tcp_subr.c revision 1.153
1 /* $NetBSD: tcp_subr.c,v 1.153 2003/09/08 02:06:34 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 1.153 2003/09/08 02:06:34 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 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 169 struct tcpstat tcpstat; /* tcp statistics */ 170 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 171 172 /* patchable/settable parameters for tcp */ 173 int tcp_mssdflt = TCP_MSS; 174 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 175 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 176 #if NRND > 0 177 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 178 #endif 179 int tcp_do_sack = 1; /* selective acknowledgement */ 180 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 181 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 182 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 183 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 184 #ifndef TCP_INIT_WIN 185 #define TCP_INIT_WIN 1 /* initial slow start window */ 186 #endif 187 #ifndef TCP_INIT_WIN_LOCAL 188 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 189 #endif 190 int tcp_init_win = TCP_INIT_WIN; 191 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 192 int tcp_mss_ifmtu = 0; 193 #ifdef TCP_COMPAT_42 194 int tcp_compat_42 = 1; 195 #else 196 int tcp_compat_42 = 0; 197 #endif 198 int tcp_rst_ppslim = 100; /* 100pps */ 199 200 /* tcb hash */ 201 #ifndef TCBHASHSIZE 202 #define TCBHASHSIZE 128 203 #endif 204 int tcbhashsize = TCBHASHSIZE; 205 206 /* syn hash parameters */ 207 #define TCP_SYN_HASH_SIZE 293 208 #define TCP_SYN_BUCKET_SIZE 35 209 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 210 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 211 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 212 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 213 214 int tcp_freeq __P((struct tcpcb *)); 215 216 #ifdef INET 217 void tcp_mtudisc_callback __P((struct in_addr)); 218 #endif 219 #ifdef INET6 220 void tcp6_mtudisc_callback __P((struct in6_addr *)); 221 #endif 222 223 void tcp_mtudisc __P((struct inpcb *, int)); 224 #ifdef INET6 225 void tcp6_mtudisc __P((struct in6pcb *, int)); 226 #endif 227 228 struct pool tcpcb_pool; 229 230 #ifdef TCP_CSUM_COUNTERS 231 #include <sys/device.h> 232 233 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 234 NULL, "tcp", "hwcsum bad"); 235 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 236 NULL, "tcp", "hwcsum ok"); 237 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 238 NULL, "tcp", "hwcsum data"); 239 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 240 NULL, "tcp", "swcsum"); 241 #endif /* TCP_CSUM_COUNTERS */ 242 243 #ifdef TCP_OUTPUT_COUNTERS 244 #include <sys/device.h> 245 246 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 247 NULL, "tcp", "output big header"); 248 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 249 NULL, "tcp", "output copy small"); 250 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp", "output copy big"); 252 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 NULL, "tcp", "output reference big"); 254 #endif /* TCP_OUTPUT_COUNTERS */ 255 256 #ifdef TCP_REASS_COUNTERS 257 #include <sys/device.h> 258 259 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp_reass", "calls"); 261 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 &tcp_reass_, "tcp_reass", "insert into empty queue"); 263 struct evcnt tcp_reass_iteration[8] = { 264 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 265 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 266 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 267 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 268 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 269 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 270 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 271 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 272 }; 273 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 274 &tcp_reass_, "tcp_reass", "prepend to first"); 275 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 276 &tcp_reass_, "tcp_reass", "prepend"); 277 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 278 &tcp_reass_, "tcp_reass", "insert"); 279 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 &tcp_reass_, "tcp_reass", "insert at tail"); 281 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 &tcp_reass_, "tcp_reass", "append"); 283 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 &tcp_reass_, "tcp_reass", "append to tail fragment"); 285 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 &tcp_reass_, "tcp_reass", "overlap at end"); 287 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 &tcp_reass_, "tcp_reass", "overlap at start"); 289 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 &tcp_reass_, "tcp_reass", "duplicate segment"); 291 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 &tcp_reass_, "tcp_reass", "duplicate fragment"); 293 294 #endif /* TCP_REASS_COUNTERS */ 295 296 #ifdef MBUFTRACE 297 struct mowner tcp_mowner = { "tcp" }; 298 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 299 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 300 #endif 301 302 /* 303 * Tcp initialization 304 */ 305 void 306 tcp_init() 307 { 308 int hlen; 309 310 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 311 NULL); 312 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 313 314 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 315 #ifdef INET6 316 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 317 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 318 #endif 319 if (max_protohdr < hlen) 320 max_protohdr = hlen; 321 if (max_linkhdr + hlen > MHLEN) 322 panic("tcp_init"); 323 324 #ifdef INET 325 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 326 #endif 327 #ifdef INET6 328 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 329 #endif 330 331 /* Initialize timer state. */ 332 tcp_timer_init(); 333 334 /* Initialize the compressed state engine. */ 335 syn_cache_init(); 336 337 #ifdef TCP_CSUM_COUNTERS 338 evcnt_attach_static(&tcp_hwcsum_bad); 339 evcnt_attach_static(&tcp_hwcsum_ok); 340 evcnt_attach_static(&tcp_hwcsum_data); 341 evcnt_attach_static(&tcp_swcsum); 342 #endif /* TCP_CSUM_COUNTERS */ 343 344 #ifdef TCP_OUTPUT_COUNTERS 345 evcnt_attach_static(&tcp_output_bigheader); 346 evcnt_attach_static(&tcp_output_copysmall); 347 evcnt_attach_static(&tcp_output_copybig); 348 evcnt_attach_static(&tcp_output_refbig); 349 #endif /* TCP_OUTPUT_COUNTERS */ 350 351 #ifdef TCP_REASS_COUNTERS 352 evcnt_attach_static(&tcp_reass_); 353 evcnt_attach_static(&tcp_reass_empty); 354 evcnt_attach_static(&tcp_reass_iteration[0]); 355 evcnt_attach_static(&tcp_reass_iteration[1]); 356 evcnt_attach_static(&tcp_reass_iteration[2]); 357 evcnt_attach_static(&tcp_reass_iteration[3]); 358 evcnt_attach_static(&tcp_reass_iteration[4]); 359 evcnt_attach_static(&tcp_reass_iteration[5]); 360 evcnt_attach_static(&tcp_reass_iteration[6]); 361 evcnt_attach_static(&tcp_reass_iteration[7]); 362 evcnt_attach_static(&tcp_reass_prependfirst); 363 evcnt_attach_static(&tcp_reass_prepend); 364 evcnt_attach_static(&tcp_reass_insert); 365 evcnt_attach_static(&tcp_reass_inserttail); 366 evcnt_attach_static(&tcp_reass_append); 367 evcnt_attach_static(&tcp_reass_appendtail); 368 evcnt_attach_static(&tcp_reass_overlaptail); 369 evcnt_attach_static(&tcp_reass_overlapfront); 370 evcnt_attach_static(&tcp_reass_segdup); 371 evcnt_attach_static(&tcp_reass_fragdup); 372 #endif /* TCP_REASS_COUNTERS */ 373 374 MOWNER_ATTACH(&tcp_tx_mowner); 375 MOWNER_ATTACH(&tcp_rx_mowner); 376 MOWNER_ATTACH(&tcp_mowner); 377 } 378 379 /* 380 * Create template to be used to send tcp packets on a connection. 381 * Call after host entry created, allocates an mbuf and fills 382 * in a skeletal tcp/ip header, minimizing the amount of work 383 * necessary when the connection is used. 384 */ 385 struct mbuf * 386 tcp_template(tp) 387 struct tcpcb *tp; 388 { 389 struct inpcb *inp = tp->t_inpcb; 390 #ifdef INET6 391 struct in6pcb *in6p = tp->t_in6pcb; 392 #endif 393 struct tcphdr *n; 394 struct mbuf *m; 395 int hlen; 396 397 switch (tp->t_family) { 398 case AF_INET: 399 hlen = sizeof(struct ip); 400 if (inp) 401 break; 402 #ifdef INET6 403 if (in6p) { 404 /* mapped addr case */ 405 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 406 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 407 break; 408 } 409 #endif 410 return NULL; /*EINVAL*/ 411 #ifdef INET6 412 case AF_INET6: 413 hlen = sizeof(struct ip6_hdr); 414 if (in6p) { 415 /* more sainty check? */ 416 break; 417 } 418 return NULL; /*EINVAL*/ 419 #endif 420 default: 421 hlen = 0; /*pacify gcc*/ 422 return NULL; /*EAFNOSUPPORT*/ 423 } 424 #ifdef DIAGNOSTIC 425 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 426 panic("mclbytes too small for t_template"); 427 #endif 428 m = tp->t_template; 429 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 430 ; 431 else { 432 if (m) 433 m_freem(m); 434 m = tp->t_template = NULL; 435 MGETHDR(m, M_DONTWAIT, MT_HEADER); 436 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 437 MCLGET(m, M_DONTWAIT); 438 if ((m->m_flags & M_EXT) == 0) { 439 m_free(m); 440 m = NULL; 441 } 442 } 443 if (m == NULL) 444 return NULL; 445 MCLAIM(m, &tcp_mowner); 446 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 447 } 448 449 bzero(mtod(m, caddr_t), m->m_len); 450 451 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 452 453 switch (tp->t_family) { 454 case AF_INET: 455 { 456 struct ipovly *ipov; 457 mtod(m, struct ip *)->ip_v = 4; 458 mtod(m, struct ip *)->ip_hl = hlen >> 2; 459 ipov = mtod(m, struct ipovly *); 460 ipov->ih_pr = IPPROTO_TCP; 461 ipov->ih_len = htons(sizeof(struct tcphdr)); 462 if (inp) { 463 ipov->ih_src = inp->inp_laddr; 464 ipov->ih_dst = inp->inp_faddr; 465 } 466 #ifdef INET6 467 else if (in6p) { 468 /* mapped addr case */ 469 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 470 sizeof(ipov->ih_src)); 471 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 472 sizeof(ipov->ih_dst)); 473 } 474 #endif 475 /* 476 * Compute the pseudo-header portion of the checksum 477 * now. We incrementally add in the TCP option and 478 * payload lengths later, and then compute the TCP 479 * checksum right before the packet is sent off onto 480 * the wire. 481 */ 482 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 483 ipov->ih_dst.s_addr, 484 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 485 break; 486 } 487 #ifdef INET6 488 case AF_INET6: 489 { 490 struct ip6_hdr *ip6; 491 mtod(m, struct ip *)->ip_v = 6; 492 ip6 = mtod(m, struct ip6_hdr *); 493 ip6->ip6_nxt = IPPROTO_TCP; 494 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 495 ip6->ip6_src = in6p->in6p_laddr; 496 ip6->ip6_dst = in6p->in6p_faddr; 497 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 498 if (ip6_auto_flowlabel) { 499 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 500 ip6->ip6_flow |= 501 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 502 } 503 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 504 ip6->ip6_vfc |= IPV6_VERSION; 505 506 /* 507 * Compute the pseudo-header portion of the checksum 508 * now. We incrementally add in the TCP option and 509 * payload lengths later, and then compute the TCP 510 * checksum right before the packet is sent off onto 511 * the wire. 512 */ 513 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 514 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 515 htonl(IPPROTO_TCP)); 516 break; 517 } 518 #endif 519 } 520 if (inp) { 521 n->th_sport = inp->inp_lport; 522 n->th_dport = inp->inp_fport; 523 } 524 #ifdef INET6 525 else if (in6p) { 526 n->th_sport = in6p->in6p_lport; 527 n->th_dport = in6p->in6p_fport; 528 } 529 #endif 530 n->th_seq = 0; 531 n->th_ack = 0; 532 n->th_x2 = 0; 533 n->th_off = 5; 534 n->th_flags = 0; 535 n->th_win = 0; 536 n->th_urp = 0; 537 return (m); 538 } 539 540 /* 541 * Send a single message to the TCP at address specified by 542 * the given TCP/IP header. If m == 0, then we make a copy 543 * of the tcpiphdr at ti and send directly to the addressed host. 544 * This is used to force keep alive messages out using the TCP 545 * template for a connection tp->t_template. If flags are given 546 * then we send a message back to the TCP which originated the 547 * segment ti, and discard the mbuf containing it and any other 548 * attached mbufs. 549 * 550 * In any case the ack and sequence number of the transmitted 551 * segment are as specified by the parameters. 552 */ 553 int 554 tcp_respond(tp, template, m, th0, ack, seq, flags) 555 struct tcpcb *tp; 556 struct mbuf *template; 557 struct mbuf *m; 558 struct tcphdr *th0; 559 tcp_seq ack, seq; 560 int flags; 561 { 562 struct route *ro; 563 int error, tlen, win = 0; 564 int hlen; 565 struct ip *ip; 566 #ifdef INET6 567 struct ip6_hdr *ip6; 568 #endif 569 int family; /* family on packet, not inpcb/in6pcb! */ 570 struct tcphdr *th; 571 struct socket *so; 572 573 if (tp != NULL && (flags & TH_RST) == 0) { 574 #ifdef DIAGNOSTIC 575 if (tp->t_inpcb && tp->t_in6pcb) 576 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 577 #endif 578 #ifdef INET 579 if (tp->t_inpcb) 580 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 581 #endif 582 #ifdef INET6 583 if (tp->t_in6pcb) 584 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 585 #endif 586 } 587 588 th = NULL; /* Quell uninitialized warning */ 589 ip = NULL; 590 #ifdef INET6 591 ip6 = NULL; 592 #endif 593 if (m == 0) { 594 if (!template) 595 return EINVAL; 596 597 /* get family information from template */ 598 switch (mtod(template, struct ip *)->ip_v) { 599 case 4: 600 family = AF_INET; 601 hlen = sizeof(struct ip); 602 break; 603 #ifdef INET6 604 case 6: 605 family = AF_INET6; 606 hlen = sizeof(struct ip6_hdr); 607 break; 608 #endif 609 default: 610 return EAFNOSUPPORT; 611 } 612 613 MGETHDR(m, M_DONTWAIT, MT_HEADER); 614 if (m) { 615 MCLAIM(m, &tcp_tx_mowner); 616 MCLGET(m, M_DONTWAIT); 617 if ((m->m_flags & M_EXT) == 0) { 618 m_free(m); 619 m = NULL; 620 } 621 } 622 if (m == NULL) 623 return (ENOBUFS); 624 625 if (tcp_compat_42) 626 tlen = 1; 627 else 628 tlen = 0; 629 630 m->m_data += max_linkhdr; 631 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 632 template->m_len); 633 switch (family) { 634 case AF_INET: 635 ip = mtod(m, struct ip *); 636 th = (struct tcphdr *)(ip + 1); 637 break; 638 #ifdef INET6 639 case AF_INET6: 640 ip6 = mtod(m, struct ip6_hdr *); 641 th = (struct tcphdr *)(ip6 + 1); 642 break; 643 #endif 644 #if 0 645 default: 646 /* noone will visit here */ 647 m_freem(m); 648 return EAFNOSUPPORT; 649 #endif 650 } 651 flags = TH_ACK; 652 } else { 653 654 if ((m->m_flags & M_PKTHDR) == 0) { 655 #if 0 656 printf("non PKTHDR to tcp_respond\n"); 657 #endif 658 m_freem(m); 659 return EINVAL; 660 } 661 #ifdef DIAGNOSTIC 662 if (!th0) 663 panic("th0 == NULL in tcp_respond"); 664 #endif 665 666 /* get family information from m */ 667 switch (mtod(m, struct ip *)->ip_v) { 668 case 4: 669 family = AF_INET; 670 hlen = sizeof(struct ip); 671 ip = mtod(m, struct ip *); 672 break; 673 #ifdef INET6 674 case 6: 675 family = AF_INET6; 676 hlen = sizeof(struct ip6_hdr); 677 ip6 = mtod(m, struct ip6_hdr *); 678 break; 679 #endif 680 default: 681 m_freem(m); 682 return EAFNOSUPPORT; 683 } 684 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 685 tlen = sizeof(*th0); 686 else 687 tlen = th0->th_off << 2; 688 689 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 690 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 691 m->m_len = hlen + tlen; 692 m_freem(m->m_next); 693 m->m_next = NULL; 694 } else { 695 struct mbuf *n; 696 697 #ifdef DIAGNOSTIC 698 if (max_linkhdr + hlen + tlen > MCLBYTES) { 699 m_freem(m); 700 return EMSGSIZE; 701 } 702 #endif 703 MGETHDR(n, M_DONTWAIT, MT_HEADER); 704 if (n && max_linkhdr + hlen + tlen > MHLEN) { 705 MCLGET(n, M_DONTWAIT); 706 if ((n->m_flags & M_EXT) == 0) { 707 m_freem(n); 708 n = NULL; 709 } 710 } 711 if (!n) { 712 m_freem(m); 713 return ENOBUFS; 714 } 715 716 MCLAIM(n, &tcp_tx_mowner); 717 n->m_data += max_linkhdr; 718 n->m_len = hlen + tlen; 719 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 720 m_copyback(n, hlen, tlen, (caddr_t)th0); 721 722 m_freem(m); 723 m = n; 724 n = NULL; 725 } 726 727 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 728 switch (family) { 729 case AF_INET: 730 ip = mtod(m, struct ip *); 731 th = (struct tcphdr *)(ip + 1); 732 ip->ip_p = IPPROTO_TCP; 733 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 734 ip->ip_p = IPPROTO_TCP; 735 break; 736 #ifdef INET6 737 case AF_INET6: 738 ip6 = mtod(m, struct ip6_hdr *); 739 th = (struct tcphdr *)(ip6 + 1); 740 ip6->ip6_nxt = IPPROTO_TCP; 741 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 742 ip6->ip6_nxt = IPPROTO_TCP; 743 break; 744 #endif 745 #if 0 746 default: 747 /* noone will visit here */ 748 m_freem(m); 749 return EAFNOSUPPORT; 750 #endif 751 } 752 xchg(th->th_dport, th->th_sport, u_int16_t); 753 #undef xchg 754 tlen = 0; /*be friendly with the following code*/ 755 } 756 th->th_seq = htonl(seq); 757 th->th_ack = htonl(ack); 758 th->th_x2 = 0; 759 if ((flags & TH_SYN) == 0) { 760 if (tp) 761 win >>= tp->rcv_scale; 762 if (win > TCP_MAXWIN) 763 win = TCP_MAXWIN; 764 th->th_win = htons((u_int16_t)win); 765 th->th_off = sizeof (struct tcphdr) >> 2; 766 tlen += sizeof(*th); 767 } else 768 tlen += th->th_off << 2; 769 m->m_len = hlen + tlen; 770 m->m_pkthdr.len = hlen + tlen; 771 m->m_pkthdr.rcvif = (struct ifnet *) 0; 772 th->th_flags = flags; 773 th->th_urp = 0; 774 775 switch (family) { 776 #ifdef INET 777 case AF_INET: 778 { 779 struct ipovly *ipov = (struct ipovly *)ip; 780 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 781 ipov->ih_len = htons((u_int16_t)tlen); 782 783 th->th_sum = 0; 784 th->th_sum = in_cksum(m, hlen + tlen); 785 ip->ip_len = htons(hlen + tlen); 786 ip->ip_ttl = ip_defttl; 787 break; 788 } 789 #endif 790 #ifdef INET6 791 case AF_INET6: 792 { 793 th->th_sum = 0; 794 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 795 tlen); 796 ip6->ip6_plen = ntohs(tlen); 797 if (tp && tp->t_in6pcb) { 798 struct ifnet *oifp; 799 ro = (struct route *)&tp->t_in6pcb->in6p_route; 800 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 801 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 802 } else 803 ip6->ip6_hlim = ip6_defhlim; 804 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 805 if (ip6_auto_flowlabel) { 806 ip6->ip6_flow |= 807 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 808 } 809 break; 810 } 811 #endif 812 } 813 814 if (tp && tp->t_inpcb) 815 so = tp->t_inpcb->inp_socket; 816 #ifdef INET6 817 else if (tp && tp->t_in6pcb) 818 so = tp->t_in6pcb->in6p_socket; 819 #endif 820 else 821 so = NULL; 822 823 if (tp != NULL && tp->t_inpcb != NULL) { 824 ro = &tp->t_inpcb->inp_route; 825 #ifdef DIAGNOSTIC 826 if (family != AF_INET) 827 panic("tcp_respond: address family mismatch"); 828 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 829 panic("tcp_respond: ip_dst %x != inp_faddr %x", 830 ntohl(ip->ip_dst.s_addr), 831 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 832 } 833 #endif 834 } 835 #ifdef INET6 836 else if (tp != NULL && tp->t_in6pcb != NULL) { 837 ro = (struct route *)&tp->t_in6pcb->in6p_route; 838 #ifdef DIAGNOSTIC 839 if (family == AF_INET) { 840 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 841 panic("tcp_respond: not mapped addr"); 842 if (bcmp(&ip->ip_dst, 843 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 844 sizeof(ip->ip_dst)) != 0) { 845 panic("tcp_respond: ip_dst != in6p_faddr"); 846 } 847 } else if (family == AF_INET6) { 848 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 849 &tp->t_in6pcb->in6p_faddr)) 850 panic("tcp_respond: ip6_dst != in6p_faddr"); 851 } else 852 panic("tcp_respond: address family mismatch"); 853 #endif 854 } 855 #endif 856 else 857 ro = NULL; 858 859 switch (family) { 860 #ifdef INET 861 case AF_INET: 862 error = ip_output(m, NULL, ro, 863 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 864 (struct ip_moptions *)0, so); 865 break; 866 #endif 867 #ifdef INET6 868 case AF_INET6: 869 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 870 (struct ip6_moptions *)0, so, NULL); 871 break; 872 #endif 873 default: 874 error = EAFNOSUPPORT; 875 break; 876 } 877 878 return (error); 879 } 880 881 /* 882 * Create a new TCP control block, making an 883 * empty reassembly queue and hooking it to the argument 884 * protocol control block. 885 */ 886 struct tcpcb * 887 tcp_newtcpcb(family, aux) 888 int family; /* selects inpcb, or in6pcb */ 889 void *aux; 890 { 891 struct tcpcb *tp; 892 int i; 893 894 switch (family) { 895 case PF_INET: 896 break; 897 #ifdef INET6 898 case PF_INET6: 899 break; 900 #endif 901 default: 902 return NULL; 903 } 904 905 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 906 if (tp == NULL) 907 return (NULL); 908 bzero((caddr_t)tp, sizeof(struct tcpcb)); 909 TAILQ_INIT(&tp->segq); 910 TAILQ_INIT(&tp->timeq); 911 tp->t_family = family; /* may be overridden later on */ 912 tp->t_peermss = tcp_mssdflt; 913 tp->t_ourmss = tcp_mssdflt; 914 tp->t_segsz = tcp_mssdflt; 915 LIST_INIT(&tp->t_sc); 916 917 tp->t_lastm = NULL; 918 tp->t_lastoff = 0; 919 920 callout_init(&tp->t_delack_ch); 921 for (i = 0; i < TCPT_NTIMERS; i++) 922 TCP_TIMER_INIT(tp, i); 923 924 tp->t_flags = 0; 925 if (tcp_do_rfc1323 && tcp_do_win_scale) 926 tp->t_flags |= TF_REQ_SCALE; 927 if (tcp_do_rfc1323 && tcp_do_timestamps) 928 tp->t_flags |= TF_REQ_TSTMP; 929 if (tcp_do_sack == 2) 930 tp->t_flags |= TF_WILL_SACK; 931 else if (tcp_do_sack == 1) 932 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 933 tp->t_flags |= TF_CANT_TXSACK; 934 switch (family) { 935 case PF_INET: 936 tp->t_inpcb = (struct inpcb *)aux; 937 tp->t_mtudisc = ip_mtudisc; 938 break; 939 #ifdef INET6 940 case PF_INET6: 941 tp->t_in6pcb = (struct in6pcb *)aux; 942 /* for IPv6, always try to run path MTU discovery */ 943 tp->t_mtudisc = 1; 944 break; 945 #endif 946 } 947 /* 948 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 949 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 950 * reasonable initial retransmit time. 951 */ 952 tp->t_srtt = TCPTV_SRTTBASE; 953 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 954 tp->t_rttmin = TCPTV_MIN; 955 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 956 TCPTV_MIN, TCPTV_REXMTMAX); 957 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 958 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 959 if (family == AF_INET) { 960 struct inpcb *inp = (struct inpcb *)aux; 961 inp->inp_ip.ip_ttl = ip_defttl; 962 inp->inp_ppcb = (caddr_t)tp; 963 } 964 #ifdef INET6 965 else if (family == AF_INET6) { 966 struct in6pcb *in6p = (struct in6pcb *)aux; 967 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 968 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 969 : NULL); 970 in6p->in6p_ppcb = (caddr_t)tp; 971 } 972 #endif 973 974 /* 975 * Initialize our timebase. When we send timestamps, we take 976 * the delta from tcp_now -- this means each connection always 977 * gets a timebase of 0, which makes it, among other things, 978 * more difficult to determine how long a system has been up, 979 * and thus how many TCP sequence increments have occurred. 980 */ 981 tp->ts_timebase = tcp_now; 982 983 return (tp); 984 } 985 986 /* 987 * Drop a TCP connection, reporting 988 * the specified error. If connection is synchronized, 989 * then send a RST to peer. 990 */ 991 struct tcpcb * 992 tcp_drop(tp, errno) 993 struct tcpcb *tp; 994 int errno; 995 { 996 struct socket *so = NULL; 997 998 #ifdef DIAGNOSTIC 999 if (tp->t_inpcb && tp->t_in6pcb) 1000 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1001 #endif 1002 #ifdef INET 1003 if (tp->t_inpcb) 1004 so = tp->t_inpcb->inp_socket; 1005 #endif 1006 #ifdef INET6 1007 if (tp->t_in6pcb) 1008 so = tp->t_in6pcb->in6p_socket; 1009 #endif 1010 if (!so) 1011 return NULL; 1012 1013 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1014 tp->t_state = TCPS_CLOSED; 1015 (void) tcp_output(tp); 1016 tcpstat.tcps_drops++; 1017 } else 1018 tcpstat.tcps_conndrops++; 1019 if (errno == ETIMEDOUT && tp->t_softerror) 1020 errno = tp->t_softerror; 1021 so->so_error = errno; 1022 return (tcp_close(tp)); 1023 } 1024 1025 /* 1026 * Return whether this tcpcb is marked as dead, indicating 1027 * to the calling timer function that no further action should 1028 * be taken, as we are about to release this tcpcb. The release 1029 * of the storage will be done if this is the last timer running. 1030 * 1031 * This is typically called from the callout handler function before 1032 * callout_ack() is done, therefore we need to test the number of 1033 * running timer functions against 1 below, not 0. 1034 */ 1035 int 1036 tcp_isdead(tp) 1037 struct tcpcb *tp; 1038 { 1039 int dead = (tp->t_flags & TF_DEAD); 1040 1041 if (__predict_false(dead)) { 1042 if (tcp_timers_invoking(tp) > 1) 1043 /* not quite there yet -- count separately? */ 1044 return dead; 1045 tcpstat.tcps_delayed_free++; 1046 pool_put(&tcpcb_pool, tp); 1047 } 1048 return dead; 1049 } 1050 1051 /* 1052 * Close a TCP control block: 1053 * discard all space held by the tcp 1054 * discard internet protocol block 1055 * wake up any sleepers 1056 */ 1057 struct tcpcb * 1058 tcp_close(tp) 1059 struct tcpcb *tp; 1060 { 1061 struct inpcb *inp; 1062 #ifdef INET6 1063 struct in6pcb *in6p; 1064 #endif 1065 struct socket *so; 1066 #ifdef RTV_RTT 1067 struct rtentry *rt; 1068 #endif 1069 struct route *ro; 1070 1071 inp = tp->t_inpcb; 1072 #ifdef INET6 1073 in6p = tp->t_in6pcb; 1074 #endif 1075 so = NULL; 1076 ro = NULL; 1077 if (inp) { 1078 so = inp->inp_socket; 1079 ro = &inp->inp_route; 1080 } 1081 #ifdef INET6 1082 else if (in6p) { 1083 so = in6p->in6p_socket; 1084 ro = (struct route *)&in6p->in6p_route; 1085 } 1086 #endif 1087 1088 #ifdef RTV_RTT 1089 /* 1090 * If we sent enough data to get some meaningful characteristics, 1091 * save them in the routing entry. 'Enough' is arbitrarily 1092 * defined as the sendpipesize (default 4K) * 16. This would 1093 * give us 16 rtt samples assuming we only get one sample per 1094 * window (the usual case on a long haul net). 16 samples is 1095 * enough for the srtt filter to converge to within 5% of the correct 1096 * value; fewer samples and we could save a very bogus rtt. 1097 * 1098 * Don't update the default route's characteristics and don't 1099 * update anything that the user "locked". 1100 */ 1101 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1102 ro && (rt = ro->ro_rt) && 1103 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1104 u_long i = 0; 1105 1106 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1107 i = tp->t_srtt * 1108 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1109 if (rt->rt_rmx.rmx_rtt && i) 1110 /* 1111 * filter this update to half the old & half 1112 * the new values, converting scale. 1113 * See route.h and tcp_var.h for a 1114 * description of the scaling constants. 1115 */ 1116 rt->rt_rmx.rmx_rtt = 1117 (rt->rt_rmx.rmx_rtt + i) / 2; 1118 else 1119 rt->rt_rmx.rmx_rtt = i; 1120 } 1121 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1122 i = tp->t_rttvar * 1123 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1124 if (rt->rt_rmx.rmx_rttvar && i) 1125 rt->rt_rmx.rmx_rttvar = 1126 (rt->rt_rmx.rmx_rttvar + i) / 2; 1127 else 1128 rt->rt_rmx.rmx_rttvar = i; 1129 } 1130 /* 1131 * update the pipelimit (ssthresh) if it has been updated 1132 * already or if a pipesize was specified & the threshhold 1133 * got below half the pipesize. I.e., wait for bad news 1134 * before we start updating, then update on both good 1135 * and bad news. 1136 */ 1137 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1138 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1139 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1140 /* 1141 * convert the limit from user data bytes to 1142 * packets then to packet data bytes. 1143 */ 1144 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1145 if (i < 2) 1146 i = 2; 1147 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1148 if (rt->rt_rmx.rmx_ssthresh) 1149 rt->rt_rmx.rmx_ssthresh = 1150 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1151 else 1152 rt->rt_rmx.rmx_ssthresh = i; 1153 } 1154 } 1155 #endif /* RTV_RTT */ 1156 /* free the reassembly queue, if any */ 1157 TCP_REASS_LOCK(tp); 1158 (void) tcp_freeq(tp); 1159 TCP_REASS_UNLOCK(tp); 1160 1161 tcp_canceltimers(tp); 1162 TCP_CLEAR_DELACK(tp); 1163 syn_cache_cleanup(tp); 1164 1165 if (tp->t_template) { 1166 m_free(tp->t_template); 1167 tp->t_template = NULL; 1168 } 1169 if (tcp_timers_invoking(tp)) 1170 tp->t_flags |= TF_DEAD; 1171 else 1172 pool_put(&tcpcb_pool, tp); 1173 1174 if (inp) { 1175 inp->inp_ppcb = 0; 1176 soisdisconnected(so); 1177 in_pcbdetach(inp); 1178 } 1179 #ifdef INET6 1180 else if (in6p) { 1181 in6p->in6p_ppcb = 0; 1182 soisdisconnected(so); 1183 in6_pcbdetach(in6p); 1184 } 1185 #endif 1186 tcpstat.tcps_closed++; 1187 return ((struct tcpcb *)0); 1188 } 1189 1190 int 1191 tcp_freeq(tp) 1192 struct tcpcb *tp; 1193 { 1194 struct ipqent *qe; 1195 int rv = 0; 1196 #ifdef TCPREASS_DEBUG 1197 int i = 0; 1198 #endif 1199 1200 TCP_REASS_LOCK_CHECK(tp); 1201 1202 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1203 #ifdef TCPREASS_DEBUG 1204 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1205 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1206 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1207 #endif 1208 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1209 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1210 m_freem(qe->ipqe_m); 1211 pool_put(&ipqent_pool, qe); 1212 rv = 1; 1213 } 1214 return (rv); 1215 } 1216 1217 /* 1218 * Protocol drain routine. Called when memory is in short supply. 1219 */ 1220 void 1221 tcp_drain() 1222 { 1223 struct inpcb_hdr *inph; 1224 struct inpcb *inp; 1225 struct tcpcb *tp; 1226 1227 /* 1228 * Free the sequence queue of all TCP connections. 1229 */ 1230 inph = CIRCLEQ_FIRST(&tcbtable.inpt_queue); 1231 if (inp) /* XXX */ 1232 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1233 switch (inph->inph_af) { 1234 case AF_INET: 1235 tp = intotcpcb((struct inpcb *)inph); 1236 break; 1237 #ifdef INET6 1238 case AF_INET6: 1239 tp = in6totcpcb((struct in6pcb *)inph); 1240 break; 1241 #endif 1242 default: 1243 tp = NULL; 1244 break; 1245 } 1246 if (tp != NULL) { 1247 /* 1248 * We may be called from a device's interrupt 1249 * context. If the tcpcb is already busy, 1250 * just bail out now. 1251 */ 1252 if (tcp_reass_lock_try(tp) == 0) 1253 continue; 1254 if (tcp_freeq(tp)) 1255 tcpstat.tcps_connsdrained++; 1256 TCP_REASS_UNLOCK(tp); 1257 } 1258 } 1259 } 1260 1261 /* 1262 * Notify a tcp user of an asynchronous error; 1263 * store error as soft error, but wake up user 1264 * (for now, won't do anything until can select for soft error). 1265 */ 1266 void 1267 tcp_notify(inp, error) 1268 struct inpcb *inp; 1269 int error; 1270 { 1271 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1272 struct socket *so = inp->inp_socket; 1273 1274 /* 1275 * Ignore some errors if we are hooked up. 1276 * If connection hasn't completed, has retransmitted several times, 1277 * and receives a second error, give up now. This is better 1278 * than waiting a long time to establish a connection that 1279 * can never complete. 1280 */ 1281 if (tp->t_state == TCPS_ESTABLISHED && 1282 (error == EHOSTUNREACH || error == ENETUNREACH || 1283 error == EHOSTDOWN)) { 1284 return; 1285 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1286 tp->t_rxtshift > 3 && tp->t_softerror) 1287 so->so_error = error; 1288 else 1289 tp->t_softerror = error; 1290 wakeup((caddr_t) &so->so_timeo); 1291 sorwakeup(so); 1292 sowwakeup(so); 1293 } 1294 1295 #ifdef INET6 1296 void 1297 tcp6_notify(in6p, error) 1298 struct in6pcb *in6p; 1299 int error; 1300 { 1301 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1302 struct socket *so = in6p->in6p_socket; 1303 1304 /* 1305 * Ignore some errors if we are hooked up. 1306 * If connection hasn't completed, has retransmitted several times, 1307 * and receives a second error, give up now. This is better 1308 * than waiting a long time to establish a connection that 1309 * can never complete. 1310 */ 1311 if (tp->t_state == TCPS_ESTABLISHED && 1312 (error == EHOSTUNREACH || error == ENETUNREACH || 1313 error == EHOSTDOWN)) { 1314 return; 1315 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1316 tp->t_rxtshift > 3 && tp->t_softerror) 1317 so->so_error = error; 1318 else 1319 tp->t_softerror = error; 1320 wakeup((caddr_t) &so->so_timeo); 1321 sorwakeup(so); 1322 sowwakeup(so); 1323 } 1324 #endif 1325 1326 #ifdef INET6 1327 void 1328 tcp6_ctlinput(cmd, sa, d) 1329 int cmd; 1330 struct sockaddr *sa; 1331 void *d; 1332 { 1333 struct tcphdr th; 1334 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1335 int nmatch; 1336 struct ip6_hdr *ip6; 1337 const struct sockaddr_in6 *sa6_src = NULL; 1338 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1339 struct mbuf *m; 1340 int off; 1341 1342 if (sa->sa_family != AF_INET6 || 1343 sa->sa_len != sizeof(struct sockaddr_in6)) 1344 return; 1345 if ((unsigned)cmd >= PRC_NCMDS) 1346 return; 1347 else if (cmd == PRC_QUENCH) { 1348 /* XXX there's no PRC_QUENCH in IPv6 */ 1349 notify = tcp6_quench; 1350 } else if (PRC_IS_REDIRECT(cmd)) 1351 notify = in6_rtchange, d = NULL; 1352 else if (cmd == PRC_MSGSIZE) 1353 ; /* special code is present, see below */ 1354 else if (cmd == PRC_HOSTDEAD) 1355 d = NULL; 1356 else if (inet6ctlerrmap[cmd] == 0) 1357 return; 1358 1359 /* if the parameter is from icmp6, decode it. */ 1360 if (d != NULL) { 1361 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1362 m = ip6cp->ip6c_m; 1363 ip6 = ip6cp->ip6c_ip6; 1364 off = ip6cp->ip6c_off; 1365 sa6_src = ip6cp->ip6c_src; 1366 } else { 1367 m = NULL; 1368 ip6 = NULL; 1369 sa6_src = &sa6_any; 1370 } 1371 1372 if (ip6) { 1373 /* 1374 * XXX: We assume that when ip6 is non NULL, 1375 * M and OFF are valid. 1376 */ 1377 1378 /* check if we can safely examine src and dst ports */ 1379 if (m->m_pkthdr.len < off + sizeof(th)) { 1380 if (cmd == PRC_MSGSIZE) 1381 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1382 return; 1383 } 1384 1385 bzero(&th, sizeof(th)); 1386 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1387 1388 if (cmd == PRC_MSGSIZE) { 1389 int valid = 0; 1390 1391 /* 1392 * Check to see if we have a valid TCP connection 1393 * corresponding to the address in the ICMPv6 message 1394 * payload. 1395 */ 1396 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1397 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1398 th.th_sport, 0)) 1399 valid++; 1400 1401 /* 1402 * Depending on the value of "valid" and routing table 1403 * size (mtudisc_{hi,lo}wat), we will: 1404 * - recalcurate the new MTU and create the 1405 * corresponding routing entry, or 1406 * - ignore the MTU change notification. 1407 */ 1408 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1409 1410 /* 1411 * no need to call in6_pcbnotify, it should have been 1412 * called via callback if necessary 1413 */ 1414 return; 1415 } 1416 1417 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1418 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1419 if (nmatch == 0 && syn_cache_count && 1420 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1421 inet6ctlerrmap[cmd] == ENETUNREACH || 1422 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1423 syn_cache_unreach((struct sockaddr *)sa6_src, 1424 sa, &th); 1425 } else { 1426 (void) in6_pcbnotify(&tcbtable, sa, 0, 1427 (struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1428 } 1429 } 1430 #endif 1431 1432 #ifdef INET 1433 /* assumes that ip header and tcp header are contiguous on mbuf */ 1434 void * 1435 tcp_ctlinput(cmd, sa, v) 1436 int cmd; 1437 struct sockaddr *sa; 1438 void *v; 1439 { 1440 struct ip *ip = v; 1441 struct tcphdr *th; 1442 struct icmp *icp; 1443 extern const int inetctlerrmap[]; 1444 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1445 int errno; 1446 int nmatch; 1447 #ifdef INET6 1448 struct in6_addr src6, dst6; 1449 #endif 1450 1451 if (sa->sa_family != AF_INET || 1452 sa->sa_len != sizeof(struct sockaddr_in)) 1453 return NULL; 1454 if ((unsigned)cmd >= PRC_NCMDS) 1455 return NULL; 1456 errno = inetctlerrmap[cmd]; 1457 if (cmd == PRC_QUENCH) 1458 notify = tcp_quench; 1459 else if (PRC_IS_REDIRECT(cmd)) 1460 notify = in_rtchange, ip = 0; 1461 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1462 /* 1463 * Check to see if we have a valid TCP connection 1464 * corresponding to the address in the ICMP message 1465 * payload. 1466 * 1467 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1468 */ 1469 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1470 #ifdef INET6 1471 memset(&src6, 0, sizeof(src6)); 1472 memset(&dst6, 0, sizeof(dst6)); 1473 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1474 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1475 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1476 #endif 1477 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1478 ip->ip_src, th->th_sport) != NULL) 1479 ; 1480 #ifdef INET6 1481 else if (in6_pcblookup_connect(&tcbtable, &dst6, 1482 th->th_dport, &src6, th->th_sport, 0) != NULL) 1483 ; 1484 #endif 1485 else 1486 return NULL; 1487 1488 /* 1489 * Now that we've validated that we are actually communicating 1490 * with the host indicated in the ICMP message, locate the 1491 * ICMP header, recalculate the new MTU, and create the 1492 * corresponding routing entry. 1493 */ 1494 icp = (struct icmp *)((caddr_t)ip - 1495 offsetof(struct icmp, icmp_ip)); 1496 icmp_mtudisc(icp, ip->ip_dst); 1497 1498 return NULL; 1499 } else if (cmd == PRC_HOSTDEAD) 1500 ip = 0; 1501 else if (errno == 0) 1502 return NULL; 1503 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1504 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1505 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1506 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1507 if (nmatch == 0 && syn_cache_count && 1508 (inetctlerrmap[cmd] == EHOSTUNREACH || 1509 inetctlerrmap[cmd] == ENETUNREACH || 1510 inetctlerrmap[cmd] == EHOSTDOWN)) { 1511 struct sockaddr_in sin; 1512 bzero(&sin, sizeof(sin)); 1513 sin.sin_len = sizeof(sin); 1514 sin.sin_family = AF_INET; 1515 sin.sin_port = th->th_sport; 1516 sin.sin_addr = ip->ip_src; 1517 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1518 } 1519 1520 /* XXX mapped address case */ 1521 } else 1522 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1523 notify); 1524 return NULL; 1525 } 1526 1527 /* 1528 * When a source quence is received, we are being notifed of congestion. 1529 * Close the congestion window down to the Loss Window (one segment). 1530 * We will gradually open it again as we proceed. 1531 */ 1532 void 1533 tcp_quench(inp, errno) 1534 struct inpcb *inp; 1535 int errno; 1536 { 1537 struct tcpcb *tp = intotcpcb(inp); 1538 1539 if (tp) 1540 tp->snd_cwnd = tp->t_segsz; 1541 } 1542 #endif 1543 1544 #ifdef INET6 1545 void 1546 tcp6_quench(in6p, errno) 1547 struct in6pcb *in6p; 1548 int errno; 1549 { 1550 struct tcpcb *tp = in6totcpcb(in6p); 1551 1552 if (tp) 1553 tp->snd_cwnd = tp->t_segsz; 1554 } 1555 #endif 1556 1557 #ifdef INET 1558 /* 1559 * Path MTU Discovery handlers. 1560 */ 1561 void 1562 tcp_mtudisc_callback(faddr) 1563 struct in_addr faddr; 1564 { 1565 #ifdef INET6 1566 struct in6_addr in6; 1567 #endif 1568 1569 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1570 #ifdef INET6 1571 memset(&in6, 0, sizeof(in6)); 1572 in6.s6_addr16[5] = 0xffff; 1573 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1574 tcp6_mtudisc_callback(&in6); 1575 #endif 1576 } 1577 1578 /* 1579 * On receipt of path MTU corrections, flush old route and replace it 1580 * with the new one. Retransmit all unacknowledged packets, to ensure 1581 * that all packets will be received. 1582 */ 1583 void 1584 tcp_mtudisc(inp, errno) 1585 struct inpcb *inp; 1586 int errno; 1587 { 1588 struct tcpcb *tp = intotcpcb(inp); 1589 struct rtentry *rt = in_pcbrtentry(inp); 1590 1591 if (tp != 0) { 1592 if (rt != 0) { 1593 /* 1594 * If this was not a host route, remove and realloc. 1595 */ 1596 if ((rt->rt_flags & RTF_HOST) == 0) { 1597 in_rtchange(inp, errno); 1598 if ((rt = in_pcbrtentry(inp)) == 0) 1599 return; 1600 } 1601 1602 /* 1603 * Slow start out of the error condition. We 1604 * use the MTU because we know it's smaller 1605 * than the previously transmitted segment. 1606 * 1607 * Note: This is more conservative than the 1608 * suggestion in draft-floyd-incr-init-win-03. 1609 */ 1610 if (rt->rt_rmx.rmx_mtu != 0) 1611 tp->snd_cwnd = 1612 TCP_INITIAL_WINDOW(tcp_init_win, 1613 rt->rt_rmx.rmx_mtu); 1614 } 1615 1616 /* 1617 * Resend unacknowledged packets. 1618 */ 1619 tp->snd_nxt = tp->snd_una; 1620 tcp_output(tp); 1621 } 1622 } 1623 #endif 1624 1625 #ifdef INET6 1626 /* 1627 * Path MTU Discovery handlers. 1628 */ 1629 void 1630 tcp6_mtudisc_callback(faddr) 1631 struct in6_addr *faddr; 1632 { 1633 struct sockaddr_in6 sin6; 1634 1635 bzero(&sin6, sizeof(sin6)); 1636 sin6.sin6_family = AF_INET6; 1637 sin6.sin6_len = sizeof(struct sockaddr_in6); 1638 sin6.sin6_addr = *faddr; 1639 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1640 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1641 } 1642 1643 void 1644 tcp6_mtudisc(in6p, errno) 1645 struct in6pcb *in6p; 1646 int errno; 1647 { 1648 struct tcpcb *tp = in6totcpcb(in6p); 1649 struct rtentry *rt = in6_pcbrtentry(in6p); 1650 1651 if (tp != 0) { 1652 if (rt != 0) { 1653 /* 1654 * If this was not a host route, remove and realloc. 1655 */ 1656 if ((rt->rt_flags & RTF_HOST) == 0) { 1657 in6_rtchange(in6p, errno); 1658 if ((rt = in6_pcbrtentry(in6p)) == 0) 1659 return; 1660 } 1661 1662 /* 1663 * Slow start out of the error condition. We 1664 * use the MTU because we know it's smaller 1665 * than the previously transmitted segment. 1666 * 1667 * Note: This is more conservative than the 1668 * suggestion in draft-floyd-incr-init-win-03. 1669 */ 1670 if (rt->rt_rmx.rmx_mtu != 0) 1671 tp->snd_cwnd = 1672 TCP_INITIAL_WINDOW(tcp_init_win, 1673 rt->rt_rmx.rmx_mtu); 1674 } 1675 1676 /* 1677 * Resend unacknowledged packets. 1678 */ 1679 tp->snd_nxt = tp->snd_una; 1680 tcp_output(tp); 1681 } 1682 } 1683 #endif /* INET6 */ 1684 1685 /* 1686 * Compute the MSS to advertise to the peer. Called only during 1687 * the 3-way handshake. If we are the server (peer initiated 1688 * connection), we are called with a pointer to the interface 1689 * on which the SYN packet arrived. If we are the client (we 1690 * initiated connection), we are called with a pointer to the 1691 * interface out which this connection should go. 1692 * 1693 * NOTE: Do not subtract IP option/extension header size nor IPsec 1694 * header size from MSS advertisement. MSS option must hold the maximum 1695 * segment size we can accept, so it must always be: 1696 * max(if mtu) - ip header - tcp header 1697 */ 1698 u_long 1699 tcp_mss_to_advertise(ifp, af) 1700 const struct ifnet *ifp; 1701 int af; 1702 { 1703 extern u_long in_maxmtu; 1704 u_long mss = 0; 1705 u_long hdrsiz; 1706 1707 /* 1708 * In order to avoid defeating path MTU discovery on the peer, 1709 * we advertise the max MTU of all attached networks as our MSS, 1710 * per RFC 1191, section 3.1. 1711 * 1712 * We provide the option to advertise just the MTU of 1713 * the interface on which we hope this connection will 1714 * be receiving. If we are responding to a SYN, we 1715 * will have a pretty good idea about this, but when 1716 * initiating a connection there is a bit more doubt. 1717 * 1718 * We also need to ensure that loopback has a large enough 1719 * MSS, as the loopback MTU is never included in in_maxmtu. 1720 */ 1721 1722 if (ifp != NULL) 1723 switch (af) { 1724 case AF_INET: 1725 mss = ifp->if_mtu; 1726 break; 1727 #ifdef INET6 1728 case AF_INET6: 1729 mss = IN6_LINKMTU(ifp); 1730 break; 1731 #endif 1732 } 1733 1734 if (tcp_mss_ifmtu == 0) 1735 switch (af) { 1736 case AF_INET: 1737 mss = max(in_maxmtu, mss); 1738 break; 1739 #ifdef INET6 1740 case AF_INET6: 1741 mss = max(in6_maxmtu, mss); 1742 break; 1743 #endif 1744 } 1745 1746 switch (af) { 1747 case AF_INET: 1748 hdrsiz = sizeof(struct ip); 1749 break; 1750 #ifdef INET6 1751 case AF_INET6: 1752 hdrsiz = sizeof(struct ip6_hdr); 1753 break; 1754 #endif 1755 default: 1756 hdrsiz = 0; 1757 break; 1758 } 1759 hdrsiz += sizeof(struct tcphdr); 1760 if (mss > hdrsiz) 1761 mss -= hdrsiz; 1762 1763 mss = max(tcp_mssdflt, mss); 1764 return (mss); 1765 } 1766 1767 /* 1768 * Set connection variables based on the peer's advertised MSS. 1769 * We are passed the TCPCB for the actual connection. If we 1770 * are the server, we are called by the compressed state engine 1771 * when the 3-way handshake is complete. If we are the client, 1772 * we are called when we receive the SYN,ACK from the server. 1773 * 1774 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1775 * before this routine is called! 1776 */ 1777 void 1778 tcp_mss_from_peer(tp, offer) 1779 struct tcpcb *tp; 1780 int offer; 1781 { 1782 struct socket *so; 1783 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1784 struct rtentry *rt; 1785 #endif 1786 u_long bufsize; 1787 int mss; 1788 1789 #ifdef DIAGNOSTIC 1790 if (tp->t_inpcb && tp->t_in6pcb) 1791 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1792 #endif 1793 so = NULL; 1794 rt = NULL; 1795 #ifdef INET 1796 if (tp->t_inpcb) { 1797 so = tp->t_inpcb->inp_socket; 1798 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1799 rt = in_pcbrtentry(tp->t_inpcb); 1800 #endif 1801 } 1802 #endif 1803 #ifdef INET6 1804 if (tp->t_in6pcb) { 1805 so = tp->t_in6pcb->in6p_socket; 1806 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1807 rt = in6_pcbrtentry(tp->t_in6pcb); 1808 #endif 1809 } 1810 #endif 1811 1812 /* 1813 * As per RFC1122, use the default MSS value, unless they 1814 * sent us an offer. Do not accept offers less than 32 bytes. 1815 */ 1816 mss = tcp_mssdflt; 1817 if (offer) 1818 mss = offer; 1819 mss = max(mss, 32); /* sanity */ 1820 tp->t_peermss = mss; 1821 mss -= tcp_optlen(tp); 1822 #ifdef INET 1823 if (tp->t_inpcb) 1824 mss -= ip_optlen(tp->t_inpcb); 1825 #endif 1826 #ifdef INET6 1827 if (tp->t_in6pcb) 1828 mss -= ip6_optlen(tp->t_in6pcb); 1829 #endif 1830 1831 /* 1832 * If there's a pipesize, change the socket buffer to that size. 1833 * Make the socket buffer an integral number of MSS units. If 1834 * the MSS is larger than the socket buffer, artificially decrease 1835 * the MSS. 1836 */ 1837 #ifdef RTV_SPIPE 1838 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1839 bufsize = rt->rt_rmx.rmx_sendpipe; 1840 else 1841 #endif 1842 bufsize = so->so_snd.sb_hiwat; 1843 if (bufsize < mss) 1844 mss = bufsize; 1845 else { 1846 bufsize = roundup(bufsize, mss); 1847 if (bufsize > sb_max) 1848 bufsize = sb_max; 1849 (void) sbreserve(&so->so_snd, bufsize); 1850 } 1851 tp->t_segsz = mss; 1852 1853 #ifdef RTV_SSTHRESH 1854 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1855 /* 1856 * There's some sort of gateway or interface buffer 1857 * limit on the path. Use this to set the slow 1858 * start threshold, but set the threshold to no less 1859 * than 2 * MSS. 1860 */ 1861 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1862 } 1863 #endif 1864 } 1865 1866 /* 1867 * Processing necessary when a TCP connection is established. 1868 */ 1869 void 1870 tcp_established(tp) 1871 struct tcpcb *tp; 1872 { 1873 struct socket *so; 1874 #ifdef RTV_RPIPE 1875 struct rtentry *rt; 1876 #endif 1877 u_long bufsize; 1878 1879 #ifdef DIAGNOSTIC 1880 if (tp->t_inpcb && tp->t_in6pcb) 1881 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1882 #endif 1883 so = NULL; 1884 rt = NULL; 1885 #ifdef INET 1886 if (tp->t_inpcb) { 1887 so = tp->t_inpcb->inp_socket; 1888 #if defined(RTV_RPIPE) 1889 rt = in_pcbrtentry(tp->t_inpcb); 1890 #endif 1891 } 1892 #endif 1893 #ifdef INET6 1894 if (tp->t_in6pcb) { 1895 so = tp->t_in6pcb->in6p_socket; 1896 #if defined(RTV_RPIPE) 1897 rt = in6_pcbrtentry(tp->t_in6pcb); 1898 #endif 1899 } 1900 #endif 1901 1902 tp->t_state = TCPS_ESTABLISHED; 1903 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1904 1905 #ifdef RTV_RPIPE 1906 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1907 bufsize = rt->rt_rmx.rmx_recvpipe; 1908 else 1909 #endif 1910 bufsize = so->so_rcv.sb_hiwat; 1911 if (bufsize > tp->t_ourmss) { 1912 bufsize = roundup(bufsize, tp->t_ourmss); 1913 if (bufsize > sb_max) 1914 bufsize = sb_max; 1915 (void) sbreserve(&so->so_rcv, bufsize); 1916 } 1917 } 1918 1919 /* 1920 * Check if there's an initial rtt or rttvar. Convert from the 1921 * route-table units to scaled multiples of the slow timeout timer. 1922 * Called only during the 3-way handshake. 1923 */ 1924 void 1925 tcp_rmx_rtt(tp) 1926 struct tcpcb *tp; 1927 { 1928 #ifdef RTV_RTT 1929 struct rtentry *rt = NULL; 1930 int rtt; 1931 1932 #ifdef DIAGNOSTIC 1933 if (tp->t_inpcb && tp->t_in6pcb) 1934 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1935 #endif 1936 #ifdef INET 1937 if (tp->t_inpcb) 1938 rt = in_pcbrtentry(tp->t_inpcb); 1939 #endif 1940 #ifdef INET6 1941 if (tp->t_in6pcb) 1942 rt = in6_pcbrtentry(tp->t_in6pcb); 1943 #endif 1944 if (rt == NULL) 1945 return; 1946 1947 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1948 /* 1949 * XXX The lock bit for MTU indicates that the value 1950 * is also a minimum value; this is subject to time. 1951 */ 1952 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1953 TCPT_RANGESET(tp->t_rttmin, 1954 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1955 TCPTV_MIN, TCPTV_REXMTMAX); 1956 tp->t_srtt = rtt / 1957 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1958 if (rt->rt_rmx.rmx_rttvar) { 1959 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1960 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1961 (TCP_RTTVAR_SHIFT + 2)); 1962 } else { 1963 /* Default variation is +- 1 rtt */ 1964 tp->t_rttvar = 1965 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1966 } 1967 TCPT_RANGESET(tp->t_rxtcur, 1968 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1969 tp->t_rttmin, TCPTV_REXMTMAX); 1970 } 1971 #endif 1972 } 1973 1974 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1975 #if NRND > 0 1976 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1977 #endif 1978 1979 /* 1980 * Get a new sequence value given a tcp control block 1981 */ 1982 tcp_seq 1983 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1984 { 1985 1986 #ifdef INET 1987 if (tp->t_inpcb != NULL) { 1988 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1989 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1990 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1991 addin)); 1992 } 1993 #endif 1994 #ifdef INET6 1995 if (tp->t_in6pcb != NULL) { 1996 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 1997 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 1998 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 1999 addin)); 2000 } 2001 #endif 2002 /* Not possible. */ 2003 panic("tcp_new_iss"); 2004 } 2005 2006 /* 2007 * This routine actually generates a new TCP initial sequence number. 2008 */ 2009 tcp_seq 2010 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2011 size_t addrsz, tcp_seq addin) 2012 { 2013 tcp_seq tcp_iss; 2014 2015 #if NRND > 0 2016 static int beenhere; 2017 2018 /* 2019 * If we haven't been here before, initialize our cryptographic 2020 * hash secret. 2021 */ 2022 if (beenhere == 0) { 2023 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2024 RND_EXTRACT_ANY); 2025 beenhere = 1; 2026 } 2027 2028 if (tcp_do_rfc1948) { 2029 MD5_CTX ctx; 2030 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2031 2032 /* 2033 * Compute the base value of the ISS. It is a hash 2034 * of (saddr, sport, daddr, dport, secret). 2035 */ 2036 MD5Init(&ctx); 2037 2038 MD5Update(&ctx, (u_char *) laddr, addrsz); 2039 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2040 2041 MD5Update(&ctx, (u_char *) faddr, addrsz); 2042 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2043 2044 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2045 2046 MD5Final(hash, &ctx); 2047 2048 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2049 2050 /* 2051 * Now increment our "timer", and add it in to 2052 * the computed value. 2053 * 2054 * XXX Use `addin'? 2055 * XXX TCP_ISSINCR too large to use? 2056 */ 2057 tcp_iss_seq += TCP_ISSINCR; 2058 #ifdef TCPISS_DEBUG 2059 printf("ISS hash 0x%08x, ", tcp_iss); 2060 #endif 2061 tcp_iss += tcp_iss_seq + addin; 2062 #ifdef TCPISS_DEBUG 2063 printf("new ISS 0x%08x\n", tcp_iss); 2064 #endif 2065 } else 2066 #endif /* NRND > 0 */ 2067 { 2068 /* 2069 * Randomize. 2070 */ 2071 #if NRND > 0 2072 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2073 #else 2074 tcp_iss = arc4random(); 2075 #endif 2076 2077 /* 2078 * If we were asked to add some amount to a known value, 2079 * we will take a random value obtained above, mask off 2080 * the upper bits, and add in the known value. We also 2081 * add in a constant to ensure that we are at least a 2082 * certain distance from the original value. 2083 * 2084 * This is used when an old connection is in timed wait 2085 * and we have a new one coming in, for instance. 2086 */ 2087 if (addin != 0) { 2088 #ifdef TCPISS_DEBUG 2089 printf("Random %08x, ", tcp_iss); 2090 #endif 2091 tcp_iss &= TCP_ISS_RANDOM_MASK; 2092 tcp_iss += addin + TCP_ISSINCR; 2093 #ifdef TCPISS_DEBUG 2094 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2095 #endif 2096 } else { 2097 tcp_iss &= TCP_ISS_RANDOM_MASK; 2098 tcp_iss += tcp_iss_seq; 2099 tcp_iss_seq += TCP_ISSINCR; 2100 #ifdef TCPISS_DEBUG 2101 printf("ISS %08x\n", tcp_iss); 2102 #endif 2103 } 2104 } 2105 2106 if (tcp_compat_42) { 2107 /* 2108 * Limit it to the positive range for really old TCP 2109 * implementations. 2110 * Just AND off the top bit instead of checking if 2111 * is set first - saves a branch 50% of the time. 2112 */ 2113 tcp_iss &= 0x7fffffff; /* XXX */ 2114 } 2115 2116 return (tcp_iss); 2117 } 2118 2119 #if defined(IPSEC) || defined(FAST_IPSEC) 2120 /* compute ESP/AH header size for TCP, including outer IP header. */ 2121 size_t 2122 ipsec4_hdrsiz_tcp(tp) 2123 struct tcpcb *tp; 2124 { 2125 struct inpcb *inp; 2126 size_t hdrsiz; 2127 2128 /* XXX mapped addr case (tp->t_in6pcb) */ 2129 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2130 return 0; 2131 switch (tp->t_family) { 2132 case AF_INET: 2133 /* XXX: should use currect direction. */ 2134 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2135 break; 2136 default: 2137 hdrsiz = 0; 2138 break; 2139 } 2140 2141 return hdrsiz; 2142 } 2143 2144 #ifdef INET6 2145 size_t 2146 ipsec6_hdrsiz_tcp(tp) 2147 struct tcpcb *tp; 2148 { 2149 struct in6pcb *in6p; 2150 size_t hdrsiz; 2151 2152 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2153 return 0; 2154 switch (tp->t_family) { 2155 case AF_INET6: 2156 /* XXX: should use currect direction. */ 2157 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2158 break; 2159 case AF_INET: 2160 /* mapped address case - tricky */ 2161 default: 2162 hdrsiz = 0; 2163 break; 2164 } 2165 2166 return hdrsiz; 2167 } 2168 #endif 2169 #endif /*IPSEC*/ 2170 2171 /* 2172 * Determine the length of the TCP options for this connection. 2173 * 2174 * XXX: What do we do for SACK, when we add that? Just reserve 2175 * all of the space? Otherwise we can't exactly be incrementing 2176 * cwnd by an amount that varies depending on the amount we last 2177 * had to SACK! 2178 */ 2179 2180 u_int 2181 tcp_optlen(tp) 2182 struct tcpcb *tp; 2183 { 2184 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2185 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2186 return TCPOLEN_TSTAMP_APPA; 2187 else 2188 return 0; 2189 } 2190
Indexes created Sun Sep 21 20:09:37 GMT 2025