1 /* $NetBSD: tcp_subr.c,v 1.211 2007/02/17 22:34:12 dyoung 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 1.211 2007/02/17 22:34:12 dyoung 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_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 179 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 180 #if NRND > 0 181 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 182 #endif 183 int tcp_do_sack = 1; /* selective acknowledgement */ 184 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 185 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 186 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 187 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 188 #ifndef TCP_INIT_WIN 189 #define TCP_INIT_WIN 0 /* initial slow start window */ 190 #endif 191 #ifndef TCP_INIT_WIN_LOCAL 192 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 193 #endif 194 int tcp_init_win = TCP_INIT_WIN; 195 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 196 int tcp_mss_ifmtu = 0; 197 #ifdef TCP_COMPAT_42 198 int tcp_compat_42 = 1; 199 #else 200 int tcp_compat_42 = 0; 201 #endif 202 int tcp_rst_ppslim = 100; /* 100pps */ 203 int tcp_ackdrop_ppslim = 100; /* 100pps */ 204 int tcp_do_loopback_cksum = 0; 205 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 206 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 207 int tcp_sack_tp_maxholes = 32; 208 int tcp_sack_globalmaxholes = 1024; 209 int tcp_sack_globalholes = 0; 210 int tcp_ecn_maxretries = 1; 211 212 /* tcb hash */ 213 #ifndef TCBHASHSIZE 214 #define TCBHASHSIZE 128 215 #endif 216 int tcbhashsize = TCBHASHSIZE; 217 218 /* syn hash parameters */ 219 #define TCP_SYN_HASH_SIZE 293 220 #define TCP_SYN_BUCKET_SIZE 35 221 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 222 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 223 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 224 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 225 226 int tcp_freeq(struct tcpcb *); 227 228 #ifdef INET 229 void tcp_mtudisc_callback(struct in_addr); 230 #endif 231 #ifdef INET6 232 void tcp6_mtudisc_callback(struct in6_addr *); 233 #endif 234 235 #ifdef INET6 236 void tcp6_mtudisc(struct in6pcb *, int); 237 #endif 238 239 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL); 240 241 #ifdef TCP_CSUM_COUNTERS 242 #include <sys/device.h> 243 244 #if defined(INET) 245 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 246 NULL, "tcp", "hwcsum bad"); 247 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "hwcsum ok"); 249 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 250 NULL, "tcp", "hwcsum data"); 251 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 252 NULL, "tcp", "swcsum"); 253 254 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 255 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 256 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 257 EVCNT_ATTACH_STATIC(tcp_swcsum); 258 #endif /* defined(INET) */ 259 260 #if defined(INET6) 261 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 NULL, "tcp6", "hwcsum bad"); 263 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "hwcsum ok"); 265 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 NULL, "tcp6", "hwcsum data"); 267 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 268 NULL, "tcp6", "swcsum"); 269 270 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 271 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 272 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 273 EVCNT_ATTACH_STATIC(tcp6_swcsum); 274 #endif /* defined(INET6) */ 275 #endif /* TCP_CSUM_COUNTERS */ 276 277 278 #ifdef TCP_OUTPUT_COUNTERS 279 #include <sys/device.h> 280 281 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 NULL, "tcp", "output big header"); 283 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output predict hit"); 285 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output predict miss"); 287 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output copy small"); 289 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 NULL, "tcp", "output copy big"); 291 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 NULL, "tcp", "output reference big"); 293 294 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 295 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 296 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 297 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 298 EVCNT_ATTACH_STATIC(tcp_output_copybig); 299 EVCNT_ATTACH_STATIC(tcp_output_refbig); 300 301 #endif /* TCP_OUTPUT_COUNTERS */ 302 303 #ifdef TCP_REASS_COUNTERS 304 #include <sys/device.h> 305 306 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 307 NULL, "tcp_reass", "calls"); 308 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 309 &tcp_reass_, "tcp_reass", "insert into empty queue"); 310 struct evcnt tcp_reass_iteration[8] = { 311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 318 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 319 }; 320 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 321 &tcp_reass_, "tcp_reass", "prepend to first"); 322 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "prepend"); 324 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "insert"); 326 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "insert at tail"); 328 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "append"); 330 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "append to tail fragment"); 332 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "overlap at end"); 334 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "overlap at start"); 336 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 337 &tcp_reass_, "tcp_reass", "duplicate segment"); 338 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 339 &tcp_reass_, "tcp_reass", "duplicate fragment"); 340 341 EVCNT_ATTACH_STATIC(tcp_reass_); 342 EVCNT_ATTACH_STATIC(tcp_reass_empty); 343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 349 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 350 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 351 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 352 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 353 EVCNT_ATTACH_STATIC(tcp_reass_insert); 354 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 355 EVCNT_ATTACH_STATIC(tcp_reass_append); 356 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 357 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 358 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 359 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 360 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 361 362 #endif /* TCP_REASS_COUNTERS */ 363 364 #ifdef MBUFTRACE 365 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 366 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 367 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 368 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 369 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 370 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 371 #endif 372 373 /* 374 * Tcp initialization 375 */ 376 void 377 tcp_init(void) 378 { 379 int hlen; 380 381 /* Initialize the TCPCB template. */ 382 tcp_tcpcb_template(); 383 384 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 385 386 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 387 #ifdef INET6 388 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 389 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 390 #endif 391 if (max_protohdr < hlen) 392 max_protohdr = hlen; 393 if (max_linkhdr + hlen > MHLEN) 394 panic("tcp_init"); 395 396 #ifdef INET 397 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 398 #endif 399 #ifdef INET6 400 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 401 #endif 402 403 /* Initialize timer state. */ 404 tcp_timer_init(); 405 406 /* Initialize the compressed state engine. */ 407 syn_cache_init(); 408 409 /* Initialize the congestion control algorithms. */ 410 tcp_congctl_init(); 411 412 MOWNER_ATTACH(&tcp_tx_mowner); 413 MOWNER_ATTACH(&tcp_rx_mowner); 414 MOWNER_ATTACH(&tcp_reass_mowner); 415 MOWNER_ATTACH(&tcp_sock_mowner); 416 MOWNER_ATTACH(&tcp_sock_tx_mowner); 417 MOWNER_ATTACH(&tcp_sock_rx_mowner); 418 MOWNER_ATTACH(&tcp_mowner); 419 } 420 421 /* 422 * Create template to be used to send tcp packets on a connection. 423 * Call after host entry created, allocates an mbuf and fills 424 * in a skeletal tcp/ip header, minimizing the amount of work 425 * necessary when the connection is used. 426 */ 427 struct mbuf * 428 tcp_template(struct tcpcb *tp) 429 { 430 struct inpcb *inp = tp->t_inpcb; 431 #ifdef INET6 432 struct in6pcb *in6p = tp->t_in6pcb; 433 #endif 434 struct tcphdr *n; 435 struct mbuf *m; 436 int hlen; 437 438 switch (tp->t_family) { 439 case AF_INET: 440 hlen = sizeof(struct ip); 441 if (inp) 442 break; 443 #ifdef INET6 444 if (in6p) { 445 /* mapped addr case */ 446 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 447 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 448 break; 449 } 450 #endif 451 return NULL; /*EINVAL*/ 452 #ifdef INET6 453 case AF_INET6: 454 hlen = sizeof(struct ip6_hdr); 455 if (in6p) { 456 /* more sainty check? */ 457 break; 458 } 459 return NULL; /*EINVAL*/ 460 #endif 461 default: 462 hlen = 0; /*pacify gcc*/ 463 return NULL; /*EAFNOSUPPORT*/ 464 } 465 #ifdef DIAGNOSTIC 466 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 467 panic("mclbytes too small for t_template"); 468 #endif 469 m = tp->t_template; 470 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 471 ; 472 else { 473 if (m) 474 m_freem(m); 475 m = tp->t_template = NULL; 476 MGETHDR(m, M_DONTWAIT, MT_HEADER); 477 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 478 MCLGET(m, M_DONTWAIT); 479 if ((m->m_flags & M_EXT) == 0) { 480 m_free(m); 481 m = NULL; 482 } 483 } 484 if (m == NULL) 485 return NULL; 486 MCLAIM(m, &tcp_mowner); 487 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 488 } 489 490 bzero(mtod(m, caddr_t), m->m_len); 491 492 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 493 494 switch (tp->t_family) { 495 case AF_INET: 496 { 497 struct ipovly *ipov; 498 mtod(m, struct ip *)->ip_v = 4; 499 mtod(m, struct ip *)->ip_hl = hlen >> 2; 500 ipov = mtod(m, struct ipovly *); 501 ipov->ih_pr = IPPROTO_TCP; 502 ipov->ih_len = htons(sizeof(struct tcphdr)); 503 if (inp) { 504 ipov->ih_src = inp->inp_laddr; 505 ipov->ih_dst = inp->inp_faddr; 506 } 507 #ifdef INET6 508 else if (in6p) { 509 /* mapped addr case */ 510 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 511 sizeof(ipov->ih_src)); 512 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 513 sizeof(ipov->ih_dst)); 514 } 515 #endif 516 /* 517 * Compute the pseudo-header portion of the checksum 518 * now. We incrementally add in the TCP option and 519 * payload lengths later, and then compute the TCP 520 * checksum right before the packet is sent off onto 521 * the wire. 522 */ 523 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 524 ipov->ih_dst.s_addr, 525 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 526 break; 527 } 528 #ifdef INET6 529 case AF_INET6: 530 { 531 struct ip6_hdr *ip6; 532 mtod(m, struct ip *)->ip_v = 6; 533 ip6 = mtod(m, struct ip6_hdr *); 534 ip6->ip6_nxt = IPPROTO_TCP; 535 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 536 ip6->ip6_src = in6p->in6p_laddr; 537 ip6->ip6_dst = in6p->in6p_faddr; 538 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 539 if (ip6_auto_flowlabel) { 540 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 541 ip6->ip6_flow |= 542 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 543 } 544 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 545 ip6->ip6_vfc |= IPV6_VERSION; 546 547 /* 548 * Compute the pseudo-header portion of the checksum 549 * now. We incrementally add in the TCP option and 550 * payload lengths later, and then compute the TCP 551 * checksum right before the packet is sent off onto 552 * the wire. 553 */ 554 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 555 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 556 htonl(IPPROTO_TCP)); 557 break; 558 } 559 #endif 560 } 561 if (inp) { 562 n->th_sport = inp->inp_lport; 563 n->th_dport = inp->inp_fport; 564 } 565 #ifdef INET6 566 else if (in6p) { 567 n->th_sport = in6p->in6p_lport; 568 n->th_dport = in6p->in6p_fport; 569 } 570 #endif 571 n->th_seq = 0; 572 n->th_ack = 0; 573 n->th_x2 = 0; 574 n->th_off = 5; 575 n->th_flags = 0; 576 n->th_win = 0; 577 n->th_urp = 0; 578 return (m); 579 } 580 581 /* 582 * Send a single message to the TCP at address specified by 583 * the given TCP/IP header. If m == 0, then we make a copy 584 * of the tcpiphdr at ti and send directly to the addressed host. 585 * This is used to force keep alive messages out using the TCP 586 * template for a connection tp->t_template. If flags are given 587 * then we send a message back to the TCP which originated the 588 * segment ti, and discard the mbuf containing it and any other 589 * attached mbufs. 590 * 591 * In any case the ack and sequence number of the transmitted 592 * segment are as specified by the parameters. 593 */ 594 int 595 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 596 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 597 { 598 struct route *ro; 599 int error, tlen, win = 0; 600 int hlen; 601 struct ip *ip; 602 #ifdef INET6 603 struct ip6_hdr *ip6; 604 #endif 605 int family; /* family on packet, not inpcb/in6pcb! */ 606 struct tcphdr *th; 607 struct socket *so; 608 609 if (tp != NULL && (flags & TH_RST) == 0) { 610 #ifdef DIAGNOSTIC 611 if (tp->t_inpcb && tp->t_in6pcb) 612 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 613 #endif 614 #ifdef INET 615 if (tp->t_inpcb) 616 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 617 #endif 618 #ifdef INET6 619 if (tp->t_in6pcb) 620 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 621 #endif 622 } 623 624 th = NULL; /* Quell uninitialized warning */ 625 ip = NULL; 626 #ifdef INET6 627 ip6 = NULL; 628 #endif 629 if (m == 0) { 630 if (!template) 631 return EINVAL; 632 633 /* get family information from template */ 634 switch (mtod(template, struct ip *)->ip_v) { 635 case 4: 636 family = AF_INET; 637 hlen = sizeof(struct ip); 638 break; 639 #ifdef INET6 640 case 6: 641 family = AF_INET6; 642 hlen = sizeof(struct ip6_hdr); 643 break; 644 #endif 645 default: 646 return EAFNOSUPPORT; 647 } 648 649 MGETHDR(m, M_DONTWAIT, MT_HEADER); 650 if (m) { 651 MCLAIM(m, &tcp_tx_mowner); 652 MCLGET(m, M_DONTWAIT); 653 if ((m->m_flags & M_EXT) == 0) { 654 m_free(m); 655 m = NULL; 656 } 657 } 658 if (m == NULL) 659 return (ENOBUFS); 660 661 if (tcp_compat_42) 662 tlen = 1; 663 else 664 tlen = 0; 665 666 m->m_data += max_linkhdr; 667 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 668 template->m_len); 669 switch (family) { 670 case AF_INET: 671 ip = mtod(m, struct ip *); 672 th = (struct tcphdr *)(ip + 1); 673 break; 674 #ifdef INET6 675 case AF_INET6: 676 ip6 = mtod(m, struct ip6_hdr *); 677 th = (struct tcphdr *)(ip6 + 1); 678 break; 679 #endif 680 #if 0 681 default: 682 /* noone will visit here */ 683 m_freem(m); 684 return EAFNOSUPPORT; 685 #endif 686 } 687 flags = TH_ACK; 688 } else { 689 690 if ((m->m_flags & M_PKTHDR) == 0) { 691 #if 0 692 printf("non PKTHDR to tcp_respond\n"); 693 #endif 694 m_freem(m); 695 return EINVAL; 696 } 697 #ifdef DIAGNOSTIC 698 if (!th0) 699 panic("th0 == NULL in tcp_respond"); 700 #endif 701 702 /* get family information from m */ 703 switch (mtod(m, struct ip *)->ip_v) { 704 case 4: 705 family = AF_INET; 706 hlen = sizeof(struct ip); 707 ip = mtod(m, struct ip *); 708 break; 709 #ifdef INET6 710 case 6: 711 family = AF_INET6; 712 hlen = sizeof(struct ip6_hdr); 713 ip6 = mtod(m, struct ip6_hdr *); 714 break; 715 #endif 716 default: 717 m_freem(m); 718 return EAFNOSUPPORT; 719 } 720 /* clear h/w csum flags inherited from rx packet */ 721 m->m_pkthdr.csum_flags = 0; 722 723 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 724 tlen = sizeof(*th0); 725 else 726 tlen = th0->th_off << 2; 727 728 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 729 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 730 m->m_len = hlen + tlen; 731 m_freem(m->m_next); 732 m->m_next = NULL; 733 } else { 734 struct mbuf *n; 735 736 #ifdef DIAGNOSTIC 737 if (max_linkhdr + hlen + tlen > MCLBYTES) { 738 m_freem(m); 739 return EMSGSIZE; 740 } 741 #endif 742 MGETHDR(n, M_DONTWAIT, MT_HEADER); 743 if (n && max_linkhdr + hlen + tlen > MHLEN) { 744 MCLGET(n, M_DONTWAIT); 745 if ((n->m_flags & M_EXT) == 0) { 746 m_freem(n); 747 n = NULL; 748 } 749 } 750 if (!n) { 751 m_freem(m); 752 return ENOBUFS; 753 } 754 755 MCLAIM(n, &tcp_tx_mowner); 756 n->m_data += max_linkhdr; 757 n->m_len = hlen + tlen; 758 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 759 m_copyback(n, hlen, tlen, (caddr_t)th0); 760 761 m_freem(m); 762 m = n; 763 n = NULL; 764 } 765 766 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 767 switch (family) { 768 case AF_INET: 769 ip = mtod(m, struct ip *); 770 th = (struct tcphdr *)(ip + 1); 771 ip->ip_p = IPPROTO_TCP; 772 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 773 ip->ip_p = IPPROTO_TCP; 774 break; 775 #ifdef INET6 776 case AF_INET6: 777 ip6 = mtod(m, struct ip6_hdr *); 778 th = (struct tcphdr *)(ip6 + 1); 779 ip6->ip6_nxt = IPPROTO_TCP; 780 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 781 ip6->ip6_nxt = IPPROTO_TCP; 782 break; 783 #endif 784 #if 0 785 default: 786 /* noone will visit here */ 787 m_freem(m); 788 return EAFNOSUPPORT; 789 #endif 790 } 791 xchg(th->th_dport, th->th_sport, u_int16_t); 792 #undef xchg 793 tlen = 0; /*be friendly with the following code*/ 794 } 795 th->th_seq = htonl(seq); 796 th->th_ack = htonl(ack); 797 th->th_x2 = 0; 798 if ((flags & TH_SYN) == 0) { 799 if (tp) 800 win >>= tp->rcv_scale; 801 if (win > TCP_MAXWIN) 802 win = TCP_MAXWIN; 803 th->th_win = htons((u_int16_t)win); 804 th->th_off = sizeof (struct tcphdr) >> 2; 805 tlen += sizeof(*th); 806 } else 807 tlen += th->th_off << 2; 808 m->m_len = hlen + tlen; 809 m->m_pkthdr.len = hlen + tlen; 810 m->m_pkthdr.rcvif = (struct ifnet *) 0; 811 th->th_flags = flags; 812 th->th_urp = 0; 813 814 switch (family) { 815 #ifdef INET 816 case AF_INET: 817 { 818 struct ipovly *ipov = (struct ipovly *)ip; 819 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 820 ipov->ih_len = htons((u_int16_t)tlen); 821 822 th->th_sum = 0; 823 th->th_sum = in_cksum(m, hlen + tlen); 824 ip->ip_len = htons(hlen + tlen); 825 ip->ip_ttl = ip_defttl; 826 break; 827 } 828 #endif 829 #ifdef INET6 830 case AF_INET6: 831 { 832 th->th_sum = 0; 833 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 834 tlen); 835 ip6->ip6_plen = htons(tlen); 836 if (tp && tp->t_in6pcb) { 837 struct ifnet *oifp; 838 ro = (struct route *)&tp->t_in6pcb->in6p_route; 839 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 840 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 841 } else 842 ip6->ip6_hlim = ip6_defhlim; 843 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 844 if (ip6_auto_flowlabel) { 845 ip6->ip6_flow |= 846 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 847 } 848 break; 849 } 850 #endif 851 } 852 853 if (tp && tp->t_inpcb) 854 so = tp->t_inpcb->inp_socket; 855 #ifdef INET6 856 else if (tp && tp->t_in6pcb) 857 so = tp->t_in6pcb->in6p_socket; 858 #endif 859 else 860 so = NULL; 861 862 if (tp != NULL && tp->t_inpcb != NULL) { 863 ro = &tp->t_inpcb->inp_route; 864 #ifdef DIAGNOSTIC 865 if (family != AF_INET) 866 panic("tcp_respond: address family mismatch"); 867 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 868 panic("tcp_respond: ip_dst %x != inp_faddr %x", 869 ntohl(ip->ip_dst.s_addr), 870 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 871 } 872 #endif 873 } 874 #ifdef INET6 875 else if (tp != NULL && tp->t_in6pcb != NULL) { 876 ro = (struct route *)&tp->t_in6pcb->in6p_route; 877 #ifdef DIAGNOSTIC 878 if (family == AF_INET) { 879 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 880 panic("tcp_respond: not mapped addr"); 881 if (bcmp(&ip->ip_dst, 882 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 883 sizeof(ip->ip_dst)) != 0) { 884 panic("tcp_respond: ip_dst != in6p_faddr"); 885 } 886 } else if (family == AF_INET6) { 887 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 888 &tp->t_in6pcb->in6p_faddr)) 889 panic("tcp_respond: ip6_dst != in6p_faddr"); 890 } else 891 panic("tcp_respond: address family mismatch"); 892 #endif 893 } 894 #endif 895 else 896 ro = NULL; 897 898 switch (family) { 899 #ifdef INET 900 case AF_INET: 901 error = ip_output(m, NULL, ro, 902 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 903 (struct ip_moptions *)0, so); 904 break; 905 #endif 906 #ifdef INET6 907 case AF_INET6: 908 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 909 (struct ip6_moptions *)0, so, NULL); 910 break; 911 #endif 912 default: 913 error = EAFNOSUPPORT; 914 break; 915 } 916 917 return (error); 918 } 919 920 /* 921 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 922 * a bunch of members individually, we maintain this template for the 923 * static and mostly-static components of the TCPCB, and copy it into 924 * the new TCPCB instead. 925 */ 926 static struct tcpcb tcpcb_template = { 927 /* 928 * If TCP_NTIMERS ever changes, we'll need to update this 929 * initializer. 930 */ 931 .t_timer = { 932 CALLOUT_INITIALIZER, 933 CALLOUT_INITIALIZER, 934 CALLOUT_INITIALIZER, 935 CALLOUT_INITIALIZER, 936 }, 937 .t_delack_ch = CALLOUT_INITIALIZER, 938 939 .t_srtt = TCPTV_SRTTBASE, 940 .t_rttmin = TCPTV_MIN, 941 942 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 943 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 944 .snd_numholes = 0, 945 946 .t_partialacks = -1, 947 .t_bytes_acked = 0, 948 }; 949 950 /* 951 * Updates the TCPCB template whenever a parameter that would affect 952 * the template is changed. 953 */ 954 void 955 tcp_tcpcb_template(void) 956 { 957 struct tcpcb *tp = &tcpcb_template; 958 int flags; 959 960 tp->t_peermss = tcp_mssdflt; 961 tp->t_ourmss = tcp_mssdflt; 962 tp->t_segsz = tcp_mssdflt; 963 964 flags = 0; 965 if (tcp_do_rfc1323 && tcp_do_win_scale) 966 flags |= TF_REQ_SCALE; 967 if (tcp_do_rfc1323 && tcp_do_timestamps) 968 flags |= TF_REQ_TSTMP; 969 tp->t_flags = flags; 970 971 /* 972 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 973 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 974 * reasonable initial retransmit time. 975 */ 976 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 977 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 978 TCPTV_MIN, TCPTV_REXMTMAX); 979 } 980 981 /* 982 * Create a new TCP control block, making an 983 * empty reassembly queue and hooking it to the argument 984 * protocol control block. 985 */ 986 /* family selects inpcb, or in6pcb */ 987 struct tcpcb * 988 tcp_newtcpcb(int family, void *aux) 989 { 990 struct tcpcb *tp; 991 int i; 992 993 /* XXX Consider using a pool_cache for speed. */ 994 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 995 if (tp == NULL) 996 return (NULL); 997 memcpy(tp, &tcpcb_template, sizeof(*tp)); 998 TAILQ_INIT(&tp->segq); 999 TAILQ_INIT(&tp->timeq); 1000 tp->t_family = family; /* may be overridden later on */ 1001 TAILQ_INIT(&tp->snd_holes); 1002 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1003 1004 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1005 for (i = 0; i < TCPT_NTIMERS; i++) 1006 TCP_TIMER_INIT(tp, i); 1007 1008 switch (family) { 1009 case AF_INET: 1010 { 1011 struct inpcb *inp = (struct inpcb *)aux; 1012 1013 inp->inp_ip.ip_ttl = ip_defttl; 1014 inp->inp_ppcb = (caddr_t)tp; 1015 1016 tp->t_inpcb = inp; 1017 tp->t_mtudisc = ip_mtudisc; 1018 break; 1019 } 1020 #ifdef INET6 1021 case AF_INET6: 1022 { 1023 struct in6pcb *in6p = (struct in6pcb *)aux; 1024 1025 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1026 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1027 : NULL); 1028 in6p->in6p_ppcb = (caddr_t)tp; 1029 1030 tp->t_in6pcb = in6p; 1031 /* for IPv6, always try to run path MTU discovery */ 1032 tp->t_mtudisc = 1; 1033 break; 1034 } 1035 #endif /* INET6 */ 1036 default: 1037 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1038 return (NULL); 1039 } 1040 1041 /* 1042 * Initialize our timebase. When we send timestamps, we take 1043 * the delta from tcp_now -- this means each connection always 1044 * gets a timebase of 0, which makes it, among other things, 1045 * more difficult to determine how long a system has been up, 1046 * and thus how many TCP sequence increments have occurred. 1047 */ 1048 tp->ts_timebase = tcp_now; 1049 1050 tp->t_congctl = tcp_congctl_global; 1051 tp->t_congctl->refcnt++; 1052 1053 return (tp); 1054 } 1055 1056 /* 1057 * Drop a TCP connection, reporting 1058 * the specified error. If connection is synchronized, 1059 * then send a RST to peer. 1060 */ 1061 struct tcpcb * 1062 tcp_drop(struct tcpcb *tp, int errno) 1063 { 1064 struct socket *so = NULL; 1065 1066 #ifdef DIAGNOSTIC 1067 if (tp->t_inpcb && tp->t_in6pcb) 1068 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1069 #endif 1070 #ifdef INET 1071 if (tp->t_inpcb) 1072 so = tp->t_inpcb->inp_socket; 1073 #endif 1074 #ifdef INET6 1075 if (tp->t_in6pcb) 1076 so = tp->t_in6pcb->in6p_socket; 1077 #endif 1078 if (!so) 1079 return NULL; 1080 1081 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1082 tp->t_state = TCPS_CLOSED; 1083 (void) tcp_output(tp); 1084 tcpstat.tcps_drops++; 1085 } else 1086 tcpstat.tcps_conndrops++; 1087 if (errno == ETIMEDOUT && tp->t_softerror) 1088 errno = tp->t_softerror; 1089 so->so_error = errno; 1090 return (tcp_close(tp)); 1091 } 1092 1093 /* 1094 * Return whether this tcpcb is marked as dead, indicating 1095 * to the calling timer function that no further action should 1096 * be taken, as we are about to release this tcpcb. The release 1097 * of the storage will be done if this is the last timer running. 1098 * 1099 * This should be called from the callout handler function after 1100 * callout_ack() is done, so that the number of invoking timer 1101 * functions is 0. 1102 */ 1103 int 1104 tcp_isdead(struct tcpcb *tp) 1105 { 1106 int dead = (tp->t_flags & TF_DEAD); 1107 1108 if (__predict_false(dead)) { 1109 if (tcp_timers_invoking(tp) > 0) 1110 /* not quite there yet -- count separately? */ 1111 return dead; 1112 tcpstat.tcps_delayed_free++; 1113 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_timer.c */ 1114 } 1115 return dead; 1116 } 1117 1118 /* 1119 * Close a TCP control block: 1120 * discard all space held by the tcp 1121 * discard internet protocol block 1122 * wake up any sleepers 1123 */ 1124 struct tcpcb * 1125 tcp_close(struct tcpcb *tp) 1126 { 1127 struct inpcb *inp; 1128 #ifdef INET6 1129 struct in6pcb *in6p; 1130 #endif 1131 struct socket *so; 1132 #ifdef RTV_RTT 1133 struct rtentry *rt; 1134 #endif 1135 struct route *ro; 1136 1137 inp = tp->t_inpcb; 1138 #ifdef INET6 1139 in6p = tp->t_in6pcb; 1140 #endif 1141 so = NULL; 1142 ro = NULL; 1143 if (inp) { 1144 so = inp->inp_socket; 1145 ro = &inp->inp_route; 1146 } 1147 #ifdef INET6 1148 else if (in6p) { 1149 so = in6p->in6p_socket; 1150 ro = (struct route *)&in6p->in6p_route; 1151 } 1152 #endif 1153 1154 #ifdef RTV_RTT 1155 /* 1156 * If we sent enough data to get some meaningful characteristics, 1157 * save them in the routing entry. 'Enough' is arbitrarily 1158 * defined as the sendpipesize (default 4K) * 16. This would 1159 * give us 16 rtt samples assuming we only get one sample per 1160 * window (the usual case on a long haul net). 16 samples is 1161 * enough for the srtt filter to converge to within 5% of the correct 1162 * value; fewer samples and we could save a very bogus rtt. 1163 * 1164 * Don't update the default route's characteristics and don't 1165 * update anything that the user "locked". 1166 */ 1167 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1168 ro && (rt = ro->ro_rt) && 1169 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1170 u_long i = 0; 1171 1172 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1173 i = tp->t_srtt * 1174 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1175 if (rt->rt_rmx.rmx_rtt && i) 1176 /* 1177 * filter this update to half the old & half 1178 * the new values, converting scale. 1179 * See route.h and tcp_var.h for a 1180 * description of the scaling constants. 1181 */ 1182 rt->rt_rmx.rmx_rtt = 1183 (rt->rt_rmx.rmx_rtt + i) / 2; 1184 else 1185 rt->rt_rmx.rmx_rtt = i; 1186 } 1187 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1188 i = tp->t_rttvar * 1189 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1190 if (rt->rt_rmx.rmx_rttvar && i) 1191 rt->rt_rmx.rmx_rttvar = 1192 (rt->rt_rmx.rmx_rttvar + i) / 2; 1193 else 1194 rt->rt_rmx.rmx_rttvar = i; 1195 } 1196 /* 1197 * update the pipelimit (ssthresh) if it has been updated 1198 * already or if a pipesize was specified & the threshhold 1199 * got below half the pipesize. I.e., wait for bad news 1200 * before we start updating, then update on both good 1201 * and bad news. 1202 */ 1203 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1204 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1205 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1206 /* 1207 * convert the limit from user data bytes to 1208 * packets then to packet data bytes. 1209 */ 1210 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1211 if (i < 2) 1212 i = 2; 1213 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1214 if (rt->rt_rmx.rmx_ssthresh) 1215 rt->rt_rmx.rmx_ssthresh = 1216 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1217 else 1218 rt->rt_rmx.rmx_ssthresh = i; 1219 } 1220 } 1221 #endif /* RTV_RTT */ 1222 /* free the reassembly queue, if any */ 1223 TCP_REASS_LOCK(tp); 1224 (void) tcp_freeq(tp); 1225 TCP_REASS_UNLOCK(tp); 1226 1227 /* free the SACK holes list. */ 1228 tcp_free_sackholes(tp); 1229 1230 tp->t_congctl->refcnt--; 1231 1232 tcp_canceltimers(tp); 1233 TCP_CLEAR_DELACK(tp); 1234 syn_cache_cleanup(tp); 1235 1236 if (tp->t_template) { 1237 m_free(tp->t_template); 1238 tp->t_template = NULL; 1239 } 1240 if (tcp_timers_invoking(tp)) 1241 tp->t_flags |= TF_DEAD; 1242 else 1243 pool_put(&tcpcb_pool, tp); 1244 1245 if (inp) { 1246 inp->inp_ppcb = 0; 1247 soisdisconnected(so); 1248 in_pcbdetach(inp); 1249 } 1250 #ifdef INET6 1251 else if (in6p) { 1252 in6p->in6p_ppcb = 0; 1253 soisdisconnected(so); 1254 in6_pcbdetach(in6p); 1255 } 1256 #endif 1257 tcpstat.tcps_closed++; 1258 return ((struct tcpcb *)0); 1259 } 1260 1261 int 1262 tcp_freeq(tp) 1263 struct tcpcb *tp; 1264 { 1265 struct ipqent *qe; 1266 int rv = 0; 1267 #ifdef TCPREASS_DEBUG 1268 int i = 0; 1269 #endif 1270 1271 TCP_REASS_LOCK_CHECK(tp); 1272 1273 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1274 #ifdef TCPREASS_DEBUG 1275 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1276 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1277 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1278 #endif 1279 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1280 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1281 m_freem(qe->ipqe_m); 1282 tcpipqent_free(qe); 1283 rv = 1; 1284 } 1285 tp->t_segqlen = 0; 1286 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1287 return (rv); 1288 } 1289 1290 /* 1291 * Protocol drain routine. Called when memory is in short supply. 1292 */ 1293 void 1294 tcp_drain(void) 1295 { 1296 struct inpcb_hdr *inph; 1297 struct tcpcb *tp; 1298 1299 /* 1300 * Free the sequence queue of all TCP connections. 1301 */ 1302 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1303 switch (inph->inph_af) { 1304 case AF_INET: 1305 tp = intotcpcb((struct inpcb *)inph); 1306 break; 1307 #ifdef INET6 1308 case AF_INET6: 1309 tp = in6totcpcb((struct in6pcb *)inph); 1310 break; 1311 #endif 1312 default: 1313 tp = NULL; 1314 break; 1315 } 1316 if (tp != NULL) { 1317 /* 1318 * We may be called from a device's interrupt 1319 * context. If the tcpcb is already busy, 1320 * just bail out now. 1321 */ 1322 if (tcp_reass_lock_try(tp) == 0) 1323 continue; 1324 if (tcp_freeq(tp)) 1325 tcpstat.tcps_connsdrained++; 1326 TCP_REASS_UNLOCK(tp); 1327 } 1328 } 1329 } 1330 1331 /* 1332 * Notify a tcp user of an asynchronous error; 1333 * store error as soft error, but wake up user 1334 * (for now, won't do anything until can select for soft error). 1335 */ 1336 void 1337 tcp_notify(struct inpcb *inp, int error) 1338 { 1339 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1340 struct socket *so = inp->inp_socket; 1341 1342 /* 1343 * Ignore some errors if we are hooked up. 1344 * If connection hasn't completed, has retransmitted several times, 1345 * and receives a second error, give up now. This is better 1346 * than waiting a long time to establish a connection that 1347 * can never complete. 1348 */ 1349 if (tp->t_state == TCPS_ESTABLISHED && 1350 (error == EHOSTUNREACH || error == ENETUNREACH || 1351 error == EHOSTDOWN)) { 1352 return; 1353 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1354 tp->t_rxtshift > 3 && tp->t_softerror) 1355 so->so_error = error; 1356 else 1357 tp->t_softerror = error; 1358 wakeup((caddr_t) &so->so_timeo); 1359 sorwakeup(so); 1360 sowwakeup(so); 1361 } 1362 1363 #ifdef INET6 1364 void 1365 tcp6_notify(struct in6pcb *in6p, int error) 1366 { 1367 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1368 struct socket *so = in6p->in6p_socket; 1369 1370 /* 1371 * Ignore some errors if we are hooked up. 1372 * If connection hasn't completed, has retransmitted several times, 1373 * and receives a second error, give up now. This is better 1374 * than waiting a long time to establish a connection that 1375 * can never complete. 1376 */ 1377 if (tp->t_state == TCPS_ESTABLISHED && 1378 (error == EHOSTUNREACH || error == ENETUNREACH || 1379 error == EHOSTDOWN)) { 1380 return; 1381 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1382 tp->t_rxtshift > 3 && tp->t_softerror) 1383 so->so_error = error; 1384 else 1385 tp->t_softerror = error; 1386 wakeup((caddr_t) &so->so_timeo); 1387 sorwakeup(so); 1388 sowwakeup(so); 1389 } 1390 #endif 1391 1392 #ifdef INET6 1393 void 1394 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1395 { 1396 struct tcphdr th; 1397 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1398 int nmatch; 1399 struct ip6_hdr *ip6; 1400 const struct sockaddr_in6 *sa6_src = NULL; 1401 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1402 struct mbuf *m; 1403 int off; 1404 1405 if (sa->sa_family != AF_INET6 || 1406 sa->sa_len != sizeof(struct sockaddr_in6)) 1407 return; 1408 if ((unsigned)cmd >= PRC_NCMDS) 1409 return; 1410 else if (cmd == PRC_QUENCH) { 1411 /* 1412 * Don't honor ICMP Source Quench messages meant for 1413 * TCP connections. 1414 */ 1415 return; 1416 } else if (PRC_IS_REDIRECT(cmd)) 1417 notify = in6_rtchange, d = NULL; 1418 else if (cmd == PRC_MSGSIZE) 1419 ; /* special code is present, see below */ 1420 else if (cmd == PRC_HOSTDEAD) 1421 d = NULL; 1422 else if (inet6ctlerrmap[cmd] == 0) 1423 return; 1424 1425 /* if the parameter is from icmp6, decode it. */ 1426 if (d != NULL) { 1427 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1428 m = ip6cp->ip6c_m; 1429 ip6 = ip6cp->ip6c_ip6; 1430 off = ip6cp->ip6c_off; 1431 sa6_src = ip6cp->ip6c_src; 1432 } else { 1433 m = NULL; 1434 ip6 = NULL; 1435 sa6_src = &sa6_any; 1436 off = 0; 1437 } 1438 1439 if (ip6) { 1440 /* 1441 * XXX: We assume that when ip6 is non NULL, 1442 * M and OFF are valid. 1443 */ 1444 1445 /* check if we can safely examine src and dst ports */ 1446 if (m->m_pkthdr.len < off + sizeof(th)) { 1447 if (cmd == PRC_MSGSIZE) 1448 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1449 return; 1450 } 1451 1452 bzero(&th, sizeof(th)); 1453 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1454 1455 if (cmd == PRC_MSGSIZE) { 1456 int valid = 0; 1457 1458 /* 1459 * Check to see if we have a valid TCP connection 1460 * corresponding to the address in the ICMPv6 message 1461 * payload. 1462 */ 1463 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1464 th.th_dport, 1465 (const struct in6_addr *)&sa6_src->sin6_addr, 1466 th.th_sport, 0)) 1467 valid++; 1468 1469 /* 1470 * Depending on the value of "valid" and routing table 1471 * size (mtudisc_{hi,lo}wat), we will: 1472 * - recalcurate the new MTU and create the 1473 * corresponding routing entry, or 1474 * - ignore the MTU change notification. 1475 */ 1476 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1477 1478 /* 1479 * no need to call in6_pcbnotify, it should have been 1480 * called via callback if necessary 1481 */ 1482 return; 1483 } 1484 1485 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1486 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1487 if (nmatch == 0 && syn_cache_count && 1488 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1489 inet6ctlerrmap[cmd] == ENETUNREACH || 1490 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1491 syn_cache_unreach((const struct sockaddr *)sa6_src, 1492 sa, &th); 1493 } else { 1494 (void) in6_pcbnotify(&tcbtable, sa, 0, 1495 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1496 } 1497 } 1498 #endif 1499 1500 #ifdef INET 1501 /* assumes that ip header and tcp header are contiguous on mbuf */ 1502 void * 1503 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1504 { 1505 struct ip *ip = v; 1506 struct tcphdr *th; 1507 struct icmp *icp; 1508 extern const int inetctlerrmap[]; 1509 void (*notify)(struct inpcb *, int) = tcp_notify; 1510 int errno; 1511 int nmatch; 1512 struct tcpcb *tp; 1513 u_int mtu; 1514 tcp_seq seq; 1515 struct inpcb *inp; 1516 #ifdef INET6 1517 struct in6pcb *in6p; 1518 struct in6_addr src6, dst6; 1519 #endif 1520 1521 if (sa->sa_family != AF_INET || 1522 sa->sa_len != sizeof(struct sockaddr_in)) 1523 return NULL; 1524 if ((unsigned)cmd >= PRC_NCMDS) 1525 return NULL; 1526 errno = inetctlerrmap[cmd]; 1527 if (cmd == PRC_QUENCH) 1528 /* 1529 * Don't honor ICMP Source Quench messages meant for 1530 * TCP connections. 1531 */ 1532 return NULL; 1533 else if (PRC_IS_REDIRECT(cmd)) 1534 notify = in_rtchange, ip = 0; 1535 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1536 /* 1537 * Check to see if we have a valid TCP connection 1538 * corresponding to the address in the ICMP message 1539 * payload. 1540 * 1541 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1542 */ 1543 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1544 #ifdef INET6 1545 memset(&src6, 0, sizeof(src6)); 1546 memset(&dst6, 0, sizeof(dst6)); 1547 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1548 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1549 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1550 #endif 1551 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1552 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1553 #ifdef INET6 1554 in6p = NULL; 1555 #else 1556 ; 1557 #endif 1558 #ifdef INET6 1559 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1560 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1561 ; 1562 #endif 1563 else 1564 return NULL; 1565 1566 /* 1567 * Now that we've validated that we are actually communicating 1568 * with the host indicated in the ICMP message, locate the 1569 * ICMP header, recalculate the new MTU, and create the 1570 * corresponding routing entry. 1571 */ 1572 icp = (struct icmp *)((caddr_t)ip - 1573 offsetof(struct icmp, icmp_ip)); 1574 if (inp) { 1575 if ((tp = intotcpcb(inp)) == NULL) 1576 return NULL; 1577 } 1578 #ifdef INET6 1579 else if (in6p) { 1580 if ((tp = in6totcpcb(in6p)) == NULL) 1581 return NULL; 1582 } 1583 #endif 1584 else 1585 return NULL; 1586 seq = ntohl(th->th_seq); 1587 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1588 return NULL; 1589 /* 1590 * If the ICMP message advertises a Next-Hop MTU 1591 * equal or larger than the maximum packet size we have 1592 * ever sent, drop the message. 1593 */ 1594 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1595 if (mtu >= tp->t_pmtud_mtu_sent) 1596 return NULL; 1597 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1598 /* 1599 * Calculate new MTU, and create corresponding 1600 * route (traditional PMTUD). 1601 */ 1602 tp->t_flags &= ~TF_PMTUD_PEND; 1603 icmp_mtudisc(icp, ip->ip_dst); 1604 } else { 1605 /* 1606 * Record the information got in the ICMP 1607 * message; act on it later. 1608 * If we had already recorded an ICMP message, 1609 * replace the old one only if the new message 1610 * refers to an older TCP segment 1611 */ 1612 if (tp->t_flags & TF_PMTUD_PEND) { 1613 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1614 return NULL; 1615 } else 1616 tp->t_flags |= TF_PMTUD_PEND; 1617 tp->t_pmtud_th_seq = seq; 1618 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1619 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1620 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1621 } 1622 return NULL; 1623 } else if (cmd == PRC_HOSTDEAD) 1624 ip = 0; 1625 else if (errno == 0) 1626 return NULL; 1627 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1628 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1629 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1630 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1631 if (nmatch == 0 && syn_cache_count && 1632 (inetctlerrmap[cmd] == EHOSTUNREACH || 1633 inetctlerrmap[cmd] == ENETUNREACH || 1634 inetctlerrmap[cmd] == EHOSTDOWN)) { 1635 struct sockaddr_in sin; 1636 bzero(&sin, sizeof(sin)); 1637 sin.sin_len = sizeof(sin); 1638 sin.sin_family = AF_INET; 1639 sin.sin_port = th->th_sport; 1640 sin.sin_addr = ip->ip_src; 1641 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1642 } 1643 1644 /* XXX mapped address case */ 1645 } else 1646 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1647 notify); 1648 return NULL; 1649 } 1650 1651 /* 1652 * When a source quench is received, we are being notified of congestion. 1653 * Close the congestion window down to the Loss Window (one segment). 1654 * We will gradually open it again as we proceed. 1655 */ 1656 void 1657 tcp_quench(struct inpcb *inp, int errno) 1658 { 1659 struct tcpcb *tp = intotcpcb(inp); 1660 1661 if (tp) { 1662 tp->snd_cwnd = tp->t_segsz; 1663 tp->t_bytes_acked = 0; 1664 } 1665 } 1666 #endif 1667 1668 #ifdef INET6 1669 void 1670 tcp6_quench(struct in6pcb *in6p, int errno) 1671 { 1672 struct tcpcb *tp = in6totcpcb(in6p); 1673 1674 if (tp) { 1675 tp->snd_cwnd = tp->t_segsz; 1676 tp->t_bytes_acked = 0; 1677 } 1678 } 1679 #endif 1680 1681 #ifdef INET 1682 /* 1683 * Path MTU Discovery handlers. 1684 */ 1685 void 1686 tcp_mtudisc_callback(struct in_addr faddr) 1687 { 1688 #ifdef INET6 1689 struct in6_addr in6; 1690 #endif 1691 1692 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1693 #ifdef INET6 1694 memset(&in6, 0, sizeof(in6)); 1695 in6.s6_addr16[5] = 0xffff; 1696 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1697 tcp6_mtudisc_callback(&in6); 1698 #endif 1699 } 1700 1701 /* 1702 * On receipt of path MTU corrections, flush old route and replace it 1703 * with the new one. Retransmit all unacknowledged packets, to ensure 1704 * that all packets will be received. 1705 */ 1706 void 1707 tcp_mtudisc(struct inpcb *inp, int errno) 1708 { 1709 struct tcpcb *tp = intotcpcb(inp); 1710 struct rtentry *rt = in_pcbrtentry(inp); 1711 1712 if (tp != 0) { 1713 if (rt != 0) { 1714 /* 1715 * If this was not a host route, remove and realloc. 1716 */ 1717 if ((rt->rt_flags & RTF_HOST) == 0) { 1718 in_rtchange(inp, errno); 1719 if ((rt = in_pcbrtentry(inp)) == 0) 1720 return; 1721 } 1722 1723 /* 1724 * Slow start out of the error condition. We 1725 * use the MTU because we know it's smaller 1726 * than the previously transmitted segment. 1727 * 1728 * Note: This is more conservative than the 1729 * suggestion in draft-floyd-incr-init-win-03. 1730 */ 1731 if (rt->rt_rmx.rmx_mtu != 0) 1732 tp->snd_cwnd = 1733 TCP_INITIAL_WINDOW(tcp_init_win, 1734 rt->rt_rmx.rmx_mtu); 1735 } 1736 1737 /* 1738 * Resend unacknowledged packets. 1739 */ 1740 tp->snd_nxt = tp->snd_una; 1741 tcp_output(tp); 1742 } 1743 } 1744 #endif 1745 1746 #ifdef INET6 1747 /* 1748 * Path MTU Discovery handlers. 1749 */ 1750 void 1751 tcp6_mtudisc_callback(struct in6_addr *faddr) 1752 { 1753 struct sockaddr_in6 sin6; 1754 1755 bzero(&sin6, sizeof(sin6)); 1756 sin6.sin6_family = AF_INET6; 1757 sin6.sin6_len = sizeof(struct sockaddr_in6); 1758 sin6.sin6_addr = *faddr; 1759 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1760 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1761 } 1762 1763 void 1764 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1765 { 1766 struct tcpcb *tp = in6totcpcb(in6p); 1767 struct rtentry *rt = in6_pcbrtentry(in6p); 1768 1769 if (tp != 0) { 1770 if (rt != 0) { 1771 /* 1772 * If this was not a host route, remove and realloc. 1773 */ 1774 if ((rt->rt_flags & RTF_HOST) == 0) { 1775 in6_rtchange(in6p, errno); 1776 if ((rt = in6_pcbrtentry(in6p)) == 0) 1777 return; 1778 } 1779 1780 /* 1781 * Slow start out of the error condition. We 1782 * use the MTU because we know it's smaller 1783 * than the previously transmitted segment. 1784 * 1785 * Note: This is more conservative than the 1786 * suggestion in draft-floyd-incr-init-win-03. 1787 */ 1788 if (rt->rt_rmx.rmx_mtu != 0) 1789 tp->snd_cwnd = 1790 TCP_INITIAL_WINDOW(tcp_init_win, 1791 rt->rt_rmx.rmx_mtu); 1792 } 1793 1794 /* 1795 * Resend unacknowledged packets. 1796 */ 1797 tp->snd_nxt = tp->snd_una; 1798 tcp_output(tp); 1799 } 1800 } 1801 #endif /* INET6 */ 1802 1803 /* 1804 * Compute the MSS to advertise to the peer. Called only during 1805 * the 3-way handshake. If we are the server (peer initiated 1806 * connection), we are called with a pointer to the interface 1807 * on which the SYN packet arrived. If we are the client (we 1808 * initiated connection), we are called with a pointer to the 1809 * interface out which this connection should go. 1810 * 1811 * NOTE: Do not subtract IP option/extension header size nor IPsec 1812 * header size from MSS advertisement. MSS option must hold the maximum 1813 * segment size we can accept, so it must always be: 1814 * max(if mtu) - ip header - tcp header 1815 */ 1816 u_long 1817 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1818 { 1819 extern u_long in_maxmtu; 1820 u_long mss = 0; 1821 u_long hdrsiz; 1822 1823 /* 1824 * In order to avoid defeating path MTU discovery on the peer, 1825 * we advertise the max MTU of all attached networks as our MSS, 1826 * per RFC 1191, section 3.1. 1827 * 1828 * We provide the option to advertise just the MTU of 1829 * the interface on which we hope this connection will 1830 * be receiving. If we are responding to a SYN, we 1831 * will have a pretty good idea about this, but when 1832 * initiating a connection there is a bit more doubt. 1833 * 1834 * We also need to ensure that loopback has a large enough 1835 * MSS, as the loopback MTU is never included in in_maxmtu. 1836 */ 1837 1838 if (ifp != NULL) 1839 switch (af) { 1840 case AF_INET: 1841 mss = ifp->if_mtu; 1842 break; 1843 #ifdef INET6 1844 case AF_INET6: 1845 mss = IN6_LINKMTU(ifp); 1846 break; 1847 #endif 1848 } 1849 1850 if (tcp_mss_ifmtu == 0) 1851 switch (af) { 1852 case AF_INET: 1853 mss = max(in_maxmtu, mss); 1854 break; 1855 #ifdef INET6 1856 case AF_INET6: 1857 mss = max(in6_maxmtu, mss); 1858 break; 1859 #endif 1860 } 1861 1862 switch (af) { 1863 case AF_INET: 1864 hdrsiz = sizeof(struct ip); 1865 break; 1866 #ifdef INET6 1867 case AF_INET6: 1868 hdrsiz = sizeof(struct ip6_hdr); 1869 break; 1870 #endif 1871 default: 1872 hdrsiz = 0; 1873 break; 1874 } 1875 hdrsiz += sizeof(struct tcphdr); 1876 if (mss > hdrsiz) 1877 mss -= hdrsiz; 1878 1879 mss = max(tcp_mssdflt, mss); 1880 return (mss); 1881 } 1882 1883 /* 1884 * Set connection variables based on the peer's advertised MSS. 1885 * We are passed the TCPCB for the actual connection. If we 1886 * are the server, we are called by the compressed state engine 1887 * when the 3-way handshake is complete. If we are the client, 1888 * we are called when we receive the SYN,ACK from the server. 1889 * 1890 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1891 * before this routine is called! 1892 */ 1893 void 1894 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1895 { 1896 struct socket *so; 1897 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1898 struct rtentry *rt; 1899 #endif 1900 u_long bufsize; 1901 int mss; 1902 1903 #ifdef DIAGNOSTIC 1904 if (tp->t_inpcb && tp->t_in6pcb) 1905 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1906 #endif 1907 so = NULL; 1908 rt = NULL; 1909 #ifdef INET 1910 if (tp->t_inpcb) { 1911 so = tp->t_inpcb->inp_socket; 1912 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1913 rt = in_pcbrtentry(tp->t_inpcb); 1914 #endif 1915 } 1916 #endif 1917 #ifdef INET6 1918 if (tp->t_in6pcb) { 1919 so = tp->t_in6pcb->in6p_socket; 1920 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1921 rt = in6_pcbrtentry(tp->t_in6pcb); 1922 #endif 1923 } 1924 #endif 1925 1926 /* 1927 * As per RFC1122, use the default MSS value, unless they 1928 * sent us an offer. Do not accept offers less than 256 bytes. 1929 */ 1930 mss = tcp_mssdflt; 1931 if (offer) 1932 mss = offer; 1933 mss = max(mss, 256); /* sanity */ 1934 tp->t_peermss = mss; 1935 mss -= tcp_optlen(tp); 1936 #ifdef INET 1937 if (tp->t_inpcb) 1938 mss -= ip_optlen(tp->t_inpcb); 1939 #endif 1940 #ifdef INET6 1941 if (tp->t_in6pcb) 1942 mss -= ip6_optlen(tp->t_in6pcb); 1943 #endif 1944 1945 /* 1946 * If there's a pipesize, change the socket buffer to that size. 1947 * Make the socket buffer an integral number of MSS units. If 1948 * the MSS is larger than the socket buffer, artificially decrease 1949 * the MSS. 1950 */ 1951 #ifdef RTV_SPIPE 1952 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1953 bufsize = rt->rt_rmx.rmx_sendpipe; 1954 else 1955 #endif 1956 { 1957 KASSERT(so != NULL); 1958 bufsize = so->so_snd.sb_hiwat; 1959 } 1960 if (bufsize < mss) 1961 mss = bufsize; 1962 else { 1963 bufsize = roundup(bufsize, mss); 1964 if (bufsize > sb_max) 1965 bufsize = sb_max; 1966 (void) sbreserve(&so->so_snd, bufsize, so); 1967 } 1968 tp->t_segsz = mss; 1969 1970 #ifdef RTV_SSTHRESH 1971 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1972 /* 1973 * There's some sort of gateway or interface buffer 1974 * limit on the path. Use this to set the slow 1975 * start threshold, but set the threshold to no less 1976 * than 2 * MSS. 1977 */ 1978 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1979 } 1980 #endif 1981 } 1982 1983 /* 1984 * Processing necessary when a TCP connection is established. 1985 */ 1986 void 1987 tcp_established(struct tcpcb *tp) 1988 { 1989 struct socket *so; 1990 #ifdef RTV_RPIPE 1991 struct rtentry *rt; 1992 #endif 1993 u_long bufsize; 1994 1995 #ifdef DIAGNOSTIC 1996 if (tp->t_inpcb && tp->t_in6pcb) 1997 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1998 #endif 1999 so = NULL; 2000 rt = NULL; 2001 #ifdef INET 2002 if (tp->t_inpcb) { 2003 so = tp->t_inpcb->inp_socket; 2004 #if defined(RTV_RPIPE) 2005 rt = in_pcbrtentry(tp->t_inpcb); 2006 #endif 2007 } 2008 #endif 2009 #ifdef INET6 2010 if (tp->t_in6pcb) { 2011 so = tp->t_in6pcb->in6p_socket; 2012 #if defined(RTV_RPIPE) 2013 rt = in6_pcbrtentry(tp->t_in6pcb); 2014 #endif 2015 } 2016 #endif 2017 2018 tp->t_state = TCPS_ESTABLISHED; 2019 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 2020 2021 #ifdef RTV_RPIPE 2022 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2023 bufsize = rt->rt_rmx.rmx_recvpipe; 2024 else 2025 #endif 2026 { 2027 KASSERT(so != NULL); 2028 bufsize = so->so_rcv.sb_hiwat; 2029 } 2030 if (bufsize > tp->t_ourmss) { 2031 bufsize = roundup(bufsize, tp->t_ourmss); 2032 if (bufsize > sb_max) 2033 bufsize = sb_max; 2034 (void) sbreserve(&so->so_rcv, bufsize, so); 2035 } 2036 } 2037 2038 /* 2039 * Check if there's an initial rtt or rttvar. Convert from the 2040 * route-table units to scaled multiples of the slow timeout timer. 2041 * Called only during the 3-way handshake. 2042 */ 2043 void 2044 tcp_rmx_rtt(struct tcpcb *tp) 2045 { 2046 #ifdef RTV_RTT 2047 struct rtentry *rt = NULL; 2048 int rtt; 2049 2050 #ifdef DIAGNOSTIC 2051 if (tp->t_inpcb && tp->t_in6pcb) 2052 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2053 #endif 2054 #ifdef INET 2055 if (tp->t_inpcb) 2056 rt = in_pcbrtentry(tp->t_inpcb); 2057 #endif 2058 #ifdef INET6 2059 if (tp->t_in6pcb) 2060 rt = in6_pcbrtentry(tp->t_in6pcb); 2061 #endif 2062 if (rt == NULL) 2063 return; 2064 2065 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2066 /* 2067 * XXX The lock bit for MTU indicates that the value 2068 * is also a minimum value; this is subject to time. 2069 */ 2070 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2071 TCPT_RANGESET(tp->t_rttmin, 2072 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2073 TCPTV_MIN, TCPTV_REXMTMAX); 2074 tp->t_srtt = rtt / 2075 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2076 if (rt->rt_rmx.rmx_rttvar) { 2077 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2078 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2079 (TCP_RTTVAR_SHIFT + 2)); 2080 } else { 2081 /* Default variation is +- 1 rtt */ 2082 tp->t_rttvar = 2083 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2084 } 2085 TCPT_RANGESET(tp->t_rxtcur, 2086 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2087 tp->t_rttmin, TCPTV_REXMTMAX); 2088 } 2089 #endif 2090 } 2091 2092 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2093 #if NRND > 0 2094 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2095 #endif 2096 2097 /* 2098 * Get a new sequence value given a tcp control block 2099 */ 2100 tcp_seq 2101 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2102 { 2103 2104 #ifdef INET 2105 if (tp->t_inpcb != NULL) { 2106 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2107 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2108 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2109 addin)); 2110 } 2111 #endif 2112 #ifdef INET6 2113 if (tp->t_in6pcb != NULL) { 2114 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2115 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2116 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2117 addin)); 2118 } 2119 #endif 2120 /* Not possible. */ 2121 panic("tcp_new_iss"); 2122 } 2123 2124 /* 2125 * This routine actually generates a new TCP initial sequence number. 2126 */ 2127 tcp_seq 2128 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2129 size_t addrsz, tcp_seq addin) 2130 { 2131 tcp_seq tcp_iss; 2132 2133 #if NRND > 0 2134 static int beenhere; 2135 2136 /* 2137 * If we haven't been here before, initialize our cryptographic 2138 * hash secret. 2139 */ 2140 if (beenhere == 0) { 2141 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2142 RND_EXTRACT_ANY); 2143 beenhere = 1; 2144 } 2145 2146 if (tcp_do_rfc1948) { 2147 MD5_CTX ctx; 2148 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2149 2150 /* 2151 * Compute the base value of the ISS. It is a hash 2152 * of (saddr, sport, daddr, dport, secret). 2153 */ 2154 MD5Init(&ctx); 2155 2156 MD5Update(&ctx, (u_char *) laddr, addrsz); 2157 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2158 2159 MD5Update(&ctx, (u_char *) faddr, addrsz); 2160 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2161 2162 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2163 2164 MD5Final(hash, &ctx); 2165 2166 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2167 2168 /* 2169 * Now increment our "timer", and add it in to 2170 * the computed value. 2171 * 2172 * XXX Use `addin'? 2173 * XXX TCP_ISSINCR too large to use? 2174 */ 2175 tcp_iss_seq += TCP_ISSINCR; 2176 #ifdef TCPISS_DEBUG 2177 printf("ISS hash 0x%08x, ", tcp_iss); 2178 #endif 2179 tcp_iss += tcp_iss_seq + addin; 2180 #ifdef TCPISS_DEBUG 2181 printf("new ISS 0x%08x\n", tcp_iss); 2182 #endif 2183 } else 2184 #endif /* NRND > 0 */ 2185 { 2186 /* 2187 * Randomize. 2188 */ 2189 #if NRND > 0 2190 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2191 #else 2192 tcp_iss = arc4random(); 2193 #endif 2194 2195 /* 2196 * If we were asked to add some amount to a known value, 2197 * we will take a random value obtained above, mask off 2198 * the upper bits, and add in the known value. We also 2199 * add in a constant to ensure that we are at least a 2200 * certain distance from the original value. 2201 * 2202 * This is used when an old connection is in timed wait 2203 * and we have a new one coming in, for instance. 2204 */ 2205 if (addin != 0) { 2206 #ifdef TCPISS_DEBUG 2207 printf("Random %08x, ", tcp_iss); 2208 #endif 2209 tcp_iss &= TCP_ISS_RANDOM_MASK; 2210 tcp_iss += addin + TCP_ISSINCR; 2211 #ifdef TCPISS_DEBUG 2212 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2213 #endif 2214 } else { 2215 tcp_iss &= TCP_ISS_RANDOM_MASK; 2216 tcp_iss += tcp_iss_seq; 2217 tcp_iss_seq += TCP_ISSINCR; 2218 #ifdef TCPISS_DEBUG 2219 printf("ISS %08x\n", tcp_iss); 2220 #endif 2221 } 2222 } 2223 2224 if (tcp_compat_42) { 2225 /* 2226 * Limit it to the positive range for really old TCP 2227 * implementations. 2228 * Just AND off the top bit instead of checking if 2229 * is set first - saves a branch 50% of the time. 2230 */ 2231 tcp_iss &= 0x7fffffff; /* XXX */ 2232 } 2233 2234 return (tcp_iss); 2235 } 2236 2237 #if defined(IPSEC) || defined(FAST_IPSEC) 2238 /* compute ESP/AH header size for TCP, including outer IP header. */ 2239 size_t 2240 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2241 { 2242 struct inpcb *inp; 2243 size_t hdrsiz; 2244 2245 /* XXX mapped addr case (tp->t_in6pcb) */ 2246 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2247 return 0; 2248 switch (tp->t_family) { 2249 case AF_INET: 2250 /* XXX: should use currect direction. */ 2251 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2252 break; 2253 default: 2254 hdrsiz = 0; 2255 break; 2256 } 2257 2258 return hdrsiz; 2259 } 2260 2261 #ifdef INET6 2262 size_t 2263 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2264 { 2265 struct in6pcb *in6p; 2266 size_t hdrsiz; 2267 2268 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2269 return 0; 2270 switch (tp->t_family) { 2271 case AF_INET6: 2272 /* XXX: should use currect direction. */ 2273 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2274 break; 2275 case AF_INET: 2276 /* mapped address case - tricky */ 2277 default: 2278 hdrsiz = 0; 2279 break; 2280 } 2281 2282 return hdrsiz; 2283 } 2284 #endif 2285 #endif /*IPSEC*/ 2286 2287 /* 2288 * Determine the length of the TCP options for this connection. 2289 * 2290 * XXX: What do we do for SACK, when we add that? Just reserve 2291 * all of the space? Otherwise we can't exactly be incrementing 2292 * cwnd by an amount that varies depending on the amount we last 2293 * had to SACK! 2294 */ 2295 2296 u_int 2297 tcp_optlen(struct tcpcb *tp) 2298 { 2299 u_int optlen; 2300 2301 optlen = 0; 2302 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2303 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2304 optlen += TCPOLEN_TSTAMP_APPA; 2305 2306 #ifdef TCP_SIGNATURE 2307 if (tp->t_flags & TF_SIGNATURE) 2308 optlen += TCPOLEN_SIGNATURE + 2; 2309 #endif /* TCP_SIGNATURE */ 2310 2311 return optlen; 2312 } 2313 2314 u_int 2315 tcp_hdrsz(struct tcpcb *tp) 2316 { 2317 u_int hlen; 2318 2319 switch (tp->t_family) { 2320 #ifdef INET6 2321 case AF_INET6: 2322 hlen = sizeof(struct ip6_hdr); 2323 break; 2324 #endif 2325 case AF_INET: 2326 hlen = sizeof(struct ip); 2327 break; 2328 default: 2329 hlen = 0; 2330 break; 2331 } 2332 hlen += sizeof(struct tcphdr); 2333 2334 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2335 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2336 hlen += TCPOLEN_TSTAMP_APPA; 2337 #ifdef TCP_SIGNATURE 2338 if (tp->t_flags & TF_SIGNATURE) 2339 hlen += TCPOLEN_SIGLEN; 2340 #endif 2341 return hlen; 2342 } 2343