tcp_subr.c revision 1.178
1 /* $NetBSD: tcp_subr.c,v 1.178 2005/02/02 21:41:55 perry 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 1.178 2005/02/02 21:41:55 perry 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 #include <netkey/key.h> 159 #endif /*IPSEC*/ 160 161 #ifdef FAST_IPSEC 162 #include <netipsec/ipsec.h> 163 #include <netipsec/xform.h> 164 #ifdef INET6 165 #include <netipsec/ipsec6.h> 166 #endif 167 #include <netipsec/key.h> 168 #endif /* FAST_IPSEC*/ 169 170 171 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 172 struct tcpstat tcpstat; /* tcp statistics */ 173 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 174 175 /* patchable/settable parameters for tcp */ 176 int tcp_mssdflt = TCP_MSS; 177 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 178 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 179 #if NRND > 0 180 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 181 #endif 182 int tcp_do_sack = 1; /* selective acknowledgement */ 183 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 184 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 185 int tcp_do_newreno = 1; /* Use the New Reno algorithms */ 186 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 187 #ifndef TCP_INIT_WIN 188 #define TCP_INIT_WIN 0 /* initial slow start window */ 189 #endif 190 #ifndef TCP_INIT_WIN_LOCAL 191 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 192 #endif 193 int tcp_init_win = TCP_INIT_WIN; 194 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 195 int tcp_mss_ifmtu = 0; 196 #ifdef TCP_COMPAT_42 197 int tcp_compat_42 = 1; 198 #else 199 int tcp_compat_42 = 0; 200 #endif 201 int tcp_rst_ppslim = 100; /* 100pps */ 202 int tcp_ackdrop_ppslim = 100; /* 100pps */ 203 int tcp_do_loopback_cksum = 0; 204 205 /* tcb hash */ 206 #ifndef TCBHASHSIZE 207 #define TCBHASHSIZE 128 208 #endif 209 int tcbhashsize = TCBHASHSIZE; 210 211 /* syn hash parameters */ 212 #define TCP_SYN_HASH_SIZE 293 213 #define TCP_SYN_BUCKET_SIZE 35 214 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 215 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 216 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 217 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 218 219 int tcp_freeq(struct tcpcb *); 220 221 #ifdef INET 222 void tcp_mtudisc_callback(struct in_addr); 223 #endif 224 #ifdef INET6 225 void tcp6_mtudisc_callback(struct in6_addr *); 226 #endif 227 228 void tcp_mtudisc(struct inpcb *, int); 229 #ifdef INET6 230 void tcp6_mtudisc(struct in6pcb *, int); 231 #endif 232 233 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL); 234 235 #ifdef TCP_CSUM_COUNTERS 236 #include <sys/device.h> 237 238 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 239 NULL, "tcp", "hwcsum bad"); 240 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 241 NULL, "tcp", "hwcsum ok"); 242 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 243 NULL, "tcp", "hwcsum data"); 244 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 245 NULL, "tcp", "swcsum"); 246 247 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 248 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 249 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 250 EVCNT_ATTACH_STATIC(tcp_swcsum); 251 #endif /* TCP_CSUM_COUNTERS */ 252 253 254 #ifdef TCP_OUTPUT_COUNTERS 255 #include <sys/device.h> 256 257 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 258 NULL, "tcp", "output big header"); 259 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp", "output predict hit"); 261 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 NULL, "tcp", "output predict miss"); 263 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp", "output copy small"); 265 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 NULL, "tcp", "output copy big"); 267 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 268 NULL, "tcp", "output reference big"); 269 270 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 271 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 272 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 273 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 274 EVCNT_ATTACH_STATIC(tcp_output_copybig); 275 EVCNT_ATTACH_STATIC(tcp_output_refbig); 276 277 #endif /* TCP_OUTPUT_COUNTERS */ 278 279 #ifdef TCP_REASS_COUNTERS 280 #include <sys/device.h> 281 282 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 NULL, "tcp_reass", "calls"); 284 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 285 &tcp_reass_, "tcp_reass", "insert into empty queue"); 286 struct evcnt tcp_reass_iteration[8] = { 287 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 288 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 289 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 290 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 291 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 292 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 293 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 294 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 295 }; 296 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 297 &tcp_reass_, "tcp_reass", "prepend to first"); 298 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 299 &tcp_reass_, "tcp_reass", "prepend"); 300 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 301 &tcp_reass_, "tcp_reass", "insert"); 302 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 303 &tcp_reass_, "tcp_reass", "insert at tail"); 304 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 305 &tcp_reass_, "tcp_reass", "append"); 306 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 307 &tcp_reass_, "tcp_reass", "append to tail fragment"); 308 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 309 &tcp_reass_, "tcp_reass", "overlap at end"); 310 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 311 &tcp_reass_, "tcp_reass", "overlap at start"); 312 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 313 &tcp_reass_, "tcp_reass", "duplicate segment"); 314 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 315 &tcp_reass_, "tcp_reass", "duplicate fragment"); 316 317 EVCNT_ATTACH_STATIC(tcp_reass_); 318 EVCNT_ATTACH_STATIC(tcp_reass_empty); 319 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 320 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 321 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 322 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 323 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 324 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 325 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 326 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 327 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 328 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 329 EVCNT_ATTACH_STATIC(tcp_reass_insert); 330 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 331 EVCNT_ATTACH_STATIC(tcp_reass_append); 332 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 333 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 334 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 335 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 336 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 337 338 #endif /* TCP_REASS_COUNTERS */ 339 340 #ifdef MBUFTRACE 341 struct mowner tcp_mowner = { "tcp" }; 342 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 343 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 344 #endif 345 346 /* 347 * Tcp initialization 348 */ 349 void 350 tcp_init() 351 { 352 int hlen; 353 354 /* Initialize the TCPCB template. */ 355 tcp_tcpcb_template(); 356 357 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 358 359 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 360 #ifdef INET6 361 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 362 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 363 #endif 364 if (max_protohdr < hlen) 365 max_protohdr = hlen; 366 if (max_linkhdr + hlen > MHLEN) 367 panic("tcp_init"); 368 369 #ifdef INET 370 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 371 #endif 372 #ifdef INET6 373 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 374 #endif 375 376 /* Initialize timer state. */ 377 tcp_timer_init(); 378 379 /* Initialize the compressed state engine. */ 380 syn_cache_init(); 381 382 MOWNER_ATTACH(&tcp_tx_mowner); 383 MOWNER_ATTACH(&tcp_rx_mowner); 384 MOWNER_ATTACH(&tcp_mowner); 385 } 386 387 /* 388 * Create template to be used to send tcp packets on a connection. 389 * Call after host entry created, allocates an mbuf and fills 390 * in a skeletal tcp/ip header, minimizing the amount of work 391 * necessary when the connection is used. 392 */ 393 struct mbuf * 394 tcp_template(tp) 395 struct tcpcb *tp; 396 { 397 struct inpcb *inp = tp->t_inpcb; 398 #ifdef INET6 399 struct in6pcb *in6p = tp->t_in6pcb; 400 #endif 401 struct tcphdr *n; 402 struct mbuf *m; 403 int hlen; 404 405 switch (tp->t_family) { 406 case AF_INET: 407 hlen = sizeof(struct ip); 408 if (inp) 409 break; 410 #ifdef INET6 411 if (in6p) { 412 /* mapped addr case */ 413 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 414 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 415 break; 416 } 417 #endif 418 return NULL; /*EINVAL*/ 419 #ifdef INET6 420 case AF_INET6: 421 hlen = sizeof(struct ip6_hdr); 422 if (in6p) { 423 /* more sainty check? */ 424 break; 425 } 426 return NULL; /*EINVAL*/ 427 #endif 428 default: 429 hlen = 0; /*pacify gcc*/ 430 return NULL; /*EAFNOSUPPORT*/ 431 } 432 #ifdef DIAGNOSTIC 433 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 434 panic("mclbytes too small for t_template"); 435 #endif 436 m = tp->t_template; 437 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 438 ; 439 else { 440 if (m) 441 m_freem(m); 442 m = tp->t_template = NULL; 443 MGETHDR(m, M_DONTWAIT, MT_HEADER); 444 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 445 MCLGET(m, M_DONTWAIT); 446 if ((m->m_flags & M_EXT) == 0) { 447 m_free(m); 448 m = NULL; 449 } 450 } 451 if (m == NULL) 452 return NULL; 453 MCLAIM(m, &tcp_mowner); 454 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 455 } 456 457 bzero(mtod(m, caddr_t), m->m_len); 458 459 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 460 461 switch (tp->t_family) { 462 case AF_INET: 463 { 464 struct ipovly *ipov; 465 mtod(m, struct ip *)->ip_v = 4; 466 mtod(m, struct ip *)->ip_hl = hlen >> 2; 467 ipov = mtod(m, struct ipovly *); 468 ipov->ih_pr = IPPROTO_TCP; 469 ipov->ih_len = htons(sizeof(struct tcphdr)); 470 if (inp) { 471 ipov->ih_src = inp->inp_laddr; 472 ipov->ih_dst = inp->inp_faddr; 473 } 474 #ifdef INET6 475 else if (in6p) { 476 /* mapped addr case */ 477 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 478 sizeof(ipov->ih_src)); 479 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 480 sizeof(ipov->ih_dst)); 481 } 482 #endif 483 /* 484 * Compute the pseudo-header portion of the checksum 485 * now. We incrementally add in the TCP option and 486 * payload lengths later, and then compute the TCP 487 * checksum right before the packet is sent off onto 488 * the wire. 489 */ 490 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 491 ipov->ih_dst.s_addr, 492 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 493 break; 494 } 495 #ifdef INET6 496 case AF_INET6: 497 { 498 struct ip6_hdr *ip6; 499 mtod(m, struct ip *)->ip_v = 6; 500 ip6 = mtod(m, struct ip6_hdr *); 501 ip6->ip6_nxt = IPPROTO_TCP; 502 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 503 ip6->ip6_src = in6p->in6p_laddr; 504 ip6->ip6_dst = in6p->in6p_faddr; 505 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 506 if (ip6_auto_flowlabel) { 507 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 508 ip6->ip6_flow |= 509 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 510 } 511 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 512 ip6->ip6_vfc |= IPV6_VERSION; 513 514 /* 515 * Compute the pseudo-header portion of the checksum 516 * now. We incrementally add in the TCP option and 517 * payload lengths later, and then compute the TCP 518 * checksum right before the packet is sent off onto 519 * the wire. 520 */ 521 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 522 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 523 htonl(IPPROTO_TCP)); 524 break; 525 } 526 #endif 527 } 528 if (inp) { 529 n->th_sport = inp->inp_lport; 530 n->th_dport = inp->inp_fport; 531 } 532 #ifdef INET6 533 else if (in6p) { 534 n->th_sport = in6p->in6p_lport; 535 n->th_dport = in6p->in6p_fport; 536 } 537 #endif 538 n->th_seq = 0; 539 n->th_ack = 0; 540 n->th_x2 = 0; 541 n->th_off = 5; 542 n->th_flags = 0; 543 n->th_win = 0; 544 n->th_urp = 0; 545 return (m); 546 } 547 548 /* 549 * Send a single message to the TCP at address specified by 550 * the given TCP/IP header. If m == 0, then we make a copy 551 * of the tcpiphdr at ti and send directly to the addressed host. 552 * This is used to force keep alive messages out using the TCP 553 * template for a connection tp->t_template. If flags are given 554 * then we send a message back to the TCP which originated the 555 * segment ti, and discard the mbuf containing it and any other 556 * attached mbufs. 557 * 558 * In any case the ack and sequence number of the transmitted 559 * segment are as specified by the parameters. 560 */ 561 int 562 tcp_respond(tp, template, m, th0, ack, seq, flags) 563 struct tcpcb *tp; 564 struct mbuf *template; 565 struct mbuf *m; 566 struct tcphdr *th0; 567 tcp_seq ack, seq; 568 int flags; 569 { 570 struct route *ro; 571 int error, tlen, win = 0; 572 int hlen; 573 struct ip *ip; 574 #ifdef INET6 575 struct ip6_hdr *ip6; 576 #endif 577 int family; /* family on packet, not inpcb/in6pcb! */ 578 struct tcphdr *th; 579 struct socket *so; 580 581 if (tp != NULL && (flags & TH_RST) == 0) { 582 #ifdef DIAGNOSTIC 583 if (tp->t_inpcb && tp->t_in6pcb) 584 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 585 #endif 586 #ifdef INET 587 if (tp->t_inpcb) 588 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 589 #endif 590 #ifdef INET6 591 if (tp->t_in6pcb) 592 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 593 #endif 594 } 595 596 th = NULL; /* Quell uninitialized warning */ 597 ip = NULL; 598 #ifdef INET6 599 ip6 = NULL; 600 #endif 601 if (m == 0) { 602 if (!template) 603 return EINVAL; 604 605 /* get family information from template */ 606 switch (mtod(template, struct ip *)->ip_v) { 607 case 4: 608 family = AF_INET; 609 hlen = sizeof(struct ip); 610 break; 611 #ifdef INET6 612 case 6: 613 family = AF_INET6; 614 hlen = sizeof(struct ip6_hdr); 615 break; 616 #endif 617 default: 618 return EAFNOSUPPORT; 619 } 620 621 MGETHDR(m, M_DONTWAIT, MT_HEADER); 622 if (m) { 623 MCLAIM(m, &tcp_tx_mowner); 624 MCLGET(m, M_DONTWAIT); 625 if ((m->m_flags & M_EXT) == 0) { 626 m_free(m); 627 m = NULL; 628 } 629 } 630 if (m == NULL) 631 return (ENOBUFS); 632 633 if (tcp_compat_42) 634 tlen = 1; 635 else 636 tlen = 0; 637 638 m->m_data += max_linkhdr; 639 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 640 template->m_len); 641 switch (family) { 642 case AF_INET: 643 ip = mtod(m, struct ip *); 644 th = (struct tcphdr *)(ip + 1); 645 break; 646 #ifdef INET6 647 case AF_INET6: 648 ip6 = mtod(m, struct ip6_hdr *); 649 th = (struct tcphdr *)(ip6 + 1); 650 break; 651 #endif 652 #if 0 653 default: 654 /* noone will visit here */ 655 m_freem(m); 656 return EAFNOSUPPORT; 657 #endif 658 } 659 flags = TH_ACK; 660 } else { 661 662 if ((m->m_flags & M_PKTHDR) == 0) { 663 #if 0 664 printf("non PKTHDR to tcp_respond\n"); 665 #endif 666 m_freem(m); 667 return EINVAL; 668 } 669 #ifdef DIAGNOSTIC 670 if (!th0) 671 panic("th0 == NULL in tcp_respond"); 672 #endif 673 674 /* get family information from m */ 675 switch (mtod(m, struct ip *)->ip_v) { 676 case 4: 677 family = AF_INET; 678 hlen = sizeof(struct ip); 679 ip = mtod(m, struct ip *); 680 break; 681 #ifdef INET6 682 case 6: 683 family = AF_INET6; 684 hlen = sizeof(struct ip6_hdr); 685 ip6 = mtod(m, struct ip6_hdr *); 686 break; 687 #endif 688 default: 689 m_freem(m); 690 return EAFNOSUPPORT; 691 } 692 /* clear h/w csum flags inherited from rx packet */ 693 m->m_pkthdr.csum_flags = 0; 694 695 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 696 tlen = sizeof(*th0); 697 else 698 tlen = th0->th_off << 2; 699 700 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 701 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 702 m->m_len = hlen + tlen; 703 m_freem(m->m_next); 704 m->m_next = NULL; 705 } else { 706 struct mbuf *n; 707 708 #ifdef DIAGNOSTIC 709 if (max_linkhdr + hlen + tlen > MCLBYTES) { 710 m_freem(m); 711 return EMSGSIZE; 712 } 713 #endif 714 MGETHDR(n, M_DONTWAIT, MT_HEADER); 715 if (n && max_linkhdr + hlen + tlen > MHLEN) { 716 MCLGET(n, M_DONTWAIT); 717 if ((n->m_flags & M_EXT) == 0) { 718 m_freem(n); 719 n = NULL; 720 } 721 } 722 if (!n) { 723 m_freem(m); 724 return ENOBUFS; 725 } 726 727 MCLAIM(n, &tcp_tx_mowner); 728 n->m_data += max_linkhdr; 729 n->m_len = hlen + tlen; 730 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 731 m_copyback(n, hlen, tlen, (caddr_t)th0); 732 733 m_freem(m); 734 m = n; 735 n = NULL; 736 } 737 738 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 739 switch (family) { 740 case AF_INET: 741 ip = mtod(m, struct ip *); 742 th = (struct tcphdr *)(ip + 1); 743 ip->ip_p = IPPROTO_TCP; 744 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 745 ip->ip_p = IPPROTO_TCP; 746 break; 747 #ifdef INET6 748 case AF_INET6: 749 ip6 = mtod(m, struct ip6_hdr *); 750 th = (struct tcphdr *)(ip6 + 1); 751 ip6->ip6_nxt = IPPROTO_TCP; 752 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 753 ip6->ip6_nxt = IPPROTO_TCP; 754 break; 755 #endif 756 #if 0 757 default: 758 /* noone will visit here */ 759 m_freem(m); 760 return EAFNOSUPPORT; 761 #endif 762 } 763 xchg(th->th_dport, th->th_sport, u_int16_t); 764 #undef xchg 765 tlen = 0; /*be friendly with the following code*/ 766 } 767 th->th_seq = htonl(seq); 768 th->th_ack = htonl(ack); 769 th->th_x2 = 0; 770 if ((flags & TH_SYN) == 0) { 771 if (tp) 772 win >>= tp->rcv_scale; 773 if (win > TCP_MAXWIN) 774 win = TCP_MAXWIN; 775 th->th_win = htons((u_int16_t)win); 776 th->th_off = sizeof (struct tcphdr) >> 2; 777 tlen += sizeof(*th); 778 } else 779 tlen += th->th_off << 2; 780 m->m_len = hlen + tlen; 781 m->m_pkthdr.len = hlen + tlen; 782 m->m_pkthdr.rcvif = (struct ifnet *) 0; 783 th->th_flags = flags; 784 th->th_urp = 0; 785 786 switch (family) { 787 #ifdef INET 788 case AF_INET: 789 { 790 struct ipovly *ipov = (struct ipovly *)ip; 791 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 792 ipov->ih_len = htons((u_int16_t)tlen); 793 794 th->th_sum = 0; 795 th->th_sum = in_cksum(m, hlen + tlen); 796 ip->ip_len = htons(hlen + tlen); 797 ip->ip_ttl = ip_defttl; 798 break; 799 } 800 #endif 801 #ifdef INET6 802 case AF_INET6: 803 { 804 th->th_sum = 0; 805 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 806 tlen); 807 ip6->ip6_plen = ntohs(tlen); 808 if (tp && tp->t_in6pcb) { 809 struct ifnet *oifp; 810 ro = (struct route *)&tp->t_in6pcb->in6p_route; 811 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 812 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 813 } else 814 ip6->ip6_hlim = ip6_defhlim; 815 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 816 if (ip6_auto_flowlabel) { 817 ip6->ip6_flow |= 818 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 819 } 820 break; 821 } 822 #endif 823 } 824 825 if (tp && tp->t_inpcb) 826 so = tp->t_inpcb->inp_socket; 827 #ifdef INET6 828 else if (tp && tp->t_in6pcb) 829 so = tp->t_in6pcb->in6p_socket; 830 #endif 831 else 832 so = NULL; 833 834 if (tp != NULL && tp->t_inpcb != NULL) { 835 ro = &tp->t_inpcb->inp_route; 836 #ifdef DIAGNOSTIC 837 if (family != AF_INET) 838 panic("tcp_respond: address family mismatch"); 839 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 840 panic("tcp_respond: ip_dst %x != inp_faddr %x", 841 ntohl(ip->ip_dst.s_addr), 842 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 843 } 844 #endif 845 } 846 #ifdef INET6 847 else if (tp != NULL && tp->t_in6pcb != NULL) { 848 ro = (struct route *)&tp->t_in6pcb->in6p_route; 849 #ifdef DIAGNOSTIC 850 if (family == AF_INET) { 851 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 852 panic("tcp_respond: not mapped addr"); 853 if (bcmp(&ip->ip_dst, 854 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 855 sizeof(ip->ip_dst)) != 0) { 856 panic("tcp_respond: ip_dst != in6p_faddr"); 857 } 858 } else if (family == AF_INET6) { 859 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 860 &tp->t_in6pcb->in6p_faddr)) 861 panic("tcp_respond: ip6_dst != in6p_faddr"); 862 } else 863 panic("tcp_respond: address family mismatch"); 864 #endif 865 } 866 #endif 867 else 868 ro = NULL; 869 870 switch (family) { 871 #ifdef INET 872 case AF_INET: 873 error = ip_output(m, NULL, ro, 874 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 875 (struct ip_moptions *)0, so); 876 break; 877 #endif 878 #ifdef INET6 879 case AF_INET6: 880 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 881 (struct ip6_moptions *)0, so, NULL); 882 break; 883 #endif 884 default: 885 error = EAFNOSUPPORT; 886 break; 887 } 888 889 return (error); 890 } 891 892 /* 893 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 894 * a bunch of members individually, we maintain this template for the 895 * static and mostly-static components of the TCPCB, and copy it into 896 * the new TCPCB instead. 897 */ 898 static struct tcpcb tcpcb_template = { 899 /* 900 * If TCP_NTIMERS ever changes, we'll need to update this 901 * initializer. 902 */ 903 .t_timer = { 904 CALLOUT_INITIALIZER, 905 CALLOUT_INITIALIZER, 906 CALLOUT_INITIALIZER, 907 CALLOUT_INITIALIZER, 908 }, 909 .t_delack_ch = CALLOUT_INITIALIZER, 910 911 .t_srtt = TCPTV_SRTTBASE, 912 .t_rttmin = TCPTV_MIN, 913 914 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 915 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 916 }; 917 918 /* 919 * Updates the TCPCB template whenever a parameter that would affect 920 * the template is changed. 921 */ 922 void 923 tcp_tcpcb_template(void) 924 { 925 struct tcpcb *tp = &tcpcb_template; 926 int flags; 927 928 tp->t_peermss = tcp_mssdflt; 929 tp->t_ourmss = tcp_mssdflt; 930 tp->t_segsz = tcp_mssdflt; 931 932 flags = 0; 933 if (tcp_do_rfc1323 && tcp_do_win_scale) 934 flags |= TF_REQ_SCALE; 935 if (tcp_do_rfc1323 && tcp_do_timestamps) 936 flags |= TF_REQ_TSTMP; 937 if (tcp_do_sack == 2) 938 flags |= TF_WILL_SACK; 939 else if (tcp_do_sack == 1) 940 flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 941 flags |= TF_CANT_TXSACK; 942 tp->t_flags = flags; 943 944 /* 945 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 946 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 947 * reasonable initial retransmit time. 948 */ 949 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 950 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 951 TCPTV_MIN, TCPTV_REXMTMAX); 952 } 953 954 /* 955 * Create a new TCP control block, making an 956 * empty reassembly queue and hooking it to the argument 957 * protocol control block. 958 */ 959 struct tcpcb * 960 tcp_newtcpcb(family, aux) 961 int family; /* selects inpcb, or in6pcb */ 962 void *aux; 963 { 964 struct tcpcb *tp; 965 int i; 966 967 /* XXX Consider using a pool_cache for speed. */ 968 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 969 if (tp == NULL) 970 return (NULL); 971 memcpy(tp, &tcpcb_template, sizeof(*tp)); 972 TAILQ_INIT(&tp->segq); 973 TAILQ_INIT(&tp->timeq); 974 tp->t_family = family; /* may be overridden later on */ 975 LIST_INIT(&tp->t_sc); /* XXX can template this */ 976 977 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 978 for (i = 0; i < TCPT_NTIMERS; i++) 979 TCP_TIMER_INIT(tp, i); 980 981 switch (family) { 982 case AF_INET: 983 { 984 struct inpcb *inp = (struct inpcb *)aux; 985 986 inp->inp_ip.ip_ttl = ip_defttl; 987 inp->inp_ppcb = (caddr_t)tp; 988 989 tp->t_inpcb = inp; 990 tp->t_mtudisc = ip_mtudisc; 991 break; 992 } 993 #ifdef INET6 994 case AF_INET6: 995 { 996 struct in6pcb *in6p = (struct in6pcb *)aux; 997 998 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 999 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1000 : NULL); 1001 in6p->in6p_ppcb = (caddr_t)tp; 1002 1003 tp->t_in6pcb = in6p; 1004 /* for IPv6, always try to run path MTU discovery */ 1005 tp->t_mtudisc = 1; 1006 break; 1007 } 1008 #endif /* INET6 */ 1009 default: 1010 pool_put(&tcpcb_pool, tp); 1011 return (NULL); 1012 } 1013 1014 /* 1015 * Initialize our timebase. When we send timestamps, we take 1016 * the delta from tcp_now -- this means each connection always 1017 * gets a timebase of 0, which makes it, among other things, 1018 * more difficult to determine how long a system has been up, 1019 * and thus how many TCP sequence increments have occurred. 1020 */ 1021 tp->ts_timebase = tcp_now; 1022 1023 return (tp); 1024 } 1025 1026 /* 1027 * Drop a TCP connection, reporting 1028 * the specified error. If connection is synchronized, 1029 * then send a RST to peer. 1030 */ 1031 struct tcpcb * 1032 tcp_drop(tp, errno) 1033 struct tcpcb *tp; 1034 int errno; 1035 { 1036 struct socket *so = NULL; 1037 1038 #ifdef DIAGNOSTIC 1039 if (tp->t_inpcb && tp->t_in6pcb) 1040 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1041 #endif 1042 #ifdef INET 1043 if (tp->t_inpcb) 1044 so = tp->t_inpcb->inp_socket; 1045 #endif 1046 #ifdef INET6 1047 if (tp->t_in6pcb) 1048 so = tp->t_in6pcb->in6p_socket; 1049 #endif 1050 if (!so) 1051 return NULL; 1052 1053 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1054 tp->t_state = TCPS_CLOSED; 1055 (void) tcp_output(tp); 1056 tcpstat.tcps_drops++; 1057 } else 1058 tcpstat.tcps_conndrops++; 1059 if (errno == ETIMEDOUT && tp->t_softerror) 1060 errno = tp->t_softerror; 1061 so->so_error = errno; 1062 return (tcp_close(tp)); 1063 } 1064 1065 /* 1066 * Return whether this tcpcb is marked as dead, indicating 1067 * to the calling timer function that no further action should 1068 * be taken, as we are about to release this tcpcb. The release 1069 * of the storage will be done if this is the last timer running. 1070 * 1071 * This should be called from the callout handler function after 1072 * callout_ack() is done, so that the number of invoking timer 1073 * functions is 0. 1074 */ 1075 int 1076 tcp_isdead(tp) 1077 struct tcpcb *tp; 1078 { 1079 int dead = (tp->t_flags & TF_DEAD); 1080 1081 if (__predict_false(dead)) { 1082 if (tcp_timers_invoking(tp) > 0) 1083 /* not quite there yet -- count separately? */ 1084 return dead; 1085 tcpstat.tcps_delayed_free++; 1086 pool_put(&tcpcb_pool, tp); 1087 } 1088 return dead; 1089 } 1090 1091 /* 1092 * Close a TCP control block: 1093 * discard all space held by the tcp 1094 * discard internet protocol block 1095 * wake up any sleepers 1096 */ 1097 struct tcpcb * 1098 tcp_close(tp) 1099 struct tcpcb *tp; 1100 { 1101 struct inpcb *inp; 1102 #ifdef INET6 1103 struct in6pcb *in6p; 1104 #endif 1105 struct socket *so; 1106 #ifdef RTV_RTT 1107 struct rtentry *rt; 1108 #endif 1109 struct route *ro; 1110 1111 inp = tp->t_inpcb; 1112 #ifdef INET6 1113 in6p = tp->t_in6pcb; 1114 #endif 1115 so = NULL; 1116 ro = NULL; 1117 if (inp) { 1118 so = inp->inp_socket; 1119 ro = &inp->inp_route; 1120 } 1121 #ifdef INET6 1122 else if (in6p) { 1123 so = in6p->in6p_socket; 1124 ro = (struct route *)&in6p->in6p_route; 1125 } 1126 #endif 1127 1128 #ifdef RTV_RTT 1129 /* 1130 * If we sent enough data to get some meaningful characteristics, 1131 * save them in the routing entry. 'Enough' is arbitrarily 1132 * defined as the sendpipesize (default 4K) * 16. This would 1133 * give us 16 rtt samples assuming we only get one sample per 1134 * window (the usual case on a long haul net). 16 samples is 1135 * enough for the srtt filter to converge to within 5% of the correct 1136 * value; fewer samples and we could save a very bogus rtt. 1137 * 1138 * Don't update the default route's characteristics and don't 1139 * update anything that the user "locked". 1140 */ 1141 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1142 ro && (rt = ro->ro_rt) && 1143 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1144 u_long i = 0; 1145 1146 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1147 i = tp->t_srtt * 1148 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1149 if (rt->rt_rmx.rmx_rtt && i) 1150 /* 1151 * filter this update to half the old & half 1152 * the new values, converting scale. 1153 * See route.h and tcp_var.h for a 1154 * description of the scaling constants. 1155 */ 1156 rt->rt_rmx.rmx_rtt = 1157 (rt->rt_rmx.rmx_rtt + i) / 2; 1158 else 1159 rt->rt_rmx.rmx_rtt = i; 1160 } 1161 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1162 i = tp->t_rttvar * 1163 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1164 if (rt->rt_rmx.rmx_rttvar && i) 1165 rt->rt_rmx.rmx_rttvar = 1166 (rt->rt_rmx.rmx_rttvar + i) / 2; 1167 else 1168 rt->rt_rmx.rmx_rttvar = i; 1169 } 1170 /* 1171 * update the pipelimit (ssthresh) if it has been updated 1172 * already or if a pipesize was specified & the threshhold 1173 * got below half the pipesize. I.e., wait for bad news 1174 * before we start updating, then update on both good 1175 * and bad news. 1176 */ 1177 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1178 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1179 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1180 /* 1181 * convert the limit from user data bytes to 1182 * packets then to packet data bytes. 1183 */ 1184 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1185 if (i < 2) 1186 i = 2; 1187 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1188 if (rt->rt_rmx.rmx_ssthresh) 1189 rt->rt_rmx.rmx_ssthresh = 1190 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1191 else 1192 rt->rt_rmx.rmx_ssthresh = i; 1193 } 1194 } 1195 #endif /* RTV_RTT */ 1196 /* free the reassembly queue, if any */ 1197 TCP_REASS_LOCK(tp); 1198 (void) tcp_freeq(tp); 1199 TCP_REASS_UNLOCK(tp); 1200 1201 tcp_canceltimers(tp); 1202 TCP_CLEAR_DELACK(tp); 1203 syn_cache_cleanup(tp); 1204 1205 if (tp->t_template) { 1206 m_free(tp->t_template); 1207 tp->t_template = NULL; 1208 } 1209 if (tcp_timers_invoking(tp)) 1210 tp->t_flags |= TF_DEAD; 1211 else 1212 pool_put(&tcpcb_pool, tp); 1213 1214 if (inp) { 1215 inp->inp_ppcb = 0; 1216 soisdisconnected(so); 1217 in_pcbdetach(inp); 1218 } 1219 #ifdef INET6 1220 else if (in6p) { 1221 in6p->in6p_ppcb = 0; 1222 soisdisconnected(so); 1223 in6_pcbdetach(in6p); 1224 } 1225 #endif 1226 tcpstat.tcps_closed++; 1227 return ((struct tcpcb *)0); 1228 } 1229 1230 int 1231 tcp_freeq(tp) 1232 struct tcpcb *tp; 1233 { 1234 struct ipqent *qe; 1235 int rv = 0; 1236 #ifdef TCPREASS_DEBUG 1237 int i = 0; 1238 #endif 1239 1240 TCP_REASS_LOCK_CHECK(tp); 1241 1242 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1243 #ifdef TCPREASS_DEBUG 1244 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1245 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1246 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1247 #endif 1248 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1249 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1250 m_freem(qe->ipqe_m); 1251 pool_put(&tcpipqent_pool, qe); 1252 rv = 1; 1253 } 1254 return (rv); 1255 } 1256 1257 /* 1258 * Protocol drain routine. Called when memory is in short supply. 1259 */ 1260 void 1261 tcp_drain() 1262 { 1263 struct inpcb_hdr *inph; 1264 struct tcpcb *tp; 1265 1266 /* 1267 * Free the sequence queue of all TCP connections. 1268 */ 1269 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1270 switch (inph->inph_af) { 1271 case AF_INET: 1272 tp = intotcpcb((struct inpcb *)inph); 1273 break; 1274 #ifdef INET6 1275 case AF_INET6: 1276 tp = in6totcpcb((struct in6pcb *)inph); 1277 break; 1278 #endif 1279 default: 1280 tp = NULL; 1281 break; 1282 } 1283 if (tp != NULL) { 1284 /* 1285 * We may be called from a device's interrupt 1286 * context. If the tcpcb is already busy, 1287 * just bail out now. 1288 */ 1289 if (tcp_reass_lock_try(tp) == 0) 1290 continue; 1291 if (tcp_freeq(tp)) 1292 tcpstat.tcps_connsdrained++; 1293 TCP_REASS_UNLOCK(tp); 1294 } 1295 } 1296 } 1297 1298 /* 1299 * Notify a tcp user of an asynchronous error; 1300 * store error as soft error, but wake up user 1301 * (for now, won't do anything until can select for soft error). 1302 */ 1303 void 1304 tcp_notify(inp, error) 1305 struct inpcb *inp; 1306 int error; 1307 { 1308 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1309 struct socket *so = inp->inp_socket; 1310 1311 /* 1312 * Ignore some errors if we are hooked up. 1313 * If connection hasn't completed, has retransmitted several times, 1314 * and receives a second error, give up now. This is better 1315 * than waiting a long time to establish a connection that 1316 * can never complete. 1317 */ 1318 if (tp->t_state == TCPS_ESTABLISHED && 1319 (error == EHOSTUNREACH || error == ENETUNREACH || 1320 error == EHOSTDOWN)) { 1321 return; 1322 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1323 tp->t_rxtshift > 3 && tp->t_softerror) 1324 so->so_error = error; 1325 else 1326 tp->t_softerror = error; 1327 wakeup((caddr_t) &so->so_timeo); 1328 sorwakeup(so); 1329 sowwakeup(so); 1330 } 1331 1332 #ifdef INET6 1333 void 1334 tcp6_notify(in6p, error) 1335 struct in6pcb *in6p; 1336 int error; 1337 { 1338 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1339 struct socket *so = in6p->in6p_socket; 1340 1341 /* 1342 * Ignore some errors if we are hooked up. 1343 * If connection hasn't completed, has retransmitted several times, 1344 * and receives a second error, give up now. This is better 1345 * than waiting a long time to establish a connection that 1346 * can never complete. 1347 */ 1348 if (tp->t_state == TCPS_ESTABLISHED && 1349 (error == EHOSTUNREACH || error == ENETUNREACH || 1350 error == EHOSTDOWN)) { 1351 return; 1352 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1353 tp->t_rxtshift > 3 && tp->t_softerror) 1354 so->so_error = error; 1355 else 1356 tp->t_softerror = error; 1357 wakeup((caddr_t) &so->so_timeo); 1358 sorwakeup(so); 1359 sowwakeup(so); 1360 } 1361 #endif 1362 1363 #ifdef INET6 1364 void 1365 tcp6_ctlinput(cmd, sa, d) 1366 int cmd; 1367 struct sockaddr *sa; 1368 void *d; 1369 { 1370 struct tcphdr th; 1371 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1372 int nmatch; 1373 struct ip6_hdr *ip6; 1374 const struct sockaddr_in6 *sa6_src = NULL; 1375 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1376 struct mbuf *m; 1377 int off; 1378 1379 if (sa->sa_family != AF_INET6 || 1380 sa->sa_len != sizeof(struct sockaddr_in6)) 1381 return; 1382 if ((unsigned)cmd >= PRC_NCMDS) 1383 return; 1384 else if (cmd == PRC_QUENCH) { 1385 /* XXX there's no PRC_QUENCH in IPv6 */ 1386 notify = tcp6_quench; 1387 } else if (PRC_IS_REDIRECT(cmd)) 1388 notify = in6_rtchange, d = NULL; 1389 else if (cmd == PRC_MSGSIZE) 1390 ; /* special code is present, see below */ 1391 else if (cmd == PRC_HOSTDEAD) 1392 d = NULL; 1393 else if (inet6ctlerrmap[cmd] == 0) 1394 return; 1395 1396 /* if the parameter is from icmp6, decode it. */ 1397 if (d != NULL) { 1398 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1399 m = ip6cp->ip6c_m; 1400 ip6 = ip6cp->ip6c_ip6; 1401 off = ip6cp->ip6c_off; 1402 sa6_src = ip6cp->ip6c_src; 1403 } else { 1404 m = NULL; 1405 ip6 = NULL; 1406 sa6_src = &sa6_any; 1407 off = 0; 1408 } 1409 1410 if (ip6) { 1411 /* 1412 * XXX: We assume that when ip6 is non NULL, 1413 * M and OFF are valid. 1414 */ 1415 1416 /* check if we can safely examine src and dst ports */ 1417 if (m->m_pkthdr.len < off + sizeof(th)) { 1418 if (cmd == PRC_MSGSIZE) 1419 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1420 return; 1421 } 1422 1423 bzero(&th, sizeof(th)); 1424 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1425 1426 if (cmd == PRC_MSGSIZE) { 1427 int valid = 0; 1428 1429 /* 1430 * Check to see if we have a valid TCP connection 1431 * corresponding to the address in the ICMPv6 message 1432 * payload. 1433 */ 1434 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1435 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1436 th.th_sport, 0)) 1437 valid++; 1438 1439 /* 1440 * Depending on the value of "valid" and routing table 1441 * size (mtudisc_{hi,lo}wat), we will: 1442 * - recalcurate the new MTU and create the 1443 * corresponding routing entry, or 1444 * - ignore the MTU change notification. 1445 */ 1446 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1447 1448 /* 1449 * no need to call in6_pcbnotify, it should have been 1450 * called via callback if necessary 1451 */ 1452 return; 1453 } 1454 1455 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1456 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1457 if (nmatch == 0 && syn_cache_count && 1458 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1459 inet6ctlerrmap[cmd] == ENETUNREACH || 1460 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1461 syn_cache_unreach((struct sockaddr *)sa6_src, 1462 sa, &th); 1463 } else { 1464 (void) in6_pcbnotify(&tcbtable, sa, 0, 1465 (struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1466 } 1467 } 1468 #endif 1469 1470 #ifdef INET 1471 /* assumes that ip header and tcp header are contiguous on mbuf */ 1472 void * 1473 tcp_ctlinput(cmd, sa, v) 1474 int cmd; 1475 struct sockaddr *sa; 1476 void *v; 1477 { 1478 struct ip *ip = v; 1479 struct tcphdr *th; 1480 struct icmp *icp; 1481 extern const int inetctlerrmap[]; 1482 void (*notify)(struct inpcb *, int) = tcp_notify; 1483 int errno; 1484 int nmatch; 1485 #ifdef INET6 1486 struct in6_addr src6, dst6; 1487 #endif 1488 1489 if (sa->sa_family != AF_INET || 1490 sa->sa_len != sizeof(struct sockaddr_in)) 1491 return NULL; 1492 if ((unsigned)cmd >= PRC_NCMDS) 1493 return NULL; 1494 errno = inetctlerrmap[cmd]; 1495 if (cmd == PRC_QUENCH) 1496 notify = tcp_quench; 1497 else if (PRC_IS_REDIRECT(cmd)) 1498 notify = in_rtchange, ip = 0; 1499 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1500 /* 1501 * Check to see if we have a valid TCP connection 1502 * corresponding to the address in the ICMP message 1503 * payload. 1504 * 1505 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1506 */ 1507 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1508 #ifdef INET6 1509 memset(&src6, 0, sizeof(src6)); 1510 memset(&dst6, 0, sizeof(dst6)); 1511 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1512 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1513 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1514 #endif 1515 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1516 ip->ip_src, th->th_sport) != NULL) 1517 ; 1518 #ifdef INET6 1519 else if (in6_pcblookup_connect(&tcbtable, &dst6, 1520 th->th_dport, &src6, th->th_sport, 0) != NULL) 1521 ; 1522 #endif 1523 else 1524 return NULL; 1525 1526 /* 1527 * Now that we've validated that we are actually communicating 1528 * with the host indicated in the ICMP message, locate the 1529 * ICMP header, recalculate the new MTU, and create the 1530 * corresponding routing entry. 1531 */ 1532 icp = (struct icmp *)((caddr_t)ip - 1533 offsetof(struct icmp, icmp_ip)); 1534 icmp_mtudisc(icp, ip->ip_dst); 1535 1536 return NULL; 1537 } else if (cmd == PRC_HOSTDEAD) 1538 ip = 0; 1539 else if (errno == 0) 1540 return NULL; 1541 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1542 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1543 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1544 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1545 if (nmatch == 0 && syn_cache_count && 1546 (inetctlerrmap[cmd] == EHOSTUNREACH || 1547 inetctlerrmap[cmd] == ENETUNREACH || 1548 inetctlerrmap[cmd] == EHOSTDOWN)) { 1549 struct sockaddr_in sin; 1550 bzero(&sin, sizeof(sin)); 1551 sin.sin_len = sizeof(sin); 1552 sin.sin_family = AF_INET; 1553 sin.sin_port = th->th_sport; 1554 sin.sin_addr = ip->ip_src; 1555 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1556 } 1557 1558 /* XXX mapped address case */ 1559 } else 1560 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1561 notify); 1562 return NULL; 1563 } 1564 1565 /* 1566 * When a source quence is received, we are being notifed of congestion. 1567 * Close the congestion window down to the Loss Window (one segment). 1568 * We will gradually open it again as we proceed. 1569 */ 1570 void 1571 tcp_quench(inp, errno) 1572 struct inpcb *inp; 1573 int errno; 1574 { 1575 struct tcpcb *tp = intotcpcb(inp); 1576 1577 if (tp) 1578 tp->snd_cwnd = tp->t_segsz; 1579 } 1580 #endif 1581 1582 #ifdef INET6 1583 void 1584 tcp6_quench(in6p, errno) 1585 struct in6pcb *in6p; 1586 int errno; 1587 { 1588 struct tcpcb *tp = in6totcpcb(in6p); 1589 1590 if (tp) 1591 tp->snd_cwnd = tp->t_segsz; 1592 } 1593 #endif 1594 1595 #ifdef INET 1596 /* 1597 * Path MTU Discovery handlers. 1598 */ 1599 void 1600 tcp_mtudisc_callback(faddr) 1601 struct in_addr faddr; 1602 { 1603 #ifdef INET6 1604 struct in6_addr in6; 1605 #endif 1606 1607 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1608 #ifdef INET6 1609 memset(&in6, 0, sizeof(in6)); 1610 in6.s6_addr16[5] = 0xffff; 1611 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1612 tcp6_mtudisc_callback(&in6); 1613 #endif 1614 } 1615 1616 /* 1617 * On receipt of path MTU corrections, flush old route and replace it 1618 * with the new one. Retransmit all unacknowledged packets, to ensure 1619 * that all packets will be received. 1620 */ 1621 void 1622 tcp_mtudisc(inp, errno) 1623 struct inpcb *inp; 1624 int errno; 1625 { 1626 struct tcpcb *tp = intotcpcb(inp); 1627 struct rtentry *rt = in_pcbrtentry(inp); 1628 1629 if (tp != 0) { 1630 if (rt != 0) { 1631 /* 1632 * If this was not a host route, remove and realloc. 1633 */ 1634 if ((rt->rt_flags & RTF_HOST) == 0) { 1635 in_rtchange(inp, errno); 1636 if ((rt = in_pcbrtentry(inp)) == 0) 1637 return; 1638 } 1639 1640 /* 1641 * Slow start out of the error condition. We 1642 * use the MTU because we know it's smaller 1643 * than the previously transmitted segment. 1644 * 1645 * Note: This is more conservative than the 1646 * suggestion in draft-floyd-incr-init-win-03. 1647 */ 1648 if (rt->rt_rmx.rmx_mtu != 0) 1649 tp->snd_cwnd = 1650 TCP_INITIAL_WINDOW(tcp_init_win, 1651 rt->rt_rmx.rmx_mtu); 1652 } 1653 1654 /* 1655 * Resend unacknowledged packets. 1656 */ 1657 tp->snd_nxt = tp->snd_una; 1658 tcp_output(tp); 1659 } 1660 } 1661 #endif 1662 1663 #ifdef INET6 1664 /* 1665 * Path MTU Discovery handlers. 1666 */ 1667 void 1668 tcp6_mtudisc_callback(faddr) 1669 struct in6_addr *faddr; 1670 { 1671 struct sockaddr_in6 sin6; 1672 1673 bzero(&sin6, sizeof(sin6)); 1674 sin6.sin6_family = AF_INET6; 1675 sin6.sin6_len = sizeof(struct sockaddr_in6); 1676 sin6.sin6_addr = *faddr; 1677 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1678 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1679 } 1680 1681 void 1682 tcp6_mtudisc(in6p, errno) 1683 struct in6pcb *in6p; 1684 int errno; 1685 { 1686 struct tcpcb *tp = in6totcpcb(in6p); 1687 struct rtentry *rt = in6_pcbrtentry(in6p); 1688 1689 if (tp != 0) { 1690 if (rt != 0) { 1691 /* 1692 * If this was not a host route, remove and realloc. 1693 */ 1694 if ((rt->rt_flags & RTF_HOST) == 0) { 1695 in6_rtchange(in6p, errno); 1696 if ((rt = in6_pcbrtentry(in6p)) == 0) 1697 return; 1698 } 1699 1700 /* 1701 * Slow start out of the error condition. We 1702 * use the MTU because we know it's smaller 1703 * than the previously transmitted segment. 1704 * 1705 * Note: This is more conservative than the 1706 * suggestion in draft-floyd-incr-init-win-03. 1707 */ 1708 if (rt->rt_rmx.rmx_mtu != 0) 1709 tp->snd_cwnd = 1710 TCP_INITIAL_WINDOW(tcp_init_win, 1711 rt->rt_rmx.rmx_mtu); 1712 } 1713 1714 /* 1715 * Resend unacknowledged packets. 1716 */ 1717 tp->snd_nxt = tp->snd_una; 1718 tcp_output(tp); 1719 } 1720 } 1721 #endif /* INET6 */ 1722 1723 /* 1724 * Compute the MSS to advertise to the peer. Called only during 1725 * the 3-way handshake. If we are the server (peer initiated 1726 * connection), we are called with a pointer to the interface 1727 * on which the SYN packet arrived. If we are the client (we 1728 * initiated connection), we are called with a pointer to the 1729 * interface out which this connection should go. 1730 * 1731 * NOTE: Do not subtract IP option/extension header size nor IPsec 1732 * header size from MSS advertisement. MSS option must hold the maximum 1733 * segment size we can accept, so it must always be: 1734 * max(if mtu) - ip header - tcp header 1735 */ 1736 u_long 1737 tcp_mss_to_advertise(ifp, af) 1738 const struct ifnet *ifp; 1739 int af; 1740 { 1741 extern u_long in_maxmtu; 1742 u_long mss = 0; 1743 u_long hdrsiz; 1744 1745 /* 1746 * In order to avoid defeating path MTU discovery on the peer, 1747 * we advertise the max MTU of all attached networks as our MSS, 1748 * per RFC 1191, section 3.1. 1749 * 1750 * We provide the option to advertise just the MTU of 1751 * the interface on which we hope this connection will 1752 * be receiving. If we are responding to a SYN, we 1753 * will have a pretty good idea about this, but when 1754 * initiating a connection there is a bit more doubt. 1755 * 1756 * We also need to ensure that loopback has a large enough 1757 * MSS, as the loopback MTU is never included in in_maxmtu. 1758 */ 1759 1760 if (ifp != NULL) 1761 switch (af) { 1762 case AF_INET: 1763 mss = ifp->if_mtu; 1764 break; 1765 #ifdef INET6 1766 case AF_INET6: 1767 mss = IN6_LINKMTU(ifp); 1768 break; 1769 #endif 1770 } 1771 1772 if (tcp_mss_ifmtu == 0) 1773 switch (af) { 1774 case AF_INET: 1775 mss = max(in_maxmtu, mss); 1776 break; 1777 #ifdef INET6 1778 case AF_INET6: 1779 mss = max(in6_maxmtu, mss); 1780 break; 1781 #endif 1782 } 1783 1784 switch (af) { 1785 case AF_INET: 1786 hdrsiz = sizeof(struct ip); 1787 break; 1788 #ifdef INET6 1789 case AF_INET6: 1790 hdrsiz = sizeof(struct ip6_hdr); 1791 break; 1792 #endif 1793 default: 1794 hdrsiz = 0; 1795 break; 1796 } 1797 hdrsiz += sizeof(struct tcphdr); 1798 if (mss > hdrsiz) 1799 mss -= hdrsiz; 1800 1801 mss = max(tcp_mssdflt, mss); 1802 return (mss); 1803 } 1804 1805 /* 1806 * Set connection variables based on the peer's advertised MSS. 1807 * We are passed the TCPCB for the actual connection. If we 1808 * are the server, we are called by the compressed state engine 1809 * when the 3-way handshake is complete. If we are the client, 1810 * we are called when we receive the SYN,ACK from the server. 1811 * 1812 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1813 * before this routine is called! 1814 */ 1815 void 1816 tcp_mss_from_peer(tp, offer) 1817 struct tcpcb *tp; 1818 int offer; 1819 { 1820 struct socket *so; 1821 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1822 struct rtentry *rt; 1823 #endif 1824 u_long bufsize; 1825 int mss; 1826 1827 #ifdef DIAGNOSTIC 1828 if (tp->t_inpcb && tp->t_in6pcb) 1829 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1830 #endif 1831 so = NULL; 1832 rt = NULL; 1833 #ifdef INET 1834 if (tp->t_inpcb) { 1835 so = tp->t_inpcb->inp_socket; 1836 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1837 rt = in_pcbrtentry(tp->t_inpcb); 1838 #endif 1839 } 1840 #endif 1841 #ifdef INET6 1842 if (tp->t_in6pcb) { 1843 so = tp->t_in6pcb->in6p_socket; 1844 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1845 rt = in6_pcbrtentry(tp->t_in6pcb); 1846 #endif 1847 } 1848 #endif 1849 1850 /* 1851 * As per RFC1122, use the default MSS value, unless they 1852 * sent us an offer. Do not accept offers less than 256 bytes. 1853 */ 1854 mss = tcp_mssdflt; 1855 if (offer) 1856 mss = offer; 1857 mss = max(mss, 256); /* sanity */ 1858 tp->t_peermss = mss; 1859 mss -= tcp_optlen(tp); 1860 #ifdef INET 1861 if (tp->t_inpcb) 1862 mss -= ip_optlen(tp->t_inpcb); 1863 #endif 1864 #ifdef INET6 1865 if (tp->t_in6pcb) 1866 mss -= ip6_optlen(tp->t_in6pcb); 1867 #endif 1868 1869 /* 1870 * If there's a pipesize, change the socket buffer to that size. 1871 * Make the socket buffer an integral number of MSS units. If 1872 * the MSS is larger than the socket buffer, artificially decrease 1873 * the MSS. 1874 */ 1875 #ifdef RTV_SPIPE 1876 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1877 bufsize = rt->rt_rmx.rmx_sendpipe; 1878 else 1879 #endif 1880 bufsize = so->so_snd.sb_hiwat; 1881 if (bufsize < mss) 1882 mss = bufsize; 1883 else { 1884 bufsize = roundup(bufsize, mss); 1885 if (bufsize > sb_max) 1886 bufsize = sb_max; 1887 (void) sbreserve(&so->so_snd, bufsize, so); 1888 } 1889 tp->t_segsz = mss; 1890 1891 #ifdef RTV_SSTHRESH 1892 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1893 /* 1894 * There's some sort of gateway or interface buffer 1895 * limit on the path. Use this to set the slow 1896 * start threshold, but set the threshold to no less 1897 * than 2 * MSS. 1898 */ 1899 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1900 } 1901 #endif 1902 } 1903 1904 /* 1905 * Processing necessary when a TCP connection is established. 1906 */ 1907 void 1908 tcp_established(tp) 1909 struct tcpcb *tp; 1910 { 1911 struct socket *so; 1912 #ifdef RTV_RPIPE 1913 struct rtentry *rt; 1914 #endif 1915 u_long bufsize; 1916 1917 #ifdef DIAGNOSTIC 1918 if (tp->t_inpcb && tp->t_in6pcb) 1919 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1920 #endif 1921 so = NULL; 1922 rt = NULL; 1923 #ifdef INET 1924 if (tp->t_inpcb) { 1925 so = tp->t_inpcb->inp_socket; 1926 #if defined(RTV_RPIPE) 1927 rt = in_pcbrtentry(tp->t_inpcb); 1928 #endif 1929 } 1930 #endif 1931 #ifdef INET6 1932 if (tp->t_in6pcb) { 1933 so = tp->t_in6pcb->in6p_socket; 1934 #if defined(RTV_RPIPE) 1935 rt = in6_pcbrtentry(tp->t_in6pcb); 1936 #endif 1937 } 1938 #endif 1939 1940 tp->t_state = TCPS_ESTABLISHED; 1941 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1942 1943 #ifdef RTV_RPIPE 1944 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1945 bufsize = rt->rt_rmx.rmx_recvpipe; 1946 else 1947 #endif 1948 bufsize = so->so_rcv.sb_hiwat; 1949 if (bufsize > tp->t_ourmss) { 1950 bufsize = roundup(bufsize, tp->t_ourmss); 1951 if (bufsize > sb_max) 1952 bufsize = sb_max; 1953 (void) sbreserve(&so->so_rcv, bufsize, so); 1954 } 1955 } 1956 1957 /* 1958 * Check if there's an initial rtt or rttvar. Convert from the 1959 * route-table units to scaled multiples of the slow timeout timer. 1960 * Called only during the 3-way handshake. 1961 */ 1962 void 1963 tcp_rmx_rtt(tp) 1964 struct tcpcb *tp; 1965 { 1966 #ifdef RTV_RTT 1967 struct rtentry *rt = NULL; 1968 int rtt; 1969 1970 #ifdef DIAGNOSTIC 1971 if (tp->t_inpcb && tp->t_in6pcb) 1972 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1973 #endif 1974 #ifdef INET 1975 if (tp->t_inpcb) 1976 rt = in_pcbrtentry(tp->t_inpcb); 1977 #endif 1978 #ifdef INET6 1979 if (tp->t_in6pcb) 1980 rt = in6_pcbrtentry(tp->t_in6pcb); 1981 #endif 1982 if (rt == NULL) 1983 return; 1984 1985 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1986 /* 1987 * XXX The lock bit for MTU indicates that the value 1988 * is also a minimum value; this is subject to time. 1989 */ 1990 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1991 TCPT_RANGESET(tp->t_rttmin, 1992 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1993 TCPTV_MIN, TCPTV_REXMTMAX); 1994 tp->t_srtt = rtt / 1995 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1996 if (rt->rt_rmx.rmx_rttvar) { 1997 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1998 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1999 (TCP_RTTVAR_SHIFT + 2)); 2000 } else { 2001 /* Default variation is +- 1 rtt */ 2002 tp->t_rttvar = 2003 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2004 } 2005 TCPT_RANGESET(tp->t_rxtcur, 2006 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2007 tp->t_rttmin, TCPTV_REXMTMAX); 2008 } 2009 #endif 2010 } 2011 2012 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2013 #if NRND > 0 2014 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2015 #endif 2016 2017 /* 2018 * Get a new sequence value given a tcp control block 2019 */ 2020 tcp_seq 2021 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2022 { 2023 2024 #ifdef INET 2025 if (tp->t_inpcb != NULL) { 2026 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2027 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2028 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2029 addin)); 2030 } 2031 #endif 2032 #ifdef INET6 2033 if (tp->t_in6pcb != NULL) { 2034 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2035 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2036 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2037 addin)); 2038 } 2039 #endif 2040 /* Not possible. */ 2041 panic("tcp_new_iss"); 2042 } 2043 2044 /* 2045 * This routine actually generates a new TCP initial sequence number. 2046 */ 2047 tcp_seq 2048 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2049 size_t addrsz, tcp_seq addin) 2050 { 2051 tcp_seq tcp_iss; 2052 2053 #if NRND > 0 2054 static int beenhere; 2055 2056 /* 2057 * If we haven't been here before, initialize our cryptographic 2058 * hash secret. 2059 */ 2060 if (beenhere == 0) { 2061 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2062 RND_EXTRACT_ANY); 2063 beenhere = 1; 2064 } 2065 2066 if (tcp_do_rfc1948) { 2067 MD5_CTX ctx; 2068 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2069 2070 /* 2071 * Compute the base value of the ISS. It is a hash 2072 * of (saddr, sport, daddr, dport, secret). 2073 */ 2074 MD5Init(&ctx); 2075 2076 MD5Update(&ctx, (u_char *) laddr, addrsz); 2077 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2078 2079 MD5Update(&ctx, (u_char *) faddr, addrsz); 2080 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2081 2082 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2083 2084 MD5Final(hash, &ctx); 2085 2086 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2087 2088 /* 2089 * Now increment our "timer", and add it in to 2090 * the computed value. 2091 * 2092 * XXX Use `addin'? 2093 * XXX TCP_ISSINCR too large to use? 2094 */ 2095 tcp_iss_seq += TCP_ISSINCR; 2096 #ifdef TCPISS_DEBUG 2097 printf("ISS hash 0x%08x, ", tcp_iss); 2098 #endif 2099 tcp_iss += tcp_iss_seq + addin; 2100 #ifdef TCPISS_DEBUG 2101 printf("new ISS 0x%08x\n", tcp_iss); 2102 #endif 2103 } else 2104 #endif /* NRND > 0 */ 2105 { 2106 /* 2107 * Randomize. 2108 */ 2109 #if NRND > 0 2110 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2111 #else 2112 tcp_iss = arc4random(); 2113 #endif 2114 2115 /* 2116 * If we were asked to add some amount to a known value, 2117 * we will take a random value obtained above, mask off 2118 * the upper bits, and add in the known value. We also 2119 * add in a constant to ensure that we are at least a 2120 * certain distance from the original value. 2121 * 2122 * This is used when an old connection is in timed wait 2123 * and we have a new one coming in, for instance. 2124 */ 2125 if (addin != 0) { 2126 #ifdef TCPISS_DEBUG 2127 printf("Random %08x, ", tcp_iss); 2128 #endif 2129 tcp_iss &= TCP_ISS_RANDOM_MASK; 2130 tcp_iss += addin + TCP_ISSINCR; 2131 #ifdef TCPISS_DEBUG 2132 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2133 #endif 2134 } else { 2135 tcp_iss &= TCP_ISS_RANDOM_MASK; 2136 tcp_iss += tcp_iss_seq; 2137 tcp_iss_seq += TCP_ISSINCR; 2138 #ifdef TCPISS_DEBUG 2139 printf("ISS %08x\n", tcp_iss); 2140 #endif 2141 } 2142 } 2143 2144 if (tcp_compat_42) { 2145 /* 2146 * Limit it to the positive range for really old TCP 2147 * implementations. 2148 * Just AND off the top bit instead of checking if 2149 * is set first - saves a branch 50% of the time. 2150 */ 2151 tcp_iss &= 0x7fffffff; /* XXX */ 2152 } 2153 2154 return (tcp_iss); 2155 } 2156 2157 #if defined(IPSEC) || defined(FAST_IPSEC) 2158 /* compute ESP/AH header size for TCP, including outer IP header. */ 2159 size_t 2160 ipsec4_hdrsiz_tcp(tp) 2161 struct tcpcb *tp; 2162 { 2163 struct inpcb *inp; 2164 size_t hdrsiz; 2165 2166 /* XXX mapped addr case (tp->t_in6pcb) */ 2167 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2168 return 0; 2169 switch (tp->t_family) { 2170 case AF_INET: 2171 /* XXX: should use currect direction. */ 2172 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2173 break; 2174 default: 2175 hdrsiz = 0; 2176 break; 2177 } 2178 2179 return hdrsiz; 2180 } 2181 2182 #ifdef INET6 2183 size_t 2184 ipsec6_hdrsiz_tcp(tp) 2185 struct tcpcb *tp; 2186 { 2187 struct in6pcb *in6p; 2188 size_t hdrsiz; 2189 2190 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2191 return 0; 2192 switch (tp->t_family) { 2193 case AF_INET6: 2194 /* XXX: should use currect direction. */ 2195 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2196 break; 2197 case AF_INET: 2198 /* mapped address case - tricky */ 2199 default: 2200 hdrsiz = 0; 2201 break; 2202 } 2203 2204 return hdrsiz; 2205 } 2206 #endif 2207 #endif /*IPSEC*/ 2208 2209 /* 2210 * Determine the length of the TCP options for this connection. 2211 * 2212 * XXX: What do we do for SACK, when we add that? Just reserve 2213 * all of the space? Otherwise we can't exactly be incrementing 2214 * cwnd by an amount that varies depending on the amount we last 2215 * had to SACK! 2216 */ 2217 2218 u_int 2219 tcp_optlen(tp) 2220 struct tcpcb *tp; 2221 { 2222 u_int optlen; 2223 2224 optlen = 0; 2225 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2226 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2227 optlen += TCPOLEN_TSTAMP_APPA; 2228 2229 #ifdef TCP_SIGNATURE 2230 #if defined(INET6) && defined(FAST_IPSEC) 2231 if (tp->t_family == AF_INET) 2232 #endif 2233 if (tp->t_flags & TF_SIGNATURE) 2234 optlen += TCPOLEN_SIGNATURE + 2; 2235 #endif /* TCP_SIGNATURE */ 2236 2237 return optlen; 2238 } 2239
Indexes created Sun Sep 21 16:09:51 GMT 2025