tcp_subr.c revision 1.187
1 /* $NetBSD: tcp_subr.c,v 1.187 2005/03/16 00:39:57 yamt 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 1.187 2005/03/16 00:39:57 yamt 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(void) 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(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 mtod(m, struct ip *)->ip_hl = hlen >> 2; 466 ipov = mtod(m, struct ipovly *); 467 ipov->ih_pr = IPPROTO_TCP; 468 ipov->ih_len = htons(sizeof(struct tcphdr)); 469 if (inp) { 470 ipov->ih_src = inp->inp_laddr; 471 ipov->ih_dst = inp->inp_faddr; 472 } 473 #ifdef INET6 474 else if (in6p) { 475 /* mapped addr case */ 476 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 477 sizeof(ipov->ih_src)); 478 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 479 sizeof(ipov->ih_dst)); 480 } 481 #endif 482 /* 483 * Compute the pseudo-header portion of the checksum 484 * now. We incrementally add in the TCP option and 485 * payload lengths later, and then compute the TCP 486 * checksum right before the packet is sent off onto 487 * the wire. 488 */ 489 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 490 ipov->ih_dst.s_addr, 491 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 492 break; 493 } 494 #ifdef INET6 495 case AF_INET6: 496 { 497 struct ip6_hdr *ip6; 498 mtod(m, struct ip *)->ip_v = 6; 499 ip6 = mtod(m, struct ip6_hdr *); 500 ip6->ip6_nxt = IPPROTO_TCP; 501 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 502 ip6->ip6_src = in6p->in6p_laddr; 503 ip6->ip6_dst = in6p->in6p_faddr; 504 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 505 if (ip6_auto_flowlabel) { 506 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 507 ip6->ip6_flow |= 508 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 509 } 510 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 511 ip6->ip6_vfc |= IPV6_VERSION; 512 513 /* 514 * Compute the pseudo-header portion of the checksum 515 * now. We incrementally add in the TCP option and 516 * payload lengths later, and then compute the TCP 517 * checksum right before the packet is sent off onto 518 * the wire. 519 */ 520 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 521 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 522 htonl(IPPROTO_TCP)); 523 break; 524 } 525 #endif 526 } 527 if (inp) { 528 n->th_sport = inp->inp_lport; 529 n->th_dport = inp->inp_fport; 530 } 531 #ifdef INET6 532 else if (in6p) { 533 n->th_sport = in6p->in6p_lport; 534 n->th_dport = in6p->in6p_fport; 535 } 536 #endif 537 n->th_seq = 0; 538 n->th_ack = 0; 539 n->th_x2 = 0; 540 n->th_off = 5; 541 n->th_flags = 0; 542 n->th_win = 0; 543 n->th_urp = 0; 544 return (m); 545 } 546 547 /* 548 * Send a single message to the TCP at address specified by 549 * the given TCP/IP header. If m == 0, then we make a copy 550 * of the tcpiphdr at ti and send directly to the addressed host. 551 * This is used to force keep alive messages out using the TCP 552 * template for a connection tp->t_template. If flags are given 553 * then we send a message back to the TCP which originated the 554 * segment ti, and discard the mbuf containing it and any other 555 * attached mbufs. 556 * 557 * In any case the ack and sequence number of the transmitted 558 * segment are as specified by the parameters. 559 */ 560 int 561 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 562 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 563 { 564 struct route *ro; 565 int error, tlen, win = 0; 566 int hlen; 567 struct ip *ip; 568 #ifdef INET6 569 struct ip6_hdr *ip6; 570 #endif 571 int family; /* family on packet, not inpcb/in6pcb! */ 572 struct tcphdr *th; 573 struct socket *so; 574 575 if (tp != NULL && (flags & TH_RST) == 0) { 576 #ifdef DIAGNOSTIC 577 if (tp->t_inpcb && tp->t_in6pcb) 578 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 579 #endif 580 #ifdef INET 581 if (tp->t_inpcb) 582 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 583 #endif 584 #ifdef INET6 585 if (tp->t_in6pcb) 586 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 587 #endif 588 } 589 590 th = NULL; /* Quell uninitialized warning */ 591 ip = NULL; 592 #ifdef INET6 593 ip6 = NULL; 594 #endif 595 if (m == 0) { 596 if (!template) 597 return EINVAL; 598 599 /* get family information from template */ 600 switch (mtod(template, struct ip *)->ip_v) { 601 case 4: 602 family = AF_INET; 603 hlen = sizeof(struct ip); 604 break; 605 #ifdef INET6 606 case 6: 607 family = AF_INET6; 608 hlen = sizeof(struct ip6_hdr); 609 break; 610 #endif 611 default: 612 return EAFNOSUPPORT; 613 } 614 615 MGETHDR(m, M_DONTWAIT, MT_HEADER); 616 if (m) { 617 MCLAIM(m, &tcp_tx_mowner); 618 MCLGET(m, M_DONTWAIT); 619 if ((m->m_flags & M_EXT) == 0) { 620 m_free(m); 621 m = NULL; 622 } 623 } 624 if (m == NULL) 625 return (ENOBUFS); 626 627 if (tcp_compat_42) 628 tlen = 1; 629 else 630 tlen = 0; 631 632 m->m_data += max_linkhdr; 633 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 634 template->m_len); 635 switch (family) { 636 case AF_INET: 637 ip = mtod(m, struct ip *); 638 th = (struct tcphdr *)(ip + 1); 639 break; 640 #ifdef INET6 641 case AF_INET6: 642 ip6 = mtod(m, struct ip6_hdr *); 643 th = (struct tcphdr *)(ip6 + 1); 644 break; 645 #endif 646 #if 0 647 default: 648 /* noone will visit here */ 649 m_freem(m); 650 return EAFNOSUPPORT; 651 #endif 652 } 653 flags = TH_ACK; 654 } else { 655 656 if ((m->m_flags & M_PKTHDR) == 0) { 657 #if 0 658 printf("non PKTHDR to tcp_respond\n"); 659 #endif 660 m_freem(m); 661 return EINVAL; 662 } 663 #ifdef DIAGNOSTIC 664 if (!th0) 665 panic("th0 == NULL in tcp_respond"); 666 #endif 667 668 /* get family information from m */ 669 switch (mtod(m, struct ip *)->ip_v) { 670 case 4: 671 family = AF_INET; 672 hlen = sizeof(struct ip); 673 ip = mtod(m, struct ip *); 674 break; 675 #ifdef INET6 676 case 6: 677 family = AF_INET6; 678 hlen = sizeof(struct ip6_hdr); 679 ip6 = mtod(m, struct ip6_hdr *); 680 break; 681 #endif 682 default: 683 m_freem(m); 684 return EAFNOSUPPORT; 685 } 686 /* clear h/w csum flags inherited from rx packet */ 687 m->m_pkthdr.csum_flags = 0; 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 = htons(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_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 813 } 814 break; 815 } 816 #endif 817 } 818 819 if (tp && tp->t_inpcb) 820 so = tp->t_inpcb->inp_socket; 821 #ifdef INET6 822 else if (tp && tp->t_in6pcb) 823 so = tp->t_in6pcb->in6p_socket; 824 #endif 825 else 826 so = NULL; 827 828 if (tp != NULL && tp->t_inpcb != NULL) { 829 ro = &tp->t_inpcb->inp_route; 830 #ifdef DIAGNOSTIC 831 if (family != AF_INET) 832 panic("tcp_respond: address family mismatch"); 833 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 834 panic("tcp_respond: ip_dst %x != inp_faddr %x", 835 ntohl(ip->ip_dst.s_addr), 836 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 837 } 838 #endif 839 } 840 #ifdef INET6 841 else if (tp != NULL && tp->t_in6pcb != NULL) { 842 ro = (struct route *)&tp->t_in6pcb->in6p_route; 843 #ifdef DIAGNOSTIC 844 if (family == AF_INET) { 845 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 846 panic("tcp_respond: not mapped addr"); 847 if (bcmp(&ip->ip_dst, 848 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 849 sizeof(ip->ip_dst)) != 0) { 850 panic("tcp_respond: ip_dst != in6p_faddr"); 851 } 852 } else if (family == AF_INET6) { 853 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 854 &tp->t_in6pcb->in6p_faddr)) 855 panic("tcp_respond: ip6_dst != in6p_faddr"); 856 } else 857 panic("tcp_respond: address family mismatch"); 858 #endif 859 } 860 #endif 861 else 862 ro = NULL; 863 864 switch (family) { 865 #ifdef INET 866 case AF_INET: 867 error = ip_output(m, NULL, ro, 868 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 869 (struct ip_moptions *)0, so); 870 break; 871 #endif 872 #ifdef INET6 873 case AF_INET6: 874 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 875 (struct ip6_moptions *)0, so, NULL); 876 break; 877 #endif 878 default: 879 error = EAFNOSUPPORT; 880 break; 881 } 882 883 return (error); 884 } 885 886 /* 887 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 888 * a bunch of members individually, we maintain this template for the 889 * static and mostly-static components of the TCPCB, and copy it into 890 * the new TCPCB instead. 891 */ 892 static struct tcpcb tcpcb_template = { 893 /* 894 * If TCP_NTIMERS ever changes, we'll need to update this 895 * initializer. 896 */ 897 .t_timer = { 898 CALLOUT_INITIALIZER, 899 CALLOUT_INITIALIZER, 900 CALLOUT_INITIALIZER, 901 CALLOUT_INITIALIZER, 902 }, 903 .t_delack_ch = CALLOUT_INITIALIZER, 904 905 .t_srtt = TCPTV_SRTTBASE, 906 .t_rttmin = TCPTV_MIN, 907 908 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 909 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 910 911 .t_partialacks = -1, 912 }; 913 914 /* 915 * Updates the TCPCB template whenever a parameter that would affect 916 * the template is changed. 917 */ 918 void 919 tcp_tcpcb_template(void) 920 { 921 struct tcpcb *tp = &tcpcb_template; 922 int flags; 923 924 tp->t_peermss = tcp_mssdflt; 925 tp->t_ourmss = tcp_mssdflt; 926 tp->t_segsz = tcp_mssdflt; 927 928 flags = 0; 929 if (tcp_do_rfc1323 && tcp_do_win_scale) 930 flags |= TF_REQ_SCALE; 931 if (tcp_do_rfc1323 && tcp_do_timestamps) 932 flags |= TF_REQ_TSTMP; 933 tp->t_flags = flags; 934 935 /* 936 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 937 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 938 * reasonable initial retransmit time. 939 */ 940 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 941 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 942 TCPTV_MIN, TCPTV_REXMTMAX); 943 } 944 945 /* 946 * Create a new TCP control block, making an 947 * empty reassembly queue and hooking it to the argument 948 * protocol control block. 949 */ 950 /* family selects inpcb, or in6pcb */ 951 struct tcpcb * 952 tcp_newtcpcb(int family, void *aux) 953 { 954 struct tcpcb *tp; 955 int i; 956 957 /* XXX Consider using a pool_cache for speed. */ 958 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 959 if (tp == NULL) 960 return (NULL); 961 memcpy(tp, &tcpcb_template, sizeof(*tp)); 962 TAILQ_INIT(&tp->segq); 963 TAILQ_INIT(&tp->timeq); 964 tp->t_family = family; /* may be overridden later on */ 965 TAILQ_INIT(&tp->snd_holes); 966 LIST_INIT(&tp->t_sc); /* XXX can template this */ 967 968 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 969 for (i = 0; i < TCPT_NTIMERS; i++) 970 TCP_TIMER_INIT(tp, i); 971 972 switch (family) { 973 case AF_INET: 974 { 975 struct inpcb *inp = (struct inpcb *)aux; 976 977 inp->inp_ip.ip_ttl = ip_defttl; 978 inp->inp_ppcb = (caddr_t)tp; 979 980 tp->t_inpcb = inp; 981 tp->t_mtudisc = ip_mtudisc; 982 break; 983 } 984 #ifdef INET6 985 case AF_INET6: 986 { 987 struct in6pcb *in6p = (struct in6pcb *)aux; 988 989 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 990 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 991 : NULL); 992 in6p->in6p_ppcb = (caddr_t)tp; 993 994 tp->t_in6pcb = in6p; 995 /* for IPv6, always try to run path MTU discovery */ 996 tp->t_mtudisc = 1; 997 break; 998 } 999 #endif /* INET6 */ 1000 default: 1001 pool_put(&tcpcb_pool, tp); 1002 return (NULL); 1003 } 1004 1005 /* 1006 * Initialize our timebase. When we send timestamps, we take 1007 * the delta from tcp_now -- this means each connection always 1008 * gets a timebase of 0, which makes it, among other things, 1009 * more difficult to determine how long a system has been up, 1010 * and thus how many TCP sequence increments have occurred. 1011 */ 1012 tp->ts_timebase = tcp_now; 1013 1014 return (tp); 1015 } 1016 1017 /* 1018 * Drop a TCP connection, reporting 1019 * the specified error. If connection is synchronized, 1020 * then send a RST to peer. 1021 */ 1022 struct tcpcb * 1023 tcp_drop(struct tcpcb *tp, int errno) 1024 { 1025 struct socket *so = NULL; 1026 1027 #ifdef DIAGNOSTIC 1028 if (tp->t_inpcb && tp->t_in6pcb) 1029 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1030 #endif 1031 #ifdef INET 1032 if (tp->t_inpcb) 1033 so = tp->t_inpcb->inp_socket; 1034 #endif 1035 #ifdef INET6 1036 if (tp->t_in6pcb) 1037 so = tp->t_in6pcb->in6p_socket; 1038 #endif 1039 if (!so) 1040 return NULL; 1041 1042 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1043 tp->t_state = TCPS_CLOSED; 1044 (void) tcp_output(tp); 1045 tcpstat.tcps_drops++; 1046 } else 1047 tcpstat.tcps_conndrops++; 1048 if (errno == ETIMEDOUT && tp->t_softerror) 1049 errno = tp->t_softerror; 1050 so->so_error = errno; 1051 return (tcp_close(tp)); 1052 } 1053 1054 /* 1055 * Return whether this tcpcb is marked as dead, indicating 1056 * to the calling timer function that no further action should 1057 * be taken, as we are about to release this tcpcb. The release 1058 * of the storage will be done if this is the last timer running. 1059 * 1060 * This should be called from the callout handler function after 1061 * callout_ack() is done, so that the number of invoking timer 1062 * functions is 0. 1063 */ 1064 int 1065 tcp_isdead(struct tcpcb *tp) 1066 { 1067 int dead = (tp->t_flags & TF_DEAD); 1068 1069 if (__predict_false(dead)) { 1070 if (tcp_timers_invoking(tp) > 0) 1071 /* not quite there yet -- count separately? */ 1072 return dead; 1073 tcpstat.tcps_delayed_free++; 1074 pool_put(&tcpcb_pool, tp); 1075 } 1076 return dead; 1077 } 1078 1079 /* 1080 * Close a TCP control block: 1081 * discard all space held by the tcp 1082 * discard internet protocol block 1083 * wake up any sleepers 1084 */ 1085 struct tcpcb * 1086 tcp_close(struct tcpcb *tp) 1087 { 1088 struct inpcb *inp; 1089 #ifdef INET6 1090 struct in6pcb *in6p; 1091 #endif 1092 struct socket *so; 1093 #ifdef RTV_RTT 1094 struct rtentry *rt; 1095 #endif 1096 struct route *ro; 1097 1098 inp = tp->t_inpcb; 1099 #ifdef INET6 1100 in6p = tp->t_in6pcb; 1101 #endif 1102 so = NULL; 1103 ro = NULL; 1104 if (inp) { 1105 so = inp->inp_socket; 1106 ro = &inp->inp_route; 1107 } 1108 #ifdef INET6 1109 else if (in6p) { 1110 so = in6p->in6p_socket; 1111 ro = (struct route *)&in6p->in6p_route; 1112 } 1113 #endif 1114 1115 #ifdef RTV_RTT 1116 /* 1117 * If we sent enough data to get some meaningful characteristics, 1118 * save them in the routing entry. 'Enough' is arbitrarily 1119 * defined as the sendpipesize (default 4K) * 16. This would 1120 * give us 16 rtt samples assuming we only get one sample per 1121 * window (the usual case on a long haul net). 16 samples is 1122 * enough for the srtt filter to converge to within 5% of the correct 1123 * value; fewer samples and we could save a very bogus rtt. 1124 * 1125 * Don't update the default route's characteristics and don't 1126 * update anything that the user "locked". 1127 */ 1128 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1129 ro && (rt = ro->ro_rt) && 1130 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1131 u_long i = 0; 1132 1133 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1134 i = tp->t_srtt * 1135 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1136 if (rt->rt_rmx.rmx_rtt && i) 1137 /* 1138 * filter this update to half the old & half 1139 * the new values, converting scale. 1140 * See route.h and tcp_var.h for a 1141 * description of the scaling constants. 1142 */ 1143 rt->rt_rmx.rmx_rtt = 1144 (rt->rt_rmx.rmx_rtt + i) / 2; 1145 else 1146 rt->rt_rmx.rmx_rtt = i; 1147 } 1148 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1149 i = tp->t_rttvar * 1150 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1151 if (rt->rt_rmx.rmx_rttvar && i) 1152 rt->rt_rmx.rmx_rttvar = 1153 (rt->rt_rmx.rmx_rttvar + i) / 2; 1154 else 1155 rt->rt_rmx.rmx_rttvar = i; 1156 } 1157 /* 1158 * update the pipelimit (ssthresh) if it has been updated 1159 * already or if a pipesize was specified & the threshhold 1160 * got below half the pipesize. I.e., wait for bad news 1161 * before we start updating, then update on both good 1162 * and bad news. 1163 */ 1164 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1165 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1166 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1167 /* 1168 * convert the limit from user data bytes to 1169 * packets then to packet data bytes. 1170 */ 1171 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1172 if (i < 2) 1173 i = 2; 1174 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1175 if (rt->rt_rmx.rmx_ssthresh) 1176 rt->rt_rmx.rmx_ssthresh = 1177 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1178 else 1179 rt->rt_rmx.rmx_ssthresh = i; 1180 } 1181 } 1182 #endif /* RTV_RTT */ 1183 /* free the reassembly queue, if any */ 1184 TCP_REASS_LOCK(tp); 1185 (void) tcp_freeq(tp); 1186 TCP_REASS_UNLOCK(tp); 1187 1188 /* free the SACK holes list. */ 1189 tcp_free_sackholes(tp); 1190 1191 tcp_canceltimers(tp); 1192 TCP_CLEAR_DELACK(tp); 1193 syn_cache_cleanup(tp); 1194 1195 if (tp->t_template) { 1196 m_free(tp->t_template); 1197 tp->t_template = NULL; 1198 } 1199 if (tcp_timers_invoking(tp)) 1200 tp->t_flags |= TF_DEAD; 1201 else 1202 pool_put(&tcpcb_pool, tp); 1203 1204 if (inp) { 1205 inp->inp_ppcb = 0; 1206 soisdisconnected(so); 1207 in_pcbdetach(inp); 1208 } 1209 #ifdef INET6 1210 else if (in6p) { 1211 in6p->in6p_ppcb = 0; 1212 soisdisconnected(so); 1213 in6_pcbdetach(in6p); 1214 } 1215 #endif 1216 tcpstat.tcps_closed++; 1217 return ((struct tcpcb *)0); 1218 } 1219 1220 int 1221 tcp_freeq(tp) 1222 struct tcpcb *tp; 1223 { 1224 struct ipqent *qe; 1225 int rv = 0; 1226 #ifdef TCPREASS_DEBUG 1227 int i = 0; 1228 #endif 1229 1230 TCP_REASS_LOCK_CHECK(tp); 1231 1232 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1233 #ifdef TCPREASS_DEBUG 1234 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1235 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1236 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1237 #endif 1238 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1239 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1240 m_freem(qe->ipqe_m); 1241 pool_put(&tcpipqent_pool, qe); 1242 rv = 1; 1243 } 1244 tp->t_segqlen = 0; 1245 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1246 return (rv); 1247 } 1248 1249 /* 1250 * Protocol drain routine. Called when memory is in short supply. 1251 */ 1252 void 1253 tcp_drain(void) 1254 { 1255 struct inpcb_hdr *inph; 1256 struct tcpcb *tp; 1257 1258 /* 1259 * Free the sequence queue of all TCP connections. 1260 */ 1261 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1262 switch (inph->inph_af) { 1263 case AF_INET: 1264 tp = intotcpcb((struct inpcb *)inph); 1265 break; 1266 #ifdef INET6 1267 case AF_INET6: 1268 tp = in6totcpcb((struct in6pcb *)inph); 1269 break; 1270 #endif 1271 default: 1272 tp = NULL; 1273 break; 1274 } 1275 if (tp != NULL) { 1276 /* 1277 * We may be called from a device's interrupt 1278 * context. If the tcpcb is already busy, 1279 * just bail out now. 1280 */ 1281 if (tcp_reass_lock_try(tp) == 0) 1282 continue; 1283 if (tcp_freeq(tp)) 1284 tcpstat.tcps_connsdrained++; 1285 TCP_REASS_UNLOCK(tp); 1286 } 1287 } 1288 } 1289 1290 /* 1291 * Notify a tcp user of an asynchronous error; 1292 * store error as soft error, but wake up user 1293 * (for now, won't do anything until can select for soft error). 1294 */ 1295 void 1296 tcp_notify(struct inpcb *inp, 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(struct in6pcb *in6p, int error) 1325 { 1326 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1327 struct socket *so = in6p->in6p_socket; 1328 1329 /* 1330 * Ignore some errors if we are hooked up. 1331 * If connection hasn't completed, has retransmitted several times, 1332 * and receives a second error, give up now. This is better 1333 * than waiting a long time to establish a connection that 1334 * can never complete. 1335 */ 1336 if (tp->t_state == TCPS_ESTABLISHED && 1337 (error == EHOSTUNREACH || error == ENETUNREACH || 1338 error == EHOSTDOWN)) { 1339 return; 1340 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1341 tp->t_rxtshift > 3 && tp->t_softerror) 1342 so->so_error = error; 1343 else 1344 tp->t_softerror = error; 1345 wakeup((caddr_t) &so->so_timeo); 1346 sorwakeup(so); 1347 sowwakeup(so); 1348 } 1349 #endif 1350 1351 #ifdef INET6 1352 void 1353 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1354 { 1355 struct tcphdr th; 1356 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1357 int nmatch; 1358 struct ip6_hdr *ip6; 1359 const struct sockaddr_in6 *sa6_src = NULL; 1360 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1361 struct mbuf *m; 1362 int off; 1363 1364 if (sa->sa_family != AF_INET6 || 1365 sa->sa_len != sizeof(struct sockaddr_in6)) 1366 return; 1367 if ((unsigned)cmd >= PRC_NCMDS) 1368 return; 1369 else if (cmd == PRC_QUENCH) { 1370 /* XXX there's no PRC_QUENCH in IPv6 */ 1371 notify = tcp6_quench; 1372 } else if (PRC_IS_REDIRECT(cmd)) 1373 notify = in6_rtchange, d = NULL; 1374 else if (cmd == PRC_MSGSIZE) 1375 ; /* special code is present, see below */ 1376 else if (cmd == PRC_HOSTDEAD) 1377 d = NULL; 1378 else if (inet6ctlerrmap[cmd] == 0) 1379 return; 1380 1381 /* if the parameter is from icmp6, decode it. */ 1382 if (d != NULL) { 1383 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1384 m = ip6cp->ip6c_m; 1385 ip6 = ip6cp->ip6c_ip6; 1386 off = ip6cp->ip6c_off; 1387 sa6_src = ip6cp->ip6c_src; 1388 } else { 1389 m = NULL; 1390 ip6 = NULL; 1391 sa6_src = &sa6_any; 1392 off = 0; 1393 } 1394 1395 if (ip6) { 1396 /* 1397 * XXX: We assume that when ip6 is non NULL, 1398 * M and OFF are valid. 1399 */ 1400 1401 /* check if we can safely examine src and dst ports */ 1402 if (m->m_pkthdr.len < off + sizeof(th)) { 1403 if (cmd == PRC_MSGSIZE) 1404 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1405 return; 1406 } 1407 1408 bzero(&th, sizeof(th)); 1409 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1410 1411 if (cmd == PRC_MSGSIZE) { 1412 int valid = 0; 1413 1414 /* 1415 * Check to see if we have a valid TCP connection 1416 * corresponding to the address in the ICMPv6 message 1417 * payload. 1418 */ 1419 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1420 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1421 th.th_sport, 0)) 1422 valid++; 1423 1424 /* 1425 * Depending on the value of "valid" and routing table 1426 * size (mtudisc_{hi,lo}wat), we will: 1427 * - recalcurate the new MTU and create the 1428 * corresponding routing entry, or 1429 * - ignore the MTU change notification. 1430 */ 1431 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1432 1433 /* 1434 * no need to call in6_pcbnotify, it should have been 1435 * called via callback if necessary 1436 */ 1437 return; 1438 } 1439 1440 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1441 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1442 if (nmatch == 0 && syn_cache_count && 1443 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1444 inet6ctlerrmap[cmd] == ENETUNREACH || 1445 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1446 syn_cache_unreach((struct sockaddr *)sa6_src, 1447 sa, &th); 1448 } else { 1449 (void) in6_pcbnotify(&tcbtable, sa, 0, 1450 (struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1451 } 1452 } 1453 #endif 1454 1455 #ifdef INET 1456 /* assumes that ip header and tcp header are contiguous on mbuf */ 1457 void * 1458 tcp_ctlinput(int cmd, struct sockaddr *sa, void *v) 1459 { 1460 struct ip *ip = v; 1461 struct tcphdr *th; 1462 struct icmp *icp; 1463 extern const int inetctlerrmap[]; 1464 void (*notify)(struct inpcb *, int) = tcp_notify; 1465 int errno; 1466 int nmatch; 1467 #ifdef INET6 1468 struct in6_addr src6, dst6; 1469 #endif 1470 1471 if (sa->sa_family != AF_INET || 1472 sa->sa_len != sizeof(struct sockaddr_in)) 1473 return NULL; 1474 if ((unsigned)cmd >= PRC_NCMDS) 1475 return NULL; 1476 errno = inetctlerrmap[cmd]; 1477 if (cmd == PRC_QUENCH) 1478 notify = tcp_quench; 1479 else if (PRC_IS_REDIRECT(cmd)) 1480 notify = in_rtchange, ip = 0; 1481 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1482 /* 1483 * Check to see if we have a valid TCP connection 1484 * corresponding to the address in the ICMP message 1485 * payload. 1486 * 1487 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1488 */ 1489 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1490 #ifdef INET6 1491 memset(&src6, 0, sizeof(src6)); 1492 memset(&dst6, 0, sizeof(dst6)); 1493 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1494 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1495 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1496 #endif 1497 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1498 ip->ip_src, th->th_sport) != NULL) 1499 ; 1500 #ifdef INET6 1501 else if (in6_pcblookup_connect(&tcbtable, &dst6, 1502 th->th_dport, &src6, th->th_sport, 0) != NULL) 1503 ; 1504 #endif 1505 else 1506 return NULL; 1507 1508 /* 1509 * Now that we've validated that we are actually communicating 1510 * with the host indicated in the ICMP message, locate the 1511 * ICMP header, recalculate the new MTU, and create the 1512 * corresponding routing entry. 1513 */ 1514 icp = (struct icmp *)((caddr_t)ip - 1515 offsetof(struct icmp, icmp_ip)); 1516 icmp_mtudisc(icp, ip->ip_dst); 1517 1518 return NULL; 1519 } else if (cmd == PRC_HOSTDEAD) 1520 ip = 0; 1521 else if (errno == 0) 1522 return NULL; 1523 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1524 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1525 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1526 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1527 if (nmatch == 0 && syn_cache_count && 1528 (inetctlerrmap[cmd] == EHOSTUNREACH || 1529 inetctlerrmap[cmd] == ENETUNREACH || 1530 inetctlerrmap[cmd] == EHOSTDOWN)) { 1531 struct sockaddr_in sin; 1532 bzero(&sin, sizeof(sin)); 1533 sin.sin_len = sizeof(sin); 1534 sin.sin_family = AF_INET; 1535 sin.sin_port = th->th_sport; 1536 sin.sin_addr = ip->ip_src; 1537 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1538 } 1539 1540 /* XXX mapped address case */ 1541 } else 1542 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1543 notify); 1544 return NULL; 1545 } 1546 1547 /* 1548 * When a source quench is received, we are being notified of congestion. 1549 * Close the congestion window down to the Loss Window (one segment). 1550 * We will gradually open it again as we proceed. 1551 */ 1552 void 1553 tcp_quench(struct inpcb *inp, int errno) 1554 { 1555 struct tcpcb *tp = intotcpcb(inp); 1556 1557 if (tp) 1558 tp->snd_cwnd = tp->t_segsz; 1559 } 1560 #endif 1561 1562 #ifdef INET6 1563 void 1564 tcp6_quench(struct in6pcb *in6p, int errno) 1565 { 1566 struct tcpcb *tp = in6totcpcb(in6p); 1567 1568 if (tp) 1569 tp->snd_cwnd = tp->t_segsz; 1570 } 1571 #endif 1572 1573 #ifdef INET 1574 /* 1575 * Path MTU Discovery handlers. 1576 */ 1577 void 1578 tcp_mtudisc_callback(struct in_addr faddr) 1579 { 1580 #ifdef INET6 1581 struct in6_addr in6; 1582 #endif 1583 1584 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1585 #ifdef INET6 1586 memset(&in6, 0, sizeof(in6)); 1587 in6.s6_addr16[5] = 0xffff; 1588 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1589 tcp6_mtudisc_callback(&in6); 1590 #endif 1591 } 1592 1593 /* 1594 * On receipt of path MTU corrections, flush old route and replace it 1595 * with the new one. Retransmit all unacknowledged packets, to ensure 1596 * that all packets will be received. 1597 */ 1598 void 1599 tcp_mtudisc(struct inpcb *inp, int errno) 1600 { 1601 struct tcpcb *tp = intotcpcb(inp); 1602 struct rtentry *rt = in_pcbrtentry(inp); 1603 1604 if (tp != 0) { 1605 if (rt != 0) { 1606 /* 1607 * If this was not a host route, remove and realloc. 1608 */ 1609 if ((rt->rt_flags & RTF_HOST) == 0) { 1610 in_rtchange(inp, errno); 1611 if ((rt = in_pcbrtentry(inp)) == 0) 1612 return; 1613 } 1614 1615 /* 1616 * Slow start out of the error condition. We 1617 * use the MTU because we know it's smaller 1618 * than the previously transmitted segment. 1619 * 1620 * Note: This is more conservative than the 1621 * suggestion in draft-floyd-incr-init-win-03. 1622 */ 1623 if (rt->rt_rmx.rmx_mtu != 0) 1624 tp->snd_cwnd = 1625 TCP_INITIAL_WINDOW(tcp_init_win, 1626 rt->rt_rmx.rmx_mtu); 1627 } 1628 1629 /* 1630 * Resend unacknowledged packets. 1631 */ 1632 tp->snd_nxt = tp->snd_una; 1633 tcp_output(tp); 1634 } 1635 } 1636 #endif 1637 1638 #ifdef INET6 1639 /* 1640 * Path MTU Discovery handlers. 1641 */ 1642 void 1643 tcp6_mtudisc_callback(struct in6_addr *faddr) 1644 { 1645 struct sockaddr_in6 sin6; 1646 1647 bzero(&sin6, sizeof(sin6)); 1648 sin6.sin6_family = AF_INET6; 1649 sin6.sin6_len = sizeof(struct sockaddr_in6); 1650 sin6.sin6_addr = *faddr; 1651 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1652 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1653 } 1654 1655 void 1656 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1657 { 1658 struct tcpcb *tp = in6totcpcb(in6p); 1659 struct rtentry *rt = in6_pcbrtentry(in6p); 1660 1661 if (tp != 0) { 1662 if (rt != 0) { 1663 /* 1664 * If this was not a host route, remove and realloc. 1665 */ 1666 if ((rt->rt_flags & RTF_HOST) == 0) { 1667 in6_rtchange(in6p, errno); 1668 if ((rt = in6_pcbrtentry(in6p)) == 0) 1669 return; 1670 } 1671 1672 /* 1673 * Slow start out of the error condition. We 1674 * use the MTU because we know it's smaller 1675 * than the previously transmitted segment. 1676 * 1677 * Note: This is more conservative than the 1678 * suggestion in draft-floyd-incr-init-win-03. 1679 */ 1680 if (rt->rt_rmx.rmx_mtu != 0) 1681 tp->snd_cwnd = 1682 TCP_INITIAL_WINDOW(tcp_init_win, 1683 rt->rt_rmx.rmx_mtu); 1684 } 1685 1686 /* 1687 * Resend unacknowledged packets. 1688 */ 1689 tp->snd_nxt = tp->snd_una; 1690 tcp_output(tp); 1691 } 1692 } 1693 #endif /* INET6 */ 1694 1695 /* 1696 * Compute the MSS to advertise to the peer. Called only during 1697 * the 3-way handshake. If we are the server (peer initiated 1698 * connection), we are called with a pointer to the interface 1699 * on which the SYN packet arrived. If we are the client (we 1700 * initiated connection), we are called with a pointer to the 1701 * interface out which this connection should go. 1702 * 1703 * NOTE: Do not subtract IP option/extension header size nor IPsec 1704 * header size from MSS advertisement. MSS option must hold the maximum 1705 * segment size we can accept, so it must always be: 1706 * max(if mtu) - ip header - tcp header 1707 */ 1708 u_long 1709 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1710 { 1711 extern u_long in_maxmtu; 1712 u_long mss = 0; 1713 u_long hdrsiz; 1714 1715 /* 1716 * In order to avoid defeating path MTU discovery on the peer, 1717 * we advertise the max MTU of all attached networks as our MSS, 1718 * per RFC 1191, section 3.1. 1719 * 1720 * We provide the option to advertise just the MTU of 1721 * the interface on which we hope this connection will 1722 * be receiving. If we are responding to a SYN, we 1723 * will have a pretty good idea about this, but when 1724 * initiating a connection there is a bit more doubt. 1725 * 1726 * We also need to ensure that loopback has a large enough 1727 * MSS, as the loopback MTU is never included in in_maxmtu. 1728 */ 1729 1730 if (ifp != NULL) 1731 switch (af) { 1732 case AF_INET: 1733 mss = ifp->if_mtu; 1734 break; 1735 #ifdef INET6 1736 case AF_INET6: 1737 mss = IN6_LINKMTU(ifp); 1738 break; 1739 #endif 1740 } 1741 1742 if (tcp_mss_ifmtu == 0) 1743 switch (af) { 1744 case AF_INET: 1745 mss = max(in_maxmtu, mss); 1746 break; 1747 #ifdef INET6 1748 case AF_INET6: 1749 mss = max(in6_maxmtu, mss); 1750 break; 1751 #endif 1752 } 1753 1754 switch (af) { 1755 case AF_INET: 1756 hdrsiz = sizeof(struct ip); 1757 break; 1758 #ifdef INET6 1759 case AF_INET6: 1760 hdrsiz = sizeof(struct ip6_hdr); 1761 break; 1762 #endif 1763 default: 1764 hdrsiz = 0; 1765 break; 1766 } 1767 hdrsiz += sizeof(struct tcphdr); 1768 if (mss > hdrsiz) 1769 mss -= hdrsiz; 1770 1771 mss = max(tcp_mssdflt, mss); 1772 return (mss); 1773 } 1774 1775 /* 1776 * Set connection variables based on the peer's advertised MSS. 1777 * We are passed the TCPCB for the actual connection. If we 1778 * are the server, we are called by the compressed state engine 1779 * when the 3-way handshake is complete. If we are the client, 1780 * we are called when we receive the SYN,ACK from the server. 1781 * 1782 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1783 * before this routine is called! 1784 */ 1785 void 1786 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1787 { 1788 struct socket *so; 1789 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1790 struct rtentry *rt; 1791 #endif 1792 u_long bufsize; 1793 int mss; 1794 1795 #ifdef DIAGNOSTIC 1796 if (tp->t_inpcb && tp->t_in6pcb) 1797 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1798 #endif 1799 so = NULL; 1800 rt = NULL; 1801 #ifdef INET 1802 if (tp->t_inpcb) { 1803 so = tp->t_inpcb->inp_socket; 1804 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1805 rt = in_pcbrtentry(tp->t_inpcb); 1806 #endif 1807 } 1808 #endif 1809 #ifdef INET6 1810 if (tp->t_in6pcb) { 1811 so = tp->t_in6pcb->in6p_socket; 1812 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1813 rt = in6_pcbrtentry(tp->t_in6pcb); 1814 #endif 1815 } 1816 #endif 1817 1818 /* 1819 * As per RFC1122, use the default MSS value, unless they 1820 * sent us an offer. Do not accept offers less than 256 bytes. 1821 */ 1822 mss = tcp_mssdflt; 1823 if (offer) 1824 mss = offer; 1825 mss = max(mss, 256); /* sanity */ 1826 tp->t_peermss = mss; 1827 mss -= tcp_optlen(tp); 1828 #ifdef INET 1829 if (tp->t_inpcb) 1830 mss -= ip_optlen(tp->t_inpcb); 1831 #endif 1832 #ifdef INET6 1833 if (tp->t_in6pcb) 1834 mss -= ip6_optlen(tp->t_in6pcb); 1835 #endif 1836 1837 /* 1838 * If there's a pipesize, change the socket buffer to that size. 1839 * Make the socket buffer an integral number of MSS units. If 1840 * the MSS is larger than the socket buffer, artificially decrease 1841 * the MSS. 1842 */ 1843 #ifdef RTV_SPIPE 1844 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1845 bufsize = rt->rt_rmx.rmx_sendpipe; 1846 else 1847 #endif 1848 bufsize = so->so_snd.sb_hiwat; 1849 if (bufsize < mss) 1850 mss = bufsize; 1851 else { 1852 bufsize = roundup(bufsize, mss); 1853 if (bufsize > sb_max) 1854 bufsize = sb_max; 1855 (void) sbreserve(&so->so_snd, bufsize, so); 1856 } 1857 tp->t_segsz = mss; 1858 1859 #ifdef RTV_SSTHRESH 1860 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1861 /* 1862 * There's some sort of gateway or interface buffer 1863 * limit on the path. Use this to set the slow 1864 * start threshold, but set the threshold to no less 1865 * than 2 * MSS. 1866 */ 1867 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1868 } 1869 #endif 1870 } 1871 1872 /* 1873 * Processing necessary when a TCP connection is established. 1874 */ 1875 void 1876 tcp_established(struct tcpcb *tp) 1877 { 1878 struct socket *so; 1879 #ifdef RTV_RPIPE 1880 struct rtentry *rt; 1881 #endif 1882 u_long bufsize; 1883 1884 #ifdef DIAGNOSTIC 1885 if (tp->t_inpcb && tp->t_in6pcb) 1886 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1887 #endif 1888 so = NULL; 1889 rt = NULL; 1890 #ifdef INET 1891 if (tp->t_inpcb) { 1892 so = tp->t_inpcb->inp_socket; 1893 #if defined(RTV_RPIPE) 1894 rt = in_pcbrtentry(tp->t_inpcb); 1895 #endif 1896 } 1897 #endif 1898 #ifdef INET6 1899 if (tp->t_in6pcb) { 1900 so = tp->t_in6pcb->in6p_socket; 1901 #if defined(RTV_RPIPE) 1902 rt = in6_pcbrtentry(tp->t_in6pcb); 1903 #endif 1904 } 1905 #endif 1906 1907 tp->t_state = TCPS_ESTABLISHED; 1908 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1909 1910 #ifdef RTV_RPIPE 1911 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1912 bufsize = rt->rt_rmx.rmx_recvpipe; 1913 else 1914 #endif 1915 bufsize = so->so_rcv.sb_hiwat; 1916 if (bufsize > tp->t_ourmss) { 1917 bufsize = roundup(bufsize, tp->t_ourmss); 1918 if (bufsize > sb_max) 1919 bufsize = sb_max; 1920 (void) sbreserve(&so->so_rcv, bufsize, so); 1921 } 1922 } 1923 1924 /* 1925 * Check if there's an initial rtt or rttvar. Convert from the 1926 * route-table units to scaled multiples of the slow timeout timer. 1927 * Called only during the 3-way handshake. 1928 */ 1929 void 1930 tcp_rmx_rtt(struct tcpcb *tp) 1931 { 1932 #ifdef RTV_RTT 1933 struct rtentry *rt = NULL; 1934 int rtt; 1935 1936 #ifdef DIAGNOSTIC 1937 if (tp->t_inpcb && tp->t_in6pcb) 1938 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1939 #endif 1940 #ifdef INET 1941 if (tp->t_inpcb) 1942 rt = in_pcbrtentry(tp->t_inpcb); 1943 #endif 1944 #ifdef INET6 1945 if (tp->t_in6pcb) 1946 rt = in6_pcbrtentry(tp->t_in6pcb); 1947 #endif 1948 if (rt == NULL) 1949 return; 1950 1951 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1952 /* 1953 * XXX The lock bit for MTU indicates that the value 1954 * is also a minimum value; this is subject to time. 1955 */ 1956 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1957 TCPT_RANGESET(tp->t_rttmin, 1958 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1959 TCPTV_MIN, TCPTV_REXMTMAX); 1960 tp->t_srtt = rtt / 1961 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1962 if (rt->rt_rmx.rmx_rttvar) { 1963 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1964 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1965 (TCP_RTTVAR_SHIFT + 2)); 1966 } else { 1967 /* Default variation is +- 1 rtt */ 1968 tp->t_rttvar = 1969 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1970 } 1971 TCPT_RANGESET(tp->t_rxtcur, 1972 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1973 tp->t_rttmin, TCPTV_REXMTMAX); 1974 } 1975 #endif 1976 } 1977 1978 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1979 #if NRND > 0 1980 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1981 #endif 1982 1983 /* 1984 * Get a new sequence value given a tcp control block 1985 */ 1986 tcp_seq 1987 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1988 { 1989 1990 #ifdef INET 1991 if (tp->t_inpcb != NULL) { 1992 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1993 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1994 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1995 addin)); 1996 } 1997 #endif 1998 #ifdef INET6 1999 if (tp->t_in6pcb != NULL) { 2000 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2001 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2002 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2003 addin)); 2004 } 2005 #endif 2006 /* Not possible. */ 2007 panic("tcp_new_iss"); 2008 } 2009 2010 /* 2011 * This routine actually generates a new TCP initial sequence number. 2012 */ 2013 tcp_seq 2014 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2015 size_t addrsz, tcp_seq addin) 2016 { 2017 tcp_seq tcp_iss; 2018 2019 #if NRND > 0 2020 static int beenhere; 2021 2022 /* 2023 * If we haven't been here before, initialize our cryptographic 2024 * hash secret. 2025 */ 2026 if (beenhere == 0) { 2027 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2028 RND_EXTRACT_ANY); 2029 beenhere = 1; 2030 } 2031 2032 if (tcp_do_rfc1948) { 2033 MD5_CTX ctx; 2034 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2035 2036 /* 2037 * Compute the base value of the ISS. It is a hash 2038 * of (saddr, sport, daddr, dport, secret). 2039 */ 2040 MD5Init(&ctx); 2041 2042 MD5Update(&ctx, (u_char *) laddr, addrsz); 2043 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2044 2045 MD5Update(&ctx, (u_char *) faddr, addrsz); 2046 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2047 2048 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2049 2050 MD5Final(hash, &ctx); 2051 2052 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2053 2054 /* 2055 * Now increment our "timer", and add it in to 2056 * the computed value. 2057 * 2058 * XXX Use `addin'? 2059 * XXX TCP_ISSINCR too large to use? 2060 */ 2061 tcp_iss_seq += TCP_ISSINCR; 2062 #ifdef TCPISS_DEBUG 2063 printf("ISS hash 0x%08x, ", tcp_iss); 2064 #endif 2065 tcp_iss += tcp_iss_seq + addin; 2066 #ifdef TCPISS_DEBUG 2067 printf("new ISS 0x%08x\n", tcp_iss); 2068 #endif 2069 } else 2070 #endif /* NRND > 0 */ 2071 { 2072 /* 2073 * Randomize. 2074 */ 2075 #if NRND > 0 2076 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2077 #else 2078 tcp_iss = arc4random(); 2079 #endif 2080 2081 /* 2082 * If we were asked to add some amount to a known value, 2083 * we will take a random value obtained above, mask off 2084 * the upper bits, and add in the known value. We also 2085 * add in a constant to ensure that we are at least a 2086 * certain distance from the original value. 2087 * 2088 * This is used when an old connection is in timed wait 2089 * and we have a new one coming in, for instance. 2090 */ 2091 if (addin != 0) { 2092 #ifdef TCPISS_DEBUG 2093 printf("Random %08x, ", tcp_iss); 2094 #endif 2095 tcp_iss &= TCP_ISS_RANDOM_MASK; 2096 tcp_iss += addin + TCP_ISSINCR; 2097 #ifdef TCPISS_DEBUG 2098 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2099 #endif 2100 } else { 2101 tcp_iss &= TCP_ISS_RANDOM_MASK; 2102 tcp_iss += tcp_iss_seq; 2103 tcp_iss_seq += TCP_ISSINCR; 2104 #ifdef TCPISS_DEBUG 2105 printf("ISS %08x\n", tcp_iss); 2106 #endif 2107 } 2108 } 2109 2110 if (tcp_compat_42) { 2111 /* 2112 * Limit it to the positive range for really old TCP 2113 * implementations. 2114 * Just AND off the top bit instead of checking if 2115 * is set first - saves a branch 50% of the time. 2116 */ 2117 tcp_iss &= 0x7fffffff; /* XXX */ 2118 } 2119 2120 return (tcp_iss); 2121 } 2122 2123 #if defined(IPSEC) || defined(FAST_IPSEC) 2124 /* compute ESP/AH header size for TCP, including outer IP header. */ 2125 size_t 2126 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2127 { 2128 struct inpcb *inp; 2129 size_t hdrsiz; 2130 2131 /* XXX mapped addr case (tp->t_in6pcb) */ 2132 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2133 return 0; 2134 switch (tp->t_family) { 2135 case AF_INET: 2136 /* XXX: should use currect direction. */ 2137 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2138 break; 2139 default: 2140 hdrsiz = 0; 2141 break; 2142 } 2143 2144 return hdrsiz; 2145 } 2146 2147 #ifdef INET6 2148 size_t 2149 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2150 { 2151 struct in6pcb *in6p; 2152 size_t hdrsiz; 2153 2154 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2155 return 0; 2156 switch (tp->t_family) { 2157 case AF_INET6: 2158 /* XXX: should use currect direction. */ 2159 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2160 break; 2161 case AF_INET: 2162 /* mapped address case - tricky */ 2163 default: 2164 hdrsiz = 0; 2165 break; 2166 } 2167 2168 return hdrsiz; 2169 } 2170 #endif 2171 #endif /*IPSEC*/ 2172 2173 /* 2174 * Determine the length of the TCP options for this connection. 2175 * 2176 * XXX: What do we do for SACK, when we add that? Just reserve 2177 * all of the space? Otherwise we can't exactly be incrementing 2178 * cwnd by an amount that varies depending on the amount we last 2179 * had to SACK! 2180 */ 2181 2182 u_int 2183 tcp_optlen(struct tcpcb *tp) 2184 { 2185 u_int optlen; 2186 2187 optlen = 0; 2188 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2189 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2190 optlen += TCPOLEN_TSTAMP_APPA; 2191 2192 #ifdef TCP_SIGNATURE 2193 #if defined(INET6) && defined(FAST_IPSEC) 2194 if (tp->t_family == AF_INET) 2195 #endif 2196 if (tp->t_flags & TF_SIGNATURE) 2197 optlen += TCPOLEN_SIGNATURE + 2; 2198 #endif /* TCP_SIGNATURE */ 2199 2200 return optlen; 2201 } 2202
Indexes created Sun Sep 21 16:09:51 GMT 2025