tcp_subr.c revision 1.166
1 /* $NetBSD: tcp_subr.c,v 1.166 2004/04/25 22:25:04 jonathan Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998, 2000, 2001 The NetBSD Foundation, Inc. 34 * All rights reserved. 35 * 36 * This code is derived from software contributed to The NetBSD Foundation 37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 38 * Facility, NASA Ames Research Center. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. All advertising materials mentioning features or use of this software 49 * must display the following acknowledgement: 50 * This product includes software developed by the NetBSD 51 * Foundation, Inc. and its contributors. 52 * 4. Neither the name of The NetBSD Foundation nor the names of its 53 * contributors may be used to endorse or promote products derived 54 * from this software without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 57 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 66 * POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 /* 70 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 71 * The Regents of the University of California. All rights reserved. 72 * 73 * Redistribution and use in source and binary forms, with or without 74 * modification, are permitted provided that the following conditions 75 * are met: 76 * 1. Redistributions of source code must retain the above copyright 77 * notice, this list of conditions and the following disclaimer. 78 * 2. Redistributions in binary form must reproduce the above copyright 79 * notice, this list of conditions and the following disclaimer in the 80 * documentation and/or other materials provided with the distribution. 81 * 3. Neither the name of the University nor the names of its contributors 82 * may be used to endorse or promote products derived from this software 83 * without specific prior written permission. 84 * 85 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 86 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 87 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 88 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 89 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 90 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 91 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 92 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 93 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 94 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 95 * SUCH DAMAGE. 96 * 97 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 98 */ 99 100 #include <sys/cdefs.h> 101 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.166 2004/04/25 22:25:04 jonathan Exp $"); 102 103 #include "opt_inet.h" 104 #include "opt_ipsec.h" 105 #include "opt_tcp_compat_42.h" 106 #include "opt_inet_csum.h" 107 #include "opt_mbuftrace.h" 108 #include "rnd.h" 109 110 #include <sys/param.h> 111 #include <sys/proc.h> 112 #include <sys/systm.h> 113 #include <sys/malloc.h> 114 #include <sys/mbuf.h> 115 #include <sys/socket.h> 116 #include <sys/socketvar.h> 117 #include <sys/protosw.h> 118 #include <sys/errno.h> 119 #include <sys/kernel.h> 120 #include <sys/pool.h> 121 #if NRND > 0 122 #include <sys/md5.h> 123 #include <sys/rnd.h> 124 #endif 125 126 #include <net/route.h> 127 #include <net/if.h> 128 129 #include <netinet/in.h> 130 #include <netinet/in_systm.h> 131 #include <netinet/ip.h> 132 #include <netinet/in_pcb.h> 133 #include <netinet/ip_var.h> 134 #include <netinet/ip_icmp.h> 135 136 #ifdef INET6 137 #ifndef INET 138 #include <netinet/in.h> 139 #endif 140 #include <netinet/ip6.h> 141 #include <netinet6/in6_pcb.h> 142 #include <netinet6/ip6_var.h> 143 #include <netinet6/in6_var.h> 144 #include <netinet6/ip6protosw.h> 145 #include <netinet/icmp6.h> 146 #include <netinet6/nd6.h> 147 #endif 148 149 #include <netinet/tcp.h> 150 #include <netinet/tcp_fsm.h> 151 #include <netinet/tcp_seq.h> 152 #include <netinet/tcp_timer.h> 153 #include <netinet/tcp_var.h> 154 #include <netinet/tcpip.h> 155 156 #ifdef IPSEC 157 #include <netinet6/ipsec.h> 158 #endif /*IPSEC*/ 159 160 #ifdef FAST_IPSEC 161 #include <netipsec/ipsec.h> 162 #include <netipsec/xform.h> 163 #ifdef INET6 164 #include <netipsec/ipsec6.h> 165 #endif 166 #include <netipsec/key.h> 167 #endif /* FAST_IPSEC*/ 168 169 170 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 171 struct tcpstat tcpstat; /* tcp statistics */ 172 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 173 174 /* patchable/settable parameters for tcp */ 175 int tcp_mssdflt = TCP_MSS; 176 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 177 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 178 #if NRND > 0 179 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 180 #endif 181 int tcp_do_sack = 1; /* selective acknowledgement */ 182 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 183 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 184 int tcp_do_newreno = 1; /* Use the New Reno algorithms */ 185 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 186 #ifndef TCP_INIT_WIN 187 #define TCP_INIT_WIN 0 /* initial slow start window */ 188 #endif 189 #ifndef TCP_INIT_WIN_LOCAL 190 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 191 #endif 192 int tcp_init_win = TCP_INIT_WIN; 193 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 194 int tcp_mss_ifmtu = 0; 195 #ifdef TCP_COMPAT_42 196 int tcp_compat_42 = 1; 197 #else 198 int tcp_compat_42 = 0; 199 #endif 200 int tcp_rst_ppslim = 100; /* 100pps */ 201 int tcp_ackdrop_ppslim = 100; /* 100pps */ 202 203 /* tcb hash */ 204 #ifndef TCBHASHSIZE 205 #define TCBHASHSIZE 128 206 #endif 207 int tcbhashsize = TCBHASHSIZE; 208 209 /* syn hash parameters */ 210 #define TCP_SYN_HASH_SIZE 293 211 #define TCP_SYN_BUCKET_SIZE 35 212 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 213 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 214 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 215 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 216 217 int tcp_freeq __P((struct tcpcb *)); 218 219 #ifdef INET 220 void tcp_mtudisc_callback __P((struct in_addr)); 221 #endif 222 #ifdef INET6 223 void tcp6_mtudisc_callback __P((struct in6_addr *)); 224 #endif 225 226 void tcp_mtudisc __P((struct inpcb *, int)); 227 #ifdef INET6 228 void tcp6_mtudisc __P((struct in6pcb *, int)); 229 #endif 230 231 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL); 232 233 #ifdef TCP_CSUM_COUNTERS 234 #include <sys/device.h> 235 236 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 237 NULL, "tcp", "hwcsum bad"); 238 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 239 NULL, "tcp", "hwcsum ok"); 240 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 241 NULL, "tcp", "hwcsum data"); 242 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 243 NULL, "tcp", "swcsum"); 244 #endif /* TCP_CSUM_COUNTERS */ 245 246 #ifdef TCP_OUTPUT_COUNTERS 247 #include <sys/device.h> 248 249 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 250 NULL, "tcp", "output big header"); 251 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 252 NULL, "tcp", "output predict hit"); 253 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 254 NULL, "tcp", "output predict miss"); 255 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 256 NULL, "tcp", "output copy small"); 257 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 258 NULL, "tcp", "output copy big"); 259 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp", "output reference big"); 261 #endif /* TCP_OUTPUT_COUNTERS */ 262 263 #ifdef TCP_REASS_COUNTERS 264 #include <sys/device.h> 265 266 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 NULL, "tcp_reass", "calls"); 268 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 269 &tcp_reass_, "tcp_reass", "insert into empty queue"); 270 struct evcnt tcp_reass_iteration[8] = { 271 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 272 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 273 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 274 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 275 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 276 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 277 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 278 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 279 }; 280 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 281 &tcp_reass_, "tcp_reass", "prepend to first"); 282 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 &tcp_reass_, "tcp_reass", "prepend"); 284 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 285 &tcp_reass_, "tcp_reass", "insert"); 286 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 287 &tcp_reass_, "tcp_reass", "insert at tail"); 288 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 289 &tcp_reass_, "tcp_reass", "append"); 290 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 291 &tcp_reass_, "tcp_reass", "append to tail fragment"); 292 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 293 &tcp_reass_, "tcp_reass", "overlap at end"); 294 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 295 &tcp_reass_, "tcp_reass", "overlap at start"); 296 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 297 &tcp_reass_, "tcp_reass", "duplicate segment"); 298 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 299 &tcp_reass_, "tcp_reass", "duplicate fragment"); 300 301 #endif /* TCP_REASS_COUNTERS */ 302 303 #ifdef MBUFTRACE 304 struct mowner tcp_mowner = { "tcp" }; 305 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 306 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 307 #endif 308 309 /* 310 * Tcp initialization 311 */ 312 void 313 tcp_init() 314 { 315 int hlen; 316 317 /* Initialize the TCPCB template. */ 318 tcp_tcpcb_template(); 319 320 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 321 322 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 323 #ifdef INET6 324 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 325 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 326 #endif 327 if (max_protohdr < hlen) 328 max_protohdr = hlen; 329 if (max_linkhdr + hlen > MHLEN) 330 panic("tcp_init"); 331 332 #ifdef INET 333 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 334 #endif 335 #ifdef INET6 336 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 337 #endif 338 339 /* Initialize timer state. */ 340 tcp_timer_init(); 341 342 /* Initialize the compressed state engine. */ 343 syn_cache_init(); 344 345 #ifdef TCP_CSUM_COUNTERS 346 evcnt_attach_static(&tcp_hwcsum_bad); 347 evcnt_attach_static(&tcp_hwcsum_ok); 348 evcnt_attach_static(&tcp_hwcsum_data); 349 evcnt_attach_static(&tcp_swcsum); 350 #endif /* TCP_CSUM_COUNTERS */ 351 352 #ifdef TCP_OUTPUT_COUNTERS 353 evcnt_attach_static(&tcp_output_bigheader); 354 evcnt_attach_static(&tcp_output_predict_hit); 355 evcnt_attach_static(&tcp_output_predict_miss); 356 evcnt_attach_static(&tcp_output_copysmall); 357 evcnt_attach_static(&tcp_output_copybig); 358 evcnt_attach_static(&tcp_output_refbig); 359 #endif /* TCP_OUTPUT_COUNTERS */ 360 361 #ifdef TCP_REASS_COUNTERS 362 evcnt_attach_static(&tcp_reass_); 363 evcnt_attach_static(&tcp_reass_empty); 364 evcnt_attach_static(&tcp_reass_iteration[0]); 365 evcnt_attach_static(&tcp_reass_iteration[1]); 366 evcnt_attach_static(&tcp_reass_iteration[2]); 367 evcnt_attach_static(&tcp_reass_iteration[3]); 368 evcnt_attach_static(&tcp_reass_iteration[4]); 369 evcnt_attach_static(&tcp_reass_iteration[5]); 370 evcnt_attach_static(&tcp_reass_iteration[6]); 371 evcnt_attach_static(&tcp_reass_iteration[7]); 372 evcnt_attach_static(&tcp_reass_prependfirst); 373 evcnt_attach_static(&tcp_reass_prepend); 374 evcnt_attach_static(&tcp_reass_insert); 375 evcnt_attach_static(&tcp_reass_inserttail); 376 evcnt_attach_static(&tcp_reass_append); 377 evcnt_attach_static(&tcp_reass_appendtail); 378 evcnt_attach_static(&tcp_reass_overlaptail); 379 evcnt_attach_static(&tcp_reass_overlapfront); 380 evcnt_attach_static(&tcp_reass_segdup); 381 evcnt_attach_static(&tcp_reass_fragdup); 382 #endif /* TCP_REASS_COUNTERS */ 383 384 MOWNER_ATTACH(&tcp_tx_mowner); 385 MOWNER_ATTACH(&tcp_rx_mowner); 386 MOWNER_ATTACH(&tcp_mowner); 387 } 388 389 /* 390 * Create template to be used to send tcp packets on a connection. 391 * Call after host entry created, allocates an mbuf and fills 392 * in a skeletal tcp/ip header, minimizing the amount of work 393 * necessary when the connection is used. 394 */ 395 struct mbuf * 396 tcp_template(tp) 397 struct tcpcb *tp; 398 { 399 struct inpcb *inp = tp->t_inpcb; 400 #ifdef INET6 401 struct in6pcb *in6p = tp->t_in6pcb; 402 #endif 403 struct tcphdr *n; 404 struct mbuf *m; 405 int hlen; 406 407 switch (tp->t_family) { 408 case AF_INET: 409 hlen = sizeof(struct ip); 410 if (inp) 411 break; 412 #ifdef INET6 413 if (in6p) { 414 /* mapped addr case */ 415 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 416 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 417 break; 418 } 419 #endif 420 return NULL; /*EINVAL*/ 421 #ifdef INET6 422 case AF_INET6: 423 hlen = sizeof(struct ip6_hdr); 424 if (in6p) { 425 /* more sainty check? */ 426 break; 427 } 428 return NULL; /*EINVAL*/ 429 #endif 430 default: 431 hlen = 0; /*pacify gcc*/ 432 return NULL; /*EAFNOSUPPORT*/ 433 } 434 #ifdef DIAGNOSTIC 435 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 436 panic("mclbytes too small for t_template"); 437 #endif 438 m = tp->t_template; 439 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 440 ; 441 else { 442 if (m) 443 m_freem(m); 444 m = tp->t_template = NULL; 445 MGETHDR(m, M_DONTWAIT, MT_HEADER); 446 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 447 MCLGET(m, M_DONTWAIT); 448 if ((m->m_flags & M_EXT) == 0) { 449 m_free(m); 450 m = NULL; 451 } 452 } 453 if (m == NULL) 454 return NULL; 455 MCLAIM(m, &tcp_mowner); 456 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 457 } 458 459 bzero(mtod(m, caddr_t), m->m_len); 460 461 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 462 463 switch (tp->t_family) { 464 case AF_INET: 465 { 466 struct ipovly *ipov; 467 mtod(m, struct ip *)->ip_v = 4; 468 mtod(m, struct ip *)->ip_hl = hlen >> 2; 469 ipov = mtod(m, struct ipovly *); 470 ipov->ih_pr = IPPROTO_TCP; 471 ipov->ih_len = htons(sizeof(struct tcphdr)); 472 if (inp) { 473 ipov->ih_src = inp->inp_laddr; 474 ipov->ih_dst = inp->inp_faddr; 475 } 476 #ifdef INET6 477 else if (in6p) { 478 /* mapped addr case */ 479 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 480 sizeof(ipov->ih_src)); 481 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 482 sizeof(ipov->ih_dst)); 483 } 484 #endif 485 /* 486 * Compute the pseudo-header portion of the checksum 487 * now. We incrementally add in the TCP option and 488 * payload lengths later, and then compute the TCP 489 * checksum right before the packet is sent off onto 490 * the wire. 491 */ 492 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 493 ipov->ih_dst.s_addr, 494 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 495 break; 496 } 497 #ifdef INET6 498 case AF_INET6: 499 { 500 struct ip6_hdr *ip6; 501 mtod(m, struct ip *)->ip_v = 6; 502 ip6 = mtod(m, struct ip6_hdr *); 503 ip6->ip6_nxt = IPPROTO_TCP; 504 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 505 ip6->ip6_src = in6p->in6p_laddr; 506 ip6->ip6_dst = in6p->in6p_faddr; 507 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 508 if (ip6_auto_flowlabel) { 509 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 510 ip6->ip6_flow |= 511 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 512 } 513 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 514 ip6->ip6_vfc |= IPV6_VERSION; 515 516 /* 517 * Compute the pseudo-header portion of the checksum 518 * now. We incrementally add in the TCP option and 519 * payload lengths later, and then compute the TCP 520 * checksum right before the packet is sent off onto 521 * the wire. 522 */ 523 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 524 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 525 htonl(IPPROTO_TCP)); 526 break; 527 } 528 #endif 529 } 530 if (inp) { 531 n->th_sport = inp->inp_lport; 532 n->th_dport = inp->inp_fport; 533 } 534 #ifdef INET6 535 else if (in6p) { 536 n->th_sport = in6p->in6p_lport; 537 n->th_dport = in6p->in6p_fport; 538 } 539 #endif 540 n->th_seq = 0; 541 n->th_ack = 0; 542 n->th_x2 = 0; 543 n->th_off = 5; 544 n->th_flags = 0; 545 n->th_win = 0; 546 n->th_urp = 0; 547 return (m); 548 } 549 550 /* 551 * Send a single message to the TCP at address specified by 552 * the given TCP/IP header. If m == 0, then we make a copy 553 * of the tcpiphdr at ti and send directly to the addressed host. 554 * This is used to force keep alive messages out using the TCP 555 * template for a connection tp->t_template. If flags are given 556 * then we send a message back to the TCP which originated the 557 * segment ti, and discard the mbuf containing it and any other 558 * attached mbufs. 559 * 560 * In any case the ack and sequence number of the transmitted 561 * segment are as specified by the parameters. 562 */ 563 int 564 tcp_respond(tp, template, m, th0, ack, seq, flags) 565 struct tcpcb *tp; 566 struct mbuf *template; 567 struct mbuf *m; 568 struct tcphdr *th0; 569 tcp_seq ack, seq; 570 int flags; 571 { 572 struct route *ro; 573 int error, tlen, win = 0; 574 int hlen; 575 struct ip *ip; 576 #ifdef INET6 577 struct ip6_hdr *ip6; 578 #endif 579 int family; /* family on packet, not inpcb/in6pcb! */ 580 struct tcphdr *th; 581 struct socket *so; 582 583 if (tp != NULL && (flags & TH_RST) == 0) { 584 #ifdef DIAGNOSTIC 585 if (tp->t_inpcb && tp->t_in6pcb) 586 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 587 #endif 588 #ifdef INET 589 if (tp->t_inpcb) 590 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 591 #endif 592 #ifdef INET6 593 if (tp->t_in6pcb) 594 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 595 #endif 596 } 597 598 th = NULL; /* Quell uninitialized warning */ 599 ip = NULL; 600 #ifdef INET6 601 ip6 = NULL; 602 #endif 603 if (m == 0) { 604 if (!template) 605 return EINVAL; 606 607 /* get family information from template */ 608 switch (mtod(template, struct ip *)->ip_v) { 609 case 4: 610 family = AF_INET; 611 hlen = sizeof(struct ip); 612 break; 613 #ifdef INET6 614 case 6: 615 family = AF_INET6; 616 hlen = sizeof(struct ip6_hdr); 617 break; 618 #endif 619 default: 620 return EAFNOSUPPORT; 621 } 622 623 MGETHDR(m, M_DONTWAIT, MT_HEADER); 624 if (m) { 625 MCLAIM(m, &tcp_tx_mowner); 626 MCLGET(m, M_DONTWAIT); 627 if ((m->m_flags & M_EXT) == 0) { 628 m_free(m); 629 m = NULL; 630 } 631 } 632 if (m == NULL) 633 return (ENOBUFS); 634 635 if (tcp_compat_42) 636 tlen = 1; 637 else 638 tlen = 0; 639 640 m->m_data += max_linkhdr; 641 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 642 template->m_len); 643 switch (family) { 644 case AF_INET: 645 ip = mtod(m, struct ip *); 646 th = (struct tcphdr *)(ip + 1); 647 break; 648 #ifdef INET6 649 case AF_INET6: 650 ip6 = mtod(m, struct ip6_hdr *); 651 th = (struct tcphdr *)(ip6 + 1); 652 break; 653 #endif 654 #if 0 655 default: 656 /* noone will visit here */ 657 m_freem(m); 658 return EAFNOSUPPORT; 659 #endif 660 } 661 flags = TH_ACK; 662 } else { 663 664 if ((m->m_flags & M_PKTHDR) == 0) { 665 #if 0 666 printf("non PKTHDR to tcp_respond\n"); 667 #endif 668 m_freem(m); 669 return EINVAL; 670 } 671 #ifdef DIAGNOSTIC 672 if (!th0) 673 panic("th0 == NULL in tcp_respond"); 674 #endif 675 676 /* get family information from m */ 677 switch (mtod(m, struct ip *)->ip_v) { 678 case 4: 679 family = AF_INET; 680 hlen = sizeof(struct ip); 681 ip = mtod(m, struct ip *); 682 break; 683 #ifdef INET6 684 case 6: 685 family = AF_INET6; 686 hlen = sizeof(struct ip6_hdr); 687 ip6 = mtod(m, struct ip6_hdr *); 688 break; 689 #endif 690 default: 691 m_freem(m); 692 return EAFNOSUPPORT; 693 } 694 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 695 tlen = sizeof(*th0); 696 else 697 tlen = th0->th_off << 2; 698 699 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 700 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 701 m->m_len = hlen + tlen; 702 m_freem(m->m_next); 703 m->m_next = NULL; 704 } else { 705 struct mbuf *n; 706 707 #ifdef DIAGNOSTIC 708 if (max_linkhdr + hlen + tlen > MCLBYTES) { 709 m_freem(m); 710 return EMSGSIZE; 711 } 712 #endif 713 MGETHDR(n, M_DONTWAIT, MT_HEADER); 714 if (n && max_linkhdr + hlen + tlen > MHLEN) { 715 MCLGET(n, M_DONTWAIT); 716 if ((n->m_flags & M_EXT) == 0) { 717 m_freem(n); 718 n = NULL; 719 } 720 } 721 if (!n) { 722 m_freem(m); 723 return ENOBUFS; 724 } 725 726 MCLAIM(n, &tcp_tx_mowner); 727 n->m_data += max_linkhdr; 728 n->m_len = hlen + tlen; 729 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 730 m_copyback(n, hlen, tlen, (caddr_t)th0); 731 732 m_freem(m); 733 m = n; 734 n = NULL; 735 } 736 737 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 738 switch (family) { 739 case AF_INET: 740 ip = mtod(m, struct ip *); 741 th = (struct tcphdr *)(ip + 1); 742 ip->ip_p = IPPROTO_TCP; 743 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 744 ip->ip_p = IPPROTO_TCP; 745 break; 746 #ifdef INET6 747 case AF_INET6: 748 ip6 = mtod(m, struct ip6_hdr *); 749 th = (struct tcphdr *)(ip6 + 1); 750 ip6->ip6_nxt = IPPROTO_TCP; 751 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 752 ip6->ip6_nxt = IPPROTO_TCP; 753 break; 754 #endif 755 #if 0 756 default: 757 /* noone will visit here */ 758 m_freem(m); 759 return EAFNOSUPPORT; 760 #endif 761 } 762 xchg(th->th_dport, th->th_sport, u_int16_t); 763 #undef xchg 764 tlen = 0; /*be friendly with the following code*/ 765 } 766 th->th_seq = htonl(seq); 767 th->th_ack = htonl(ack); 768 th->th_x2 = 0; 769 if ((flags & TH_SYN) == 0) { 770 if (tp) 771 win >>= tp->rcv_scale; 772 if (win > TCP_MAXWIN) 773 win = TCP_MAXWIN; 774 th->th_win = htons((u_int16_t)win); 775 th->th_off = sizeof (struct tcphdr) >> 2; 776 tlen += sizeof(*th); 777 } else 778 tlen += th->th_off << 2; 779 m->m_len = hlen + tlen; 780 m->m_pkthdr.len = hlen + tlen; 781 m->m_pkthdr.rcvif = (struct ifnet *) 0; 782 th->th_flags = flags; 783 th->th_urp = 0; 784 785 switch (family) { 786 #ifdef INET 787 case AF_INET: 788 { 789 struct ipovly *ipov = (struct ipovly *)ip; 790 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 791 ipov->ih_len = htons((u_int16_t)tlen); 792 793 th->th_sum = 0; 794 th->th_sum = in_cksum(m, hlen + tlen); 795 ip->ip_len = htons(hlen + tlen); 796 ip->ip_ttl = ip_defttl; 797 break; 798 } 799 #endif 800 #ifdef INET6 801 case AF_INET6: 802 { 803 th->th_sum = 0; 804 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 805 tlen); 806 ip6->ip6_plen = ntohs(tlen); 807 if (tp && tp->t_in6pcb) { 808 struct ifnet *oifp; 809 ro = (struct route *)&tp->t_in6pcb->in6p_route; 810 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 811 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 812 } else 813 ip6->ip6_hlim = ip6_defhlim; 814 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 815 if (ip6_auto_flowlabel) { 816 ip6->ip6_flow |= 817 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 818 } 819 break; 820 } 821 #endif 822 } 823 824 if (tp && tp->t_inpcb) 825 so = tp->t_inpcb->inp_socket; 826 #ifdef INET6 827 else if (tp && tp->t_in6pcb) 828 so = tp->t_in6pcb->in6p_socket; 829 #endif 830 else 831 so = NULL; 832 833 if (tp != NULL && tp->t_inpcb != NULL) { 834 ro = &tp->t_inpcb->inp_route; 835 #ifdef DIAGNOSTIC 836 if (family != AF_INET) 837 panic("tcp_respond: address family mismatch"); 838 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 839 panic("tcp_respond: ip_dst %x != inp_faddr %x", 840 ntohl(ip->ip_dst.s_addr), 841 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 842 } 843 #endif 844 } 845 #ifdef INET6 846 else if (tp != NULL && tp->t_in6pcb != NULL) { 847 ro = (struct route *)&tp->t_in6pcb->in6p_route; 848 #ifdef DIAGNOSTIC 849 if (family == AF_INET) { 850 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 851 panic("tcp_respond: not mapped addr"); 852 if (bcmp(&ip->ip_dst, 853 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 854 sizeof(ip->ip_dst)) != 0) { 855 panic("tcp_respond: ip_dst != in6p_faddr"); 856 } 857 } else if (family == AF_INET6) { 858 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 859 &tp->t_in6pcb->in6p_faddr)) 860 panic("tcp_respond: ip6_dst != in6p_faddr"); 861 } else 862 panic("tcp_respond: address family mismatch"); 863 #endif 864 } 865 #endif 866 else 867 ro = NULL; 868 869 switch (family) { 870 #ifdef INET 871 case AF_INET: 872 error = ip_output(m, NULL, ro, 873 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 874 (struct ip_moptions *)0, so); 875 break; 876 #endif 877 #ifdef INET6 878 case AF_INET6: 879 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 880 (struct ip6_moptions *)0, so, NULL); 881 break; 882 #endif 883 default: 884 error = EAFNOSUPPORT; 885 break; 886 } 887 888 return (error); 889 } 890 891 /* 892 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 893 * a bunch of members individually, we maintain this template for the 894 * static and mostly-static components of the TCPCB, and copy it into 895 * the new TCPCB instead. 896 */ 897 static struct tcpcb tcpcb_template = { 898 /* 899 * If TCP_NTIMERS ever changes, we'll need to update this 900 * initializer. 901 */ 902 .t_timer = { 903 CALLOUT_INITIALIZER, 904 CALLOUT_INITIALIZER, 905 CALLOUT_INITIALIZER, 906 CALLOUT_INITIALIZER, 907 }, 908 .t_delack_ch = CALLOUT_INITIALIZER, 909 910 .t_srtt = TCPTV_SRTTBASE, 911 .t_rttmin = TCPTV_MIN, 912 913 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 914 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 915 }; 916 917 /* 918 * Updates the TCPCB template whenever a parameter that would affect 919 * the template is changed. 920 */ 921 void 922 tcp_tcpcb_template(void) 923 { 924 struct tcpcb *tp = &tcpcb_template; 925 int flags; 926 927 tp->t_peermss = tcp_mssdflt; 928 tp->t_ourmss = tcp_mssdflt; 929 tp->t_segsz = tcp_mssdflt; 930 931 flags = 0; 932 if (tcp_do_rfc1323 && tcp_do_win_scale) 933 flags |= TF_REQ_SCALE; 934 if (tcp_do_rfc1323 && tcp_do_timestamps) 935 flags |= TF_REQ_TSTMP; 936 if (tcp_do_sack == 2) 937 flags |= TF_WILL_SACK; 938 else if (tcp_do_sack == 1) 939 flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 940 flags |= TF_CANT_TXSACK; 941 tp->t_flags = flags; 942 943 /* 944 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 945 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 946 * reasonable initial retransmit time. 947 */ 948 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 949 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 950 TCPTV_MIN, TCPTV_REXMTMAX); 951 } 952 953 /* 954 * Create a new TCP control block, making an 955 * empty reassembly queue and hooking it to the argument 956 * protocol control block. 957 */ 958 struct tcpcb * 959 tcp_newtcpcb(family, aux) 960 int family; /* selects inpcb, or in6pcb */ 961 void *aux; 962 { 963 struct tcpcb *tp; 964 int i; 965 966 /* XXX Consider using a pool_cache for speed. */ 967 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 968 if (tp == NULL) 969 return (NULL); 970 memcpy(tp, &tcpcb_template, sizeof(*tp)); 971 TAILQ_INIT(&tp->segq); 972 TAILQ_INIT(&tp->timeq); 973 tp->t_family = family; /* may be overridden later on */ 974 LIST_INIT(&tp->t_sc); /* XXX can template this */ 975 976 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 977 for (i = 0; i < TCPT_NTIMERS; i++) 978 TCP_TIMER_INIT(tp, i); 979 980 switch (family) { 981 case AF_INET: 982 { 983 struct inpcb *inp = (struct inpcb *)aux; 984 985 inp->inp_ip.ip_ttl = ip_defttl; 986 inp->inp_ppcb = (caddr_t)tp; 987 988 tp->t_inpcb = inp; 989 tp->t_mtudisc = ip_mtudisc; 990 break; 991 } 992 #ifdef INET6 993 case AF_INET6: 994 { 995 struct in6pcb *in6p = (struct in6pcb *)aux; 996 997 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 998 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 999 : NULL); 1000 in6p->in6p_ppcb = (caddr_t)tp; 1001 1002 tp->t_in6pcb = in6p; 1003 /* for IPv6, always try to run path MTU discovery */ 1004 tp->t_mtudisc = 1; 1005 break; 1006 } 1007 #endif /* INET6 */ 1008 default: 1009 pool_put(&tcpcb_pool, tp); 1010 return (NULL); 1011 } 1012 1013 /* 1014 * Initialize our timebase. When we send timestamps, we take 1015 * the delta from tcp_now -- this means each connection always 1016 * gets a timebase of 0, which makes it, among other things, 1017 * more difficult to determine how long a system has been up, 1018 * and thus how many TCP sequence increments have occurred. 1019 */ 1020 tp->ts_timebase = tcp_now; 1021 1022 return (tp); 1023 } 1024 1025 /* 1026 * Drop a TCP connection, reporting 1027 * the specified error. If connection is synchronized, 1028 * then send a RST to peer. 1029 */ 1030 struct tcpcb * 1031 tcp_drop(tp, errno) 1032 struct tcpcb *tp; 1033 int errno; 1034 { 1035 struct socket *so = NULL; 1036 1037 #ifdef DIAGNOSTIC 1038 if (tp->t_inpcb && tp->t_in6pcb) 1039 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1040 #endif 1041 #ifdef INET 1042 if (tp->t_inpcb) 1043 so = tp->t_inpcb->inp_socket; 1044 #endif 1045 #ifdef INET6 1046 if (tp->t_in6pcb) 1047 so = tp->t_in6pcb->in6p_socket; 1048 #endif 1049 if (!so) 1050 return NULL; 1051 1052 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1053 tp->t_state = TCPS_CLOSED; 1054 (void) tcp_output(tp); 1055 tcpstat.tcps_drops++; 1056 } else 1057 tcpstat.tcps_conndrops++; 1058 if (errno == ETIMEDOUT && tp->t_softerror) 1059 errno = tp->t_softerror; 1060 so->so_error = errno; 1061 return (tcp_close(tp)); 1062 } 1063 1064 /* 1065 * Return whether this tcpcb is marked as dead, indicating 1066 * to the calling timer function that no further action should 1067 * be taken, as we are about to release this tcpcb. The release 1068 * of the storage will be done if this is the last timer running. 1069 * 1070 * This should be called from the callout handler function after 1071 * callout_ack() is done, so that the number of invoking timer 1072 * functions is 0. 1073 */ 1074 int 1075 tcp_isdead(tp) 1076 struct tcpcb *tp; 1077 { 1078 int dead = (tp->t_flags & TF_DEAD); 1079 1080 if (__predict_false(dead)) { 1081 if (tcp_timers_invoking(tp) > 0) 1082 /* not quite there yet -- count separately? */ 1083 return dead; 1084 tcpstat.tcps_delayed_free++; 1085 pool_put(&tcpcb_pool, tp); 1086 } 1087 return dead; 1088 } 1089 1090 /* 1091 * Close a TCP control block: 1092 * discard all space held by the tcp 1093 * discard internet protocol block 1094 * wake up any sleepers 1095 */ 1096 struct tcpcb * 1097 tcp_close(tp) 1098 struct tcpcb *tp; 1099 { 1100 struct inpcb *inp; 1101 #ifdef INET6 1102 struct in6pcb *in6p; 1103 #endif 1104 struct socket *so; 1105 #ifdef RTV_RTT 1106 struct rtentry *rt; 1107 #endif 1108 struct route *ro; 1109 1110 inp = tp->t_inpcb; 1111 #ifdef INET6 1112 in6p = tp->t_in6pcb; 1113 #endif 1114 so = NULL; 1115 ro = NULL; 1116 if (inp) { 1117 so = inp->inp_socket; 1118 ro = &inp->inp_route; 1119 } 1120 #ifdef INET6 1121 else if (in6p) { 1122 so = in6p->in6p_socket; 1123 ro = (struct route *)&in6p->in6p_route; 1124 } 1125 #endif 1126 1127 #ifdef RTV_RTT 1128 /* 1129 * If we sent enough data to get some meaningful characteristics, 1130 * save them in the routing entry. 'Enough' is arbitrarily 1131 * defined as the sendpipesize (default 4K) * 16. This would 1132 * give us 16 rtt samples assuming we only get one sample per 1133 * window (the usual case on a long haul net). 16 samples is 1134 * enough for the srtt filter to converge to within 5% of the correct 1135 * value; fewer samples and we could save a very bogus rtt. 1136 * 1137 * Don't update the default route's characteristics and don't 1138 * update anything that the user "locked". 1139 */ 1140 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1141 ro && (rt = ro->ro_rt) && 1142 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1143 u_long i = 0; 1144 1145 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1146 i = tp->t_srtt * 1147 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1148 if (rt->rt_rmx.rmx_rtt && i) 1149 /* 1150 * filter this update to half the old & half 1151 * the new values, converting scale. 1152 * See route.h and tcp_var.h for a 1153 * description of the scaling constants. 1154 */ 1155 rt->rt_rmx.rmx_rtt = 1156 (rt->rt_rmx.rmx_rtt + i) / 2; 1157 else 1158 rt->rt_rmx.rmx_rtt = i; 1159 } 1160 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1161 i = tp->t_rttvar * 1162 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1163 if (rt->rt_rmx.rmx_rttvar && i) 1164 rt->rt_rmx.rmx_rttvar = 1165 (rt->rt_rmx.rmx_rttvar + i) / 2; 1166 else 1167 rt->rt_rmx.rmx_rttvar = i; 1168 } 1169 /* 1170 * update the pipelimit (ssthresh) if it has been updated 1171 * already or if a pipesize was specified & the threshhold 1172 * got below half the pipesize. I.e., wait for bad news 1173 * before we start updating, then update on both good 1174 * and bad news. 1175 */ 1176 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1177 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1178 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1179 /* 1180 * convert the limit from user data bytes to 1181 * packets then to packet data bytes. 1182 */ 1183 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1184 if (i < 2) 1185 i = 2; 1186 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1187 if (rt->rt_rmx.rmx_ssthresh) 1188 rt->rt_rmx.rmx_ssthresh = 1189 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1190 else 1191 rt->rt_rmx.rmx_ssthresh = i; 1192 } 1193 } 1194 #endif /* RTV_RTT */ 1195 /* free the reassembly queue, if any */ 1196 TCP_REASS_LOCK(tp); 1197 (void) tcp_freeq(tp); 1198 TCP_REASS_UNLOCK(tp); 1199 1200 tcp_canceltimers(tp); 1201 TCP_CLEAR_DELACK(tp); 1202 syn_cache_cleanup(tp); 1203 1204 if (tp->t_template) { 1205 m_free(tp->t_template); 1206 tp->t_template = NULL; 1207 } 1208 if (tcp_timers_invoking(tp)) 1209 tp->t_flags |= TF_DEAD; 1210 else 1211 pool_put(&tcpcb_pool, tp); 1212 1213 if (inp) { 1214 inp->inp_ppcb = 0; 1215 soisdisconnected(so); 1216 in_pcbdetach(inp); 1217 } 1218 #ifdef INET6 1219 else if (in6p) { 1220 in6p->in6p_ppcb = 0; 1221 soisdisconnected(so); 1222 in6_pcbdetach(in6p); 1223 } 1224 #endif 1225 tcpstat.tcps_closed++; 1226 return ((struct tcpcb *)0); 1227 } 1228 1229 int 1230 tcp_freeq(tp) 1231 struct tcpcb *tp; 1232 { 1233 struct ipqent *qe; 1234 int rv = 0; 1235 #ifdef TCPREASS_DEBUG 1236 int i = 0; 1237 #endif 1238 1239 TCP_REASS_LOCK_CHECK(tp); 1240 1241 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1242 #ifdef TCPREASS_DEBUG 1243 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1244 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1245 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1246 #endif 1247 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1248 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1249 m_freem(qe->ipqe_m); 1250 pool_put(&ipqent_pool, qe); 1251 rv = 1; 1252 } 1253 return (rv); 1254 } 1255 1256 /* 1257 * Protocol drain routine. Called when memory is in short supply. 1258 */ 1259 void 1260 tcp_drain() 1261 { 1262 struct inpcb_hdr *inph; 1263 struct tcpcb *tp; 1264 1265 /* 1266 * Free the sequence queue of all TCP connections. 1267 */ 1268 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1269 switch (inph->inph_af) { 1270 case AF_INET: 1271 tp = intotcpcb((struct inpcb *)inph); 1272 break; 1273 #ifdef INET6 1274 case AF_INET6: 1275 tp = in6totcpcb((struct in6pcb *)inph); 1276 break; 1277 #endif 1278 default: 1279 tp = NULL; 1280 break; 1281 } 1282 if (tp != NULL) { 1283 /* 1284 * We may be called from a device's interrupt 1285 * context. If the tcpcb is already busy, 1286 * just bail out now. 1287 */ 1288 if (tcp_reass_lock_try(tp) == 0) 1289 continue; 1290 if (tcp_freeq(tp)) 1291 tcpstat.tcps_connsdrained++; 1292 TCP_REASS_UNLOCK(tp); 1293 } 1294 } 1295 } 1296 1297 /* 1298 * Notify a tcp user of an asynchronous error; 1299 * store error as soft error, but wake up user 1300 * (for now, won't do anything until can select for soft error). 1301 */ 1302 void 1303 tcp_notify(inp, error) 1304 struct inpcb *inp; 1305 int error; 1306 { 1307 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1308 struct socket *so = inp->inp_socket; 1309 1310 /* 1311 * Ignore some errors if we are hooked up. 1312 * If connection hasn't completed, has retransmitted several times, 1313 * and receives a second error, give up now. This is better 1314 * than waiting a long time to establish a connection that 1315 * can never complete. 1316 */ 1317 if (tp->t_state == TCPS_ESTABLISHED && 1318 (error == EHOSTUNREACH || error == ENETUNREACH || 1319 error == EHOSTDOWN)) { 1320 return; 1321 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1322 tp->t_rxtshift > 3 && tp->t_softerror) 1323 so->so_error = error; 1324 else 1325 tp->t_softerror = error; 1326 wakeup((caddr_t) &so->so_timeo); 1327 sorwakeup(so); 1328 sowwakeup(so); 1329 } 1330 1331 #ifdef INET6 1332 void 1333 tcp6_notify(in6p, error) 1334 struct in6pcb *in6p; 1335 int error; 1336 { 1337 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1338 struct socket *so = in6p->in6p_socket; 1339 1340 /* 1341 * Ignore some errors if we are hooked up. 1342 * If connection hasn't completed, has retransmitted several times, 1343 * and receives a second error, give up now. This is better 1344 * than waiting a long time to establish a connection that 1345 * can never complete. 1346 */ 1347 if (tp->t_state == TCPS_ESTABLISHED && 1348 (error == EHOSTUNREACH || error == ENETUNREACH || 1349 error == EHOSTDOWN)) { 1350 return; 1351 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1352 tp->t_rxtshift > 3 && tp->t_softerror) 1353 so->so_error = error; 1354 else 1355 tp->t_softerror = error; 1356 wakeup((caddr_t) &so->so_timeo); 1357 sorwakeup(so); 1358 sowwakeup(so); 1359 } 1360 #endif 1361 1362 #ifdef INET6 1363 void 1364 tcp6_ctlinput(cmd, sa, d) 1365 int cmd; 1366 struct sockaddr *sa; 1367 void *d; 1368 { 1369 struct tcphdr th; 1370 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1371 int nmatch; 1372 struct ip6_hdr *ip6; 1373 const struct sockaddr_in6 *sa6_src = NULL; 1374 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1375 struct mbuf *m; 1376 int off; 1377 1378 if (sa->sa_family != AF_INET6 || 1379 sa->sa_len != sizeof(struct sockaddr_in6)) 1380 return; 1381 if ((unsigned)cmd >= PRC_NCMDS) 1382 return; 1383 else if (cmd == PRC_QUENCH) { 1384 /* XXX there's no PRC_QUENCH in IPv6 */ 1385 notify = tcp6_quench; 1386 } else if (PRC_IS_REDIRECT(cmd)) 1387 notify = in6_rtchange, d = NULL; 1388 else if (cmd == PRC_MSGSIZE) 1389 ; /* special code is present, see below */ 1390 else if (cmd == PRC_HOSTDEAD) 1391 d = NULL; 1392 else if (inet6ctlerrmap[cmd] == 0) 1393 return; 1394 1395 /* if the parameter is from icmp6, decode it. */ 1396 if (d != NULL) { 1397 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1398 m = ip6cp->ip6c_m; 1399 ip6 = ip6cp->ip6c_ip6; 1400 off = ip6cp->ip6c_off; 1401 sa6_src = ip6cp->ip6c_src; 1402 } else { 1403 m = NULL; 1404 ip6 = NULL; 1405 sa6_src = &sa6_any; 1406 off = 0; 1407 } 1408 1409 if (ip6) { 1410 /* 1411 * XXX: We assume that when ip6 is non NULL, 1412 * M and OFF are valid. 1413 */ 1414 1415 /* check if we can safely examine src and dst ports */ 1416 if (m->m_pkthdr.len < off + sizeof(th)) { 1417 if (cmd == PRC_MSGSIZE) 1418 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1419 return; 1420 } 1421 1422 bzero(&th, sizeof(th)); 1423 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1424 1425 if (cmd == PRC_MSGSIZE) { 1426 int valid = 0; 1427 1428 /* 1429 * Check to see if we have a valid TCP connection 1430 * corresponding to the address in the ICMPv6 message 1431 * payload. 1432 */ 1433 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1434 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1435 th.th_sport, 0)) 1436 valid++; 1437 1438 /* 1439 * Depending on the value of "valid" and routing table 1440 * size (mtudisc_{hi,lo}wat), we will: 1441 * - recalcurate the new MTU and create the 1442 * corresponding routing entry, or 1443 * - ignore the MTU change notification. 1444 */ 1445 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1446 1447 /* 1448 * no need to call in6_pcbnotify, it should have been 1449 * called via callback if necessary 1450 */ 1451 return; 1452 } 1453 1454 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1455 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1456 if (nmatch == 0 && syn_cache_count && 1457 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1458 inet6ctlerrmap[cmd] == ENETUNREACH || 1459 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1460 syn_cache_unreach((struct sockaddr *)sa6_src, 1461 sa, &th); 1462 } else { 1463 (void) in6_pcbnotify(&tcbtable, sa, 0, 1464 (struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1465 } 1466 } 1467 #endif 1468 1469 #ifdef INET 1470 /* assumes that ip header and tcp header are contiguous on mbuf */ 1471 void * 1472 tcp_ctlinput(cmd, sa, v) 1473 int cmd; 1474 struct sockaddr *sa; 1475 void *v; 1476 { 1477 struct ip *ip = v; 1478 struct tcphdr *th; 1479 struct icmp *icp; 1480 extern const int inetctlerrmap[]; 1481 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1482 int errno; 1483 int nmatch; 1484 #ifdef INET6 1485 struct in6_addr src6, dst6; 1486 #endif 1487 1488 if (sa->sa_family != AF_INET || 1489 sa->sa_len != sizeof(struct sockaddr_in)) 1490 return NULL; 1491 if ((unsigned)cmd >= PRC_NCMDS) 1492 return NULL; 1493 errno = inetctlerrmap[cmd]; 1494 if (cmd == PRC_QUENCH) 1495 notify = tcp_quench; 1496 else if (PRC_IS_REDIRECT(cmd)) 1497 notify = in_rtchange, ip = 0; 1498 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1499 /* 1500 * Check to see if we have a valid TCP connection 1501 * corresponding to the address in the ICMP message 1502 * payload. 1503 * 1504 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1505 */ 1506 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1507 #ifdef INET6 1508 memset(&src6, 0, sizeof(src6)); 1509 memset(&dst6, 0, sizeof(dst6)); 1510 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1511 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1512 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1513 #endif 1514 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1515 ip->ip_src, th->th_sport) != NULL) 1516 ; 1517 #ifdef INET6 1518 else if (in6_pcblookup_connect(&tcbtable, &dst6, 1519 th->th_dport, &src6, th->th_sport, 0) != NULL) 1520 ; 1521 #endif 1522 else 1523 return NULL; 1524 1525 /* 1526 * Now that we've validated that we are actually communicating 1527 * with the host indicated in the ICMP message, locate the 1528 * ICMP header, recalculate the new MTU, and create the 1529 * corresponding routing entry. 1530 */ 1531 icp = (struct icmp *)((caddr_t)ip - 1532 offsetof(struct icmp, icmp_ip)); 1533 icmp_mtudisc(icp, ip->ip_dst); 1534 1535 return NULL; 1536 } else if (cmd == PRC_HOSTDEAD) 1537 ip = 0; 1538 else if (errno == 0) 1539 return NULL; 1540 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1541 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1542 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1543 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1544 if (nmatch == 0 && syn_cache_count && 1545 (inetctlerrmap[cmd] == EHOSTUNREACH || 1546 inetctlerrmap[cmd] == ENETUNREACH || 1547 inetctlerrmap[cmd] == EHOSTDOWN)) { 1548 struct sockaddr_in sin; 1549 bzero(&sin, sizeof(sin)); 1550 sin.sin_len = sizeof(sin); 1551 sin.sin_family = AF_INET; 1552 sin.sin_port = th->th_sport; 1553 sin.sin_addr = ip->ip_src; 1554 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1555 } 1556 1557 /* XXX mapped address case */ 1558 } else 1559 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1560 notify); 1561 return NULL; 1562 } 1563 1564 /* 1565 * When a source quence is received, we are being notifed of congestion. 1566 * Close the congestion window down to the Loss Window (one segment). 1567 * We will gradually open it again as we proceed. 1568 */ 1569 void 1570 tcp_quench(inp, errno) 1571 struct inpcb *inp; 1572 int errno; 1573 { 1574 struct tcpcb *tp = intotcpcb(inp); 1575 1576 if (tp) 1577 tp->snd_cwnd = tp->t_segsz; 1578 } 1579 #endif 1580 1581 #ifdef INET6 1582 void 1583 tcp6_quench(in6p, errno) 1584 struct in6pcb *in6p; 1585 int errno; 1586 { 1587 struct tcpcb *tp = in6totcpcb(in6p); 1588 1589 if (tp) 1590 tp->snd_cwnd = tp->t_segsz; 1591 } 1592 #endif 1593 1594 #ifdef INET 1595 /* 1596 * Path MTU Discovery handlers. 1597 */ 1598 void 1599 tcp_mtudisc_callback(faddr) 1600 struct in_addr faddr; 1601 { 1602 #ifdef INET6 1603 struct in6_addr in6; 1604 #endif 1605 1606 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1607 #ifdef INET6 1608 memset(&in6, 0, sizeof(in6)); 1609 in6.s6_addr16[5] = 0xffff; 1610 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1611 tcp6_mtudisc_callback(&in6); 1612 #endif 1613 } 1614 1615 #ifdef TCP_SIGNATURE 1616 /* 1617 * Callback function invoked by m_apply() to digest TCP segment data 1618 * contained within an mbuf chain. 1619 */ 1620 static int 1621 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 1622 { 1623 1624 MD5Update(fstate, (u_char *)data, len); 1625 return (0); 1626 } 1627 1628 /* 1629 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 1630 * 1631 * Parameters: 1632 * m pointer to head of mbuf chain 1633 * off0 offset to TCP header within the mbuf chain 1634 * len length of TCP segment data, excluding options 1635 * optlen length of TCP segment options 1636 * buf pointer to storage for computed MD5 digest 1637 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 1638 * 1639 * We do this over ip, tcphdr, segment data, and the key in the SADB. 1640 * When called from tcp_input(), we can be sure that th_sum has been 1641 * zeroed out and verified already. 1642 * 1643 * This function is for IPv4 use only. Calling this function with an 1644 * IPv6 packet in the mbuf chain will yield undefined results. 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, int off0, int len, int optlen, 1655 u_char *buf, u_int direction) 1656 { 1657 union sockaddr_union dst; 1658 MD5_CTX ctx; 1659 int doff; 1660 struct ip *ip; 1661 struct ipovly *ipovly; 1662 struct secasvar *sav; 1663 struct tcphdr *th; 1664 u_short savecsum; 1665 struct ippseudo ippseudo; 1666 1667 1668 KASSERT(m != NULL /*, ("NULL mbuf chain")*/); 1669 KASSERT(buf != NULL /*, ("NULL signature pointer")*/); 1670 1671 /* Extract the destination from the IP header in the mbuf. */ 1672 ip = mtod(m, struct ip *); 1673 bzero(&dst, sizeof(union sockaddr_union)); 1674 dst.sa.sa_len = sizeof(struct sockaddr_in); 1675 dst.sa.sa_family = AF_INET; 1676 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 1677 ip->ip_src : ip->ip_dst; 1678 1679 /* Look up an SADB entry which matches the address of the peer. */ 1680 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1681 if (sav == NULL) { 1682 printf("%s: SADB lookup failed for %s\n", __func__, 1683 inet_ntoa(dst.sin.sin_addr)); 1684 return (EINVAL); 1685 } 1686 1687 MD5Init(&ctx); 1688 ipovly = (struct ipovly *)ip; 1689 th = (struct tcphdr *)((u_char *)ip + off0); 1690 doff = off0 + sizeof(struct tcphdr) + optlen; 1691 1692 /* 1693 * Step 1: Update MD5 hash with IP pseudo-header. 1694 * 1695 * XXX The ippseudo header MUST be digested in network byte order, 1696 * or else we'll fail the regression test. Assume all fields we've 1697 * been doing arithmetic on have been in host byte order. 1698 * XXX One cannot depend on ipovly->ih_len here. When called from 1699 * tcp_output(), the underlying ip_len member has not yet been set. 1700 */ 1701 ippseudo.ippseudo_src = ipovly->ih_src; 1702 ippseudo.ippseudo_dst = ipovly->ih_dst; 1703 ippseudo.ippseudo_pad = 0; 1704 ippseudo.ippseudo_p = IPPROTO_TCP; 1705 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 1706 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 1707 1708 /* 1709 * Step 2: Update MD5 hash with TCP header, excluding options. 1710 * The TCP checksum must be set to zero. 1711 */ 1712 savecsum = th->th_sum; 1713 th->th_sum = 0; 1714 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 1715 th->th_sum = savecsum; 1716 1717 /* 1718 * Step 3: Update MD5 hash with TCP segment data. 1719 * Use m_apply() to avoid an early m_pullup(). 1720 */ 1721 if (len > 0) 1722 m_apply(m, doff, len, tcp_signature_apply, &ctx); 1723 1724 /* 1725 * Step 4: Update MD5 hash with shared secret. 1726 */ 1727 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 1728 MD5Final(buf, &ctx); 1729 1730 key_sa_recordxfer(sav, m); 1731 KEY_FREESAV(&sav); 1732 return (0); 1733 } 1734 #endif /* TCP_SIGNATURE */ 1735 1736 /* 1737 * On receipt of path MTU corrections, flush old route and replace it 1738 * with the new one. Retransmit all unacknowledged packets, to ensure 1739 * that all packets will be received. 1740 */ 1741 void 1742 tcp_mtudisc(inp, errno) 1743 struct inpcb *inp; 1744 int errno; 1745 { 1746 struct tcpcb *tp = intotcpcb(inp); 1747 struct rtentry *rt = in_pcbrtentry(inp); 1748 1749 if (tp != 0) { 1750 if (rt != 0) { 1751 /* 1752 * If this was not a host route, remove and realloc. 1753 */ 1754 if ((rt->rt_flags & RTF_HOST) == 0) { 1755 in_rtchange(inp, errno); 1756 if ((rt = in_pcbrtentry(inp)) == 0) 1757 return; 1758 } 1759 1760 /* 1761 * Slow start out of the error condition. We 1762 * use the MTU because we know it's smaller 1763 * than the previously transmitted segment. 1764 * 1765 * Note: This is more conservative than the 1766 * suggestion in draft-floyd-incr-init-win-03. 1767 */ 1768 if (rt->rt_rmx.rmx_mtu != 0) 1769 tp->snd_cwnd = 1770 TCP_INITIAL_WINDOW(tcp_init_win, 1771 rt->rt_rmx.rmx_mtu); 1772 } 1773 1774 /* 1775 * Resend unacknowledged packets. 1776 */ 1777 tp->snd_nxt = tp->snd_una; 1778 tcp_output(tp); 1779 } 1780 } 1781 #endif 1782 1783 #ifdef INET6 1784 /* 1785 * Path MTU Discovery handlers. 1786 */ 1787 void 1788 tcp6_mtudisc_callback(faddr) 1789 struct in6_addr *faddr; 1790 { 1791 struct sockaddr_in6 sin6; 1792 1793 bzero(&sin6, sizeof(sin6)); 1794 sin6.sin6_family = AF_INET6; 1795 sin6.sin6_len = sizeof(struct sockaddr_in6); 1796 sin6.sin6_addr = *faddr; 1797 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1798 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1799 } 1800 1801 void 1802 tcp6_mtudisc(in6p, errno) 1803 struct in6pcb *in6p; 1804 int errno; 1805 { 1806 struct tcpcb *tp = in6totcpcb(in6p); 1807 struct rtentry *rt = in6_pcbrtentry(in6p); 1808 1809 if (tp != 0) { 1810 if (rt != 0) { 1811 /* 1812 * If this was not a host route, remove and realloc. 1813 */ 1814 if ((rt->rt_flags & RTF_HOST) == 0) { 1815 in6_rtchange(in6p, errno); 1816 if ((rt = in6_pcbrtentry(in6p)) == 0) 1817 return; 1818 } 1819 1820 /* 1821 * Slow start out of the error condition. We 1822 * use the MTU because we know it's smaller 1823 * than the previously transmitted segment. 1824 * 1825 * Note: This is more conservative than the 1826 * suggestion in draft-floyd-incr-init-win-03. 1827 */ 1828 if (rt->rt_rmx.rmx_mtu != 0) 1829 tp->snd_cwnd = 1830 TCP_INITIAL_WINDOW(tcp_init_win, 1831 rt->rt_rmx.rmx_mtu); 1832 } 1833 1834 /* 1835 * Resend unacknowledged packets. 1836 */ 1837 tp->snd_nxt = tp->snd_una; 1838 tcp_output(tp); 1839 } 1840 } 1841 #endif /* INET6 */ 1842 1843 /* 1844 * Compute the MSS to advertise to the peer. Called only during 1845 * the 3-way handshake. If we are the server (peer initiated 1846 * connection), we are called with a pointer to the interface 1847 * on which the SYN packet arrived. If we are the client (we 1848 * initiated connection), we are called with a pointer to the 1849 * interface out which this connection should go. 1850 * 1851 * NOTE: Do not subtract IP option/extension header size nor IPsec 1852 * header size from MSS advertisement. MSS option must hold the maximum 1853 * segment size we can accept, so it must always be: 1854 * max(if mtu) - ip header - tcp header 1855 */ 1856 u_long 1857 tcp_mss_to_advertise(ifp, af) 1858 const struct ifnet *ifp; 1859 int af; 1860 { 1861 extern u_long in_maxmtu; 1862 u_long mss = 0; 1863 u_long hdrsiz; 1864 1865 /* 1866 * In order to avoid defeating path MTU discovery on the peer, 1867 * we advertise the max MTU of all attached networks as our MSS, 1868 * per RFC 1191, section 3.1. 1869 * 1870 * We provide the option to advertise just the MTU of 1871 * the interface on which we hope this connection will 1872 * be receiving. If we are responding to a SYN, we 1873 * will have a pretty good idea about this, but when 1874 * initiating a connection there is a bit more doubt. 1875 * 1876 * We also need to ensure that loopback has a large enough 1877 * MSS, as the loopback MTU is never included in in_maxmtu. 1878 */ 1879 1880 if (ifp != NULL) 1881 switch (af) { 1882 case AF_INET: 1883 mss = ifp->if_mtu; 1884 break; 1885 #ifdef INET6 1886 case AF_INET6: 1887 mss = IN6_LINKMTU(ifp); 1888 break; 1889 #endif 1890 } 1891 1892 if (tcp_mss_ifmtu == 0) 1893 switch (af) { 1894 case AF_INET: 1895 mss = max(in_maxmtu, mss); 1896 break; 1897 #ifdef INET6 1898 case AF_INET6: 1899 mss = max(in6_maxmtu, mss); 1900 break; 1901 #endif 1902 } 1903 1904 switch (af) { 1905 case AF_INET: 1906 hdrsiz = sizeof(struct ip); 1907 break; 1908 #ifdef INET6 1909 case AF_INET6: 1910 hdrsiz = sizeof(struct ip6_hdr); 1911 break; 1912 #endif 1913 default: 1914 hdrsiz = 0; 1915 break; 1916 } 1917 hdrsiz += sizeof(struct tcphdr); 1918 if (mss > hdrsiz) 1919 mss -= hdrsiz; 1920 1921 mss = max(tcp_mssdflt, mss); 1922 return (mss); 1923 } 1924 1925 /* 1926 * Set connection variables based on the peer's advertised MSS. 1927 * We are passed the TCPCB for the actual connection. If we 1928 * are the server, we are called by the compressed state engine 1929 * when the 3-way handshake is complete. If we are the client, 1930 * we are called when we receive the SYN,ACK from the server. 1931 * 1932 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1933 * before this routine is called! 1934 */ 1935 void 1936 tcp_mss_from_peer(tp, offer) 1937 struct tcpcb *tp; 1938 int offer; 1939 { 1940 struct socket *so; 1941 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1942 struct rtentry *rt; 1943 #endif 1944 u_long bufsize; 1945 int mss; 1946 1947 #ifdef DIAGNOSTIC 1948 if (tp->t_inpcb && tp->t_in6pcb) 1949 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1950 #endif 1951 so = NULL; 1952 rt = NULL; 1953 #ifdef INET 1954 if (tp->t_inpcb) { 1955 so = tp->t_inpcb->inp_socket; 1956 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1957 rt = in_pcbrtentry(tp->t_inpcb); 1958 #endif 1959 } 1960 #endif 1961 #ifdef INET6 1962 if (tp->t_in6pcb) { 1963 so = tp->t_in6pcb->in6p_socket; 1964 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1965 rt = in6_pcbrtentry(tp->t_in6pcb); 1966 #endif 1967 } 1968 #endif 1969 1970 /* 1971 * As per RFC1122, use the default MSS value, unless they 1972 * sent us an offer. Do not accept offers less than 256 bytes. 1973 */ 1974 mss = tcp_mssdflt; 1975 if (offer) 1976 mss = offer; 1977 mss = max(mss, 256); /* sanity */ 1978 tp->t_peermss = mss; 1979 mss -= tcp_optlen(tp); 1980 #ifdef INET 1981 if (tp->t_inpcb) 1982 mss -= ip_optlen(tp->t_inpcb); 1983 #endif 1984 #ifdef INET6 1985 if (tp->t_in6pcb) 1986 mss -= ip6_optlen(tp->t_in6pcb); 1987 #endif 1988 1989 /* 1990 * If there's a pipesize, change the socket buffer to that size. 1991 * Make the socket buffer an integral number of MSS units. If 1992 * the MSS is larger than the socket buffer, artificially decrease 1993 * the MSS. 1994 */ 1995 #ifdef RTV_SPIPE 1996 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1997 bufsize = rt->rt_rmx.rmx_sendpipe; 1998 else 1999 #endif 2000 bufsize = so->so_snd.sb_hiwat; 2001 if (bufsize < mss) 2002 mss = bufsize; 2003 else { 2004 bufsize = roundup(bufsize, mss); 2005 if (bufsize > sb_max) 2006 bufsize = sb_max; 2007 (void) sbreserve(&so->so_snd, bufsize, so); 2008 } 2009 tp->t_segsz = mss; 2010 2011 #ifdef RTV_SSTHRESH 2012 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 2013 /* 2014 * There's some sort of gateway or interface buffer 2015 * limit on the path. Use this to set the slow 2016 * start threshold, but set the threshold to no less 2017 * than 2 * MSS. 2018 */ 2019 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2020 } 2021 #endif 2022 } 2023 2024 /* 2025 * Processing necessary when a TCP connection is established. 2026 */ 2027 void 2028 tcp_established(tp) 2029 struct tcpcb *tp; 2030 { 2031 struct socket *so; 2032 #ifdef RTV_RPIPE 2033 struct rtentry *rt; 2034 #endif 2035 u_long bufsize; 2036 2037 #ifdef DIAGNOSTIC 2038 if (tp->t_inpcb && tp->t_in6pcb) 2039 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2040 #endif 2041 so = NULL; 2042 rt = NULL; 2043 #ifdef INET 2044 if (tp->t_inpcb) { 2045 so = tp->t_inpcb->inp_socket; 2046 #if defined(RTV_RPIPE) 2047 rt = in_pcbrtentry(tp->t_inpcb); 2048 #endif 2049 } 2050 #endif 2051 #ifdef INET6 2052 if (tp->t_in6pcb) { 2053 so = tp->t_in6pcb->in6p_socket; 2054 #if defined(RTV_RPIPE) 2055 rt = in6_pcbrtentry(tp->t_in6pcb); 2056 #endif 2057 } 2058 #endif 2059 2060 tp->t_state = TCPS_ESTABLISHED; 2061 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 2062 2063 #ifdef RTV_RPIPE 2064 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2065 bufsize = rt->rt_rmx.rmx_recvpipe; 2066 else 2067 #endif 2068 bufsize = so->so_rcv.sb_hiwat; 2069 if (bufsize > tp->t_ourmss) { 2070 bufsize = roundup(bufsize, tp->t_ourmss); 2071 if (bufsize > sb_max) 2072 bufsize = sb_max; 2073 (void) sbreserve(&so->so_rcv, bufsize, so); 2074 } 2075 } 2076 2077 /* 2078 * Check if there's an initial rtt or rttvar. Convert from the 2079 * route-table units to scaled multiples of the slow timeout timer. 2080 * Called only during the 3-way handshake. 2081 */ 2082 void 2083 tcp_rmx_rtt(tp) 2084 struct tcpcb *tp; 2085 { 2086 #ifdef RTV_RTT 2087 struct rtentry *rt = NULL; 2088 int rtt; 2089 2090 #ifdef DIAGNOSTIC 2091 if (tp->t_inpcb && tp->t_in6pcb) 2092 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2093 #endif 2094 #ifdef INET 2095 if (tp->t_inpcb) 2096 rt = in_pcbrtentry(tp->t_inpcb); 2097 #endif 2098 #ifdef INET6 2099 if (tp->t_in6pcb) 2100 rt = in6_pcbrtentry(tp->t_in6pcb); 2101 #endif 2102 if (rt == NULL) 2103 return; 2104 2105 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2106 /* 2107 * XXX The lock bit for MTU indicates that the value 2108 * is also a minimum value; this is subject to time. 2109 */ 2110 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2111 TCPT_RANGESET(tp->t_rttmin, 2112 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2113 TCPTV_MIN, TCPTV_REXMTMAX); 2114 tp->t_srtt = rtt / 2115 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2116 if (rt->rt_rmx.rmx_rttvar) { 2117 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2118 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2119 (TCP_RTTVAR_SHIFT + 2)); 2120 } else { 2121 /* Default variation is +- 1 rtt */ 2122 tp->t_rttvar = 2123 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2124 } 2125 TCPT_RANGESET(tp->t_rxtcur, 2126 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2127 tp->t_rttmin, TCPTV_REXMTMAX); 2128 } 2129 #endif 2130 } 2131 2132 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2133 #if NRND > 0 2134 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2135 #endif 2136 2137 /* 2138 * Get a new sequence value given a tcp control block 2139 */ 2140 tcp_seq 2141 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2142 { 2143 2144 #ifdef INET 2145 if (tp->t_inpcb != NULL) { 2146 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2147 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2148 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2149 addin)); 2150 } 2151 #endif 2152 #ifdef INET6 2153 if (tp->t_in6pcb != NULL) { 2154 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2155 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2156 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2157 addin)); 2158 } 2159 #endif 2160 /* Not possible. */ 2161 panic("tcp_new_iss"); 2162 } 2163 2164 /* 2165 * This routine actually generates a new TCP initial sequence number. 2166 */ 2167 tcp_seq 2168 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2169 size_t addrsz, tcp_seq addin) 2170 { 2171 tcp_seq tcp_iss; 2172 2173 #if NRND > 0 2174 static int beenhere; 2175 2176 /* 2177 * If we haven't been here before, initialize our cryptographic 2178 * hash secret. 2179 */ 2180 if (beenhere == 0) { 2181 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2182 RND_EXTRACT_ANY); 2183 beenhere = 1; 2184 } 2185 2186 if (tcp_do_rfc1948) { 2187 MD5_CTX ctx; 2188 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2189 2190 /* 2191 * Compute the base value of the ISS. It is a hash 2192 * of (saddr, sport, daddr, dport, secret). 2193 */ 2194 MD5Init(&ctx); 2195 2196 MD5Update(&ctx, (u_char *) laddr, addrsz); 2197 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2198 2199 MD5Update(&ctx, (u_char *) faddr, addrsz); 2200 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2201 2202 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2203 2204 MD5Final(hash, &ctx); 2205 2206 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2207 2208 /* 2209 * Now increment our "timer", and add it in to 2210 * the computed value. 2211 * 2212 * XXX Use `addin'? 2213 * XXX TCP_ISSINCR too large to use? 2214 */ 2215 tcp_iss_seq += TCP_ISSINCR; 2216 #ifdef TCPISS_DEBUG 2217 printf("ISS hash 0x%08x, ", tcp_iss); 2218 #endif 2219 tcp_iss += tcp_iss_seq + addin; 2220 #ifdef TCPISS_DEBUG 2221 printf("new ISS 0x%08x\n", tcp_iss); 2222 #endif 2223 } else 2224 #endif /* NRND > 0 */ 2225 { 2226 /* 2227 * Randomize. 2228 */ 2229 #if NRND > 0 2230 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2231 #else 2232 tcp_iss = arc4random(); 2233 #endif 2234 2235 /* 2236 * If we were asked to add some amount to a known value, 2237 * we will take a random value obtained above, mask off 2238 * the upper bits, and add in the known value. We also 2239 * add in a constant to ensure that we are at least a 2240 * certain distance from the original value. 2241 * 2242 * This is used when an old connection is in timed wait 2243 * and we have a new one coming in, for instance. 2244 */ 2245 if (addin != 0) { 2246 #ifdef TCPISS_DEBUG 2247 printf("Random %08x, ", tcp_iss); 2248 #endif 2249 tcp_iss &= TCP_ISS_RANDOM_MASK; 2250 tcp_iss += addin + TCP_ISSINCR; 2251 #ifdef TCPISS_DEBUG 2252 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2253 #endif 2254 } else { 2255 tcp_iss &= TCP_ISS_RANDOM_MASK; 2256 tcp_iss += tcp_iss_seq; 2257 tcp_iss_seq += TCP_ISSINCR; 2258 #ifdef TCPISS_DEBUG 2259 printf("ISS %08x\n", tcp_iss); 2260 #endif 2261 } 2262 } 2263 2264 if (tcp_compat_42) { 2265 /* 2266 * Limit it to the positive range for really old TCP 2267 * implementations. 2268 * Just AND off the top bit instead of checking if 2269 * is set first - saves a branch 50% of the time. 2270 */ 2271 tcp_iss &= 0x7fffffff; /* XXX */ 2272 } 2273 2274 return (tcp_iss); 2275 } 2276 2277 #if defined(IPSEC) || defined(FAST_IPSEC) 2278 /* compute ESP/AH header size for TCP, including outer IP header. */ 2279 size_t 2280 ipsec4_hdrsiz_tcp(tp) 2281 struct tcpcb *tp; 2282 { 2283 struct inpcb *inp; 2284 size_t hdrsiz; 2285 2286 /* XXX mapped addr case (tp->t_in6pcb) */ 2287 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2288 return 0; 2289 switch (tp->t_family) { 2290 case AF_INET: 2291 /* XXX: should use currect direction. */ 2292 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2293 break; 2294 default: 2295 hdrsiz = 0; 2296 break; 2297 } 2298 2299 return hdrsiz; 2300 } 2301 2302 #ifdef INET6 2303 size_t 2304 ipsec6_hdrsiz_tcp(tp) 2305 struct tcpcb *tp; 2306 { 2307 struct in6pcb *in6p; 2308 size_t hdrsiz; 2309 2310 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2311 return 0; 2312 switch (tp->t_family) { 2313 case AF_INET6: 2314 /* XXX: should use currect direction. */ 2315 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2316 break; 2317 case AF_INET: 2318 /* mapped address case - tricky */ 2319 default: 2320 hdrsiz = 0; 2321 break; 2322 } 2323 2324 return hdrsiz; 2325 } 2326 #endif 2327 #endif /*IPSEC*/ 2328 2329 /* 2330 * Determine the length of the TCP options for this connection. 2331 * 2332 * XXX: What do we do for SACK, when we add that? Just reserve 2333 * all of the space? Otherwise we can't exactly be incrementing 2334 * cwnd by an amount that varies depending on the amount we last 2335 * had to SACK! 2336 */ 2337 2338 u_int 2339 tcp_optlen(tp) 2340 struct tcpcb *tp; 2341 { 2342 u_int optlen; 2343 2344 optlen = 0; 2345 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2346 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2347 optlen += TCPOLEN_TSTAMP_APPA; 2348 2349 #ifdef TCP_SIGNATURE 2350 #ifdef INET6 2351 /* XXX: FreeBSD uses ((tp->t_inpcb->inp_vflag & INP_IPV6) == 0) */ 2352 if (tp->t_family == AF_INET) 2353 #endif 2354 if (tp->t_flags & TF_SIGNATURE) 2355 optlen += TCPOLEN_SIGNATURE + 2; 2356 #endif /* TCP_SIGNATURE */ 2357 2358 return optlen; 2359 } 2360
Indexes created Mon Sep 22 05:09:51 GMT 2025