tcp_subr.c revision 1.216
1 /* $NetBSD: tcp_subr.c,v 1.216 2007/07/09 21:11:12 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 1.216 2007/07/09 21:11:12 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 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 383 384 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 385 #ifdef INET6 386 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 387 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 388 #endif 389 if (max_protohdr < hlen) 390 max_protohdr = hlen; 391 if (max_linkhdr + hlen > MHLEN) 392 panic("tcp_init"); 393 394 #ifdef INET 395 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 396 #endif 397 #ifdef INET6 398 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 399 #endif 400 401 /* Initialize timer state. */ 402 tcp_timer_init(); 403 404 /* Initialize the compressed state engine. */ 405 syn_cache_init(); 406 407 /* Initialize the congestion control algorithms. */ 408 tcp_congctl_init(); 409 410 /* Initialize the TCPCB template. */ 411 tcp_tcpcb_template(); 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, ro, 0, NULL, 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 .t_srtt = TCPTV_SRTTBASE, 928 .t_rttmin = TCPTV_MIN, 929 930 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 931 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 932 .snd_numholes = 0, 933 934 .t_partialacks = -1, 935 .t_bytes_acked = 0, 936 }; 937 938 /* 939 * Updates the TCPCB template whenever a parameter that would affect 940 * the template is changed. 941 */ 942 void 943 tcp_tcpcb_template(void) 944 { 945 struct tcpcb *tp = &tcpcb_template; 946 int flags; 947 948 tp->t_peermss = tcp_mssdflt; 949 tp->t_ourmss = tcp_mssdflt; 950 tp->t_segsz = tcp_mssdflt; 951 952 flags = 0; 953 if (tcp_do_rfc1323 && tcp_do_win_scale) 954 flags |= TF_REQ_SCALE; 955 if (tcp_do_rfc1323 && tcp_do_timestamps) 956 flags |= TF_REQ_TSTMP; 957 tp->t_flags = flags; 958 959 /* 960 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 961 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 962 * reasonable initial retransmit time. 963 */ 964 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 965 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 966 TCPTV_MIN, TCPTV_REXMTMAX); 967 968 /* Keep Alive */ 969 tp->t_keepinit = tcp_keepinit; 970 tp->t_keepidle = tcp_keepidle; 971 tp->t_keepintvl = tcp_keepintvl; 972 tp->t_keepcnt = tcp_keepcnt; 973 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 974 } 975 976 /* 977 * Create a new TCP control block, making an 978 * empty reassembly queue and hooking it to the argument 979 * protocol control block. 980 */ 981 /* family selects inpcb, or in6pcb */ 982 struct tcpcb * 983 tcp_newtcpcb(int family, void *aux) 984 { 985 struct tcpcb *tp; 986 int i; 987 988 /* XXX Consider using a pool_cache for speed. */ 989 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 990 if (tp == NULL) 991 return (NULL); 992 memcpy(tp, &tcpcb_template, sizeof(*tp)); 993 TAILQ_INIT(&tp->segq); 994 TAILQ_INIT(&tp->timeq); 995 tp->t_family = family; /* may be overridden later on */ 996 TAILQ_INIT(&tp->snd_holes); 997 LIST_INIT(&tp->t_sc); /* XXX can template this */ 998 999 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1000 for (i = 0; i < TCPT_NTIMERS; i++) { 1001 callout_init(&tp->t_timer[i], 0); 1002 TCP_TIMER_INIT(tp, i); 1003 } 1004 callout_init(&tp->t_delack_ch, 0); 1005 1006 switch (family) { 1007 case AF_INET: 1008 { 1009 struct inpcb *inp = (struct inpcb *)aux; 1010 1011 inp->inp_ip.ip_ttl = ip_defttl; 1012 inp->inp_ppcb = (void *)tp; 1013 1014 tp->t_inpcb = inp; 1015 tp->t_mtudisc = ip_mtudisc; 1016 break; 1017 } 1018 #ifdef INET6 1019 case AF_INET6: 1020 { 1021 struct in6pcb *in6p = (struct in6pcb *)aux; 1022 1023 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1024 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1025 : NULL); 1026 in6p->in6p_ppcb = (void *)tp; 1027 1028 tp->t_in6pcb = in6p; 1029 /* for IPv6, always try to run path MTU discovery */ 1030 tp->t_mtudisc = 1; 1031 break; 1032 } 1033 #endif /* INET6 */ 1034 default: 1035 for (i = 0; i < TCPT_NTIMERS; i++) 1036 callout_destroy(&tp->t_timer[i]); 1037 callout_destroy(&tp->t_delack_ch); 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 i, 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 for (i = 0; i < TCPT_NTIMERS; i++) 1115 callout_destroy(&tp->t_timer[i]); 1116 callout_destroy(&tp->t_delack_ch); 1117 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_timer.c */ 1118 } 1119 return dead; 1120 } 1121 1122 /* 1123 * Close a TCP control block: 1124 * discard all space held by the tcp 1125 * discard internet protocol block 1126 * wake up any sleepers 1127 */ 1128 struct tcpcb * 1129 tcp_close(struct tcpcb *tp) 1130 { 1131 struct inpcb *inp; 1132 #ifdef INET6 1133 struct in6pcb *in6p; 1134 #endif 1135 struct socket *so; 1136 #ifdef RTV_RTT 1137 struct rtentry *rt; 1138 #endif 1139 struct route *ro; 1140 int j; 1141 1142 inp = tp->t_inpcb; 1143 #ifdef INET6 1144 in6p = tp->t_in6pcb; 1145 #endif 1146 so = NULL; 1147 ro = NULL; 1148 if (inp) { 1149 so = inp->inp_socket; 1150 ro = &inp->inp_route; 1151 } 1152 #ifdef INET6 1153 else if (in6p) { 1154 so = in6p->in6p_socket; 1155 ro = (struct route *)&in6p->in6p_route; 1156 } 1157 #endif 1158 1159 #ifdef RTV_RTT 1160 /* 1161 * If we sent enough data to get some meaningful characteristics, 1162 * save them in the routing entry. 'Enough' is arbitrarily 1163 * defined as the sendpipesize (default 4K) * 16. This would 1164 * give us 16 rtt samples assuming we only get one sample per 1165 * window (the usual case on a long haul net). 16 samples is 1166 * enough for the srtt filter to converge to within 5% of the correct 1167 * value; fewer samples and we could save a very bogus rtt. 1168 * 1169 * Don't update the default route's characteristics and don't 1170 * update anything that the user "locked". 1171 */ 1172 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1173 ro && (rt = ro->ro_rt) && 1174 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1175 u_long i = 0; 1176 1177 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1178 i = tp->t_srtt * 1179 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1180 if (rt->rt_rmx.rmx_rtt && i) 1181 /* 1182 * filter this update to half the old & half 1183 * the new values, converting scale. 1184 * See route.h and tcp_var.h for a 1185 * description of the scaling constants. 1186 */ 1187 rt->rt_rmx.rmx_rtt = 1188 (rt->rt_rmx.rmx_rtt + i) / 2; 1189 else 1190 rt->rt_rmx.rmx_rtt = i; 1191 } 1192 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1193 i = tp->t_rttvar * 1194 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1195 if (rt->rt_rmx.rmx_rttvar && i) 1196 rt->rt_rmx.rmx_rttvar = 1197 (rt->rt_rmx.rmx_rttvar + i) / 2; 1198 else 1199 rt->rt_rmx.rmx_rttvar = i; 1200 } 1201 /* 1202 * update the pipelimit (ssthresh) if it has been updated 1203 * already or if a pipesize was specified & the threshhold 1204 * got below half the pipesize. I.e., wait for bad news 1205 * before we start updating, then update on both good 1206 * and bad news. 1207 */ 1208 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1209 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1210 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1211 /* 1212 * convert the limit from user data bytes to 1213 * packets then to packet data bytes. 1214 */ 1215 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1216 if (i < 2) 1217 i = 2; 1218 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1219 if (rt->rt_rmx.rmx_ssthresh) 1220 rt->rt_rmx.rmx_ssthresh = 1221 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1222 else 1223 rt->rt_rmx.rmx_ssthresh = i; 1224 } 1225 } 1226 #endif /* RTV_RTT */ 1227 /* free the reassembly queue, if any */ 1228 TCP_REASS_LOCK(tp); 1229 (void) tcp_freeq(tp); 1230 TCP_REASS_UNLOCK(tp); 1231 1232 /* free the SACK holes list. */ 1233 tcp_free_sackholes(tp); 1234 1235 tp->t_congctl->refcnt--; 1236 1237 tcp_canceltimers(tp); 1238 TCP_CLEAR_DELACK(tp); 1239 syn_cache_cleanup(tp); 1240 1241 if (tp->t_template) { 1242 m_free(tp->t_template); 1243 tp->t_template = NULL; 1244 } 1245 if (tcp_timers_invoking(tp)) 1246 tp->t_flags |= TF_DEAD; 1247 else { 1248 for (j = 0; j < TCPT_NTIMERS; j++) 1249 callout_destroy(&tp->t_timer[j]); 1250 callout_destroy(&tp->t_delack_ch); 1251 pool_put(&tcpcb_pool, tp); 1252 } 1253 1254 if (inp) { 1255 inp->inp_ppcb = 0; 1256 soisdisconnected(so); 1257 in_pcbdetach(inp); 1258 } 1259 #ifdef INET6 1260 else if (in6p) { 1261 in6p->in6p_ppcb = 0; 1262 soisdisconnected(so); 1263 in6_pcbdetach(in6p); 1264 } 1265 #endif 1266 tcpstat.tcps_closed++; 1267 return ((struct tcpcb *)0); 1268 } 1269 1270 int 1271 tcp_freeq(tp) 1272 struct tcpcb *tp; 1273 { 1274 struct ipqent *qe; 1275 int rv = 0; 1276 #ifdef TCPREASS_DEBUG 1277 int i = 0; 1278 #endif 1279 1280 TCP_REASS_LOCK_CHECK(tp); 1281 1282 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1283 #ifdef TCPREASS_DEBUG 1284 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1285 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1286 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1287 #endif 1288 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1289 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1290 m_freem(qe->ipqe_m); 1291 tcpipqent_free(qe); 1292 rv = 1; 1293 } 1294 tp->t_segqlen = 0; 1295 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1296 return (rv); 1297 } 1298 1299 /* 1300 * Protocol drain routine. Called when memory is in short supply. 1301 */ 1302 void 1303 tcp_drain(void) 1304 { 1305 struct inpcb_hdr *inph; 1306 struct tcpcb *tp; 1307 1308 /* 1309 * Free the sequence queue of all TCP connections. 1310 */ 1311 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1312 switch (inph->inph_af) { 1313 case AF_INET: 1314 tp = intotcpcb((struct inpcb *)inph); 1315 break; 1316 #ifdef INET6 1317 case AF_INET6: 1318 tp = in6totcpcb((struct in6pcb *)inph); 1319 break; 1320 #endif 1321 default: 1322 tp = NULL; 1323 break; 1324 } 1325 if (tp != NULL) { 1326 /* 1327 * We may be called from a device's interrupt 1328 * context. If the tcpcb is already busy, 1329 * just bail out now. 1330 */ 1331 if (tcp_reass_lock_try(tp) == 0) 1332 continue; 1333 if (tcp_freeq(tp)) 1334 tcpstat.tcps_connsdrained++; 1335 TCP_REASS_UNLOCK(tp); 1336 } 1337 } 1338 } 1339 1340 /* 1341 * Notify a tcp user of an asynchronous error; 1342 * store error as soft error, but wake up user 1343 * (for now, won't do anything until can select for soft error). 1344 */ 1345 void 1346 tcp_notify(struct inpcb *inp, int error) 1347 { 1348 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1349 struct socket *so = inp->inp_socket; 1350 1351 /* 1352 * Ignore some errors if we are hooked up. 1353 * If connection hasn't completed, has retransmitted several times, 1354 * and receives a second error, give up now. This is better 1355 * than waiting a long time to establish a connection that 1356 * can never complete. 1357 */ 1358 if (tp->t_state == TCPS_ESTABLISHED && 1359 (error == EHOSTUNREACH || error == ENETUNREACH || 1360 error == EHOSTDOWN)) { 1361 return; 1362 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1363 tp->t_rxtshift > 3 && tp->t_softerror) 1364 so->so_error = error; 1365 else 1366 tp->t_softerror = error; 1367 wakeup((void *) &so->so_timeo); 1368 sorwakeup(so); 1369 sowwakeup(so); 1370 } 1371 1372 #ifdef INET6 1373 void 1374 tcp6_notify(struct in6pcb *in6p, int error) 1375 { 1376 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1377 struct socket *so = in6p->in6p_socket; 1378 1379 /* 1380 * Ignore some errors if we are hooked up. 1381 * If connection hasn't completed, has retransmitted several times, 1382 * and receives a second error, give up now. This is better 1383 * than waiting a long time to establish a connection that 1384 * can never complete. 1385 */ 1386 if (tp->t_state == TCPS_ESTABLISHED && 1387 (error == EHOSTUNREACH || error == ENETUNREACH || 1388 error == EHOSTDOWN)) { 1389 return; 1390 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1391 tp->t_rxtshift > 3 && tp->t_softerror) 1392 so->so_error = error; 1393 else 1394 tp->t_softerror = error; 1395 wakeup((void *) &so->so_timeo); 1396 sorwakeup(so); 1397 sowwakeup(so); 1398 } 1399 #endif 1400 1401 #ifdef INET6 1402 void 1403 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1404 { 1405 struct tcphdr th; 1406 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1407 int nmatch; 1408 struct ip6_hdr *ip6; 1409 const struct sockaddr_in6 *sa6_src = NULL; 1410 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1411 struct mbuf *m; 1412 int off; 1413 1414 if (sa->sa_family != AF_INET6 || 1415 sa->sa_len != sizeof(struct sockaddr_in6)) 1416 return; 1417 if ((unsigned)cmd >= PRC_NCMDS) 1418 return; 1419 else if (cmd == PRC_QUENCH) { 1420 /* 1421 * Don't honor ICMP Source Quench messages meant for 1422 * TCP connections. 1423 */ 1424 return; 1425 } else if (PRC_IS_REDIRECT(cmd)) 1426 notify = in6_rtchange, d = NULL; 1427 else if (cmd == PRC_MSGSIZE) 1428 ; /* special code is present, see below */ 1429 else if (cmd == PRC_HOSTDEAD) 1430 d = NULL; 1431 else if (inet6ctlerrmap[cmd] == 0) 1432 return; 1433 1434 /* if the parameter is from icmp6, decode it. */ 1435 if (d != NULL) { 1436 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1437 m = ip6cp->ip6c_m; 1438 ip6 = ip6cp->ip6c_ip6; 1439 off = ip6cp->ip6c_off; 1440 sa6_src = ip6cp->ip6c_src; 1441 } else { 1442 m = NULL; 1443 ip6 = NULL; 1444 sa6_src = &sa6_any; 1445 off = 0; 1446 } 1447 1448 if (ip6) { 1449 /* 1450 * XXX: We assume that when ip6 is non NULL, 1451 * M and OFF are valid. 1452 */ 1453 1454 /* check if we can safely examine src and dst ports */ 1455 if (m->m_pkthdr.len < off + sizeof(th)) { 1456 if (cmd == PRC_MSGSIZE) 1457 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1458 return; 1459 } 1460 1461 bzero(&th, sizeof(th)); 1462 m_copydata(m, off, sizeof(th), (void *)&th); 1463 1464 if (cmd == PRC_MSGSIZE) { 1465 int valid = 0; 1466 1467 /* 1468 * Check to see if we have a valid TCP connection 1469 * corresponding to the address in the ICMPv6 message 1470 * payload. 1471 */ 1472 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1473 th.th_dport, 1474 (const struct in6_addr *)&sa6_src->sin6_addr, 1475 th.th_sport, 0)) 1476 valid++; 1477 1478 /* 1479 * Depending on the value of "valid" and routing table 1480 * size (mtudisc_{hi,lo}wat), we will: 1481 * - recalcurate the new MTU and create the 1482 * corresponding routing entry, or 1483 * - ignore the MTU change notification. 1484 */ 1485 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1486 1487 /* 1488 * no need to call in6_pcbnotify, it should have been 1489 * called via callback if necessary 1490 */ 1491 return; 1492 } 1493 1494 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1495 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1496 if (nmatch == 0 && syn_cache_count && 1497 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1498 inet6ctlerrmap[cmd] == ENETUNREACH || 1499 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1500 syn_cache_unreach((const struct sockaddr *)sa6_src, 1501 sa, &th); 1502 } else { 1503 (void) in6_pcbnotify(&tcbtable, sa, 0, 1504 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1505 } 1506 } 1507 #endif 1508 1509 #ifdef INET 1510 /* assumes that ip header and tcp header are contiguous on mbuf */ 1511 void * 1512 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1513 { 1514 struct ip *ip = v; 1515 struct tcphdr *th; 1516 struct icmp *icp; 1517 extern const int inetctlerrmap[]; 1518 void (*notify)(struct inpcb *, int) = tcp_notify; 1519 int errno; 1520 int nmatch; 1521 struct tcpcb *tp; 1522 u_int mtu; 1523 tcp_seq seq; 1524 struct inpcb *inp; 1525 #ifdef INET6 1526 struct in6pcb *in6p; 1527 struct in6_addr src6, dst6; 1528 #endif 1529 1530 if (sa->sa_family != AF_INET || 1531 sa->sa_len != sizeof(struct sockaddr_in)) 1532 return NULL; 1533 if ((unsigned)cmd >= PRC_NCMDS) 1534 return NULL; 1535 errno = inetctlerrmap[cmd]; 1536 if (cmd == PRC_QUENCH) 1537 /* 1538 * Don't honor ICMP Source Quench messages meant for 1539 * TCP connections. 1540 */ 1541 return NULL; 1542 else if (PRC_IS_REDIRECT(cmd)) 1543 notify = in_rtchange, ip = 0; 1544 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1545 /* 1546 * Check to see if we have a valid TCP connection 1547 * corresponding to the address in the ICMP message 1548 * payload. 1549 * 1550 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1551 */ 1552 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1553 #ifdef INET6 1554 memset(&src6, 0, sizeof(src6)); 1555 memset(&dst6, 0, sizeof(dst6)); 1556 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1557 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1558 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1559 #endif 1560 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1561 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1562 #ifdef INET6 1563 in6p = NULL; 1564 #else 1565 ; 1566 #endif 1567 #ifdef INET6 1568 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1569 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1570 ; 1571 #endif 1572 else 1573 return NULL; 1574 1575 /* 1576 * Now that we've validated that we are actually communicating 1577 * with the host indicated in the ICMP message, locate the 1578 * ICMP header, recalculate the new MTU, and create the 1579 * corresponding routing entry. 1580 */ 1581 icp = (struct icmp *)((char *)ip - 1582 offsetof(struct icmp, icmp_ip)); 1583 if (inp) { 1584 if ((tp = intotcpcb(inp)) == NULL) 1585 return NULL; 1586 } 1587 #ifdef INET6 1588 else if (in6p) { 1589 if ((tp = in6totcpcb(in6p)) == NULL) 1590 return NULL; 1591 } 1592 #endif 1593 else 1594 return NULL; 1595 seq = ntohl(th->th_seq); 1596 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1597 return NULL; 1598 /* 1599 * If the ICMP message advertises a Next-Hop MTU 1600 * equal or larger than the maximum packet size we have 1601 * ever sent, drop the message. 1602 */ 1603 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1604 if (mtu >= tp->t_pmtud_mtu_sent) 1605 return NULL; 1606 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1607 /* 1608 * Calculate new MTU, and create corresponding 1609 * route (traditional PMTUD). 1610 */ 1611 tp->t_flags &= ~TF_PMTUD_PEND; 1612 icmp_mtudisc(icp, ip->ip_dst); 1613 } else { 1614 /* 1615 * Record the information got in the ICMP 1616 * message; act on it later. 1617 * If we had already recorded an ICMP message, 1618 * replace the old one only if the new message 1619 * refers to an older TCP segment 1620 */ 1621 if (tp->t_flags & TF_PMTUD_PEND) { 1622 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1623 return NULL; 1624 } else 1625 tp->t_flags |= TF_PMTUD_PEND; 1626 tp->t_pmtud_th_seq = seq; 1627 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1628 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1629 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1630 } 1631 return NULL; 1632 } else if (cmd == PRC_HOSTDEAD) 1633 ip = 0; 1634 else if (errno == 0) 1635 return NULL; 1636 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1637 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1638 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1639 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1640 if (nmatch == 0 && syn_cache_count && 1641 (inetctlerrmap[cmd] == EHOSTUNREACH || 1642 inetctlerrmap[cmd] == ENETUNREACH || 1643 inetctlerrmap[cmd] == EHOSTDOWN)) { 1644 struct sockaddr_in sin; 1645 bzero(&sin, sizeof(sin)); 1646 sin.sin_len = sizeof(sin); 1647 sin.sin_family = AF_INET; 1648 sin.sin_port = th->th_sport; 1649 sin.sin_addr = ip->ip_src; 1650 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1651 } 1652 1653 /* XXX mapped address case */ 1654 } else 1655 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1656 notify); 1657 return NULL; 1658 } 1659 1660 /* 1661 * When a source quench is received, we are being notified of congestion. 1662 * Close the congestion window down to the Loss Window (one segment). 1663 * We will gradually open it again as we proceed. 1664 */ 1665 void 1666 tcp_quench(struct inpcb *inp, int errno) 1667 { 1668 struct tcpcb *tp = intotcpcb(inp); 1669 1670 if (tp) { 1671 tp->snd_cwnd = tp->t_segsz; 1672 tp->t_bytes_acked = 0; 1673 } 1674 } 1675 #endif 1676 1677 #ifdef INET6 1678 void 1679 tcp6_quench(struct in6pcb *in6p, int errno) 1680 { 1681 struct tcpcb *tp = in6totcpcb(in6p); 1682 1683 if (tp) { 1684 tp->snd_cwnd = tp->t_segsz; 1685 tp->t_bytes_acked = 0; 1686 } 1687 } 1688 #endif 1689 1690 #ifdef INET 1691 /* 1692 * Path MTU Discovery handlers. 1693 */ 1694 void 1695 tcp_mtudisc_callback(struct in_addr faddr) 1696 { 1697 #ifdef INET6 1698 struct in6_addr in6; 1699 #endif 1700 1701 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1702 #ifdef INET6 1703 memset(&in6, 0, sizeof(in6)); 1704 in6.s6_addr16[5] = 0xffff; 1705 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1706 tcp6_mtudisc_callback(&in6); 1707 #endif 1708 } 1709 1710 /* 1711 * On receipt of path MTU corrections, flush old route and replace it 1712 * with the new one. Retransmit all unacknowledged packets, to ensure 1713 * that all packets will be received. 1714 */ 1715 void 1716 tcp_mtudisc(struct inpcb *inp, int errno) 1717 { 1718 struct tcpcb *tp = intotcpcb(inp); 1719 struct rtentry *rt = in_pcbrtentry(inp); 1720 1721 if (tp != 0) { 1722 if (rt != 0) { 1723 /* 1724 * If this was not a host route, remove and realloc. 1725 */ 1726 if ((rt->rt_flags & RTF_HOST) == 0) { 1727 in_rtchange(inp, errno); 1728 if ((rt = in_pcbrtentry(inp)) == 0) 1729 return; 1730 } 1731 1732 /* 1733 * Slow start out of the error condition. We 1734 * use the MTU because we know it's smaller 1735 * than the previously transmitted segment. 1736 * 1737 * Note: This is more conservative than the 1738 * suggestion in draft-floyd-incr-init-win-03. 1739 */ 1740 if (rt->rt_rmx.rmx_mtu != 0) 1741 tp->snd_cwnd = 1742 TCP_INITIAL_WINDOW(tcp_init_win, 1743 rt->rt_rmx.rmx_mtu); 1744 } 1745 1746 /* 1747 * Resend unacknowledged packets. 1748 */ 1749 tp->snd_nxt = tp->snd_una; 1750 tcp_output(tp); 1751 } 1752 } 1753 #endif 1754 1755 #ifdef INET6 1756 /* 1757 * Path MTU Discovery handlers. 1758 */ 1759 void 1760 tcp6_mtudisc_callback(struct in6_addr *faddr) 1761 { 1762 struct sockaddr_in6 sin6; 1763 1764 bzero(&sin6, sizeof(sin6)); 1765 sin6.sin6_family = AF_INET6; 1766 sin6.sin6_len = sizeof(struct sockaddr_in6); 1767 sin6.sin6_addr = *faddr; 1768 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1769 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1770 } 1771 1772 void 1773 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1774 { 1775 struct tcpcb *tp = in6totcpcb(in6p); 1776 struct rtentry *rt = in6_pcbrtentry(in6p); 1777 1778 if (tp != 0) { 1779 if (rt != 0) { 1780 /* 1781 * If this was not a host route, remove and realloc. 1782 */ 1783 if ((rt->rt_flags & RTF_HOST) == 0) { 1784 in6_rtchange(in6p, errno); 1785 if ((rt = in6_pcbrtentry(in6p)) == 0) 1786 return; 1787 } 1788 1789 /* 1790 * Slow start out of the error condition. We 1791 * use the MTU because we know it's smaller 1792 * than the previously transmitted segment. 1793 * 1794 * Note: This is more conservative than the 1795 * suggestion in draft-floyd-incr-init-win-03. 1796 */ 1797 if (rt->rt_rmx.rmx_mtu != 0) 1798 tp->snd_cwnd = 1799 TCP_INITIAL_WINDOW(tcp_init_win, 1800 rt->rt_rmx.rmx_mtu); 1801 } 1802 1803 /* 1804 * Resend unacknowledged packets. 1805 */ 1806 tp->snd_nxt = tp->snd_una; 1807 tcp_output(tp); 1808 } 1809 } 1810 #endif /* INET6 */ 1811 1812 /* 1813 * Compute the MSS to advertise to the peer. Called only during 1814 * the 3-way handshake. If we are the server (peer initiated 1815 * connection), we are called with a pointer to the interface 1816 * on which the SYN packet arrived. If we are the client (we 1817 * initiated connection), we are called with a pointer to the 1818 * interface out which this connection should go. 1819 * 1820 * NOTE: Do not subtract IP option/extension header size nor IPsec 1821 * header size from MSS advertisement. MSS option must hold the maximum 1822 * segment size we can accept, so it must always be: 1823 * max(if mtu) - ip header - tcp header 1824 */ 1825 u_long 1826 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1827 { 1828 extern u_long in_maxmtu; 1829 u_long mss = 0; 1830 u_long hdrsiz; 1831 1832 /* 1833 * In order to avoid defeating path MTU discovery on the peer, 1834 * we advertise the max MTU of all attached networks as our MSS, 1835 * per RFC 1191, section 3.1. 1836 * 1837 * We provide the option to advertise just the MTU of 1838 * the interface on which we hope this connection will 1839 * be receiving. If we are responding to a SYN, we 1840 * will have a pretty good idea about this, but when 1841 * initiating a connection there is a bit more doubt. 1842 * 1843 * We also need to ensure that loopback has a large enough 1844 * MSS, as the loopback MTU is never included in in_maxmtu. 1845 */ 1846 1847 if (ifp != NULL) 1848 switch (af) { 1849 case AF_INET: 1850 mss = ifp->if_mtu; 1851 break; 1852 #ifdef INET6 1853 case AF_INET6: 1854 mss = IN6_LINKMTU(ifp); 1855 break; 1856 #endif 1857 } 1858 1859 if (tcp_mss_ifmtu == 0) 1860 switch (af) { 1861 case AF_INET: 1862 mss = max(in_maxmtu, mss); 1863 break; 1864 #ifdef INET6 1865 case AF_INET6: 1866 mss = max(in6_maxmtu, mss); 1867 break; 1868 #endif 1869 } 1870 1871 switch (af) { 1872 case AF_INET: 1873 hdrsiz = sizeof(struct ip); 1874 break; 1875 #ifdef INET6 1876 case AF_INET6: 1877 hdrsiz = sizeof(struct ip6_hdr); 1878 break; 1879 #endif 1880 default: 1881 hdrsiz = 0; 1882 break; 1883 } 1884 hdrsiz += sizeof(struct tcphdr); 1885 if (mss > hdrsiz) 1886 mss -= hdrsiz; 1887 1888 mss = max(tcp_mssdflt, mss); 1889 return (mss); 1890 } 1891 1892 /* 1893 * Set connection variables based on the peer's advertised MSS. 1894 * We are passed the TCPCB for the actual connection. If we 1895 * are the server, we are called by the compressed state engine 1896 * when the 3-way handshake is complete. If we are the client, 1897 * we are called when we receive the SYN,ACK from the server. 1898 * 1899 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1900 * before this routine is called! 1901 */ 1902 void 1903 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1904 { 1905 struct socket *so; 1906 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1907 struct rtentry *rt; 1908 #endif 1909 u_long bufsize; 1910 int mss; 1911 1912 #ifdef DIAGNOSTIC 1913 if (tp->t_inpcb && tp->t_in6pcb) 1914 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1915 #endif 1916 so = NULL; 1917 rt = NULL; 1918 #ifdef INET 1919 if (tp->t_inpcb) { 1920 so = tp->t_inpcb->inp_socket; 1921 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1922 rt = in_pcbrtentry(tp->t_inpcb); 1923 #endif 1924 } 1925 #endif 1926 #ifdef INET6 1927 if (tp->t_in6pcb) { 1928 so = tp->t_in6pcb->in6p_socket; 1929 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1930 rt = in6_pcbrtentry(tp->t_in6pcb); 1931 #endif 1932 } 1933 #endif 1934 1935 /* 1936 * As per RFC1122, use the default MSS value, unless they 1937 * sent us an offer. Do not accept offers less than 256 bytes. 1938 */ 1939 mss = tcp_mssdflt; 1940 if (offer) 1941 mss = offer; 1942 mss = max(mss, 256); /* sanity */ 1943 tp->t_peermss = mss; 1944 mss -= tcp_optlen(tp); 1945 #ifdef INET 1946 if (tp->t_inpcb) 1947 mss -= ip_optlen(tp->t_inpcb); 1948 #endif 1949 #ifdef INET6 1950 if (tp->t_in6pcb) 1951 mss -= ip6_optlen(tp->t_in6pcb); 1952 #endif 1953 1954 /* 1955 * If there's a pipesize, change the socket buffer to that size. 1956 * Make the socket buffer an integral number of MSS units. If 1957 * the MSS is larger than the socket buffer, artificially decrease 1958 * the MSS. 1959 */ 1960 #ifdef RTV_SPIPE 1961 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1962 bufsize = rt->rt_rmx.rmx_sendpipe; 1963 else 1964 #endif 1965 { 1966 KASSERT(so != NULL); 1967 bufsize = so->so_snd.sb_hiwat; 1968 } 1969 if (bufsize < mss) 1970 mss = bufsize; 1971 else { 1972 bufsize = roundup(bufsize, mss); 1973 if (bufsize > sb_max) 1974 bufsize = sb_max; 1975 (void) sbreserve(&so->so_snd, bufsize, so); 1976 } 1977 tp->t_segsz = mss; 1978 1979 #ifdef RTV_SSTHRESH 1980 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1981 /* 1982 * There's some sort of gateway or interface buffer 1983 * limit on the path. Use this to set the slow 1984 * start threshold, but set the threshold to no less 1985 * than 2 * MSS. 1986 */ 1987 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1988 } 1989 #endif 1990 } 1991 1992 /* 1993 * Processing necessary when a TCP connection is established. 1994 */ 1995 void 1996 tcp_established(struct tcpcb *tp) 1997 { 1998 struct socket *so; 1999 #ifdef RTV_RPIPE 2000 struct rtentry *rt; 2001 #endif 2002 u_long bufsize; 2003 2004 #ifdef DIAGNOSTIC 2005 if (tp->t_inpcb && tp->t_in6pcb) 2006 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2007 #endif 2008 so = NULL; 2009 rt = NULL; 2010 #ifdef INET 2011 if (tp->t_inpcb) { 2012 so = tp->t_inpcb->inp_socket; 2013 #if defined(RTV_RPIPE) 2014 rt = in_pcbrtentry(tp->t_inpcb); 2015 #endif 2016 } 2017 #endif 2018 #ifdef INET6 2019 if (tp->t_in6pcb) { 2020 so = tp->t_in6pcb->in6p_socket; 2021 #if defined(RTV_RPIPE) 2022 rt = in6_pcbrtentry(tp->t_in6pcb); 2023 #endif 2024 } 2025 #endif 2026 2027 tp->t_state = TCPS_ESTABLISHED; 2028 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2029 2030 #ifdef RTV_RPIPE 2031 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2032 bufsize = rt->rt_rmx.rmx_recvpipe; 2033 else 2034 #endif 2035 { 2036 KASSERT(so != NULL); 2037 bufsize = so->so_rcv.sb_hiwat; 2038 } 2039 if (bufsize > tp->t_ourmss) { 2040 bufsize = roundup(bufsize, tp->t_ourmss); 2041 if (bufsize > sb_max) 2042 bufsize = sb_max; 2043 (void) sbreserve(&so->so_rcv, bufsize, so); 2044 } 2045 } 2046 2047 /* 2048 * Check if there's an initial rtt or rttvar. Convert from the 2049 * route-table units to scaled multiples of the slow timeout timer. 2050 * Called only during the 3-way handshake. 2051 */ 2052 void 2053 tcp_rmx_rtt(struct tcpcb *tp) 2054 { 2055 #ifdef RTV_RTT 2056 struct rtentry *rt = NULL; 2057 int rtt; 2058 2059 #ifdef DIAGNOSTIC 2060 if (tp->t_inpcb && tp->t_in6pcb) 2061 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2062 #endif 2063 #ifdef INET 2064 if (tp->t_inpcb) 2065 rt = in_pcbrtentry(tp->t_inpcb); 2066 #endif 2067 #ifdef INET6 2068 if (tp->t_in6pcb) 2069 rt = in6_pcbrtentry(tp->t_in6pcb); 2070 #endif 2071 if (rt == NULL) 2072 return; 2073 2074 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2075 /* 2076 * XXX The lock bit for MTU indicates that the value 2077 * is also a minimum value; this is subject to time. 2078 */ 2079 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2080 TCPT_RANGESET(tp->t_rttmin, 2081 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2082 TCPTV_MIN, TCPTV_REXMTMAX); 2083 tp->t_srtt = rtt / 2084 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2085 if (rt->rt_rmx.rmx_rttvar) { 2086 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2087 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2088 (TCP_RTTVAR_SHIFT + 2)); 2089 } else { 2090 /* Default variation is +- 1 rtt */ 2091 tp->t_rttvar = 2092 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2093 } 2094 TCPT_RANGESET(tp->t_rxtcur, 2095 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2096 tp->t_rttmin, TCPTV_REXMTMAX); 2097 } 2098 #endif 2099 } 2100 2101 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2102 #if NRND > 0 2103 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2104 #endif 2105 2106 /* 2107 * Get a new sequence value given a tcp control block 2108 */ 2109 tcp_seq 2110 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2111 { 2112 2113 #ifdef INET 2114 if (tp->t_inpcb != NULL) { 2115 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2116 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2117 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2118 addin)); 2119 } 2120 #endif 2121 #ifdef INET6 2122 if (tp->t_in6pcb != NULL) { 2123 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2124 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2125 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2126 addin)); 2127 } 2128 #endif 2129 /* Not possible. */ 2130 panic("tcp_new_iss"); 2131 } 2132 2133 /* 2134 * This routine actually generates a new TCP initial sequence number. 2135 */ 2136 tcp_seq 2137 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2138 size_t addrsz, tcp_seq addin) 2139 { 2140 tcp_seq tcp_iss; 2141 2142 #if NRND > 0 2143 static int beenhere; 2144 2145 /* 2146 * If we haven't been here before, initialize our cryptographic 2147 * hash secret. 2148 */ 2149 if (beenhere == 0) { 2150 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2151 RND_EXTRACT_ANY); 2152 beenhere = 1; 2153 } 2154 2155 if (tcp_do_rfc1948) { 2156 MD5_CTX ctx; 2157 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2158 2159 /* 2160 * Compute the base value of the ISS. It is a hash 2161 * of (saddr, sport, daddr, dport, secret). 2162 */ 2163 MD5Init(&ctx); 2164 2165 MD5Update(&ctx, (u_char *) laddr, addrsz); 2166 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2167 2168 MD5Update(&ctx, (u_char *) faddr, addrsz); 2169 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2170 2171 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2172 2173 MD5Final(hash, &ctx); 2174 2175 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2176 2177 /* 2178 * Now increment our "timer", and add it in to 2179 * the computed value. 2180 * 2181 * XXX Use `addin'? 2182 * XXX TCP_ISSINCR too large to use? 2183 */ 2184 tcp_iss_seq += TCP_ISSINCR; 2185 #ifdef TCPISS_DEBUG 2186 printf("ISS hash 0x%08x, ", tcp_iss); 2187 #endif 2188 tcp_iss += tcp_iss_seq + addin; 2189 #ifdef TCPISS_DEBUG 2190 printf("new ISS 0x%08x\n", tcp_iss); 2191 #endif 2192 } else 2193 #endif /* NRND > 0 */ 2194 { 2195 /* 2196 * Randomize. 2197 */ 2198 #if NRND > 0 2199 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2200 #else 2201 tcp_iss = arc4random(); 2202 #endif 2203 2204 /* 2205 * If we were asked to add some amount to a known value, 2206 * we will take a random value obtained above, mask off 2207 * the upper bits, and add in the known value. We also 2208 * add in a constant to ensure that we are at least a 2209 * certain distance from the original value. 2210 * 2211 * This is used when an old connection is in timed wait 2212 * and we have a new one coming in, for instance. 2213 */ 2214 if (addin != 0) { 2215 #ifdef TCPISS_DEBUG 2216 printf("Random %08x, ", tcp_iss); 2217 #endif 2218 tcp_iss &= TCP_ISS_RANDOM_MASK; 2219 tcp_iss += addin + TCP_ISSINCR; 2220 #ifdef TCPISS_DEBUG 2221 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2222 #endif 2223 } else { 2224 tcp_iss &= TCP_ISS_RANDOM_MASK; 2225 tcp_iss += tcp_iss_seq; 2226 tcp_iss_seq += TCP_ISSINCR; 2227 #ifdef TCPISS_DEBUG 2228 printf("ISS %08x\n", tcp_iss); 2229 #endif 2230 } 2231 } 2232 2233 if (tcp_compat_42) { 2234 /* 2235 * Limit it to the positive range for really old TCP 2236 * implementations. 2237 * Just AND off the top bit instead of checking if 2238 * is set first - saves a branch 50% of the time. 2239 */ 2240 tcp_iss &= 0x7fffffff; /* XXX */ 2241 } 2242 2243 return (tcp_iss); 2244 } 2245 2246 #if defined(IPSEC) || defined(FAST_IPSEC) 2247 /* compute ESP/AH header size for TCP, including outer IP header. */ 2248 size_t 2249 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2250 { 2251 struct inpcb *inp; 2252 size_t hdrsiz; 2253 2254 /* XXX mapped addr case (tp->t_in6pcb) */ 2255 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2256 return 0; 2257 switch (tp->t_family) { 2258 case AF_INET: 2259 /* XXX: should use currect direction. */ 2260 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2261 break; 2262 default: 2263 hdrsiz = 0; 2264 break; 2265 } 2266 2267 return hdrsiz; 2268 } 2269 2270 #ifdef INET6 2271 size_t 2272 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2273 { 2274 struct in6pcb *in6p; 2275 size_t hdrsiz; 2276 2277 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2278 return 0; 2279 switch (tp->t_family) { 2280 case AF_INET6: 2281 /* XXX: should use currect direction. */ 2282 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2283 break; 2284 case AF_INET: 2285 /* mapped address case - tricky */ 2286 default: 2287 hdrsiz = 0; 2288 break; 2289 } 2290 2291 return hdrsiz; 2292 } 2293 #endif 2294 #endif /*IPSEC*/ 2295 2296 /* 2297 * Determine the length of the TCP options for this connection. 2298 * 2299 * XXX: What do we do for SACK, when we add that? Just reserve 2300 * all of the space? Otherwise we can't exactly be incrementing 2301 * cwnd by an amount that varies depending on the amount we last 2302 * had to SACK! 2303 */ 2304 2305 u_int 2306 tcp_optlen(struct tcpcb *tp) 2307 { 2308 u_int optlen; 2309 2310 optlen = 0; 2311 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2312 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2313 optlen += TCPOLEN_TSTAMP_APPA; 2314 2315 #ifdef TCP_SIGNATURE 2316 if (tp->t_flags & TF_SIGNATURE) 2317 optlen += TCPOLEN_SIGNATURE + 2; 2318 #endif /* TCP_SIGNATURE */ 2319 2320 return optlen; 2321 } 2322 2323 u_int 2324 tcp_hdrsz(struct tcpcb *tp) 2325 { 2326 u_int hlen; 2327 2328 switch (tp->t_family) { 2329 #ifdef INET6 2330 case AF_INET6: 2331 hlen = sizeof(struct ip6_hdr); 2332 break; 2333 #endif 2334 case AF_INET: 2335 hlen = sizeof(struct ip); 2336 break; 2337 default: 2338 hlen = 0; 2339 break; 2340 } 2341 hlen += sizeof(struct tcphdr); 2342 2343 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2344 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2345 hlen += TCPOLEN_TSTAMP_APPA; 2346 #ifdef TCP_SIGNATURE 2347 if (tp->t_flags & TF_SIGNATURE) 2348 hlen += TCPOLEN_SIGLEN; 2349 #endif 2350 return hlen; 2351 } 2352
Indexes created Mon Sep 22 13:09:51 GMT 2025