1 /* $NetBSD: tcp_subr.c,v 1.205 2006/10/13 18:28:06 dogcow 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 1.205 2006/10/13 18:28:06 dogcow 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_sack_tp_maxholes = 32; 206 int tcp_sack_globalmaxholes = 1024; 207 int tcp_sack_globalholes = 0; 208 int tcp_ecn_maxretries = 1; 209 210 /* tcb hash */ 211 #ifndef TCBHASHSIZE 212 #define TCBHASHSIZE 128 213 #endif 214 int tcbhashsize = TCBHASHSIZE; 215 216 /* syn hash parameters */ 217 #define TCP_SYN_HASH_SIZE 293 218 #define TCP_SYN_BUCKET_SIZE 35 219 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 220 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 221 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 222 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 223 224 int tcp_freeq(struct tcpcb *); 225 226 #ifdef INET 227 void tcp_mtudisc_callback(struct in_addr); 228 #endif 229 #ifdef INET6 230 void tcp6_mtudisc_callback(struct in6_addr *); 231 #endif 232 233 #ifdef INET6 234 void tcp6_mtudisc(struct in6pcb *, int); 235 #endif 236 237 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL); 238 239 #ifdef TCP_CSUM_COUNTERS 240 #include <sys/device.h> 241 242 #if defined(INET) 243 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "hwcsum bad"); 245 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 246 NULL, "tcp", "hwcsum ok"); 247 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "hwcsum data"); 249 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 250 NULL, "tcp", "swcsum"); 251 252 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 253 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 254 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 255 EVCNT_ATTACH_STATIC(tcp_swcsum); 256 #endif /* defined(INET) */ 257 258 #if defined(INET6) 259 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp6", "hwcsum bad"); 261 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 NULL, "tcp6", "hwcsum ok"); 263 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "hwcsum data"); 265 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 NULL, "tcp6", "swcsum"); 267 268 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 269 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 270 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 271 EVCNT_ATTACH_STATIC(tcp6_swcsum); 272 #endif /* defined(INET6) */ 273 #endif /* TCP_CSUM_COUNTERS */ 274 275 276 #ifdef TCP_OUTPUT_COUNTERS 277 #include <sys/device.h> 278 279 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 NULL, "tcp", "output big header"); 281 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 NULL, "tcp", "output predict hit"); 283 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output predict miss"); 285 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output copy small"); 287 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output copy big"); 289 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 NULL, "tcp", "output reference big"); 291 292 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 293 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 294 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 295 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 296 EVCNT_ATTACH_STATIC(tcp_output_copybig); 297 EVCNT_ATTACH_STATIC(tcp_output_refbig); 298 299 #endif /* TCP_OUTPUT_COUNTERS */ 300 301 #ifdef TCP_REASS_COUNTERS 302 #include <sys/device.h> 303 304 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 305 NULL, "tcp_reass", "calls"); 306 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 307 &tcp_reass_, "tcp_reass", "insert into empty queue"); 308 struct evcnt tcp_reass_iteration[8] = { 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 317 }; 318 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 319 &tcp_reass_, "tcp_reass", "prepend to first"); 320 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 321 &tcp_reass_, "tcp_reass", "prepend"); 322 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "insert"); 324 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "insert at tail"); 326 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "append"); 328 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "append to tail fragment"); 330 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "overlap at end"); 332 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "overlap at start"); 334 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "duplicate segment"); 336 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 337 &tcp_reass_, "tcp_reass", "duplicate fragment"); 338 339 EVCNT_ATTACH_STATIC(tcp_reass_); 340 EVCNT_ATTACH_STATIC(tcp_reass_empty); 341 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 342 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 349 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 350 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 351 EVCNT_ATTACH_STATIC(tcp_reass_insert); 352 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 353 EVCNT_ATTACH_STATIC(tcp_reass_append); 354 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 355 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 356 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 357 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 358 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 359 360 #endif /* TCP_REASS_COUNTERS */ 361 362 #ifdef MBUFTRACE 363 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 364 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 365 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 366 #endif 367 368 /* 369 * Tcp initialization 370 */ 371 void 372 tcp_init(void) 373 { 374 int hlen; 375 376 /* Initialize the TCPCB template. */ 377 tcp_tcpcb_template(); 378 379 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 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 MOWNER_ATTACH(&tcp_tx_mowner); 408 MOWNER_ATTACH(&tcp_rx_mowner); 409 MOWNER_ATTACH(&tcp_mowner); 410 } 411 412 /* 413 * Create template to be used to send tcp packets on a connection. 414 * Call after host entry created, allocates an mbuf and fills 415 * in a skeletal tcp/ip header, minimizing the amount of work 416 * necessary when the connection is used. 417 */ 418 struct mbuf * 419 tcp_template(struct tcpcb *tp) 420 { 421 struct inpcb *inp = tp->t_inpcb; 422 #ifdef INET6 423 struct in6pcb *in6p = tp->t_in6pcb; 424 #endif 425 struct tcphdr *n; 426 struct mbuf *m; 427 int hlen; 428 429 switch (tp->t_family) { 430 case AF_INET: 431 hlen = sizeof(struct ip); 432 if (inp) 433 break; 434 #ifdef INET6 435 if (in6p) { 436 /* mapped addr case */ 437 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 438 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 439 break; 440 } 441 #endif 442 return NULL; /*EINVAL*/ 443 #ifdef INET6 444 case AF_INET6: 445 hlen = sizeof(struct ip6_hdr); 446 if (in6p) { 447 /* more sainty check? */ 448 break; 449 } 450 return NULL; /*EINVAL*/ 451 #endif 452 default: 453 hlen = 0; /*pacify gcc*/ 454 return NULL; /*EAFNOSUPPORT*/ 455 } 456 #ifdef DIAGNOSTIC 457 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 458 panic("mclbytes too small for t_template"); 459 #endif 460 m = tp->t_template; 461 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 462 ; 463 else { 464 if (m) 465 m_freem(m); 466 m = tp->t_template = NULL; 467 MGETHDR(m, M_DONTWAIT, MT_HEADER); 468 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 469 MCLGET(m, M_DONTWAIT); 470 if ((m->m_flags & M_EXT) == 0) { 471 m_free(m); 472 m = NULL; 473 } 474 } 475 if (m == NULL) 476 return NULL; 477 MCLAIM(m, &tcp_mowner); 478 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 479 } 480 481 bzero(mtod(m, caddr_t), m->m_len); 482 483 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 484 485 switch (tp->t_family) { 486 case AF_INET: 487 { 488 struct ipovly *ipov; 489 mtod(m, struct ip *)->ip_v = 4; 490 mtod(m, struct ip *)->ip_hl = hlen >> 2; 491 ipov = mtod(m, struct ipovly *); 492 ipov->ih_pr = IPPROTO_TCP; 493 ipov->ih_len = htons(sizeof(struct tcphdr)); 494 if (inp) { 495 ipov->ih_src = inp->inp_laddr; 496 ipov->ih_dst = inp->inp_faddr; 497 } 498 #ifdef INET6 499 else if (in6p) { 500 /* mapped addr case */ 501 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 502 sizeof(ipov->ih_src)); 503 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 504 sizeof(ipov->ih_dst)); 505 } 506 #endif 507 /* 508 * Compute the pseudo-header portion of the checksum 509 * now. We incrementally add in the TCP option and 510 * payload lengths later, and then compute the TCP 511 * checksum right before the packet is sent off onto 512 * the wire. 513 */ 514 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 515 ipov->ih_dst.s_addr, 516 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 517 break; 518 } 519 #ifdef INET6 520 case AF_INET6: 521 { 522 struct ip6_hdr *ip6; 523 mtod(m, struct ip *)->ip_v = 6; 524 ip6 = mtod(m, struct ip6_hdr *); 525 ip6->ip6_nxt = IPPROTO_TCP; 526 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 527 ip6->ip6_src = in6p->in6p_laddr; 528 ip6->ip6_dst = in6p->in6p_faddr; 529 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 530 if (ip6_auto_flowlabel) { 531 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 532 ip6->ip6_flow |= 533 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 534 } 535 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 536 ip6->ip6_vfc |= IPV6_VERSION; 537 538 /* 539 * Compute the pseudo-header portion of the checksum 540 * now. We incrementally add in the TCP option and 541 * payload lengths later, and then compute the TCP 542 * checksum right before the packet is sent off onto 543 * the wire. 544 */ 545 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 546 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 547 htonl(IPPROTO_TCP)); 548 break; 549 } 550 #endif 551 } 552 if (inp) { 553 n->th_sport = inp->inp_lport; 554 n->th_dport = inp->inp_fport; 555 } 556 #ifdef INET6 557 else if (in6p) { 558 n->th_sport = in6p->in6p_lport; 559 n->th_dport = in6p->in6p_fport; 560 } 561 #endif 562 n->th_seq = 0; 563 n->th_ack = 0; 564 n->th_x2 = 0; 565 n->th_off = 5; 566 n->th_flags = 0; 567 n->th_win = 0; 568 n->th_urp = 0; 569 return (m); 570 } 571 572 /* 573 * Send a single message to the TCP at address specified by 574 * the given TCP/IP header. If m == 0, then we make a copy 575 * of the tcpiphdr at ti and send directly to the addressed host. 576 * This is used to force keep alive messages out using the TCP 577 * template for a connection tp->t_template. If flags are given 578 * then we send a message back to the TCP which originated the 579 * segment ti, and discard the mbuf containing it and any other 580 * attached mbufs. 581 * 582 * In any case the ack and sequence number of the transmitted 583 * segment are as specified by the parameters. 584 */ 585 int 586 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 587 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 588 { 589 struct route *ro; 590 int error, tlen, win = 0; 591 int hlen; 592 struct ip *ip; 593 #ifdef INET6 594 struct ip6_hdr *ip6; 595 #endif 596 int family; /* family on packet, not inpcb/in6pcb! */ 597 struct tcphdr *th; 598 struct socket *so; 599 600 if (tp != NULL && (flags & TH_RST) == 0) { 601 #ifdef DIAGNOSTIC 602 if (tp->t_inpcb && tp->t_in6pcb) 603 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 604 #endif 605 #ifdef INET 606 if (tp->t_inpcb) 607 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 608 #endif 609 #ifdef INET6 610 if (tp->t_in6pcb) 611 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 612 #endif 613 } 614 615 th = NULL; /* Quell uninitialized warning */ 616 ip = NULL; 617 #ifdef INET6 618 ip6 = NULL; 619 #endif 620 if (m == 0) { 621 if (!template) 622 return EINVAL; 623 624 /* get family information from template */ 625 switch (mtod(template, struct ip *)->ip_v) { 626 case 4: 627 family = AF_INET; 628 hlen = sizeof(struct ip); 629 break; 630 #ifdef INET6 631 case 6: 632 family = AF_INET6; 633 hlen = sizeof(struct ip6_hdr); 634 break; 635 #endif 636 default: 637 return EAFNOSUPPORT; 638 } 639 640 MGETHDR(m, M_DONTWAIT, MT_HEADER); 641 if (m) { 642 MCLAIM(m, &tcp_tx_mowner); 643 MCLGET(m, M_DONTWAIT); 644 if ((m->m_flags & M_EXT) == 0) { 645 m_free(m); 646 m = NULL; 647 } 648 } 649 if (m == NULL) 650 return (ENOBUFS); 651 652 if (tcp_compat_42) 653 tlen = 1; 654 else 655 tlen = 0; 656 657 m->m_data += max_linkhdr; 658 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 659 template->m_len); 660 switch (family) { 661 case AF_INET: 662 ip = mtod(m, struct ip *); 663 th = (struct tcphdr *)(ip + 1); 664 break; 665 #ifdef INET6 666 case AF_INET6: 667 ip6 = mtod(m, struct ip6_hdr *); 668 th = (struct tcphdr *)(ip6 + 1); 669 break; 670 #endif 671 #if 0 672 default: 673 /* noone will visit here */ 674 m_freem(m); 675 return EAFNOSUPPORT; 676 #endif 677 } 678 flags = TH_ACK; 679 } else { 680 681 if ((m->m_flags & M_PKTHDR) == 0) { 682 #if 0 683 printf("non PKTHDR to tcp_respond\n"); 684 #endif 685 m_freem(m); 686 return EINVAL; 687 } 688 #ifdef DIAGNOSTIC 689 if (!th0) 690 panic("th0 == NULL in tcp_respond"); 691 #endif 692 693 /* get family information from m */ 694 switch (mtod(m, struct ip *)->ip_v) { 695 case 4: 696 family = AF_INET; 697 hlen = sizeof(struct ip); 698 ip = mtod(m, struct ip *); 699 break; 700 #ifdef INET6 701 case 6: 702 family = AF_INET6; 703 hlen = sizeof(struct ip6_hdr); 704 ip6 = mtod(m, struct ip6_hdr *); 705 break; 706 #endif 707 default: 708 m_freem(m); 709 return EAFNOSUPPORT; 710 } 711 /* clear h/w csum flags inherited from rx packet */ 712 m->m_pkthdr.csum_flags = 0; 713 714 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 715 tlen = sizeof(*th0); 716 else 717 tlen = th0->th_off << 2; 718 719 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 720 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 721 m->m_len = hlen + tlen; 722 m_freem(m->m_next); 723 m->m_next = NULL; 724 } else { 725 struct mbuf *n; 726 727 #ifdef DIAGNOSTIC 728 if (max_linkhdr + hlen + tlen > MCLBYTES) { 729 m_freem(m); 730 return EMSGSIZE; 731 } 732 #endif 733 MGETHDR(n, M_DONTWAIT, MT_HEADER); 734 if (n && max_linkhdr + hlen + tlen > MHLEN) { 735 MCLGET(n, M_DONTWAIT); 736 if ((n->m_flags & M_EXT) == 0) { 737 m_freem(n); 738 n = NULL; 739 } 740 } 741 if (!n) { 742 m_freem(m); 743 return ENOBUFS; 744 } 745 746 MCLAIM(n, &tcp_tx_mowner); 747 n->m_data += max_linkhdr; 748 n->m_len = hlen + tlen; 749 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 750 m_copyback(n, hlen, tlen, (caddr_t)th0); 751 752 m_freem(m); 753 m = n; 754 n = NULL; 755 } 756 757 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 758 switch (family) { 759 case AF_INET: 760 ip = mtod(m, struct ip *); 761 th = (struct tcphdr *)(ip + 1); 762 ip->ip_p = IPPROTO_TCP; 763 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 764 ip->ip_p = IPPROTO_TCP; 765 break; 766 #ifdef INET6 767 case AF_INET6: 768 ip6 = mtod(m, struct ip6_hdr *); 769 th = (struct tcphdr *)(ip6 + 1); 770 ip6->ip6_nxt = IPPROTO_TCP; 771 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 772 ip6->ip6_nxt = IPPROTO_TCP; 773 break; 774 #endif 775 #if 0 776 default: 777 /* noone will visit here */ 778 m_freem(m); 779 return EAFNOSUPPORT; 780 #endif 781 } 782 xchg(th->th_dport, th->th_sport, u_int16_t); 783 #undef xchg 784 tlen = 0; /*be friendly with the following code*/ 785 } 786 th->th_seq = htonl(seq); 787 th->th_ack = htonl(ack); 788 th->th_x2 = 0; 789 if ((flags & TH_SYN) == 0) { 790 if (tp) 791 win >>= tp->rcv_scale; 792 if (win > TCP_MAXWIN) 793 win = TCP_MAXWIN; 794 th->th_win = htons((u_int16_t)win); 795 th->th_off = sizeof (struct tcphdr) >> 2; 796 tlen += sizeof(*th); 797 } else 798 tlen += th->th_off << 2; 799 m->m_len = hlen + tlen; 800 m->m_pkthdr.len = hlen + tlen; 801 m->m_pkthdr.rcvif = (struct ifnet *) 0; 802 th->th_flags = flags; 803 th->th_urp = 0; 804 805 switch (family) { 806 #ifdef INET 807 case AF_INET: 808 { 809 struct ipovly *ipov = (struct ipovly *)ip; 810 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 811 ipov->ih_len = htons((u_int16_t)tlen); 812 813 th->th_sum = 0; 814 th->th_sum = in_cksum(m, hlen + tlen); 815 ip->ip_len = htons(hlen + tlen); 816 ip->ip_ttl = ip_defttl; 817 break; 818 } 819 #endif 820 #ifdef INET6 821 case AF_INET6: 822 { 823 th->th_sum = 0; 824 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 825 tlen); 826 ip6->ip6_plen = htons(tlen); 827 if (tp && tp->t_in6pcb) { 828 struct ifnet *oifp; 829 ro = (struct route *)&tp->t_in6pcb->in6p_route; 830 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 831 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 832 } else 833 ip6->ip6_hlim = ip6_defhlim; 834 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 835 if (ip6_auto_flowlabel) { 836 ip6->ip6_flow |= 837 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 838 } 839 break; 840 } 841 #endif 842 } 843 844 if (tp && tp->t_inpcb) 845 so = tp->t_inpcb->inp_socket; 846 #ifdef INET6 847 else if (tp && tp->t_in6pcb) 848 so = tp->t_in6pcb->in6p_socket; 849 #endif 850 else 851 so = NULL; 852 853 if (tp != NULL && tp->t_inpcb != NULL) { 854 ro = &tp->t_inpcb->inp_route; 855 #ifdef DIAGNOSTIC 856 if (family != AF_INET) 857 panic("tcp_respond: address family mismatch"); 858 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 859 panic("tcp_respond: ip_dst %x != inp_faddr %x", 860 ntohl(ip->ip_dst.s_addr), 861 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 862 } 863 #endif 864 } 865 #ifdef INET6 866 else if (tp != NULL && tp->t_in6pcb != NULL) { 867 ro = (struct route *)&tp->t_in6pcb->in6p_route; 868 #ifdef DIAGNOSTIC 869 if (family == AF_INET) { 870 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 871 panic("tcp_respond: not mapped addr"); 872 if (bcmp(&ip->ip_dst, 873 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 874 sizeof(ip->ip_dst)) != 0) { 875 panic("tcp_respond: ip_dst != in6p_faddr"); 876 } 877 } else if (family == AF_INET6) { 878 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 879 &tp->t_in6pcb->in6p_faddr)) 880 panic("tcp_respond: ip6_dst != in6p_faddr"); 881 } else 882 panic("tcp_respond: address family mismatch"); 883 #endif 884 } 885 #endif 886 else 887 ro = NULL; 888 889 switch (family) { 890 #ifdef INET 891 case AF_INET: 892 error = ip_output(m, NULL, ro, 893 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 894 (struct ip_moptions *)0, so); 895 break; 896 #endif 897 #ifdef INET6 898 case AF_INET6: 899 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 900 (struct ip6_moptions *)0, so, NULL); 901 break; 902 #endif 903 default: 904 error = EAFNOSUPPORT; 905 break; 906 } 907 908 return (error); 909 } 910 911 /* 912 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 913 * a bunch of members individually, we maintain this template for the 914 * static and mostly-static components of the TCPCB, and copy it into 915 * the new TCPCB instead. 916 */ 917 static struct tcpcb tcpcb_template = { 918 /* 919 * If TCP_NTIMERS ever changes, we'll need to update this 920 * initializer. 921 */ 922 .t_timer = { 923 CALLOUT_INITIALIZER, 924 CALLOUT_INITIALIZER, 925 CALLOUT_INITIALIZER, 926 CALLOUT_INITIALIZER, 927 }, 928 .t_delack_ch = CALLOUT_INITIALIZER, 929 930 .t_srtt = TCPTV_SRTTBASE, 931 .t_rttmin = TCPTV_MIN, 932 933 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 934 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 935 .snd_numholes = 0, 936 937 .t_partialacks = -1, 938 }; 939 940 /* 941 * Updates the TCPCB template whenever a parameter that would affect 942 * the template is changed. 943 */ 944 void 945 tcp_tcpcb_template(void) 946 { 947 struct tcpcb *tp = &tcpcb_template; 948 int flags; 949 950 tp->t_peermss = tcp_mssdflt; 951 tp->t_ourmss = tcp_mssdflt; 952 tp->t_segsz = tcp_mssdflt; 953 954 flags = 0; 955 if (tcp_do_rfc1323 && tcp_do_win_scale) 956 flags |= TF_REQ_SCALE; 957 if (tcp_do_rfc1323 && tcp_do_timestamps) 958 flags |= TF_REQ_TSTMP; 959 tp->t_flags = flags; 960 961 /* 962 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 963 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 964 * reasonable initial retransmit time. 965 */ 966 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 967 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 968 TCPTV_MIN, TCPTV_REXMTMAX); 969 } 970 971 /* 972 * Create a new TCP control block, making an 973 * empty reassembly queue and hooking it to the argument 974 * protocol control block. 975 */ 976 /* family selects inpcb, or in6pcb */ 977 struct tcpcb * 978 tcp_newtcpcb(int family, void *aux) 979 { 980 struct tcpcb *tp; 981 int i; 982 983 /* XXX Consider using a pool_cache for speed. */ 984 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 985 if (tp == NULL) 986 return (NULL); 987 memcpy(tp, &tcpcb_template, sizeof(*tp)); 988 TAILQ_INIT(&tp->segq); 989 TAILQ_INIT(&tp->timeq); 990 tp->t_family = family; /* may be overridden later on */ 991 TAILQ_INIT(&tp->snd_holes); 992 LIST_INIT(&tp->t_sc); /* XXX can template this */ 993 994 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 995 for (i = 0; i < TCPT_NTIMERS; i++) 996 TCP_TIMER_INIT(tp, i); 997 998 switch (family) { 999 case AF_INET: 1000 { 1001 struct inpcb *inp = (struct inpcb *)aux; 1002 1003 inp->inp_ip.ip_ttl = ip_defttl; 1004 inp->inp_ppcb = (caddr_t)tp; 1005 1006 tp->t_inpcb = inp; 1007 tp->t_mtudisc = ip_mtudisc; 1008 break; 1009 } 1010 #ifdef INET6 1011 case AF_INET6: 1012 { 1013 struct in6pcb *in6p = (struct in6pcb *)aux; 1014 1015 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1016 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1017 : NULL); 1018 in6p->in6p_ppcb = (caddr_t)tp; 1019 1020 tp->t_in6pcb = in6p; 1021 /* for IPv6, always try to run path MTU discovery */ 1022 tp->t_mtudisc = 1; 1023 break; 1024 } 1025 #endif /* INET6 */ 1026 default: 1027 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1028 return (NULL); 1029 } 1030 1031 /* 1032 * Initialize our timebase. When we send timestamps, we take 1033 * the delta from tcp_now -- this means each connection always 1034 * gets a timebase of 0, which makes it, among other things, 1035 * more difficult to determine how long a system has been up, 1036 * and thus how many TCP sequence increments have occurred. 1037 */ 1038 tp->ts_timebase = tcp_now; 1039 1040 tp->t_congctl = tcp_congctl_global; 1041 tp->t_congctl->refcnt++; 1042 1043 return (tp); 1044 } 1045 1046 /* 1047 * Drop a TCP connection, reporting 1048 * the specified error. If connection is synchronized, 1049 * then send a RST to peer. 1050 */ 1051 struct tcpcb * 1052 tcp_drop(struct tcpcb *tp, int errno) 1053 { 1054 struct socket *so = NULL; 1055 1056 #ifdef DIAGNOSTIC 1057 if (tp->t_inpcb && tp->t_in6pcb) 1058 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1059 #endif 1060 #ifdef INET 1061 if (tp->t_inpcb) 1062 so = tp->t_inpcb->inp_socket; 1063 #endif 1064 #ifdef INET6 1065 if (tp->t_in6pcb) 1066 so = tp->t_in6pcb->in6p_socket; 1067 #endif 1068 if (!so) 1069 return NULL; 1070 1071 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1072 tp->t_state = TCPS_CLOSED; 1073 (void) tcp_output(tp); 1074 tcpstat.tcps_drops++; 1075 } else 1076 tcpstat.tcps_conndrops++; 1077 if (errno == ETIMEDOUT && tp->t_softerror) 1078 errno = tp->t_softerror; 1079 so->so_error = errno; 1080 return (tcp_close(tp)); 1081 } 1082 1083 /* 1084 * Return whether this tcpcb is marked as dead, indicating 1085 * to the calling timer function that no further action should 1086 * be taken, as we are about to release this tcpcb. The release 1087 * of the storage will be done if this is the last timer running. 1088 * 1089 * This should be called from the callout handler function after 1090 * callout_ack() is done, so that the number of invoking timer 1091 * functions is 0. 1092 */ 1093 int 1094 tcp_isdead(struct tcpcb *tp) 1095 { 1096 int dead = (tp->t_flags & TF_DEAD); 1097 1098 if (__predict_false(dead)) { 1099 if (tcp_timers_invoking(tp) > 0) 1100 /* not quite there yet -- count separately? */ 1101 return dead; 1102 tcpstat.tcps_delayed_free++; 1103 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_timer.c */ 1104 } 1105 return dead; 1106 } 1107 1108 /* 1109 * Close a TCP control block: 1110 * discard all space held by the tcp 1111 * discard internet protocol block 1112 * wake up any sleepers 1113 */ 1114 struct tcpcb * 1115 tcp_close(struct tcpcb *tp) 1116 { 1117 struct inpcb *inp; 1118 #ifdef INET6 1119 struct in6pcb *in6p; 1120 #endif 1121 struct socket *so; 1122 #ifdef RTV_RTT 1123 struct rtentry *rt; 1124 #endif 1125 struct route *ro; 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 = ro->ro_rt) && 1159 !in_nullhost(satosin(rt_key(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 1220 tp->t_congctl->refcnt--; 1221 1222 tcp_canceltimers(tp); 1223 TCP_CLEAR_DELACK(tp); 1224 syn_cache_cleanup(tp); 1225 1226 if (tp->t_template) { 1227 m_free(tp->t_template); 1228 tp->t_template = NULL; 1229 } 1230 if (tcp_timers_invoking(tp)) 1231 tp->t_flags |= TF_DEAD; 1232 else 1233 pool_put(&tcpcb_pool, tp); 1234 1235 if (inp) { 1236 inp->inp_ppcb = 0; 1237 soisdisconnected(so); 1238 in_pcbdetach(inp); 1239 } 1240 #ifdef INET6 1241 else if (in6p) { 1242 in6p->in6p_ppcb = 0; 1243 soisdisconnected(so); 1244 in6_pcbdetach(in6p); 1245 } 1246 #endif 1247 tcpstat.tcps_closed++; 1248 return ((struct tcpcb *)0); 1249 } 1250 1251 int 1252 tcp_freeq(tp) 1253 struct tcpcb *tp; 1254 { 1255 struct ipqent *qe; 1256 int rv = 0; 1257 #ifdef TCPREASS_DEBUG 1258 int i = 0; 1259 #endif 1260 1261 TCP_REASS_LOCK_CHECK(tp); 1262 1263 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1264 #ifdef TCPREASS_DEBUG 1265 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1266 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1267 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1268 #endif 1269 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1270 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1271 m_freem(qe->ipqe_m); 1272 tcpipqent_free(qe); 1273 rv = 1; 1274 } 1275 tp->t_segqlen = 0; 1276 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1277 return (rv); 1278 } 1279 1280 /* 1281 * Protocol drain routine. Called when memory is in short supply. 1282 */ 1283 void 1284 tcp_drain(void) 1285 { 1286 struct inpcb_hdr *inph; 1287 struct tcpcb *tp; 1288 1289 /* 1290 * Free the sequence queue of all TCP connections. 1291 */ 1292 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1293 switch (inph->inph_af) { 1294 case AF_INET: 1295 tp = intotcpcb((struct inpcb *)inph); 1296 break; 1297 #ifdef INET6 1298 case AF_INET6: 1299 tp = in6totcpcb((struct in6pcb *)inph); 1300 break; 1301 #endif 1302 default: 1303 tp = NULL; 1304 break; 1305 } 1306 if (tp != NULL) { 1307 /* 1308 * We may be called from a device's interrupt 1309 * context. If the tcpcb is already busy, 1310 * just bail out now. 1311 */ 1312 if (tcp_reass_lock_try(tp) == 0) 1313 continue; 1314 if (tcp_freeq(tp)) 1315 tcpstat.tcps_connsdrained++; 1316 TCP_REASS_UNLOCK(tp); 1317 } 1318 } 1319 } 1320 1321 /* 1322 * Notify a tcp user of an asynchronous error; 1323 * store error as soft error, but wake up user 1324 * (for now, won't do anything until can select for soft error). 1325 */ 1326 void 1327 tcp_notify(struct inpcb *inp, int error) 1328 { 1329 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1330 struct socket *so = inp->inp_socket; 1331 1332 /* 1333 * Ignore some errors if we are hooked up. 1334 * If connection hasn't completed, has retransmitted several times, 1335 * and receives a second error, give up now. This is better 1336 * than waiting a long time to establish a connection that 1337 * can never complete. 1338 */ 1339 if (tp->t_state == TCPS_ESTABLISHED && 1340 (error == EHOSTUNREACH || error == ENETUNREACH || 1341 error == EHOSTDOWN)) { 1342 return; 1343 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1344 tp->t_rxtshift > 3 && tp->t_softerror) 1345 so->so_error = error; 1346 else 1347 tp->t_softerror = error; 1348 wakeup((caddr_t) &so->so_timeo); 1349 sorwakeup(so); 1350 sowwakeup(so); 1351 } 1352 1353 #ifdef INET6 1354 void 1355 tcp6_notify(struct in6pcb *in6p, int error) 1356 { 1357 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1358 struct socket *so = in6p->in6p_socket; 1359 1360 /* 1361 * Ignore some errors if we are hooked up. 1362 * If connection hasn't completed, has retransmitted several times, 1363 * and receives a second error, give up now. This is better 1364 * than waiting a long time to establish a connection that 1365 * can never complete. 1366 */ 1367 if (tp->t_state == TCPS_ESTABLISHED && 1368 (error == EHOSTUNREACH || error == ENETUNREACH || 1369 error == EHOSTDOWN)) { 1370 return; 1371 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1372 tp->t_rxtshift > 3 && tp->t_softerror) 1373 so->so_error = error; 1374 else 1375 tp->t_softerror = error; 1376 wakeup((caddr_t) &so->so_timeo); 1377 sorwakeup(so); 1378 sowwakeup(so); 1379 } 1380 #endif 1381 1382 #ifdef INET6 1383 void 1384 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1385 { 1386 struct tcphdr th; 1387 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1388 int nmatch; 1389 struct ip6_hdr *ip6; 1390 const struct sockaddr_in6 *sa6_src = NULL; 1391 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1392 struct mbuf *m; 1393 int off; 1394 1395 if (sa->sa_family != AF_INET6 || 1396 sa->sa_len != sizeof(struct sockaddr_in6)) 1397 return; 1398 if ((unsigned)cmd >= PRC_NCMDS) 1399 return; 1400 else if (cmd == PRC_QUENCH) { 1401 /* 1402 * Don't honor ICMP Source Quench messages meant for 1403 * TCP connections. 1404 */ 1405 return; 1406 } else if (PRC_IS_REDIRECT(cmd)) 1407 notify = in6_rtchange, d = NULL; 1408 else if (cmd == PRC_MSGSIZE) 1409 ; /* special code is present, see below */ 1410 else if (cmd == PRC_HOSTDEAD) 1411 d = NULL; 1412 else if (inet6ctlerrmap[cmd] == 0) 1413 return; 1414 1415 /* if the parameter is from icmp6, decode it. */ 1416 if (d != NULL) { 1417 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1418 m = ip6cp->ip6c_m; 1419 ip6 = ip6cp->ip6c_ip6; 1420 off = ip6cp->ip6c_off; 1421 sa6_src = ip6cp->ip6c_src; 1422 } else { 1423 m = NULL; 1424 ip6 = NULL; 1425 sa6_src = &sa6_any; 1426 off = 0; 1427 } 1428 1429 if (ip6) { 1430 /* 1431 * XXX: We assume that when ip6 is non NULL, 1432 * M and OFF are valid. 1433 */ 1434 1435 /* check if we can safely examine src and dst ports */ 1436 if (m->m_pkthdr.len < off + sizeof(th)) { 1437 if (cmd == PRC_MSGSIZE) 1438 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1439 return; 1440 } 1441 1442 bzero(&th, sizeof(th)); 1443 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1444 1445 if (cmd == PRC_MSGSIZE) { 1446 int valid = 0; 1447 1448 /* 1449 * Check to see if we have a valid TCP connection 1450 * corresponding to the address in the ICMPv6 message 1451 * payload. 1452 */ 1453 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1454 th.th_dport, (const struct in6_addr *)&sa6_src->sin6_addr, 1455 th.th_sport, 0)) 1456 valid++; 1457 1458 /* 1459 * Depending on the value of "valid" and routing table 1460 * size (mtudisc_{hi,lo}wat), we will: 1461 * - recalcurate the new MTU and create the 1462 * corresponding routing entry, or 1463 * - ignore the MTU change notification. 1464 */ 1465 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1466 1467 /* 1468 * no need to call in6_pcbnotify, it should have been 1469 * called via callback if necessary 1470 */ 1471 return; 1472 } 1473 1474 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1475 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1476 if (nmatch == 0 && syn_cache_count && 1477 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1478 inet6ctlerrmap[cmd] == ENETUNREACH || 1479 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1480 syn_cache_unreach((const struct sockaddr *)sa6_src, 1481 sa, &th); 1482 } else { 1483 (void) in6_pcbnotify(&tcbtable, sa, 0, 1484 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1485 } 1486 } 1487 #endif 1488 1489 #ifdef INET 1490 /* assumes that ip header and tcp header are contiguous on mbuf */ 1491 void * 1492 tcp_ctlinput(int cmd, struct sockaddr *sa, void *v) 1493 { 1494 struct ip *ip = v; 1495 struct tcphdr *th; 1496 struct icmp *icp; 1497 extern const int inetctlerrmap[]; 1498 void (*notify)(struct inpcb *, int) = tcp_notify; 1499 int errno; 1500 int nmatch; 1501 struct tcpcb *tp; 1502 u_int mtu; 1503 tcp_seq seq; 1504 struct inpcb *inp; 1505 #ifdef INET6 1506 struct in6pcb *in6p; 1507 struct in6_addr src6, dst6; 1508 #endif 1509 1510 if (sa->sa_family != AF_INET || 1511 sa->sa_len != sizeof(struct sockaddr_in)) 1512 return NULL; 1513 if ((unsigned)cmd >= PRC_NCMDS) 1514 return NULL; 1515 errno = inetctlerrmap[cmd]; 1516 if (cmd == PRC_QUENCH) 1517 /* 1518 * Don't honor ICMP Source Quench messages meant for 1519 * TCP connections. 1520 */ 1521 return NULL; 1522 else if (PRC_IS_REDIRECT(cmd)) 1523 notify = in_rtchange, ip = 0; 1524 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1525 /* 1526 * Check to see if we have a valid TCP connection 1527 * corresponding to the address in the ICMP message 1528 * payload. 1529 * 1530 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1531 */ 1532 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1533 #ifdef INET6 1534 memset(&src6, 0, sizeof(src6)); 1535 memset(&dst6, 0, sizeof(dst6)); 1536 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1537 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1538 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1539 #endif 1540 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1541 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1542 #ifdef INET6 1543 in6p = NULL; 1544 #else 1545 ; 1546 #endif 1547 #ifdef INET6 1548 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1549 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1550 ; 1551 #endif 1552 else 1553 return NULL; 1554 1555 /* 1556 * Now that we've validated that we are actually communicating 1557 * with the host indicated in the ICMP message, locate the 1558 * ICMP header, recalculate the new MTU, and create the 1559 * corresponding routing entry. 1560 */ 1561 icp = (struct icmp *)((caddr_t)ip - 1562 offsetof(struct icmp, icmp_ip)); 1563 if (inp) { 1564 if ((tp = intotcpcb(inp)) == NULL) 1565 return NULL; 1566 } 1567 #ifdef INET6 1568 else if (in6p) { 1569 if ((tp = in6totcpcb(in6p)) == NULL) 1570 return NULL; 1571 } 1572 #endif 1573 else 1574 return NULL; 1575 seq = ntohl(th->th_seq); 1576 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1577 return NULL; 1578 /* 1579 * If the ICMP message advertises a Next-Hop MTU 1580 * equal or larger than the maximum packet size we have 1581 * ever sent, drop the message. 1582 */ 1583 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1584 if (mtu >= tp->t_pmtud_mtu_sent) 1585 return NULL; 1586 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1587 /* 1588 * Calculate new MTU, and create corresponding 1589 * route (traditional PMTUD). 1590 */ 1591 tp->t_flags &= ~TF_PMTUD_PEND; 1592 icmp_mtudisc(icp, ip->ip_dst); 1593 } else { 1594 /* 1595 * Record the information got in the ICMP 1596 * message; act on it later. 1597 * If we had already recorded an ICMP message, 1598 * replace the old one only if the new message 1599 * refers to an older TCP segment 1600 */ 1601 if (tp->t_flags & TF_PMTUD_PEND) { 1602 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1603 return NULL; 1604 } else 1605 tp->t_flags |= TF_PMTUD_PEND; 1606 tp->t_pmtud_th_seq = seq; 1607 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1608 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1609 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1610 } 1611 return NULL; 1612 } else if (cmd == PRC_HOSTDEAD) 1613 ip = 0; 1614 else if (errno == 0) 1615 return NULL; 1616 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1617 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1618 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1619 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1620 if (nmatch == 0 && syn_cache_count && 1621 (inetctlerrmap[cmd] == EHOSTUNREACH || 1622 inetctlerrmap[cmd] == ENETUNREACH || 1623 inetctlerrmap[cmd] == EHOSTDOWN)) { 1624 struct sockaddr_in sin; 1625 bzero(&sin, sizeof(sin)); 1626 sin.sin_len = sizeof(sin); 1627 sin.sin_family = AF_INET; 1628 sin.sin_port = th->th_sport; 1629 sin.sin_addr = ip->ip_src; 1630 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1631 } 1632 1633 /* XXX mapped address case */ 1634 } else 1635 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1636 notify); 1637 return NULL; 1638 } 1639 1640 /* 1641 * When a source quench is received, we are being notified of congestion. 1642 * Close the congestion window down to the Loss Window (one segment). 1643 * We will gradually open it again as we proceed. 1644 */ 1645 void 1646 tcp_quench(struct inpcb *inp, int errno __unused) 1647 { 1648 struct tcpcb *tp = intotcpcb(inp); 1649 1650 if (tp) 1651 tp->snd_cwnd = tp->t_segsz; 1652 } 1653 #endif 1654 1655 #ifdef INET6 1656 void 1657 tcp6_quench(struct in6pcb *in6p, int errno __unused) 1658 { 1659 struct tcpcb *tp = in6totcpcb(in6p); 1660 1661 if (tp) 1662 tp->snd_cwnd = tp->t_segsz; 1663 } 1664 #endif 1665 1666 #ifdef INET 1667 /* 1668 * Path MTU Discovery handlers. 1669 */ 1670 void 1671 tcp_mtudisc_callback(struct in_addr faddr) 1672 { 1673 #ifdef INET6 1674 struct in6_addr in6; 1675 #endif 1676 1677 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1678 #ifdef INET6 1679 memset(&in6, 0, sizeof(in6)); 1680 in6.s6_addr16[5] = 0xffff; 1681 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1682 tcp6_mtudisc_callback(&in6); 1683 #endif 1684 } 1685 1686 /* 1687 * On receipt of path MTU corrections, flush old route and replace it 1688 * with the new one. Retransmit all unacknowledged packets, to ensure 1689 * that all packets will be received. 1690 */ 1691 void 1692 tcp_mtudisc(struct inpcb *inp, int errno) 1693 { 1694 struct tcpcb *tp = intotcpcb(inp); 1695 struct rtentry *rt = in_pcbrtentry(inp); 1696 1697 if (tp != 0) { 1698 if (rt != 0) { 1699 /* 1700 * If this was not a host route, remove and realloc. 1701 */ 1702 if ((rt->rt_flags & RTF_HOST) == 0) { 1703 in_rtchange(inp, errno); 1704 if ((rt = in_pcbrtentry(inp)) == 0) 1705 return; 1706 } 1707 1708 /* 1709 * Slow start out of the error condition. We 1710 * use the MTU because we know it's smaller 1711 * than the previously transmitted segment. 1712 * 1713 * Note: This is more conservative than the 1714 * suggestion in draft-floyd-incr-init-win-03. 1715 */ 1716 if (rt->rt_rmx.rmx_mtu != 0) 1717 tp->snd_cwnd = 1718 TCP_INITIAL_WINDOW(tcp_init_win, 1719 rt->rt_rmx.rmx_mtu); 1720 } 1721 1722 /* 1723 * Resend unacknowledged packets. 1724 */ 1725 tp->snd_nxt = tp->snd_una; 1726 tcp_output(tp); 1727 } 1728 } 1729 #endif 1730 1731 #ifdef INET6 1732 /* 1733 * Path MTU Discovery handlers. 1734 */ 1735 void 1736 tcp6_mtudisc_callback(struct in6_addr *faddr) 1737 { 1738 struct sockaddr_in6 sin6; 1739 1740 bzero(&sin6, sizeof(sin6)); 1741 sin6.sin6_family = AF_INET6; 1742 sin6.sin6_len = sizeof(struct sockaddr_in6); 1743 sin6.sin6_addr = *faddr; 1744 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1745 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1746 } 1747 1748 void 1749 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1750 { 1751 struct tcpcb *tp = in6totcpcb(in6p); 1752 struct rtentry *rt = in6_pcbrtentry(in6p); 1753 1754 if (tp != 0) { 1755 if (rt != 0) { 1756 /* 1757 * If this was not a host route, remove and realloc. 1758 */ 1759 if ((rt->rt_flags & RTF_HOST) == 0) { 1760 in6_rtchange(in6p, errno); 1761 if ((rt = in6_pcbrtentry(in6p)) == 0) 1762 return; 1763 } 1764 1765 /* 1766 * Slow start out of the error condition. We 1767 * use the MTU because we know it's smaller 1768 * than the previously transmitted segment. 1769 * 1770 * Note: This is more conservative than the 1771 * suggestion in draft-floyd-incr-init-win-03. 1772 */ 1773 if (rt->rt_rmx.rmx_mtu != 0) 1774 tp->snd_cwnd = 1775 TCP_INITIAL_WINDOW(tcp_init_win, 1776 rt->rt_rmx.rmx_mtu); 1777 } 1778 1779 /* 1780 * Resend unacknowledged packets. 1781 */ 1782 tp->snd_nxt = tp->snd_una; 1783 tcp_output(tp); 1784 } 1785 } 1786 #endif /* INET6 */ 1787 1788 /* 1789 * Compute the MSS to advertise to the peer. Called only during 1790 * the 3-way handshake. If we are the server (peer initiated 1791 * connection), we are called with a pointer to the interface 1792 * on which the SYN packet arrived. If we are the client (we 1793 * initiated connection), we are called with a pointer to the 1794 * interface out which this connection should go. 1795 * 1796 * NOTE: Do not subtract IP option/extension header size nor IPsec 1797 * header size from MSS advertisement. MSS option must hold the maximum 1798 * segment size we can accept, so it must always be: 1799 * max(if mtu) - ip header - tcp header 1800 */ 1801 u_long 1802 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1803 { 1804 extern u_long in_maxmtu; 1805 u_long mss = 0; 1806 u_long hdrsiz; 1807 1808 /* 1809 * In order to avoid defeating path MTU discovery on the peer, 1810 * we advertise the max MTU of all attached networks as our MSS, 1811 * per RFC 1191, section 3.1. 1812 * 1813 * We provide the option to advertise just the MTU of 1814 * the interface on which we hope this connection will 1815 * be receiving. If we are responding to a SYN, we 1816 * will have a pretty good idea about this, but when 1817 * initiating a connection there is a bit more doubt. 1818 * 1819 * We also need to ensure that loopback has a large enough 1820 * MSS, as the loopback MTU is never included in in_maxmtu. 1821 */ 1822 1823 if (ifp != NULL) 1824 switch (af) { 1825 case AF_INET: 1826 mss = ifp->if_mtu; 1827 break; 1828 #ifdef INET6 1829 case AF_INET6: 1830 mss = IN6_LINKMTU(ifp); 1831 break; 1832 #endif 1833 } 1834 1835 if (tcp_mss_ifmtu == 0) 1836 switch (af) { 1837 case AF_INET: 1838 mss = max(in_maxmtu, mss); 1839 break; 1840 #ifdef INET6 1841 case AF_INET6: 1842 mss = max(in6_maxmtu, mss); 1843 break; 1844 #endif 1845 } 1846 1847 switch (af) { 1848 case AF_INET: 1849 hdrsiz = sizeof(struct ip); 1850 break; 1851 #ifdef INET6 1852 case AF_INET6: 1853 hdrsiz = sizeof(struct ip6_hdr); 1854 break; 1855 #endif 1856 default: 1857 hdrsiz = 0; 1858 break; 1859 } 1860 hdrsiz += sizeof(struct tcphdr); 1861 if (mss > hdrsiz) 1862 mss -= hdrsiz; 1863 1864 mss = max(tcp_mssdflt, mss); 1865 return (mss); 1866 } 1867 1868 /* 1869 * Set connection variables based on the peer's advertised MSS. 1870 * We are passed the TCPCB for the actual connection. If we 1871 * are the server, we are called by the compressed state engine 1872 * when the 3-way handshake is complete. If we are the client, 1873 * we are called when we receive the SYN,ACK from the server. 1874 * 1875 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1876 * before this routine is called! 1877 */ 1878 void 1879 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1880 { 1881 struct socket *so; 1882 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1883 struct rtentry *rt; 1884 #endif 1885 u_long bufsize; 1886 int mss; 1887 1888 #ifdef DIAGNOSTIC 1889 if (tp->t_inpcb && tp->t_in6pcb) 1890 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1891 #endif 1892 so = NULL; 1893 rt = NULL; 1894 #ifdef INET 1895 if (tp->t_inpcb) { 1896 so = tp->t_inpcb->inp_socket; 1897 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1898 rt = in_pcbrtentry(tp->t_inpcb); 1899 #endif 1900 } 1901 #endif 1902 #ifdef INET6 1903 if (tp->t_in6pcb) { 1904 so = tp->t_in6pcb->in6p_socket; 1905 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1906 rt = in6_pcbrtentry(tp->t_in6pcb); 1907 #endif 1908 } 1909 #endif 1910 1911 /* 1912 * As per RFC1122, use the default MSS value, unless they 1913 * sent us an offer. Do not accept offers less than 256 bytes. 1914 */ 1915 mss = tcp_mssdflt; 1916 if (offer) 1917 mss = offer; 1918 mss = max(mss, 256); /* sanity */ 1919 tp->t_peermss = mss; 1920 mss -= tcp_optlen(tp); 1921 #ifdef INET 1922 if (tp->t_inpcb) 1923 mss -= ip_optlen(tp->t_inpcb); 1924 #endif 1925 #ifdef INET6 1926 if (tp->t_in6pcb) 1927 mss -= ip6_optlen(tp->t_in6pcb); 1928 #endif 1929 1930 /* 1931 * If there's a pipesize, change the socket buffer to that size. 1932 * Make the socket buffer an integral number of MSS units. If 1933 * the MSS is larger than the socket buffer, artificially decrease 1934 * the MSS. 1935 */ 1936 #ifdef RTV_SPIPE 1937 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1938 bufsize = rt->rt_rmx.rmx_sendpipe; 1939 else 1940 #endif 1941 { 1942 KASSERT(so != NULL); 1943 bufsize = so->so_snd.sb_hiwat; 1944 } 1945 if (bufsize < mss) 1946 mss = bufsize; 1947 else { 1948 bufsize = roundup(bufsize, mss); 1949 if (bufsize > sb_max) 1950 bufsize = sb_max; 1951 (void) sbreserve(&so->so_snd, bufsize, so); 1952 } 1953 tp->t_segsz = mss; 1954 1955 #ifdef RTV_SSTHRESH 1956 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1957 /* 1958 * There's some sort of gateway or interface buffer 1959 * limit on the path. Use this to set the slow 1960 * start threshold, but set the threshold to no less 1961 * than 2 * MSS. 1962 */ 1963 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1964 } 1965 #endif 1966 } 1967 1968 /* 1969 * Processing necessary when a TCP connection is established. 1970 */ 1971 void 1972 tcp_established(struct tcpcb *tp) 1973 { 1974 struct socket *so; 1975 #ifdef RTV_RPIPE 1976 struct rtentry *rt; 1977 #endif 1978 u_long bufsize; 1979 1980 #ifdef DIAGNOSTIC 1981 if (tp->t_inpcb && tp->t_in6pcb) 1982 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1983 #endif 1984 so = NULL; 1985 rt = NULL; 1986 #ifdef INET 1987 if (tp->t_inpcb) { 1988 so = tp->t_inpcb->inp_socket; 1989 #if defined(RTV_RPIPE) 1990 rt = in_pcbrtentry(tp->t_inpcb); 1991 #endif 1992 } 1993 #endif 1994 #ifdef INET6 1995 if (tp->t_in6pcb) { 1996 so = tp->t_in6pcb->in6p_socket; 1997 #if defined(RTV_RPIPE) 1998 rt = in6_pcbrtentry(tp->t_in6pcb); 1999 #endif 2000 } 2001 #endif 2002 2003 tp->t_state = TCPS_ESTABLISHED; 2004 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 2005 2006 #ifdef RTV_RPIPE 2007 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2008 bufsize = rt->rt_rmx.rmx_recvpipe; 2009 else 2010 #endif 2011 { 2012 KASSERT(so != NULL); 2013 bufsize = so->so_rcv.sb_hiwat; 2014 } 2015 if (bufsize > tp->t_ourmss) { 2016 bufsize = roundup(bufsize, tp->t_ourmss); 2017 if (bufsize > sb_max) 2018 bufsize = sb_max; 2019 (void) sbreserve(&so->so_rcv, bufsize, so); 2020 } 2021 } 2022 2023 /* 2024 * Check if there's an initial rtt or rttvar. Convert from the 2025 * route-table units to scaled multiples of the slow timeout timer. 2026 * Called only during the 3-way handshake. 2027 */ 2028 void 2029 tcp_rmx_rtt(struct tcpcb *tp) 2030 { 2031 #ifdef RTV_RTT 2032 struct rtentry *rt = NULL; 2033 int rtt; 2034 2035 #ifdef DIAGNOSTIC 2036 if (tp->t_inpcb && tp->t_in6pcb) 2037 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2038 #endif 2039 #ifdef INET 2040 if (tp->t_inpcb) 2041 rt = in_pcbrtentry(tp->t_inpcb); 2042 #endif 2043 #ifdef INET6 2044 if (tp->t_in6pcb) 2045 rt = in6_pcbrtentry(tp->t_in6pcb); 2046 #endif 2047 if (rt == NULL) 2048 return; 2049 2050 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2051 /* 2052 * XXX The lock bit for MTU indicates that the value 2053 * is also a minimum value; this is subject to time. 2054 */ 2055 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2056 TCPT_RANGESET(tp->t_rttmin, 2057 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2058 TCPTV_MIN, TCPTV_REXMTMAX); 2059 tp->t_srtt = rtt / 2060 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2061 if (rt->rt_rmx.rmx_rttvar) { 2062 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2063 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2064 (TCP_RTTVAR_SHIFT + 2)); 2065 } else { 2066 /* Default variation is +- 1 rtt */ 2067 tp->t_rttvar = 2068 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2069 } 2070 TCPT_RANGESET(tp->t_rxtcur, 2071 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2072 tp->t_rttmin, TCPTV_REXMTMAX); 2073 } 2074 #endif 2075 } 2076 2077 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2078 #if NRND > 0 2079 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2080 #endif 2081 2082 /* 2083 * Get a new sequence value given a tcp control block 2084 */ 2085 tcp_seq 2086 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2087 { 2088 2089 #ifdef INET 2090 if (tp->t_inpcb != NULL) { 2091 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2092 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2093 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2094 addin)); 2095 } 2096 #endif 2097 #ifdef INET6 2098 if (tp->t_in6pcb != NULL) { 2099 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2100 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2101 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2102 addin)); 2103 } 2104 #endif 2105 /* Not possible. */ 2106 panic("tcp_new_iss"); 2107 } 2108 2109 /* 2110 * This routine actually generates a new TCP initial sequence number. 2111 */ 2112 tcp_seq 2113 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2114 size_t addrsz, tcp_seq addin) 2115 { 2116 tcp_seq tcp_iss; 2117 2118 #if NRND > 0 2119 static int beenhere; 2120 2121 /* 2122 * If we haven't been here before, initialize our cryptographic 2123 * hash secret. 2124 */ 2125 if (beenhere == 0) { 2126 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2127 RND_EXTRACT_ANY); 2128 beenhere = 1; 2129 } 2130 2131 if (tcp_do_rfc1948) { 2132 MD5_CTX ctx; 2133 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2134 2135 /* 2136 * Compute the base value of the ISS. It is a hash 2137 * of (saddr, sport, daddr, dport, secret). 2138 */ 2139 MD5Init(&ctx); 2140 2141 MD5Update(&ctx, (u_char *) laddr, addrsz); 2142 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2143 2144 MD5Update(&ctx, (u_char *) faddr, addrsz); 2145 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2146 2147 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2148 2149 MD5Final(hash, &ctx); 2150 2151 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2152 2153 /* 2154 * Now increment our "timer", and add it in to 2155 * the computed value. 2156 * 2157 * XXX Use `addin'? 2158 * XXX TCP_ISSINCR too large to use? 2159 */ 2160 tcp_iss_seq += TCP_ISSINCR; 2161 #ifdef TCPISS_DEBUG 2162 printf("ISS hash 0x%08x, ", tcp_iss); 2163 #endif 2164 tcp_iss += tcp_iss_seq + addin; 2165 #ifdef TCPISS_DEBUG 2166 printf("new ISS 0x%08x\n", tcp_iss); 2167 #endif 2168 } else 2169 #else 2170 /* for -Wunused */ 2171 do { if (&laddr) {} } while (/* CONSTCOND */ 0); 2172 do { if (&faddr) {} } while (/* CONSTCOND */ 0); 2173 do { if (&lport) {} } while (/* CONSTCOND */ 0); 2174 do { if (&fport) {} } while (/* CONSTCOND */ 0); 2175 do { if (&addrsz) {} } while (/* CONSTCOND */ 0); 2176 #endif /* NRND > 0 */ 2177 { 2178 /* 2179 * Randomize. 2180 */ 2181 #if NRND > 0 2182 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2183 #else 2184 tcp_iss = arc4random(); 2185 #endif 2186 2187 /* 2188 * If we were asked to add some amount to a known value, 2189 * we will take a random value obtained above, mask off 2190 * the upper bits, and add in the known value. We also 2191 * add in a constant to ensure that we are at least a 2192 * certain distance from the original value. 2193 * 2194 * This is used when an old connection is in timed wait 2195 * and we have a new one coming in, for instance. 2196 */ 2197 if (addin != 0) { 2198 #ifdef TCPISS_DEBUG 2199 printf("Random %08x, ", tcp_iss); 2200 #endif 2201 tcp_iss &= TCP_ISS_RANDOM_MASK; 2202 tcp_iss += addin + TCP_ISSINCR; 2203 #ifdef TCPISS_DEBUG 2204 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2205 #endif 2206 } else { 2207 tcp_iss &= TCP_ISS_RANDOM_MASK; 2208 tcp_iss += tcp_iss_seq; 2209 tcp_iss_seq += TCP_ISSINCR; 2210 #ifdef TCPISS_DEBUG 2211 printf("ISS %08x\n", tcp_iss); 2212 #endif 2213 } 2214 } 2215 2216 if (tcp_compat_42) { 2217 /* 2218 * Limit it to the positive range for really old TCP 2219 * implementations. 2220 * Just AND off the top bit instead of checking if 2221 * is set first - saves a branch 50% of the time. 2222 */ 2223 tcp_iss &= 0x7fffffff; /* XXX */ 2224 } 2225 2226 return (tcp_iss); 2227 } 2228 2229 #if defined(IPSEC) || defined(FAST_IPSEC) 2230 /* compute ESP/AH header size for TCP, including outer IP header. */ 2231 size_t 2232 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2233 { 2234 struct inpcb *inp; 2235 size_t hdrsiz; 2236 2237 /* XXX mapped addr case (tp->t_in6pcb) */ 2238 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2239 return 0; 2240 switch (tp->t_family) { 2241 case AF_INET: 2242 /* XXX: should use currect direction. */ 2243 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2244 break; 2245 default: 2246 hdrsiz = 0; 2247 break; 2248 } 2249 2250 return hdrsiz; 2251 } 2252 2253 #ifdef INET6 2254 size_t 2255 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2256 { 2257 struct in6pcb *in6p; 2258 size_t hdrsiz; 2259 2260 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2261 return 0; 2262 switch (tp->t_family) { 2263 case AF_INET6: 2264 /* XXX: should use currect direction. */ 2265 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2266 break; 2267 case AF_INET: 2268 /* mapped address case - tricky */ 2269 default: 2270 hdrsiz = 0; 2271 break; 2272 } 2273 2274 return hdrsiz; 2275 } 2276 #endif 2277 #endif /*IPSEC*/ 2278 2279 /* 2280 * Determine the length of the TCP options for this connection. 2281 * 2282 * XXX: What do we do for SACK, when we add that? Just reserve 2283 * all of the space? Otherwise we can't exactly be incrementing 2284 * cwnd by an amount that varies depending on the amount we last 2285 * had to SACK! 2286 */ 2287 2288 u_int 2289 tcp_optlen(struct tcpcb *tp) 2290 { 2291 u_int optlen; 2292 2293 optlen = 0; 2294 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2295 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2296 optlen += TCPOLEN_TSTAMP_APPA; 2297 2298 #ifdef TCP_SIGNATURE 2299 #if defined(INET6) && defined(FAST_IPSEC) 2300 if (tp->t_family == AF_INET) 2301 #endif 2302 if (tp->t_flags & TF_SIGNATURE) 2303 optlen += TCPOLEN_SIGNATURE + 2; 2304 #endif /* TCP_SIGNATURE */ 2305 2306 return optlen; 2307 } 2308 2309 u_int 2310 tcp_hdrsz(struct tcpcb *tp) 2311 { 2312 u_int hlen; 2313 2314 switch (tp->t_family) { 2315 #ifdef INET6 2316 case AF_INET6: 2317 hlen = sizeof(struct ip6_hdr); 2318 break; 2319 #endif 2320 case AF_INET: 2321 hlen = sizeof(struct ip); 2322 break; 2323 default: 2324 hlen = 0; 2325 break; 2326 } 2327 hlen += sizeof(struct tcphdr); 2328 2329 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2330 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2331 hlen += TCPOLEN_TSTAMP_APPA; 2332 #ifdef TCP_SIGNATURE 2333 if (tp->t_flags & TF_SIGNATURE) 2334 hlen += TCPOLEN_SIGLEN; 2335 #endif 2336 return hlen; 2337 } 2338