tcp_subr.c revision 1.213
1 /* $NetBSD: tcp_subr.c,v 1.213 2007/03/12 18:18:36 ad 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 1.213 2007/03/12 18:18:36 ad 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 IPL_SOFTNET); 241 242 #ifdef TCP_CSUM_COUNTERS 243 #include <sys/device.h> 244 245 #if defined(INET) 246 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 247 NULL, "tcp", "hwcsum bad"); 248 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 249 NULL, "tcp", "hwcsum ok"); 250 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp", "hwcsum data"); 252 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 NULL, "tcp", "swcsum"); 254 255 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 256 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 257 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 258 EVCNT_ATTACH_STATIC(tcp_swcsum); 259 #endif /* defined(INET) */ 260 261 #if defined(INET6) 262 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 263 NULL, "tcp6", "hwcsum bad"); 264 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 265 NULL, "tcp6", "hwcsum ok"); 266 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 NULL, "tcp6", "hwcsum data"); 268 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 269 NULL, "tcp6", "swcsum"); 270 271 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 272 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 273 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 274 EVCNT_ATTACH_STATIC(tcp6_swcsum); 275 #endif /* defined(INET6) */ 276 #endif /* TCP_CSUM_COUNTERS */ 277 278 279 #ifdef TCP_OUTPUT_COUNTERS 280 #include <sys/device.h> 281 282 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 NULL, "tcp", "output big header"); 284 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 285 NULL, "tcp", "output predict hit"); 286 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 287 NULL, "tcp", "output predict miss"); 288 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 289 NULL, "tcp", "output copy small"); 290 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 291 NULL, "tcp", "output copy big"); 292 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 293 NULL, "tcp", "output reference big"); 294 295 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 296 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 297 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 298 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 299 EVCNT_ATTACH_STATIC(tcp_output_copybig); 300 EVCNT_ATTACH_STATIC(tcp_output_refbig); 301 302 #endif /* TCP_OUTPUT_COUNTERS */ 303 304 #ifdef TCP_REASS_COUNTERS 305 #include <sys/device.h> 306 307 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 308 NULL, "tcp_reass", "calls"); 309 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 310 &tcp_reass_, "tcp_reass", "insert into empty queue"); 311 struct evcnt tcp_reass_iteration[8] = { 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 318 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 319 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 320 }; 321 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 322 &tcp_reass_, "tcp_reass", "prepend to first"); 323 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 324 &tcp_reass_, "tcp_reass", "prepend"); 325 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 326 &tcp_reass_, "tcp_reass", "insert"); 327 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 328 &tcp_reass_, "tcp_reass", "insert at tail"); 329 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 330 &tcp_reass_, "tcp_reass", "append"); 331 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 332 &tcp_reass_, "tcp_reass", "append to tail fragment"); 333 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 334 &tcp_reass_, "tcp_reass", "overlap at end"); 335 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 336 &tcp_reass_, "tcp_reass", "overlap at start"); 337 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 338 &tcp_reass_, "tcp_reass", "duplicate segment"); 339 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 340 &tcp_reass_, "tcp_reass", "duplicate fragment"); 341 342 EVCNT_ATTACH_STATIC(tcp_reass_); 343 EVCNT_ATTACH_STATIC(tcp_reass_empty); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 349 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 350 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 351 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 352 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 353 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 354 EVCNT_ATTACH_STATIC(tcp_reass_insert); 355 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 356 EVCNT_ATTACH_STATIC(tcp_reass_append); 357 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 358 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 359 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 360 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 361 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 362 363 #endif /* TCP_REASS_COUNTERS */ 364 365 #ifdef MBUFTRACE 366 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 367 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 368 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 369 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 370 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 371 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 372 #endif 373 374 /* 375 * Tcp initialization 376 */ 377 void 378 tcp_init(void) 379 { 380 int hlen; 381 382 /* Initialize the TCPCB template. */ 383 tcp_tcpcb_template(); 384 385 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 386 387 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 388 #ifdef INET6 389 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 390 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 391 #endif 392 if (max_protohdr < hlen) 393 max_protohdr = hlen; 394 if (max_linkhdr + hlen > MHLEN) 395 panic("tcp_init"); 396 397 #ifdef INET 398 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 399 #endif 400 #ifdef INET6 401 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 402 #endif 403 404 /* Initialize timer state. */ 405 tcp_timer_init(); 406 407 /* Initialize the compressed state engine. */ 408 syn_cache_init(); 409 410 /* Initialize the congestion control algorithms. */ 411 tcp_congctl_init(); 412 413 MOWNER_ATTACH(&tcp_tx_mowner); 414 MOWNER_ATTACH(&tcp_rx_mowner); 415 MOWNER_ATTACH(&tcp_reass_mowner); 416 MOWNER_ATTACH(&tcp_sock_mowner); 417 MOWNER_ATTACH(&tcp_sock_tx_mowner); 418 MOWNER_ATTACH(&tcp_sock_rx_mowner); 419 MOWNER_ATTACH(&tcp_mowner); 420 } 421 422 /* 423 * Create template to be used to send tcp packets on a connection. 424 * Call after host entry created, allocates an mbuf and fills 425 * in a skeletal tcp/ip header, minimizing the amount of work 426 * necessary when the connection is used. 427 */ 428 struct mbuf * 429 tcp_template(struct tcpcb *tp) 430 { 431 struct inpcb *inp = tp->t_inpcb; 432 #ifdef INET6 433 struct in6pcb *in6p = tp->t_in6pcb; 434 #endif 435 struct tcphdr *n; 436 struct mbuf *m; 437 int hlen; 438 439 switch (tp->t_family) { 440 case AF_INET: 441 hlen = sizeof(struct ip); 442 if (inp) 443 break; 444 #ifdef INET6 445 if (in6p) { 446 /* mapped addr case */ 447 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 448 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 449 break; 450 } 451 #endif 452 return NULL; /*EINVAL*/ 453 #ifdef INET6 454 case AF_INET6: 455 hlen = sizeof(struct ip6_hdr); 456 if (in6p) { 457 /* more sainty check? */ 458 break; 459 } 460 return NULL; /*EINVAL*/ 461 #endif 462 default: 463 hlen = 0; /*pacify gcc*/ 464 return NULL; /*EAFNOSUPPORT*/ 465 } 466 #ifdef DIAGNOSTIC 467 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 468 panic("mclbytes too small for t_template"); 469 #endif 470 m = tp->t_template; 471 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 472 ; 473 else { 474 if (m) 475 m_freem(m); 476 m = tp->t_template = NULL; 477 MGETHDR(m, M_DONTWAIT, MT_HEADER); 478 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 479 MCLGET(m, M_DONTWAIT); 480 if ((m->m_flags & M_EXT) == 0) { 481 m_free(m); 482 m = NULL; 483 } 484 } 485 if (m == NULL) 486 return NULL; 487 MCLAIM(m, &tcp_mowner); 488 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 489 } 490 491 bzero(mtod(m, void *), m->m_len); 492 493 n = (struct tcphdr *)(mtod(m, char *) + hlen); 494 495 switch (tp->t_family) { 496 case AF_INET: 497 { 498 struct ipovly *ipov; 499 mtod(m, struct ip *)->ip_v = 4; 500 mtod(m, struct ip *)->ip_hl = hlen >> 2; 501 ipov = mtod(m, struct ipovly *); 502 ipov->ih_pr = IPPROTO_TCP; 503 ipov->ih_len = htons(sizeof(struct tcphdr)); 504 if (inp) { 505 ipov->ih_src = inp->inp_laddr; 506 ipov->ih_dst = inp->inp_faddr; 507 } 508 #ifdef INET6 509 else if (in6p) { 510 /* mapped addr case */ 511 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 512 sizeof(ipov->ih_src)); 513 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 514 sizeof(ipov->ih_dst)); 515 } 516 #endif 517 /* 518 * Compute the pseudo-header portion of the checksum 519 * now. We incrementally add in the TCP option and 520 * payload lengths later, and then compute the TCP 521 * checksum right before the packet is sent off onto 522 * the wire. 523 */ 524 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 525 ipov->ih_dst.s_addr, 526 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 527 break; 528 } 529 #ifdef INET6 530 case AF_INET6: 531 { 532 struct ip6_hdr *ip6; 533 mtod(m, struct ip *)->ip_v = 6; 534 ip6 = mtod(m, struct ip6_hdr *); 535 ip6->ip6_nxt = IPPROTO_TCP; 536 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 537 ip6->ip6_src = in6p->in6p_laddr; 538 ip6->ip6_dst = in6p->in6p_faddr; 539 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 540 if (ip6_auto_flowlabel) { 541 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 542 ip6->ip6_flow |= 543 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 544 } 545 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 546 ip6->ip6_vfc |= IPV6_VERSION; 547 548 /* 549 * Compute the pseudo-header portion of the checksum 550 * now. We incrementally add in the TCP option and 551 * payload lengths later, and then compute the TCP 552 * checksum right before the packet is sent off onto 553 * the wire. 554 */ 555 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 556 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 557 htonl(IPPROTO_TCP)); 558 break; 559 } 560 #endif 561 } 562 if (inp) { 563 n->th_sport = inp->inp_lport; 564 n->th_dport = inp->inp_fport; 565 } 566 #ifdef INET6 567 else if (in6p) { 568 n->th_sport = in6p->in6p_lport; 569 n->th_dport = in6p->in6p_fport; 570 } 571 #endif 572 n->th_seq = 0; 573 n->th_ack = 0; 574 n->th_x2 = 0; 575 n->th_off = 5; 576 n->th_flags = 0; 577 n->th_win = 0; 578 n->th_urp = 0; 579 return (m); 580 } 581 582 /* 583 * Send a single message to the TCP at address specified by 584 * the given TCP/IP header. If m == 0, then we make a copy 585 * of the tcpiphdr at ti and send directly to the addressed host. 586 * This is used to force keep alive messages out using the TCP 587 * template for a connection tp->t_template. If flags are given 588 * then we send a message back to the TCP which originated the 589 * segment ti, and discard the mbuf containing it and any other 590 * attached mbufs. 591 * 592 * In any case the ack and sequence number of the transmitted 593 * segment are as specified by the parameters. 594 */ 595 int 596 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 597 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 598 { 599 struct route *ro; 600 int error, tlen, win = 0; 601 int hlen; 602 struct ip *ip; 603 #ifdef INET6 604 struct ip6_hdr *ip6; 605 #endif 606 int family; /* family on packet, not inpcb/in6pcb! */ 607 struct tcphdr *th; 608 struct socket *so; 609 610 if (tp != NULL && (flags & TH_RST) == 0) { 611 #ifdef DIAGNOSTIC 612 if (tp->t_inpcb && tp->t_in6pcb) 613 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 614 #endif 615 #ifdef INET 616 if (tp->t_inpcb) 617 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 618 #endif 619 #ifdef INET6 620 if (tp->t_in6pcb) 621 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 622 #endif 623 } 624 625 th = NULL; /* Quell uninitialized warning */ 626 ip = NULL; 627 #ifdef INET6 628 ip6 = NULL; 629 #endif 630 if (m == 0) { 631 if (!template) 632 return EINVAL; 633 634 /* get family information from template */ 635 switch (mtod(template, struct ip *)->ip_v) { 636 case 4: 637 family = AF_INET; 638 hlen = sizeof(struct ip); 639 break; 640 #ifdef INET6 641 case 6: 642 family = AF_INET6; 643 hlen = sizeof(struct ip6_hdr); 644 break; 645 #endif 646 default: 647 return EAFNOSUPPORT; 648 } 649 650 MGETHDR(m, M_DONTWAIT, MT_HEADER); 651 if (m) { 652 MCLAIM(m, &tcp_tx_mowner); 653 MCLGET(m, M_DONTWAIT); 654 if ((m->m_flags & M_EXT) == 0) { 655 m_free(m); 656 m = NULL; 657 } 658 } 659 if (m == NULL) 660 return (ENOBUFS); 661 662 if (tcp_compat_42) 663 tlen = 1; 664 else 665 tlen = 0; 666 667 m->m_data += max_linkhdr; 668 bcopy(mtod(template, void *), mtod(m, void *), 669 template->m_len); 670 switch (family) { 671 case AF_INET: 672 ip = mtod(m, struct ip *); 673 th = (struct tcphdr *)(ip + 1); 674 break; 675 #ifdef INET6 676 case AF_INET6: 677 ip6 = mtod(m, struct ip6_hdr *); 678 th = (struct tcphdr *)(ip6 + 1); 679 break; 680 #endif 681 #if 0 682 default: 683 /* noone will visit here */ 684 m_freem(m); 685 return EAFNOSUPPORT; 686 #endif 687 } 688 flags = TH_ACK; 689 } else { 690 691 if ((m->m_flags & M_PKTHDR) == 0) { 692 #if 0 693 printf("non PKTHDR to tcp_respond\n"); 694 #endif 695 m_freem(m); 696 return EINVAL; 697 } 698 #ifdef DIAGNOSTIC 699 if (!th0) 700 panic("th0 == NULL in tcp_respond"); 701 #endif 702 703 /* get family information from m */ 704 switch (mtod(m, struct ip *)->ip_v) { 705 case 4: 706 family = AF_INET; 707 hlen = sizeof(struct ip); 708 ip = mtod(m, struct ip *); 709 break; 710 #ifdef INET6 711 case 6: 712 family = AF_INET6; 713 hlen = sizeof(struct ip6_hdr); 714 ip6 = mtod(m, struct ip6_hdr *); 715 break; 716 #endif 717 default: 718 m_freem(m); 719 return EAFNOSUPPORT; 720 } 721 /* clear h/w csum flags inherited from rx packet */ 722 m->m_pkthdr.csum_flags = 0; 723 724 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 725 tlen = sizeof(*th0); 726 else 727 tlen = th0->th_off << 2; 728 729 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 730 mtod(m, char *) + hlen == (char *)th0) { 731 m->m_len = hlen + tlen; 732 m_freem(m->m_next); 733 m->m_next = NULL; 734 } else { 735 struct mbuf *n; 736 737 #ifdef DIAGNOSTIC 738 if (max_linkhdr + hlen + tlen > MCLBYTES) { 739 m_freem(m); 740 return EMSGSIZE; 741 } 742 #endif 743 MGETHDR(n, M_DONTWAIT, MT_HEADER); 744 if (n && max_linkhdr + hlen + tlen > MHLEN) { 745 MCLGET(n, M_DONTWAIT); 746 if ((n->m_flags & M_EXT) == 0) { 747 m_freem(n); 748 n = NULL; 749 } 750 } 751 if (!n) { 752 m_freem(m); 753 return ENOBUFS; 754 } 755 756 MCLAIM(n, &tcp_tx_mowner); 757 n->m_data += max_linkhdr; 758 n->m_len = hlen + tlen; 759 m_copyback(n, 0, hlen, mtod(m, void *)); 760 m_copyback(n, hlen, tlen, (void *)th0); 761 762 m_freem(m); 763 m = n; 764 n = NULL; 765 } 766 767 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 768 switch (family) { 769 case AF_INET: 770 ip = mtod(m, struct ip *); 771 th = (struct tcphdr *)(ip + 1); 772 ip->ip_p = IPPROTO_TCP; 773 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 774 ip->ip_p = IPPROTO_TCP; 775 break; 776 #ifdef INET6 777 case AF_INET6: 778 ip6 = mtod(m, struct ip6_hdr *); 779 th = (struct tcphdr *)(ip6 + 1); 780 ip6->ip6_nxt = IPPROTO_TCP; 781 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 782 ip6->ip6_nxt = IPPROTO_TCP; 783 break; 784 #endif 785 #if 0 786 default: 787 /* noone will visit here */ 788 m_freem(m); 789 return EAFNOSUPPORT; 790 #endif 791 } 792 xchg(th->th_dport, th->th_sport, u_int16_t); 793 #undef xchg 794 tlen = 0; /*be friendly with the following code*/ 795 } 796 th->th_seq = htonl(seq); 797 th->th_ack = htonl(ack); 798 th->th_x2 = 0; 799 if ((flags & TH_SYN) == 0) { 800 if (tp) 801 win >>= tp->rcv_scale; 802 if (win > TCP_MAXWIN) 803 win = TCP_MAXWIN; 804 th->th_win = htons((u_int16_t)win); 805 th->th_off = sizeof (struct tcphdr) >> 2; 806 tlen += sizeof(*th); 807 } else 808 tlen += th->th_off << 2; 809 m->m_len = hlen + tlen; 810 m->m_pkthdr.len = hlen + tlen; 811 m->m_pkthdr.rcvif = (struct ifnet *) 0; 812 th->th_flags = flags; 813 th->th_urp = 0; 814 815 switch (family) { 816 #ifdef INET 817 case AF_INET: 818 { 819 struct ipovly *ipov = (struct ipovly *)ip; 820 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 821 ipov->ih_len = htons((u_int16_t)tlen); 822 823 th->th_sum = 0; 824 th->th_sum = in_cksum(m, hlen + tlen); 825 ip->ip_len = htons(hlen + tlen); 826 ip->ip_ttl = ip_defttl; 827 break; 828 } 829 #endif 830 #ifdef INET6 831 case AF_INET6: 832 { 833 th->th_sum = 0; 834 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 835 tlen); 836 ip6->ip6_plen = htons(tlen); 837 if (tp && tp->t_in6pcb) { 838 struct ifnet *oifp; 839 ro = (struct route *)&tp->t_in6pcb->in6p_route; 840 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 841 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 842 } else 843 ip6->ip6_hlim = ip6_defhlim; 844 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 845 if (ip6_auto_flowlabel) { 846 ip6->ip6_flow |= 847 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 848 } 849 break; 850 } 851 #endif 852 } 853 854 if (tp && tp->t_inpcb) 855 so = tp->t_inpcb->inp_socket; 856 #ifdef INET6 857 else if (tp && tp->t_in6pcb) 858 so = tp->t_in6pcb->in6p_socket; 859 #endif 860 else 861 so = NULL; 862 863 if (tp != NULL && tp->t_inpcb != NULL) { 864 ro = &tp->t_inpcb->inp_route; 865 #ifdef DIAGNOSTIC 866 if (family != AF_INET) 867 panic("tcp_respond: address family mismatch"); 868 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 869 panic("tcp_respond: ip_dst %x != inp_faddr %x", 870 ntohl(ip->ip_dst.s_addr), 871 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 872 } 873 #endif 874 } 875 #ifdef INET6 876 else if (tp != NULL && tp->t_in6pcb != NULL) { 877 ro = (struct route *)&tp->t_in6pcb->in6p_route; 878 #ifdef DIAGNOSTIC 879 if (family == AF_INET) { 880 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 881 panic("tcp_respond: not mapped addr"); 882 if (bcmp(&ip->ip_dst, 883 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 884 sizeof(ip->ip_dst)) != 0) { 885 panic("tcp_respond: ip_dst != in6p_faddr"); 886 } 887 } else if (family == AF_INET6) { 888 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 889 &tp->t_in6pcb->in6p_faddr)) 890 panic("tcp_respond: ip6_dst != in6p_faddr"); 891 } else 892 panic("tcp_respond: address family mismatch"); 893 #endif 894 } 895 #endif 896 else 897 ro = NULL; 898 899 switch (family) { 900 #ifdef INET 901 case AF_INET: 902 error = ip_output(m, NULL, ro, 903 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 904 (struct ip_moptions *)0, so); 905 break; 906 #endif 907 #ifdef INET6 908 case AF_INET6: 909 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 910 (struct ip6_moptions *)0, 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 /* 929 * If TCP_NTIMERS ever changes, we'll need to update this 930 * initializer. 931 */ 932 .t_timer = { 933 CALLOUT_INITIALIZER, 934 CALLOUT_INITIALIZER, 935 CALLOUT_INITIALIZER, 936 CALLOUT_INITIALIZER, 937 }, 938 .t_delack_ch = CALLOUT_INITIALIZER, 939 940 .t_srtt = TCPTV_SRTTBASE, 941 .t_rttmin = TCPTV_MIN, 942 943 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 944 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 945 .snd_numholes = 0, 946 947 .t_partialacks = -1, 948 .t_bytes_acked = 0, 949 }; 950 951 /* 952 * Updates the TCPCB template whenever a parameter that would affect 953 * the template is changed. 954 */ 955 void 956 tcp_tcpcb_template(void) 957 { 958 struct tcpcb *tp = &tcpcb_template; 959 int flags; 960 961 tp->t_peermss = tcp_mssdflt; 962 tp->t_ourmss = tcp_mssdflt; 963 tp->t_segsz = tcp_mssdflt; 964 965 flags = 0; 966 if (tcp_do_rfc1323 && tcp_do_win_scale) 967 flags |= TF_REQ_SCALE; 968 if (tcp_do_rfc1323 && tcp_do_timestamps) 969 flags |= TF_REQ_TSTMP; 970 tp->t_flags = flags; 971 972 /* 973 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 974 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 975 * reasonable initial retransmit time. 976 */ 977 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 978 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 979 TCPTV_MIN, TCPTV_REXMTMAX); 980 } 981 982 /* 983 * Create a new TCP control block, making an 984 * empty reassembly queue and hooking it to the argument 985 * protocol control block. 986 */ 987 /* family selects inpcb, or in6pcb */ 988 struct tcpcb * 989 tcp_newtcpcb(int family, void *aux) 990 { 991 struct tcpcb *tp; 992 int i; 993 994 /* XXX Consider using a pool_cache for speed. */ 995 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 996 if (tp == NULL) 997 return (NULL); 998 memcpy(tp, &tcpcb_template, sizeof(*tp)); 999 TAILQ_INIT(&tp->segq); 1000 TAILQ_INIT(&tp->timeq); 1001 tp->t_family = family; /* may be overridden later on */ 1002 TAILQ_INIT(&tp->snd_holes); 1003 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1004 1005 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1006 for (i = 0; i < TCPT_NTIMERS; i++) 1007 TCP_TIMER_INIT(tp, i); 1008 1009 switch (family) { 1010 case AF_INET: 1011 { 1012 struct inpcb *inp = (struct inpcb *)aux; 1013 1014 inp->inp_ip.ip_ttl = ip_defttl; 1015 inp->inp_ppcb = (void *)tp; 1016 1017 tp->t_inpcb = inp; 1018 tp->t_mtudisc = ip_mtudisc; 1019 break; 1020 } 1021 #ifdef INET6 1022 case AF_INET6: 1023 { 1024 struct in6pcb *in6p = (struct in6pcb *)aux; 1025 1026 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1027 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1028 : NULL); 1029 in6p->in6p_ppcb = (void *)tp; 1030 1031 tp->t_in6pcb = in6p; 1032 /* for IPv6, always try to run path MTU discovery */ 1033 tp->t_mtudisc = 1; 1034 break; 1035 } 1036 #endif /* INET6 */ 1037 default: 1038 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1039 return (NULL); 1040 } 1041 1042 /* 1043 * Initialize our timebase. When we send timestamps, we take 1044 * the delta from tcp_now -- this means each connection always 1045 * gets a timebase of 0, which makes it, among other things, 1046 * more difficult to determine how long a system has been up, 1047 * and thus how many TCP sequence increments have occurred. 1048 */ 1049 tp->ts_timebase = tcp_now; 1050 1051 tp->t_congctl = tcp_congctl_global; 1052 tp->t_congctl->refcnt++; 1053 1054 return (tp); 1055 } 1056 1057 /* 1058 * Drop a TCP connection, reporting 1059 * the specified error. If connection is synchronized, 1060 * then send a RST to peer. 1061 */ 1062 struct tcpcb * 1063 tcp_drop(struct tcpcb *tp, int errno) 1064 { 1065 struct socket *so = NULL; 1066 1067 #ifdef DIAGNOSTIC 1068 if (tp->t_inpcb && tp->t_in6pcb) 1069 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1070 #endif 1071 #ifdef INET 1072 if (tp->t_inpcb) 1073 so = tp->t_inpcb->inp_socket; 1074 #endif 1075 #ifdef INET6 1076 if (tp->t_in6pcb) 1077 so = tp->t_in6pcb->in6p_socket; 1078 #endif 1079 if (!so) 1080 return NULL; 1081 1082 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1083 tp->t_state = TCPS_CLOSED; 1084 (void) tcp_output(tp); 1085 tcpstat.tcps_drops++; 1086 } else 1087 tcpstat.tcps_conndrops++; 1088 if (errno == ETIMEDOUT && tp->t_softerror) 1089 errno = tp->t_softerror; 1090 so->so_error = errno; 1091 return (tcp_close(tp)); 1092 } 1093 1094 /* 1095 * Return whether this tcpcb is marked as dead, indicating 1096 * to the calling timer function that no further action should 1097 * be taken, as we are about to release this tcpcb. The release 1098 * of the storage will be done if this is the last timer running. 1099 * 1100 * This should be called from the callout handler function after 1101 * callout_ack() is done, so that the number of invoking timer 1102 * functions is 0. 1103 */ 1104 int 1105 tcp_isdead(struct tcpcb *tp) 1106 { 1107 int dead = (tp->t_flags & TF_DEAD); 1108 1109 if (__predict_false(dead)) { 1110 if (tcp_timers_invoking(tp) > 0) 1111 /* not quite there yet -- count separately? */ 1112 return dead; 1113 tcpstat.tcps_delayed_free++; 1114 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_timer.c */ 1115 } 1116 return dead; 1117 } 1118 1119 /* 1120 * Close a TCP control block: 1121 * discard all space held by the tcp 1122 * discard internet protocol block 1123 * wake up any sleepers 1124 */ 1125 struct tcpcb * 1126 tcp_close(struct tcpcb *tp) 1127 { 1128 struct inpcb *inp; 1129 #ifdef INET6 1130 struct in6pcb *in6p; 1131 #endif 1132 struct socket *so; 1133 #ifdef RTV_RTT 1134 struct rtentry *rt; 1135 #endif 1136 struct route *ro; 1137 1138 inp = tp->t_inpcb; 1139 #ifdef INET6 1140 in6p = tp->t_in6pcb; 1141 #endif 1142 so = NULL; 1143 ro = NULL; 1144 if (inp) { 1145 so = inp->inp_socket; 1146 ro = &inp->inp_route; 1147 } 1148 #ifdef INET6 1149 else if (in6p) { 1150 so = in6p->in6p_socket; 1151 ro = (struct route *)&in6p->in6p_route; 1152 } 1153 #endif 1154 1155 #ifdef RTV_RTT 1156 /* 1157 * If we sent enough data to get some meaningful characteristics, 1158 * save them in the routing entry. 'Enough' is arbitrarily 1159 * defined as the sendpipesize (default 4K) * 16. This would 1160 * give us 16 rtt samples assuming we only get one sample per 1161 * window (the usual case on a long haul net). 16 samples is 1162 * enough for the srtt filter to converge to within 5% of the correct 1163 * value; fewer samples and we could save a very bogus rtt. 1164 * 1165 * Don't update the default route's characteristics and don't 1166 * update anything that the user "locked". 1167 */ 1168 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1169 ro && (rt = ro->ro_rt) && 1170 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1171 u_long i = 0; 1172 1173 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1174 i = tp->t_srtt * 1175 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1176 if (rt->rt_rmx.rmx_rtt && i) 1177 /* 1178 * filter this update to half the old & half 1179 * the new values, converting scale. 1180 * See route.h and tcp_var.h for a 1181 * description of the scaling constants. 1182 */ 1183 rt->rt_rmx.rmx_rtt = 1184 (rt->rt_rmx.rmx_rtt + i) / 2; 1185 else 1186 rt->rt_rmx.rmx_rtt = i; 1187 } 1188 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1189 i = tp->t_rttvar * 1190 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1191 if (rt->rt_rmx.rmx_rttvar && i) 1192 rt->rt_rmx.rmx_rttvar = 1193 (rt->rt_rmx.rmx_rttvar + i) / 2; 1194 else 1195 rt->rt_rmx.rmx_rttvar = i; 1196 } 1197 /* 1198 * update the pipelimit (ssthresh) if it has been updated 1199 * already or if a pipesize was specified & the threshhold 1200 * got below half the pipesize. I.e., wait for bad news 1201 * before we start updating, then update on both good 1202 * and bad news. 1203 */ 1204 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1205 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1206 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1207 /* 1208 * convert the limit from user data bytes to 1209 * packets then to packet data bytes. 1210 */ 1211 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1212 if (i < 2) 1213 i = 2; 1214 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1215 if (rt->rt_rmx.rmx_ssthresh) 1216 rt->rt_rmx.rmx_ssthresh = 1217 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1218 else 1219 rt->rt_rmx.rmx_ssthresh = i; 1220 } 1221 } 1222 #endif /* RTV_RTT */ 1223 /* free the reassembly queue, if any */ 1224 TCP_REASS_LOCK(tp); 1225 (void) tcp_freeq(tp); 1226 TCP_REASS_UNLOCK(tp); 1227 1228 /* free the SACK holes list. */ 1229 tcp_free_sackholes(tp); 1230 1231 tp->t_congctl->refcnt--; 1232 1233 tcp_canceltimers(tp); 1234 TCP_CLEAR_DELACK(tp); 1235 syn_cache_cleanup(tp); 1236 1237 if (tp->t_template) { 1238 m_free(tp->t_template); 1239 tp->t_template = NULL; 1240 } 1241 if (tcp_timers_invoking(tp)) 1242 tp->t_flags |= TF_DEAD; 1243 else 1244 pool_put(&tcpcb_pool, tp); 1245 1246 if (inp) { 1247 inp->inp_ppcb = 0; 1248 soisdisconnected(so); 1249 in_pcbdetach(inp); 1250 } 1251 #ifdef INET6 1252 else if (in6p) { 1253 in6p->in6p_ppcb = 0; 1254 soisdisconnected(so); 1255 in6_pcbdetach(in6p); 1256 } 1257 #endif 1258 tcpstat.tcps_closed++; 1259 return ((struct tcpcb *)0); 1260 } 1261 1262 int 1263 tcp_freeq(tp) 1264 struct tcpcb *tp; 1265 { 1266 struct ipqent *qe; 1267 int rv = 0; 1268 #ifdef TCPREASS_DEBUG 1269 int i = 0; 1270 #endif 1271 1272 TCP_REASS_LOCK_CHECK(tp); 1273 1274 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1275 #ifdef TCPREASS_DEBUG 1276 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1277 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1278 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1279 #endif 1280 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1281 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1282 m_freem(qe->ipqe_m); 1283 tcpipqent_free(qe); 1284 rv = 1; 1285 } 1286 tp->t_segqlen = 0; 1287 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1288 return (rv); 1289 } 1290 1291 /* 1292 * Protocol drain routine. Called when memory is in short supply. 1293 */ 1294 void 1295 tcp_drain(void) 1296 { 1297 struct inpcb_hdr *inph; 1298 struct tcpcb *tp; 1299 1300 /* 1301 * Free the sequence queue of all TCP connections. 1302 */ 1303 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1304 switch (inph->inph_af) { 1305 case AF_INET: 1306 tp = intotcpcb((struct inpcb *)inph); 1307 break; 1308 #ifdef INET6 1309 case AF_INET6: 1310 tp = in6totcpcb((struct in6pcb *)inph); 1311 break; 1312 #endif 1313 default: 1314 tp = NULL; 1315 break; 1316 } 1317 if (tp != NULL) { 1318 /* 1319 * We may be called from a device's interrupt 1320 * context. If the tcpcb is already busy, 1321 * just bail out now. 1322 */ 1323 if (tcp_reass_lock_try(tp) == 0) 1324 continue; 1325 if (tcp_freeq(tp)) 1326 tcpstat.tcps_connsdrained++; 1327 TCP_REASS_UNLOCK(tp); 1328 } 1329 } 1330 } 1331 1332 /* 1333 * Notify a tcp user of an asynchronous error; 1334 * store error as soft error, but wake up user 1335 * (for now, won't do anything until can select for soft error). 1336 */ 1337 void 1338 tcp_notify(struct inpcb *inp, int error) 1339 { 1340 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1341 struct socket *so = inp->inp_socket; 1342 1343 /* 1344 * Ignore some errors if we are hooked up. 1345 * If connection hasn't completed, has retransmitted several times, 1346 * and receives a second error, give up now. This is better 1347 * than waiting a long time to establish a connection that 1348 * can never complete. 1349 */ 1350 if (tp->t_state == TCPS_ESTABLISHED && 1351 (error == EHOSTUNREACH || error == ENETUNREACH || 1352 error == EHOSTDOWN)) { 1353 return; 1354 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1355 tp->t_rxtshift > 3 && tp->t_softerror) 1356 so->so_error = error; 1357 else 1358 tp->t_softerror = error; 1359 wakeup((void *) &so->so_timeo); 1360 sorwakeup(so); 1361 sowwakeup(so); 1362 } 1363 1364 #ifdef INET6 1365 void 1366 tcp6_notify(struct in6pcb *in6p, int error) 1367 { 1368 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1369 struct socket *so = in6p->in6p_socket; 1370 1371 /* 1372 * Ignore some errors if we are hooked up. 1373 * If connection hasn't completed, has retransmitted several times, 1374 * and receives a second error, give up now. This is better 1375 * than waiting a long time to establish a connection that 1376 * can never complete. 1377 */ 1378 if (tp->t_state == TCPS_ESTABLISHED && 1379 (error == EHOSTUNREACH || error == ENETUNREACH || 1380 error == EHOSTDOWN)) { 1381 return; 1382 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1383 tp->t_rxtshift > 3 && tp->t_softerror) 1384 so->so_error = error; 1385 else 1386 tp->t_softerror = error; 1387 wakeup((void *) &so->so_timeo); 1388 sorwakeup(so); 1389 sowwakeup(so); 1390 } 1391 #endif 1392 1393 #ifdef INET6 1394 void 1395 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1396 { 1397 struct tcphdr th; 1398 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1399 int nmatch; 1400 struct ip6_hdr *ip6; 1401 const struct sockaddr_in6 *sa6_src = NULL; 1402 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1403 struct mbuf *m; 1404 int off; 1405 1406 if (sa->sa_family != AF_INET6 || 1407 sa->sa_len != sizeof(struct sockaddr_in6)) 1408 return; 1409 if ((unsigned)cmd >= PRC_NCMDS) 1410 return; 1411 else if (cmd == PRC_QUENCH) { 1412 /* 1413 * Don't honor ICMP Source Quench messages meant for 1414 * TCP connections. 1415 */ 1416 return; 1417 } else if (PRC_IS_REDIRECT(cmd)) 1418 notify = in6_rtchange, d = NULL; 1419 else if (cmd == PRC_MSGSIZE) 1420 ; /* special code is present, see below */ 1421 else if (cmd == PRC_HOSTDEAD) 1422 d = NULL; 1423 else if (inet6ctlerrmap[cmd] == 0) 1424 return; 1425 1426 /* if the parameter is from icmp6, decode it. */ 1427 if (d != NULL) { 1428 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1429 m = ip6cp->ip6c_m; 1430 ip6 = ip6cp->ip6c_ip6; 1431 off = ip6cp->ip6c_off; 1432 sa6_src = ip6cp->ip6c_src; 1433 } else { 1434 m = NULL; 1435 ip6 = NULL; 1436 sa6_src = &sa6_any; 1437 off = 0; 1438 } 1439 1440 if (ip6) { 1441 /* 1442 * XXX: We assume that when ip6 is non NULL, 1443 * M and OFF are valid. 1444 */ 1445 1446 /* check if we can safely examine src and dst ports */ 1447 if (m->m_pkthdr.len < off + sizeof(th)) { 1448 if (cmd == PRC_MSGSIZE) 1449 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1450 return; 1451 } 1452 1453 bzero(&th, sizeof(th)); 1454 m_copydata(m, off, sizeof(th), (void *)&th); 1455 1456 if (cmd == PRC_MSGSIZE) { 1457 int valid = 0; 1458 1459 /* 1460 * Check to see if we have a valid TCP connection 1461 * corresponding to the address in the ICMPv6 message 1462 * payload. 1463 */ 1464 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1465 th.th_dport, 1466 (const struct in6_addr *)&sa6_src->sin6_addr, 1467 th.th_sport, 0)) 1468 valid++; 1469 1470 /* 1471 * Depending on the value of "valid" and routing table 1472 * size (mtudisc_{hi,lo}wat), we will: 1473 * - recalcurate the new MTU and create the 1474 * corresponding routing entry, or 1475 * - ignore the MTU change notification. 1476 */ 1477 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1478 1479 /* 1480 * no need to call in6_pcbnotify, it should have been 1481 * called via callback if necessary 1482 */ 1483 return; 1484 } 1485 1486 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1487 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1488 if (nmatch == 0 && syn_cache_count && 1489 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1490 inet6ctlerrmap[cmd] == ENETUNREACH || 1491 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1492 syn_cache_unreach((const struct sockaddr *)sa6_src, 1493 sa, &th); 1494 } else { 1495 (void) in6_pcbnotify(&tcbtable, sa, 0, 1496 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1497 } 1498 } 1499 #endif 1500 1501 #ifdef INET 1502 /* assumes that ip header and tcp header are contiguous on mbuf */ 1503 void * 1504 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1505 { 1506 struct ip *ip = v; 1507 struct tcphdr *th; 1508 struct icmp *icp; 1509 extern const int inetctlerrmap[]; 1510 void (*notify)(struct inpcb *, int) = tcp_notify; 1511 int errno; 1512 int nmatch; 1513 struct tcpcb *tp; 1514 u_int mtu; 1515 tcp_seq seq; 1516 struct inpcb *inp; 1517 #ifdef INET6 1518 struct in6pcb *in6p; 1519 struct in6_addr src6, dst6; 1520 #endif 1521 1522 if (sa->sa_family != AF_INET || 1523 sa->sa_len != sizeof(struct sockaddr_in)) 1524 return NULL; 1525 if ((unsigned)cmd >= PRC_NCMDS) 1526 return NULL; 1527 errno = inetctlerrmap[cmd]; 1528 if (cmd == PRC_QUENCH) 1529 /* 1530 * Don't honor ICMP Source Quench messages meant for 1531 * TCP connections. 1532 */ 1533 return NULL; 1534 else if (PRC_IS_REDIRECT(cmd)) 1535 notify = in_rtchange, ip = 0; 1536 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1537 /* 1538 * Check to see if we have a valid TCP connection 1539 * corresponding to the address in the ICMP message 1540 * payload. 1541 * 1542 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1543 */ 1544 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1545 #ifdef INET6 1546 memset(&src6, 0, sizeof(src6)); 1547 memset(&dst6, 0, sizeof(dst6)); 1548 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1549 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1550 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1551 #endif 1552 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1553 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1554 #ifdef INET6 1555 in6p = NULL; 1556 #else 1557 ; 1558 #endif 1559 #ifdef INET6 1560 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1561 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1562 ; 1563 #endif 1564 else 1565 return NULL; 1566 1567 /* 1568 * Now that we've validated that we are actually communicating 1569 * with the host indicated in the ICMP message, locate the 1570 * ICMP header, recalculate the new MTU, and create the 1571 * corresponding routing entry. 1572 */ 1573 icp = (struct icmp *)((char *)ip - 1574 offsetof(struct icmp, icmp_ip)); 1575 if (inp) { 1576 if ((tp = intotcpcb(inp)) == NULL) 1577 return NULL; 1578 } 1579 #ifdef INET6 1580 else if (in6p) { 1581 if ((tp = in6totcpcb(in6p)) == NULL) 1582 return NULL; 1583 } 1584 #endif 1585 else 1586 return NULL; 1587 seq = ntohl(th->th_seq); 1588 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1589 return NULL; 1590 /* 1591 * If the ICMP message advertises a Next-Hop MTU 1592 * equal or larger than the maximum packet size we have 1593 * ever sent, drop the message. 1594 */ 1595 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1596 if (mtu >= tp->t_pmtud_mtu_sent) 1597 return NULL; 1598 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1599 /* 1600 * Calculate new MTU, and create corresponding 1601 * route (traditional PMTUD). 1602 */ 1603 tp->t_flags &= ~TF_PMTUD_PEND; 1604 icmp_mtudisc(icp, ip->ip_dst); 1605 } else { 1606 /* 1607 * Record the information got in the ICMP 1608 * message; act on it later. 1609 * If we had already recorded an ICMP message, 1610 * replace the old one only if the new message 1611 * refers to an older TCP segment 1612 */ 1613 if (tp->t_flags & TF_PMTUD_PEND) { 1614 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1615 return NULL; 1616 } else 1617 tp->t_flags |= TF_PMTUD_PEND; 1618 tp->t_pmtud_th_seq = seq; 1619 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1620 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1621 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1622 } 1623 return NULL; 1624 } else if (cmd == PRC_HOSTDEAD) 1625 ip = 0; 1626 else if (errno == 0) 1627 return NULL; 1628 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1629 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1630 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1631 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1632 if (nmatch == 0 && syn_cache_count && 1633 (inetctlerrmap[cmd] == EHOSTUNREACH || 1634 inetctlerrmap[cmd] == ENETUNREACH || 1635 inetctlerrmap[cmd] == EHOSTDOWN)) { 1636 struct sockaddr_in sin; 1637 bzero(&sin, sizeof(sin)); 1638 sin.sin_len = sizeof(sin); 1639 sin.sin_family = AF_INET; 1640 sin.sin_port = th->th_sport; 1641 sin.sin_addr = ip->ip_src; 1642 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1643 } 1644 1645 /* XXX mapped address case */ 1646 } else 1647 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1648 notify); 1649 return NULL; 1650 } 1651 1652 /* 1653 * When a source quench is received, we are being notified of congestion. 1654 * Close the congestion window down to the Loss Window (one segment). 1655 * We will gradually open it again as we proceed. 1656 */ 1657 void 1658 tcp_quench(struct inpcb *inp, int errno) 1659 { 1660 struct tcpcb *tp = intotcpcb(inp); 1661 1662 if (tp) { 1663 tp->snd_cwnd = tp->t_segsz; 1664 tp->t_bytes_acked = 0; 1665 } 1666 } 1667 #endif 1668 1669 #ifdef INET6 1670 void 1671 tcp6_quench(struct in6pcb *in6p, int errno) 1672 { 1673 struct tcpcb *tp = in6totcpcb(in6p); 1674 1675 if (tp) { 1676 tp->snd_cwnd = tp->t_segsz; 1677 tp->t_bytes_acked = 0; 1678 } 1679 } 1680 #endif 1681 1682 #ifdef INET 1683 /* 1684 * Path MTU Discovery handlers. 1685 */ 1686 void 1687 tcp_mtudisc_callback(struct in_addr faddr) 1688 { 1689 #ifdef INET6 1690 struct in6_addr in6; 1691 #endif 1692 1693 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1694 #ifdef INET6 1695 memset(&in6, 0, sizeof(in6)); 1696 in6.s6_addr16[5] = 0xffff; 1697 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1698 tcp6_mtudisc_callback(&in6); 1699 #endif 1700 } 1701 1702 /* 1703 * On receipt of path MTU corrections, flush old route and replace it 1704 * with the new one. Retransmit all unacknowledged packets, to ensure 1705 * that all packets will be received. 1706 */ 1707 void 1708 tcp_mtudisc(struct inpcb *inp, int errno) 1709 { 1710 struct tcpcb *tp = intotcpcb(inp); 1711 struct rtentry *rt = in_pcbrtentry(inp); 1712 1713 if (tp != 0) { 1714 if (rt != 0) { 1715 /* 1716 * If this was not a host route, remove and realloc. 1717 */ 1718 if ((rt->rt_flags & RTF_HOST) == 0) { 1719 in_rtchange(inp, errno); 1720 if ((rt = in_pcbrtentry(inp)) == 0) 1721 return; 1722 } 1723 1724 /* 1725 * Slow start out of the error condition. We 1726 * use the MTU because we know it's smaller 1727 * than the previously transmitted segment. 1728 * 1729 * Note: This is more conservative than the 1730 * suggestion in draft-floyd-incr-init-win-03. 1731 */ 1732 if (rt->rt_rmx.rmx_mtu != 0) 1733 tp->snd_cwnd = 1734 TCP_INITIAL_WINDOW(tcp_init_win, 1735 rt->rt_rmx.rmx_mtu); 1736 } 1737 1738 /* 1739 * Resend unacknowledged packets. 1740 */ 1741 tp->snd_nxt = tp->snd_una; 1742 tcp_output(tp); 1743 } 1744 } 1745 #endif 1746 1747 #ifdef INET6 1748 /* 1749 * Path MTU Discovery handlers. 1750 */ 1751 void 1752 tcp6_mtudisc_callback(struct in6_addr *faddr) 1753 { 1754 struct sockaddr_in6 sin6; 1755 1756 bzero(&sin6, sizeof(sin6)); 1757 sin6.sin6_family = AF_INET6; 1758 sin6.sin6_len = sizeof(struct sockaddr_in6); 1759 sin6.sin6_addr = *faddr; 1760 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1761 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1762 } 1763 1764 void 1765 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1766 { 1767 struct tcpcb *tp = in6totcpcb(in6p); 1768 struct rtentry *rt = in6_pcbrtentry(in6p); 1769 1770 if (tp != 0) { 1771 if (rt != 0) { 1772 /* 1773 * If this was not a host route, remove and realloc. 1774 */ 1775 if ((rt->rt_flags & RTF_HOST) == 0) { 1776 in6_rtchange(in6p, errno); 1777 if ((rt = in6_pcbrtentry(in6p)) == 0) 1778 return; 1779 } 1780 1781 /* 1782 * Slow start out of the error condition. We 1783 * use the MTU because we know it's smaller 1784 * than the previously transmitted segment. 1785 * 1786 * Note: This is more conservative than the 1787 * suggestion in draft-floyd-incr-init-win-03. 1788 */ 1789 if (rt->rt_rmx.rmx_mtu != 0) 1790 tp->snd_cwnd = 1791 TCP_INITIAL_WINDOW(tcp_init_win, 1792 rt->rt_rmx.rmx_mtu); 1793 } 1794 1795 /* 1796 * Resend unacknowledged packets. 1797 */ 1798 tp->snd_nxt = tp->snd_una; 1799 tcp_output(tp); 1800 } 1801 } 1802 #endif /* INET6 */ 1803 1804 /* 1805 * Compute the MSS to advertise to the peer. Called only during 1806 * the 3-way handshake. If we are the server (peer initiated 1807 * connection), we are called with a pointer to the interface 1808 * on which the SYN packet arrived. If we are the client (we 1809 * initiated connection), we are called with a pointer to the 1810 * interface out which this connection should go. 1811 * 1812 * NOTE: Do not subtract IP option/extension header size nor IPsec 1813 * header size from MSS advertisement. MSS option must hold the maximum 1814 * segment size we can accept, so it must always be: 1815 * max(if mtu) - ip header - tcp header 1816 */ 1817 u_long 1818 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1819 { 1820 extern u_long in_maxmtu; 1821 u_long mss = 0; 1822 u_long hdrsiz; 1823 1824 /* 1825 * In order to avoid defeating path MTU discovery on the peer, 1826 * we advertise the max MTU of all attached networks as our MSS, 1827 * per RFC 1191, section 3.1. 1828 * 1829 * We provide the option to advertise just the MTU of 1830 * the interface on which we hope this connection will 1831 * be receiving. If we are responding to a SYN, we 1832 * will have a pretty good idea about this, but when 1833 * initiating a connection there is a bit more doubt. 1834 * 1835 * We also need to ensure that loopback has a large enough 1836 * MSS, as the loopback MTU is never included in in_maxmtu. 1837 */ 1838 1839 if (ifp != NULL) 1840 switch (af) { 1841 case AF_INET: 1842 mss = ifp->if_mtu; 1843 break; 1844 #ifdef INET6 1845 case AF_INET6: 1846 mss = IN6_LINKMTU(ifp); 1847 break; 1848 #endif 1849 } 1850 1851 if (tcp_mss_ifmtu == 0) 1852 switch (af) { 1853 case AF_INET: 1854 mss = max(in_maxmtu, mss); 1855 break; 1856 #ifdef INET6 1857 case AF_INET6: 1858 mss = max(in6_maxmtu, mss); 1859 break; 1860 #endif 1861 } 1862 1863 switch (af) { 1864 case AF_INET: 1865 hdrsiz = sizeof(struct ip); 1866 break; 1867 #ifdef INET6 1868 case AF_INET6: 1869 hdrsiz = sizeof(struct ip6_hdr); 1870 break; 1871 #endif 1872 default: 1873 hdrsiz = 0; 1874 break; 1875 } 1876 hdrsiz += sizeof(struct tcphdr); 1877 if (mss > hdrsiz) 1878 mss -= hdrsiz; 1879 1880 mss = max(tcp_mssdflt, mss); 1881 return (mss); 1882 } 1883 1884 /* 1885 * Set connection variables based on the peer's advertised MSS. 1886 * We are passed the TCPCB for the actual connection. If we 1887 * are the server, we are called by the compressed state engine 1888 * when the 3-way handshake is complete. If we are the client, 1889 * we are called when we receive the SYN,ACK from the server. 1890 * 1891 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1892 * before this routine is called! 1893 */ 1894 void 1895 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1896 { 1897 struct socket *so; 1898 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1899 struct rtentry *rt; 1900 #endif 1901 u_long bufsize; 1902 int mss; 1903 1904 #ifdef DIAGNOSTIC 1905 if (tp->t_inpcb && tp->t_in6pcb) 1906 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1907 #endif 1908 so = NULL; 1909 rt = NULL; 1910 #ifdef INET 1911 if (tp->t_inpcb) { 1912 so = tp->t_inpcb->inp_socket; 1913 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1914 rt = in_pcbrtentry(tp->t_inpcb); 1915 #endif 1916 } 1917 #endif 1918 #ifdef INET6 1919 if (tp->t_in6pcb) { 1920 so = tp->t_in6pcb->in6p_socket; 1921 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1922 rt = in6_pcbrtentry(tp->t_in6pcb); 1923 #endif 1924 } 1925 #endif 1926 1927 /* 1928 * As per RFC1122, use the default MSS value, unless they 1929 * sent us an offer. Do not accept offers less than 256 bytes. 1930 */ 1931 mss = tcp_mssdflt; 1932 if (offer) 1933 mss = offer; 1934 mss = max(mss, 256); /* sanity */ 1935 tp->t_peermss = mss; 1936 mss -= tcp_optlen(tp); 1937 #ifdef INET 1938 if (tp->t_inpcb) 1939 mss -= ip_optlen(tp->t_inpcb); 1940 #endif 1941 #ifdef INET6 1942 if (tp->t_in6pcb) 1943 mss -= ip6_optlen(tp->t_in6pcb); 1944 #endif 1945 1946 /* 1947 * If there's a pipesize, change the socket buffer to that size. 1948 * Make the socket buffer an integral number of MSS units. If 1949 * the MSS is larger than the socket buffer, artificially decrease 1950 * the MSS. 1951 */ 1952 #ifdef RTV_SPIPE 1953 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1954 bufsize = rt->rt_rmx.rmx_sendpipe; 1955 else 1956 #endif 1957 { 1958 KASSERT(so != NULL); 1959 bufsize = so->so_snd.sb_hiwat; 1960 } 1961 if (bufsize < mss) 1962 mss = bufsize; 1963 else { 1964 bufsize = roundup(bufsize, mss); 1965 if (bufsize > sb_max) 1966 bufsize = sb_max; 1967 (void) sbreserve(&so->so_snd, bufsize, so); 1968 } 1969 tp->t_segsz = mss; 1970 1971 #ifdef RTV_SSTHRESH 1972 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1973 /* 1974 * There's some sort of gateway or interface buffer 1975 * limit on the path. Use this to set the slow 1976 * start threshold, but set the threshold to no less 1977 * than 2 * MSS. 1978 */ 1979 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1980 } 1981 #endif 1982 } 1983 1984 /* 1985 * Processing necessary when a TCP connection is established. 1986 */ 1987 void 1988 tcp_established(struct tcpcb *tp) 1989 { 1990 struct socket *so; 1991 #ifdef RTV_RPIPE 1992 struct rtentry *rt; 1993 #endif 1994 u_long bufsize; 1995 1996 #ifdef DIAGNOSTIC 1997 if (tp->t_inpcb && tp->t_in6pcb) 1998 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1999 #endif 2000 so = NULL; 2001 rt = NULL; 2002 #ifdef INET 2003 if (tp->t_inpcb) { 2004 so = tp->t_inpcb->inp_socket; 2005 #if defined(RTV_RPIPE) 2006 rt = in_pcbrtentry(tp->t_inpcb); 2007 #endif 2008 } 2009 #endif 2010 #ifdef INET6 2011 if (tp->t_in6pcb) { 2012 so = tp->t_in6pcb->in6p_socket; 2013 #if defined(RTV_RPIPE) 2014 rt = in6_pcbrtentry(tp->t_in6pcb); 2015 #endif 2016 } 2017 #endif 2018 2019 tp->t_state = TCPS_ESTABLISHED; 2020 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 2021 2022 #ifdef RTV_RPIPE 2023 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2024 bufsize = rt->rt_rmx.rmx_recvpipe; 2025 else 2026 #endif 2027 { 2028 KASSERT(so != NULL); 2029 bufsize = so->so_rcv.sb_hiwat; 2030 } 2031 if (bufsize > tp->t_ourmss) { 2032 bufsize = roundup(bufsize, tp->t_ourmss); 2033 if (bufsize > sb_max) 2034 bufsize = sb_max; 2035 (void) sbreserve(&so->so_rcv, bufsize, so); 2036 } 2037 } 2038 2039 /* 2040 * Check if there's an initial rtt or rttvar. Convert from the 2041 * route-table units to scaled multiples of the slow timeout timer. 2042 * Called only during the 3-way handshake. 2043 */ 2044 void 2045 tcp_rmx_rtt(struct tcpcb *tp) 2046 { 2047 #ifdef RTV_RTT 2048 struct rtentry *rt = NULL; 2049 int rtt; 2050 2051 #ifdef DIAGNOSTIC 2052 if (tp->t_inpcb && tp->t_in6pcb) 2053 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2054 #endif 2055 #ifdef INET 2056 if (tp->t_inpcb) 2057 rt = in_pcbrtentry(tp->t_inpcb); 2058 #endif 2059 #ifdef INET6 2060 if (tp->t_in6pcb) 2061 rt = in6_pcbrtentry(tp->t_in6pcb); 2062 #endif 2063 if (rt == NULL) 2064 return; 2065 2066 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2067 /* 2068 * XXX The lock bit for MTU indicates that the value 2069 * is also a minimum value; this is subject to time. 2070 */ 2071 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2072 TCPT_RANGESET(tp->t_rttmin, 2073 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2074 TCPTV_MIN, TCPTV_REXMTMAX); 2075 tp->t_srtt = rtt / 2076 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2077 if (rt->rt_rmx.rmx_rttvar) { 2078 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2079 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2080 (TCP_RTTVAR_SHIFT + 2)); 2081 } else { 2082 /* Default variation is +- 1 rtt */ 2083 tp->t_rttvar = 2084 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2085 } 2086 TCPT_RANGESET(tp->t_rxtcur, 2087 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2088 tp->t_rttmin, TCPTV_REXMTMAX); 2089 } 2090 #endif 2091 } 2092 2093 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2094 #if NRND > 0 2095 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2096 #endif 2097 2098 /* 2099 * Get a new sequence value given a tcp control block 2100 */ 2101 tcp_seq 2102 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2103 { 2104 2105 #ifdef INET 2106 if (tp->t_inpcb != NULL) { 2107 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2108 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2109 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2110 addin)); 2111 } 2112 #endif 2113 #ifdef INET6 2114 if (tp->t_in6pcb != NULL) { 2115 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2116 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2117 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2118 addin)); 2119 } 2120 #endif 2121 /* Not possible. */ 2122 panic("tcp_new_iss"); 2123 } 2124 2125 /* 2126 * This routine actually generates a new TCP initial sequence number. 2127 */ 2128 tcp_seq 2129 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2130 size_t addrsz, tcp_seq addin) 2131 { 2132 tcp_seq tcp_iss; 2133 2134 #if NRND > 0 2135 static int beenhere; 2136 2137 /* 2138 * If we haven't been here before, initialize our cryptographic 2139 * hash secret. 2140 */ 2141 if (beenhere == 0) { 2142 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2143 RND_EXTRACT_ANY); 2144 beenhere = 1; 2145 } 2146 2147 if (tcp_do_rfc1948) { 2148 MD5_CTX ctx; 2149 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2150 2151 /* 2152 * Compute the base value of the ISS. It is a hash 2153 * of (saddr, sport, daddr, dport, secret). 2154 */ 2155 MD5Init(&ctx); 2156 2157 MD5Update(&ctx, (u_char *) laddr, addrsz); 2158 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2159 2160 MD5Update(&ctx, (u_char *) faddr, addrsz); 2161 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2162 2163 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2164 2165 MD5Final(hash, &ctx); 2166 2167 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2168 2169 /* 2170 * Now increment our "timer", and add it in to 2171 * the computed value. 2172 * 2173 * XXX Use `addin'? 2174 * XXX TCP_ISSINCR too large to use? 2175 */ 2176 tcp_iss_seq += TCP_ISSINCR; 2177 #ifdef TCPISS_DEBUG 2178 printf("ISS hash 0x%08x, ", tcp_iss); 2179 #endif 2180 tcp_iss += tcp_iss_seq + addin; 2181 #ifdef TCPISS_DEBUG 2182 printf("new ISS 0x%08x\n", tcp_iss); 2183 #endif 2184 } else 2185 #endif /* NRND > 0 */ 2186 { 2187 /* 2188 * Randomize. 2189 */ 2190 #if NRND > 0 2191 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2192 #else 2193 tcp_iss = arc4random(); 2194 #endif 2195 2196 /* 2197 * If we were asked to add some amount to a known value, 2198 * we will take a random value obtained above, mask off 2199 * the upper bits, and add in the known value. We also 2200 * add in a constant to ensure that we are at least a 2201 * certain distance from the original value. 2202 * 2203 * This is used when an old connection is in timed wait 2204 * and we have a new one coming in, for instance. 2205 */ 2206 if (addin != 0) { 2207 #ifdef TCPISS_DEBUG 2208 printf("Random %08x, ", tcp_iss); 2209 #endif 2210 tcp_iss &= TCP_ISS_RANDOM_MASK; 2211 tcp_iss += addin + TCP_ISSINCR; 2212 #ifdef TCPISS_DEBUG 2213 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2214 #endif 2215 } else { 2216 tcp_iss &= TCP_ISS_RANDOM_MASK; 2217 tcp_iss += tcp_iss_seq; 2218 tcp_iss_seq += TCP_ISSINCR; 2219 #ifdef TCPISS_DEBUG 2220 printf("ISS %08x\n", tcp_iss); 2221 #endif 2222 } 2223 } 2224 2225 if (tcp_compat_42) { 2226 /* 2227 * Limit it to the positive range for really old TCP 2228 * implementations. 2229 * Just AND off the top bit instead of checking if 2230 * is set first - saves a branch 50% of the time. 2231 */ 2232 tcp_iss &= 0x7fffffff; /* XXX */ 2233 } 2234 2235 return (tcp_iss); 2236 } 2237 2238 #if defined(IPSEC) || defined(FAST_IPSEC) 2239 /* compute ESP/AH header size for TCP, including outer IP header. */ 2240 size_t 2241 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2242 { 2243 struct inpcb *inp; 2244 size_t hdrsiz; 2245 2246 /* XXX mapped addr case (tp->t_in6pcb) */ 2247 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2248 return 0; 2249 switch (tp->t_family) { 2250 case AF_INET: 2251 /* XXX: should use currect direction. */ 2252 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2253 break; 2254 default: 2255 hdrsiz = 0; 2256 break; 2257 } 2258 2259 return hdrsiz; 2260 } 2261 2262 #ifdef INET6 2263 size_t 2264 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2265 { 2266 struct in6pcb *in6p; 2267 size_t hdrsiz; 2268 2269 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2270 return 0; 2271 switch (tp->t_family) { 2272 case AF_INET6: 2273 /* XXX: should use currect direction. */ 2274 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2275 break; 2276 case AF_INET: 2277 /* mapped address case - tricky */ 2278 default: 2279 hdrsiz = 0; 2280 break; 2281 } 2282 2283 return hdrsiz; 2284 } 2285 #endif 2286 #endif /*IPSEC*/ 2287 2288 /* 2289 * Determine the length of the TCP options for this connection. 2290 * 2291 * XXX: What do we do for SACK, when we add that? Just reserve 2292 * all of the space? Otherwise we can't exactly be incrementing 2293 * cwnd by an amount that varies depending on the amount we last 2294 * had to SACK! 2295 */ 2296 2297 u_int 2298 tcp_optlen(struct tcpcb *tp) 2299 { 2300 u_int optlen; 2301 2302 optlen = 0; 2303 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2304 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2305 optlen += TCPOLEN_TSTAMP_APPA; 2306 2307 #ifdef TCP_SIGNATURE 2308 if (tp->t_flags & TF_SIGNATURE) 2309 optlen += TCPOLEN_SIGNATURE + 2; 2310 #endif /* TCP_SIGNATURE */ 2311 2312 return optlen; 2313 } 2314 2315 u_int 2316 tcp_hdrsz(struct tcpcb *tp) 2317 { 2318 u_int hlen; 2319 2320 switch (tp->t_family) { 2321 #ifdef INET6 2322 case AF_INET6: 2323 hlen = sizeof(struct ip6_hdr); 2324 break; 2325 #endif 2326 case AF_INET: 2327 hlen = sizeof(struct ip); 2328 break; 2329 default: 2330 hlen = 0; 2331 break; 2332 } 2333 hlen += sizeof(struct tcphdr); 2334 2335 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2336 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2337 hlen += TCPOLEN_TSTAMP_APPA; 2338 #ifdef TCP_SIGNATURE 2339 if (tp->t_flags & TF_SIGNATURE) 2340 hlen += TCPOLEN_SIGLEN; 2341 #endif 2342 return hlen; 2343 } 2344
Indexes created Mon Sep 22 13:09:51 GMT 2025