1 /* $NetBSD: tcp_subr.c,v 1.218 2007/08/02 02:42:41 rmind 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 1.218 2007/08/02 02:42:41 rmind 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/tcp_congctl.h> 155 #include <netinet/tcpip.h> 156 157 #ifdef IPSEC 158 #include <netinet6/ipsec.h> 159 #include <netkey/key.h> 160 #endif /*IPSEC*/ 161 162 #ifdef FAST_IPSEC 163 #include <netipsec/ipsec.h> 164 #include <netipsec/xform.h> 165 #ifdef INET6 166 #include <netipsec/ipsec6.h> 167 #endif 168 #include <netipsec/key.h> 169 #endif /* FAST_IPSEC*/ 170 171 172 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 173 struct tcpstat tcpstat; /* tcp statistics */ 174 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 175 176 /* patchable/settable parameters for tcp */ 177 int tcp_mssdflt = TCP_MSS; 178 int tcp_minmss = TCP_MINMSS; 179 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 180 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 181 #if NRND > 0 182 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 183 #endif 184 int tcp_do_sack = 1; /* selective acknowledgement */ 185 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 186 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 187 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 188 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 189 #ifndef TCP_INIT_WIN 190 #define TCP_INIT_WIN 0 /* initial slow start window */ 191 #endif 192 #ifndef TCP_INIT_WIN_LOCAL 193 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 194 #endif 195 int tcp_init_win = TCP_INIT_WIN; 196 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 197 int tcp_mss_ifmtu = 0; 198 #ifdef TCP_COMPAT_42 199 int tcp_compat_42 = 1; 200 #else 201 int tcp_compat_42 = 0; 202 #endif 203 int tcp_rst_ppslim = 100; /* 100pps */ 204 int tcp_ackdrop_ppslim = 100; /* 100pps */ 205 int tcp_do_loopback_cksum = 0; 206 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 207 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 208 int tcp_sack_tp_maxholes = 32; 209 int tcp_sack_globalmaxholes = 1024; 210 int tcp_sack_globalholes = 0; 211 int tcp_ecn_maxretries = 1; 212 213 /* tcb hash */ 214 #ifndef TCBHASHSIZE 215 #define TCBHASHSIZE 128 216 #endif 217 int tcbhashsize = TCBHASHSIZE; 218 219 /* syn hash parameters */ 220 #define TCP_SYN_HASH_SIZE 293 221 #define TCP_SYN_BUCKET_SIZE 35 222 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 223 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 224 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 225 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 226 227 int tcp_freeq(struct tcpcb *); 228 229 #ifdef INET 230 void tcp_mtudisc_callback(struct in_addr); 231 #endif 232 #ifdef INET6 233 void tcp6_mtudisc_callback(struct in6_addr *); 234 #endif 235 236 #ifdef INET6 237 void tcp6_mtudisc(struct in6pcb *, int); 238 #endif 239 240 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL, 241 IPL_SOFTNET); 242 243 #ifdef TCP_CSUM_COUNTERS 244 #include <sys/device.h> 245 246 #if defined(INET) 247 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "hwcsum bad"); 249 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 250 NULL, "tcp", "hwcsum ok"); 251 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 252 NULL, "tcp", "hwcsum data"); 253 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 254 NULL, "tcp", "swcsum"); 255 256 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 257 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 258 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 259 EVCNT_ATTACH_STATIC(tcp_swcsum); 260 #endif /* defined(INET) */ 261 262 #if defined(INET6) 263 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "hwcsum bad"); 265 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 NULL, "tcp6", "hwcsum ok"); 267 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 268 NULL, "tcp6", "hwcsum data"); 269 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 270 NULL, "tcp6", "swcsum"); 271 272 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 273 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 274 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 275 EVCNT_ATTACH_STATIC(tcp6_swcsum); 276 #endif /* defined(INET6) */ 277 #endif /* TCP_CSUM_COUNTERS */ 278 279 280 #ifdef TCP_OUTPUT_COUNTERS 281 #include <sys/device.h> 282 283 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output big header"); 285 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output predict hit"); 287 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output predict miss"); 289 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 NULL, "tcp", "output copy small"); 291 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 NULL, "tcp", "output copy big"); 293 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 294 NULL, "tcp", "output reference big"); 295 296 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 297 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 298 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 299 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 300 EVCNT_ATTACH_STATIC(tcp_output_copybig); 301 EVCNT_ATTACH_STATIC(tcp_output_refbig); 302 303 #endif /* TCP_OUTPUT_COUNTERS */ 304 305 #ifdef TCP_REASS_COUNTERS 306 #include <sys/device.h> 307 308 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 309 NULL, "tcp_reass", "calls"); 310 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 311 &tcp_reass_, "tcp_reass", "insert into empty queue"); 312 struct evcnt tcp_reass_iteration[8] = { 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 318 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 319 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 321 }; 322 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "prepend to first"); 324 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "prepend"); 326 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "insert"); 328 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "insert at tail"); 330 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "append"); 332 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "append to tail fragment"); 334 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "overlap at end"); 336 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 337 &tcp_reass_, "tcp_reass", "overlap at start"); 338 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 339 &tcp_reass_, "tcp_reass", "duplicate segment"); 340 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 341 &tcp_reass_, "tcp_reass", "duplicate fragment"); 342 343 EVCNT_ATTACH_STATIC(tcp_reass_); 344 EVCNT_ATTACH_STATIC(tcp_reass_empty); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 349 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 350 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 351 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 352 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 353 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 354 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 355 EVCNT_ATTACH_STATIC(tcp_reass_insert); 356 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 357 EVCNT_ATTACH_STATIC(tcp_reass_append); 358 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 359 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 360 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 361 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 362 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 363 364 #endif /* TCP_REASS_COUNTERS */ 365 366 #ifdef MBUFTRACE 367 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 368 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 369 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 370 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 371 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 372 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 373 #endif 374 375 /* 376 * Tcp initialization 377 */ 378 void 379 tcp_init(void) 380 { 381 int hlen; 382 383 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 384 385 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 386 #ifdef INET6 387 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 388 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 389 #endif 390 if (max_protohdr < hlen) 391 max_protohdr = hlen; 392 if (max_linkhdr + hlen > MHLEN) 393 panic("tcp_init"); 394 395 #ifdef INET 396 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 397 #endif 398 #ifdef INET6 399 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 400 #endif 401 402 /* Initialize timer state. */ 403 tcp_timer_init(); 404 405 /* Initialize the compressed state engine. */ 406 syn_cache_init(); 407 408 /* Initialize the congestion control algorithms. */ 409 tcp_congctl_init(); 410 411 /* Initialize the TCPCB template. */ 412 tcp_tcpcb_template(); 413 414 MOWNER_ATTACH(&tcp_tx_mowner); 415 MOWNER_ATTACH(&tcp_rx_mowner); 416 MOWNER_ATTACH(&tcp_reass_mowner); 417 MOWNER_ATTACH(&tcp_sock_mowner); 418 MOWNER_ATTACH(&tcp_sock_tx_mowner); 419 MOWNER_ATTACH(&tcp_sock_rx_mowner); 420 MOWNER_ATTACH(&tcp_mowner); 421 } 422 423 /* 424 * Create template to be used to send tcp packets on a connection. 425 * Call after host entry created, allocates an mbuf and fills 426 * in a skeletal tcp/ip header, minimizing the amount of work 427 * necessary when the connection is used. 428 */ 429 struct mbuf * 430 tcp_template(struct tcpcb *tp) 431 { 432 struct inpcb *inp = tp->t_inpcb; 433 #ifdef INET6 434 struct in6pcb *in6p = tp->t_in6pcb; 435 #endif 436 struct tcphdr *n; 437 struct mbuf *m; 438 int hlen; 439 440 switch (tp->t_family) { 441 case AF_INET: 442 hlen = sizeof(struct ip); 443 if (inp) 444 break; 445 #ifdef INET6 446 if (in6p) { 447 /* mapped addr case */ 448 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 449 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 450 break; 451 } 452 #endif 453 return NULL; /*EINVAL*/ 454 #ifdef INET6 455 case AF_INET6: 456 hlen = sizeof(struct ip6_hdr); 457 if (in6p) { 458 /* more sainty check? */ 459 break; 460 } 461 return NULL; /*EINVAL*/ 462 #endif 463 default: 464 hlen = 0; /*pacify gcc*/ 465 return NULL; /*EAFNOSUPPORT*/ 466 } 467 #ifdef DIAGNOSTIC 468 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 469 panic("mclbytes too small for t_template"); 470 #endif 471 m = tp->t_template; 472 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 473 ; 474 else { 475 if (m) 476 m_freem(m); 477 m = tp->t_template = NULL; 478 MGETHDR(m, M_DONTWAIT, MT_HEADER); 479 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 480 MCLGET(m, M_DONTWAIT); 481 if ((m->m_flags & M_EXT) == 0) { 482 m_free(m); 483 m = NULL; 484 } 485 } 486 if (m == NULL) 487 return NULL; 488 MCLAIM(m, &tcp_mowner); 489 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 490 } 491 492 bzero(mtod(m, void *), m->m_len); 493 494 n = (struct tcphdr *)(mtod(m, char *) + hlen); 495 496 switch (tp->t_family) { 497 case AF_INET: 498 { 499 struct ipovly *ipov; 500 mtod(m, struct ip *)->ip_v = 4; 501 mtod(m, struct ip *)->ip_hl = hlen >> 2; 502 ipov = mtod(m, struct ipovly *); 503 ipov->ih_pr = IPPROTO_TCP; 504 ipov->ih_len = htons(sizeof(struct tcphdr)); 505 if (inp) { 506 ipov->ih_src = inp->inp_laddr; 507 ipov->ih_dst = inp->inp_faddr; 508 } 509 #ifdef INET6 510 else if (in6p) { 511 /* mapped addr case */ 512 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 513 sizeof(ipov->ih_src)); 514 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 515 sizeof(ipov->ih_dst)); 516 } 517 #endif 518 /* 519 * Compute the pseudo-header portion of the checksum 520 * now. We incrementally add in the TCP option and 521 * payload lengths later, and then compute the TCP 522 * checksum right before the packet is sent off onto 523 * the wire. 524 */ 525 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 526 ipov->ih_dst.s_addr, 527 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 528 break; 529 } 530 #ifdef INET6 531 case AF_INET6: 532 { 533 struct ip6_hdr *ip6; 534 mtod(m, struct ip *)->ip_v = 6; 535 ip6 = mtod(m, struct ip6_hdr *); 536 ip6->ip6_nxt = IPPROTO_TCP; 537 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 538 ip6->ip6_src = in6p->in6p_laddr; 539 ip6->ip6_dst = in6p->in6p_faddr; 540 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 541 if (ip6_auto_flowlabel) { 542 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 543 ip6->ip6_flow |= 544 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 545 } 546 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 547 ip6->ip6_vfc |= IPV6_VERSION; 548 549 /* 550 * Compute the pseudo-header portion of the checksum 551 * now. We incrementally add in the TCP option and 552 * payload lengths later, and then compute the TCP 553 * checksum right before the packet is sent off onto 554 * the wire. 555 */ 556 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 557 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 558 htonl(IPPROTO_TCP)); 559 break; 560 } 561 #endif 562 } 563 if (inp) { 564 n->th_sport = inp->inp_lport; 565 n->th_dport = inp->inp_fport; 566 } 567 #ifdef INET6 568 else if (in6p) { 569 n->th_sport = in6p->in6p_lport; 570 n->th_dport = in6p->in6p_fport; 571 } 572 #endif 573 n->th_seq = 0; 574 n->th_ack = 0; 575 n->th_x2 = 0; 576 n->th_off = 5; 577 n->th_flags = 0; 578 n->th_win = 0; 579 n->th_urp = 0; 580 return (m); 581 } 582 583 /* 584 * Send a single message to the TCP at address specified by 585 * the given TCP/IP header. If m == 0, then we make a copy 586 * of the tcpiphdr at ti and send directly to the addressed host. 587 * This is used to force keep alive messages out using the TCP 588 * template for a connection tp->t_template. If flags are given 589 * then we send a message back to the TCP which originated the 590 * segment ti, and discard the mbuf containing it and any other 591 * attached mbufs. 592 * 593 * In any case the ack and sequence number of the transmitted 594 * segment are as specified by the parameters. 595 */ 596 int 597 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 598 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 599 { 600 struct route *ro; 601 int error, tlen, win = 0; 602 int hlen; 603 struct ip *ip; 604 #ifdef INET6 605 struct ip6_hdr *ip6; 606 #endif 607 int family; /* family on packet, not inpcb/in6pcb! */ 608 struct tcphdr *th; 609 struct socket *so; 610 611 if (tp != NULL && (flags & TH_RST) == 0) { 612 #ifdef DIAGNOSTIC 613 if (tp->t_inpcb && tp->t_in6pcb) 614 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 615 #endif 616 #ifdef INET 617 if (tp->t_inpcb) 618 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 619 #endif 620 #ifdef INET6 621 if (tp->t_in6pcb) 622 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 623 #endif 624 } 625 626 th = NULL; /* Quell uninitialized warning */ 627 ip = NULL; 628 #ifdef INET6 629 ip6 = NULL; 630 #endif 631 if (m == 0) { 632 if (!template) 633 return EINVAL; 634 635 /* get family information from template */ 636 switch (mtod(template, struct ip *)->ip_v) { 637 case 4: 638 family = AF_INET; 639 hlen = sizeof(struct ip); 640 break; 641 #ifdef INET6 642 case 6: 643 family = AF_INET6; 644 hlen = sizeof(struct ip6_hdr); 645 break; 646 #endif 647 default: 648 return EAFNOSUPPORT; 649 } 650 651 MGETHDR(m, M_DONTWAIT, MT_HEADER); 652 if (m) { 653 MCLAIM(m, &tcp_tx_mowner); 654 MCLGET(m, M_DONTWAIT); 655 if ((m->m_flags & M_EXT) == 0) { 656 m_free(m); 657 m = NULL; 658 } 659 } 660 if (m == NULL) 661 return (ENOBUFS); 662 663 if (tcp_compat_42) 664 tlen = 1; 665 else 666 tlen = 0; 667 668 m->m_data += max_linkhdr; 669 bcopy(mtod(template, void *), mtod(m, void *), 670 template->m_len); 671 switch (family) { 672 case AF_INET: 673 ip = mtod(m, struct ip *); 674 th = (struct tcphdr *)(ip + 1); 675 break; 676 #ifdef INET6 677 case AF_INET6: 678 ip6 = mtod(m, struct ip6_hdr *); 679 th = (struct tcphdr *)(ip6 + 1); 680 break; 681 #endif 682 #if 0 683 default: 684 /* noone will visit here */ 685 m_freem(m); 686 return EAFNOSUPPORT; 687 #endif 688 } 689 flags = TH_ACK; 690 } else { 691 692 if ((m->m_flags & M_PKTHDR) == 0) { 693 #if 0 694 printf("non PKTHDR to tcp_respond\n"); 695 #endif 696 m_freem(m); 697 return EINVAL; 698 } 699 #ifdef DIAGNOSTIC 700 if (!th0) 701 panic("th0 == NULL in tcp_respond"); 702 #endif 703 704 /* get family information from m */ 705 switch (mtod(m, struct ip *)->ip_v) { 706 case 4: 707 family = AF_INET; 708 hlen = sizeof(struct ip); 709 ip = mtod(m, struct ip *); 710 break; 711 #ifdef INET6 712 case 6: 713 family = AF_INET6; 714 hlen = sizeof(struct ip6_hdr); 715 ip6 = mtod(m, struct ip6_hdr *); 716 break; 717 #endif 718 default: 719 m_freem(m); 720 return EAFNOSUPPORT; 721 } 722 /* clear h/w csum flags inherited from rx packet */ 723 m->m_pkthdr.csum_flags = 0; 724 725 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 726 tlen = sizeof(*th0); 727 else 728 tlen = th0->th_off << 2; 729 730 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 731 mtod(m, char *) + hlen == (char *)th0) { 732 m->m_len = hlen + tlen; 733 m_freem(m->m_next); 734 m->m_next = NULL; 735 } else { 736 struct mbuf *n; 737 738 #ifdef DIAGNOSTIC 739 if (max_linkhdr + hlen + tlen > MCLBYTES) { 740 m_freem(m); 741 return EMSGSIZE; 742 } 743 #endif 744 MGETHDR(n, M_DONTWAIT, MT_HEADER); 745 if (n && max_linkhdr + hlen + tlen > MHLEN) { 746 MCLGET(n, M_DONTWAIT); 747 if ((n->m_flags & M_EXT) == 0) { 748 m_freem(n); 749 n = NULL; 750 } 751 } 752 if (!n) { 753 m_freem(m); 754 return ENOBUFS; 755 } 756 757 MCLAIM(n, &tcp_tx_mowner); 758 n->m_data += max_linkhdr; 759 n->m_len = hlen + tlen; 760 m_copyback(n, 0, hlen, mtod(m, void *)); 761 m_copyback(n, hlen, tlen, (void *)th0); 762 763 m_freem(m); 764 m = n; 765 n = NULL; 766 } 767 768 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 769 switch (family) { 770 case AF_INET: 771 ip = mtod(m, struct ip *); 772 th = (struct tcphdr *)(ip + 1); 773 ip->ip_p = IPPROTO_TCP; 774 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 775 ip->ip_p = IPPROTO_TCP; 776 break; 777 #ifdef INET6 778 case AF_INET6: 779 ip6 = mtod(m, struct ip6_hdr *); 780 th = (struct tcphdr *)(ip6 + 1); 781 ip6->ip6_nxt = IPPROTO_TCP; 782 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 783 ip6->ip6_nxt = IPPROTO_TCP; 784 break; 785 #endif 786 #if 0 787 default: 788 /* noone will visit here */ 789 m_freem(m); 790 return EAFNOSUPPORT; 791 #endif 792 } 793 xchg(th->th_dport, th->th_sport, u_int16_t); 794 #undef xchg 795 tlen = 0; /*be friendly with the following code*/ 796 } 797 th->th_seq = htonl(seq); 798 th->th_ack = htonl(ack); 799 th->th_x2 = 0; 800 if ((flags & TH_SYN) == 0) { 801 if (tp) 802 win >>= tp->rcv_scale; 803 if (win > TCP_MAXWIN) 804 win = TCP_MAXWIN; 805 th->th_win = htons((u_int16_t)win); 806 th->th_off = sizeof (struct tcphdr) >> 2; 807 tlen += sizeof(*th); 808 } else 809 tlen += th->th_off << 2; 810 m->m_len = hlen + tlen; 811 m->m_pkthdr.len = hlen + tlen; 812 m->m_pkthdr.rcvif = (struct ifnet *) 0; 813 th->th_flags = flags; 814 th->th_urp = 0; 815 816 switch (family) { 817 #ifdef INET 818 case AF_INET: 819 { 820 struct ipovly *ipov = (struct ipovly *)ip; 821 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 822 ipov->ih_len = htons((u_int16_t)tlen); 823 824 th->th_sum = 0; 825 th->th_sum = in_cksum(m, hlen + tlen); 826 ip->ip_len = htons(hlen + tlen); 827 ip->ip_ttl = ip_defttl; 828 break; 829 } 830 #endif 831 #ifdef INET6 832 case AF_INET6: 833 { 834 th->th_sum = 0; 835 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 836 tlen); 837 ip6->ip6_plen = htons(tlen); 838 if (tp && tp->t_in6pcb) { 839 struct ifnet *oifp; 840 ro = (struct route *)&tp->t_in6pcb->in6p_route; 841 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 842 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 843 } else 844 ip6->ip6_hlim = ip6_defhlim; 845 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 846 if (ip6_auto_flowlabel) { 847 ip6->ip6_flow |= 848 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 849 } 850 break; 851 } 852 #endif 853 } 854 855 if (tp && tp->t_inpcb) 856 so = tp->t_inpcb->inp_socket; 857 #ifdef INET6 858 else if (tp && tp->t_in6pcb) 859 so = tp->t_in6pcb->in6p_socket; 860 #endif 861 else 862 so = NULL; 863 864 if (tp != NULL && tp->t_inpcb != NULL) { 865 ro = &tp->t_inpcb->inp_route; 866 #ifdef DIAGNOSTIC 867 if (family != AF_INET) 868 panic("tcp_respond: address family mismatch"); 869 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 870 panic("tcp_respond: ip_dst %x != inp_faddr %x", 871 ntohl(ip->ip_dst.s_addr), 872 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 873 } 874 #endif 875 } 876 #ifdef INET6 877 else if (tp != NULL && tp->t_in6pcb != NULL) { 878 ro = (struct route *)&tp->t_in6pcb->in6p_route; 879 #ifdef DIAGNOSTIC 880 if (family == AF_INET) { 881 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 882 panic("tcp_respond: not mapped addr"); 883 if (bcmp(&ip->ip_dst, 884 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 885 sizeof(ip->ip_dst)) != 0) { 886 panic("tcp_respond: ip_dst != in6p_faddr"); 887 } 888 } else if (family == AF_INET6) { 889 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 890 &tp->t_in6pcb->in6p_faddr)) 891 panic("tcp_respond: ip6_dst != in6p_faddr"); 892 } else 893 panic("tcp_respond: address family mismatch"); 894 #endif 895 } 896 #endif 897 else 898 ro = NULL; 899 900 switch (family) { 901 #ifdef INET 902 case AF_INET: 903 error = ip_output(m, NULL, ro, 904 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 905 (struct ip_moptions *)0, so); 906 break; 907 #endif 908 #ifdef INET6 909 case AF_INET6: 910 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL); 911 break; 912 #endif 913 default: 914 error = EAFNOSUPPORT; 915 break; 916 } 917 918 return (error); 919 } 920 921 /* 922 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 923 * a bunch of members individually, we maintain this template for the 924 * static and mostly-static components of the TCPCB, and copy it into 925 * the new TCPCB instead. 926 */ 927 static struct tcpcb tcpcb_template = { 928 .t_srtt = TCPTV_SRTTBASE, 929 .t_rttmin = TCPTV_MIN, 930 931 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 932 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 933 .snd_numholes = 0, 934 935 .t_partialacks = -1, 936 .t_bytes_acked = 0, 937 }; 938 939 /* 940 * Updates the TCPCB template whenever a parameter that would affect 941 * the template is changed. 942 */ 943 void 944 tcp_tcpcb_template(void) 945 { 946 struct tcpcb *tp = &tcpcb_template; 947 int flags; 948 949 tp->t_peermss = tcp_mssdflt; 950 tp->t_ourmss = tcp_mssdflt; 951 tp->t_segsz = tcp_mssdflt; 952 953 flags = 0; 954 if (tcp_do_rfc1323 && tcp_do_win_scale) 955 flags |= TF_REQ_SCALE; 956 if (tcp_do_rfc1323 && tcp_do_timestamps) 957 flags |= TF_REQ_TSTMP; 958 tp->t_flags = flags; 959 960 /* 961 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 962 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 963 * reasonable initial retransmit time. 964 */ 965 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 966 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 967 TCPTV_MIN, TCPTV_REXMTMAX); 968 969 /* Keep Alive */ 970 tp->t_keepinit = tcp_keepinit; 971 tp->t_keepidle = tcp_keepidle; 972 tp->t_keepintvl = tcp_keepintvl; 973 tp->t_keepcnt = tcp_keepcnt; 974 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 975 } 976 977 /* 978 * Create a new TCP control block, making an 979 * empty reassembly queue and hooking it to the argument 980 * protocol control block. 981 */ 982 /* family selects inpcb, or in6pcb */ 983 struct tcpcb * 984 tcp_newtcpcb(int family, void *aux) 985 { 986 struct tcpcb *tp; 987 int i; 988 989 /* XXX Consider using a pool_cache for speed. */ 990 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 991 if (tp == NULL) 992 return (NULL); 993 memcpy(tp, &tcpcb_template, sizeof(*tp)); 994 TAILQ_INIT(&tp->segq); 995 TAILQ_INIT(&tp->timeq); 996 tp->t_family = family; /* may be overridden later on */ 997 TAILQ_INIT(&tp->snd_holes); 998 LIST_INIT(&tp->t_sc); /* XXX can template this */ 999 1000 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1001 for (i = 0; i < TCPT_NTIMERS; i++) { 1002 callout_init(&tp->t_timer[i], 0); 1003 TCP_TIMER_INIT(tp, i); 1004 } 1005 callout_init(&tp->t_delack_ch, 0); 1006 1007 switch (family) { 1008 case AF_INET: 1009 { 1010 struct inpcb *inp = (struct inpcb *)aux; 1011 1012 inp->inp_ip.ip_ttl = ip_defttl; 1013 inp->inp_ppcb = (void *)tp; 1014 1015 tp->t_inpcb = inp; 1016 tp->t_mtudisc = ip_mtudisc; 1017 break; 1018 } 1019 #ifdef INET6 1020 case AF_INET6: 1021 { 1022 struct in6pcb *in6p = (struct in6pcb *)aux; 1023 1024 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1025 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1026 : NULL); 1027 in6p->in6p_ppcb = (void *)tp; 1028 1029 tp->t_in6pcb = in6p; 1030 /* for IPv6, always try to run path MTU discovery */ 1031 tp->t_mtudisc = 1; 1032 break; 1033 } 1034 #endif /* INET6 */ 1035 default: 1036 for (i = 0; i < TCPT_NTIMERS; i++) 1037 callout_destroy(&tp->t_timer[i]); 1038 callout_destroy(&tp->t_delack_ch); 1039 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1040 return (NULL); 1041 } 1042 1043 /* 1044 * Initialize our timebase. When we send timestamps, we take 1045 * the delta from tcp_now -- this means each connection always 1046 * gets a timebase of 0, which makes it, among other things, 1047 * more difficult to determine how long a system has been up, 1048 * and thus how many TCP sequence increments have occurred. 1049 */ 1050 tp->ts_timebase = tcp_now; 1051 1052 tp->t_congctl = tcp_congctl_global; 1053 tp->t_congctl->refcnt++; 1054 1055 return (tp); 1056 } 1057 1058 /* 1059 * Drop a TCP connection, reporting 1060 * the specified error. If connection is synchronized, 1061 * then send a RST to peer. 1062 */ 1063 struct tcpcb * 1064 tcp_drop(struct tcpcb *tp, int errno) 1065 { 1066 struct socket *so = NULL; 1067 1068 #ifdef DIAGNOSTIC 1069 if (tp->t_inpcb && tp->t_in6pcb) 1070 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1071 #endif 1072 #ifdef INET 1073 if (tp->t_inpcb) 1074 so = tp->t_inpcb->inp_socket; 1075 #endif 1076 #ifdef INET6 1077 if (tp->t_in6pcb) 1078 so = tp->t_in6pcb->in6p_socket; 1079 #endif 1080 if (!so) 1081 return NULL; 1082 1083 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1084 tp->t_state = TCPS_CLOSED; 1085 (void) tcp_output(tp); 1086 tcpstat.tcps_drops++; 1087 } else 1088 tcpstat.tcps_conndrops++; 1089 if (errno == ETIMEDOUT && tp->t_softerror) 1090 errno = tp->t_softerror; 1091 so->so_error = errno; 1092 return (tcp_close(tp)); 1093 } 1094 1095 /* 1096 * Return whether this tcpcb is marked as dead, indicating 1097 * to the calling timer function that no further action should 1098 * be taken, as we are about to release this tcpcb. The release 1099 * of the storage will be done if this is the last timer running. 1100 * 1101 * This should be called from the callout handler function after 1102 * callout_ack() is done, so that the number of invoking timer 1103 * functions is 0. 1104 */ 1105 int 1106 tcp_isdead(struct tcpcb *tp) 1107 { 1108 int i, dead = (tp->t_flags & TF_DEAD); 1109 1110 if (__predict_false(dead)) { 1111 if (tcp_timers_invoking(tp) > 0) 1112 /* not quite there yet -- count separately? */ 1113 return dead; 1114 tcpstat.tcps_delayed_free++; 1115 for (i = 0; i < TCPT_NTIMERS; i++) 1116 callout_destroy(&tp->t_timer[i]); 1117 callout_destroy(&tp->t_delack_ch); 1118 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_timer.c */ 1119 } 1120 return dead; 1121 } 1122 1123 /* 1124 * Close a TCP control block: 1125 * discard all space held by the tcp 1126 * discard internet protocol block 1127 * wake up any sleepers 1128 */ 1129 struct tcpcb * 1130 tcp_close(struct tcpcb *tp) 1131 { 1132 struct inpcb *inp; 1133 #ifdef INET6 1134 struct in6pcb *in6p; 1135 #endif 1136 struct socket *so; 1137 #ifdef RTV_RTT 1138 struct rtentry *rt; 1139 #endif 1140 struct route *ro; 1141 int j; 1142 1143 inp = tp->t_inpcb; 1144 #ifdef INET6 1145 in6p = tp->t_in6pcb; 1146 #endif 1147 so = NULL; 1148 ro = NULL; 1149 if (inp) { 1150 so = inp->inp_socket; 1151 ro = &inp->inp_route; 1152 } 1153 #ifdef INET6 1154 else if (in6p) { 1155 so = in6p->in6p_socket; 1156 ro = (struct route *)&in6p->in6p_route; 1157 } 1158 #endif 1159 1160 #ifdef RTV_RTT 1161 /* 1162 * If we sent enough data to get some meaningful characteristics, 1163 * save them in the routing entry. 'Enough' is arbitrarily 1164 * defined as the sendpipesize (default 4K) * 16. This would 1165 * give us 16 rtt samples assuming we only get one sample per 1166 * window (the usual case on a long haul net). 16 samples is 1167 * enough for the srtt filter to converge to within 5% of the correct 1168 * value; fewer samples and we could save a very bogus rtt. 1169 * 1170 * Don't update the default route's characteristics and don't 1171 * update anything that the user "locked". 1172 */ 1173 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1174 ro && (rt = ro->ro_rt) && 1175 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1176 u_long i = 0; 1177 1178 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1179 i = tp->t_srtt * 1180 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1181 if (rt->rt_rmx.rmx_rtt && i) 1182 /* 1183 * filter this update to half the old & half 1184 * the new values, converting scale. 1185 * See route.h and tcp_var.h for a 1186 * description of the scaling constants. 1187 */ 1188 rt->rt_rmx.rmx_rtt = 1189 (rt->rt_rmx.rmx_rtt + i) / 2; 1190 else 1191 rt->rt_rmx.rmx_rtt = i; 1192 } 1193 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1194 i = tp->t_rttvar * 1195 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1196 if (rt->rt_rmx.rmx_rttvar && i) 1197 rt->rt_rmx.rmx_rttvar = 1198 (rt->rt_rmx.rmx_rttvar + i) / 2; 1199 else 1200 rt->rt_rmx.rmx_rttvar = i; 1201 } 1202 /* 1203 * update the pipelimit (ssthresh) if it has been updated 1204 * already or if a pipesize was specified & the threshhold 1205 * got below half the pipesize. I.e., wait for bad news 1206 * before we start updating, then update on both good 1207 * and bad news. 1208 */ 1209 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1210 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1211 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1212 /* 1213 * convert the limit from user data bytes to 1214 * packets then to packet data bytes. 1215 */ 1216 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1217 if (i < 2) 1218 i = 2; 1219 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1220 if (rt->rt_rmx.rmx_ssthresh) 1221 rt->rt_rmx.rmx_ssthresh = 1222 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1223 else 1224 rt->rt_rmx.rmx_ssthresh = i; 1225 } 1226 } 1227 #endif /* RTV_RTT */ 1228 /* free the reassembly queue, if any */ 1229 TCP_REASS_LOCK(tp); 1230 (void) tcp_freeq(tp); 1231 TCP_REASS_UNLOCK(tp); 1232 1233 /* free the SACK holes list. */ 1234 tcp_free_sackholes(tp); 1235 1236 tp->t_congctl->refcnt--; 1237 1238 tcp_canceltimers(tp); 1239 TCP_CLEAR_DELACK(tp); 1240 syn_cache_cleanup(tp); 1241 1242 if (tp->t_template) { 1243 m_free(tp->t_template); 1244 tp->t_template = NULL; 1245 } 1246 if (tcp_timers_invoking(tp)) 1247 tp->t_flags |= TF_DEAD; 1248 else { 1249 for (j = 0; j < TCPT_NTIMERS; j++) 1250 callout_destroy(&tp->t_timer[j]); 1251 callout_destroy(&tp->t_delack_ch); 1252 pool_put(&tcpcb_pool, tp); 1253 } 1254 1255 if (inp) { 1256 inp->inp_ppcb = 0; 1257 soisdisconnected(so); 1258 in_pcbdetach(inp); 1259 } 1260 #ifdef INET6 1261 else if (in6p) { 1262 in6p->in6p_ppcb = 0; 1263 soisdisconnected(so); 1264 in6_pcbdetach(in6p); 1265 } 1266 #endif 1267 tcpstat.tcps_closed++; 1268 return ((struct tcpcb *)0); 1269 } 1270 1271 int 1272 tcp_freeq(tp) 1273 struct tcpcb *tp; 1274 { 1275 struct ipqent *qe; 1276 int rv = 0; 1277 #ifdef TCPREASS_DEBUG 1278 int i = 0; 1279 #endif 1280 1281 TCP_REASS_LOCK_CHECK(tp); 1282 1283 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1284 #ifdef TCPREASS_DEBUG 1285 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1286 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1287 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1288 #endif 1289 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1290 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1291 m_freem(qe->ipqe_m); 1292 tcpipqent_free(qe); 1293 rv = 1; 1294 } 1295 tp->t_segqlen = 0; 1296 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1297 return (rv); 1298 } 1299 1300 /* 1301 * Protocol drain routine. Called when memory is in short supply. 1302 */ 1303 void 1304 tcp_drain(void) 1305 { 1306 struct inpcb_hdr *inph; 1307 struct tcpcb *tp; 1308 1309 /* 1310 * Free the sequence queue of all TCP connections. 1311 */ 1312 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1313 switch (inph->inph_af) { 1314 case AF_INET: 1315 tp = intotcpcb((struct inpcb *)inph); 1316 break; 1317 #ifdef INET6 1318 case AF_INET6: 1319 tp = in6totcpcb((struct in6pcb *)inph); 1320 break; 1321 #endif 1322 default: 1323 tp = NULL; 1324 break; 1325 } 1326 if (tp != NULL) { 1327 /* 1328 * We may be called from a device's interrupt 1329 * context. If the tcpcb is already busy, 1330 * just bail out now. 1331 */ 1332 if (tcp_reass_lock_try(tp) == 0) 1333 continue; 1334 if (tcp_freeq(tp)) 1335 tcpstat.tcps_connsdrained++; 1336 TCP_REASS_UNLOCK(tp); 1337 } 1338 } 1339 } 1340 1341 /* 1342 * Notify a tcp user of an asynchronous error; 1343 * store error as soft error, but wake up user 1344 * (for now, won't do anything until can select for soft error). 1345 */ 1346 void 1347 tcp_notify(struct inpcb *inp, int error) 1348 { 1349 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1350 struct socket *so = inp->inp_socket; 1351 1352 /* 1353 * Ignore some errors if we are hooked up. 1354 * If connection hasn't completed, has retransmitted several times, 1355 * and receives a second error, give up now. This is better 1356 * than waiting a long time to establish a connection that 1357 * can never complete. 1358 */ 1359 if (tp->t_state == TCPS_ESTABLISHED && 1360 (error == EHOSTUNREACH || error == ENETUNREACH || 1361 error == EHOSTDOWN)) { 1362 return; 1363 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1364 tp->t_rxtshift > 3 && tp->t_softerror) 1365 so->so_error = error; 1366 else 1367 tp->t_softerror = error; 1368 wakeup((void *) &so->so_timeo); 1369 sorwakeup(so); 1370 sowwakeup(so); 1371 } 1372 1373 #ifdef INET6 1374 void 1375 tcp6_notify(struct in6pcb *in6p, int error) 1376 { 1377 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1378 struct socket *so = in6p->in6p_socket; 1379 1380 /* 1381 * Ignore some errors if we are hooked up. 1382 * If connection hasn't completed, has retransmitted several times, 1383 * and receives a second error, give up now. This is better 1384 * than waiting a long time to establish a connection that 1385 * can never complete. 1386 */ 1387 if (tp->t_state == TCPS_ESTABLISHED && 1388 (error == EHOSTUNREACH || error == ENETUNREACH || 1389 error == EHOSTDOWN)) { 1390 return; 1391 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1392 tp->t_rxtshift > 3 && tp->t_softerror) 1393 so->so_error = error; 1394 else 1395 tp->t_softerror = error; 1396 wakeup((void *) &so->so_timeo); 1397 sorwakeup(so); 1398 sowwakeup(so); 1399 } 1400 #endif 1401 1402 #ifdef INET6 1403 void 1404 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1405 { 1406 struct tcphdr th; 1407 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1408 int nmatch; 1409 struct ip6_hdr *ip6; 1410 const struct sockaddr_in6 *sa6_src = NULL; 1411 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1412 struct mbuf *m; 1413 int off; 1414 1415 if (sa->sa_family != AF_INET6 || 1416 sa->sa_len != sizeof(struct sockaddr_in6)) 1417 return; 1418 if ((unsigned)cmd >= PRC_NCMDS) 1419 return; 1420 else if (cmd == PRC_QUENCH) { 1421 /* 1422 * Don't honor ICMP Source Quench messages meant for 1423 * TCP connections. 1424 */ 1425 return; 1426 } else if (PRC_IS_REDIRECT(cmd)) 1427 notify = in6_rtchange, d = NULL; 1428 else if (cmd == PRC_MSGSIZE) 1429 ; /* special code is present, see below */ 1430 else if (cmd == PRC_HOSTDEAD) 1431 d = NULL; 1432 else if (inet6ctlerrmap[cmd] == 0) 1433 return; 1434 1435 /* if the parameter is from icmp6, decode it. */ 1436 if (d != NULL) { 1437 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1438 m = ip6cp->ip6c_m; 1439 ip6 = ip6cp->ip6c_ip6; 1440 off = ip6cp->ip6c_off; 1441 sa6_src = ip6cp->ip6c_src; 1442 } else { 1443 m = NULL; 1444 ip6 = NULL; 1445 sa6_src = &sa6_any; 1446 off = 0; 1447 } 1448 1449 if (ip6) { 1450 /* 1451 * XXX: We assume that when ip6 is non NULL, 1452 * M and OFF are valid. 1453 */ 1454 1455 /* check if we can safely examine src and dst ports */ 1456 if (m->m_pkthdr.len < off + sizeof(th)) { 1457 if (cmd == PRC_MSGSIZE) 1458 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1459 return; 1460 } 1461 1462 bzero(&th, sizeof(th)); 1463 m_copydata(m, off, sizeof(th), (void *)&th); 1464 1465 if (cmd == PRC_MSGSIZE) { 1466 int valid = 0; 1467 1468 /* 1469 * Check to see if we have a valid TCP connection 1470 * corresponding to the address in the ICMPv6 message 1471 * payload. 1472 */ 1473 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1474 th.th_dport, 1475 (const struct in6_addr *)&sa6_src->sin6_addr, 1476 th.th_sport, 0)) 1477 valid++; 1478 1479 /* 1480 * Depending on the value of "valid" and routing table 1481 * size (mtudisc_{hi,lo}wat), we will: 1482 * - recalcurate the new MTU and create the 1483 * corresponding routing entry, or 1484 * - ignore the MTU change notification. 1485 */ 1486 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1487 1488 /* 1489 * no need to call in6_pcbnotify, it should have been 1490 * called via callback if necessary 1491 */ 1492 return; 1493 } 1494 1495 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1496 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1497 if (nmatch == 0 && syn_cache_count && 1498 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1499 inet6ctlerrmap[cmd] == ENETUNREACH || 1500 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1501 syn_cache_unreach((const struct sockaddr *)sa6_src, 1502 sa, &th); 1503 } else { 1504 (void) in6_pcbnotify(&tcbtable, sa, 0, 1505 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1506 } 1507 } 1508 #endif 1509 1510 #ifdef INET 1511 /* assumes that ip header and tcp header are contiguous on mbuf */ 1512 void * 1513 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1514 { 1515 struct ip *ip = v; 1516 struct tcphdr *th; 1517 struct icmp *icp; 1518 extern const int inetctlerrmap[]; 1519 void (*notify)(struct inpcb *, int) = tcp_notify; 1520 int errno; 1521 int nmatch; 1522 struct tcpcb *tp; 1523 u_int mtu; 1524 tcp_seq seq; 1525 struct inpcb *inp; 1526 #ifdef INET6 1527 struct in6pcb *in6p; 1528 struct in6_addr src6, dst6; 1529 #endif 1530 1531 if (sa->sa_family != AF_INET || 1532 sa->sa_len != sizeof(struct sockaddr_in)) 1533 return NULL; 1534 if ((unsigned)cmd >= PRC_NCMDS) 1535 return NULL; 1536 errno = inetctlerrmap[cmd]; 1537 if (cmd == PRC_QUENCH) 1538 /* 1539 * Don't honor ICMP Source Quench messages meant for 1540 * TCP connections. 1541 */ 1542 return NULL; 1543 else if (PRC_IS_REDIRECT(cmd)) 1544 notify = in_rtchange, ip = 0; 1545 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1546 /* 1547 * Check to see if we have a valid TCP connection 1548 * corresponding to the address in the ICMP message 1549 * payload. 1550 * 1551 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1552 */ 1553 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1554 #ifdef INET6 1555 memset(&src6, 0, sizeof(src6)); 1556 memset(&dst6, 0, sizeof(dst6)); 1557 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1558 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1559 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1560 #endif 1561 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1562 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1563 #ifdef INET6 1564 in6p = NULL; 1565 #else 1566 ; 1567 #endif 1568 #ifdef INET6 1569 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1570 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1571 ; 1572 #endif 1573 else 1574 return NULL; 1575 1576 /* 1577 * Now that we've validated that we are actually communicating 1578 * with the host indicated in the ICMP message, locate the 1579 * ICMP header, recalculate the new MTU, and create the 1580 * corresponding routing entry. 1581 */ 1582 icp = (struct icmp *)((char *)ip - 1583 offsetof(struct icmp, icmp_ip)); 1584 if (inp) { 1585 if ((tp = intotcpcb(inp)) == NULL) 1586 return NULL; 1587 } 1588 #ifdef INET6 1589 else if (in6p) { 1590 if ((tp = in6totcpcb(in6p)) == NULL) 1591 return NULL; 1592 } 1593 #endif 1594 else 1595 return NULL; 1596 seq = ntohl(th->th_seq); 1597 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1598 return NULL; 1599 /* 1600 * If the ICMP message advertises a Next-Hop MTU 1601 * equal or larger than the maximum packet size we have 1602 * ever sent, drop the message. 1603 */ 1604 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1605 if (mtu >= tp->t_pmtud_mtu_sent) 1606 return NULL; 1607 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1608 /* 1609 * Calculate new MTU, and create corresponding 1610 * route (traditional PMTUD). 1611 */ 1612 tp->t_flags &= ~TF_PMTUD_PEND; 1613 icmp_mtudisc(icp, ip->ip_dst); 1614 } else { 1615 /* 1616 * Record the information got in the ICMP 1617 * message; act on it later. 1618 * If we had already recorded an ICMP message, 1619 * replace the old one only if the new message 1620 * refers to an older TCP segment 1621 */ 1622 if (tp->t_flags & TF_PMTUD_PEND) { 1623 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1624 return NULL; 1625 } else 1626 tp->t_flags |= TF_PMTUD_PEND; 1627 tp->t_pmtud_th_seq = seq; 1628 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1629 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1630 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1631 } 1632 return NULL; 1633 } else if (cmd == PRC_HOSTDEAD) 1634 ip = 0; 1635 else if (errno == 0) 1636 return NULL; 1637 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1638 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1639 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1640 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1641 if (nmatch == 0 && syn_cache_count && 1642 (inetctlerrmap[cmd] == EHOSTUNREACH || 1643 inetctlerrmap[cmd] == ENETUNREACH || 1644 inetctlerrmap[cmd] == EHOSTDOWN)) { 1645 struct sockaddr_in sin; 1646 bzero(&sin, sizeof(sin)); 1647 sin.sin_len = sizeof(sin); 1648 sin.sin_family = AF_INET; 1649 sin.sin_port = th->th_sport; 1650 sin.sin_addr = ip->ip_src; 1651 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1652 } 1653 1654 /* XXX mapped address case */ 1655 } else 1656 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1657 notify); 1658 return NULL; 1659 } 1660 1661 /* 1662 * When a source quench is received, we are being notified of congestion. 1663 * Close the congestion window down to the Loss Window (one segment). 1664 * We will gradually open it again as we proceed. 1665 */ 1666 void 1667 tcp_quench(struct inpcb *inp, int errno) 1668 { 1669 struct tcpcb *tp = intotcpcb(inp); 1670 1671 if (tp) { 1672 tp->snd_cwnd = tp->t_segsz; 1673 tp->t_bytes_acked = 0; 1674 } 1675 } 1676 #endif 1677 1678 #ifdef INET6 1679 void 1680 tcp6_quench(struct in6pcb *in6p, int errno) 1681 { 1682 struct tcpcb *tp = in6totcpcb(in6p); 1683 1684 if (tp) { 1685 tp->snd_cwnd = tp->t_segsz; 1686 tp->t_bytes_acked = 0; 1687 } 1688 } 1689 #endif 1690 1691 #ifdef INET 1692 /* 1693 * Path MTU Discovery handlers. 1694 */ 1695 void 1696 tcp_mtudisc_callback(struct in_addr faddr) 1697 { 1698 #ifdef INET6 1699 struct in6_addr in6; 1700 #endif 1701 1702 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1703 #ifdef INET6 1704 memset(&in6, 0, sizeof(in6)); 1705 in6.s6_addr16[5] = 0xffff; 1706 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1707 tcp6_mtudisc_callback(&in6); 1708 #endif 1709 } 1710 1711 /* 1712 * On receipt of path MTU corrections, flush old route and replace it 1713 * with the new one. Retransmit all unacknowledged packets, to ensure 1714 * that all packets will be received. 1715 */ 1716 void 1717 tcp_mtudisc(struct inpcb *inp, int errno) 1718 { 1719 struct tcpcb *tp = intotcpcb(inp); 1720 struct rtentry *rt = in_pcbrtentry(inp); 1721 1722 if (tp != 0) { 1723 if (rt != 0) { 1724 /* 1725 * If this was not a host route, remove and realloc. 1726 */ 1727 if ((rt->rt_flags & RTF_HOST) == 0) { 1728 in_rtchange(inp, errno); 1729 if ((rt = in_pcbrtentry(inp)) == 0) 1730 return; 1731 } 1732 1733 /* 1734 * Slow start out of the error condition. We 1735 * use the MTU because we know it's smaller 1736 * than the previously transmitted segment. 1737 * 1738 * Note: This is more conservative than the 1739 * suggestion in draft-floyd-incr-init-win-03. 1740 */ 1741 if (rt->rt_rmx.rmx_mtu != 0) 1742 tp->snd_cwnd = 1743 TCP_INITIAL_WINDOW(tcp_init_win, 1744 rt->rt_rmx.rmx_mtu); 1745 } 1746 1747 /* 1748 * Resend unacknowledged packets. 1749 */ 1750 tp->snd_nxt = tp->snd_una; 1751 tcp_output(tp); 1752 } 1753 } 1754 #endif 1755 1756 #ifdef INET6 1757 /* 1758 * Path MTU Discovery handlers. 1759 */ 1760 void 1761 tcp6_mtudisc_callback(struct in6_addr *faddr) 1762 { 1763 struct sockaddr_in6 sin6; 1764 1765 bzero(&sin6, sizeof(sin6)); 1766 sin6.sin6_family = AF_INET6; 1767 sin6.sin6_len = sizeof(struct sockaddr_in6); 1768 sin6.sin6_addr = *faddr; 1769 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1770 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1771 } 1772 1773 void 1774 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1775 { 1776 struct tcpcb *tp = in6totcpcb(in6p); 1777 struct rtentry *rt = in6_pcbrtentry(in6p); 1778 1779 if (tp != 0) { 1780 if (rt != 0) { 1781 /* 1782 * If this was not a host route, remove and realloc. 1783 */ 1784 if ((rt->rt_flags & RTF_HOST) == 0) { 1785 in6_rtchange(in6p, errno); 1786 if ((rt = in6_pcbrtentry(in6p)) == 0) 1787 return; 1788 } 1789 1790 /* 1791 * Slow start out of the error condition. We 1792 * use the MTU because we know it's smaller 1793 * than the previously transmitted segment. 1794 * 1795 * Note: This is more conservative than the 1796 * suggestion in draft-floyd-incr-init-win-03. 1797 */ 1798 if (rt->rt_rmx.rmx_mtu != 0) 1799 tp->snd_cwnd = 1800 TCP_INITIAL_WINDOW(tcp_init_win, 1801 rt->rt_rmx.rmx_mtu); 1802 } 1803 1804 /* 1805 * Resend unacknowledged packets. 1806 */ 1807 tp->snd_nxt = tp->snd_una; 1808 tcp_output(tp); 1809 } 1810 } 1811 #endif /* INET6 */ 1812 1813 /* 1814 * Compute the MSS to advertise to the peer. Called only during 1815 * the 3-way handshake. If we are the server (peer initiated 1816 * connection), we are called with a pointer to the interface 1817 * on which the SYN packet arrived. If we are the client (we 1818 * initiated connection), we are called with a pointer to the 1819 * interface out which this connection should go. 1820 * 1821 * NOTE: Do not subtract IP option/extension header size nor IPsec 1822 * header size from MSS advertisement. MSS option must hold the maximum 1823 * segment size we can accept, so it must always be: 1824 * max(if mtu) - ip header - tcp header 1825 */ 1826 u_long 1827 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1828 { 1829 extern u_long in_maxmtu; 1830 u_long mss = 0; 1831 u_long hdrsiz; 1832 1833 /* 1834 * In order to avoid defeating path MTU discovery on the peer, 1835 * we advertise the max MTU of all attached networks as our MSS, 1836 * per RFC 1191, section 3.1. 1837 * 1838 * We provide the option to advertise just the MTU of 1839 * the interface on which we hope this connection will 1840 * be receiving. If we are responding to a SYN, we 1841 * will have a pretty good idea about this, but when 1842 * initiating a connection there is a bit more doubt. 1843 * 1844 * We also need to ensure that loopback has a large enough 1845 * MSS, as the loopback MTU is never included in in_maxmtu. 1846 */ 1847 1848 if (ifp != NULL) 1849 switch (af) { 1850 case AF_INET: 1851 mss = ifp->if_mtu; 1852 break; 1853 #ifdef INET6 1854 case AF_INET6: 1855 mss = IN6_LINKMTU(ifp); 1856 break; 1857 #endif 1858 } 1859 1860 if (tcp_mss_ifmtu == 0) 1861 switch (af) { 1862 case AF_INET: 1863 mss = max(in_maxmtu, mss); 1864 break; 1865 #ifdef INET6 1866 case AF_INET6: 1867 mss = max(in6_maxmtu, mss); 1868 break; 1869 #endif 1870 } 1871 1872 switch (af) { 1873 case AF_INET: 1874 hdrsiz = sizeof(struct ip); 1875 break; 1876 #ifdef INET6 1877 case AF_INET6: 1878 hdrsiz = sizeof(struct ip6_hdr); 1879 break; 1880 #endif 1881 default: 1882 hdrsiz = 0; 1883 break; 1884 } 1885 hdrsiz += sizeof(struct tcphdr); 1886 if (mss > hdrsiz) 1887 mss -= hdrsiz; 1888 1889 mss = max(tcp_mssdflt, mss); 1890 return (mss); 1891 } 1892 1893 /* 1894 * Set connection variables based on the peer's advertised MSS. 1895 * We are passed the TCPCB for the actual connection. If we 1896 * are the server, we are called by the compressed state engine 1897 * when the 3-way handshake is complete. If we are the client, 1898 * we are called when we receive the SYN,ACK from the server. 1899 * 1900 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1901 * before this routine is called! 1902 */ 1903 void 1904 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1905 { 1906 struct socket *so; 1907 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1908 struct rtentry *rt; 1909 #endif 1910 u_long bufsize; 1911 int mss; 1912 1913 #ifdef DIAGNOSTIC 1914 if (tp->t_inpcb && tp->t_in6pcb) 1915 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1916 #endif 1917 so = NULL; 1918 rt = NULL; 1919 #ifdef INET 1920 if (tp->t_inpcb) { 1921 so = tp->t_inpcb->inp_socket; 1922 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1923 rt = in_pcbrtentry(tp->t_inpcb); 1924 #endif 1925 } 1926 #endif 1927 #ifdef INET6 1928 if (tp->t_in6pcb) { 1929 so = tp->t_in6pcb->in6p_socket; 1930 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1931 rt = in6_pcbrtentry(tp->t_in6pcb); 1932 #endif 1933 } 1934 #endif 1935 1936 /* 1937 * As per RFC1122, use the default MSS value, unless they 1938 * sent us an offer. Do not accept offers less than 256 bytes. 1939 */ 1940 mss = tcp_mssdflt; 1941 if (offer) 1942 mss = offer; 1943 mss = max(mss, 256); /* sanity */ 1944 tp->t_peermss = mss; 1945 mss -= tcp_optlen(tp); 1946 #ifdef INET 1947 if (tp->t_inpcb) 1948 mss -= ip_optlen(tp->t_inpcb); 1949 #endif 1950 #ifdef INET6 1951 if (tp->t_in6pcb) 1952 mss -= ip6_optlen(tp->t_in6pcb); 1953 #endif 1954 1955 /* 1956 * If there's a pipesize, change the socket buffer to that size. 1957 * Make the socket buffer an integral number of MSS units. If 1958 * the MSS is larger than the socket buffer, artificially decrease 1959 * the MSS. 1960 */ 1961 #ifdef RTV_SPIPE 1962 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1963 bufsize = rt->rt_rmx.rmx_sendpipe; 1964 else 1965 #endif 1966 { 1967 KASSERT(so != NULL); 1968 bufsize = so->so_snd.sb_hiwat; 1969 } 1970 if (bufsize < mss) 1971 mss = bufsize; 1972 else { 1973 bufsize = roundup(bufsize, mss); 1974 if (bufsize > sb_max) 1975 bufsize = sb_max; 1976 (void) sbreserve(&so->so_snd, bufsize, so); 1977 } 1978 tp->t_segsz = mss; 1979 1980 #ifdef RTV_SSTHRESH 1981 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1982 /* 1983 * There's some sort of gateway or interface buffer 1984 * limit on the path. Use this to set the slow 1985 * start threshold, but set the threshold to no less 1986 * than 2 * MSS. 1987 */ 1988 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1989 } 1990 #endif 1991 } 1992 1993 /* 1994 * Processing necessary when a TCP connection is established. 1995 */ 1996 void 1997 tcp_established(struct tcpcb *tp) 1998 { 1999 struct socket *so; 2000 #ifdef RTV_RPIPE 2001 struct rtentry *rt; 2002 #endif 2003 u_long bufsize; 2004 2005 #ifdef DIAGNOSTIC 2006 if (tp->t_inpcb && tp->t_in6pcb) 2007 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2008 #endif 2009 so = NULL; 2010 rt = NULL; 2011 #ifdef INET 2012 if (tp->t_inpcb) { 2013 so = tp->t_inpcb->inp_socket; 2014 #if defined(RTV_RPIPE) 2015 rt = in_pcbrtentry(tp->t_inpcb); 2016 #endif 2017 } 2018 #endif 2019 #ifdef INET6 2020 if (tp->t_in6pcb) { 2021 so = tp->t_in6pcb->in6p_socket; 2022 #if defined(RTV_RPIPE) 2023 rt = in6_pcbrtentry(tp->t_in6pcb); 2024 #endif 2025 } 2026 #endif 2027 2028 tp->t_state = TCPS_ESTABLISHED; 2029 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2030 2031 #ifdef RTV_RPIPE 2032 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2033 bufsize = rt->rt_rmx.rmx_recvpipe; 2034 else 2035 #endif 2036 { 2037 KASSERT(so != NULL); 2038 bufsize = so->so_rcv.sb_hiwat; 2039 } 2040 if (bufsize > tp->t_ourmss) { 2041 bufsize = roundup(bufsize, tp->t_ourmss); 2042 if (bufsize > sb_max) 2043 bufsize = sb_max; 2044 (void) sbreserve(&so->so_rcv, bufsize, so); 2045 } 2046 } 2047 2048 /* 2049 * Check if there's an initial rtt or rttvar. Convert from the 2050 * route-table units to scaled multiples of the slow timeout timer. 2051 * Called only during the 3-way handshake. 2052 */ 2053 void 2054 tcp_rmx_rtt(struct tcpcb *tp) 2055 { 2056 #ifdef RTV_RTT 2057 struct rtentry *rt = NULL; 2058 int rtt; 2059 2060 #ifdef DIAGNOSTIC 2061 if (tp->t_inpcb && tp->t_in6pcb) 2062 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2063 #endif 2064 #ifdef INET 2065 if (tp->t_inpcb) 2066 rt = in_pcbrtentry(tp->t_inpcb); 2067 #endif 2068 #ifdef INET6 2069 if (tp->t_in6pcb) 2070 rt = in6_pcbrtentry(tp->t_in6pcb); 2071 #endif 2072 if (rt == NULL) 2073 return; 2074 2075 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2076 /* 2077 * XXX The lock bit for MTU indicates that the value 2078 * is also a minimum value; this is subject to time. 2079 */ 2080 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2081 TCPT_RANGESET(tp->t_rttmin, 2082 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2083 TCPTV_MIN, TCPTV_REXMTMAX); 2084 tp->t_srtt = rtt / 2085 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2086 if (rt->rt_rmx.rmx_rttvar) { 2087 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2088 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2089 (TCP_RTTVAR_SHIFT + 2)); 2090 } else { 2091 /* Default variation is +- 1 rtt */ 2092 tp->t_rttvar = 2093 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2094 } 2095 TCPT_RANGESET(tp->t_rxtcur, 2096 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2097 tp->t_rttmin, TCPTV_REXMTMAX); 2098 } 2099 #endif 2100 } 2101 2102 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2103 #if NRND > 0 2104 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2105 #endif 2106 2107 /* 2108 * Get a new sequence value given a tcp control block 2109 */ 2110 tcp_seq 2111 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2112 { 2113 2114 #ifdef INET 2115 if (tp->t_inpcb != NULL) { 2116 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2117 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2118 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2119 addin)); 2120 } 2121 #endif 2122 #ifdef INET6 2123 if (tp->t_in6pcb != NULL) { 2124 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2125 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2126 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2127 addin)); 2128 } 2129 #endif 2130 /* Not possible. */ 2131 panic("tcp_new_iss"); 2132 } 2133 2134 /* 2135 * This routine actually generates a new TCP initial sequence number. 2136 */ 2137 tcp_seq 2138 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2139 size_t addrsz, tcp_seq addin) 2140 { 2141 tcp_seq tcp_iss; 2142 2143 #if NRND > 0 2144 static int beenhere; 2145 2146 /* 2147 * If we haven't been here before, initialize our cryptographic 2148 * hash secret. 2149 */ 2150 if (beenhere == 0) { 2151 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2152 RND_EXTRACT_ANY); 2153 beenhere = 1; 2154 } 2155 2156 if (tcp_do_rfc1948) { 2157 MD5_CTX ctx; 2158 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2159 2160 /* 2161 * Compute the base value of the ISS. It is a hash 2162 * of (saddr, sport, daddr, dport, secret). 2163 */ 2164 MD5Init(&ctx); 2165 2166 MD5Update(&ctx, (u_char *) laddr, addrsz); 2167 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2168 2169 MD5Update(&ctx, (u_char *) faddr, addrsz); 2170 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2171 2172 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2173 2174 MD5Final(hash, &ctx); 2175 2176 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2177 2178 /* 2179 * Now increment our "timer", and add it in to 2180 * the computed value. 2181 * 2182 * XXX Use `addin'? 2183 * XXX TCP_ISSINCR too large to use? 2184 */ 2185 tcp_iss_seq += TCP_ISSINCR; 2186 #ifdef TCPISS_DEBUG 2187 printf("ISS hash 0x%08x, ", tcp_iss); 2188 #endif 2189 tcp_iss += tcp_iss_seq + addin; 2190 #ifdef TCPISS_DEBUG 2191 printf("new ISS 0x%08x\n", tcp_iss); 2192 #endif 2193 } else 2194 #endif /* NRND > 0 */ 2195 { 2196 /* 2197 * Randomize. 2198 */ 2199 #if NRND > 0 2200 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2201 #else 2202 tcp_iss = arc4random(); 2203 #endif 2204 2205 /* 2206 * If we were asked to add some amount to a known value, 2207 * we will take a random value obtained above, mask off 2208 * the upper bits, and add in the known value. We also 2209 * add in a constant to ensure that we are at least a 2210 * certain distance from the original value. 2211 * 2212 * This is used when an old connection is in timed wait 2213 * and we have a new one coming in, for instance. 2214 */ 2215 if (addin != 0) { 2216 #ifdef TCPISS_DEBUG 2217 printf("Random %08x, ", tcp_iss); 2218 #endif 2219 tcp_iss &= TCP_ISS_RANDOM_MASK; 2220 tcp_iss += addin + TCP_ISSINCR; 2221 #ifdef TCPISS_DEBUG 2222 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2223 #endif 2224 } else { 2225 tcp_iss &= TCP_ISS_RANDOM_MASK; 2226 tcp_iss += tcp_iss_seq; 2227 tcp_iss_seq += TCP_ISSINCR; 2228 #ifdef TCPISS_DEBUG 2229 printf("ISS %08x\n", tcp_iss); 2230 #endif 2231 } 2232 } 2233 2234 if (tcp_compat_42) { 2235 /* 2236 * Limit it to the positive range for really old TCP 2237 * implementations. 2238 * Just AND off the top bit instead of checking if 2239 * is set first - saves a branch 50% of the time. 2240 */ 2241 tcp_iss &= 0x7fffffff; /* XXX */ 2242 } 2243 2244 return (tcp_iss); 2245 } 2246 2247 #if defined(IPSEC) || defined(FAST_IPSEC) 2248 /* compute ESP/AH header size for TCP, including outer IP header. */ 2249 size_t 2250 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2251 { 2252 struct inpcb *inp; 2253 size_t hdrsiz; 2254 2255 /* XXX mapped addr case (tp->t_in6pcb) */ 2256 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2257 return 0; 2258 switch (tp->t_family) { 2259 case AF_INET: 2260 /* XXX: should use currect direction. */ 2261 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2262 break; 2263 default: 2264 hdrsiz = 0; 2265 break; 2266 } 2267 2268 return hdrsiz; 2269 } 2270 2271 #ifdef INET6 2272 size_t 2273 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2274 { 2275 struct in6pcb *in6p; 2276 size_t hdrsiz; 2277 2278 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2279 return 0; 2280 switch (tp->t_family) { 2281 case AF_INET6: 2282 /* XXX: should use currect direction. */ 2283 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2284 break; 2285 case AF_INET: 2286 /* mapped address case - tricky */ 2287 default: 2288 hdrsiz = 0; 2289 break; 2290 } 2291 2292 return hdrsiz; 2293 } 2294 #endif 2295 #endif /*IPSEC*/ 2296 2297 /* 2298 * Determine the length of the TCP options for this connection. 2299 * 2300 * XXX: What do we do for SACK, when we add that? Just reserve 2301 * all of the space? Otherwise we can't exactly be incrementing 2302 * cwnd by an amount that varies depending on the amount we last 2303 * had to SACK! 2304 */ 2305 2306 u_int 2307 tcp_optlen(struct tcpcb *tp) 2308 { 2309 u_int optlen; 2310 2311 optlen = 0; 2312 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2313 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2314 optlen += TCPOLEN_TSTAMP_APPA; 2315 2316 #ifdef TCP_SIGNATURE 2317 if (tp->t_flags & TF_SIGNATURE) 2318 optlen += TCPOLEN_SIGNATURE + 2; 2319 #endif /* TCP_SIGNATURE */ 2320 2321 return optlen; 2322 } 2323 2324 u_int 2325 tcp_hdrsz(struct tcpcb *tp) 2326 { 2327 u_int hlen; 2328 2329 switch (tp->t_family) { 2330 #ifdef INET6 2331 case AF_INET6: 2332 hlen = sizeof(struct ip6_hdr); 2333 break; 2334 #endif 2335 case AF_INET: 2336 hlen = sizeof(struct ip); 2337 break; 2338 default: 2339 hlen = 0; 2340 break; 2341 } 2342 hlen += sizeof(struct tcphdr); 2343 2344 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2345 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2346 hlen += TCPOLEN_TSTAMP_APPA; 2347 #ifdef TCP_SIGNATURE 2348 if (tp->t_flags & TF_SIGNATURE) 2349 hlen += TCPOLEN_SIGLEN; 2350 #endif 2351 return hlen; 2352 } 2353