tcp_subr.c revision 1.152
1 /* $NetBSD: tcp_subr.c,v 1.152 2003/09/06 03:36:31 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 1.152 2003/09/06 03:36:31 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 ipov = mtod(m, struct ipovly *); 459 ipov->ih_pr = IPPROTO_TCP; 460 ipov->ih_len = htons(sizeof(struct tcphdr)); 461 if (inp) { 462 ipov->ih_src = inp->inp_laddr; 463 ipov->ih_dst = inp->inp_faddr; 464 } 465 #ifdef INET6 466 else if (in6p) { 467 /* mapped addr case */ 468 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 469 sizeof(ipov->ih_src)); 470 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 471 sizeof(ipov->ih_dst)); 472 } 473 #endif 474 /* 475 * Compute the pseudo-header portion of the checksum 476 * now. We incrementally add in the TCP option and 477 * payload lengths later, and then compute the TCP 478 * checksum right before the packet is sent off onto 479 * the wire. 480 */ 481 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 482 ipov->ih_dst.s_addr, 483 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 484 break; 485 } 486 #ifdef INET6 487 case AF_INET6: 488 { 489 struct ip6_hdr *ip6; 490 mtod(m, struct ip *)->ip_v = 6; 491 ip6 = mtod(m, struct ip6_hdr *); 492 ip6->ip6_nxt = IPPROTO_TCP; 493 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 494 ip6->ip6_src = in6p->in6p_laddr; 495 ip6->ip6_dst = in6p->in6p_faddr; 496 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 497 if (ip6_auto_flowlabel) { 498 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 499 ip6->ip6_flow |= 500 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 501 } 502 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 503 ip6->ip6_vfc |= IPV6_VERSION; 504 505 /* 506 * Compute the pseudo-header portion of the checksum 507 * now. We incrementally add in the TCP option and 508 * payload lengths later, and then compute the TCP 509 * checksum right before the packet is sent off onto 510 * the wire. 511 */ 512 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 513 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 514 htonl(IPPROTO_TCP)); 515 break; 516 } 517 #endif 518 } 519 if (inp) { 520 n->th_sport = inp->inp_lport; 521 n->th_dport = inp->inp_fport; 522 } 523 #ifdef INET6 524 else if (in6p) { 525 n->th_sport = in6p->in6p_lport; 526 n->th_dport = in6p->in6p_fport; 527 } 528 #endif 529 n->th_seq = 0; 530 n->th_ack = 0; 531 n->th_x2 = 0; 532 n->th_off = 5; 533 n->th_flags = 0; 534 n->th_win = 0; 535 n->th_urp = 0; 536 return (m); 537 } 538 539 /* 540 * Send a single message to the TCP at address specified by 541 * the given TCP/IP header. If m == 0, then we make a copy 542 * of the tcpiphdr at ti and send directly to the addressed host. 543 * This is used to force keep alive messages out using the TCP 544 * template for a connection tp->t_template. If flags are given 545 * then we send a message back to the TCP which originated the 546 * segment ti, and discard the mbuf containing it and any other 547 * attached mbufs. 548 * 549 * In any case the ack and sequence number of the transmitted 550 * segment are as specified by the parameters. 551 */ 552 int 553 tcp_respond(tp, template, m, th0, ack, seq, flags) 554 struct tcpcb *tp; 555 struct mbuf *template; 556 struct mbuf *m; 557 struct tcphdr *th0; 558 tcp_seq ack, seq; 559 int flags; 560 { 561 struct route *ro; 562 int error, tlen, win = 0; 563 int hlen; 564 struct ip *ip; 565 #ifdef INET6 566 struct ip6_hdr *ip6; 567 #endif 568 int family; /* family on packet, not inpcb/in6pcb! */ 569 struct tcphdr *th; 570 struct socket *so; 571 572 if (tp != NULL && (flags & TH_RST) == 0) { 573 #ifdef DIAGNOSTIC 574 if (tp->t_inpcb && tp->t_in6pcb) 575 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 576 #endif 577 #ifdef INET 578 if (tp->t_inpcb) 579 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 580 #endif 581 #ifdef INET6 582 if (tp->t_in6pcb) 583 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 584 #endif 585 } 586 587 th = NULL; /* Quell uninitialized warning */ 588 ip = NULL; 589 #ifdef INET6 590 ip6 = NULL; 591 #endif 592 if (m == 0) { 593 if (!template) 594 return EINVAL; 595 596 /* get family information from template */ 597 switch (mtod(template, struct ip *)->ip_v) { 598 case 4: 599 family = AF_INET; 600 hlen = sizeof(struct ip); 601 break; 602 #ifdef INET6 603 case 6: 604 family = AF_INET6; 605 hlen = sizeof(struct ip6_hdr); 606 break; 607 #endif 608 default: 609 return EAFNOSUPPORT; 610 } 611 612 MGETHDR(m, M_DONTWAIT, MT_HEADER); 613 if (m) { 614 MCLAIM(m, &tcp_tx_mowner); 615 MCLGET(m, M_DONTWAIT); 616 if ((m->m_flags & M_EXT) == 0) { 617 m_free(m); 618 m = NULL; 619 } 620 } 621 if (m == NULL) 622 return (ENOBUFS); 623 624 if (tcp_compat_42) 625 tlen = 1; 626 else 627 tlen = 0; 628 629 m->m_data += max_linkhdr; 630 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 631 template->m_len); 632 switch (family) { 633 case AF_INET: 634 ip = mtod(m, struct ip *); 635 th = (struct tcphdr *)(ip + 1); 636 break; 637 #ifdef INET6 638 case AF_INET6: 639 ip6 = mtod(m, struct ip6_hdr *); 640 th = (struct tcphdr *)(ip6 + 1); 641 break; 642 #endif 643 #if 0 644 default: 645 /* noone will visit here */ 646 m_freem(m); 647 return EAFNOSUPPORT; 648 #endif 649 } 650 flags = TH_ACK; 651 } else { 652 653 if ((m->m_flags & M_PKTHDR) == 0) { 654 #if 0 655 printf("non PKTHDR to tcp_respond\n"); 656 #endif 657 m_freem(m); 658 return EINVAL; 659 } 660 #ifdef DIAGNOSTIC 661 if (!th0) 662 panic("th0 == NULL in tcp_respond"); 663 #endif 664 665 /* get family information from m */ 666 switch (mtod(m, struct ip *)->ip_v) { 667 case 4: 668 family = AF_INET; 669 hlen = sizeof(struct ip); 670 ip = mtod(m, struct ip *); 671 break; 672 #ifdef INET6 673 case 6: 674 family = AF_INET6; 675 hlen = sizeof(struct ip6_hdr); 676 ip6 = mtod(m, struct ip6_hdr *); 677 break; 678 #endif 679 default: 680 m_freem(m); 681 return EAFNOSUPPORT; 682 } 683 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 684 tlen = sizeof(*th0); 685 else 686 tlen = th0->th_off << 2; 687 688 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 689 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 690 m->m_len = hlen + tlen; 691 m_freem(m->m_next); 692 m->m_next = NULL; 693 } else { 694 struct mbuf *n; 695 696 #ifdef DIAGNOSTIC 697 if (max_linkhdr + hlen + tlen > MCLBYTES) { 698 m_freem(m); 699 return EMSGSIZE; 700 } 701 #endif 702 MGETHDR(n, M_DONTWAIT, MT_HEADER); 703 if (n && max_linkhdr + hlen + tlen > MHLEN) { 704 MCLGET(n, M_DONTWAIT); 705 if ((n->m_flags & M_EXT) == 0) { 706 m_freem(n); 707 n = NULL; 708 } 709 } 710 if (!n) { 711 m_freem(m); 712 return ENOBUFS; 713 } 714 715 MCLAIM(n, &tcp_tx_mowner); 716 n->m_data += max_linkhdr; 717 n->m_len = hlen + tlen; 718 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 719 m_copyback(n, hlen, tlen, (caddr_t)th0); 720 721 m_freem(m); 722 m = n; 723 n = NULL; 724 } 725 726 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 727 switch (family) { 728 case AF_INET: 729 ip = mtod(m, struct ip *); 730 th = (struct tcphdr *)(ip + 1); 731 ip->ip_p = IPPROTO_TCP; 732 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 733 ip->ip_p = IPPROTO_TCP; 734 break; 735 #ifdef INET6 736 case AF_INET6: 737 ip6 = mtod(m, struct ip6_hdr *); 738 th = (struct tcphdr *)(ip6 + 1); 739 ip6->ip6_nxt = IPPROTO_TCP; 740 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 741 ip6->ip6_nxt = IPPROTO_TCP; 742 break; 743 #endif 744 #if 0 745 default: 746 /* noone will visit here */ 747 m_freem(m); 748 return EAFNOSUPPORT; 749 #endif 750 } 751 xchg(th->th_dport, th->th_sport, u_int16_t); 752 #undef xchg 753 tlen = 0; /*be friendly with the following code*/ 754 } 755 th->th_seq = htonl(seq); 756 th->th_ack = htonl(ack); 757 th->th_x2 = 0; 758 if ((flags & TH_SYN) == 0) { 759 if (tp) 760 win >>= tp->rcv_scale; 761 if (win > TCP_MAXWIN) 762 win = TCP_MAXWIN; 763 th->th_win = htons((u_int16_t)win); 764 th->th_off = sizeof (struct tcphdr) >> 2; 765 tlen += sizeof(*th); 766 } else 767 tlen += th->th_off << 2; 768 m->m_len = hlen + tlen; 769 m->m_pkthdr.len = hlen + tlen; 770 m->m_pkthdr.rcvif = (struct ifnet *) 0; 771 th->th_flags = flags; 772 th->th_urp = 0; 773 774 switch (family) { 775 #ifdef INET 776 case AF_INET: 777 { 778 struct ipovly *ipov = (struct ipovly *)ip; 779 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 780 ipov->ih_len = htons((u_int16_t)tlen); 781 782 th->th_sum = 0; 783 th->th_sum = in_cksum(m, hlen + tlen); 784 ip->ip_len = htons(hlen + tlen); 785 ip->ip_ttl = ip_defttl; 786 break; 787 } 788 #endif 789 #ifdef INET6 790 case AF_INET6: 791 { 792 th->th_sum = 0; 793 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 794 tlen); 795 ip6->ip6_plen = ntohs(tlen); 796 if (tp && tp->t_in6pcb) { 797 struct ifnet *oifp; 798 ro = (struct route *)&tp->t_in6pcb->in6p_route; 799 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 800 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 801 } else 802 ip6->ip6_hlim = ip6_defhlim; 803 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 804 if (ip6_auto_flowlabel) { 805 ip6->ip6_flow |= 806 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 807 } 808 break; 809 } 810 #endif 811 } 812 813 if (tp && tp->t_inpcb) 814 so = tp->t_inpcb->inp_socket; 815 #ifdef INET6 816 else if (tp && tp->t_in6pcb) 817 so = tp->t_in6pcb->in6p_socket; 818 #endif 819 else 820 so = NULL; 821 822 if (tp != NULL && tp->t_inpcb != NULL) { 823 ro = &tp->t_inpcb->inp_route; 824 #ifdef DIAGNOSTIC 825 if (family != AF_INET) 826 panic("tcp_respond: address family mismatch"); 827 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 828 panic("tcp_respond: ip_dst %x != inp_faddr %x", 829 ntohl(ip->ip_dst.s_addr), 830 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 831 } 832 #endif 833 } 834 #ifdef INET6 835 else if (tp != NULL && tp->t_in6pcb != NULL) { 836 ro = (struct route *)&tp->t_in6pcb->in6p_route; 837 #ifdef DIAGNOSTIC 838 if (family == AF_INET) { 839 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 840 panic("tcp_respond: not mapped addr"); 841 if (bcmp(&ip->ip_dst, 842 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 843 sizeof(ip->ip_dst)) != 0) { 844 panic("tcp_respond: ip_dst != in6p_faddr"); 845 } 846 } else if (family == AF_INET6) { 847 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 848 &tp->t_in6pcb->in6p_faddr)) 849 panic("tcp_respond: ip6_dst != in6p_faddr"); 850 } else 851 panic("tcp_respond: address family mismatch"); 852 #endif 853 } 854 #endif 855 else 856 ro = NULL; 857 858 switch (family) { 859 #ifdef INET 860 case AF_INET: 861 error = ip_output(m, NULL, ro, 862 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 863 (struct ip_moptions *)0, so); 864 break; 865 #endif 866 #ifdef INET6 867 case AF_INET6: 868 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 869 (struct ip6_moptions *)0, so, NULL); 870 break; 871 #endif 872 default: 873 error = EAFNOSUPPORT; 874 break; 875 } 876 877 return (error); 878 } 879 880 /* 881 * Create a new TCP control block, making an 882 * empty reassembly queue and hooking it to the argument 883 * protocol control block. 884 */ 885 struct tcpcb * 886 tcp_newtcpcb(family, aux) 887 int family; /* selects inpcb, or in6pcb */ 888 void *aux; 889 { 890 struct tcpcb *tp; 891 int i; 892 893 switch (family) { 894 case PF_INET: 895 break; 896 #ifdef INET6 897 case PF_INET6: 898 break; 899 #endif 900 default: 901 return NULL; 902 } 903 904 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 905 if (tp == NULL) 906 return (NULL); 907 bzero((caddr_t)tp, sizeof(struct tcpcb)); 908 TAILQ_INIT(&tp->segq); 909 TAILQ_INIT(&tp->timeq); 910 tp->t_family = family; /* may be overridden later on */ 911 tp->t_peermss = tcp_mssdflt; 912 tp->t_ourmss = tcp_mssdflt; 913 tp->t_segsz = tcp_mssdflt; 914 LIST_INIT(&tp->t_sc); 915 916 tp->t_lastm = NULL; 917 tp->t_lastoff = 0; 918 919 callout_init(&tp->t_delack_ch); 920 for (i = 0; i < TCPT_NTIMERS; i++) 921 TCP_TIMER_INIT(tp, i); 922 923 tp->t_flags = 0; 924 if (tcp_do_rfc1323 && tcp_do_win_scale) 925 tp->t_flags |= TF_REQ_SCALE; 926 if (tcp_do_rfc1323 && tcp_do_timestamps) 927 tp->t_flags |= TF_REQ_TSTMP; 928 if (tcp_do_sack == 2) 929 tp->t_flags |= TF_WILL_SACK; 930 else if (tcp_do_sack == 1) 931 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 932 tp->t_flags |= TF_CANT_TXSACK; 933 switch (family) { 934 case PF_INET: 935 tp->t_inpcb = (struct inpcb *)aux; 936 tp->t_mtudisc = ip_mtudisc; 937 break; 938 #ifdef INET6 939 case PF_INET6: 940 tp->t_in6pcb = (struct in6pcb *)aux; 941 /* for IPv6, always try to run path MTU discovery */ 942 tp->t_mtudisc = 1; 943 break; 944 #endif 945 } 946 /* 947 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 948 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 949 * reasonable initial retransmit time. 950 */ 951 tp->t_srtt = TCPTV_SRTTBASE; 952 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 953 tp->t_rttmin = TCPTV_MIN; 954 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 955 TCPTV_MIN, TCPTV_REXMTMAX); 956 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 957 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 958 if (family == AF_INET) { 959 struct inpcb *inp = (struct inpcb *)aux; 960 inp->inp_ip.ip_ttl = ip_defttl; 961 inp->inp_ppcb = (caddr_t)tp; 962 } 963 #ifdef INET6 964 else if (family == AF_INET6) { 965 struct in6pcb *in6p = (struct in6pcb *)aux; 966 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 967 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 968 : NULL); 969 in6p->in6p_ppcb = (caddr_t)tp; 970 } 971 #endif 972 973 /* 974 * Initialize our timebase. When we send timestamps, we take 975 * the delta from tcp_now -- this means each connection always 976 * gets a timebase of 0, which makes it, among other things, 977 * more difficult to determine how long a system has been up, 978 * and thus how many TCP sequence increments have occurred. 979 */ 980 tp->ts_timebase = tcp_now; 981 982 return (tp); 983 } 984 985 /* 986 * Drop a TCP connection, reporting 987 * the specified error. If connection is synchronized, 988 * then send a RST to peer. 989 */ 990 struct tcpcb * 991 tcp_drop(tp, errno) 992 struct tcpcb *tp; 993 int errno; 994 { 995 struct socket *so = NULL; 996 997 #ifdef DIAGNOSTIC 998 if (tp->t_inpcb && tp->t_in6pcb) 999 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1000 #endif 1001 #ifdef INET 1002 if (tp->t_inpcb) 1003 so = tp->t_inpcb->inp_socket; 1004 #endif 1005 #ifdef INET6 1006 if (tp->t_in6pcb) 1007 so = tp->t_in6pcb->in6p_socket; 1008 #endif 1009 if (!so) 1010 return NULL; 1011 1012 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1013 tp->t_state = TCPS_CLOSED; 1014 (void) tcp_output(tp); 1015 tcpstat.tcps_drops++; 1016 } else 1017 tcpstat.tcps_conndrops++; 1018 if (errno == ETIMEDOUT && tp->t_softerror) 1019 errno = tp->t_softerror; 1020 so->so_error = errno; 1021 return (tcp_close(tp)); 1022 } 1023 1024 /* 1025 * Return whether this tcpcb is marked as dead, indicating 1026 * to the calling timer function that no further action should 1027 * be taken, as we are about to release this tcpcb. The release 1028 * of the storage will be done if this is the last timer running. 1029 * 1030 * This is typically called from the callout handler function before 1031 * callout_ack() is done, therefore we need to test the number of 1032 * running timer functions against 1 below, not 0. 1033 */ 1034 int 1035 tcp_isdead(tp) 1036 struct tcpcb *tp; 1037 { 1038 int dead = (tp->t_flags & TF_DEAD); 1039 1040 if (__predict_false(dead)) { 1041 if (tcp_timers_invoking(tp) > 1) 1042 /* not quite there yet -- count separately? */ 1043 return dead; 1044 tcpstat.tcps_delayed_free++; 1045 pool_put(&tcpcb_pool, tp); 1046 } 1047 return dead; 1048 } 1049 1050 /* 1051 * Close a TCP control block: 1052 * discard all space held by the tcp 1053 * discard internet protocol block 1054 * wake up any sleepers 1055 */ 1056 struct tcpcb * 1057 tcp_close(tp) 1058 struct tcpcb *tp; 1059 { 1060 struct inpcb *inp; 1061 #ifdef INET6 1062 struct in6pcb *in6p; 1063 #endif 1064 struct socket *so; 1065 #ifdef RTV_RTT 1066 struct rtentry *rt; 1067 #endif 1068 struct route *ro; 1069 1070 inp = tp->t_inpcb; 1071 #ifdef INET6 1072 in6p = tp->t_in6pcb; 1073 #endif 1074 so = NULL; 1075 ro = NULL; 1076 if (inp) { 1077 so = inp->inp_socket; 1078 ro = &inp->inp_route; 1079 } 1080 #ifdef INET6 1081 else if (in6p) { 1082 so = in6p->in6p_socket; 1083 ro = (struct route *)&in6p->in6p_route; 1084 } 1085 #endif 1086 1087 #ifdef RTV_RTT 1088 /* 1089 * If we sent enough data to get some meaningful characteristics, 1090 * save them in the routing entry. 'Enough' is arbitrarily 1091 * defined as the sendpipesize (default 4K) * 16. This would 1092 * give us 16 rtt samples assuming we only get one sample per 1093 * window (the usual case on a long haul net). 16 samples is 1094 * enough for the srtt filter to converge to within 5% of the correct 1095 * value; fewer samples and we could save a very bogus rtt. 1096 * 1097 * Don't update the default route's characteristics and don't 1098 * update anything that the user "locked". 1099 */ 1100 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1101 ro && (rt = ro->ro_rt) && 1102 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1103 u_long i = 0; 1104 1105 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1106 i = tp->t_srtt * 1107 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1108 if (rt->rt_rmx.rmx_rtt && i) 1109 /* 1110 * filter this update to half the old & half 1111 * the new values, converting scale. 1112 * See route.h and tcp_var.h for a 1113 * description of the scaling constants. 1114 */ 1115 rt->rt_rmx.rmx_rtt = 1116 (rt->rt_rmx.rmx_rtt + i) / 2; 1117 else 1118 rt->rt_rmx.rmx_rtt = i; 1119 } 1120 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1121 i = tp->t_rttvar * 1122 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1123 if (rt->rt_rmx.rmx_rttvar && i) 1124 rt->rt_rmx.rmx_rttvar = 1125 (rt->rt_rmx.rmx_rttvar + i) / 2; 1126 else 1127 rt->rt_rmx.rmx_rttvar = i; 1128 } 1129 /* 1130 * update the pipelimit (ssthresh) if it has been updated 1131 * already or if a pipesize was specified & the threshhold 1132 * got below half the pipesize. I.e., wait for bad news 1133 * before we start updating, then update on both good 1134 * and bad news. 1135 */ 1136 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1137 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1138 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1139 /* 1140 * convert the limit from user data bytes to 1141 * packets then to packet data bytes. 1142 */ 1143 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1144 if (i < 2) 1145 i = 2; 1146 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1147 if (rt->rt_rmx.rmx_ssthresh) 1148 rt->rt_rmx.rmx_ssthresh = 1149 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1150 else 1151 rt->rt_rmx.rmx_ssthresh = i; 1152 } 1153 } 1154 #endif /* RTV_RTT */ 1155 /* free the reassembly queue, if any */ 1156 TCP_REASS_LOCK(tp); 1157 (void) tcp_freeq(tp); 1158 TCP_REASS_UNLOCK(tp); 1159 1160 tcp_canceltimers(tp); 1161 TCP_CLEAR_DELACK(tp); 1162 syn_cache_cleanup(tp); 1163 1164 if (tp->t_template) { 1165 m_free(tp->t_template); 1166 tp->t_template = NULL; 1167 } 1168 if (tcp_timers_invoking(tp)) 1169 tp->t_flags |= TF_DEAD; 1170 else 1171 pool_put(&tcpcb_pool, tp); 1172 1173 if (inp) { 1174 inp->inp_ppcb = 0; 1175 soisdisconnected(so); 1176 in_pcbdetach(inp); 1177 } 1178 #ifdef INET6 1179 else if (in6p) { 1180 in6p->in6p_ppcb = 0; 1181 soisdisconnected(so); 1182 in6_pcbdetach(in6p); 1183 } 1184 #endif 1185 tcpstat.tcps_closed++; 1186 return ((struct tcpcb *)0); 1187 } 1188 1189 int 1190 tcp_freeq(tp) 1191 struct tcpcb *tp; 1192 { 1193 struct ipqent *qe; 1194 int rv = 0; 1195 #ifdef TCPREASS_DEBUG 1196 int i = 0; 1197 #endif 1198 1199 TCP_REASS_LOCK_CHECK(tp); 1200 1201 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1202 #ifdef TCPREASS_DEBUG 1203 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1204 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1205 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1206 #endif 1207 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1208 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1209 m_freem(qe->ipqe_m); 1210 pool_put(&ipqent_pool, qe); 1211 rv = 1; 1212 } 1213 return (rv); 1214 } 1215 1216 /* 1217 * Protocol drain routine. Called when memory is in short supply. 1218 */ 1219 void 1220 tcp_drain() 1221 { 1222 struct inpcb_hdr *inph; 1223 struct inpcb *inp; 1224 struct tcpcb *tp; 1225 1226 /* 1227 * Free the sequence queue of all TCP connections. 1228 */ 1229 inph = CIRCLEQ_FIRST(&tcbtable.inpt_queue); 1230 if (inp) /* XXX */ 1231 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1232 switch (inph->inph_af) { 1233 case AF_INET: 1234 tp = intotcpcb((struct inpcb *)inph); 1235 break; 1236 #ifdef INET6 1237 case AF_INET6: 1238 tp = in6totcpcb((struct in6pcb *)inph); 1239 break; 1240 #endif 1241 default: 1242 tp = NULL; 1243 break; 1244 } 1245 if (tp != NULL) { 1246 /* 1247 * We may be called from a device's interrupt 1248 * context. If the tcpcb is already busy, 1249 * just bail out now. 1250 */ 1251 if (tcp_reass_lock_try(tp) == 0) 1252 continue; 1253 if (tcp_freeq(tp)) 1254 tcpstat.tcps_connsdrained++; 1255 TCP_REASS_UNLOCK(tp); 1256 } 1257 } 1258 } 1259 1260 /* 1261 * Notify a tcp user of an asynchronous error; 1262 * store error as soft error, but wake up user 1263 * (for now, won't do anything until can select for soft error). 1264 */ 1265 void 1266 tcp_notify(inp, error) 1267 struct inpcb *inp; 1268 int error; 1269 { 1270 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1271 struct socket *so = inp->inp_socket; 1272 1273 /* 1274 * Ignore some errors if we are hooked up. 1275 * If connection hasn't completed, has retransmitted several times, 1276 * and receives a second error, give up now. This is better 1277 * than waiting a long time to establish a connection that 1278 * can never complete. 1279 */ 1280 if (tp->t_state == TCPS_ESTABLISHED && 1281 (error == EHOSTUNREACH || error == ENETUNREACH || 1282 error == EHOSTDOWN)) { 1283 return; 1284 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1285 tp->t_rxtshift > 3 && tp->t_softerror) 1286 so->so_error = error; 1287 else 1288 tp->t_softerror = error; 1289 wakeup((caddr_t) &so->so_timeo); 1290 sorwakeup(so); 1291 sowwakeup(so); 1292 } 1293 1294 #ifdef INET6 1295 void 1296 tcp6_notify(in6p, error) 1297 struct in6pcb *in6p; 1298 int error; 1299 { 1300 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1301 struct socket *so = in6p->in6p_socket; 1302 1303 /* 1304 * Ignore some errors if we are hooked up. 1305 * If connection hasn't completed, has retransmitted several times, 1306 * and receives a second error, give up now. This is better 1307 * than waiting a long time to establish a connection that 1308 * can never complete. 1309 */ 1310 if (tp->t_state == TCPS_ESTABLISHED && 1311 (error == EHOSTUNREACH || error == ENETUNREACH || 1312 error == EHOSTDOWN)) { 1313 return; 1314 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1315 tp->t_rxtshift > 3 && tp->t_softerror) 1316 so->so_error = error; 1317 else 1318 tp->t_softerror = error; 1319 wakeup((caddr_t) &so->so_timeo); 1320 sorwakeup(so); 1321 sowwakeup(so); 1322 } 1323 #endif 1324 1325 #ifdef INET6 1326 void 1327 tcp6_ctlinput(cmd, sa, d) 1328 int cmd; 1329 struct sockaddr *sa; 1330 void *d; 1331 { 1332 struct tcphdr th; 1333 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1334 int nmatch; 1335 struct ip6_hdr *ip6; 1336 const struct sockaddr_in6 *sa6_src = NULL; 1337 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1338 struct mbuf *m; 1339 int off; 1340 1341 if (sa->sa_family != AF_INET6 || 1342 sa->sa_len != sizeof(struct sockaddr_in6)) 1343 return; 1344 if ((unsigned)cmd >= PRC_NCMDS) 1345 return; 1346 else if (cmd == PRC_QUENCH) { 1347 /* XXX there's no PRC_QUENCH in IPv6 */ 1348 notify = tcp6_quench; 1349 } else if (PRC_IS_REDIRECT(cmd)) 1350 notify = in6_rtchange, d = NULL; 1351 else if (cmd == PRC_MSGSIZE) 1352 ; /* special code is present, see below */ 1353 else if (cmd == PRC_HOSTDEAD) 1354 d = NULL; 1355 else if (inet6ctlerrmap[cmd] == 0) 1356 return; 1357 1358 /* if the parameter is from icmp6, decode it. */ 1359 if (d != NULL) { 1360 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1361 m = ip6cp->ip6c_m; 1362 ip6 = ip6cp->ip6c_ip6; 1363 off = ip6cp->ip6c_off; 1364 sa6_src = ip6cp->ip6c_src; 1365 } else { 1366 m = NULL; 1367 ip6 = NULL; 1368 sa6_src = &sa6_any; 1369 } 1370 1371 if (ip6) { 1372 /* 1373 * XXX: We assume that when ip6 is non NULL, 1374 * M and OFF are valid. 1375 */ 1376 1377 /* check if we can safely examine src and dst ports */ 1378 if (m->m_pkthdr.len < off + sizeof(th)) { 1379 if (cmd == PRC_MSGSIZE) 1380 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1381 return; 1382 } 1383 1384 bzero(&th, sizeof(th)); 1385 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1386 1387 if (cmd == PRC_MSGSIZE) { 1388 int valid = 0; 1389 1390 /* 1391 * Check to see if we have a valid TCP connection 1392 * corresponding to the address in the ICMPv6 message 1393 * payload. 1394 */ 1395 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1396 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1397 th.th_sport, 0)) 1398 valid++; 1399 1400 /* 1401 * Depending on the value of "valid" and routing table 1402 * size (mtudisc_{hi,lo}wat), we will: 1403 * - recalcurate the new MTU and create the 1404 * corresponding routing entry, or 1405 * - ignore the MTU change notification. 1406 */ 1407 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1408 1409 /* 1410 * no need to call in6_pcbnotify, it should have been 1411 * called via callback if necessary 1412 */ 1413 return; 1414 } 1415 1416 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1417 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1418 if (nmatch == 0 && syn_cache_count && 1419 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1420 inet6ctlerrmap[cmd] == ENETUNREACH || 1421 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1422 syn_cache_unreach((struct sockaddr *)sa6_src, 1423 sa, &th); 1424 } else { 1425 (void) in6_pcbnotify(&tcbtable, sa, 0, 1426 (struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1427 } 1428 } 1429 #endif 1430 1431 #ifdef INET 1432 /* assumes that ip header and tcp header are contiguous on mbuf */ 1433 void * 1434 tcp_ctlinput(cmd, sa, v) 1435 int cmd; 1436 struct sockaddr *sa; 1437 void *v; 1438 { 1439 struct ip *ip = v; 1440 struct tcphdr *th; 1441 struct icmp *icp; 1442 extern const int inetctlerrmap[]; 1443 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1444 int errno; 1445 int nmatch; 1446 #ifdef INET6 1447 struct in6_addr src6, dst6; 1448 #endif 1449 1450 if (sa->sa_family != AF_INET || 1451 sa->sa_len != sizeof(struct sockaddr_in)) 1452 return NULL; 1453 if ((unsigned)cmd >= PRC_NCMDS) 1454 return NULL; 1455 errno = inetctlerrmap[cmd]; 1456 if (cmd == PRC_QUENCH) 1457 notify = tcp_quench; 1458 else if (PRC_IS_REDIRECT(cmd)) 1459 notify = in_rtchange, ip = 0; 1460 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1461 /* 1462 * Check to see if we have a valid TCP connection 1463 * corresponding to the address in the ICMP message 1464 * payload. 1465 * 1466 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1467 */ 1468 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1469 #ifdef INET6 1470 memset(&src6, 0, sizeof(src6)); 1471 memset(&dst6, 0, sizeof(dst6)); 1472 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1473 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1474 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1475 #endif 1476 if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport, 1477 ip->ip_src, th->th_sport) != NULL) 1478 ; 1479 #ifdef INET6 1480 else if (in6_pcblookup_connect(&tcbtable, &dst6, 1481 th->th_dport, &src6, th->th_sport, 0) != NULL) 1482 ; 1483 #endif 1484 else 1485 return NULL; 1486 1487 /* 1488 * Now that we've validated that we are actually communicating 1489 * with the host indicated in the ICMP message, locate the 1490 * ICMP header, recalculate the new MTU, and create the 1491 * corresponding routing entry. 1492 */ 1493 icp = (struct icmp *)((caddr_t)ip - 1494 offsetof(struct icmp, icmp_ip)); 1495 icmp_mtudisc(icp, ip->ip_dst); 1496 1497 return NULL; 1498 } else if (cmd == PRC_HOSTDEAD) 1499 ip = 0; 1500 else if (errno == 0) 1501 return NULL; 1502 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1503 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1504 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1505 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1506 if (nmatch == 0 && syn_cache_count && 1507 (inetctlerrmap[cmd] == EHOSTUNREACH || 1508 inetctlerrmap[cmd] == ENETUNREACH || 1509 inetctlerrmap[cmd] == EHOSTDOWN)) { 1510 struct sockaddr_in sin; 1511 bzero(&sin, sizeof(sin)); 1512 sin.sin_len = sizeof(sin); 1513 sin.sin_family = AF_INET; 1514 sin.sin_port = th->th_sport; 1515 sin.sin_addr = ip->ip_src; 1516 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1517 } 1518 1519 /* XXX mapped address case */ 1520 } else 1521 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1522 notify); 1523 return NULL; 1524 } 1525 1526 /* 1527 * When a source quence is received, we are being notifed of congestion. 1528 * Close the congestion window down to the Loss Window (one segment). 1529 * We will gradually open it again as we proceed. 1530 */ 1531 void 1532 tcp_quench(inp, errno) 1533 struct inpcb *inp; 1534 int errno; 1535 { 1536 struct tcpcb *tp = intotcpcb(inp); 1537 1538 if (tp) 1539 tp->snd_cwnd = tp->t_segsz; 1540 } 1541 #endif 1542 1543 #ifdef INET6 1544 void 1545 tcp6_quench(in6p, errno) 1546 struct in6pcb *in6p; 1547 int errno; 1548 { 1549 struct tcpcb *tp = in6totcpcb(in6p); 1550 1551 if (tp) 1552 tp->snd_cwnd = tp->t_segsz; 1553 } 1554 #endif 1555 1556 #ifdef INET 1557 /* 1558 * Path MTU Discovery handlers. 1559 */ 1560 void 1561 tcp_mtudisc_callback(faddr) 1562 struct in_addr faddr; 1563 { 1564 #ifdef INET6 1565 struct in6_addr in6; 1566 #endif 1567 1568 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1569 #ifdef INET6 1570 memset(&in6, 0, sizeof(in6)); 1571 in6.s6_addr16[5] = 0xffff; 1572 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1573 tcp6_mtudisc_callback(&in6); 1574 #endif 1575 } 1576 1577 /* 1578 * On receipt of path MTU corrections, flush old route and replace it 1579 * with the new one. Retransmit all unacknowledged packets, to ensure 1580 * that all packets will be received. 1581 */ 1582 void 1583 tcp_mtudisc(inp, errno) 1584 struct inpcb *inp; 1585 int errno; 1586 { 1587 struct tcpcb *tp = intotcpcb(inp); 1588 struct rtentry *rt = in_pcbrtentry(inp); 1589 1590 if (tp != 0) { 1591 if (rt != 0) { 1592 /* 1593 * If this was not a host route, remove and realloc. 1594 */ 1595 if ((rt->rt_flags & RTF_HOST) == 0) { 1596 in_rtchange(inp, errno); 1597 if ((rt = in_pcbrtentry(inp)) == 0) 1598 return; 1599 } 1600 1601 /* 1602 * Slow start out of the error condition. We 1603 * use the MTU because we know it's smaller 1604 * than the previously transmitted segment. 1605 * 1606 * Note: This is more conservative than the 1607 * suggestion in draft-floyd-incr-init-win-03. 1608 */ 1609 if (rt->rt_rmx.rmx_mtu != 0) 1610 tp->snd_cwnd = 1611 TCP_INITIAL_WINDOW(tcp_init_win, 1612 rt->rt_rmx.rmx_mtu); 1613 } 1614 1615 /* 1616 * Resend unacknowledged packets. 1617 */ 1618 tp->snd_nxt = tp->snd_una; 1619 tcp_output(tp); 1620 } 1621 } 1622 #endif 1623 1624 #ifdef INET6 1625 /* 1626 * Path MTU Discovery handlers. 1627 */ 1628 void 1629 tcp6_mtudisc_callback(faddr) 1630 struct in6_addr *faddr; 1631 { 1632 struct sockaddr_in6 sin6; 1633 1634 bzero(&sin6, sizeof(sin6)); 1635 sin6.sin6_family = AF_INET6; 1636 sin6.sin6_len = sizeof(struct sockaddr_in6); 1637 sin6.sin6_addr = *faddr; 1638 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1639 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1640 } 1641 1642 void 1643 tcp6_mtudisc(in6p, errno) 1644 struct in6pcb *in6p; 1645 int errno; 1646 { 1647 struct tcpcb *tp = in6totcpcb(in6p); 1648 struct rtentry *rt = in6_pcbrtentry(in6p); 1649 1650 if (tp != 0) { 1651 if (rt != 0) { 1652 /* 1653 * If this was not a host route, remove and realloc. 1654 */ 1655 if ((rt->rt_flags & RTF_HOST) == 0) { 1656 in6_rtchange(in6p, errno); 1657 if ((rt = in6_pcbrtentry(in6p)) == 0) 1658 return; 1659 } 1660 1661 /* 1662 * Slow start out of the error condition. We 1663 * use the MTU because we know it's smaller 1664 * than the previously transmitted segment. 1665 * 1666 * Note: This is more conservative than the 1667 * suggestion in draft-floyd-incr-init-win-03. 1668 */ 1669 if (rt->rt_rmx.rmx_mtu != 0) 1670 tp->snd_cwnd = 1671 TCP_INITIAL_WINDOW(tcp_init_win, 1672 rt->rt_rmx.rmx_mtu); 1673 } 1674 1675 /* 1676 * Resend unacknowledged packets. 1677 */ 1678 tp->snd_nxt = tp->snd_una; 1679 tcp_output(tp); 1680 } 1681 } 1682 #endif /* INET6 */ 1683 1684 /* 1685 * Compute the MSS to advertise to the peer. Called only during 1686 * the 3-way handshake. If we are the server (peer initiated 1687 * connection), we are called with a pointer to the interface 1688 * on which the SYN packet arrived. If we are the client (we 1689 * initiated connection), we are called with a pointer to the 1690 * interface out which this connection should go. 1691 * 1692 * NOTE: Do not subtract IP option/extension header size nor IPsec 1693 * header size from MSS advertisement. MSS option must hold the maximum 1694 * segment size we can accept, so it must always be: 1695 * max(if mtu) - ip header - tcp header 1696 */ 1697 u_long 1698 tcp_mss_to_advertise(ifp, af) 1699 const struct ifnet *ifp; 1700 int af; 1701 { 1702 extern u_long in_maxmtu; 1703 u_long mss = 0; 1704 u_long hdrsiz; 1705 1706 /* 1707 * In order to avoid defeating path MTU discovery on the peer, 1708 * we advertise the max MTU of all attached networks as our MSS, 1709 * per RFC 1191, section 3.1. 1710 * 1711 * We provide the option to advertise just the MTU of 1712 * the interface on which we hope this connection will 1713 * be receiving. If we are responding to a SYN, we 1714 * will have a pretty good idea about this, but when 1715 * initiating a connection there is a bit more doubt. 1716 * 1717 * We also need to ensure that loopback has a large enough 1718 * MSS, as the loopback MTU is never included in in_maxmtu. 1719 */ 1720 1721 if (ifp != NULL) 1722 switch (af) { 1723 case AF_INET: 1724 mss = ifp->if_mtu; 1725 break; 1726 #ifdef INET6 1727 case AF_INET6: 1728 mss = IN6_LINKMTU(ifp); 1729 break; 1730 #endif 1731 } 1732 1733 if (tcp_mss_ifmtu == 0) 1734 switch (af) { 1735 case AF_INET: 1736 mss = max(in_maxmtu, mss); 1737 break; 1738 #ifdef INET6 1739 case AF_INET6: 1740 mss = max(in6_maxmtu, mss); 1741 break; 1742 #endif 1743 } 1744 1745 switch (af) { 1746 case AF_INET: 1747 hdrsiz = sizeof(struct ip); 1748 break; 1749 #ifdef INET6 1750 case AF_INET6: 1751 hdrsiz = sizeof(struct ip6_hdr); 1752 break; 1753 #endif 1754 default: 1755 hdrsiz = 0; 1756 break; 1757 } 1758 hdrsiz += sizeof(struct tcphdr); 1759 if (mss > hdrsiz) 1760 mss -= hdrsiz; 1761 1762 mss = max(tcp_mssdflt, mss); 1763 return (mss); 1764 } 1765 1766 /* 1767 * Set connection variables based on the peer's advertised MSS. 1768 * We are passed the TCPCB for the actual connection. If we 1769 * are the server, we are called by the compressed state engine 1770 * when the 3-way handshake is complete. If we are the client, 1771 * we are called when we receive the SYN,ACK from the server. 1772 * 1773 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1774 * before this routine is called! 1775 */ 1776 void 1777 tcp_mss_from_peer(tp, offer) 1778 struct tcpcb *tp; 1779 int offer; 1780 { 1781 struct socket *so; 1782 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1783 struct rtentry *rt; 1784 #endif 1785 u_long bufsize; 1786 int mss; 1787 1788 #ifdef DIAGNOSTIC 1789 if (tp->t_inpcb && tp->t_in6pcb) 1790 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1791 #endif 1792 so = NULL; 1793 rt = NULL; 1794 #ifdef INET 1795 if (tp->t_inpcb) { 1796 so = tp->t_inpcb->inp_socket; 1797 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1798 rt = in_pcbrtentry(tp->t_inpcb); 1799 #endif 1800 } 1801 #endif 1802 #ifdef INET6 1803 if (tp->t_in6pcb) { 1804 so = tp->t_in6pcb->in6p_socket; 1805 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1806 rt = in6_pcbrtentry(tp->t_in6pcb); 1807 #endif 1808 } 1809 #endif 1810 1811 /* 1812 * As per RFC1122, use the default MSS value, unless they 1813 * sent us an offer. Do not accept offers less than 32 bytes. 1814 */ 1815 mss = tcp_mssdflt; 1816 if (offer) 1817 mss = offer; 1818 mss = max(mss, 32); /* sanity */ 1819 tp->t_peermss = mss; 1820 mss -= tcp_optlen(tp); 1821 #ifdef INET 1822 if (tp->t_inpcb) 1823 mss -= ip_optlen(tp->t_inpcb); 1824 #endif 1825 #ifdef INET6 1826 if (tp->t_in6pcb) 1827 mss -= ip6_optlen(tp->t_in6pcb); 1828 #endif 1829 1830 /* 1831 * If there's a pipesize, change the socket buffer to that size. 1832 * Make the socket buffer an integral number of MSS units. If 1833 * the MSS is larger than the socket buffer, artificially decrease 1834 * the MSS. 1835 */ 1836 #ifdef RTV_SPIPE 1837 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1838 bufsize = rt->rt_rmx.rmx_sendpipe; 1839 else 1840 #endif 1841 bufsize = so->so_snd.sb_hiwat; 1842 if (bufsize < mss) 1843 mss = bufsize; 1844 else { 1845 bufsize = roundup(bufsize, mss); 1846 if (bufsize > sb_max) 1847 bufsize = sb_max; 1848 (void) sbreserve(&so->so_snd, bufsize); 1849 } 1850 tp->t_segsz = mss; 1851 1852 #ifdef RTV_SSTHRESH 1853 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1854 /* 1855 * There's some sort of gateway or interface buffer 1856 * limit on the path. Use this to set the slow 1857 * start threshold, but set the threshold to no less 1858 * than 2 * MSS. 1859 */ 1860 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1861 } 1862 #endif 1863 } 1864 1865 /* 1866 * Processing necessary when a TCP connection is established. 1867 */ 1868 void 1869 tcp_established(tp) 1870 struct tcpcb *tp; 1871 { 1872 struct socket *so; 1873 #ifdef RTV_RPIPE 1874 struct rtentry *rt; 1875 #endif 1876 u_long bufsize; 1877 1878 #ifdef DIAGNOSTIC 1879 if (tp->t_inpcb && tp->t_in6pcb) 1880 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1881 #endif 1882 so = NULL; 1883 rt = NULL; 1884 #ifdef INET 1885 if (tp->t_inpcb) { 1886 so = tp->t_inpcb->inp_socket; 1887 #if defined(RTV_RPIPE) 1888 rt = in_pcbrtentry(tp->t_inpcb); 1889 #endif 1890 } 1891 #endif 1892 #ifdef INET6 1893 if (tp->t_in6pcb) { 1894 so = tp->t_in6pcb->in6p_socket; 1895 #if defined(RTV_RPIPE) 1896 rt = in6_pcbrtentry(tp->t_in6pcb); 1897 #endif 1898 } 1899 #endif 1900 1901 tp->t_state = TCPS_ESTABLISHED; 1902 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1903 1904 #ifdef RTV_RPIPE 1905 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1906 bufsize = rt->rt_rmx.rmx_recvpipe; 1907 else 1908 #endif 1909 bufsize = so->so_rcv.sb_hiwat; 1910 if (bufsize > tp->t_ourmss) { 1911 bufsize = roundup(bufsize, tp->t_ourmss); 1912 if (bufsize > sb_max) 1913 bufsize = sb_max; 1914 (void) sbreserve(&so->so_rcv, bufsize); 1915 } 1916 } 1917 1918 /* 1919 * Check if there's an initial rtt or rttvar. Convert from the 1920 * route-table units to scaled multiples of the slow timeout timer. 1921 * Called only during the 3-way handshake. 1922 */ 1923 void 1924 tcp_rmx_rtt(tp) 1925 struct tcpcb *tp; 1926 { 1927 #ifdef RTV_RTT 1928 struct rtentry *rt = NULL; 1929 int rtt; 1930 1931 #ifdef DIAGNOSTIC 1932 if (tp->t_inpcb && tp->t_in6pcb) 1933 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1934 #endif 1935 #ifdef INET 1936 if (tp->t_inpcb) 1937 rt = in_pcbrtentry(tp->t_inpcb); 1938 #endif 1939 #ifdef INET6 1940 if (tp->t_in6pcb) 1941 rt = in6_pcbrtentry(tp->t_in6pcb); 1942 #endif 1943 if (rt == NULL) 1944 return; 1945 1946 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1947 /* 1948 * XXX The lock bit for MTU indicates that the value 1949 * is also a minimum value; this is subject to time. 1950 */ 1951 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1952 TCPT_RANGESET(tp->t_rttmin, 1953 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1954 TCPTV_MIN, TCPTV_REXMTMAX); 1955 tp->t_srtt = rtt / 1956 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1957 if (rt->rt_rmx.rmx_rttvar) { 1958 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1959 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1960 (TCP_RTTVAR_SHIFT + 2)); 1961 } else { 1962 /* Default variation is +- 1 rtt */ 1963 tp->t_rttvar = 1964 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1965 } 1966 TCPT_RANGESET(tp->t_rxtcur, 1967 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1968 tp->t_rttmin, TCPTV_REXMTMAX); 1969 } 1970 #endif 1971 } 1972 1973 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1974 #if NRND > 0 1975 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1976 #endif 1977 1978 /* 1979 * Get a new sequence value given a tcp control block 1980 */ 1981 tcp_seq 1982 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1983 { 1984 1985 #ifdef INET 1986 if (tp->t_inpcb != NULL) { 1987 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1988 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1989 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1990 addin)); 1991 } 1992 #endif 1993 #ifdef INET6 1994 if (tp->t_in6pcb != NULL) { 1995 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 1996 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 1997 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 1998 addin)); 1999 } 2000 #endif 2001 /* Not possible. */ 2002 panic("tcp_new_iss"); 2003 } 2004 2005 /* 2006 * This routine actually generates a new TCP initial sequence number. 2007 */ 2008 tcp_seq 2009 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2010 size_t addrsz, tcp_seq addin) 2011 { 2012 tcp_seq tcp_iss; 2013 2014 #if NRND > 0 2015 static int beenhere; 2016 2017 /* 2018 * If we haven't been here before, initialize our cryptographic 2019 * hash secret. 2020 */ 2021 if (beenhere == 0) { 2022 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2023 RND_EXTRACT_ANY); 2024 beenhere = 1; 2025 } 2026 2027 if (tcp_do_rfc1948) { 2028 MD5_CTX ctx; 2029 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2030 2031 /* 2032 * Compute the base value of the ISS. It is a hash 2033 * of (saddr, sport, daddr, dport, secret). 2034 */ 2035 MD5Init(&ctx); 2036 2037 MD5Update(&ctx, (u_char *) laddr, addrsz); 2038 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2039 2040 MD5Update(&ctx, (u_char *) faddr, addrsz); 2041 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2042 2043 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2044 2045 MD5Final(hash, &ctx); 2046 2047 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2048 2049 /* 2050 * Now increment our "timer", and add it in to 2051 * the computed value. 2052 * 2053 * XXX Use `addin'? 2054 * XXX TCP_ISSINCR too large to use? 2055 */ 2056 tcp_iss_seq += TCP_ISSINCR; 2057 #ifdef TCPISS_DEBUG 2058 printf("ISS hash 0x%08x, ", tcp_iss); 2059 #endif 2060 tcp_iss += tcp_iss_seq + addin; 2061 #ifdef TCPISS_DEBUG 2062 printf("new ISS 0x%08x\n", tcp_iss); 2063 #endif 2064 } else 2065 #endif /* NRND > 0 */ 2066 { 2067 /* 2068 * Randomize. 2069 */ 2070 #if NRND > 0 2071 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2072 #else 2073 tcp_iss = arc4random(); 2074 #endif 2075 2076 /* 2077 * If we were asked to add some amount to a known value, 2078 * we will take a random value obtained above, mask off 2079 * the upper bits, and add in the known value. We also 2080 * add in a constant to ensure that we are at least a 2081 * certain distance from the original value. 2082 * 2083 * This is used when an old connection is in timed wait 2084 * and we have a new one coming in, for instance. 2085 */ 2086 if (addin != 0) { 2087 #ifdef TCPISS_DEBUG 2088 printf("Random %08x, ", tcp_iss); 2089 #endif 2090 tcp_iss &= TCP_ISS_RANDOM_MASK; 2091 tcp_iss += addin + TCP_ISSINCR; 2092 #ifdef TCPISS_DEBUG 2093 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2094 #endif 2095 } else { 2096 tcp_iss &= TCP_ISS_RANDOM_MASK; 2097 tcp_iss += tcp_iss_seq; 2098 tcp_iss_seq += TCP_ISSINCR; 2099 #ifdef TCPISS_DEBUG 2100 printf("ISS %08x\n", tcp_iss); 2101 #endif 2102 } 2103 } 2104 2105 if (tcp_compat_42) { 2106 /* 2107 * Limit it to the positive range for really old TCP 2108 * implementations. 2109 * Just AND off the top bit instead of checking if 2110 * is set first - saves a branch 50% of the time. 2111 */ 2112 tcp_iss &= 0x7fffffff; /* XXX */ 2113 } 2114 2115 return (tcp_iss); 2116 } 2117 2118 #if defined(IPSEC) || defined(FAST_IPSEC) 2119 /* compute ESP/AH header size for TCP, including outer IP header. */ 2120 size_t 2121 ipsec4_hdrsiz_tcp(tp) 2122 struct tcpcb *tp; 2123 { 2124 struct inpcb *inp; 2125 size_t hdrsiz; 2126 2127 /* XXX mapped addr case (tp->t_in6pcb) */ 2128 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2129 return 0; 2130 switch (tp->t_family) { 2131 case AF_INET: 2132 /* XXX: should use currect direction. */ 2133 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2134 break; 2135 default: 2136 hdrsiz = 0; 2137 break; 2138 } 2139 2140 return hdrsiz; 2141 } 2142 2143 #ifdef INET6 2144 size_t 2145 ipsec6_hdrsiz_tcp(tp) 2146 struct tcpcb *tp; 2147 { 2148 struct in6pcb *in6p; 2149 size_t hdrsiz; 2150 2151 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2152 return 0; 2153 switch (tp->t_family) { 2154 case AF_INET6: 2155 /* XXX: should use currect direction. */ 2156 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2157 break; 2158 case AF_INET: 2159 /* mapped address case - tricky */ 2160 default: 2161 hdrsiz = 0; 2162 break; 2163 } 2164 2165 return hdrsiz; 2166 } 2167 #endif 2168 #endif /*IPSEC*/ 2169 2170 /* 2171 * Determine the length of the TCP options for this connection. 2172 * 2173 * XXX: What do we do for SACK, when we add that? Just reserve 2174 * all of the space? Otherwise we can't exactly be incrementing 2175 * cwnd by an amount that varies depending on the amount we last 2176 * had to SACK! 2177 */ 2178 2179 u_int 2180 tcp_optlen(tp) 2181 struct tcpcb *tp; 2182 { 2183 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2184 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2185 return TCPOLEN_TSTAMP_APPA; 2186 else 2187 return 0; 2188 } 2189
Indexes created Mon Sep 22 05:09:51 GMT 2025