp->t_ourmss); 1940 if (bufsize > sb_max) 1941 bufsize = sb_max; 1942 (void) sbreserve(&so->so_rcv, bufsize); 1943 } 1944 } 1945 1946 /* 1947 * Check if there's an initial rtt or rttvar. Convert from the 1948 * route-table units to scaled multiples of the slow timeout timer. 1949 * Called only during the 3-way handshake. 1950 */ 1951 void 1952 tcp_rmx_rtt(tp) 1953 struct tcpcb *tp; 1954 { 1955 #ifdef RTV_RTT 1956 struct rtentry *rt = NULL; 1957 int rtt; 1958 1959 #ifdef DIAGNOSTIC 1960 if (tp->t_inpcb && tp->t_in6pcb) 1961 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1962 #endif 1963 #ifdef INET 1964 if (tp->t_inpcb) 1965 rt = in_pcbrtentry(tp->t_inpcb); 1966 #endif 1967 #ifdef INET6 1968 if (tp->t_in6pcb) 1969 rt = in6_pcbrtentry(tp->t_in6pcb); 1970 #endif 1971 if (rt == NULL) 1972 return; 1973 1974 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1975 /* 1976 * XXX The lock bit for MTU indicates that the value 1977 * is also a minimum value; this is subject to time. 1978 */ 1979 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1980 TCPT_RANGESET(tp->t_rttmin, 1981 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1982 TCPTV_MIN, TCPTV_REXMTMAX); 1983 tp->t_srtt = rtt / 1984 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1985 if (rt->rt_rmx.rmx_rttvar) { 1986 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1987 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1988 (TCP_RTTVAR_SHIFT + 2)); 1989 } else { 1990 /* Default variation is +- 1 rtt */ 1991 tp->t_rttvar = 1992 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1993 } 1994 TCPT_RANGESET(tp->t_rxtcur, 1995 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1996 tp->t_rttmin, TCPTV_REXMTMAX); 1997 } 1998 #endif 1999 } 2000 2001 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2002 #if NRND > 0 2003 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2004 #endif 2005 2006 /* 2007 * Get a new sequence value given a tcp control block 2008 */ 2009 tcp_seq 2010 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2011 { 2012 2013 #ifdef INET 2014 if (tp->t_inpcb != NULL) { 2015 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2016 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2017 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2018 addin)); 2019 } 2020 #endif 2021 #ifdef INET6 2022 if (tp->t_in6pcb != NULL) { 2023 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2024 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2025 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2026 addin)); 2027 } 2028 #endif 2029 /* Not possible. */ 2030 panic("tcp_new_iss"); 2031 } 2032 2033 /* 2034 * This routine actually generates a new TCP initial sequence number. 2035 */ 2036 tcp_seq 2037 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2038 size_t addrsz, tcp_seq addin) 2039 { 2040 tcp_seq tcp_iss; 2041 2042 #if NRND > 0 2043 static int beenhere; 2044 2045 /* 2046 * If we haven't been here before, initialize our cryptographic 2047 * hash secret. 2048 */ 2049 if (beenhere == 0) { 2050 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2051 RND_EXTRACT_ANY); 2052 beenhere = 1; 2053 } 2054 2055 if (tcp_do_rfc1948) { 2056 MD5_CTX ctx; 2057 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2058 2059 /* 2060 * Compute the base value of the ISS. It is a hash 2061 * of (saddr, sport, daddr, dport, secret). 2062 */ 2063 MD5Init(&ctx); 2064 2065 MD5Update(&ctx, (u_char *) laddr, addrsz); 2066 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2067 2068 MD5Update(&ctx, (u_char *) faddr, addrsz); 2069 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2070 2071 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2072 2073 MD5Final(hash, &ctx); 2074 2075 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2076 2077 /* 2078 * Now increment our "timer", and add it in to 2079 * the computed value. 2080 * 2081 * XXX Use `addin'? 2082 * XXX TCP_ISSINCR too large to use? 2083 */ 2084 tcp_iss_seq += TCP_ISSINCR; 2085 #ifdef TCPISS_DEBUG 2086 printf("ISS hash 0x%08x, ", tcp_iss); 2087 #endif 2088 tcp_iss += tcp_iss_seq + addin; 2089 #ifdef TCPISS_DEBUG 2090 printf("new ISS 0x%08x\n", tcp_iss); 2091 #endif 2092 } else 2093 #endif /* NRND > 0 */ 2094 { 2095 /* 2096 * Randomize. 2097 */ 2098 #if NRND > 0 2099 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2100 #else 2101 tcp_iss = arc4random(); 2102 #endif 2103 2104 /* 2105 * If we were asked to add some amount to a known value, 2106 * we will take a random value obtained above, mask off 2107 * the upper bits, and add in the known value. We also 2108 * add in a constant to ensure that we are at least a 2109 * certain distance from the original value. 2110 * 2111 * This is used when an old connection is in timed wait 2112 * and we have a new one coming in, for instance. 2113 */ 2114 if (addin != 0) { 2115 #ifdef TCPISS_DEBUG 2116 printf("Random %08x, ", tcp_iss); 2117 #endif 2118 tcp_iss &= TCP_ISS_RANDOM_MASK; 2119 tcp_iss += addin + TCP_ISSINCR; 2120 #ifdef TCPISS_DEBUG 2121 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2122 #endif 2123 } else { 2124 tcp_iss &= TCP_ISS_RANDOM_MASK; 2125 tcp_iss += tcp_iss_seq; 2126 tcp_iss_seq += TCP_ISSINCR; 2127 #ifdef TCPISS_DEBUG 2128 printf("ISS %08x\n", tcp_iss); 2129 #endif 2130 } 2131 } 2132 2133 if (tcp_compat_42) { 2134 /* 2135 * Limit it to the positive range for really old TCP 2136 * implementations. 2137 * Just AND off the top bit instead of checking if 2138 * is set first - saves a branch 50% of the time. 2139 */ 2140 tcp_iss &= 0x7fffffff; /* XXX */ 2141 } 2142 2143 return (tcp_iss); 2144 } 2145 2146 #if defined(IPSEC) || defined(FAST_IPSEC) 2147 /* compute ESP/AH header size for TCP, including outer IP header. */ 2148 size_t 2149 ipsec4_hdrsiz_tcp(tp) 2150 struct tcpcb *tp; 2151 { 2152 struct inpcb *inp; 2153 size_t hdrsiz; 2154 2155 /* XXX mapped addr case (tp->t_in6pcb) */ 2156 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2157 return 0; 2158 switch (tp->t_family) { 2159 case AF_INET: 2160 /* XXX: should use currect direction. */ 2161 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2162 break; 2163 default: 2164 hdrsiz = 0; 2165 break; 2166 } 2167 2168 return hdrsiz; 2169 } 2170 2171 #ifdef INET6 2172 size_t 2173 ipsec6_hdrsiz_tcp(tp) 2174 struct tcpcb *tp; 2175 { 2176 struct in6pcb *in6p; 2177 size_t hdrsiz; 2178 2179 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2180 return 0; 2181 switch (tp->t_family) { 2182 case AF_INET6: 2183 /* XXX: should use currect direction. */ 2184 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2185 break; 2186 case AF_INET: 2187 /* mapped address case - tricky */ 2188 default: 2189 hdrsiz = 0; 2190 break; 2191 } 2192 2193 return hdrsiz; 2194 } 2195 #endif 2196 #endif /*IPSEC*/ 2197 2198 /* 2199 * Determine the length of the TCP options for this connection. 2200 * 2201 * XXX: What do we do for SACK, when we add that? Just reserve 2202 * all of the space? Otherwise we can't exactly be incrementing 2203 * cwnd by an amount that varies depending on the amount we last 2204 * had to SACK! 2205 */ 2206 2207 u_int 2208 tcp_optlen(tp) 2209 struct tcpcb *tp; 2210 { 2211 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2212 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2213 return TCPOLEN_TSTAMP_APPA; 2214 else 2215 return 0; 2216 } 2217
Indexes created Sun Sep 21 16:09:51 GMT 2025