1 /* $NetBSD: tcp_subr.c,v 1.129 2002/05/28 10:17:27 itojun 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. All advertising materials mentioning features or use of this software 82 * must display the following acknowledgement: 83 * This product includes software developed by the University of 84 * California, Berkeley and its contributors. 85 * 4. Neither the name of the University nor the names of its contributors 86 * may be used to endorse or promote products derived from this software 87 * without specific prior written permission. 88 * 89 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 90 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 91 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 92 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 93 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 94 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 95 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 96 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 97 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 98 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 99 * SUCH DAMAGE. 100 * 101 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 102 */ 103 104 #include <sys/cdefs.h> 105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.129 2002/05/28 10:17:27 itojun Exp $"); 106 107 #include "opt_inet.h" 108 #include "opt_ipsec.h" 109 #include "opt_tcp_compat_42.h" 110 #include "opt_inet_csum.h" 111 #include "rnd.h" 112 113 #include <sys/param.h> 114 #include <sys/proc.h> 115 #include <sys/systm.h> 116 #include <sys/malloc.h> 117 #include <sys/mbuf.h> 118 #include <sys/socket.h> 119 #include <sys/socketvar.h> 120 #include <sys/protosw.h> 121 #include <sys/errno.h> 122 #include <sys/kernel.h> 123 #include <sys/pool.h> 124 #if NRND > 0 125 #include <sys/md5.h> 126 #include <sys/rnd.h> 127 #endif 128 129 #include <net/route.h> 130 #include <net/if.h> 131 132 #include <netinet/in.h> 133 #include <netinet/in_systm.h> 134 #include <netinet/ip.h> 135 #include <netinet/in_pcb.h> 136 #include <netinet/ip_var.h> 137 #include <netinet/ip_icmp.h> 138 139 #ifdef INET6 140 #ifndef INET 141 #include <netinet/in.h> 142 #endif 143 #include <netinet/ip6.h> 144 #include <netinet6/in6_pcb.h> 145 #include <netinet6/ip6_var.h> 146 #include <netinet6/in6_var.h> 147 #include <netinet6/ip6protosw.h> 148 #include <netinet/icmp6.h> 149 #endif 150 151 #include <netinet/tcp.h> 152 #include <netinet/tcp_fsm.h> 153 #include <netinet/tcp_seq.h> 154 #include <netinet/tcp_timer.h> 155 #include <netinet/tcp_var.h> 156 #include <netinet/tcpip.h> 157 158 #ifdef IPSEC 159 #include <netinet6/ipsec.h> 160 #endif /*IPSEC*/ 161 162 #ifdef INET6 163 struct in6pcb tcb6; 164 #endif 165 166 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 167 struct tcpstat tcpstat; /* tcp statistics */ 168 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 169 170 /* patchable/settable parameters for tcp */ 171 int tcp_mssdflt = TCP_MSS; 172 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 173 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 174 #if NRND > 0 175 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 176 #endif 177 int tcp_do_sack = 1; /* selective acknowledgement */ 178 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 179 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 180 int tcp_do_newreno = 0; /* Use the New Reno algorithms */ 181 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 182 int tcp_init_win = 1; 183 int tcp_mss_ifmtu = 0; 184 #ifdef TCP_COMPAT_42 185 int tcp_compat_42 = 1; 186 #else 187 int tcp_compat_42 = 0; 188 #endif 189 int tcp_rst_ppslim = 100; /* 100pps */ 190 191 /* tcb hash */ 192 #ifndef TCBHASHSIZE 193 #define TCBHASHSIZE 128 194 #endif 195 int tcbhashsize = TCBHASHSIZE; 196 197 /* syn hash parameters */ 198 #define TCP_SYN_HASH_SIZE 293 199 #define TCP_SYN_BUCKET_SIZE 35 200 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 201 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 202 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 203 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 204 205 int tcp_freeq __P((struct tcpcb *)); 206 207 #ifdef INET 208 void tcp_mtudisc_callback __P((struct in_addr)); 209 #endif 210 #ifdef INET6 211 void tcp6_mtudisc_callback __P((struct in6_addr *)); 212 #endif 213 214 void tcp_mtudisc __P((struct inpcb *, int)); 215 #ifdef INET6 216 void tcp6_mtudisc __P((struct in6pcb *, int)); 217 #endif 218 219 struct pool tcpcb_pool; 220 221 #ifdef TCP_CSUM_COUNTERS 222 #include <sys/device.h> 223 224 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 225 NULL, "tcp", "hwcsum bad"); 226 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 227 NULL, "tcp", "hwcsum ok"); 228 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 229 NULL, "tcp", "hwcsum data"); 230 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 231 NULL, "tcp", "swcsum"); 232 #endif /* TCP_CSUM_COUNTERS */ 233 234 #ifdef TCP_OUTPUT_COUNTERS 235 #include <sys/device.h> 236 237 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 238 NULL, "tcp", "output big header"); 239 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 240 NULL, "tcp", "output copy small"); 241 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "output copy big"); 243 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "output reference big"); 245 #endif /* TCP_OUTPUT_COUNTERS */ 246 247 #ifdef TCP_REASS_COUNTERS 248 #include <sys/device.h> 249 250 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 251 NULL, "tcp_reass", "calls"); 252 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 253 &tcp_reass_, "tcp_reass", "insert into empty queue"); 254 struct evcnt tcp_reass_iteration[8] = { 255 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 256 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 257 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 258 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 259 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 260 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 261 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 262 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 263 }; 264 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 265 &tcp_reass_, "tcp_reass", "prepend to first"); 266 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 &tcp_reass_, "tcp_reass", "prepend"); 268 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 269 &tcp_reass_, "tcp_reass", "insert"); 270 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 271 &tcp_reass_, "tcp_reass", "insert at tail"); 272 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 273 &tcp_reass_, "tcp_reass", "append"); 274 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 275 &tcp_reass_, "tcp_reass", "append to tail fragment"); 276 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 277 &tcp_reass_, "tcp_reass", "overlap at end"); 278 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 279 &tcp_reass_, "tcp_reass", "overlap at start"); 280 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 281 &tcp_reass_, "tcp_reass", "duplicate segment"); 282 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 &tcp_reass_, "tcp_reass", "duplicate fragment"); 284 285 #endif /* TCP_REASS_COUNTERS */ 286 287 /* 288 * Tcp initialization 289 */ 290 void 291 tcp_init() 292 { 293 int hlen; 294 295 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 296 NULL); 297 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 298 #ifdef INET6 299 tcb6.in6p_next = tcb6.in6p_prev = &tcb6; 300 #endif 301 302 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 303 #ifdef INET6 304 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 305 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 306 #endif 307 if (max_protohdr < hlen) 308 max_protohdr = hlen; 309 if (max_linkhdr + hlen > MHLEN) 310 panic("tcp_init"); 311 312 #ifdef INET 313 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 314 #endif 315 #ifdef INET6 316 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 317 #endif 318 319 /* Initialize timer state. */ 320 tcp_timer_init(); 321 322 /* Initialize the compressed state engine. */ 323 syn_cache_init(); 324 325 #ifdef TCP_CSUM_COUNTERS 326 evcnt_attach_static(&tcp_hwcsum_bad); 327 evcnt_attach_static(&tcp_hwcsum_ok); 328 evcnt_attach_static(&tcp_hwcsum_data); 329 evcnt_attach_static(&tcp_swcsum); 330 #endif /* TCP_CSUM_COUNTERS */ 331 332 #ifdef TCP_OUTPUT_COUNTERS 333 evcnt_attach_static(&tcp_output_bigheader); 334 evcnt_attach_static(&tcp_output_copysmall); 335 evcnt_attach_static(&tcp_output_copybig); 336 evcnt_attach_static(&tcp_output_refbig); 337 #endif /* TCP_OUTPUT_COUNTERS */ 338 339 #ifdef TCP_REASS_COUNTERS 340 evcnt_attach_static(&tcp_reass_); 341 evcnt_attach_static(&tcp_reass_empty); 342 evcnt_attach_static(&tcp_reass_iteration[0]); 343 evcnt_attach_static(&tcp_reass_iteration[1]); 344 evcnt_attach_static(&tcp_reass_iteration[2]); 345 evcnt_attach_static(&tcp_reass_iteration[3]); 346 evcnt_attach_static(&tcp_reass_iteration[4]); 347 evcnt_attach_static(&tcp_reass_iteration[5]); 348 evcnt_attach_static(&tcp_reass_iteration[6]); 349 evcnt_attach_static(&tcp_reass_iteration[7]); 350 evcnt_attach_static(&tcp_reass_prependfirst); 351 evcnt_attach_static(&tcp_reass_prepend); 352 evcnt_attach_static(&tcp_reass_insert); 353 evcnt_attach_static(&tcp_reass_inserttail); 354 evcnt_attach_static(&tcp_reass_append); 355 evcnt_attach_static(&tcp_reass_appendtail); 356 evcnt_attach_static(&tcp_reass_overlaptail); 357 evcnt_attach_static(&tcp_reass_overlapfront); 358 evcnt_attach_static(&tcp_reass_segdup); 359 evcnt_attach_static(&tcp_reass_fragdup); 360 #endif /* TCP_REASS_COUNTERS */ 361 } 362 363 /* 364 * Create template to be used to send tcp packets on a connection. 365 * Call after host entry created, allocates an mbuf and fills 366 * in a skeletal tcp/ip header, minimizing the amount of work 367 * necessary when the connection is used. 368 */ 369 struct mbuf * 370 tcp_template(tp) 371 struct tcpcb *tp; 372 { 373 struct inpcb *inp = tp->t_inpcb; 374 #ifdef INET6 375 struct in6pcb *in6p = tp->t_in6pcb; 376 #endif 377 struct tcphdr *n; 378 struct mbuf *m; 379 int hlen; 380 381 switch (tp->t_family) { 382 case AF_INET: 383 hlen = sizeof(struct ip); 384 if (inp) 385 break; 386 #ifdef INET6 387 if (in6p) { 388 /* mapped addr case */ 389 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 390 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 391 break; 392 } 393 #endif 394 return NULL; /*EINVAL*/ 395 #ifdef INET6 396 case AF_INET6: 397 hlen = sizeof(struct ip6_hdr); 398 if (in6p) { 399 /* more sainty check? */ 400 break; 401 } 402 return NULL; /*EINVAL*/ 403 #endif 404 default: 405 hlen = 0; /*pacify gcc*/ 406 return NULL; /*EAFNOSUPPORT*/ 407 } 408 #ifdef DIAGNOSTIC 409 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 410 panic("mclbytes too small for t_template"); 411 #endif 412 m = tp->t_template; 413 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 414 ; 415 else { 416 if (m) 417 m_freem(m); 418 m = tp->t_template = NULL; 419 MGETHDR(m, M_DONTWAIT, MT_HEADER); 420 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 421 MCLGET(m, M_DONTWAIT); 422 if ((m->m_flags & M_EXT) == 0) { 423 m_free(m); 424 m = NULL; 425 } 426 } 427 if (m == NULL) 428 return NULL; 429 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 430 } 431 432 bzero(mtod(m, caddr_t), m->m_len); 433 434 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 435 436 switch (tp->t_family) { 437 case AF_INET: 438 { 439 struct ipovly *ipov; 440 mtod(m, struct ip *)->ip_v = 4; 441 ipov = mtod(m, struct ipovly *); 442 ipov->ih_pr = IPPROTO_TCP; 443 ipov->ih_len = htons(sizeof(struct tcphdr)); 444 if (inp) { 445 ipov->ih_src = inp->inp_laddr; 446 ipov->ih_dst = inp->inp_faddr; 447 } 448 #ifdef INET6 449 else if (in6p) { 450 /* mapped addr case */ 451 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 452 sizeof(ipov->ih_src)); 453 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 454 sizeof(ipov->ih_dst)); 455 } 456 #endif 457 /* 458 * Compute the pseudo-header portion of the checksum 459 * now. We incrementally add in the TCP option and 460 * payload lengths later, and then compute the TCP 461 * checksum right before the packet is sent off onto 462 * the wire. 463 */ 464 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 465 ipov->ih_dst.s_addr, 466 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 467 break; 468 } 469 #ifdef INET6 470 case AF_INET6: 471 { 472 struct ip6_hdr *ip6; 473 mtod(m, struct ip *)->ip_v = 6; 474 ip6 = mtod(m, struct ip6_hdr *); 475 ip6->ip6_nxt = IPPROTO_TCP; 476 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 477 ip6->ip6_src = in6p->in6p_laddr; 478 ip6->ip6_dst = in6p->in6p_faddr; 479 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 480 if (ip6_auto_flowlabel) { 481 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 482 ip6->ip6_flow |= 483 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 484 } 485 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 486 ip6->ip6_vfc |= IPV6_VERSION; 487 488 /* 489 * Compute the pseudo-header portion of the checksum 490 * now. We incrementally add in the TCP option and 491 * payload lengths later, and then compute the TCP 492 * checksum right before the packet is sent off onto 493 * the wire. 494 */ 495 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 496 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 497 htonl(IPPROTO_TCP)); 498 break; 499 } 500 #endif 501 } 502 if (inp) { 503 n->th_sport = inp->inp_lport; 504 n->th_dport = inp->inp_fport; 505 } 506 #ifdef INET6 507 else if (in6p) { 508 n->th_sport = in6p->in6p_lport; 509 n->th_dport = in6p->in6p_fport; 510 } 511 #endif 512 n->th_seq = 0; 513 n->th_ack = 0; 514 n->th_x2 = 0; 515 n->th_off = 5; 516 n->th_flags = 0; 517 n->th_win = 0; 518 n->th_urp = 0; 519 return (m); 520 } 521 522 /* 523 * Send a single message to the TCP at address specified by 524 * the given TCP/IP header. If m == 0, then we make a copy 525 * of the tcpiphdr at ti and send directly to the addressed host. 526 * This is used to force keep alive messages out using the TCP 527 * template for a connection tp->t_template. If flags are given 528 * then we send a message back to the TCP which originated the 529 * segment ti, and discard the mbuf containing it and any other 530 * attached mbufs. 531 * 532 * In any case the ack and sequence number of the transmitted 533 * segment are as specified by the parameters. 534 */ 535 int 536 tcp_respond(tp, template, m, th0, ack, seq, flags) 537 struct tcpcb *tp; 538 struct mbuf *template; 539 struct mbuf *m; 540 struct tcphdr *th0; 541 tcp_seq ack, seq; 542 int flags; 543 { 544 struct route *ro; 545 int error, tlen, win = 0; 546 int hlen; 547 struct ip *ip; 548 #ifdef INET6 549 struct ip6_hdr *ip6; 550 #endif 551 int family; /* family on packet, not inpcb/in6pcb! */ 552 struct tcphdr *th; 553 554 if (tp != NULL && (flags & TH_RST) == 0) { 555 #ifdef DIAGNOSTIC 556 if (tp->t_inpcb && tp->t_in6pcb) 557 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 558 #endif 559 #ifdef INET 560 if (tp->t_inpcb) 561 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 562 #endif 563 #ifdef INET6 564 if (tp->t_in6pcb) 565 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 566 #endif 567 } 568 569 ip = NULL; 570 #ifdef INET6 571 ip6 = NULL; 572 #endif 573 if (m == 0) { 574 if (!template) 575 return EINVAL; 576 577 /* get family information from template */ 578 switch (mtod(template, struct ip *)->ip_v) { 579 case 4: 580 family = AF_INET; 581 hlen = sizeof(struct ip); 582 break; 583 #ifdef INET6 584 case 6: 585 family = AF_INET6; 586 hlen = sizeof(struct ip6_hdr); 587 break; 588 #endif 589 default: 590 return EAFNOSUPPORT; 591 } 592 593 MGETHDR(m, M_DONTWAIT, MT_HEADER); 594 if (m) { 595 MCLGET(m, M_DONTWAIT); 596 if ((m->m_flags & M_EXT) == 0) { 597 m_free(m); 598 m = NULL; 599 } 600 } 601 if (m == NULL) 602 return (ENOBUFS); 603 604 if (tcp_compat_42) 605 tlen = 1; 606 else 607 tlen = 0; 608 609 m->m_data += max_linkhdr; 610 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 611 template->m_len); 612 switch (family) { 613 case AF_INET: 614 ip = mtod(m, struct ip *); 615 th = (struct tcphdr *)(ip + 1); 616 break; 617 #ifdef INET6 618 case AF_INET6: 619 ip6 = mtod(m, struct ip6_hdr *); 620 th = (struct tcphdr *)(ip6 + 1); 621 break; 622 #endif 623 #if 0 624 default: 625 /* noone will visit here */ 626 m_freem(m); 627 return EAFNOSUPPORT; 628 #endif 629 } 630 flags = TH_ACK; 631 } else { 632 633 if ((m->m_flags & M_PKTHDR) == 0) { 634 #if 0 635 printf("non PKTHDR to tcp_respond\n"); 636 #endif 637 m_freem(m); 638 return EINVAL; 639 } 640 #ifdef DIAGNOSTIC 641 if (!th0) 642 panic("th0 == NULL in tcp_respond"); 643 #endif 644 645 /* get family information from m */ 646 switch (mtod(m, struct ip *)->ip_v) { 647 case 4: 648 family = AF_INET; 649 hlen = sizeof(struct ip); 650 ip = mtod(m, struct ip *); 651 break; 652 #ifdef INET6 653 case 6: 654 family = AF_INET6; 655 hlen = sizeof(struct ip6_hdr); 656 ip6 = mtod(m, struct ip6_hdr *); 657 break; 658 #endif 659 default: 660 m_freem(m); 661 return EAFNOSUPPORT; 662 } 663 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 664 tlen = sizeof(*th0); 665 else 666 tlen = th0->th_off << 2; 667 668 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 669 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 670 m->m_len = hlen + tlen; 671 m_freem(m->m_next); 672 m->m_next = NULL; 673 } else { 674 struct mbuf *n; 675 676 #ifdef DIAGNOSTIC 677 if (max_linkhdr + hlen + tlen > MCLBYTES) { 678 m_freem(m); 679 return EMSGSIZE; 680 } 681 #endif 682 MGETHDR(n, M_DONTWAIT, MT_HEADER); 683 if (n && max_linkhdr + hlen + tlen > MHLEN) { 684 MCLGET(n, M_DONTWAIT); 685 if ((n->m_flags & M_EXT) == 0) { 686 m_freem(n); 687 n = NULL; 688 } 689 } 690 if (!n) { 691 m_freem(m); 692 return ENOBUFS; 693 } 694 695 n->m_data += max_linkhdr; 696 n->m_len = hlen + tlen; 697 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 698 m_copyback(n, hlen, tlen, (caddr_t)th0); 699 700 m_freem(m); 701 m = n; 702 n = NULL; 703 } 704 705 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 706 switch (family) { 707 case AF_INET: 708 ip = mtod(m, struct ip *); 709 th = (struct tcphdr *)(ip + 1); 710 ip->ip_p = IPPROTO_TCP; 711 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 712 ip->ip_p = IPPROTO_TCP; 713 break; 714 #ifdef INET6 715 case AF_INET6: 716 ip6 = mtod(m, struct ip6_hdr *); 717 th = (struct tcphdr *)(ip6 + 1); 718 ip6->ip6_nxt = IPPROTO_TCP; 719 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 720 ip6->ip6_nxt = IPPROTO_TCP; 721 break; 722 #endif 723 #if 0 724 default: 725 /* noone will visit here */ 726 m_freem(m); 727 return EAFNOSUPPORT; 728 #endif 729 } 730 xchg(th->th_dport, th->th_sport, u_int16_t); 731 #undef xchg 732 tlen = 0; /*be friendly with the following code*/ 733 } 734 th->th_seq = htonl(seq); 735 th->th_ack = htonl(ack); 736 th->th_x2 = 0; 737 if ((flags & TH_SYN) == 0) { 738 if (tp) 739 win >>= tp->rcv_scale; 740 if (win > TCP_MAXWIN) 741 win = TCP_MAXWIN; 742 th->th_win = htons((u_int16_t)win); 743 th->th_off = sizeof (struct tcphdr) >> 2; 744 tlen += sizeof(*th); 745 } else 746 tlen += th->th_off << 2; 747 m->m_len = hlen + tlen; 748 m->m_pkthdr.len = hlen + tlen; 749 m->m_pkthdr.rcvif = (struct ifnet *) 0; 750 th->th_flags = flags; 751 th->th_urp = 0; 752 753 switch (family) { 754 #ifdef INET 755 case AF_INET: 756 { 757 struct ipovly *ipov = (struct ipovly *)ip; 758 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 759 ipov->ih_len = htons((u_int16_t)tlen); 760 761 th->th_sum = 0; 762 th->th_sum = in_cksum(m, hlen + tlen); 763 ip->ip_len = hlen + tlen; /*will be flipped on output*/ 764 ip->ip_ttl = ip_defttl; 765 break; 766 } 767 #endif 768 #ifdef INET6 769 case AF_INET6: 770 { 771 th->th_sum = 0; 772 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 773 tlen); 774 ip6->ip6_plen = ntohs(tlen); 775 if (tp && tp->t_in6pcb) { 776 struct ifnet *oifp; 777 ro = (struct route *)&tp->t_in6pcb->in6p_route; 778 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 779 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 780 } else 781 ip6->ip6_hlim = ip6_defhlim; 782 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 783 if (ip6_auto_flowlabel) { 784 ip6->ip6_flow |= 785 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 786 } 787 break; 788 } 789 #endif 790 } 791 792 #ifdef IPSEC 793 (void)ipsec_setsocket(m, NULL); 794 #endif /*IPSEC*/ 795 796 if (tp != NULL && tp->t_inpcb != NULL) { 797 ro = &tp->t_inpcb->inp_route; 798 #ifdef IPSEC 799 if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) { 800 m_freem(m); 801 return ENOBUFS; 802 } 803 #endif 804 #ifdef DIAGNOSTIC 805 if (family != AF_INET) 806 panic("tcp_respond: address family mismatch"); 807 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 808 panic("tcp_respond: ip_dst %x != inp_faddr %x", 809 ntohl(ip->ip_dst.s_addr), 810 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 811 } 812 #endif 813 } 814 #ifdef INET6 815 else if (tp != NULL && tp->t_in6pcb != NULL) { 816 ro = (struct route *)&tp->t_in6pcb->in6p_route; 817 #ifdef IPSEC 818 if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) { 819 m_freem(m); 820 return ENOBUFS; 821 } 822 #endif 823 #ifdef DIAGNOSTIC 824 if (family == AF_INET) { 825 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 826 panic("tcp_respond: not mapped addr"); 827 if (bcmp(&ip->ip_dst, 828 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 829 sizeof(ip->ip_dst)) != 0) { 830 panic("tcp_respond: ip_dst != in6p_faddr"); 831 } 832 } else if (family == AF_INET6) { 833 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr)) 834 panic("tcp_respond: ip6_dst != in6p_faddr"); 835 } else 836 panic("tcp_respond: address family mismatch"); 837 #endif 838 } 839 #endif 840 else 841 ro = NULL; 842 843 switch (family) { 844 #ifdef INET 845 case AF_INET: 846 error = ip_output(m, NULL, ro, 847 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 848 NULL); 849 break; 850 #endif 851 #ifdef INET6 852 case AF_INET6: 853 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL, 854 NULL); 855 break; 856 #endif 857 default: 858 error = EAFNOSUPPORT; 859 break; 860 } 861 862 return (error); 863 } 864 865 /* 866 * Create a new TCP control block, making an 867 * empty reassembly queue and hooking it to the argument 868 * protocol control block. 869 */ 870 struct tcpcb * 871 tcp_newtcpcb(family, aux) 872 int family; /* selects inpcb, or in6pcb */ 873 void *aux; 874 { 875 struct tcpcb *tp; 876 int i; 877 878 switch (family) { 879 case PF_INET: 880 break; 881 #ifdef INET6 882 case PF_INET6: 883 break; 884 #endif 885 default: 886 return NULL; 887 } 888 889 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 890 if (tp == NULL) 891 return (NULL); 892 bzero((caddr_t)tp, sizeof(struct tcpcb)); 893 TAILQ_INIT(&tp->segq); 894 TAILQ_INIT(&tp->timeq); 895 tp->t_family = family; /* may be overridden later on */ 896 tp->t_peermss = tcp_mssdflt; 897 tp->t_ourmss = tcp_mssdflt; 898 tp->t_segsz = tcp_mssdflt; 899 LIST_INIT(&tp->t_sc); 900 901 callout_init(&tp->t_delack_ch); 902 for (i = 0; i < TCPT_NTIMERS; i++) 903 TCP_TIMER_INIT(tp, i); 904 905 tp->t_flags = 0; 906 if (tcp_do_rfc1323 && tcp_do_win_scale) 907 tp->t_flags |= TF_REQ_SCALE; 908 if (tcp_do_rfc1323 && tcp_do_timestamps) 909 tp->t_flags |= TF_REQ_TSTMP; 910 if (tcp_do_sack == 2) 911 tp->t_flags |= TF_WILL_SACK; 912 else if (tcp_do_sack == 1) 913 tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK; 914 tp->t_flags |= TF_CANT_TXSACK; 915 switch (family) { 916 case PF_INET: 917 tp->t_inpcb = (struct inpcb *)aux; 918 tp->t_mtudisc = ip_mtudisc; 919 break; 920 #ifdef INET6 921 case PF_INET6: 922 tp->t_in6pcb = (struct in6pcb *)aux; 923 /* for IPv6, always try to run path MTU discovery */ 924 tp->t_mtudisc = 1; 925 break; 926 #endif 927 } 928 /* 929 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 930 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 931 * reasonable initial retransmit time. 932 */ 933 tp->t_srtt = TCPTV_SRTTBASE; 934 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 935 tp->t_rttmin = TCPTV_MIN; 936 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 937 TCPTV_MIN, TCPTV_REXMTMAX); 938 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 939 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 940 if (family == AF_INET) { 941 struct inpcb *inp = (struct inpcb *)aux; 942 inp->inp_ip.ip_ttl = ip_defttl; 943 inp->inp_ppcb = (caddr_t)tp; 944 } 945 #ifdef INET6 946 else if (family == AF_INET6) { 947 struct in6pcb *in6p = (struct in6pcb *)aux; 948 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 949 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 950 : NULL); 951 in6p->in6p_ppcb = (caddr_t)tp; 952 } 953 #endif 954 955 /* 956 * Initialize our timebase. When we send timestamps, we take 957 * the delta from tcp_now -- this means each connection always 958 * gets a timebase of 0, which makes it, among other things, 959 * more difficult to determine how long a system has been up, 960 * and thus how many TCP sequence increments have occurred. 961 */ 962 tp->ts_timebase = tcp_now; 963 964 return (tp); 965 } 966 967 /* 968 * Drop a TCP connection, reporting 969 * the specified error. If connection is synchronized, 970 * then send a RST to peer. 971 */ 972 struct tcpcb * 973 tcp_drop(tp, errno) 974 struct tcpcb *tp; 975 int errno; 976 { 977 struct socket *so = NULL; 978 979 #ifdef DIAGNOSTIC 980 if (tp->t_inpcb && tp->t_in6pcb) 981 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 982 #endif 983 #ifdef INET 984 if (tp->t_inpcb) 985 so = tp->t_inpcb->inp_socket; 986 #endif 987 #ifdef INET6 988 if (tp->t_in6pcb) 989 so = tp->t_in6pcb->in6p_socket; 990 #endif 991 if (!so) 992 return NULL; 993 994 if (TCPS_HAVERCVDSYN(tp->t_state)) { 995 tp->t_state = TCPS_CLOSED; 996 (void) tcp_output(tp); 997 tcpstat.tcps_drops++; 998 } else 999 tcpstat.tcps_conndrops++; 1000 if (errno == ETIMEDOUT && tp->t_softerror) 1001 errno = tp->t_softerror; 1002 so->so_error = errno; 1003 return (tcp_close(tp)); 1004 } 1005 1006 /* 1007 * Close a TCP control block: 1008 * discard all space held by the tcp 1009 * discard internet protocol block 1010 * wake up any sleepers 1011 */ 1012 struct tcpcb * 1013 tcp_close(tp) 1014 struct tcpcb *tp; 1015 { 1016 struct inpcb *inp; 1017 #ifdef INET6 1018 struct in6pcb *in6p; 1019 #endif 1020 struct socket *so; 1021 #ifdef RTV_RTT 1022 struct rtentry *rt; 1023 #endif 1024 struct route *ro; 1025 1026 inp = tp->t_inpcb; 1027 #ifdef INET6 1028 in6p = tp->t_in6pcb; 1029 #endif 1030 so = NULL; 1031 ro = NULL; 1032 if (inp) { 1033 so = inp->inp_socket; 1034 ro = &inp->inp_route; 1035 } 1036 #ifdef INET6 1037 else if (in6p) { 1038 so = in6p->in6p_socket; 1039 ro = (struct route *)&in6p->in6p_route; 1040 } 1041 #endif 1042 1043 #ifdef RTV_RTT 1044 /* 1045 * If we sent enough data to get some meaningful characteristics, 1046 * save them in the routing entry. 'Enough' is arbitrarily 1047 * defined as the sendpipesize (default 4K) * 16. This would 1048 * give us 16 rtt samples assuming we only get one sample per 1049 * window (the usual case on a long haul net). 16 samples is 1050 * enough for the srtt filter to converge to within 5% of the correct 1051 * value; fewer samples and we could save a very bogus rtt. 1052 * 1053 * Don't update the default route's characteristics and don't 1054 * update anything that the user "locked". 1055 */ 1056 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1057 ro && (rt = ro->ro_rt) && 1058 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1059 u_long i = 0; 1060 1061 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1062 i = tp->t_srtt * 1063 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1064 if (rt->rt_rmx.rmx_rtt && i) 1065 /* 1066 * filter this update to half the old & half 1067 * the new values, converting scale. 1068 * See route.h and tcp_var.h for a 1069 * description of the scaling constants. 1070 */ 1071 rt->rt_rmx.rmx_rtt = 1072 (rt->rt_rmx.rmx_rtt + i) / 2; 1073 else 1074 rt->rt_rmx.rmx_rtt = i; 1075 } 1076 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1077 i = tp->t_rttvar * 1078 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1079 if (rt->rt_rmx.rmx_rttvar && i) 1080 rt->rt_rmx.rmx_rttvar = 1081 (rt->rt_rmx.rmx_rttvar + i) / 2; 1082 else 1083 rt->rt_rmx.rmx_rttvar = i; 1084 } 1085 /* 1086 * update the pipelimit (ssthresh) if it has been updated 1087 * already or if a pipesize was specified & the threshhold 1088 * got below half the pipesize. I.e., wait for bad news 1089 * before we start updating, then update on both good 1090 * and bad news. 1091 */ 1092 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1093 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1094 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1095 /* 1096 * convert the limit from user data bytes to 1097 * packets then to packet data bytes. 1098 */ 1099 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1100 if (i < 2) 1101 i = 2; 1102 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1103 if (rt->rt_rmx.rmx_ssthresh) 1104 rt->rt_rmx.rmx_ssthresh = 1105 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1106 else 1107 rt->rt_rmx.rmx_ssthresh = i; 1108 } 1109 } 1110 #endif /* RTV_RTT */ 1111 /* free the reassembly queue, if any */ 1112 TCP_REASS_LOCK(tp); 1113 (void) tcp_freeq(tp); 1114 TCP_REASS_UNLOCK(tp); 1115 1116 tcp_canceltimers(tp); 1117 TCP_CLEAR_DELACK(tp); 1118 syn_cache_cleanup(tp); 1119 1120 if (tp->t_template) { 1121 m_free(tp->t_template); 1122 tp->t_template = NULL; 1123 } 1124 pool_put(&tcpcb_pool, tp); 1125 if (inp) { 1126 inp->inp_ppcb = 0; 1127 soisdisconnected(so); 1128 in_pcbdetach(inp); 1129 } 1130 #ifdef INET6 1131 else if (in6p) { 1132 in6p->in6p_ppcb = 0; 1133 soisdisconnected(so); 1134 in6_pcbdetach(in6p); 1135 } 1136 #endif 1137 tcpstat.tcps_closed++; 1138 return ((struct tcpcb *)0); 1139 } 1140 1141 int 1142 tcp_freeq(tp) 1143 struct tcpcb *tp; 1144 { 1145 struct ipqent *qe; 1146 int rv = 0; 1147 #ifdef TCPREASS_DEBUG 1148 int i = 0; 1149 #endif 1150 1151 TCP_REASS_LOCK_CHECK(tp); 1152 1153 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1154 #ifdef TCPREASS_DEBUG 1155 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1156 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1157 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1158 #endif 1159 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1160 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1161 m_freem(qe->ipqe_m); 1162 pool_put(&ipqent_pool, qe); 1163 rv = 1; 1164 } 1165 return (rv); 1166 } 1167 1168 /* 1169 * Protocol drain routine. Called when memory is in short supply. 1170 */ 1171 void 1172 tcp_drain() 1173 { 1174 struct inpcb *inp; 1175 struct tcpcb *tp; 1176 1177 /* 1178 * Free the sequence queue of all TCP connections. 1179 */ 1180 inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue); 1181 if (inp) /* XXX */ 1182 CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) { 1183 if ((tp = intotcpcb(inp)) != NULL) { 1184 /* 1185 * We may be called from a device's interrupt 1186 * context. If the tcpcb is already busy, 1187 * just bail out now. 1188 */ 1189 if (tcp_reass_lock_try(tp) == 0) 1190 continue; 1191 if (tcp_freeq(tp)) 1192 tcpstat.tcps_connsdrained++; 1193 TCP_REASS_UNLOCK(tp); 1194 } 1195 } 1196 } 1197 1198 #ifdef INET6 1199 void 1200 tcp6_drain() 1201 { 1202 struct in6pcb *in6p; 1203 struct tcpcb *tp; 1204 struct in6pcb *head = &tcb6; 1205 1206 /* 1207 * Free the sequence queue of all TCP connections. 1208 */ 1209 for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) { 1210 if ((tp = in6totcpcb(in6p)) != NULL) { 1211 /* 1212 * We may be called from a device's interrupt 1213 * context. If the tcpcb is already busy, 1214 * just bail out now. 1215 */ 1216 if (tcp_reass_lock_try(tp) == 0) 1217 continue; 1218 if (tcp_freeq(tp)) 1219 tcpstat.tcps_connsdrained++; 1220 TCP_REASS_UNLOCK(tp); 1221 } 1222 } 1223 } 1224 #endif 1225 1226 /* 1227 * Notify a tcp user of an asynchronous error; 1228 * store error as soft error, but wake up user 1229 * (for now, won't do anything until can select for soft error). 1230 */ 1231 void 1232 tcp_notify(inp, error) 1233 struct inpcb *inp; 1234 int error; 1235 { 1236 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1237 struct socket *so = inp->inp_socket; 1238 1239 /* 1240 * Ignore some errors if we are hooked up. 1241 * If connection hasn't completed, has retransmitted several times, 1242 * and receives a second error, give up now. This is better 1243 * than waiting a long time to establish a connection that 1244 * can never complete. 1245 */ 1246 if (tp->t_state == TCPS_ESTABLISHED && 1247 (error == EHOSTUNREACH || error == ENETUNREACH || 1248 error == EHOSTDOWN)) { 1249 return; 1250 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1251 tp->t_rxtshift > 3 && tp->t_softerror) 1252 so->so_error = error; 1253 else 1254 tp->t_softerror = error; 1255 wakeup((caddr_t) &so->so_timeo); 1256 sorwakeup(so); 1257 sowwakeup(so); 1258 } 1259 1260 #ifdef INET6 1261 void 1262 tcp6_notify(in6p, error) 1263 struct in6pcb *in6p; 1264 int error; 1265 { 1266 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1267 struct socket *so = in6p->in6p_socket; 1268 1269 /* 1270 * Ignore some errors if we are hooked up. 1271 * If connection hasn't completed, has retransmitted several times, 1272 * and receives a second error, give up now. This is better 1273 * than waiting a long time to establish a connection that 1274 * can never complete. 1275 */ 1276 if (tp->t_state == TCPS_ESTABLISHED && 1277 (error == EHOSTUNREACH || error == ENETUNREACH || 1278 error == EHOSTDOWN)) { 1279 return; 1280 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1281 tp->t_rxtshift > 3 && tp->t_softerror) 1282 so->so_error = error; 1283 else 1284 tp->t_softerror = error; 1285 wakeup((caddr_t) &so->so_timeo); 1286 sorwakeup(so); 1287 sowwakeup(so); 1288 } 1289 #endif 1290 1291 #ifdef INET6 1292 void 1293 tcp6_ctlinput(cmd, sa, d) 1294 int cmd; 1295 struct sockaddr *sa; 1296 void *d; 1297 { 1298 struct tcphdr th; 1299 void (*notify) __P((struct in6pcb *, int)) = tcp6_notify; 1300 int nmatch; 1301 struct ip6_hdr *ip6; 1302 const struct sockaddr_in6 *sa6_src = NULL; 1303 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1304 struct mbuf *m; 1305 int off; 1306 1307 if (sa->sa_family != AF_INET6 || 1308 sa->sa_len != sizeof(struct sockaddr_in6)) 1309 return; 1310 if ((unsigned)cmd >= PRC_NCMDS) 1311 return; 1312 else if (cmd == PRC_QUENCH) { 1313 /* XXX there's no PRC_QUENCH in IPv6 */ 1314 notify = tcp6_quench; 1315 } else if (PRC_IS_REDIRECT(cmd)) 1316 notify = in6_rtchange, d = NULL; 1317 else if (cmd == PRC_MSGSIZE) 1318 ; /* special code is present, see below */ 1319 else if (cmd == PRC_HOSTDEAD) 1320 d = NULL; 1321 else if (inet6ctlerrmap[cmd] == 0) 1322 return; 1323 1324 /* if the parameter is from icmp6, decode it. */ 1325 if (d != NULL) { 1326 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1327 m = ip6cp->ip6c_m; 1328 ip6 = ip6cp->ip6c_ip6; 1329 off = ip6cp->ip6c_off; 1330 sa6_src = ip6cp->ip6c_src; 1331 } else { 1332 m = NULL; 1333 ip6 = NULL; 1334 sa6_src = &sa6_any; 1335 } 1336 1337 if (ip6) { 1338 /* 1339 * XXX: We assume that when ip6 is non NULL, 1340 * M and OFF are valid. 1341 */ 1342 1343 /* check if we can safely examine src and dst ports */ 1344 if (m->m_pkthdr.len < off + sizeof(th)) { 1345 if (cmd == PRC_MSGSIZE) 1346 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1347 return; 1348 } 1349 1350 bzero(&th, sizeof(th)); 1351 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1352 1353 if (cmd == PRC_MSGSIZE) { 1354 int valid = 0; 1355 1356 /* 1357 * Check to see if we have a valid TCP connection 1358 * corresponding to the address in the ICMPv6 message 1359 * payload. 1360 */ 1361 if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr, 1362 th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr, 1363 th.th_sport, 0)) 1364 valid++; 1365 1366 /* 1367 * Depending on the value of "valid" and routing table 1368 * size (mtudisc_{hi,lo}wat), we will: 1369 * - recalcurate the new MTU and create the 1370 * corresponding routing entry, or 1371 * - ignore the MTU change notification. 1372 */ 1373 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1374 1375 /* 1376 * no need to call in6_pcbnotify, it should have been 1377 * called via callback if necessary 1378 */ 1379 return; 1380 } 1381 1382 nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport, 1383 (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1384 if (nmatch == 0 && syn_cache_count && 1385 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1386 inet6ctlerrmap[cmd] == ENETUNREACH || 1387 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1388 syn_cache_unreach((struct sockaddr *)sa6_src, 1389 sa, &th); 1390 } else { 1391 (void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src, 1392 0, cmd, NULL, notify); 1393 } 1394 } 1395 #endif 1396 1397 #ifdef INET 1398 /* assumes that ip header and tcp header are contiguous on mbuf */ 1399 void * 1400 tcp_ctlinput(cmd, sa, v) 1401 int cmd; 1402 struct sockaddr *sa; 1403 void *v; 1404 { 1405 struct ip *ip = v; 1406 struct tcphdr *th; 1407 struct icmp *icp; 1408 extern const int inetctlerrmap[]; 1409 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1410 int errno; 1411 int nmatch; 1412 1413 if (sa->sa_family != AF_INET || 1414 sa->sa_len != sizeof(struct sockaddr_in)) 1415 return NULL; 1416 if ((unsigned)cmd >= PRC_NCMDS) 1417 return NULL; 1418 errno = inetctlerrmap[cmd]; 1419 if (cmd == PRC_QUENCH) 1420 notify = tcp_quench; 1421 else if (PRC_IS_REDIRECT(cmd)) 1422 notify = in_rtchange, ip = 0; 1423 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1424 /* 1425 * Check to see if we have a valid TCP connection 1426 * corresponding to the address in the ICMP message 1427 * payload. 1428 * 1429 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1430 */ 1431 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1432 if (in_pcblookup_connect(&tcbtable, 1433 ip->ip_dst, th->th_dport, 1434 ip->ip_src, th->th_sport) == NULL) 1435 return NULL; 1436 1437 /* 1438 * Now that we've validated that we are actually communicating 1439 * with the host indicated in the ICMP message, locate the 1440 * ICMP header, recalculate the new MTU, and create the 1441 * corresponding routing entry. 1442 */ 1443 icp = (struct icmp *)((caddr_t)ip - 1444 offsetof(struct icmp, icmp_ip)); 1445 icmp_mtudisc(icp, ip->ip_dst); 1446 1447 return NULL; 1448 } else if (cmd == PRC_HOSTDEAD) 1449 ip = 0; 1450 else if (errno == 0) 1451 return NULL; 1452 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1453 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1454 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1455 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1456 if (nmatch == 0 && syn_cache_count && 1457 (inetctlerrmap[cmd] == EHOSTUNREACH || 1458 inetctlerrmap[cmd] == ENETUNREACH || 1459 inetctlerrmap[cmd] == EHOSTDOWN)) { 1460 struct sockaddr_in sin; 1461 bzero(&sin, sizeof(sin)); 1462 sin.sin_len = sizeof(sin); 1463 sin.sin_family = AF_INET; 1464 sin.sin_port = th->th_sport; 1465 sin.sin_addr = ip->ip_src; 1466 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1467 } 1468 1469 /* XXX mapped address case */ 1470 } else 1471 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1472 notify); 1473 return NULL; 1474 } 1475 1476 /* 1477 * When a source quence is received, we are being notifed of congestion. 1478 * Close the congestion window down to the Loss Window (one segment). 1479 * We will gradually open it again as we proceed. 1480 */ 1481 void 1482 tcp_quench(inp, errno) 1483 struct inpcb *inp; 1484 int errno; 1485 { 1486 struct tcpcb *tp = intotcpcb(inp); 1487 1488 if (tp) 1489 tp->snd_cwnd = tp->t_segsz; 1490 } 1491 #endif 1492 1493 #ifdef INET6 1494 void 1495 tcp6_quench(in6p, errno) 1496 struct in6pcb *in6p; 1497 int errno; 1498 { 1499 struct tcpcb *tp = in6totcpcb(in6p); 1500 1501 if (tp) 1502 tp->snd_cwnd = tp->t_segsz; 1503 } 1504 #endif 1505 1506 #ifdef INET 1507 /* 1508 * Path MTU Discovery handlers. 1509 */ 1510 void 1511 tcp_mtudisc_callback(faddr) 1512 struct in_addr faddr; 1513 { 1514 1515 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1516 } 1517 1518 /* 1519 * On receipt of path MTU corrections, flush old route and replace it 1520 * with the new one. Retransmit all unacknowledged packets, to ensure 1521 * that all packets will be received. 1522 */ 1523 void 1524 tcp_mtudisc(inp, errno) 1525 struct inpcb *inp; 1526 int errno; 1527 { 1528 struct tcpcb *tp = intotcpcb(inp); 1529 struct rtentry *rt = in_pcbrtentry(inp); 1530 1531 if (tp != 0) { 1532 if (rt != 0) { 1533 /* 1534 * If this was not a host route, remove and realloc. 1535 */ 1536 if ((rt->rt_flags & RTF_HOST) == 0) { 1537 in_rtchange(inp, errno); 1538 if ((rt = in_pcbrtentry(inp)) == 0) 1539 return; 1540 } 1541 1542 /* 1543 * Slow start out of the error condition. We 1544 * use the MTU because we know it's smaller 1545 * than the previously transmitted segment. 1546 * 1547 * Note: This is more conservative than the 1548 * suggestion in draft-floyd-incr-init-win-03. 1549 */ 1550 if (rt->rt_rmx.rmx_mtu != 0) 1551 tp->snd_cwnd = 1552 TCP_INITIAL_WINDOW(tcp_init_win, 1553 rt->rt_rmx.rmx_mtu); 1554 } 1555 1556 /* 1557 * Resend unacknowledged packets. 1558 */ 1559 tp->snd_nxt = tp->snd_una; 1560 tcp_output(tp); 1561 } 1562 } 1563 #endif 1564 1565 #ifdef INET6 1566 /* 1567 * Path MTU Discovery handlers. 1568 */ 1569 void 1570 tcp6_mtudisc_callback(faddr) 1571 struct in6_addr *faddr; 1572 { 1573 struct sockaddr_in6 sin6; 1574 1575 bzero(&sin6, sizeof(sin6)); 1576 sin6.sin6_family = AF_INET6; 1577 sin6.sin6_len = sizeof(struct sockaddr_in6); 1578 sin6.sin6_addr = *faddr; 1579 (void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0, 1580 (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1581 } 1582 1583 void 1584 tcp6_mtudisc(in6p, errno) 1585 struct in6pcb *in6p; 1586 int errno; 1587 { 1588 struct tcpcb *tp = in6totcpcb(in6p); 1589 struct rtentry *rt = in6_pcbrtentry(in6p); 1590 1591 if (tp != 0) { 1592 if (rt != 0) { 1593 /* 1594 * If this was not a host route, remove and realloc. 1595 */ 1596 if ((rt->rt_flags & RTF_HOST) == 0) { 1597 in6_rtchange(in6p, errno); 1598 if ((rt = in6_pcbrtentry(in6p)) == 0) 1599 return; 1600 } 1601 1602 /* 1603 * Slow start out of the error condition. We 1604 * use the MTU because we know it's smaller 1605 * than the previously transmitted segment. 1606 * 1607 * Note: This is more conservative than the 1608 * suggestion in draft-floyd-incr-init-win-03. 1609 */ 1610 if (rt->rt_rmx.rmx_mtu != 0) 1611 tp->snd_cwnd = 1612 TCP_INITIAL_WINDOW(tcp_init_win, 1613 rt->rt_rmx.rmx_mtu); 1614 } 1615 1616 /* 1617 * Resend unacknowledged packets. 1618 */ 1619 tp->snd_nxt = tp->snd_una; 1620 tcp_output(tp); 1621 } 1622 } 1623 #endif /* INET6 */ 1624 1625 /* 1626 * Compute the MSS to advertise to the peer. Called only during 1627 * the 3-way handshake. If we are the server (peer initiated 1628 * connection), we are called with a pointer to the interface 1629 * on which the SYN packet arrived. If we are the client (we 1630 * initiated connection), we are called with a pointer to the 1631 * interface out which this connection should go. 1632 * 1633 * NOTE: Do not subtract IP option/extension header size nor IPsec 1634 * header size from MSS advertisement. MSS option must hold the maximum 1635 * segment size we can accept, so it must always be: 1636 * max(if mtu) - ip header - tcp header 1637 */ 1638 u_long 1639 tcp_mss_to_advertise(ifp, af) 1640 const struct ifnet *ifp; 1641 int af; 1642 { 1643 extern u_long in_maxmtu; 1644 u_long mss = 0; 1645 u_long hdrsiz; 1646 1647 /* 1648 * In order to avoid defeating path MTU discovery on the peer, 1649 * we advertise the max MTU of all attached networks as our MSS, 1650 * per RFC 1191, section 3.1. 1651 * 1652 * We provide the option to advertise just the MTU of 1653 * the interface on which we hope this connection will 1654 * be receiving. If we are responding to a SYN, we 1655 * will have a pretty good idea about this, but when 1656 * initiating a connection there is a bit more doubt. 1657 * 1658 * We also need to ensure that loopback has a large enough 1659 * MSS, as the loopback MTU is never included in in_maxmtu. 1660 */ 1661 1662 if (ifp != NULL) 1663 mss = ifp->if_mtu; 1664 1665 if (tcp_mss_ifmtu == 0) 1666 switch (af) { 1667 case AF_INET: 1668 mss = max(in_maxmtu, mss); 1669 break; 1670 #ifdef INET6 1671 case AF_INET6: 1672 mss = max(in6_maxmtu, mss); 1673 break; 1674 #endif 1675 } 1676 1677 switch (af) { 1678 case AF_INET: 1679 hdrsiz = sizeof(struct ip); 1680 break; 1681 #ifdef INET6 1682 case AF_INET6: 1683 hdrsiz = sizeof(struct ip6_hdr); 1684 break; 1685 #endif 1686 default: 1687 hdrsiz = 0; 1688 break; 1689 } 1690 hdrsiz += sizeof(struct tcphdr); 1691 if (mss > hdrsiz) 1692 mss -= hdrsiz; 1693 1694 mss = max(tcp_mssdflt, mss); 1695 return (mss); 1696 } 1697 1698 /* 1699 * Set connection variables based on the peer's advertised MSS. 1700 * We are passed the TCPCB for the actual connection. If we 1701 * are the server, we are called by the compressed state engine 1702 * when the 3-way handshake is complete. If we are the client, 1703 * we are called when we receive the SYN,ACK from the server. 1704 * 1705 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1706 * before this routine is called! 1707 */ 1708 void 1709 tcp_mss_from_peer(tp, offer) 1710 struct tcpcb *tp; 1711 int offer; 1712 { 1713 struct socket *so; 1714 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1715 struct rtentry *rt; 1716 #endif 1717 u_long bufsize; 1718 int mss; 1719 1720 #ifdef DIAGNOSTIC 1721 if (tp->t_inpcb && tp->t_in6pcb) 1722 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1723 #endif 1724 so = NULL; 1725 rt = NULL; 1726 #ifdef INET 1727 if (tp->t_inpcb) { 1728 so = tp->t_inpcb->inp_socket; 1729 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1730 rt = in_pcbrtentry(tp->t_inpcb); 1731 #endif 1732 } 1733 #endif 1734 #ifdef INET6 1735 if (tp->t_in6pcb) { 1736 so = tp->t_in6pcb->in6p_socket; 1737 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1738 rt = in6_pcbrtentry(tp->t_in6pcb); 1739 #endif 1740 } 1741 #endif 1742 1743 /* 1744 * As per RFC1122, use the default MSS value, unless they 1745 * sent us an offer. Do not accept offers less than 32 bytes. 1746 */ 1747 mss = tcp_mssdflt; 1748 if (offer) 1749 mss = offer; 1750 mss = max(mss, 32); /* sanity */ 1751 tp->t_peermss = mss; 1752 mss -= tcp_optlen(tp); 1753 #ifdef INET 1754 if (tp->t_inpcb) 1755 mss -= ip_optlen(tp->t_inpcb); 1756 #endif 1757 #ifdef INET6 1758 if (tp->t_in6pcb) 1759 mss -= ip6_optlen(tp->t_in6pcb); 1760 #endif 1761 1762 /* 1763 * If there's a pipesize, change the socket buffer to that size. 1764 * Make the socket buffer an integral number of MSS units. If 1765 * the MSS is larger than the socket buffer, artificially decrease 1766 * the MSS. 1767 */ 1768 #ifdef RTV_SPIPE 1769 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1770 bufsize = rt->rt_rmx.rmx_sendpipe; 1771 else 1772 #endif 1773 bufsize = so->so_snd.sb_hiwat; 1774 if (bufsize < mss) 1775 mss = bufsize; 1776 else { 1777 bufsize = roundup(bufsize, mss); 1778 if (bufsize > sb_max) 1779 bufsize = sb_max; 1780 (void) sbreserve(&so->so_snd, bufsize); 1781 } 1782 tp->t_segsz = mss; 1783 1784 #ifdef RTV_SSTHRESH 1785 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1786 /* 1787 * There's some sort of gateway or interface buffer 1788 * limit on the path. Use this to set the slow 1789 * start threshold, but set the threshold to no less 1790 * than 2 * MSS. 1791 */ 1792 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1793 } 1794 #endif 1795 } 1796 1797 /* 1798 * Processing necessary when a TCP connection is established. 1799 */ 1800 void 1801 tcp_established(tp) 1802 struct tcpcb *tp; 1803 { 1804 struct socket *so; 1805 #ifdef RTV_RPIPE 1806 struct rtentry *rt; 1807 #endif 1808 u_long bufsize; 1809 1810 #ifdef DIAGNOSTIC 1811 if (tp->t_inpcb && tp->t_in6pcb) 1812 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1813 #endif 1814 so = NULL; 1815 rt = NULL; 1816 #ifdef INET 1817 if (tp->t_inpcb) { 1818 so = tp->t_inpcb->inp_socket; 1819 #if defined(RTV_RPIPE) 1820 rt = in_pcbrtentry(tp->t_inpcb); 1821 #endif 1822 } 1823 #endif 1824 #ifdef INET6 1825 if (tp->t_in6pcb) { 1826 so = tp->t_in6pcb->in6p_socket; 1827 #if defined(RTV_RPIPE) 1828 rt = in6_pcbrtentry(tp->t_in6pcb); 1829 #endif 1830 } 1831 #endif 1832 1833 tp->t_state = TCPS_ESTABLISHED; 1834 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1835 1836 #ifdef RTV_RPIPE 1837 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1838 bufsize = rt->rt_rmx.rmx_recvpipe; 1839 else 1840 #endif 1841 bufsize = so->so_rcv.sb_hiwat; 1842 if (bufsize > tp->t_ourmss) { 1843 bufsize = roundup(bufsize, tp->t_ourmss); 1844 if (bufsize > sb_max) 1845 bufsize = sb_max; 1846 (void) sbreserve(&so->so_rcv, bufsize); 1847 } 1848 } 1849 1850 /* 1851 * Check if there's an initial rtt or rttvar. Convert from the 1852 * route-table units to scaled multiples of the slow timeout timer. 1853 * Called only during the 3-way handshake. 1854 */ 1855 void 1856 tcp_rmx_rtt(tp) 1857 struct tcpcb *tp; 1858 { 1859 #ifdef RTV_RTT 1860 struct rtentry *rt = NULL; 1861 int rtt; 1862 1863 #ifdef DIAGNOSTIC 1864 if (tp->t_inpcb && tp->t_in6pcb) 1865 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 1866 #endif 1867 #ifdef INET 1868 if (tp->t_inpcb) 1869 rt = in_pcbrtentry(tp->t_inpcb); 1870 #endif 1871 #ifdef INET6 1872 if (tp->t_in6pcb) 1873 rt = in6_pcbrtentry(tp->t_in6pcb); 1874 #endif 1875 if (rt == NULL) 1876 return; 1877 1878 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1879 /* 1880 * XXX The lock bit for MTU indicates that the value 1881 * is also a minimum value; this is subject to time. 1882 */ 1883 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1884 TCPT_RANGESET(tp->t_rttmin, 1885 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 1886 TCPTV_MIN, TCPTV_REXMTMAX); 1887 tp->t_srtt = rtt / 1888 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1889 if (rt->rt_rmx.rmx_rttvar) { 1890 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1891 ((RTM_RTTUNIT / PR_SLOWHZ) >> 1892 (TCP_RTTVAR_SHIFT + 2)); 1893 } else { 1894 /* Default variation is +- 1 rtt */ 1895 tp->t_rttvar = 1896 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 1897 } 1898 TCPT_RANGESET(tp->t_rxtcur, 1899 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 1900 tp->t_rttmin, TCPTV_REXMTMAX); 1901 } 1902 #endif 1903 } 1904 1905 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 1906 #if NRND > 0 1907 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 1908 #endif 1909 1910 /* 1911 * Get a new sequence value given a tcp control block 1912 */ 1913 tcp_seq 1914 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 1915 { 1916 1917 #ifdef INET 1918 if (tp->t_inpcb != NULL) { 1919 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 1920 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 1921 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 1922 addin)); 1923 } 1924 #endif 1925 #ifdef INET6 1926 if (tp->t_in6pcb != NULL) { 1927 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 1928 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 1929 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 1930 addin)); 1931 } 1932 #endif 1933 /* Not possible. */ 1934 panic("tcp_new_iss"); 1935 } 1936 1937 /* 1938 * This routine actually generates a new TCP initial sequence number. 1939 */ 1940 tcp_seq 1941 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 1942 size_t addrsz, tcp_seq addin) 1943 { 1944 tcp_seq tcp_iss; 1945 1946 #if NRND > 0 1947 static int beenhere; 1948 1949 /* 1950 * If we haven't been here before, initialize our cryptographic 1951 * hash secret. 1952 */ 1953 if (beenhere == 0) { 1954 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 1955 RND_EXTRACT_ANY); 1956 beenhere = 1; 1957 } 1958 1959 if (tcp_do_rfc1948) { 1960 MD5_CTX ctx; 1961 u_int8_t hash[16]; /* XXX MD5 knowledge */ 1962 1963 /* 1964 * Compute the base value of the ISS. It is a hash 1965 * of (saddr, sport, daddr, dport, secret). 1966 */ 1967 MD5Init(&ctx); 1968 1969 MD5Update(&ctx, (u_char *) laddr, addrsz); 1970 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 1971 1972 MD5Update(&ctx, (u_char *) faddr, addrsz); 1973 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 1974 1975 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 1976 1977 MD5Final(hash, &ctx); 1978 1979 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 1980 1981 /* 1982 * Now increment our "timer", and add it in to 1983 * the computed value. 1984 * 1985 * XXX Use `addin'? 1986 * XXX TCP_ISSINCR too large to use? 1987 */ 1988 tcp_iss_seq += TCP_ISSINCR; 1989 #ifdef TCPISS_DEBUG 1990 printf("ISS hash 0x%08x, ", tcp_iss); 1991 #endif 1992 tcp_iss += tcp_iss_seq + addin; 1993 #ifdef TCPISS_DEBUG 1994 printf("new ISS 0x%08x\n", tcp_iss); 1995 #endif 1996 } else 1997 #endif /* NRND > 0 */ 1998 { 1999 /* 2000 * Randomize. 2001 */ 2002 #if NRND > 0 2003 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2004 #else 2005 tcp_iss = arc4random(); 2006 #endif 2007 2008 /* 2009 * If we were asked to add some amount to a known value, 2010 * we will take a random value obtained above, mask off 2011 * the upper bits, and add in the known value. We also 2012 * add in a constant to ensure that we are at least a 2013 * certain distance from the original value. 2014 * 2015 * This is used when an old connection is in timed wait 2016 * and we have a new one coming in, for instance. 2017 */ 2018 if (addin != 0) { 2019 #ifdef TCPISS_DEBUG 2020 printf("Random %08x, ", tcp_iss); 2021 #endif 2022 tcp_iss &= TCP_ISS_RANDOM_MASK; 2023 tcp_iss += addin + TCP_ISSINCR; 2024 #ifdef TCPISS_DEBUG 2025 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2026 #endif 2027 } else { 2028 tcp_iss &= TCP_ISS_RANDOM_MASK; 2029 tcp_iss += tcp_iss_seq; 2030 tcp_iss_seq += TCP_ISSINCR; 2031 #ifdef TCPISS_DEBUG 2032 printf("ISS %08x\n", tcp_iss); 2033 #endif 2034 } 2035 } 2036 2037 if (tcp_compat_42) { 2038 /* 2039 * Limit it to the positive range for really old TCP 2040 * implementations. 2041 */ 2042 if (tcp_iss >= 0x80000000) 2043 tcp_iss &= 0x7fffffff; /* XXX */ 2044 } 2045 2046 return (tcp_iss); 2047 } 2048 2049 #ifdef IPSEC 2050 /* compute ESP/AH header size for TCP, including outer IP header. */ 2051 size_t 2052 ipsec4_hdrsiz_tcp(tp) 2053 struct tcpcb *tp; 2054 { 2055 struct inpcb *inp; 2056 size_t hdrsiz; 2057 2058 /* XXX mapped addr case (tp->t_in6pcb) */ 2059 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2060 return 0; 2061 switch (tp->t_family) { 2062 case AF_INET: 2063 /* XXX: should use currect direction. */ 2064 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2065 break; 2066 default: 2067 hdrsiz = 0; 2068 break; 2069 } 2070 2071 return hdrsiz; 2072 } 2073 2074 #ifdef INET6 2075 size_t 2076 ipsec6_hdrsiz_tcp(tp) 2077 struct tcpcb *tp; 2078 { 2079 struct in6pcb *in6p; 2080 size_t hdrsiz; 2081 2082 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2083 return 0; 2084 switch (tp->t_family) { 2085 case AF_INET6: 2086 /* XXX: should use currect direction. */ 2087 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2088 break; 2089 case AF_INET: 2090 /* mapped address case - tricky */ 2091 default: 2092 hdrsiz = 0; 2093 break; 2094 } 2095 2096 return hdrsiz; 2097 } 2098 #endif 2099 #endif /*IPSEC*/ 2100 2101 /* 2102 * Determine the length of the TCP options for this connection. 2103 * 2104 * XXX: What do we do for SACK, when we add that? Just reserve 2105 * all of the space? Otherwise we can't exactly be incrementing 2106 * cwnd by an amount that varies depending on the amount we last 2107 * had to SACK! 2108 */ 2109 2110 u_int 2111 tcp_optlen(tp) 2112 struct tcpcb *tp; 2113 { 2114 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2115 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2116 return TCPOLEN_TSTAMP_APPA; 2117 else 2118 return 0; 2119 } 2120