1 /* $NetBSD: tcp_subr.c,v 1.236 2009/03/18 16:00:22 cegger 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, 2008 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 * 49 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 50 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 51 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 52 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 53 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 54 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 55 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 56 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 57 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 58 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 59 * POSSIBILITY OF SUCH DAMAGE. 60 */ 61 62 /* 63 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 64 * The Regents of the University of California. All rights reserved. 65 * 66 * Redistribution and use in source and binary forms, with or without 67 * modification, are permitted provided that the following conditions 68 * are met: 69 * 1. Redistributions of source code must retain the above copyright 70 * notice, this list of conditions and the following disclaimer. 71 * 2. Redistributions in binary form must reproduce the above copyright 72 * notice, this list of conditions and the following disclaimer in the 73 * documentation and/or other materials provided with the distribution. 74 * 3. Neither the name of the University nor the names of its contributors 75 * may be used to endorse or promote products derived from this software 76 * without specific prior written permission. 77 * 78 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 79 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 80 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 81 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 82 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 83 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 84 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 85 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 86 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 87 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 88 * SUCH DAMAGE. 89 * 90 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 91 */ 92 93 #include <sys/cdefs.h> 94 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.236 2009/03/18 16:00:22 cegger Exp $"); 95 96 #include "opt_inet.h" 97 #include "opt_ipsec.h" 98 #include "opt_tcp_compat_42.h" 99 #include "opt_inet_csum.h" 100 #include "opt_mbuftrace.h" 101 #include "rnd.h" 102 103 #include <sys/param.h> 104 #include <sys/proc.h> 105 #include <sys/systm.h> 106 #include <sys/malloc.h> 107 #include <sys/mbuf.h> 108 #include <sys/socket.h> 109 #include <sys/socketvar.h> 110 #include <sys/protosw.h> 111 #include <sys/errno.h> 112 #include <sys/kernel.h> 113 #include <sys/pool.h> 114 #if NRND > 0 115 #include <sys/md5.h> 116 #include <sys/rnd.h> 117 #endif 118 119 #include <net/route.h> 120 #include <net/if.h> 121 122 #include <netinet/in.h> 123 #include <netinet/in_systm.h> 124 #include <netinet/ip.h> 125 #include <netinet/in_pcb.h> 126 #include <netinet/ip_var.h> 127 #include <netinet/ip_icmp.h> 128 129 #ifdef INET6 130 #ifndef INET 131 #include <netinet/in.h> 132 #endif 133 #include <netinet/ip6.h> 134 #include <netinet6/in6_pcb.h> 135 #include <netinet6/ip6_var.h> 136 #include <netinet6/in6_var.h> 137 #include <netinet6/ip6protosw.h> 138 #include <netinet/icmp6.h> 139 #include <netinet6/nd6.h> 140 #endif 141 142 #include <netinet/tcp.h> 143 #include <netinet/tcp_fsm.h> 144 #include <netinet/tcp_seq.h> 145 #include <netinet/tcp_timer.h> 146 #include <netinet/tcp_var.h> 147 #include <netinet/tcp_private.h> 148 #include <netinet/tcp_congctl.h> 149 #include <netinet/tcpip.h> 150 151 #ifdef IPSEC 152 #include <netinet6/ipsec.h> 153 #include <netkey/key.h> 154 #endif /*IPSEC*/ 155 156 #ifdef FAST_IPSEC 157 #include <netipsec/ipsec.h> 158 #include <netipsec/xform.h> 159 #ifdef INET6 160 #include <netipsec/ipsec6.h> 161 #endif 162 #include <netipsec/key.h> 163 #endif /* FAST_IPSEC*/ 164 165 166 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 167 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 168 169 percpu_t *tcpstat_percpu; 170 171 /* patchable/settable parameters for tcp */ 172 int tcp_mssdflt = TCP_MSS; 173 int tcp_minmss = TCP_MINMSS; 174 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 175 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 176 #if NRND > 0 177 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 178 #endif 179 int tcp_do_sack = 1; /* selective acknowledgement */ 180 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 181 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 182 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 183 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 184 #ifndef TCP_INIT_WIN 185 #define TCP_INIT_WIN 0 /* initial slow start window */ 186 #endif 187 #ifndef TCP_INIT_WIN_LOCAL 188 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 189 #endif 190 int tcp_init_win = TCP_INIT_WIN; 191 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 192 int tcp_mss_ifmtu = 0; 193 #ifdef TCP_COMPAT_42 194 int tcp_compat_42 = 1; 195 #else 196 int tcp_compat_42 = 0; 197 #endif 198 int tcp_rst_ppslim = 100; /* 100pps */ 199 int tcp_ackdrop_ppslim = 100; /* 100pps */ 200 int tcp_do_loopback_cksum = 0; 201 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 202 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 203 int tcp_sack_tp_maxholes = 32; 204 int tcp_sack_globalmaxholes = 1024; 205 int tcp_sack_globalholes = 0; 206 int tcp_ecn_maxretries = 1; 207 208 /* tcb hash */ 209 #ifndef TCBHASHSIZE 210 #define TCBHASHSIZE 128 211 #endif 212 int tcbhashsize = TCBHASHSIZE; 213 214 /* syn hash parameters */ 215 #define TCP_SYN_HASH_SIZE 293 216 #define TCP_SYN_BUCKET_SIZE 35 217 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 218 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 219 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 220 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 221 222 int tcp_freeq(struct tcpcb *); 223 224 #ifdef INET 225 void tcp_mtudisc_callback(struct in_addr); 226 #endif 227 #ifdef INET6 228 void tcp6_mtudisc_callback(struct in6_addr *); 229 #endif 230 231 #ifdef INET6 232 void tcp6_mtudisc(struct in6pcb *, int); 233 #endif 234 235 static struct pool tcpcb_pool; 236 237 #ifdef TCP_CSUM_COUNTERS 238 #include <sys/device.h> 239 240 #if defined(INET) 241 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "hwcsum bad"); 243 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "hwcsum ok"); 245 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 246 NULL, "tcp", "hwcsum data"); 247 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "swcsum"); 249 250 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 251 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 252 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 253 EVCNT_ATTACH_STATIC(tcp_swcsum); 254 #endif /* defined(INET) */ 255 256 #if defined(INET6) 257 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 258 NULL, "tcp6", "hwcsum bad"); 259 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp6", "hwcsum ok"); 261 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 NULL, "tcp6", "hwcsum data"); 263 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "swcsum"); 265 266 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 267 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 268 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 269 EVCNT_ATTACH_STATIC(tcp6_swcsum); 270 #endif /* defined(INET6) */ 271 #endif /* TCP_CSUM_COUNTERS */ 272 273 274 #ifdef TCP_OUTPUT_COUNTERS 275 #include <sys/device.h> 276 277 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 278 NULL, "tcp", "output big header"); 279 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 NULL, "tcp", "output predict hit"); 281 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 NULL, "tcp", "output predict miss"); 283 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output copy small"); 285 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output copy big"); 287 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output reference big"); 289 290 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 291 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 292 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 293 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 294 EVCNT_ATTACH_STATIC(tcp_output_copybig); 295 EVCNT_ATTACH_STATIC(tcp_output_refbig); 296 297 #endif /* TCP_OUTPUT_COUNTERS */ 298 299 #ifdef TCP_REASS_COUNTERS 300 #include <sys/device.h> 301 302 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 303 NULL, "tcp_reass", "calls"); 304 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 305 &tcp_reass_, "tcp_reass", "insert into empty queue"); 306 struct evcnt tcp_reass_iteration[8] = { 307 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 308 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 315 }; 316 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 317 &tcp_reass_, "tcp_reass", "prepend to first"); 318 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 319 &tcp_reass_, "tcp_reass", "prepend"); 320 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 321 &tcp_reass_, "tcp_reass", "insert"); 322 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "insert at tail"); 324 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "append"); 326 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "append to tail fragment"); 328 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "overlap at end"); 330 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "overlap at start"); 332 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "duplicate segment"); 334 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "duplicate fragment"); 336 337 EVCNT_ATTACH_STATIC(tcp_reass_); 338 EVCNT_ATTACH_STATIC(tcp_reass_empty); 339 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 340 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 341 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 342 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 347 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 348 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 349 EVCNT_ATTACH_STATIC(tcp_reass_insert); 350 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 351 EVCNT_ATTACH_STATIC(tcp_reass_append); 352 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 353 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 354 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 355 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 356 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 357 358 #endif /* TCP_REASS_COUNTERS */ 359 360 #ifdef MBUFTRACE 361 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 362 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 363 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 364 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 365 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 366 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 367 #endif 368 369 /* 370 * Tcp initialization 371 */ 372 void 373 tcp_init(void) 374 { 375 int hlen; 376 377 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 378 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 379 NULL, IPL_SOFTNET); 380 381 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 382 #ifdef INET6 383 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 384 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 385 #endif 386 if (max_protohdr < hlen) 387 max_protohdr = hlen; 388 if (max_linkhdr + hlen > MHLEN) 389 panic("tcp_init"); 390 391 #ifdef INET 392 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 393 #endif 394 #ifdef INET6 395 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 396 #endif 397 398 /* Initialize timer state. */ 399 tcp_timer_init(); 400 401 /* Initialize the compressed state engine. */ 402 syn_cache_init(); 403 404 /* Initialize the congestion control algorithms. */ 405 tcp_congctl_init(); 406 407 /* Initialize the TCPCB template. */ 408 tcp_tcpcb_template(); 409 410 /* Initialize reassembly queue */ 411 tcpipqent_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 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS); 422 } 423 424 /* 425 * Create template to be used to send tcp packets on a connection. 426 * Call after host entry created, allocates an mbuf and fills 427 * in a skeletal tcp/ip header, minimizing the amount of work 428 * necessary when the connection is used. 429 */ 430 struct mbuf * 431 tcp_template(struct tcpcb *tp) 432 { 433 struct inpcb *inp = tp->t_inpcb; 434 #ifdef INET6 435 struct in6pcb *in6p = tp->t_in6pcb; 436 #endif 437 struct tcphdr *n; 438 struct mbuf *m; 439 int hlen; 440 441 switch (tp->t_family) { 442 case AF_INET: 443 hlen = sizeof(struct ip); 444 if (inp) 445 break; 446 #ifdef INET6 447 if (in6p) { 448 /* mapped addr case */ 449 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 450 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 451 break; 452 } 453 #endif 454 return NULL; /*EINVAL*/ 455 #ifdef INET6 456 case AF_INET6: 457 hlen = sizeof(struct ip6_hdr); 458 if (in6p) { 459 /* more sainty check? */ 460 break; 461 } 462 return NULL; /*EINVAL*/ 463 #endif 464 default: 465 hlen = 0; /*pacify gcc*/ 466 return NULL; /*EAFNOSUPPORT*/ 467 } 468 #ifdef DIAGNOSTIC 469 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 470 panic("mclbytes too small for t_template"); 471 #endif 472 m = tp->t_template; 473 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 474 ; 475 else { 476 if (m) 477 m_freem(m); 478 m = tp->t_template = NULL; 479 MGETHDR(m, M_DONTWAIT, MT_HEADER); 480 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 481 MCLGET(m, M_DONTWAIT); 482 if ((m->m_flags & M_EXT) == 0) { 483 m_free(m); 484 m = NULL; 485 } 486 } 487 if (m == NULL) 488 return NULL; 489 MCLAIM(m, &tcp_mowner); 490 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 491 } 492 493 memset(mtod(m, void *), 0, m->m_len); 494 495 n = (struct tcphdr *)(mtod(m, char *) + hlen); 496 497 switch (tp->t_family) { 498 case AF_INET: 499 { 500 struct ipovly *ipov; 501 mtod(m, struct ip *)->ip_v = 4; 502 mtod(m, struct ip *)->ip_hl = hlen >> 2; 503 ipov = mtod(m, struct ipovly *); 504 ipov->ih_pr = IPPROTO_TCP; 505 ipov->ih_len = htons(sizeof(struct tcphdr)); 506 if (inp) { 507 ipov->ih_src = inp->inp_laddr; 508 ipov->ih_dst = inp->inp_faddr; 509 } 510 #ifdef INET6 511 else if (in6p) { 512 /* mapped addr case */ 513 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 514 sizeof(ipov->ih_src)); 515 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 516 sizeof(ipov->ih_dst)); 517 } 518 #endif 519 /* 520 * Compute the pseudo-header portion of the checksum 521 * now. We incrementally add in the TCP option and 522 * payload lengths later, and then compute the TCP 523 * checksum right before the packet is sent off onto 524 * the wire. 525 */ 526 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 527 ipov->ih_dst.s_addr, 528 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 529 break; 530 } 531 #ifdef INET6 532 case AF_INET6: 533 { 534 struct ip6_hdr *ip6; 535 mtod(m, struct ip *)->ip_v = 6; 536 ip6 = mtod(m, struct ip6_hdr *); 537 ip6->ip6_nxt = IPPROTO_TCP; 538 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 539 ip6->ip6_src = in6p->in6p_laddr; 540 ip6->ip6_dst = in6p->in6p_faddr; 541 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 542 if (ip6_auto_flowlabel) { 543 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 544 ip6->ip6_flow |= 545 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 546 } 547 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 548 ip6->ip6_vfc |= IPV6_VERSION; 549 550 /* 551 * Compute the pseudo-header portion of the checksum 552 * now. We incrementally add in the TCP option and 553 * payload lengths later, and then compute the TCP 554 * checksum right before the packet is sent off onto 555 * the wire. 556 */ 557 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 558 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 559 htonl(IPPROTO_TCP)); 560 break; 561 } 562 #endif 563 } 564 if (inp) { 565 n->th_sport = inp->inp_lport; 566 n->th_dport = inp->inp_fport; 567 } 568 #ifdef INET6 569 else if (in6p) { 570 n->th_sport = in6p->in6p_lport; 571 n->th_dport = in6p->in6p_fport; 572 } 573 #endif 574 n->th_seq = 0; 575 n->th_ack = 0; 576 n->th_x2 = 0; 577 n->th_off = 5; 578 n->th_flags = 0; 579 n->th_win = 0; 580 n->th_urp = 0; 581 return (m); 582 } 583 584 /* 585 * Send a single message to the TCP at address specified by 586 * the given TCP/IP header. If m == 0, then we make a copy 587 * of the tcpiphdr at ti and send directly to the addressed host. 588 * This is used to force keep alive messages out using the TCP 589 * template for a connection tp->t_template. If flags are given 590 * then we send a message back to the TCP which originated the 591 * segment ti, and discard the mbuf containing it and any other 592 * attached mbufs. 593 * 594 * In any case the ack and sequence number of the transmitted 595 * segment are as specified by the parameters. 596 */ 597 int 598 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 599 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 600 { 601 #ifdef INET6 602 struct rtentry *rt; 603 #endif 604 struct route *ro; 605 int error, tlen, win = 0; 606 int hlen; 607 struct ip *ip; 608 #ifdef INET6 609 struct ip6_hdr *ip6; 610 #endif 611 int family; /* family on packet, not inpcb/in6pcb! */ 612 struct tcphdr *th; 613 struct socket *so; 614 615 if (tp != NULL && (flags & TH_RST) == 0) { 616 #ifdef DIAGNOSTIC 617 if (tp->t_inpcb && tp->t_in6pcb) 618 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 619 #endif 620 #ifdef INET 621 if (tp->t_inpcb) 622 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 623 #endif 624 #ifdef INET6 625 if (tp->t_in6pcb) 626 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 627 #endif 628 } 629 630 th = NULL; /* Quell uninitialized warning */ 631 ip = NULL; 632 #ifdef INET6 633 ip6 = NULL; 634 #endif 635 if (m == 0) { 636 if (!template) 637 return EINVAL; 638 639 /* get family information from template */ 640 switch (mtod(template, struct ip *)->ip_v) { 641 case 4: 642 family = AF_INET; 643 hlen = sizeof(struct ip); 644 break; 645 #ifdef INET6 646 case 6: 647 family = AF_INET6; 648 hlen = sizeof(struct ip6_hdr); 649 break; 650 #endif 651 default: 652 return EAFNOSUPPORT; 653 } 654 655 MGETHDR(m, M_DONTWAIT, MT_HEADER); 656 if (m) { 657 MCLAIM(m, &tcp_tx_mowner); 658 MCLGET(m, M_DONTWAIT); 659 if ((m->m_flags & M_EXT) == 0) { 660 m_free(m); 661 m = NULL; 662 } 663 } 664 if (m == NULL) 665 return (ENOBUFS); 666 667 if (tcp_compat_42) 668 tlen = 1; 669 else 670 tlen = 0; 671 672 m->m_data += max_linkhdr; 673 bcopy(mtod(template, void *), mtod(m, void *), 674 template->m_len); 675 switch (family) { 676 case AF_INET: 677 ip = mtod(m, struct ip *); 678 th = (struct tcphdr *)(ip + 1); 679 break; 680 #ifdef INET6 681 case AF_INET6: 682 ip6 = mtod(m, struct ip6_hdr *); 683 th = (struct tcphdr *)(ip6 + 1); 684 break; 685 #endif 686 #if 0 687 default: 688 /* noone will visit here */ 689 m_freem(m); 690 return EAFNOSUPPORT; 691 #endif 692 } 693 flags = TH_ACK; 694 } else { 695 696 if ((m->m_flags & M_PKTHDR) == 0) { 697 #if 0 698 printf("non PKTHDR to tcp_respond\n"); 699 #endif 700 m_freem(m); 701 return EINVAL; 702 } 703 #ifdef DIAGNOSTIC 704 if (!th0) 705 panic("th0 == NULL in tcp_respond"); 706 #endif 707 708 /* get family information from m */ 709 switch (mtod(m, struct ip *)->ip_v) { 710 case 4: 711 family = AF_INET; 712 hlen = sizeof(struct ip); 713 ip = mtod(m, struct ip *); 714 break; 715 #ifdef INET6 716 case 6: 717 family = AF_INET6; 718 hlen = sizeof(struct ip6_hdr); 719 ip6 = mtod(m, struct ip6_hdr *); 720 break; 721 #endif 722 default: 723 m_freem(m); 724 return EAFNOSUPPORT; 725 } 726 /* clear h/w csum flags inherited from rx packet */ 727 m->m_pkthdr.csum_flags = 0; 728 729 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 730 tlen = sizeof(*th0); 731 else 732 tlen = th0->th_off << 2; 733 734 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 735 mtod(m, char *) + hlen == (char *)th0) { 736 m->m_len = hlen + tlen; 737 m_freem(m->m_next); 738 m->m_next = NULL; 739 } else { 740 struct mbuf *n; 741 742 #ifdef DIAGNOSTIC 743 if (max_linkhdr + hlen + tlen > MCLBYTES) { 744 m_freem(m); 745 return EMSGSIZE; 746 } 747 #endif 748 MGETHDR(n, M_DONTWAIT, MT_HEADER); 749 if (n && max_linkhdr + hlen + tlen > MHLEN) { 750 MCLGET(n, M_DONTWAIT); 751 if ((n->m_flags & M_EXT) == 0) { 752 m_freem(n); 753 n = NULL; 754 } 755 } 756 if (!n) { 757 m_freem(m); 758 return ENOBUFS; 759 } 760 761 MCLAIM(n, &tcp_tx_mowner); 762 n->m_data += max_linkhdr; 763 n->m_len = hlen + tlen; 764 m_copyback(n, 0, hlen, mtod(m, void *)); 765 m_copyback(n, hlen, tlen, (void *)th0); 766 767 m_freem(m); 768 m = n; 769 n = NULL; 770 } 771 772 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 773 switch (family) { 774 case AF_INET: 775 ip = mtod(m, struct ip *); 776 th = (struct tcphdr *)(ip + 1); 777 ip->ip_p = IPPROTO_TCP; 778 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 779 ip->ip_p = IPPROTO_TCP; 780 break; 781 #ifdef INET6 782 case AF_INET6: 783 ip6 = mtod(m, struct ip6_hdr *); 784 th = (struct tcphdr *)(ip6 + 1); 785 ip6->ip6_nxt = IPPROTO_TCP; 786 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 787 ip6->ip6_nxt = IPPROTO_TCP; 788 break; 789 #endif 790 #if 0 791 default: 792 /* noone will visit here */ 793 m_freem(m); 794 return EAFNOSUPPORT; 795 #endif 796 } 797 xchg(th->th_dport, th->th_sport, u_int16_t); 798 #undef xchg 799 tlen = 0; /*be friendly with the following code*/ 800 } 801 th->th_seq = htonl(seq); 802 th->th_ack = htonl(ack); 803 th->th_x2 = 0; 804 if ((flags & TH_SYN) == 0) { 805 if (tp) 806 win >>= tp->rcv_scale; 807 if (win > TCP_MAXWIN) 808 win = TCP_MAXWIN; 809 th->th_win = htons((u_int16_t)win); 810 th->th_off = sizeof (struct tcphdr) >> 2; 811 tlen += sizeof(*th); 812 } else 813 tlen += th->th_off << 2; 814 m->m_len = hlen + tlen; 815 m->m_pkthdr.len = hlen + tlen; 816 m->m_pkthdr.rcvif = (struct ifnet *) 0; 817 th->th_flags = flags; 818 th->th_urp = 0; 819 820 switch (family) { 821 #ifdef INET 822 case AF_INET: 823 { 824 struct ipovly *ipov = (struct ipovly *)ip; 825 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1); 826 ipov->ih_len = htons((u_int16_t)tlen); 827 828 th->th_sum = 0; 829 th->th_sum = in_cksum(m, hlen + tlen); 830 ip->ip_len = htons(hlen + tlen); 831 ip->ip_ttl = ip_defttl; 832 break; 833 } 834 #endif 835 #ifdef INET6 836 case AF_INET6: 837 { 838 th->th_sum = 0; 839 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 840 tlen); 841 ip6->ip6_plen = htons(tlen); 842 if (tp && tp->t_in6pcb) { 843 struct ifnet *oifp; 844 ro = &tp->t_in6pcb->in6p_route; 845 oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 846 : NULL; 847 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 848 } else 849 ip6->ip6_hlim = ip6_defhlim; 850 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 851 if (ip6_auto_flowlabel) { 852 ip6->ip6_flow |= 853 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 854 } 855 break; 856 } 857 #endif 858 } 859 860 if (tp && tp->t_inpcb) 861 so = tp->t_inpcb->inp_socket; 862 #ifdef INET6 863 else if (tp && tp->t_in6pcb) 864 so = tp->t_in6pcb->in6p_socket; 865 #endif 866 else 867 so = NULL; 868 869 if (tp != NULL && tp->t_inpcb != NULL) { 870 ro = &tp->t_inpcb->inp_route; 871 #ifdef DIAGNOSTIC 872 if (family != AF_INET) 873 panic("tcp_respond: address family mismatch"); 874 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 875 panic("tcp_respond: ip_dst %x != inp_faddr %x", 876 ntohl(ip->ip_dst.s_addr), 877 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 878 } 879 #endif 880 } 881 #ifdef INET6 882 else if (tp != NULL && tp->t_in6pcb != NULL) { 883 ro = (struct route *)&tp->t_in6pcb->in6p_route; 884 #ifdef DIAGNOSTIC 885 if (family == AF_INET) { 886 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 887 panic("tcp_respond: not mapped addr"); 888 if (memcmp(&ip->ip_dst, 889 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 890 sizeof(ip->ip_dst)) != 0) { 891 panic("tcp_respond: ip_dst != in6p_faddr"); 892 } 893 } else if (family == AF_INET6) { 894 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 895 &tp->t_in6pcb->in6p_faddr)) 896 panic("tcp_respond: ip6_dst != in6p_faddr"); 897 } else 898 panic("tcp_respond: address family mismatch"); 899 #endif 900 } 901 #endif 902 else 903 ro = NULL; 904 905 switch (family) { 906 #ifdef INET 907 case AF_INET: 908 error = ip_output(m, NULL, ro, 909 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 910 (struct ip_moptions *)0, so); 911 break; 912 #endif 913 #ifdef INET6 914 case AF_INET6: 915 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL); 916 break; 917 #endif 918 default: 919 error = EAFNOSUPPORT; 920 break; 921 } 922 923 return (error); 924 } 925 926 /* 927 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 928 * a bunch of members individually, we maintain this template for the 929 * static and mostly-static components of the TCPCB, and copy it into 930 * the new TCPCB instead. 931 */ 932 static struct tcpcb tcpcb_template = { 933 .t_srtt = TCPTV_SRTTBASE, 934 .t_rttmin = TCPTV_MIN, 935 936 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 937 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 938 .snd_numholes = 0, 939 940 .t_partialacks = -1, 941 .t_bytes_acked = 0, 942 }; 943 944 /* 945 * Updates the TCPCB template whenever a parameter that would affect 946 * the template is changed. 947 */ 948 void 949 tcp_tcpcb_template(void) 950 { 951 struct tcpcb *tp = &tcpcb_template; 952 int flags; 953 954 tp->t_peermss = tcp_mssdflt; 955 tp->t_ourmss = tcp_mssdflt; 956 tp->t_segsz = tcp_mssdflt; 957 958 flags = 0; 959 if (tcp_do_rfc1323 && tcp_do_win_scale) 960 flags |= TF_REQ_SCALE; 961 if (tcp_do_rfc1323 && tcp_do_timestamps) 962 flags |= TF_REQ_TSTMP; 963 tp->t_flags = flags; 964 965 /* 966 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 967 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 968 * reasonable initial retransmit time. 969 */ 970 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 971 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 972 TCPTV_MIN, TCPTV_REXMTMAX); 973 974 /* Keep Alive */ 975 tp->t_keepinit = tcp_keepinit; 976 tp->t_keepidle = tcp_keepidle; 977 tp->t_keepintvl = tcp_keepintvl; 978 tp->t_keepcnt = tcp_keepcnt; 979 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 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 #ifdef INET6 992 struct rtentry *rt; 993 #endif 994 struct tcpcb *tp; 995 int i; 996 997 /* XXX Consider using a pool_cache for speed. */ 998 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 999 if (tp == NULL) 1000 return (NULL); 1001 memcpy(tp, &tcpcb_template, sizeof(*tp)); 1002 TAILQ_INIT(&tp->segq); 1003 TAILQ_INIT(&tp->timeq); 1004 tp->t_family = family; /* may be overridden later on */ 1005 TAILQ_INIT(&tp->snd_holes); 1006 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1007 1008 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1009 for (i = 0; i < TCPT_NTIMERS; i++) { 1010 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 1011 TCP_TIMER_INIT(tp, i); 1012 } 1013 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 1014 1015 switch (family) { 1016 case AF_INET: 1017 { 1018 struct inpcb *inp = (struct inpcb *)aux; 1019 1020 inp->inp_ip.ip_ttl = ip_defttl; 1021 inp->inp_ppcb = (void *)tp; 1022 1023 tp->t_inpcb = inp; 1024 tp->t_mtudisc = ip_mtudisc; 1025 break; 1026 } 1027 #ifdef INET6 1028 case AF_INET6: 1029 { 1030 struct in6pcb *in6p = (struct in6pcb *)aux; 1031 1032 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1033 (rt = rtcache_validate(&in6p->in6p_route)) != NULL 1034 ? rt->rt_ifp 1035 : NULL); 1036 in6p->in6p_ppcb = (void *)tp; 1037 1038 tp->t_in6pcb = in6p; 1039 /* for IPv6, always try to run path MTU discovery */ 1040 tp->t_mtudisc = 1; 1041 break; 1042 } 1043 #endif /* INET6 */ 1044 default: 1045 for (i = 0; i < TCPT_NTIMERS; i++) 1046 callout_destroy(&tp->t_timer[i]); 1047 callout_destroy(&tp->t_delack_ch); 1048 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1049 return (NULL); 1050 } 1051 1052 /* 1053 * Initialize our timebase. When we send timestamps, we take 1054 * the delta from tcp_now -- this means each connection always 1055 * gets a timebase of 1, which makes it, among other things, 1056 * more difficult to determine how long a system has been up, 1057 * and thus how many TCP sequence increments have occurred. 1058 * 1059 * We start with 1, because 0 doesn't work with linux, which 1060 * considers timestamp 0 in a SYN packet as a bug and disables 1061 * timestamps. 1062 */ 1063 tp->ts_timebase = tcp_now - 1; 1064 1065 tcp_congctl_select(tp, tcp_congctl_global_name); 1066 1067 return (tp); 1068 } 1069 1070 /* 1071 * Drop a TCP connection, reporting 1072 * the specified error. If connection is synchronized, 1073 * then send a RST to peer. 1074 */ 1075 struct tcpcb * 1076 tcp_drop(struct tcpcb *tp, int errno) 1077 { 1078 struct socket *so = NULL; 1079 1080 #ifdef DIAGNOSTIC 1081 if (tp->t_inpcb && tp->t_in6pcb) 1082 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1083 #endif 1084 #ifdef INET 1085 if (tp->t_inpcb) 1086 so = tp->t_inpcb->inp_socket; 1087 #endif 1088 #ifdef INET6 1089 if (tp->t_in6pcb) 1090 so = tp->t_in6pcb->in6p_socket; 1091 #endif 1092 if (!so) 1093 return NULL; 1094 1095 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1096 tp->t_state = TCPS_CLOSED; 1097 (void) tcp_output(tp); 1098 TCP_STATINC(TCP_STAT_DROPS); 1099 } else 1100 TCP_STATINC(TCP_STAT_CONNDROPS); 1101 if (errno == ETIMEDOUT && tp->t_softerror) 1102 errno = tp->t_softerror; 1103 so->so_error = errno; 1104 return (tcp_close(tp)); 1105 } 1106 1107 /* 1108 * Close a TCP control block: 1109 * discard all space held by the tcp 1110 * discard internet protocol block 1111 * wake up any sleepers 1112 */ 1113 struct tcpcb * 1114 tcp_close(struct tcpcb *tp) 1115 { 1116 struct inpcb *inp; 1117 #ifdef INET6 1118 struct in6pcb *in6p; 1119 #endif 1120 struct socket *so; 1121 #ifdef RTV_RTT 1122 struct rtentry *rt; 1123 #endif 1124 struct route *ro; 1125 int j; 1126 1127 inp = tp->t_inpcb; 1128 #ifdef INET6 1129 in6p = tp->t_in6pcb; 1130 #endif 1131 so = NULL; 1132 ro = NULL; 1133 if (inp) { 1134 so = inp->inp_socket; 1135 ro = &inp->inp_route; 1136 } 1137 #ifdef INET6 1138 else if (in6p) { 1139 so = in6p->in6p_socket; 1140 ro = (struct route *)&in6p->in6p_route; 1141 } 1142 #endif 1143 1144 #ifdef RTV_RTT 1145 /* 1146 * If we sent enough data to get some meaningful characteristics, 1147 * save them in the routing entry. 'Enough' is arbitrarily 1148 * defined as the sendpipesize (default 4K) * 16. This would 1149 * give us 16 rtt samples assuming we only get one sample per 1150 * window (the usual case on a long haul net). 16 samples is 1151 * enough for the srtt filter to converge to within 5% of the correct 1152 * value; fewer samples and we could save a very bogus rtt. 1153 * 1154 * Don't update the default route's characteristics and don't 1155 * update anything that the user "locked". 1156 */ 1157 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1158 ro && (rt = rtcache_validate(ro)) != NULL && 1159 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1160 u_long i = 0; 1161 1162 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1163 i = tp->t_srtt * 1164 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1165 if (rt->rt_rmx.rmx_rtt && i) 1166 /* 1167 * filter this update to half the old & half 1168 * the new values, converting scale. 1169 * See route.h and tcp_var.h for a 1170 * description of the scaling constants. 1171 */ 1172 rt->rt_rmx.rmx_rtt = 1173 (rt->rt_rmx.rmx_rtt + i) / 2; 1174 else 1175 rt->rt_rmx.rmx_rtt = i; 1176 } 1177 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1178 i = tp->t_rttvar * 1179 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1180 if (rt->rt_rmx.rmx_rttvar && i) 1181 rt->rt_rmx.rmx_rttvar = 1182 (rt->rt_rmx.rmx_rttvar + i) / 2; 1183 else 1184 rt->rt_rmx.rmx_rttvar = i; 1185 } 1186 /* 1187 * update the pipelimit (ssthresh) if it has been updated 1188 * already or if a pipesize was specified & the threshhold 1189 * got below half the pipesize. I.e., wait for bad news 1190 * before we start updating, then update on both good 1191 * and bad news. 1192 */ 1193 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1194 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1195 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1196 /* 1197 * convert the limit from user data bytes to 1198 * packets then to packet data bytes. 1199 */ 1200 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1201 if (i < 2) 1202 i = 2; 1203 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1204 if (rt->rt_rmx.rmx_ssthresh) 1205 rt->rt_rmx.rmx_ssthresh = 1206 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1207 else 1208 rt->rt_rmx.rmx_ssthresh = i; 1209 } 1210 } 1211 #endif /* RTV_RTT */ 1212 /* free the reassembly queue, if any */ 1213 TCP_REASS_LOCK(tp); 1214 (void) tcp_freeq(tp); 1215 TCP_REASS_UNLOCK(tp); 1216 1217 /* free the SACK holes list. */ 1218 tcp_free_sackholes(tp); 1219 tcp_congctl_release(tp); 1220 syn_cache_cleanup(tp); 1221 1222 if (tp->t_template) { 1223 m_free(tp->t_template); 1224 tp->t_template = NULL; 1225 } 1226 1227 /* 1228 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1229 * the timers can also drop the lock. We need to prevent access 1230 * to the tcpcb as it's half torn down. Flag the pcb as dead 1231 * (prevents access by timers) and only then detach it. 1232 */ 1233 tp->t_flags |= TF_DEAD; 1234 if (inp) { 1235 inp->inp_ppcb = 0; 1236 soisdisconnected(so); 1237 in_pcbdetach(inp); 1238 } 1239 #ifdef INET6 1240 else if (in6p) { 1241 in6p->in6p_ppcb = 0; 1242 soisdisconnected(so); 1243 in6_pcbdetach(in6p); 1244 } 1245 #endif 1246 /* 1247 * pcb is no longer visble elsewhere, so we can safely release 1248 * the lock in callout_halt() if needed. 1249 */ 1250 TCP_STATINC(TCP_STAT_CLOSED); 1251 for (j = 0; j < TCPT_NTIMERS; j++) { 1252 callout_halt(&tp->t_timer[j], softnet_lock); 1253 callout_destroy(&tp->t_timer[j]); 1254 } 1255 callout_halt(&tp->t_delack_ch, softnet_lock); 1256 callout_destroy(&tp->t_delack_ch); 1257 pool_put(&tcpcb_pool, tp); 1258 1259 return ((struct tcpcb *)0); 1260 } 1261 1262 int 1263 tcp_freeq(struct tcpcb *tp) 1264 { 1265 struct ipqent *qe; 1266 int rv = 0; 1267 #ifdef TCPREASS_DEBUG 1268 int i = 0; 1269 #endif 1270 1271 TCP_REASS_LOCK_CHECK(tp); 1272 1273 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1274 #ifdef TCPREASS_DEBUG 1275 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1276 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1277 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1278 #endif 1279 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1280 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1281 m_freem(qe->ipqe_m); 1282 tcpipqent_free(qe); 1283 rv = 1; 1284 } 1285 tp->t_segqlen = 0; 1286 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1287 return (rv); 1288 } 1289 1290 /* 1291 * Protocol drain routine. Called when memory is in short supply. 1292 * Don't acquire softnet_lock as can be called from hardware 1293 * interrupt handler. 1294 */ 1295 void 1296 tcp_drain(void) 1297 { 1298 struct inpcb_hdr *inph; 1299 struct tcpcb *tp; 1300 1301 KERNEL_LOCK(1, NULL); 1302 1303 /* 1304 * Free the sequence queue of all TCP connections. 1305 */ 1306 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1307 switch (inph->inph_af) { 1308 case AF_INET: 1309 tp = intotcpcb((struct inpcb *)inph); 1310 break; 1311 #ifdef INET6 1312 case AF_INET6: 1313 tp = in6totcpcb((struct in6pcb *)inph); 1314 break; 1315 #endif 1316 default: 1317 tp = NULL; 1318 break; 1319 } 1320 if (tp != NULL) { 1321 /* 1322 * We may be called from a device's interrupt 1323 * context. If the tcpcb is already busy, 1324 * just bail out now. 1325 */ 1326 if (tcp_reass_lock_try(tp) == 0) 1327 continue; 1328 if (tcp_freeq(tp)) 1329 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1330 TCP_REASS_UNLOCK(tp); 1331 } 1332 } 1333 1334 KERNEL_UNLOCK_ONE(NULL); 1335 } 1336 1337 /* 1338 * Notify a tcp user of an asynchronous error; 1339 * store error as soft error, but wake up user 1340 * (for now, won't do anything until can select for soft error). 1341 */ 1342 void 1343 tcp_notify(struct inpcb *inp, int error) 1344 { 1345 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1346 struct socket *so = inp->inp_socket; 1347 1348 /* 1349 * Ignore some errors if we are hooked up. 1350 * If connection hasn't completed, has retransmitted several times, 1351 * and receives a second error, give up now. This is better 1352 * than waiting a long time to establish a connection that 1353 * can never complete. 1354 */ 1355 if (tp->t_state == TCPS_ESTABLISHED && 1356 (error == EHOSTUNREACH || error == ENETUNREACH || 1357 error == EHOSTDOWN)) { 1358 return; 1359 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1360 tp->t_rxtshift > 3 && tp->t_softerror) 1361 so->so_error = error; 1362 else 1363 tp->t_softerror = error; 1364 cv_broadcast(&so->so_cv); 1365 sorwakeup(so); 1366 sowwakeup(so); 1367 } 1368 1369 #ifdef INET6 1370 void 1371 tcp6_notify(struct in6pcb *in6p, int error) 1372 { 1373 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1374 struct socket *so = in6p->in6p_socket; 1375 1376 /* 1377 * Ignore some errors if we are hooked up. 1378 * If connection hasn't completed, has retransmitted several times, 1379 * and receives a second error, give up now. This is better 1380 * than waiting a long time to establish a connection that 1381 * can never complete. 1382 */ 1383 if (tp->t_state == TCPS_ESTABLISHED && 1384 (error == EHOSTUNREACH || error == ENETUNREACH || 1385 error == EHOSTDOWN)) { 1386 return; 1387 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1388 tp->t_rxtshift > 3 && tp->t_softerror) 1389 so->so_error = error; 1390 else 1391 tp->t_softerror = error; 1392 cv_broadcast(&so->so_cv); 1393 sorwakeup(so); 1394 sowwakeup(so); 1395 } 1396 #endif 1397 1398 #ifdef INET6 1399 void * 1400 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1401 { 1402 struct tcphdr th; 1403 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1404 int nmatch; 1405 struct ip6_hdr *ip6; 1406 const struct sockaddr_in6 *sa6_src = NULL; 1407 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1408 struct mbuf *m; 1409 int off; 1410 1411 if (sa->sa_family != AF_INET6 || 1412 sa->sa_len != sizeof(struct sockaddr_in6)) 1413 return NULL; 1414 if ((unsigned)cmd >= PRC_NCMDS) 1415 return NULL; 1416 else if (cmd == PRC_QUENCH) { 1417 /* 1418 * Don't honor ICMP Source Quench messages meant for 1419 * TCP connections. 1420 */ 1421 return NULL; 1422 } else if (PRC_IS_REDIRECT(cmd)) 1423 notify = in6_rtchange, d = NULL; 1424 else if (cmd == PRC_MSGSIZE) 1425 ; /* special code is present, see below */ 1426 else if (cmd == PRC_HOSTDEAD) 1427 d = NULL; 1428 else if (inet6ctlerrmap[cmd] == 0) 1429 return NULL; 1430 1431 /* if the parameter is from icmp6, decode it. */ 1432 if (d != NULL) { 1433 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1434 m = ip6cp->ip6c_m; 1435 ip6 = ip6cp->ip6c_ip6; 1436 off = ip6cp->ip6c_off; 1437 sa6_src = ip6cp->ip6c_src; 1438 } else { 1439 m = NULL; 1440 ip6 = NULL; 1441 sa6_src = &sa6_any; 1442 off = 0; 1443 } 1444 1445 if (ip6) { 1446 /* 1447 * XXX: We assume that when ip6 is non NULL, 1448 * M and OFF are valid. 1449 */ 1450 1451 /* check if we can safely examine src and dst ports */ 1452 if (m->m_pkthdr.len < off + sizeof(th)) { 1453 if (cmd == PRC_MSGSIZE) 1454 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1455 return NULL; 1456 } 1457 1458 memset(&th, 0, sizeof(th)); 1459 m_copydata(m, off, sizeof(th), (void *)&th); 1460 1461 if (cmd == PRC_MSGSIZE) { 1462 int valid = 0; 1463 1464 /* 1465 * Check to see if we have a valid TCP connection 1466 * corresponding to the address in the ICMPv6 message 1467 * payload. 1468 */ 1469 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1470 th.th_dport, 1471 (const struct in6_addr *)&sa6_src->sin6_addr, 1472 th.th_sport, 0)) 1473 valid++; 1474 1475 /* 1476 * Depending on the value of "valid" and routing table 1477 * size (mtudisc_{hi,lo}wat), we will: 1478 * - recalcurate the new MTU and create the 1479 * corresponding routing entry, or 1480 * - ignore the MTU change notification. 1481 */ 1482 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1483 1484 /* 1485 * no need to call in6_pcbnotify, it should have been 1486 * called via callback if necessary 1487 */ 1488 return NULL; 1489 } 1490 1491 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1492 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1493 if (nmatch == 0 && syn_cache_count && 1494 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1495 inet6ctlerrmap[cmd] == ENETUNREACH || 1496 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1497 syn_cache_unreach((const struct sockaddr *)sa6_src, 1498 sa, &th); 1499 } else { 1500 (void) in6_pcbnotify(&tcbtable, sa, 0, 1501 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1502 } 1503 1504 return NULL; 1505 } 1506 #endif 1507 1508 #ifdef INET 1509 /* assumes that ip header and tcp header are contiguous on mbuf */ 1510 void * 1511 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1512 { 1513 struct ip *ip = v; 1514 struct tcphdr *th; 1515 struct icmp *icp; 1516 extern const int inetctlerrmap[]; 1517 void (*notify)(struct inpcb *, int) = tcp_notify; 1518 int errno; 1519 int nmatch; 1520 struct tcpcb *tp; 1521 u_int mtu; 1522 tcp_seq seq; 1523 struct inpcb *inp; 1524 #ifdef INET6 1525 struct in6pcb *in6p; 1526 struct in6_addr src6, dst6; 1527 #endif 1528 1529 if (sa->sa_family != AF_INET || 1530 sa->sa_len != sizeof(struct sockaddr_in)) 1531 return NULL; 1532 if ((unsigned)cmd >= PRC_NCMDS) 1533 return NULL; 1534 errno = inetctlerrmap[cmd]; 1535 if (cmd == PRC_QUENCH) 1536 /* 1537 * Don't honor ICMP Source Quench messages meant for 1538 * TCP connections. 1539 */ 1540 return NULL; 1541 else if (PRC_IS_REDIRECT(cmd)) 1542 notify = in_rtchange, ip = 0; 1543 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1544 /* 1545 * Check to see if we have a valid TCP connection 1546 * corresponding to the address in the ICMP message 1547 * payload. 1548 * 1549 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1550 */ 1551 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1552 #ifdef INET6 1553 memset(&src6, 0, sizeof(src6)); 1554 memset(&dst6, 0, sizeof(dst6)); 1555 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1556 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1557 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1558 #endif 1559 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1560 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1561 #ifdef INET6 1562 in6p = NULL; 1563 #else 1564 ; 1565 #endif 1566 #ifdef INET6 1567 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1568 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1569 ; 1570 #endif 1571 else 1572 return NULL; 1573 1574 /* 1575 * Now that we've validated that we are actually communicating 1576 * with the host indicated in the ICMP message, locate the 1577 * ICMP header, recalculate the new MTU, and create the 1578 * corresponding routing entry. 1579 */ 1580 icp = (struct icmp *)((char *)ip - 1581 offsetof(struct icmp, icmp_ip)); 1582 if (inp) { 1583 if ((tp = intotcpcb(inp)) == NULL) 1584 return NULL; 1585 } 1586 #ifdef INET6 1587 else if (in6p) { 1588 if ((tp = in6totcpcb(in6p)) == NULL) 1589 return NULL; 1590 } 1591 #endif 1592 else 1593 return NULL; 1594 seq = ntohl(th->th_seq); 1595 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1596 return NULL; 1597 /* 1598 * If the ICMP message advertises a Next-Hop MTU 1599 * equal or larger than the maximum packet size we have 1600 * ever sent, drop the message. 1601 */ 1602 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1603 if (mtu >= tp->t_pmtud_mtu_sent) 1604 return NULL; 1605 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1606 /* 1607 * Calculate new MTU, and create corresponding 1608 * route (traditional PMTUD). 1609 */ 1610 tp->t_flags &= ~TF_PMTUD_PEND; 1611 icmp_mtudisc(icp, ip->ip_dst); 1612 } else { 1613 /* 1614 * Record the information got in the ICMP 1615 * message; act on it later. 1616 * If we had already recorded an ICMP message, 1617 * replace the old one only if the new message 1618 * refers to an older TCP segment 1619 */ 1620 if (tp->t_flags & TF_PMTUD_PEND) { 1621 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1622 return NULL; 1623 } else 1624 tp->t_flags |= TF_PMTUD_PEND; 1625 tp->t_pmtud_th_seq = seq; 1626 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1627 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1628 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1629 } 1630 return NULL; 1631 } else if (cmd == PRC_HOSTDEAD) 1632 ip = 0; 1633 else if (errno == 0) 1634 return NULL; 1635 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1636 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1637 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1638 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1639 if (nmatch == 0 && syn_cache_count && 1640 (inetctlerrmap[cmd] == EHOSTUNREACH || 1641 inetctlerrmap[cmd] == ENETUNREACH || 1642 inetctlerrmap[cmd] == EHOSTDOWN)) { 1643 struct sockaddr_in sin; 1644 memset(&sin, 0, sizeof(sin)); 1645 sin.sin_len = sizeof(sin); 1646 sin.sin_family = AF_INET; 1647 sin.sin_port = th->th_sport; 1648 sin.sin_addr = ip->ip_src; 1649 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1650 } 1651 1652 /* XXX mapped address case */ 1653 } else 1654 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1655 notify); 1656 return NULL; 1657 } 1658 1659 /* 1660 * When a source quench is received, we are being notified of congestion. 1661 * Close the congestion window down to the Loss Window (one segment). 1662 * We will gradually open it again as we proceed. 1663 */ 1664 void 1665 tcp_quench(struct inpcb *inp, int errno) 1666 { 1667 struct tcpcb *tp = intotcpcb(inp); 1668 1669 if (tp) { 1670 tp->snd_cwnd = tp->t_segsz; 1671 tp->t_bytes_acked = 0; 1672 } 1673 } 1674 #endif 1675 1676 #ifdef INET6 1677 void 1678 tcp6_quench(struct in6pcb *in6p, int errno) 1679 { 1680 struct tcpcb *tp = in6totcpcb(in6p); 1681 1682 if (tp) { 1683 tp->snd_cwnd = tp->t_segsz; 1684 tp->t_bytes_acked = 0; 1685 } 1686 } 1687 #endif 1688 1689 #ifdef INET 1690 /* 1691 * Path MTU Discovery handlers. 1692 */ 1693 void 1694 tcp_mtudisc_callback(struct in_addr faddr) 1695 { 1696 #ifdef INET6 1697 struct in6_addr in6; 1698 #endif 1699 1700 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1701 #ifdef INET6 1702 memset(&in6, 0, sizeof(in6)); 1703 in6.s6_addr16[5] = 0xffff; 1704 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1705 tcp6_mtudisc_callback(&in6); 1706 #endif 1707 } 1708 1709 /* 1710 * On receipt of path MTU corrections, flush old route and replace it 1711 * with the new one. Retransmit all unacknowledged packets, to ensure 1712 * that all packets will be received. 1713 */ 1714 void 1715 tcp_mtudisc(struct inpcb *inp, int errno) 1716 { 1717 struct tcpcb *tp = intotcpcb(inp); 1718 struct rtentry *rt = in_pcbrtentry(inp); 1719 1720 if (tp != 0) { 1721 if (rt != 0) { 1722 /* 1723 * If this was not a host route, remove and realloc. 1724 */ 1725 if ((rt->rt_flags & RTF_HOST) == 0) { 1726 in_rtchange(inp, errno); 1727 if ((rt = in_pcbrtentry(inp)) == 0) 1728 return; 1729 } 1730 1731 /* 1732 * Slow start out of the error condition. We 1733 * use the MTU because we know it's smaller 1734 * than the previously transmitted segment. 1735 * 1736 * Note: This is more conservative than the 1737 * suggestion in draft-floyd-incr-init-win-03. 1738 */ 1739 if (rt->rt_rmx.rmx_mtu != 0) 1740 tp->snd_cwnd = 1741 TCP_INITIAL_WINDOW(tcp_init_win, 1742 rt->rt_rmx.rmx_mtu); 1743 } 1744 1745 /* 1746 * Resend unacknowledged packets. 1747 */ 1748 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1749 tcp_output(tp); 1750 } 1751 } 1752 #endif 1753 1754 #ifdef INET6 1755 /* 1756 * Path MTU Discovery handlers. 1757 */ 1758 void 1759 tcp6_mtudisc_callback(struct in6_addr *faddr) 1760 { 1761 struct sockaddr_in6 sin6; 1762 1763 memset(&sin6, 0, sizeof(sin6)); 1764 sin6.sin6_family = AF_INET6; 1765 sin6.sin6_len = sizeof(struct sockaddr_in6); 1766 sin6.sin6_addr = *faddr; 1767 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1768 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1769 } 1770 1771 void 1772 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1773 { 1774 struct tcpcb *tp = in6totcpcb(in6p); 1775 struct rtentry *rt = in6_pcbrtentry(in6p); 1776 1777 if (tp != 0) { 1778 if (rt != 0) { 1779 /* 1780 * If this was not a host route, remove and realloc. 1781 */ 1782 if ((rt->rt_flags & RTF_HOST) == 0) { 1783 in6_rtchange(in6p, errno); 1784 if ((rt = in6_pcbrtentry(in6p)) == 0) 1785 return; 1786 } 1787 1788 /* 1789 * Slow start out of the error condition. We 1790 * use the MTU because we know it's smaller 1791 * than the previously transmitted segment. 1792 * 1793 * Note: This is more conservative than the 1794 * suggestion in draft-floyd-incr-init-win-03. 1795 */ 1796 if (rt->rt_rmx.rmx_mtu != 0) 1797 tp->snd_cwnd = 1798 TCP_INITIAL_WINDOW(tcp_init_win, 1799 rt->rt_rmx.rmx_mtu); 1800 } 1801 1802 /* 1803 * Resend unacknowledged packets. 1804 */ 1805 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1806 tcp_output(tp); 1807 } 1808 } 1809 #endif /* INET6 */ 1810 1811 /* 1812 * Compute the MSS to advertise to the peer. Called only during 1813 * the 3-way handshake. If we are the server (peer initiated 1814 * connection), we are called with a pointer to the interface 1815 * on which the SYN packet arrived. If we are the client (we 1816 * initiated connection), we are called with a pointer to the 1817 * interface out which this connection should go. 1818 * 1819 * NOTE: Do not subtract IP option/extension header size nor IPsec 1820 * header size from MSS advertisement. MSS option must hold the maximum 1821 * segment size we can accept, so it must always be: 1822 * max(if mtu) - ip header - tcp header 1823 */ 1824 u_long 1825 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1826 { 1827 extern u_long in_maxmtu; 1828 u_long mss = 0; 1829 u_long hdrsiz; 1830 1831 /* 1832 * In order to avoid defeating path MTU discovery on the peer, 1833 * we advertise the max MTU of all attached networks as our MSS, 1834 * per RFC 1191, section 3.1. 1835 * 1836 * We provide the option to advertise just the MTU of 1837 * the interface on which we hope this connection will 1838 * be receiving. If we are responding to a SYN, we 1839 * will have a pretty good idea about this, but when 1840 * initiating a connection there is a bit more doubt. 1841 * 1842 * We also need to ensure that loopback has a large enough 1843 * MSS, as the loopback MTU is never included in in_maxmtu. 1844 */ 1845 1846 if (ifp != NULL) 1847 switch (af) { 1848 case AF_INET: 1849 mss = ifp->if_mtu; 1850 break; 1851 #ifdef INET6 1852 case AF_INET6: 1853 mss = IN6_LINKMTU(ifp); 1854 break; 1855 #endif 1856 } 1857 1858 if (tcp_mss_ifmtu == 0) 1859 switch (af) { 1860 case AF_INET: 1861 mss = max(in_maxmtu, mss); 1862 break; 1863 #ifdef INET6 1864 case AF_INET6: 1865 mss = max(in6_maxmtu, mss); 1866 break; 1867 #endif 1868 } 1869 1870 switch (af) { 1871 case AF_INET: 1872 hdrsiz = sizeof(struct ip); 1873 break; 1874 #ifdef INET6 1875 case AF_INET6: 1876 hdrsiz = sizeof(struct ip6_hdr); 1877 break; 1878 #endif 1879 default: 1880 hdrsiz = 0; 1881 break; 1882 } 1883 hdrsiz += sizeof(struct tcphdr); 1884 if (mss > hdrsiz) 1885 mss -= hdrsiz; 1886 1887 mss = max(tcp_mssdflt, mss); 1888 return (mss); 1889 } 1890 1891 /* 1892 * Set connection variables based on the peer's advertised MSS. 1893 * We are passed the TCPCB for the actual connection. If we 1894 * are the server, we are called by the compressed state engine 1895 * when the 3-way handshake is complete. If we are the client, 1896 * we are called when we receive the SYN,ACK from the server. 1897 * 1898 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1899 * before this routine is called! 1900 */ 1901 void 1902 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1903 { 1904 struct socket *so; 1905 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1906 struct rtentry *rt; 1907 #endif 1908 u_long bufsize; 1909 int mss; 1910 1911 #ifdef DIAGNOSTIC 1912 if (tp->t_inpcb && tp->t_in6pcb) 1913 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1914 #endif 1915 so = NULL; 1916 rt = NULL; 1917 #ifdef INET 1918 if (tp->t_inpcb) { 1919 so = tp->t_inpcb->inp_socket; 1920 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1921 rt = in_pcbrtentry(tp->t_inpcb); 1922 #endif 1923 } 1924 #endif 1925 #ifdef INET6 1926 if (tp->t_in6pcb) { 1927 so = tp->t_in6pcb->in6p_socket; 1928 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1929 rt = in6_pcbrtentry(tp->t_in6pcb); 1930 #endif 1931 } 1932 #endif 1933 1934 /* 1935 * As per RFC1122, use the default MSS value, unless they 1936 * sent us an offer. Do not accept offers less than 256 bytes. 1937 */ 1938 mss = tcp_mssdflt; 1939 if (offer) 1940 mss = offer; 1941 mss = max(mss, 256); /* sanity */ 1942 tp->t_peermss = mss; 1943 mss -= tcp_optlen(tp); 1944 #ifdef INET 1945 if (tp->t_inpcb) 1946 mss -= ip_optlen(tp->t_inpcb); 1947 #endif 1948 #ifdef INET6 1949 if (tp->t_in6pcb) 1950 mss -= ip6_optlen(tp->t_in6pcb); 1951 #endif 1952 1953 /* 1954 * If there's a pipesize, change the socket buffer to that size. 1955 * Make the socket buffer an integral number of MSS units. If 1956 * the MSS is larger than the socket buffer, artificially decrease 1957 * the MSS. 1958 */ 1959 #ifdef RTV_SPIPE 1960 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1961 bufsize = rt->rt_rmx.rmx_sendpipe; 1962 else 1963 #endif 1964 { 1965 KASSERT(so != NULL); 1966 bufsize = so->so_snd.sb_hiwat; 1967 } 1968 if (bufsize < mss) 1969 mss = bufsize; 1970 else { 1971 bufsize = roundup(bufsize, mss); 1972 if (bufsize > sb_max) 1973 bufsize = sb_max; 1974 (void) sbreserve(&so->so_snd, bufsize, so); 1975 } 1976 tp->t_segsz = mss; 1977 1978 #ifdef RTV_SSTHRESH 1979 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1980 /* 1981 * There's some sort of gateway or interface buffer 1982 * limit on the path. Use this to set the slow 1983 * start threshold, but set the threshold to no less 1984 * than 2 * MSS. 1985 */ 1986 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1987 } 1988 #endif 1989 } 1990 1991 /* 1992 * Processing necessary when a TCP connection is established. 1993 */ 1994 void 1995 tcp_established(struct tcpcb *tp) 1996 { 1997 struct socket *so; 1998 #ifdef RTV_RPIPE 1999 struct rtentry *rt; 2000 #endif 2001 u_long bufsize; 2002 2003 #ifdef DIAGNOSTIC 2004 if (tp->t_inpcb && tp->t_in6pcb) 2005 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2006 #endif 2007 so = NULL; 2008 rt = NULL; 2009 #ifdef INET 2010 if (tp->t_inpcb) { 2011 so = tp->t_inpcb->inp_socket; 2012 #if defined(RTV_RPIPE) 2013 rt = in_pcbrtentry(tp->t_inpcb); 2014 #endif 2015 } 2016 #endif 2017 #ifdef INET6 2018 if (tp->t_in6pcb) { 2019 so = tp->t_in6pcb->in6p_socket; 2020 #if defined(RTV_RPIPE) 2021 rt = in6_pcbrtentry(tp->t_in6pcb); 2022 #endif 2023 } 2024 #endif 2025 2026 tp->t_state = TCPS_ESTABLISHED; 2027 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2028 2029 #ifdef RTV_RPIPE 2030 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2031 bufsize = rt->rt_rmx.rmx_recvpipe; 2032 else 2033 #endif 2034 { 2035 KASSERT(so != NULL); 2036 bufsize = so->so_rcv.sb_hiwat; 2037 } 2038 if (bufsize > tp->t_ourmss) { 2039 bufsize = roundup(bufsize, tp->t_ourmss); 2040 if (bufsize > sb_max) 2041 bufsize = sb_max; 2042 (void) sbreserve(&so->so_rcv, bufsize, so); 2043 } 2044 } 2045 2046 /* 2047 * Check if there's an initial rtt or rttvar. Convert from the 2048 * route-table units to scaled multiples of the slow timeout timer. 2049 * Called only during the 3-way handshake. 2050 */ 2051 void 2052 tcp_rmx_rtt(struct tcpcb *tp) 2053 { 2054 #ifdef RTV_RTT 2055 struct rtentry *rt = NULL; 2056 int rtt; 2057 2058 #ifdef DIAGNOSTIC 2059 if (tp->t_inpcb && tp->t_in6pcb) 2060 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2061 #endif 2062 #ifdef INET 2063 if (tp->t_inpcb) 2064 rt = in_pcbrtentry(tp->t_inpcb); 2065 #endif 2066 #ifdef INET6 2067 if (tp->t_in6pcb) 2068 rt = in6_pcbrtentry(tp->t_in6pcb); 2069 #endif 2070 if (rt == NULL) 2071 return; 2072 2073 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2074 /* 2075 * XXX The lock bit for MTU indicates that the value 2076 * is also a minimum value; this is subject to time. 2077 */ 2078 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2079 TCPT_RANGESET(tp->t_rttmin, 2080 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2081 TCPTV_MIN, TCPTV_REXMTMAX); 2082 tp->t_srtt = rtt / 2083 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2084 if (rt->rt_rmx.rmx_rttvar) { 2085 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2086 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2087 (TCP_RTTVAR_SHIFT + 2)); 2088 } else { 2089 /* Default variation is +- 1 rtt */ 2090 tp->t_rttvar = 2091 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2092 } 2093 TCPT_RANGESET(tp->t_rxtcur, 2094 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2095 tp->t_rttmin, TCPTV_REXMTMAX); 2096 } 2097 #endif 2098 } 2099 2100 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2101 #if NRND > 0 2102 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2103 #endif 2104 2105 /* 2106 * Get a new sequence value given a tcp control block 2107 */ 2108 tcp_seq 2109 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2110 { 2111 2112 #ifdef INET 2113 if (tp->t_inpcb != NULL) { 2114 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2115 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2116 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2117 addin)); 2118 } 2119 #endif 2120 #ifdef INET6 2121 if (tp->t_in6pcb != NULL) { 2122 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2123 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2124 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2125 addin)); 2126 } 2127 #endif 2128 /* Not possible. */ 2129 panic("tcp_new_iss"); 2130 } 2131 2132 /* 2133 * This routine actually generates a new TCP initial sequence number. 2134 */ 2135 tcp_seq 2136 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2137 size_t addrsz, tcp_seq addin) 2138 { 2139 tcp_seq tcp_iss; 2140 2141 #if NRND > 0 2142 static bool tcp_iss_gotten_secret; 2143 2144 /* 2145 * If we haven't been here before, initialize our cryptographic 2146 * hash secret. 2147 */ 2148 if (tcp_iss_gotten_secret == false) { 2149 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2150 RND_EXTRACT_ANY); 2151 tcp_iss_gotten_secret = true; 2152 } 2153 2154 if (tcp_do_rfc1948) { 2155 MD5_CTX ctx; 2156 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2157 2158 /* 2159 * Compute the base value of the ISS. It is a hash 2160 * of (saddr, sport, daddr, dport, secret). 2161 */ 2162 MD5Init(&ctx); 2163 2164 MD5Update(&ctx, (u_char *) laddr, addrsz); 2165 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2166 2167 MD5Update(&ctx, (u_char *) faddr, addrsz); 2168 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2169 2170 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2171 2172 MD5Final(hash, &ctx); 2173 2174 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2175 2176 /* 2177 * Now increment our "timer", and add it in to 2178 * the computed value. 2179 * 2180 * XXX Use `addin'? 2181 * XXX TCP_ISSINCR too large to use? 2182 */ 2183 tcp_iss_seq += TCP_ISSINCR; 2184 #ifdef TCPISS_DEBUG 2185 printf("ISS hash 0x%08x, ", tcp_iss); 2186 #endif 2187 tcp_iss += tcp_iss_seq + addin; 2188 #ifdef TCPISS_DEBUG 2189 printf("new ISS 0x%08x\n", tcp_iss); 2190 #endif 2191 } else 2192 #endif /* NRND > 0 */ 2193 { 2194 /* 2195 * Randomize. 2196 */ 2197 #if NRND > 0 2198 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2199 #else 2200 tcp_iss = arc4random(); 2201 #endif 2202 2203 /* 2204 * If we were asked to add some amount to a known value, 2205 * we will take a random value obtained above, mask off 2206 * the upper bits, and add in the known value. We also 2207 * add in a constant to ensure that we are at least a 2208 * certain distance from the original value. 2209 * 2210 * This is used when an old connection is in timed wait 2211 * and we have a new one coming in, for instance. 2212 */ 2213 if (addin != 0) { 2214 #ifdef TCPISS_DEBUG 2215 printf("Random %08x, ", tcp_iss); 2216 #endif 2217 tcp_iss &= TCP_ISS_RANDOM_MASK; 2218 tcp_iss += addin + TCP_ISSINCR; 2219 #ifdef TCPISS_DEBUG 2220 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2221 #endif 2222 } else { 2223 tcp_iss &= TCP_ISS_RANDOM_MASK; 2224 tcp_iss += tcp_iss_seq; 2225 tcp_iss_seq += TCP_ISSINCR; 2226 #ifdef TCPISS_DEBUG 2227 printf("ISS %08x\n", tcp_iss); 2228 #endif 2229 } 2230 } 2231 2232 if (tcp_compat_42) { 2233 /* 2234 * Limit it to the positive range for really old TCP 2235 * implementations. 2236 * Just AND off the top bit instead of checking if 2237 * is set first - saves a branch 50% of the time. 2238 */ 2239 tcp_iss &= 0x7fffffff; /* XXX */ 2240 } 2241 2242 return (tcp_iss); 2243 } 2244 2245 #if defined(IPSEC) || defined(FAST_IPSEC) 2246 /* compute ESP/AH header size for TCP, including outer IP header. */ 2247 size_t 2248 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2249 { 2250 struct inpcb *inp; 2251 size_t hdrsiz; 2252 2253 /* XXX mapped addr case (tp->t_in6pcb) */ 2254 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2255 return 0; 2256 switch (tp->t_family) { 2257 case AF_INET: 2258 /* XXX: should use currect direction. */ 2259 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2260 break; 2261 default: 2262 hdrsiz = 0; 2263 break; 2264 } 2265 2266 return hdrsiz; 2267 } 2268 2269 #ifdef INET6 2270 size_t 2271 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2272 { 2273 struct in6pcb *in6p; 2274 size_t hdrsiz; 2275 2276 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2277 return 0; 2278 switch (tp->t_family) { 2279 case AF_INET6: 2280 /* XXX: should use currect direction. */ 2281 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2282 break; 2283 case AF_INET: 2284 /* mapped address case - tricky */ 2285 default: 2286 hdrsiz = 0; 2287 break; 2288 } 2289 2290 return hdrsiz; 2291 } 2292 #endif 2293 #endif /*IPSEC*/ 2294 2295 /* 2296 * Determine the length of the TCP options for this connection. 2297 * 2298 * XXX: What do we do for SACK, when we add that? Just reserve 2299 * all of the space? Otherwise we can't exactly be incrementing 2300 * cwnd by an amount that varies depending on the amount we last 2301 * had to SACK! 2302 */ 2303 2304 u_int 2305 tcp_optlen(struct tcpcb *tp) 2306 { 2307 u_int optlen; 2308 2309 optlen = 0; 2310 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2311 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2312 optlen += TCPOLEN_TSTAMP_APPA; 2313 2314 #ifdef TCP_SIGNATURE 2315 if (tp->t_flags & TF_SIGNATURE) 2316 optlen += TCPOLEN_SIGNATURE + 2; 2317 #endif /* TCP_SIGNATURE */ 2318 2319 return optlen; 2320 } 2321 2322 u_int 2323 tcp_hdrsz(struct tcpcb *tp) 2324 { 2325 u_int hlen; 2326 2327 switch (tp->t_family) { 2328 #ifdef INET6 2329 case AF_INET6: 2330 hlen = sizeof(struct ip6_hdr); 2331 break; 2332 #endif 2333 case AF_INET: 2334 hlen = sizeof(struct ip); 2335 break; 2336 default: 2337 hlen = 0; 2338 break; 2339 } 2340 hlen += sizeof(struct tcphdr); 2341 2342 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2343 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2344 hlen += TCPOLEN_TSTAMP_APPA; 2345 #ifdef TCP_SIGNATURE 2346 if (tp->t_flags & TF_SIGNATURE) 2347 hlen += TCPOLEN_SIGLEN; 2348 #endif 2349 return hlen; 2350 } 2351 2352 void 2353 tcp_statinc(u_int stat) 2354 { 2355 2356 KASSERT(stat < TCP_NSTATS); 2357 TCP_STATINC(stat); 2358 } 2359 2360 void 2361 tcp_statadd(u_int stat, uint64_t val) 2362 { 2363 2364 KASSERT(stat < TCP_NSTATS); 2365 TCP_STATADD(stat, val); 2366 } 2367