tcp_subr.c revision 1.170
1 /* $NetBSD: tcp_subr.c,v 1.170 2004/04/26 05:18:13 itojun Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998, 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 1.170 2004/04/26 05:18:13 itojun 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 */ 203 204 /* tcb hash */ 205 #ifndef TCBHASHSIZE 206 #define TCBHASHSIZE 128 207 #endif 208 int tcbhashsize = TCBHASHSIZE; 209 210 /* syn hash parameters */ 211 #define TCP_SYN_HASH_SIZE 293 212 #define TCP_SYN_BUCKET_SIZE 35 213 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 214 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 215 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 216 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 217 218 int tcp_freeq __P((struct tcpcb *)); 219 220 #ifdef INET 221 void tcp_mtudisc_callback __P((struct in_addr)); 222 #endif 223 #ifdef INET6 224 void tcp6_mtudisc_callback __P((struct in6_addr *)); 225 #endif 226 227 void tcp_mtudisc __P((struct inpcb *, int)); 228 #ifdef INET6 229 void tcp6_mtudisc __P((struct in6pcb *, int)); 230 #endif 231 232 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL); 233 234 #ifdef TCP_CSUM_COUNTERS 235 #include <sys/device.h> 236 237 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 238 NULL, "tcp", "hwcsum bad"); 239 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 240 NULL, "tcp", "hwcsum ok"); 241 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "hwcsum data"); 243 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "swcsum"); 245 #endif /* TCP_CSUM_COUNTERS */ 246 247 #ifdef TCP_OUTPUT_COUNTERS 248 #include <sys/device.h> 249 250 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp", "output big header"); 252 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 NULL, "tcp", "output predict hit"); 254 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 255 NULL, "tcp", "output predict miss"); 256 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 257 NULL, "tcp", "output copy small"); 258 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 259 NULL, "tcp", "output copy big"); 260 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 261 NULL, "tcp", "output reference big"); 262 #endif /* TCP_OUTPUT_COUNTERS */ 263 264 #ifdef TCP_REASS_COUNTERS 265 #include <sys/device.h> 266 267 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 268 NULL, "tcp_reass", "calls"); 269 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 270 &tcp_reass_, "tcp_reass", "insert into empty queue"); 271 struct evcnt tcp_reass_iteration[8] = { 272 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 273 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 274 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 275 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 276 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 277 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 278 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 279 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 280 }; 281 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 &tcp_reass_, "tcp_reass", "prepend to first"); 283 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 &tcp_reass_, "tcp_reass", "prepend"); 285 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 &tcp_reass_, "tcp_reass", "insert"); 287 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 &tcp_reass_, "tcp_reass", "insert at tail"); 289 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 &tcp_reass_, "tcp_reass", "append"); 291 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 &tcp_reass_, "tcp_reass", "append to tail fragment"); 293 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 294 &tcp_reass_, "tcp_reass", "overlap at end"); 295 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 296 &tcp_reass_, "tcp_reass", "overlap at start"); 297 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 298 &tcp_reass_, "tcp_reass", "duplicate segment"); 299 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 300 &tcp_reass_, "tcp_reass", "duplicate fragment"); 301 302 #endif /* TCP_REASS_COUNTERS */ 303 304 #ifdef MBUFTRACE 305 struct mowner tcp_mowner = { "tcp" }; 306 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 307 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 308 #endif 309 310 /* 311 * Tcp initialization 312 */ 313 void 314 tcp_init() 315 { 316 int hlen; 317 318 /* Initialize the TCPCB template. */ 319 tcp_tcpcb_template(); 320 321 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 322 323 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 324 #ifdef INET6 325 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 326 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 327 #endif 328 if (max_protohdr < hlen) 329 max_protohdr = hlen; 330 if (max_linkhdr + hlen > MHLEN) 331 panic("tcp_init"); 332 333 #ifdef INET 334 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 335 #endif 336 #ifdef INET6 337 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 338 #endif 339 340 /* Initialize timer state. */ 341 tcp_timer_init(); 342 343 /* Initialize the compressed state engine. */ 344 syn_cache_init(); 345 346 #ifdef TCP_CSUM_COUNTERS 347 evcnt_attach_static(&tcp_hwcsum_bad); 348 evcnt_attach_static(&tcp_hwcsum_ok); 349 evcnt_attach_static(&tcp_hwcsum_data); 350 evcnt_attach_static(&tcp_swcsum); 351 #endif /* TCP_CSUM_COUNTERS */ 352 353 #ifdef TCP_OUTPUT_COUNTERS 354 evcnt_attach_static(&tcp_output_bigheader); 355 evcnt_attach_static(&tcp_output_predict_hit); 356 evcnt_attach_static(&tcp_output_predict_miss); 357 evcnt_attach_static(&tcp_output_copysmall); 358 evcnt_attach_static(&tcp_output_copybig); 359 evcnt_attach_static(&tcp_output_refbig); 360 #endif /* TCP_OUTPUT_COUNTERS */ 361 362 #ifdef TCP_REASS_COUNTERS 363 evcnt_attach_static(&tcp_reass_); 364 evcnt_attach_static(&tcp_reass_empty); 365 evcnt_attach_static(&tcp_reass_iteration[0]); 366 evcnt_attach_static(&tcp_reass_iteration[1]); 367 evcnt_attach_static(&tcp_reass_iteration[2]); 368 evcnt_attach_static(&tcp_reass_iteration[3]); 369 evcnt_attach_static(&tcp_reass_iteration[4]); 370 evcnt_attach_static(&tcp_reass_iteration[5]); 371 evcnt_attach_static(&tcp_reass_iteration[6]); 372 evcnt_attach_static(&tcp_reass_iteration[7]); 373 evcnt_attach_static(&tcp_reass_prependfirst); 374 evcnt_attach_static(&tcp_reass_prepend); 375 evcnt_attach_static(&tcp_reass_insert); 376 evcnt_attach_static(&tcp_reass_inserttail); 377 evcnt_attach_static(&tcp_reass_append); 378 evcnt_attach_static(&tcp_reass_appendtail); 379 evcnt_attach_static(&tcp_reass_overlaptail); 380 evcnt_attach_static(&tcp_reass_overlapfront); 381 evcnt_attach_static(&tcp_reass_segdup); 382 evcnt_attach_static(&tcp_reass_fragdup); 383 #endif /* TCP_REASS_COUNTERS */ 384 385 MOWNER_ATTACH(&tcp_tx_mowner); 386 MOWNER_ATTACH(&tcp_rx_mowner); 387 MOWNER_ATTACH(&tcp_mowner); 388 } 389 390 /* 391 * Create template to be used to send tcp packets on a connection. 392 * Call after host entry created, allocates an mbuf and fills 393 * in a skeletal tcp/ip header, minimizing the amount of work 394 * necessary when the connection is used. 395 */ 396 struct mbuf * 397 tcp_template(tp) 398 struct tcpcb *tp; 399 { 400 struct inpcb *inp = tp->t_inpcb; 401 #ifdef INET6 402 struct in6pcb *in6p = tp->t_in6pcb; 403 #endif 404 struct tcphdr *n; 405 struct mbuf *m; 406 int hlen; 407 408 switch (tp->t_family) { 409 case AF_INET: 410 hlen = sizeof(struct ip); 411 if (inp) 412 break; 413 #ifdef INET6 414 if (in6p) { 415 /* mapped addr case */ 416 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 417 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 418 break; 419 } 420 #endif 421 return NULL; /*EINVAL*/ 422 #ifdef INET6 423 case AF_INET6: 424 hlen = sizeof(struct ip6_hdr); 425 if (in6p) { 426 /* more sainty check? */ 427 break; 428 } 429 return NULL; /*EINVAL*/ 430 #endif 431 default: 432 hlen = 0; /*pacify gcc*/ 433 return NULL; /*EAFNOSUPPORT*/ 434 } 435 #ifdef DIAGNOSTIC 436 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 437 panic("mclbytes too small for t_template"); 438 #endif 439 m = tp->t_template; 440 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 441 ; 442 else { 443 if (m) 444 m_freem(m); 445 m = tp->t_template = NULL; 446 MGETHDR(m, M_DONTWAIT, MT_HEADER); 447 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 448 MCLGET(m, M_DONTWAIT); 449 if ((m->m_flags & M_EXT) == 0) { 450 m_free(m); 451 m = NULL; 452 } 453 } 454 if (m == NULL) 455 return NULL; 456 MCLAIM(m, &tcp_mowner); 457 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 458 } 459 460 bzero(mtod(m, caddr_t), m->m_len); 461 462 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 463 464 switch (tp->t_family) { 465 case AF_INET: 466 { 467 struct ipovly *ipov; 468 mtod(m, struct ip *)->ip_v = 4; 469 mtod(m, struct ip *)->ip_hl = hlen >> 2; 470 ipov = mtod(m, struct ipovly *); 471 ipov->ih_pr = IPPROTO_TCP; 472 ipov->ih_len = htons(sizeof(struct tcphdr)); 473 if (inp) { 474 ipov->ih_src = inp->inp_laddr; 475 ipov->ih_dst = inp->inp_faddr; 476 } 477 #ifdef INET6 478 else if (in6p) { 479 /* mapped addr case */ 480 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 481 sizeof(ipov->ih_src)); 482 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 483 sizeof(ipov->ih_dst)); 484 } 485 #endif 486 /* 487 * Compute the pseudo-header portion of the checksum 488 * now. We incrementally add in the TCP option and 489 * payload lengths later, and then compute the TCP 490 * checksum right before the packet is sent off onto 491 * the wire. 492 */ 493 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 494 ipov->ih_dst.s_addr, 495 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 496 break; 497 } 498 #ifdef INET6 499 case AF_INET6: 500 { 501 struct ip6_hdr *ip6; 502 mtod(m, struct ip *)->ip_v = 6; 503 ip6 = mtod(m, struct ip6_hdr *); 504 ip6->ip6_nxt = IPPROTO_TCP; 505 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 506 ip6->ip6_src = in6p->in6p_laddr; 507 ip6->ip6_dst = in6p->in6p_faddr; 508 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 509 if (ip6_auto_flowlabel) { 510 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 511 ip6->ip6_flow |= 512 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 513 } 514 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 515 ip6->ip6_vfc |= IPV6_VERSION; 516 517 /* 518 * Compute the pseudo-header portion of the checksum 519 * now. We incrementally add in the TCP option and 520 * payload lengths later, and then compute the TCP 521 * checksum right before the packet is sent off onto 522 * the wire. 523 */ 524 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 525 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 526 htonl(IPPROTO_TCP)); 527 break; 528 } 529 #endif 530 } 531 if (inp) { 532 n->th_sport = inp->inp_lport; 533 n->th_dport = inp->inp_fport; 534 } 535 #ifdef INET6 536 else if (in6p) { 537 n->th_sport = in6p->in6p_lport; 538 n->th_dport = in6p->in6p_fport; 539 } 540 #endif 541 n->th_seq = 0; 542 n->th_ack = 0; 543 n->th_x2 = 0; 544 n->th_off = 5; 545 n->th_flags = 0; 546 n->th_win = 0; 547 n->th_urp = 0; 548 return (m); 549 } 550 551 /* 552 * Send a single message to the TCP at address specified by 553 * the given TCP/IP header. If m == 0, then we make a copy 554 * of the tcpiphdr at ti and send directly to the addressed host. 555 * This is used to force keep alive messages out using the TCP 556 * template for a connection tp->t_template. If flags are given 557 * then we send a message back to the TCP which originated the 558 * segment ti, and discard the mbuf containing it and any other 559 * attached mbufs. 560 * 561 * In any case the ack and sequence number of the transmitted 562 * segment are as specified by the parameters. 563 */ 564 int 565 tcp_respond(tp, template, m, th0, ack, seq, flags) 566 struct tcpcb *tp; 567 struct mbuf *template; 568 struct mbuf *m; 569 struct tcphdr *th0; 570 tcp_seq ack, seq; 571 int flags; 572 { 573 struct route *ro; 574 int error, tlen, win = 0; 575 int hlen; 576 struct ip *ip; 577 #ifdef INET6 578 struct ip6_hdr *ip6; 579 #endif 580 int family; /* family on packet, not inpcb/in6pcb! */ 581 struct tcphdr *th; 582 struct socket *so; 583 584 if (tp != NULL && (flags & TH_RST) == 0) { 585 #ifdef DIAGNOSTIC 586 if (tp->t_inpcb && tp->t_in6pcb) 587 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 588 #endif 589 #ifdef INET 590 if (tp->t_inpcb) 591 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 592 #endif 593 #ifdef INET6 594 if (tp->t_in6pcb) 595 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 596 #endif 597 } 598 599 th = NULL; /* Quell uninitialized warning */ 600 ip = NULL; 601 #ifdef INET6 602 ip6 = NULL; 603 #endif 604 if (m == 0) { 605 if (!template) 606 return EINVAL; 607 608 /* get family information from template */ 609 switch (mtod(template, struct ip *)->ip_v) { 610 case 4: 611 family = AF_INET; 612 hlen = sizeof(struct ip); 613 break; 614 #ifdef INET6 615 case 6: 616 family = AF_INET6; 617 hlen = sizeof(struct ip6_hdr); 618 break; 619 #endif 620 default: 621 return EAFNOSUPPORT; 622 } 623 624 MGETHDR(m, M_DONTWAIT, MT_HEADER); 625 if (m) { 626 MCLAIM(m, &tcp_tx_mowner); 627 MCLGET(m, M_DONTWAIT); 628 if ((m->m_flags & M_EXT) == 0) { 629 m_free(m); 630 m = NULL; 631 } 632 } 633 if (m == NULL) 634 return (ENOBUFS); 635 636 if (tcp_compat_42) 637 tlen = 1; 638 else 639 tlen = 0; 640 641 m->m_data += max_linkhdr; 642 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 643 template->m_len); 644 switch (family) { 645 case AF_INET: 646 ip = mtod(m, struct ip *); 647 th = (struct tcphdr *)(ip + 1); 648 break; 649 #ifdef INET6 650 case AF_INET6: 651 ip6 = mtod(m, struct ip6_hdr *); 652 th = (struct tcphdr *)(ip6 + 1); 653 break; 654 #endif 655 #if 0 656 default: 657 /* noone will visit here */ 658 m_freem(m); 659 return EAFNOSUPPORT; 660 #endif 661 } 662 flags = TH_ACK; 663 } else { 664 665 if ((m->m_flags & M_PKTHDR) == 0) { 666 #if 0 667 printf("non PKTHDR to tcp_respond\n"); 668 #endif 669 m_freem(m); 670 return EINVAL; 671 } 672 #ifdef DIAGNOSTIC 673 if (!th0) 674 panic("th0 == NULL in tcp_respond"); 675 #endif 676 677 /* get family information from m */ 678 switch (mtod(m, struct ip *)->ip_v) { 679 case 4: 680 family = AF_INET; 681 hlen = sizeof(struct ip); 682 ip = mtod(m, struct ip *); 683 break; 684 #ifdef INET6 685 case 6: 686 family = AF_INET6; 687 hlen = sizeof(struct ip6_hdr); 688 ip6 = mtod(m, struct ip6_hdr *); 689 break; 690 #endif 691 default: 692 m_freem(m); 693 return EAFNOSUPPORT; 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(&ipqent_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) __P((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) __P((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 #ifdef TCP_SIGNATURE 1617 /* 1618 * Callback function invoked by m_apply() to digest TCP segment data 1619 * contained within an mbuf chain. 1620 */ 1621 static int 1622 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 1623 { 1624 1625 MD5Update(fstate, (u_char *)data, len); 1626 return (0); 1627 } 1628 1629 /* 1630 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 1631 * 1632 * Parameters: 1633 * m pointer to head of mbuf chain 1634 * th pointer to tcp header 1635 * thoff offset to TCP header within the mbuf chain 1636 * (if you don't know the value, set to -1) 1637 * len length of TCP segment data, excluding options 1638 * optlen length of TCP segment options 1639 * buf pointer to storage for computed MD5 digest 1640 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 1641 * 1642 * We do this over ip, tcphdr, segment data, and the key in the SADB. 1643 * When called from tcp_input(), we can be sure that th_sum has been 1644 * zeroed out and verified already. 1645 * 1646 * Return 0 if successful, otherwise return -1. 1647 * 1648 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 1649 * search with the destination IP address, and a 'magic SPI' of 0x1000. 1650 * Another branch of this code exists which uses the SPD to specify 1651 * per-application flows, but it is unstable. 1652 */ 1653 int 1654 tcp_signature_compute(struct mbuf *m, struct tcphdr *th, int thoff, 1655 int len, int optlen, u_char *buf, u_int direction) 1656 { 1657 #ifdef FAST_IPSEC 1658 union sockaddr_union dst; 1659 #endif 1660 MD5_CTX ctx; 1661 int doff; 1662 struct ip *ip; 1663 struct ipovly *ipovly; 1664 struct ip6_hdr *ip6; 1665 struct secasvar *sav; 1666 u_int16_t savecsum; 1667 struct ippseudo ippseudo; 1668 struct ip6_ext ip6e; 1669 struct ip6_hdr_pseudo ip6pseudo; 1670 u_int8_t nxt; 1671 1672 KASSERT(m != NULL /*, ("NULL mbuf chain")*/); 1673 KASSERT(buf != NULL /*, ("NULL signature pointer")*/); 1674 1675 ip = mtod(m, struct ip *); 1676 ip6 = NULL; 1677 if (ip->ip_v == 6) { 1678 ip = NULL; 1679 ip6 = mtod(m, struct ip6_hdr *); 1680 } 1681 1682 /* look for TCP header offset - it won't take too long */ 1683 if (thoff >= 0) { 1684 nxt = IPPROTO_TCP; 1685 goto found; 1686 } 1687 1688 thoff = ip ? sizeof(*ip) : sizeof(*ip6); 1689 nxt = ip ? ip->ip_p : ip6->ip6_nxt; 1690 1691 while (1) { 1692 switch (nxt) { 1693 case IPPROTO_TCP: 1694 goto found; 1695 case IPPROTO_AH: 1696 m_copydata(m, thoff, sizeof(ip6e), (caddr_t)&ip6e); 1697 thoff += (ip6e.ip6e_len + 2) << 2; 1698 break; 1699 case IPPROTO_DSTOPTS: 1700 case IPPROTO_ROUTING: 1701 case IPPROTO_HOPOPTS: 1702 m_copydata(m, thoff, sizeof(ip6e), (caddr_t)&ip6e); 1703 thoff += (ip6e.ip6e_len + 1) << 3; 1704 break; 1705 default: 1706 return EINVAL; 1707 } 1708 1709 nxt = ip6e.ip6e_nxt; 1710 } 1711 1712 found: 1713 1714 #ifdef FAST_IPSEC 1715 /* Extract the destination from the IP header in the mbuf. */ 1716 bzero(&dst, sizeof(union sockaddr_union)); 1717 dst.sa.sa_len = sizeof(struct sockaddr_in); 1718 dst.sa.sa_family = AF_INET; 1719 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 1720 ip->ip_src : ip->ip_dst; 1721 1722 /* Look up an SADB entry which matches the address of the peer. */ 1723 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1724 #else 1725 if (ip) 1726 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 1727 (caddr_t)&ip->ip_dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1728 else 1729 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 1730 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1731 #endif 1732 if (sav == NULL) { 1733 printf("%s: SADB lookup failed\n", __func__); 1734 return (EINVAL); 1735 } 1736 1737 MD5Init(&ctx); 1738 doff = thoff + sizeof(struct tcphdr) + optlen; 1739 1740 /* 1741 * Step 1: Update MD5 hash with IP pseudo-header. 1742 * 1743 * XXX The ippseudo header MUST be digested in network byte order, 1744 * or else we'll fail the regression test. Assume all fields we've 1745 * been doing arithmetic on have been in host byte order. 1746 * XXX One cannot depend on ipovly->ih_len here. When called from 1747 * tcp_output(), the underlying ip_len member has not yet been set. 1748 */ 1749 if (ip) { 1750 memset(&ippseudo, 0, sizeof(ippseudo)); 1751 ipovly = (struct ipovly *)ip; 1752 ippseudo.ippseudo_src = ipovly->ih_src; 1753 ippseudo.ippseudo_dst = ipovly->ih_dst; 1754 ippseudo.ippseudo_pad = 0; 1755 ippseudo.ippseudo_p = IPPROTO_TCP; 1756 ippseudo.ippseudo_len = 1757 htons(len + sizeof(struct tcphdr) + optlen); 1758 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 1759 } else { 1760 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 1761 ip6pseudo.ip6ph_src = ip6->ip6_src; 1762 in6_clearscope(&ip6pseudo.ip6ph_src); 1763 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 1764 in6_clearscope(&ip6pseudo.ip6ph_dst); 1765 ip6pseudo.ip6ph_len = 1766 htonl(len + sizeof(struct tcphdr) + optlen); 1767 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 1768 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 1769 } 1770 1771 /* 1772 * Step 2: Update MD5 hash with TCP header, excluding options. 1773 * The TCP checksum must be set to zero. 1774 */ 1775 savecsum = th->th_sum; 1776 th->th_sum = 0; 1777 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 1778 th->th_sum = savecsum; 1779 1780 /* 1781 * Step 3: Update MD5 hash with TCP segment data. 1782 * Use m_apply() to avoid an early m_pullup(). 1783 */ 1784 if (len > 0) 1785 m_apply(m, doff, len, tcp_signature_apply, &ctx); 1786 1787 /* 1788 * Step 4: Update MD5 hash with shared secret. 1789 */ 1790 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 1791 MD5Final(buf, &ctx); 1792 1793 key_sa_recordxfer(sav, m); 1794 #ifdef FAST_IPSEC 1795 KEY_FREESAV(&sav); 1796 #else 1797 key_freesav(sav); 1798 #endif 1799 return (0); 1800 } 1801 #endif /* TCP_SIGNATURE */ 1802 1803 /* 1804 * On receipt of path MTU corrections, flush old route and replace it 1805 * with the new one. Retransmit all unacknowledged packets, to ensure 1806 * that all packets will be received. 1807 */ 1808 void 1809 tcp_mtudisc(inp, errno) 1810 struct inpcb *inp; 1811 int errno; 1812 { 1813 struct tcpcb *tp = intotcpcb(inp); 1814 struct rtentry *rt = in_pcbrtentry(inp); 1815 1816 if (tp != 0) { 1817 if (rt != 0) { 1818 /* 1819 * If this was not a host route, remove and realloc. 1820 */ 1821 if ((rt->rt_flags & RTF_HOST) == 0) { 1822 in_rtchange(inp, errno); 1823 if ((rt = in_pcbrtentry(inp)) == 0) 1824 return; 1825 } 1826 1827 /* 1828 * Slow start out of the error condition. We 1829 * use the MTU because we know it's smaller 1830 * than the previously transmitted segment. 1831 * 1832 * Note: This is more conservative than the 1833 * suggestion in draft-floyd-incr-init-win-03. 1834 */ 1835 if (rt->rt_rmx.rmx_mtu != 0) 1836 tp->snd_cwnd = 1837 TCP_INITIAL_WINDOW(tcp_init_win, 1838 rt->rt_rmx.rmx_mtu); 1839 } 1840 1841 /* 1842 * Resend unacknowledged packets. 1843 */ 1844 tp->snd_nxt = tp->snd_una; 1845 tcp_output(tp); 1846 } 1847 } 1848 #endif 1849 1850 #ifdef INET6 1851 /* 1852 * Path MTU Discovery handlers. 1853 */ 1854 void 1855 tcp6_mtudisc_callback(faddr) 1856 struct in6_addr *faddr; 1857 { 1858 struct sockaddr_in6 sin6; 1859 1860 bzero(&sin6, sizeof(sin6)); 1861 sin6.sin6_family = AF_INET6; 1862 sin6.sin6_len = sizeof(struct sockaddr_in6); 1863 sin6.sin6_addr = *faddr; 1864 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1865 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1866 } 1867 1868 void 1869 tcp6_mtudisc(in6p, errno) 1870 struct in6pcb *in6p; 1871 int errno; 1872 { 1873 struct tcpcb *tp = in6totcpcb(in6p); 1874 struct rtentry *rt = in6_pcbrtentry(in6p); 1875 1876 if (tp != 0) { 1877 if (rt != 0) { 1878 /* 1879 * If this was not a host route, remove and realloc. 1880 */ 1881 if ((rt->rt_flags & RTF_HOST) == 0) { 1882 in6_rtchange(in6p, errno); 1883 if ((rt = in6_pcbrtentry(in6p)) == 0) 1884 return; 1885 } 1886 1887 /* 1888 * Slow start out of the error condition. We 1889 * use the MTU because we know it's smaller 1890 * than the previously transmitted segment. 1891 * 1892 * Note: This is more conservative than the 1893 * suggestion in draft-floyd-incr-init-win-03. 1894 */ 1895 if (rt->rt_rmx.rmx_mtu != 0) 1896 tp->snd_cwnd = 1897 TCP_INITIAL_WINDOW(tcp_init_win, 1898 rt->rt_rmx.rmx_mtu); 1899 } 1900 1901 /* 1902 * Resend unacknowledged packets. 1903 */ 1904 tp->snd_nxt = tp->snd_una; 1905 tcp_output(tp); 1906 } 1907 } 1908 #endif /* INET6 */ 1909 1910 /* 1911 * Compute the MSS to advertise to the peer. Called only during 1912 * the 3-way handshake. If we are the server (peer initiated 1913 * connection), we are called with a pointer to the interface 1914 * on which the SYN packet arrived. If we are the client (we 1915 * initiated connection), we are called with a pointer to the 1916 * interface out which this connection should go. 1917 * 1918 * NOTE: Do not subtract IP option/extension header size nor IPsec 1919 * header size from MSS advertisement. MSS option must hold the maximum 1920 * segment size we can accept, so it must always be: 1921 * max(if mtu) - ip header - tcp header 1922 */ 1923 u_long 1924 tcp_mss_to_advertise(ifp, af) 1925 const struct ifnet *ifp; 1926 int af; 1927 { 1928 extern u_long in_maxmtu; 1929 u_long mss = 0; 1930 u_long hdrsiz; 1931 1932 /* 1933 * In order to avoid defeating path MTU discovery on the peer, 1934 * we advertise the max MTU of all attached networks as our MSS, 1935 * per RFC 1191, section 3.1. 1936 * 1937 * We provide the option to advertise just the MTU of 1938 * the interface on which we hope this connection will 1939 * be receiving. If we are responding to a SYN, we 1940 * will have a pretty good idea about this, but when 1941 * initiating a connection there is a bit more doubt. 1942 * 1943 * We also need to ensure that loopback has a large enough 1944 * MSS, as the loopback MTU is never included in in_maxmtu. 1945 */ 1946 1947 if (ifp != NULL) 1948 switch (af) { 1949 case AF_INET: 1950 mss = ifp->if_mtu; 1951 break; 1952 #ifdef INET6 1953 case AF_INET6: 1954 mss = IN6_LINKMTU(ifp); 1955 break; 1956 #endif 1957 } 1958 1959 if (tcp_mss_ifmtu == 0) 1960 switch (af) { 1961 case AF_INET: 1962 mss = max(in_maxmtu, mss); 1963 break; 1964 #ifdef INET6 1965 case AF_INET6: 1966 mss = max(in6_maxmtu, mss); 1967 break; 1968 #endif 1969 } 1970 1971 switch (af) { 1972 case AF_INET: 1973 hdrsiz = sizeof(struct ip); 1974 break; 1975 #ifdef INET6 1976 case AF_INET6: 1977 hdrsiz = sizeof(struct ip6_hdr); 1978 break; 1979 #endif 1980 default: 1981 hdrsiz = 0; 1982 break; 1983 } 1984 hdrsiz += sizeof(struct tcphdr); 1985 if (mss > hdrsiz) 1986 mss -= hdrsiz; 1987 1988 mss = max(tcp_mssdflt, mss); 1989 return (mss); 1990 } 1991 1992 /* 1993 * Set connection variables based on the peer's advertised MSS. 1994 * We are passed the TCPCB for the actual connection. If we 1995 * are the server, we are called by the compressed state engine 1996 * when the 3-way handshake is complete. If we are the client, 1997 * we are called when we receive the SYN,ACK from the server. 1998 * 1999 * NOTE: Our advertised MSS value must be initialized in the TCPCB 2000 * before this routine is called! 2001 */ 2002 void 2003 tcp_mss_from_peer(tp, offer) 2004 struct tcpcb *tp; 2005 int offer; 2006 { 2007 struct socket *so; 2008 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 2009 struct rtentry *rt; 2010 #endif 2011 u_long bufsize; 2012 int mss; 2013 2014 #ifdef DIAGNOSTIC 2015 if (tp->t_inpcb && tp->t_in6pcb) 2016 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 2017 #endif 2018 so = NULL; 2019 rt = NULL; 2020 #ifdef INET 2021 if (tp->t_inpcb) { 2022 so = tp->t_inpcb->inp_socket; 2023 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 2024 rt = in_pcbrtentry(tp->t_inpcb); 2025 #endif 2026 } 2027 #endif 2028 #ifdef INET6 2029 if (tp->t_in6pcb) { 2030 so = tp->t_in6pcb->in6p_socket; 2031 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 2032 rt = in6_pcbrtentry(tp->t_in6pcb); 2033 #endif 2034 } 2035 #endif 2036 2037 /* 2038 * As per RFC1122, use the default MSS value, unless they 2039 * sent us an offer. Do not accept offers less than 256 bytes. 2040 */ 2041 mss = tcp_mssdflt; 2042 if (offer) 2043 mss = offer; 2044 mss = max(mss, 256); /* sanity */ 2045 tp->t_peermss = mss; 2046 mss -= tcp_optlen(tp); 2047 #ifdef INET 2048 if (tp->t_inpcb) 2049 mss -= ip_optlen(tp->t_inpcb); 2050 #endif 2051 #ifdef INET6 2052 if (tp->t_in6pcb) 2053 mss -= ip6_optlen(tp->t_in6pcb); 2054 #endif 2055 2056 /* 2057 * If there's a pipesize, change the socket buffer to that size. 2058 * Make the socket buffer an integral number of MSS units. If 2059 * the MSS is larger than the socket buffer, artificially decrease 2060 * the MSS. 2061 */ 2062 #ifdef RTV_SPIPE 2063 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 2064 bufsize = rt->rt_rmx.rmx_sendpipe; 2065 else 2066 #endif 2067 bufsize = so->so_snd.sb_hiwat; 2068 if (bufsize < mss) 2069 mss = bufsize; 2070 else { 2071 bufsize = roundup(bufsize, mss); 2072 if (bufsize > sb_max) 2073 bufsize = sb_max; 2074 (void) sbreserve(&so->so_snd, bufsize, so); 2075 } 2076 tp->t_segsz = mss; 2077 2078 #ifdef RTV_SSTHRESH 2079 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 2080 /* 2081 * There's some sort of gateway or interface buffer 2082 * limit on the path. Use this to set the slow 2083 * start threshold, but set the threshold to no less 2084 * than 2 * MSS. 2085 */ 2086 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2087 } 2088 #endif 2089 } 2090 2091 /* 2092 * Processing necessary when a TCP connection is established. 2093 */ 2094 void 2095 tcp_established(tp) 2096 struct tcpcb *tp; 2097 { 2098 struct socket *so; 2099 #ifdef RTV_RPIPE 2100 struct rtentry *rt; 2101 #endif 2102 u_long bufsize; 2103 2104 #ifdef DIAGNOSTIC 2105 if (tp->t_inpcb && tp->t_in6pcb) 2106 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2107 #endif 2108 so = NULL; 2109 rt = NULL; 2110 #ifdef INET 2111 if (tp->t_inpcb) { 2112 so = tp->t_inpcb->inp_socket; 2113 #if defined(RTV_RPIPE) 2114 rt = in_pcbrtentry(tp->t_inpcb); 2115 #endif 2116 } 2117 #endif 2118 #ifdef INET6 2119 if (tp->t_in6pcb) { 2120 so = tp->t_in6pcb->in6p_socket; 2121 #if defined(RTV_RPIPE) 2122 rt = in6_pcbrtentry(tp->t_in6pcb); 2123 #endif 2124 } 2125 #endif 2126 2127 tp->t_state = TCPS_ESTABLISHED; 2128 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 2129 2130 #ifdef RTV_RPIPE 2131 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2132 bufsize = rt->rt_rmx.rmx_recvpipe; 2133 else 2134 #endif 2135 bufsize = so->so_rcv.sb_hiwat; 2136 if (bufsize > tp->t_ourmss) { 2137 bufsize = roundup(bufsize, tp->t_ourmss); 2138 if (bufsize > sb_max) 2139 bufsize = sb_max; 2140 (void) sbreserve(&so->so_rcv, bufsize, so); 2141 } 2142 } 2143 2144 /* 2145 * Check if there's an initial rtt or rttvar. Convert from the 2146 * route-table units to scaled multiples of the slow timeout timer. 2147 * Called only during the 3-way handshake. 2148 */ 2149 void 2150 tcp_rmx_rtt(tp) 2151 struct tcpcb *tp; 2152 { 2153 #ifdef RTV_RTT 2154 struct rtentry *rt = NULL; 2155 int rtt; 2156 2157 #ifdef DIAGNOSTIC 2158 if (tp->t_inpcb && tp->t_in6pcb) 2159 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2160 #endif 2161 #ifdef INET 2162 if (tp->t_inpcb) 2163 rt = in_pcbrtentry(tp->t_inpcb); 2164 #endif 2165 #ifdef INET6 2166 if (tp->t_in6pcb) 2167 rt = in6_pcbrtentry(tp->t_in6pcb); 2168 #endif 2169 if (rt == NULL) 2170 return; 2171 2172 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2173 /* 2174 * XXX The lock bit for MTU indicates that the value 2175 * is also a minimum value; this is subject to time. 2176 */ 2177 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2178 TCPT_RANGESET(tp->t_rttmin, 2179 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2180 TCPTV_MIN, TCPTV_REXMTMAX); 2181 tp->t_srtt = rtt / 2182 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2183 if (rt->rt_rmx.rmx_rttvar) { 2184 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2185 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2186 (TCP_RTTVAR_SHIFT + 2)); 2187 } else { 2188 /* Default variation is +- 1 rtt */ 2189 tp->t_rttvar = 2190 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2191 } 2192 TCPT_RANGESET(tp->t_rxtcur, 2193 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2194 tp->t_rttmin, TCPTV_REXMTMAX); 2195 } 2196 #endif 2197 } 2198 2199 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2200 #if NRND > 0 2201 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2202 #endif 2203 2204 /* 2205 * Get a new sequence value given a tcp control block 2206 */ 2207 tcp_seq 2208 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2209 { 2210 2211 #ifdef INET 2212 if (tp->t_inpcb != NULL) { 2213 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2214 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2215 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2216 addin)); 2217 } 2218 #endif 2219 #ifdef INET6 2220 if (tp->t_in6pcb != NULL) { 2221 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2222 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2223 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2224 addin)); 2225 } 2226 #endif 2227 /* Not possible. */ 2228 panic("tcp_new_iss"); 2229 } 2230 2231 /* 2232 * This routine actually generates a new TCP initial sequence number. 2233 */ 2234 tcp_seq 2235 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2236 size_t addrsz, tcp_seq addin) 2237 { 2238 tcp_seq tcp_iss; 2239 2240 #if NRND > 0 2241 static int beenhere; 2242 2243 /* 2244 * If we haven't been here before, initialize our cryptographic 2245 * hash secret. 2246 */ 2247 if (beenhere == 0) { 2248 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2249 RND_EXTRACT_ANY); 2250 beenhere = 1; 2251 } 2252 2253 if (tcp_do_rfc1948) { 2254 MD5_CTX ctx; 2255 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2256 2257 /* 2258 * Compute the base value of the ISS. It is a hash 2259 * of (saddr, sport, daddr, dport, secret). 2260 */ 2261 MD5Init(&ctx); 2262 2263 MD5Update(&ctx, (u_char *) laddr, addrsz); 2264 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2265 2266 MD5Update(&ctx, (u_char *) faddr, addrsz); 2267 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2268 2269 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2270 2271 MD5Final(hash, &ctx); 2272 2273 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2274 2275 /* 2276 * Now increment our "timer", and add it in to 2277 * the computed value. 2278 * 2279 * XXX Use `addin'? 2280 * XXX TCP_ISSINCR too large to use? 2281 */ 2282 tcp_iss_seq += TCP_ISSINCR; 2283 #ifdef TCPISS_DEBUG 2284 printf("ISS hash 0x%08x, ", tcp_iss); 2285 #endif 2286 tcp_iss += tcp_iss_seq + addin; 2287 #ifdef TCPISS_DEBUG 2288 printf("new ISS 0x%08x\n", tcp_iss); 2289 #endif 2290 } else 2291 #endif /* NRND > 0 */ 2292 { 2293 /* 2294 * Randomize. 2295 */ 2296 #if NRND > 0 2297 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2298 #else 2299 tcp_iss = arc4random(); 2300 #endif 2301 2302 /* 2303 * If we were asked to add some amount to a known value, 2304 * we will take a random value obtained above, mask off 2305 * the upper bits, and add in the known value. We also 2306 * add in a constant to ensure that we are at least a 2307 * certain distance from the original value. 2308 * 2309 * This is used when an old connection is in timed wait 2310 * and we have a new one coming in, for instance. 2311 */ 2312 if (addin != 0) { 2313 #ifdef TCPISS_DEBUG 2314 printf("Random %08x, ", tcp_iss); 2315 #endif 2316 tcp_iss &= TCP_ISS_RANDOM_MASK; 2317 tcp_iss += addin + TCP_ISSINCR; 2318 #ifdef TCPISS_DEBUG 2319 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2320 #endif 2321 } else { 2322 tcp_iss &= TCP_ISS_RANDOM_MASK; 2323 tcp_iss += tcp_iss_seq; 2324 tcp_iss_seq += TCP_ISSINCR; 2325 #ifdef TCPISS_DEBUG 2326 printf("ISS %08x\n", tcp_iss); 2327 #endif 2328 } 2329 } 2330 2331 if (tcp_compat_42) { 2332 /* 2333 * Limit it to the positive range for really old TCP 2334 * implementations. 2335 * Just AND off the top bit instead of checking if 2336 * is set first - saves a branch 50% of the time. 2337 */ 2338 tcp_iss &= 0x7fffffff; /* XXX */ 2339 } 2340 2341 return (tcp_iss); 2342 } 2343 2344 #if defined(IPSEC) || defined(FAST_IPSEC) 2345 /* compute ESP/AH header size for TCP, including outer IP header. */ 2346 size_t 2347 ipsec4_hdrsiz_tcp(tp) 2348 struct tcpcb *tp; 2349 { 2350 struct inpcb *inp; 2351 size_t hdrsiz; 2352 2353 /* XXX mapped addr case (tp->t_in6pcb) */ 2354 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2355 return 0; 2356 switch (tp->t_family) { 2357 case AF_INET: 2358 /* XXX: should use currect direction. */ 2359 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2360 break; 2361 default: 2362 hdrsiz = 0; 2363 break; 2364 } 2365 2366 return hdrsiz; 2367 } 2368 2369 #ifdef INET6 2370 size_t 2371 ipsec6_hdrsiz_tcp(tp) 2372 struct tcpcb *tp; 2373 { 2374 struct in6pcb *in6p; 2375 size_t hdrsiz; 2376 2377 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2378 return 0; 2379 switch (tp->t_family) { 2380 case AF_INET6: 2381 /* XXX: should use currect direction. */ 2382 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2383 break; 2384 case AF_INET: 2385 /* mapped address case - tricky */ 2386 default: 2387 hdrsiz = 0; 2388 break; 2389 } 2390 2391 return hdrsiz; 2392 } 2393 #endif 2394 #endif /*IPSEC*/ 2395 2396 /* 2397 * Determine the length of the TCP options for this connection. 2398 * 2399 * XXX: What do we do for SACK, when we add that? Just reserve 2400 * all of the space? Otherwise we can't exactly be incrementing 2401 * cwnd by an amount that varies depending on the amount we last 2402 * had to SACK! 2403 */ 2404 2405 u_int 2406 tcp_optlen(tp) 2407 struct tcpcb *tp; 2408 { 2409 u_int optlen; 2410 2411 optlen = 0; 2412 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2413 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2414 optlen += TCPOLEN_TSTAMP_APPA; 2415 2416 #ifdef TCP_SIGNATURE 2417 #if defined(INET6) && defined(FAST_IPSEC) 2418 if (tp->t_family == AF_INET) 2419 #endif 2420 if (tp->t_flags & TF_SIGNATURE) 2421 optlen += TCPOLEN_SIGNATURE + 2; 2422 #endif /* TCP_SIGNATURE */ 2423 2424 return optlen; 2425 } 2426
Indexes created Sun Sep 21 20:09:37 GMT 2025