tcp_subr.c revision 1.148
1 /* $NetBSD: tcp_subr.c,v 1.148 2003/08/22 21:53:06 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 1.148 2003/08/22 21:53:06 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 #endif /*IPSEC*/ 159 160 #ifdef FAST_IPSEC 161 #include <netipsec/ipsec.h> 162 #ifdef INET6 163 #include <netipsec/ipsec6.h> 164 #endif 165 #endif /* FAST_IPSEC*/ 166 167 168 #ifdef INET6 169 struct in6pcb tcb6; 170 #endif 171 172 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 173 struct tcpstat tcpstat; /* tcp statistics */ 174 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 175 176 /* patchable/settable parameters for tcp */ 177 int tcp_mssdflt = TCP_MSS; 178 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 179 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 180 #if NRND > 0 181 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 182 #endif 183 int tcp_do_sack = 1; /* selective acknowledgement */ 184 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 185 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 186 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 187 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 188 #ifndef TCP_INIT_WIN 189 #define TCP_INIT_WIN 1 /* initial slow start window */ 190 #endif 191 #ifndef TCP_INIT_WIN_LOCAL 192 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 193 #endif 194 int tcp_init_win = TCP_INIT_WIN; 195 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 196 int tcp_mss_ifmtu = 0; 197 #ifdef TCP_COMPAT_42 198 int tcp_compat_42 = 1; 199 #else 200 int tcp_compat_42 = 0; 201 #endif 202 int tcp_rst_ppslim = 100; /* 100pps */ 203 204 /* tcb hash */ 205 #ifndef TCBHASHSIZE 206 #define TCBHASHSIZE 128 207 #endif 208 int tcbhashsize = TCBHASHSIZE; 209 210 /* syn hash parameters */ 211 #define TCP_SYN_HASH_SIZE 293 212 #define TCP_SYN_BUCKET_SIZE 35 213 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 214 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 215 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 216 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 217 218 int tcp_freeq __P((struct tcpcb *)); 219 220 #ifdef INET 221 void tcp_mtudisc_callback __P((struct in_addr)); 222 #endif 223 #ifdef INET6 224 void tcp6_mtudisc_callback __P((struct in6_addr *)); 225 #endif 226 227 void tcp_mtudisc __P((struct inpcb *, int)); 228 #ifdef INET6 229 void tcp6_mtudisc __P((struct in6pcb *, int)); 230 #endif 231 232 struct pool tcpcb_pool; 233 234 #ifdef TCP_CSUM_COUNTERS 235 #include <sys/device.h> 236 237 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 238 NULL, "tcp", "hwcsum bad"); 239 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 240 NULL, "tcp", "hwcsum ok"); 241 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "hwcsum data"); 243 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "swcsum"); 245 #endif /* TCP_CSUM_COUNTERS */ 246 247 #ifdef TCP_OUTPUT_COUNTERS 248 #include <sys/device.h> 249 250 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp", "output big header"); 252 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 NULL, "tcp", "output copy small"); 254 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 255 NULL, "tcp", "output copy big"); 256 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 257 NULL, "tcp", "output reference big"); 258 #endif /* TCP_OUTPUT_COUNTERS */ 259 260 #ifdef TCP_REASS_COUNTERS 261 #include <sys/device.h> 262 263 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp_reass", "calls"); 265 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 &tcp_reass_, "tcp_reass", "insert into empty queue"); 267 struct evcnt tcp_reass_iteration[8] = { 268 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 269 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 270 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 271 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 272 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 273 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 274 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 275 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 276 }; 277 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 278 &tcp_reass_, "tcp_reass", "prepend to first"); 279 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 &tcp_reass_, "tcp_reass", "prepend"); 281 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 &tcp_reass_, "tcp_reass", "insert"); 283 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 &tcp_reass_, "tcp_reass", "insert at tail"); 285 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 &tcp_reass_, "tcp_reass", "append"); 287 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 &tcp_reass_, "tcp_reass", "append to tail fragment"); 289 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 &tcp_reass_, "tcp_reass", "overlap at end"); 291 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 &tcp_reass_, "tcp_reass", "overlap at start"); 293 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 294 &tcp_reass_, "tcp_reass", "duplicate segment"); 295 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 296 &tcp_reass_, "tcp_reass", "duplicate fragment"); 297 298 #endif /* TCP_REASS_COUNTERS */ 299 300 #ifdef MBUFTRACE 301 struct mowner tcp_mowner = { "tcp" }; 302 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 303 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 304 #endif 305 306 /* 307 * Tcp initialization 308 */ 309 void 310 tcp_init() 311 { 312 int hlen; 313 314 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 315 NULL); 316 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 317 #ifdef INET6 318 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 319 #endif 320 321 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 322 #ifdef INET6 323 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 324 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 325 #endif 326 if (max_protohdr < hlen) 327 max_protohdr = hlen; 328 if (max_linkhdr + hlen > MHLEN) 329 panic("tcp_init"); 330 331 #ifdef INET 332 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 333 #endif 334 #ifdef INET6 335 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 336 #endif 337 338 /* Initialize timer state. */ 339 tcp_timer_init(); 340 341 /* Initialize the compressed state engine. */ 342 syn_cache_init(); 343 344 #ifdef TCP_CSUM_COUNTERS 345 evcnt_attach_static(&tcp_hwcsum_bad); 346 evcnt_attach_static(&tcp_hwcsum_ok); 347 evcnt_attach_static(&tcp_hwcsum_data); 348 evcnt_attach_static(&tcp_swcsum); 349 #endif /* TCP_CSUM_COUNTERS */ 350 351 #ifdef TCP_OUTPUT_COUNTERS 352 evcnt_attach_static(&tcp_output_bigheader); 353 evcnt_attach_static(&tcp_output_copysmall); 354 evcnt_attach_static(&tcp_output_copybig); 355 evcnt_attach_static(&tcp_output_refbig); 356 #endif /* TCP_OUTPUT_COUNTERS */ 357 358 #ifdef TCP_REASS_COUNTERS 359 evcnt_attach_static(&tcp_reass_); 360 evcnt_attach_static(&tcp_reass_empty); 361 evcnt_attach_static(&tcp_reass_iteration[0]); 362 evcnt_attach_static(&tcp_reass_iteration[1]); 363 evcnt_attach_static(&tcp_reass_iteration[2]); 364 evcnt_attach_static(&tcp_reass_iteration[3]); 365 evcnt_attach_static(&tcp_reass_iteration[4]); 366 evcnt_attach_static(&tcp_reass_iteration[5]); 367 evcnt_attach_static(&tcp_reass_iteration[6]); 368 evcnt_attach_static(&tcp_reass_iteration[7]); 369 evcnt_attach_static(&tcp_reass_prependfirst); 370 evcnt_attach_static(&tcp_reass_prepend); 371 evcnt_attach_static(&tcp_reass_insert); 372 evcnt_attach_static(&tcp_reass_inserttail); 373 evcnt_attach_static(&tcp_reass_append); 374 evcnt_attach_static(&tcp_reass_appendtail); 375 evcnt_attach_static(&tcp_reass_overlaptail); 376 evcnt_attach_static(&tcp_reass_overlapfront); 377 evcnt_attach_static(&tcp_reass_segdup); 378 evcnt_attach_static(&tcp_reass_fragdup); 379 #endif /* TCP_REASS_COUNTERS */ 380 381 MOWNER_ATTACH(&tcp_tx_mowner); 382 MOWNER_ATTACH(&tcp_rx_mowner); 383 MOWNER_ATTACH(&tcp_mowner); 384 } 385 386 /* 387 * Create template to be used to send tcp packets on a connection. 388 * Call after host entry created, allocates an mbuf and fills 389 * in a skeletal tcp/ip header, minimizing the amount of work 390 * necessary when the connection is used. 391 */ 392 struct mbuf * 393 tcp_template(tp) 394 struct tcpcb *tp; 395 { 396 struct inpcb *inp = tp->t_inpcb; 397 #ifdef INET6 398 struct in6pcb *in6p = tp->t_in6pcb; 399 #endif 400 struct tcphdr *n; 401 struct mbuf *m; 402 int hlen; 403 404 switch (tp->t_family) { 405 case AF_INET: 406 hlen = sizeof(struct ip); 407 if (inp) 408 break; 409 #ifdef INET6 410 if (in6p) { 411 /* mapped addr case */ 412 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 413 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 414 break; 415 } 416 #endif 417 return NULL; /*EINVAL*/ 418 #ifdef INET6 419 case AF_INET6: 420 hlen = sizeof(struct ip6_hdr); 421 if (in6p) { 422 /* more sainty check? */ 423 break; 424 } 425 return NULL; /*EINVAL*/ 426 #endif 427 default: 428 hlen = 0; /*pacify gcc*/ 429 return NULL; /*EAFNOSUPPORT*/ 430 } 431 #ifdef DIAGNOSTIC 432 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 433 panic("mclbytes too small for t_template"); 434 #endif 435 m = tp->t_template; 436 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 437 ; 438 else { 439 if (m) 440 m_freem(m); 441 m = tp->t_template = NULL; 442 MGETHDR(m, M_DONTWAIT, MT_HEADER); 443 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 444 MCLGET(m, M_DONTWAIT); 445 if ((m->m_flags & M_EXT) == 0) { 446 m_free(m); 447 m = NULL; 448 } 449 } 450 if (m == NULL) 451 return NULL; 452 MCLAIM(m, &tcp_mowner); 453 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 454 } 455 456 bzero(mtod(m, caddr_t), m->m_len); 457 458 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 459 460 switch (tp->t_family) { 461 case AF_INET: 462 { 463 struct ipovly *ipov; 464 mtod(m, struct ip *)->ip_v = 4; 465 ipov = mtod(m, struct ipovly *); 466 ipov->ih_pr = IPPROTO_TCP; 467 ipov->ih_len = htons(sizeof(struct tcphdr)); 468 if (inp) { 469 ipov->ih_src = inp->inp_laddr; 470 ipov->ih_dst = inp->inp_faddr; 471 } 472 #ifdef INET6 473 else if (in6p) { 474 /* mapped addr case */ 475 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 476 sizeof(ipov->ih_src)); 477 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 478 sizeof(ipov->ih_dst)); 479 } 480 #endif 481 /* 482 * Compute the pseudo-header portion of the checksum 483 * now. We incrementally add in the TCP option and 484 * payload lengths later, and then compute the TCP 485 * checksum right before the packet is sent off onto 486 * the wire. 487 */ 488 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 489 ipov->ih_dst.s_addr, 490 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 491 break; 492 } 493 #ifdef INET6 494 case AF_INET6: 495 { 496 struct ip6_hdr *ip6; 497 mtod(m, struct ip *)->ip_v = 6; 498 ip6 = mtod(m, struct ip6_hdr *); 499 ip6->ip6_nxt = IPPROTO_TCP; 500 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 501 ip6->ip6_src = in6p->in6p_laddr; 502 ip6->ip6_dst = in6p->in6p_faddr; 503 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 504 if (ip6_auto_flowlabel) { 505 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 506 ip6->ip6_flow |= 507 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 508 } 509 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 510 ip6->ip6_vfc |= IPV6_VERSION; 511 512 /* 513 * Compute the pseudo-header portion of the checksum 514 * now. We incrementally add in the TCP option and 515 * payload lengths later, and then compute the TCP 516 * checksum right before the packet is sent off onto 517 * the wire. 518 */ 519 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 520 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 521 htonl(IPPROTO_TCP)); 522 break; 523 } 524 #endif 525 } 526 if (inp) { 527 n->th_sport = inp->inp_lport; 528 n->th_dport = inp->inp_fport; 529 } 530 #ifdef INET6 531 else if (in6p) { 532 n->th_sport = in6p->in6p_lport; 533 n->th_dport = in6p->in6p_fport; 534 } 535 #endif 536 n->th_seq = 0; 537 n->th_ack = 0; 538 n->th_x2 = 0; 539 n->th_off = 5; 540 n->th_flags = 0; 541 n->th_win = 0; 542 n->th_urp = 0; 543 return (m); 544 } 545 546 /* 547 * Send a single message to the TCP at address specified by 548 * the given TCP/IP header. If m == 0, then we make a copy 549 * of the tcpiphdr at ti and send directly to the addressed host. 550 * This is used to force keep alive messages out using the TCP 551 * template for a connection tp->t_template. If flags are given 552 * then we send a message back to the TCP which originated the 553 * segment ti, and discard the mbuf containing it and any other 554 * attached mbufs. 555 * 556 * In any case the ack and sequence number of the transmitted 557 * segment are as specified by the parameters. 558 */ 559 int 560 tcp_respond(tp, template, m, th0, ack, seq, flags) 561 struct tcpcb *tp; 562 struct mbuf *template; 563 struct mbuf *m; 564 struct tcphdr *th0; 565 tcp_seq ack, seq; 566 int flags; 567 { 568 struct route *ro; 569 int error, tlen, win = 0; 570 int hlen; 571 struct ip *ip; 572 #ifdef INET6 573 struct ip6_hdr *ip6; 574 #endif 575 int family; /* family on packet, not inpcb/in6pcb! */ 576 struct tcphdr *th; 577 struct socket *so; 578 579 if (tp != NULL && (flags & TH_RST) == 0) { 580 #ifdef DIAGNOSTIC 581 if (tp->t_inpcb && tp->t_in6pcb) 582 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 583 #endif 584 #ifdef INET 585 if (tp->t_inpcb) 586 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 587 #endif 588 #ifdef INET6 589 if (tp->t_in6pcb) 590 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 591 #endif 592 } 593 594 th = NULL; /* Quell uninitialized warning */ 595 ip = NULL; 596 #ifdef INET6 597 ip6 = NULL; 598 #endif 599 if (m == 0) { 600 if (!template) 601 return EINVAL; 602 603 /* get family information from template */ 604 switch (mtod(template, struct ip *)->ip_v) { 605 case 4: 606 family = AF_INET; 607 hlen = sizeof(struct ip); 608 break; 609 #ifdef INET6 610 case 6: 611 family = AF_INET6; 612 hlen = sizeof(struct ip6_hdr); 613 break; 614 #endif 615 default: 616 return EAFNOSUPPORT; 617 } 618 619 MGETHDR(m, M_DONTWAIT, MT_HEADER); 620 if (m) { 621 MCLAIM(m, &tcp_tx_mowner); 622 MCLGET(m, M_DONTWAIT); 623 if ((m->m_flags & M_EXT) == 0) { 624 m_free(m); 625 m = NULL; 626 } 627 } 628 if (m == NULL) 629 return (ENOBUFS); 630 631 if (tcp_compat_42) 632 tlen = 1; 633 else 634 tlen = 0; 635 636 m->m_data += max_linkhdr; 637 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 638 template->m_len); 639 switch (family) { 640 case AF_INET: 641 ip = mtod(m, struct ip *); 642 th = (struct tcphdr *)(ip + 1); 643 break; 644 #ifdef INET6 645 case AF_INET6: 646 ip6 = mtod(m, struct ip6_hdr *); 647 th = (struct tcphdr *)(ip6 + 1); 648 break; 649 #endif 650 #if 0 651 default: 652 /* noone will visit here */ 653 m_freem(m); 654 return EAFNOSUPPORT; 655 #endif 656 } 657 flags = TH_ACK; 658 } else { 659 660 if ((m->m_flags & M_PKTHDR) == 0) { 661 #if 0 662 printf("non PKTHDR to tcp_respond\n"); 663 #endif 664 m_freem(m); 665 return EINVAL; 666 } 667 #ifdef DIAGNOSTIC 668 if (!th0) 669 panic("th0 == NULL in tcp_respond"); 670 #endif 671 672 /* get family information from m */ 673 switch (mtod(m, struct ip *)->ip_v) { 674 case 4: 675 family = AF_INET; 676 hlen = sizeof(struct ip); 677 ip = mtod(m, struct ip *); 678 break; 679 #ifdef INET6 680 case 6: 681 family = AF_INET6; 682 hlen = sizeof(struct ip6_hdr); 683 ip6 = mtod(m, struct ip6_hdr *); 684 break; 685 #endif 686 default: 687 m_freem(m); 688 return EAFNOSUPPORT; 689 } 690 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 691 tlen = sizeof(*th0); 692 else 693 tlen = th0->th_off << 2; 694 695 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 696 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 697 m->m_len = hlen + tlen; 698 m_freem(m->m_next); 699 m->m_next = NULL; 700 } else { 701 struct mbuf *n; 702 703 #ifdef DIAGNOSTIC 704 if (max_linkhdr + hlen + tlen > MCLBYTES) { 705 m_freem(m); 706 return EMSGSIZE; 707 } 708 #endif 709 MGETHDR(n, M_DONTWAIT, MT_HEADER); 710 if (n && max_linkhdr + hlen + tlen > MHLEN) { 711 MCLGET(n, M_DONTWAIT); 712 if ((n->m_flags & M_EXT) == 0) { 713 m_freem(n); 714 n = NULL; 715 } 716 } 717 if (!n) { 718 m_freem(m); 719 return ENOBUFS; 720 } 721 722 MCLAIM(n, &tcp_tx_mowner); 723 n->m_data += max_linkhdr; 724 n->m_len = hlen + tlen; 725 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 726 m_copyback(n, hlen, tlen, (caddr_t)th0); 727 728 m_freem(m); 729 m = n; 730 n = NULL; 731 } 732 733 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 734 switch (family) { 735 case AF_INET: 736 ip = mtod(m, struct ip *); 737 th = (struct tcphdr *)(ip + 1); 738 ip->ip_p = IPPROTO_TCP; 739 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 740 ip->ip_p = IPPROTO_TCP; 741 break; 742 #ifdef INET6 743 case AF_INET6: 744 ip6 = mtod(m, struct ip6_hdr *); 745 th = (struct tcphdr *)(ip6 + 1); 746 ip6->ip6_nxt = IPPROTO_TCP; 747 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 748 ip6->ip6_nxt = IPPROTO_TCP; 749 break; 750 #endif 751 #if 0 752 default: 753 /* noone will visit here */ 754 m_freem(m); 755 return EAFNOSUPPORT; 756 #endif 757 } 758 xchg(th->th_dport, th->th_sport, u_int16_t); 759 #undef xchg 760 tlen = 0; /*be friendly with the following code*/ 761 } 762 th->th_seq = htonl(seq); 763 th->th_ack = htonl(ack); 764 th->th_x2 = 0; 765 if ((flags & TH_SYN) == 0) { 766 if (tp) 767 win >>= tp->rcv_scale; 768 if (win > TCP_MAXWIN) 769 win = TCP_MAXWIN; 770 th->th_win = htons((u_int16_t)win); 771 th->th_off = sizeof (struct tcphdr) >> 2; 772 tlen += sizeof(*th); 773 } else 774 tlen += th->th_off << 2; 775 m->m_len = hlen + tlen; 776 m->m_pkthdr.len = hlen + tlen; 777 m->m_pkthdr.rcvif = (struct ifnet *) 0; 778 th->th_flags = flags; 779 th->th_urp = 0; 780 781 switch (family) { 782 #ifdef INET 783 case AF_INET: 784 { 785 struct ipovly *ipov = (struct ipovly *)ip; 786 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 787 ipov->ih_len = htons((u_int16_t)tlen); 788 789 th->th_sum = 0; 790 th->th_sum = in_cksum(m, hlen + tlen); 791 ip->ip_len = htons(hlen + tlen); 792 ip->ip_ttl = ip_defttl; 793 break; 794 } 795 #endif 796 #ifdef INET6 797 case AF_INET6: 798 { 799 th->th_sum = 0; 800 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 801 tlen); 802 ip6->ip6_plen = ntohs(tlen); 803 if (tp && tp->t_in6pcb) { 804 struct ifnet *oifp; 805 ro = (struct route *)&tp->t_in6pcb->in6p_route; 806 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 807 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 808 } else 809 ip6->ip6_hlim = ip6_defhlim; 810 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 811 if (ip6_auto_flowlabel) { 812 ip6->ip6_flow |= 813 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 814 } 815 break; 816 } 817 #endif 818 } 819 820 if (tp->t_inpcb) 821 so = tp->t_inpcb->inp_socket; 822 #ifdef INET6 823 else if (tp->t_in6pcb) 824 so = tp->t_in6pcb->in6p_socket; 825 #endif 826 else 827 so = NULL; 828 #ifdef IPSEC 829 if (ipsec_setsocket(m, so) != 0) { 830 m_freem(m); 831 return ENOBUFS; 832 } 833 #endif /*IPSEC*/ 834 835 if (tp != NULL && tp->t_inpcb != NULL) { 836 ro = &tp->t_inpcb->inp_route; 837 #ifdef DIAGNOSTIC 838 if (family != AF_INET) 839 panic("tcp_respond: address family mismatch"); 840 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 841 panic("tcp_respond: ip_dst %x != inp_faddr %x", 842 ntohl(ip->ip_dst.s_addr), 843 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 844 } 845 #endif 846 } 847 #ifdef INET6 848 else if (tp != NULL && tp->t_in6pcb != NULL) { 849 ro = (struct route *)&tp->t_in6pcb->in6p_route; 850 #ifdef DIAGNOSTIC 851 if (family == AF_INET) { 852 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 853 panic("tcp_respond: not mapped addr"); 854 if (bcmp(&ip->ip_dst, 855 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 856 sizeof(ip->ip_dst)) != 0) { 857 panic("tcp_respond: ip_dst != in6p_faddr"); 858 } 859 } else if (family == AF_INET6) { 860 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 861 &tp->t_in6pcb->in6p_faddr)) 862 panic("tcp_respond: ip6_dst != in6p_faddr"); 863 } else 864 panic("tcp_respond: address family mismatch"); 865 #endif 866 } 867 #endif 868 else 869 ro = NULL; 870 871 switch (family) { 872 #ifdef INET 873 case AF_INET: 874 error = ip_output(m, NULL, ro, 875 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 876 (struct ip_moptions *)0, so); 877 break; 878 #endif 879 #ifdef INET6 880 case AF_INET6: 881 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 882 (struct ip6_moptions *)0, so, NULL); 883 break; 884 #endif 885 default: 886 error = EAFNOSUPPORT; 887 break; 888 } 889 890 return (error); 891 } 892 893 /* 894 * Create a new TCP control block, making an 895 * empty reassembly queue and hooking it to the argument 896 * protocol control block. 897 */ 898 struct tcpcb * 899 tcp_newtcpcb(family, aux) 900 int family; /* selects inpcb, or in6pcb */ 901 void *aux; 902 { 903 struct tcpcb *tp; 904 int i; 905 906 switch (family) { 907 case PF_INET: 908 break; 909 #ifdef INET6 910 case PF_INET6: 911 break; 912 #endif 913 default: 914 return NULL; 915 } 916 917 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 918 if (tp == NULL) 919 return (NULL); 920 bzero((caddr_t)tp, sizeof(struct tcpcb)); 921 TAILQ_INIT(&tp->segq); 922 TAILQ_INIT(&tp->timeq); 923 tp->t_family = family; /* may be overridden later on */ 924 tp->t_peermss = tcp_mssdflt; 925 tp->t_ourmss = tcp_mssdflt; 926 tp->t_segsz = tcp_mssdflt; 927 LIST_INIT(&tp->t_sc); 928 929 tp->t_lastm = NULL; 930 tp->t_lastoff = 0; 931 932 callout_init(&tp->t_delack_ch); 933 for (i = 0; i < TCPT_NTIMERS; i++) 934 TCP_TIMER_INIT(tp, i); 935 936 tp->t_flags = 0; 937 if (tcp_do_rfc1323 && tcp_do_win_scale) 938 tp->t_flags |= TF_REQ_SCALE; 939 if (tcp_do_rfc1323 && tcp_do_timestamps) 940 tp->t_flags |= TF_REQ_TSTMP; 941 if (tcp_do_sack == 2) 942 tp->t_flags |= TF_WILL_SACK; 943 else if (tcp_do_sack == 1) 944 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 945 tp->t_flags |= TF_CANT_TXSACK; 946 switch (family) { 947 case PF_INET: 948 tp->t_inpcb = (struct inpcb *)aux; 949 tp->t_mtudisc = ip_mtudisc; 950 break; 951 #ifdef INET6 952 case PF_INET6: 953 tp->t_in6pcb = (struct in6pcb *)aux; 954 /* for IPv6, always try to run path MTU discovery */ 955 tp->t_mtudisc = 1; 956 break; 957 #endif 958 } 959 /* 960 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 961 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 962 * reasonable initial retransmit time. 963 */ 964 tp->t_srtt = TCPTV_SRTTBASE; 965 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 966 tp->t_rttmin = TCPTV_MIN; 967 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 968 TCPTV_MIN, TCPTV_REXMTMAX); 969 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 970 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 971 if (family == AF_INET) { 972 struct inpcb *inp = (struct inpcb *)aux; 973 inp->inp_ip.ip_ttl = ip_defttl; 974 inp->inp_ppcb = (caddr_t)tp; 975 } 976 #ifdef INET6 977 else if (family == AF_INET6) { 978 struct in6pcb *in6p = (struct in6pcb *)aux; 979 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 980 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 981 : NULL); 982 in6p->in6p_ppcb = (caddr_t)tp; 983 } 984 #endif 985 986 /* 987 * Initialize our timebase. When we send timestamps, we take 988 * the delta from tcp_now -- this means each connection always 989 * gets a timebase of 0, which makes it, among other things, 990 * more difficult to determine how long a system has been up, 991 * and thus how many TCP sequence increments have occurred. 992 */ 993 tp->ts_timebase = tcp_now; 994 995 return (tp); 996 } 997 998 /* 999 * Drop a TCP connection, reporting 1000 * the specified error. If connection is synchronized, 1001 * then send a RST to peer. 1002 */ 1003 struct tcpcb * 1004 tcp_drop(tp, errno) 1005 struct tcpcb *tp; 1006 int errno; 1007 { 1008 struct socket *so = NULL; 1009 1010 #ifdef DIAGNOSTIC 1011 if (tp->t_inpcb && tp->t_in6pcb) 1012 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1013 #endif 1014 #ifdef INET 1015 if (tp->t_inpcb) 1016 so = tp->t_inpcb->inp_socket; 1017 #endif 1018 #ifdef INET6 1019 if (tp->t_in6pcb) 1020 so = tp->t_in6pcb->in6p_socket; 1021 #endif 1022 if (!so) 1023 return NULL; 1024 1025 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1026 tp->t_state = TCPS_CLOSED; 1027 (void) tcp_output(tp); 1028 tcpstat.tcps_drops++; 1029 } else 1030 tcpstat.tcps_conndrops++; 1031 if (errno == ETIMEDOUT && tp->t_softerror) 1032 errno = tp->t_softerror; 1033 so->so_error = errno; 1034 return (tcp_close(tp)); 1035 } 1036 1037 /* 1038 * Return whether this tcpcb is marked as dead, indicating 1039 * to the calling timer function that no further action should 1040 * be taken, as we are about to release this tcpcb. The release 1041 * of the storage will be done if this is the last timer running. 1042 * 1043 * This is typically called from the callout handler function before 1044 * callout_ack() is done, therefore we need to test the number of 1045 * running timer functions against 1 below, not 0. 1046 */ 1047 int 1048 tcp_isdead(tp) 1049 struct tcpcb *tp; 1050 { 1051 int dead = (tp->t_flags & TF_DEAD); 1052 1053 if (__predict_false(dead)) { 1054 if (tcp_timers_invoking(tp) > 1) 1055 /* not quite there yet -- count separately? */ 1056 return dead; 1057 tcpstat.tcps_delayed_free++; 1058 pool_put(&tcpcb_pool, tp); 1059 } 1060 return dead; 1061 } 1062 1063 /* 1064 * Close a TCP control block: 1065 * discard all space held by the tcp 1066 * discard internet protocol block 1067 * wake up any sleepers 1068 */ 1069 struct tcpcb * 1070 tcp_close(tp) 1071 struct tcpcb *tp; 1072 { 1073 struct inpcb *inp; 1074 #ifdef INET6 1075 struct in6pcb *in6p; 1076 #endif 1077 struct socket *so; 1078 #ifdef RTV_RTT 1079 struct rtentry *rt; 1080 #endif 1081 struct route *ro; 1082 1083 inp = tp->t_inpcb; 1084 #ifdef INET6 1085 in6p = tp->t_in6pcb; 1086 #endif 1087 so = NULL; 1088 ro = NULL; 1089 if (inp) { 1090 so = inp->inp_socket; 1091 ro = &inp->inp_route; 1092 } 1093 #ifdef INET6 1094 else if (in6p) { 1095 so = in6p->in6p_socket; 1096 ro = (struct route *)&in6p->in6p_route; 1097 } 1098 #endif 1099 1100 #ifdef RTV_RTT 1101 /* 1102 * If we sent enough data to get some meaningful characteristics, 1103 * save them in the routing entry. 'Enough' is arbitrarily 1104 * defined as the sendpipesize (default 4K) * 16. This would 1105 * give us 16 rtt samples assuming we only get one sample per 1106 * window (the usual case on a long haul net). 16 samples is 1107 * enough for the srtt filter to converge to within 5% of the correct 1108 * value; fewer samples and we could save a very bogus rtt. 1109 * 1110 * Don't update the default route's characteristics and don't 1111 * update anything that the user "locked". 1112 */ 1113 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1114 ro && (rt = ro->ro_rt) && 1115 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1116 u_long i = 0; 1117 1118 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1119 i = tp->t_srtt * 1120 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1121 if (rt->rt_rmx.rmx_rtt && i) 1122 /* 1123 * filter this update to half the old & half 1124 * the new values, converting scale. 1125 * See route.h and tcp_var.h for a 1126 * description of the scaling constants. 1127 */ 1128 rt->rt_rmx.rmx_rtt = 1129 (rt->rt_rmx.rmx_rtt + i) / 2; 1130 else 1131 rt->rt_rmx.rmx_rtt = i; 1132 } 1133 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1134 i = tp->t_rttvar * 1135 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1136 if (rt->rt_rmx.rmx_rttvar && i) 1137 rt->rt_rmx.rmx_rttvar = 1138 (rt->rt_rmx.rmx_rttvar + i) / 2; 1139 else 1140 rt->rt_rmx.rmx_rttvar = i; 1141 } 1142 /* 1143 * update the pipelimit (ssthresh) if it has been updated 1144 * already or if a pipesize was specified & the threshhold 1145 * got below half the pipesize. I.e., wait for bad news 1146 * before we start updating, then update on both good 1147 * and bad news. 1148 */ 1149 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1150 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1151 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1152 /* 1153 * convert the limit from user data bytes to 1154 * packets then to packet data bytes. 1155 */ 1156 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1157 if (i < 2) 1158 i = 2; 1159 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1160 if (rt->rt_rmx.rmx_ssthresh) 1161 rt->rt_rmx.rmx_ssthresh = 1162 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1163 else 1164 rt->rt_rmx.rmx_ssthresh = i; 1165 } 1166 } 1167 #endif /* RTV_RTT */ 1168 /* free the reassembly queue, if any */ 1169 TCP_REASS_LOCK(tp); 1170 (void) tcp_freeq(tp); 1171 TCP_REASS_UNLOCK(tp); 1172 1173 tcp_canceltimers(tp); 1174 TCP_CLEAR_DELACK(tp); 1175 syn_cache_cleanup(tp); 1176 1177 if (tp->t_template) { 1178 m_free(tp->t_template); 1179 tp->t_template = NULL; 1180 } 1181 if (tcp_timers_invoking(tp)) 1182 tp->t_flags |= TF_DEAD; 1183 else 1184 pool_put(&tcpcb_pool, tp); 1185 1186 if (inp) { 1187 inp->inp_ppcb = 0; 1188 soisdisconnected(so); 1189 in_pcbdetach(inp); 1190 } 1191 #ifdef INET6 1192 else if (in6p) { 1193 in6p->in6p_ppcb = 0; 1194 soisdisconnected(so); 1195 in6_pcbdetach(in6p); 1196 } 1197 #endif 1198 tcpstat.tcps_closed++; 1199 return ((struct tcpcb *)0); 1200 } 1201 1202 int 1203 tcp_freeq(tp) 1204 struct tcpcb *tp; 1205 { 1206 struct ipqent *qe; 1207 int rv = 0; 1208 #ifdef TCPREASS_DEBUG 1209 int i = 0; 1210 #endif 1211 1212 TCP_REASS_LOCK_CHECK(tp); 1213 1214 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1215 #ifdef TCPREASS_DEBUG 1216 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1217 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1218 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1219 #endif 1220 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1221 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1222 m_freem(qe->ipqe_m); 1223 pool_put(&ipqent_pool, qe); 1224 rv = 1; 1225 } 1226 return (rv); 1227 } 1228 1229 /* 1230 * Protocol drain routine. Called when memory is in short supply. 1231 */ 1232 void 1233 tcp_drain() 1234 { 1235 struct inpcb *inp; 1236 struct tcpcb *tp; 1237 1238 /* 1239 * Free the sequence queue of all TCP connections. 1240 */ 1241 inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue); 1242 if (inp) /* XXX */ 1243 CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) { 1244 if ((tp = intotcpcb(inp)) != NULL) { 1245 /* 1246 * We may be called from a device's interrupt 1247 * context. If the tcpcb is already busy, 1248 * just bail out now. 1249 */ 1250 if (tcp_reass_lock_try(tp) == 0) 1251 continue; 1252 if (tcp_freeq(tp)) 1253 tcpstat.tcps_connsdrained++; 1254 TCP_REASS_UNLOCK(tp); 1255 } 1256 } 1257 } 1258 1259 #ifdef INET6 1260 void 1261 tcp6_drain() 1262 { 1263 struct in6pcb *in6p; 1264 struct tcpcb *tp; 1265 struct in6pcb *head = &tcb6; 1266 1267 /* 1268 * Free the sequence queue of all TCP connections. 1269 */ 1270 for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) { 1271 if ((tp = in6totcpcb(in6p)) != NULL) { 1272 /* 1273 * We may be called from a device's interrupt 1274 * context. If the tcpcb is already busy, 1275 * just bail out now. 1276 */ 1277 if (tcp_reass_lock_try(tp) == 0) 1278 continue; 1279 if (tcp_freeq(tp)) 1280 tcpstat.tcps_connsdrained++; 1281 TCP_REASS_UNLOCK(tp); 1282 } 1283 } 1284 } 1285 #endif 1286 1287 /* 1288 * Notify a tcp user of an asynchronous error; 1289 * store error as soft error, but wake up user 1290 * (for now, won't do anything until can select for soft error). 1291 */ 1292 void 1293 tcp_notify(inp, error) 1294 struct inpcb *inp; 1295 int error; 1296 { 1297 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1298 struct socket *so = inp->inp_socket; 1299 1300 /* 1301 * Ignore some errors if we are hooked up. 1302 * If connection hasn't completed, has retransmitted several times, 1303 * and receives a second error, give up now. This is better 1304 * than waiting a long time to establish a connection that 1305 * can never complete. 1306 */ 1307 if (tp->t_state == TCPS_ESTABLISHED && 1308 (error == EHOSTUNREACH || error == ENETUNREACH || 1309 error == EHOSTDOWN)) { 1310 return; 1311 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1312 tp->t_rxtshift > 3 && tp->t_softerror) 1313 so->so_error = error; 1314 else 1315 tp->t_softerror = error; 1316 wakeup((caddr_t) &so->so_timeo); 1317 sorwakeup(so); 1318 sowwakeup(so); 1319 } 1320 1321 #ifdef INET6 1322 void 1323 tcp6_notify(in6p, error) 1324 struct in6pcb *in6p; 1325 int error; 1326 { 1327 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1328 struct socket *so = in6p->in6p_socket; 1329 1330 /* 1331 * Ignore some errors if we are hooked up. 1332 * If connection hasn't completed, has retransmitted several times, 1333 * and receives a second error, give up now. This is better 1334 * than waiting a long time to establish a connection that 1335 * can never complete. 1336 */ 1337 if (tp->t_state == TCPS_ESTABLISHED && 1338 (error == EHOSTUNREACH || error == ENETUNREACH || 1339 error == EHOSTDOWN)) { 1340 return; 1341 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1342 tp->t_rxtshift > 3 && tp->t_softerror) 1343 so->so_error = error; 1344 else 1345 tp->t_softerror = error; 1346 wakeup((caddr_t) &so->so_timeo); 1347 sorwakeup(so); 1348 sowwakeup(so); 1349 } 1350 #endif 1351 1352 #ifdef INET6 1353 void 1354 tcp6_ctlinput(cmd, sa, d) 1355 int cmd; 1356 struct sockaddr *sa; 1357 void *d; 1358 { 1359 struct tcphdr th; 1360 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1361 int nmatch; 1362 struct ip6_hdr *ip6; 1363 const struct sockaddr_in6 *sa6_src = NULL; 1364 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1365 struct mbuf *m; 1366 int off; 1367 1368 if (sa->sa_family != AF_INET6 || 1369 sa->sa_len != sizeof(struct sockaddr_in6)) 1370 return; 1371 if ((unsigned)cmd >= PRC_NCMDS) 1372 return; 1373 else if (cmd == PRC_QUENCH) { 1374 /* XXX there's no PRC_QUENCH in IPv6 */ 1375 notify = tcp6_quench; 1376 } else if (PRC_IS_REDIRECT(cmd)) 1377 notify = in6_rtchange, d = NULL; 1378 else if (cmd == PRC_MSGSIZE) 1379 ; /* special code is present, see below */ 1380 else if (cmd == PRC_HOSTDEAD) 1381 d = NULL; 1382 else if (inet6ctlerrmap[cmd] == 0) 1383 return; 1384 1385 /* if the parameter is from icmp6, decode it. */ 1386 if (d != NULL) { 1387 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1388 m = ip6cp->ip6c_m; 1389 ip6 = ip6cp->ip6c_ip6; 1390 off = ip6cp->ip6c_off; 1391 sa6_src = ip6cp->ip6c_src; 1392 } else { 1393 m = NULL; 1394 ip6 = NULL; 1395 sa6_src = &sa6_any; 1396 } 1397 1398 if (ip6) { 1399 /* 1400 * XXX: We assume that when ip6 is non NULL, 1401 * M and OFF are valid. 1402 */ 1403 1404 /* check if we can safely examine src and dst ports */ 1405 if (m->m_pkthdr.len < off + sizeof(th)) { 1406 if (cmd == PRC_MSGSIZE) 1407 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1408 return; 1409 } 1410 1411 bzero(&th, sizeof(th)); 1412 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1413 1414 if (cmd == PRC_MSGSIZE) { 1415 int valid = 0; 1416 1417 /* 1418 * Check to see if we have a valid TCP connection 1419 * corresponding to the address in the ICMPv6 message 1420 * payload. 1421 */ 1422 if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr, 1423 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1424 th.th_sport, 0)) 1425 valid++; 1426 1427 /* 1428 * Depending on the value of "valid" and routing table 1429 * size (mtudisc_{hi,lo}wat), we will: 1430 * - recalcurate the new MTU and create the 1431 * corresponding routing entry, or 1432 * - ignore the MTU change notification. 1433 */ 1434 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1435 1436 /* 1437 * no need to call in6_pcbnotify, it should have been 1438 * called via callback if necessary 1439 */ 1440 return; 1441 } 1442 1443 nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport, 1444 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1445 if (nmatch == 0 && syn_cache_count && 1446 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1447 inet6ctlerrmap[cmd] == ENETUNREACH || 1448 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1449 syn_cache_unreach((struct sockaddr *)sa6_src, 1450 sa, &th); 1451 } else { 1452 (void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src, 1453 0, cmd, NULL, notify); 1454 } 1455 } 1456 #endif 1457 1458 #ifdef INET 1459 /* assumes that ip header and tcp header are contiguous on mbuf */ 1460 void * 1461 tcp_ctlinput(cmd, sa, v) 1462 int cmd; 1463 struct sockaddr *sa; 1464 void *v; 1465 { 1466 struct ip *ip = v; 1467 struct tcphdr *th; 1468 struct icmp *icp; 1469 extern const int inetctlerrmap[]; 1470 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1471 int errno; 1472 int nmatch; 1473 #ifdef INET6 1474 struct in6_addr src6, dst6; 1475 #endif 1476 1477 if (sa->sa_family != AF_INET || 1478 sa->sa_len != sizeof(struct sockaddr_in)) 1479 return NULL; 1480 if ((unsigned)cmd >= PRC_NCMDS) 1481 return NULL; 1482 errno = inetctlerrmap[cmd]; 1483 if (cmd == PRC_QUENCH) 1484 notify = tcp_quench; 1485 else if (PRC_IS_REDIRECT(cmd)) 1486 notify = in_rtchange, ip = 0; 1487 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1488 /* 1489 * Check to see if we have a valid TCP connection 1490 * corresponding to the address in the ICMP message 1491 * payload. 1492 * 1493 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1494 */ 1495 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1496 #ifdef INET6 1497 memset(&src6, 0, sizeof(src6)); 1498 memset(&dst6, 0, sizeof(dst6)); 1499 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1500 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1501 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1502 #endif 1503 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1504 ip->ip_src, th->th_sport) != NULL) 1505 ; 1506 #ifdef INET6 1507 else if (in6_pcblookup_connect(&tcb6, &dst6, 1508 th->th_dport, &src6, th->th_sport, 0) != NULL) 1509 ; 1510 #endif 1511 else 1512 return NULL; 1513 1514 /* 1515 * Now that we've validated that we are actually communicating 1516 * with the host indicated in the ICMP message, locate the 1517 * ICMP header, recalculate the new MTU, and create the 1518 * corresponding routing entry. 1519 */ 1520 icp = (struct icmp *)((caddr_t)ip - 1521 offsetof(struct icmp, icmp_ip)); 1522 icmp_mtudisc(icp, ip->ip_dst); 1523 1524 return NULL; 1525 } else if (cmd == PRC_HOSTDEAD) 1526 ip = 0; 1527 else if (errno == 0) 1528 return NULL; 1529 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1530 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1531 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1532 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1533 if (nmatch == 0 && syn_cache_count && 1534 (inetctlerrmap[cmd] == EHOSTUNREACH || 1535 inetctlerrmap[cmd] == ENETUNREACH || 1536 inetctlerrmap[cmd] == EHOSTDOWN)) { 1537 struct sockaddr_in sin; 1538 bzero(&sin, sizeof(sin)); 1539 sin.sin_len = sizeof(sin); 1540 sin.sin_family = AF_INET; 1541 sin.sin_port = th->th_sport; 1542 sin.sin_addr = ip->ip_src; 1543 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1544 } 1545 1546 /* XXX mapped address case */ 1547 } else 1548 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1549 notify); 1550 return NULL; 1551 } 1552 1553 /* 1554 * When a source quence is received, we are being notifed of congestion. 1555 * Close the congestion window down to the Loss Window (one segment). 1556 * We will gradually open it again as we proceed. 1557 */ 1558 void 1559 tcp_quench(inp, errno) 1560 struct inpcb *inp; 1561 int errno; 1562 { 1563 struct tcpcb *tp = intotcpcb(inp); 1564 1565 if (tp) 1566 tp->snd_cwnd = tp->t_segsz; 1567 } 1568 #endif 1569 1570 #ifdef INET6 1571 void 1572 tcp6_quench(in6p, errno) 1573 struct in6pcb *in6p; 1574 int errno; 1575 { 1576 struct tcpcb *tp = in6totcpcb(in6p); 1577 1578 if (tp) 1579 tp->snd_cwnd = tp->t_segsz; 1580 } 1581 #endif 1582 1583 #ifdef INET 1584 /* 1585 * Path MTU Discovery handlers. 1586 */ 1587 void 1588 tcp_mtudisc_callback(faddr) 1589 struct in_addr faddr; 1590 { 1591 #ifdef INET6 1592 struct in6_addr in6; 1593 #endif 1594 1595 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1596 #ifdef INET6 1597 memset(&in6, 0, sizeof(in6)); 1598 in6.s6_addr16[5] = 0xffff; 1599 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1600 tcp6_mtudisc_callback(&in6); 1601 #endif 1602 } 1603 1604 /* 1605 * On receipt of path MTU corrections, flush old route and replace it 1606 * with the new one. Retransmit all unacknowledged packets, to ensure 1607 * that all packets will be received. 1608 */ 1609 void 1610 tcp_mtudisc(inp, errno) 1611 struct inpcb *inp; 1612 int errno; 1613 { 1614 struct tcpcb *tp = intotcpcb(inp); 1615 struct rtentry *rt = in_pcbrtentry(inp); 1616 1617 if (tp != 0) { 1618 if (rt != 0) { 1619 /* 1620 * If this was not a host route, remove and realloc. 1621 */ 1622 if ((rt->rt_flags & RTF_HOST) == 0) { 1623 in_rtchange(inp, errno); 1624 if ((rt = in_pcbrtentry(inp)) == 0) 1625 return; 1626 } 1627 1628 /* 1629 * Slow start out of the error condition. We 1630 * use the MTU because we know it's smaller 1631 * than the previously transmitted segment. 1632 * 1633 * Note: This is more conservative than the 1634 * suggestion in draft-floyd-incr-init-win-03. 1635 */ 1636 if (rt->rt_rmx.rmx_mtu != 0) 1637 tp->snd_cwnd = 1638 TCP_INITIAL_WINDOW(tcp_init_win, 1639 rt->rt_rmx.rmx_mtu); 1640 } 1641 1642 /* 1643 * Resend unacknowledged packets. 1644 */ 1645 tp->snd_nxt = tp->snd_una; 1646 tcp_output(tp); 1647 } 1648 } 1649 #endif 1650 1651 #ifdef INET6 1652 /* 1653 * Path MTU Discovery handlers. 1654 */ 1655 void 1656 tcp6_mtudisc_callback(faddr) 1657 struct in6_addr *faddr; 1658 { 1659 struct sockaddr_in6 sin6; 1660 1661 bzero(&sin6, sizeof(sin6)); 1662 sin6.sin6_family = AF_INET6; 1663 sin6.sin6_len = sizeof(struct sockaddr_in6); 1664 sin6.sin6_addr = *faddr; 1665 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1666 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1667 } 1668 1669 void 1670 tcp6_mtudisc(in6p, errno) 1671 struct in6pcb *in6p; 1672 int errno; 1673 { 1674 struct tcpcb *tp = in6totcpcb(in6p); 1675 struct rtentry *rt = in6_pcbrtentry(in6p); 1676 1677 if (tp != 0) { 1678 if (rt != 0) { 1679 /* 1680 * If this was not a host route, remove and realloc. 1681 */ 1682 if ((rt->rt_flags & RTF_HOST) == 0) { 1683 in6_rtchange(in6p, errno); 1684 if ((rt = in6_pcbrtentry(in6p)) == 0) 1685 return; 1686 } 1687 1688 /* 1689 * Slow start out of the error condition. We 1690 * use the MTU because we know it's smaller 1691 * than the previously transmitted segment. 1692 * 1693 * Note: This is more conservative than the 1694 * suggestion in draft-floyd-incr-init-win-03. 1695 */ 1696 if (rt->rt_rmx.rmx_mtu != 0) 1697 tp->snd_cwnd = 1698 TCP_INITIAL_WINDOW(tcp_init_win, 1699 rt->rt_rmx.rmx_mtu); 1700 } 1701 1702 /* 1703 * Resend unacknowledged packets. 1704 */ 1705 tp->snd_nxt = tp->snd_una; 1706 tcp_output(tp); 1707 } 1708 } 1709 #endif /* INET6 */ 1710 1711 /* 1712 * Compute the MSS to advertise to the peer. Called only during 1713 * the 3-way handshake. If we are the server (peer initiated 1714 * connection), we are called with a pointer to the interface 1715 * on which the SYN packet arrived. If we are the client (we 1716 * initiated connection), we are called with a pointer to the 1717 * interface out which this connection should go. 1718 * 1719 * NOTE: Do not subtract IP option/extension header size nor IPsec 1720 * header size from MSS advertisement. MSS option must hold the maximum 1721 * segment size we can accept, so it must always be: 1722 * max(if mtu) - ip header - tcp header 1723 */ 1724 u_long 1725 tcp_mss_to_advertise(ifp, af) 1726 const struct ifnet *ifp; 1727 int af; 1728 { 1729 extern u_long in_maxmtu; 1730 u_long mss = 0; 1731 u_long hdrsiz; 1732 1733 /* 1734 * In order to avoid defeating path MTU discovery on the peer, 1735 * we advertise the max MTU of all attached networks as our MSS, 1736 * per RFC 1191, section 3.1. 1737 * 1738 * We provide the option to advertise just the MTU of 1739 * the interface on which we hope this connection will 1740 * be receiving. If we are responding to a SYN, we 1741 * will have a pretty good idea about this, but when 1742 * initiating a connection there is a bit more doubt. 1743 * 1744 * We also need to ensure that loopback has a large enough 1745 * MSS, as the loopback MTU is never included in in_maxmtu. 1746 */ 1747 1748 if (ifp != NULL) 1749 switch (af) { 1750 case AF_INET: 1751 mss = ifp->if_mtu; 1752 break; 1753 #ifdef INET6 1754 case AF_INET6: 1755 mss = IN6_LINKMTU(ifp); 1756 break; 1757 #endif 1758 } 1759 1760 if (tcp_mss_ifmtu == 0) 1761 switch (af) { 1762 case AF_INET: 1763 mss = max(in_maxmtu, mss); 1764 break; 1765 #ifdef INET6 1766 case AF_INET6: 1767 mss = max(in6_maxmtu, mss); 1768 break; 1769 #endif 1770 } 1771 1772 switch (af) { 1773 case AF_INET: 1774 hdrsiz = sizeof(struct ip); 1775 break; 1776 #ifdef INET6 1777 case AF_INET6: 1778 hdrsiz = sizeof(struct ip6_hdr); 1779 break; 1780 #endif 1781 default: 1782 hdrsiz = 0; 1783 break; 1784 } 1785 hdrsiz += sizeof(struct tcphdr); 1786 if (mss > hdrsiz) 1787 mss -= hdrsiz; 1788 1789 mss = max(tcp_mssdflt, mss); 1790 return (mss); 1791 } 1792 1793 /* 1794 * Set connection variables based on the peer's advertised MSS. 1795 * We are passed the TCPCB for the actual connection. If we 1796 * are the server, we are called by the compressed state engine 1797 * when the 3-way handshake is complete. If we are the client, 1798 * we are called when we receive the SYN,ACK from the server. 1799 * 1800 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1801 * before this routine is called! 1802 */ 1803 void 1804 tcp_mss_from_peer(tp, offer) 1805 struct tcpcb *tp; 1806 int offer; 1807 { 1808 struct socket *so; 1809 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1810 struct rtentry *rt; 1811 #endif 1812 u_long bufsize; 1813 int mss; 1814 1815 #ifdef DIAGNOSTIC 1816 if (tp->t_inpcb && tp->t_in6pcb) 1817 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1818 #endif 1819 so = NULL; 1820 rt = NULL; 1821 #ifdef INET 1822 if (tp->t_inpcb) { 1823 so = tp->t_inpcb->inp_socket; 1824 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1825 rt = in_pcbrtentry(tp->t_inpcb); 1826 #endif 1827 } 1828 #endif 1829 #ifdef INET6 1830 if (tp->t_in6pcb) { 1831 so = tp->t_in6pcb->in6p_socket; 1832 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1833 rt = in6_pcbrtentry(tp->t_in6pcb); 1834 #endif 1835 } 1836 #endif 1837 1838 /* 1839 * As per RFC1122, use the default MSS value, unless they 1840 * sent us an offer. Do not accept offers less than 32 bytes. 1841 */ 1842 mss = tcp_mssdflt; 1843 if (offer) 1844 mss = offer; 1845 mss = max(mss, 32); /* sanity */ 1846 tp->t_peermss = mss; 1847 mss -= tcp_optlen(tp); 1848 #ifdef INET 1849 if (tp->t_inpcb) 1850 mss -= ip_optlen(tp->t_inpcb); 1851 #endif 1852 #ifdef INET6 1853 if (tp->t_in6pcb) 1854 mss -= ip6_optlen(tp->t_in6pcb); 1855 #endif 1856 1857 /* 1858 * If there's a pipesize, change the socket buffer to that size. 1859 * Make the socket buffer an integral number of MSS units. If 1860 * the MSS is larger than the socket buffer, artificially decrease 1861 * the MSS. 1862 */ 1863 #ifdef RTV_SPIPE 1864 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1865 bufsize = rt->rt_rmx.rmx_sendpipe; 1866 else 1867 #endif 1868 bufsize = so->so_snd.sb_hiwat; 1869 if (bufsize < mss) 1870 mss = bufsize; 1871 else { 1872 bufsize = roundup(bufsize, mss); 1873 if (bufsize > sb_max) 1874 bufsize = sb_max; 1875 (void) sbreserve(&so->so_snd, bufsize); 1876 } 1877 tp->t_segsz = mss; 1878 1879 #ifdef RTV_SSTHRESH 1880 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1881 /* 1882 * There's some sort of gateway or interface buffer 1883 * limit on the path. Use this to set the slow 1884 * start threshold, but set the threshold to no less 1885 * than 2 * MSS. 1886 */ 1887 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1888 } 1889 #endif 1890 } 1891 1892 /* 1893 * Processing necessary when a TCP connection is established. 1894 */ 1895 void 1896 tcp_established(tp) 1897 struct tcpcb *tp; 1898 { 1899 struct socket *so; 1900 #ifdef RTV_RPIPE 1901 struct rtentry *rt; 1902 #endif 1903 u_long bufsize; 1904 1905 #ifdef DIAGNOSTIC 1906 if (tp->t_inpcb && tp->t_in6pcb) 1907 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1908 #endif 1909 so = NULL; 1910 rt = NULL; 1911 #ifdef INET 1912 if (tp->t_inpcb) { 1913 so = tp->t_inpcb->inp_socket; 1914 #if defined(RTV_RPIPE) 1915 rt = in_pcbrtentry(tp->t_inpcb); 1916 #endif 1917 } 1918 #endif 1919 #ifdef INET6 1920 if (tp->t_in6pcb) { 1921 so = tp->t_in6pcb->in6p_socket; 1922 #if defined(RTV_RPIPE) 1923 rt = in6_pcbrtentry(tp->t_in6pcb); 1924 #endif 1925 } 1926 #endif 1927 1928 tp->t_state = TCPS_ESTABLISHED; 1929 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1930 1931 #ifdef RTV_RPIPE 1932 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1933 bufsize = rt->rt_rmx.rmx_recvpipe; 1934 else 1935 #endif 1936 bufsize = so->so_rcv.sb_hiwat; 1937 if (bufsize > tp->t_ourmss) { 1938 bufsize = roundup(bufsize, tp->t_ourmss); 1939 if (bufsize > sb_max) 1940 bufsize = sb_max; 1941 (void) sbreserve(&so->so_rcv, bufsize); 1942 } 1943 } 1944 1945 /* 1946 * Check if there's an initial rtt or rttvar. Convert from the 1947 * route-table units to scaled multiples of the slow timeout timer. 1948 * Called only during the 3-way handshake. 1949 */ 1950 void 1951 tcp_rmx_rtt(tp) 1952 struct tcpcb *tp; 1953 { 1954 #ifdef RTV_RTT 1955 struct rtentry *rt = NULL; 1956 int rtt; 1957 1958 #ifdef DIAGNOSTIC 1959 if (tp->t_inpcb && tp->t_in6pcb) 1960 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1961 #endif 1962 #ifdef INET 1963 if (tp->t_inpcb) 1964 rt = in_pcbrtentry(tp->t_inpcb); 1965 #endif 1966 #ifdef INET6 1967 if (tp->t_in6pcb) 1968 rt = in6_pcbrtentry(tp->t_in6pcb); 1969 #endif 1970 if (rt == NULL) 1971 return; 1972 1973 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1974 /* 1975 * XXX The lock bit for MTU indicates that the value 1976 * is also a minimum value; this is subject to time. 1977 */ 1978 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1979 TCPT_RANGESET(tp->t_rttmin, 1980 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1981 TCPTV_MIN, TCPTV_REXMTMAX); 1982 tp->t_srtt = rtt / 1983 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1984 if (rt->rt_rmx.rmx_rttvar) { 1985 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1986 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1987 (TCP_RTTVAR_SHIFT + 2)); 1988 } else { 1989 /* Default variation is +- 1 rtt */ 1990 tp->t_rttvar = 1991 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1992 } 1993 TCPT_RANGESET(tp->t_rxtcur, 1994 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1995 tp->t_rttmin, TCPTV_REXMTMAX); 1996 } 1997 #endif 1998 } 1999 2000 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2001 #if NRND > 0 2002 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2003 #endif 2004 2005 /* 2006 * Get a new sequence value given a tcp control block 2007 */ 2008 tcp_seq 2009 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2010 { 2011 2012 #ifdef INET 2013 if (tp->t_inpcb != NULL) { 2014 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2015 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2016 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2017 addin)); 2018 } 2019 #endif 2020 #ifdef INET6 2021 if (tp->t_in6pcb != NULL) { 2022 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2023 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2024 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2025 addin)); 2026 } 2027 #endif 2028 /* Not possible. */ 2029 panic("tcp_new_iss"); 2030 } 2031 2032 /* 2033 * This routine actually generates a new TCP initial sequence number. 2034 */ 2035 tcp_seq 2036 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2037 size_t addrsz, tcp_seq addin) 2038 { 2039 tcp_seq tcp_iss; 2040 2041 #if NRND > 0 2042 static int beenhere; 2043 2044 /* 2045 * If we haven't been here before, initialize our cryptographic 2046 * hash secret. 2047 */ 2048 if (beenhere == 0) { 2049 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2050 RND_EXTRACT_ANY); 2051 beenhere = 1; 2052 } 2053 2054 if (tcp_do_rfc1948) { 2055 MD5_CTX ctx; 2056 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2057 2058 /* 2059 * Compute the base value of the ISS. It is a hash 2060 * of (saddr, sport, daddr, dport, secret). 2061 */ 2062 MD5Init(&ctx); 2063 2064 MD5Update(&ctx, (u_char *) laddr, addrsz); 2065 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2066 2067 MD5Update(&ctx, (u_char *) faddr, addrsz); 2068 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2069 2070 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2071 2072 MD5Final(hash, &ctx); 2073 2074 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2075 2076 /* 2077 * Now increment our "timer", and add it in to 2078 * the computed value. 2079 * 2080 * XXX Use `addin'? 2081 * XXX TCP_ISSINCR too large to use? 2082 */ 2083 tcp_iss_seq += TCP_ISSINCR; 2084 #ifdef TCPISS_DEBUG 2085 printf("ISS hash 0x%08x, ", tcp_iss); 2086 #endif 2087 tcp_iss += tcp_iss_seq + addin; 2088 #ifdef TCPISS_DEBUG 2089 printf("new ISS 0x%08x\n", tcp_iss); 2090 #endif 2091 } else 2092 #endif /* NRND > 0 */ 2093 { 2094 /* 2095 * Randomize. 2096 */ 2097 #if NRND > 0 2098 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2099 #else 2100 tcp_iss = arc4random(); 2101 #endif 2102 2103 /* 2104 * If we were asked to add some amount to a known value, 2105 * we will take a random value obtained above, mask off 2106 * the upper bits, and add in the known value. We also 2107 * add in a constant to ensure that we are at least a 2108 * certain distance from the original value. 2109 * 2110 * This is used when an old connection is in timed wait 2111 * and we have a new one coming in, for instance. 2112 */ 2113 if (addin != 0) { 2114 #ifdef TCPISS_DEBUG 2115 printf("Random %08x, ", tcp_iss); 2116 #endif 2117 tcp_iss &= TCP_ISS_RANDOM_MASK; 2118 tcp_iss += addin + TCP_ISSINCR; 2119 #ifdef TCPISS_DEBUG 2120 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2121 #endif 2122 } else { 2123 tcp_iss &= TCP_ISS_RANDOM_MASK; 2124 tcp_iss += tcp_iss_seq; 2125 tcp_iss_seq += TCP_ISSINCR; 2126 #ifdef TCPISS_DEBUG 2127 printf("ISS %08x\n", tcp_iss); 2128 #endif 2129 } 2130 } 2131 2132 if (tcp_compat_42) { 2133 /* 2134 * Limit it to the positive range for really old TCP 2135 * implementations. 2136 * Just AND off the top bit instead of checking if 2137 * is set first - saves a branch 50% of the time. 2138 */ 2139 tcp_iss &= 0x7fffffff; /* XXX */ 2140 } 2141 2142 return (tcp_iss); 2143 } 2144 2145 #if defined(IPSEC) || defined(FAST_IPSEC) 2146 /* compute ESP/AH header size for TCP, including outer IP header. */ 2147 size_t 2148 ipsec4_hdrsiz_tcp(tp) 2149 struct tcpcb *tp; 2150 { 2151 struct inpcb *inp; 2152 size_t hdrsiz; 2153 2154 /* XXX mapped addr case (tp->t_in6pcb) */ 2155 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2156 return 0; 2157 switch (tp->t_family) { 2158 case AF_INET: 2159 /* XXX: should use currect direction. */ 2160 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2161 break; 2162 default: 2163 hdrsiz = 0; 2164 break; 2165 } 2166 2167 return hdrsiz; 2168 } 2169 2170 #ifdef INET6 2171 size_t 2172 ipsec6_hdrsiz_tcp(tp) 2173 struct tcpcb *tp; 2174 { 2175 struct in6pcb *in6p; 2176 size_t hdrsiz; 2177 2178 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2179 return 0; 2180 switch (tp->t_family) { 2181 case AF_INET6: 2182 /* XXX: should use currect direction. */ 2183 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2184 break; 2185 case AF_INET: 2186 /* mapped address case - tricky */ 2187 default: 2188 hdrsiz = 0; 2189 break; 2190 } 2191 2192 return hdrsiz; 2193 } 2194 #endif 2195 #endif /*IPSEC*/ 2196 2197 /* 2198 * Determine the length of the TCP options for this connection. 2199 * 2200 * XXX: What do we do for SACK, when we add that? Just reserve 2201 * all of the space? Otherwise we can't exactly be incrementing 2202 * cwnd by an amount that varies depending on the amount we last 2203 * had to SACK! 2204 */ 2205 2206 u_int 2207 tcp_optlen(tp) 2208 struct tcpcb *tp; 2209 { 2210 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2211 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2212 return TCPOLEN_TSTAMP_APPA; 2213 else 2214 return 0; 2215 } 2216
Indexes created Mon Sep 22 13:09:51 GMT 2025