tcp_subr.c revision 1.221
1 /* $NetBSD: tcp_subr.c,v 1.221 2008/01/14 04:19:10 dyoung 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 1.221 2008/01/14 04:19:10 dyoung 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_minmss = TCP_MINMSS; 179 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 180 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 181 #if NRND > 0 182 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 183 #endif 184 int tcp_do_sack = 1; /* selective acknowledgement */ 185 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 186 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 187 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 188 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 189 #ifndef TCP_INIT_WIN 190 #define TCP_INIT_WIN 0 /* initial slow start window */ 191 #endif 192 #ifndef TCP_INIT_WIN_LOCAL 193 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 194 #endif 195 int tcp_init_win = TCP_INIT_WIN; 196 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 197 int tcp_mss_ifmtu = 0; 198 #ifdef TCP_COMPAT_42 199 int tcp_compat_42 = 1; 200 #else 201 int tcp_compat_42 = 0; 202 #endif 203 int tcp_rst_ppslim = 100; /* 100pps */ 204 int tcp_ackdrop_ppslim = 100; /* 100pps */ 205 int tcp_do_loopback_cksum = 0; 206 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 207 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 208 int tcp_sack_tp_maxholes = 32; 209 int tcp_sack_globalmaxholes = 1024; 210 int tcp_sack_globalholes = 0; 211 int tcp_ecn_maxretries = 1; 212 213 /* tcb hash */ 214 #ifndef TCBHASHSIZE 215 #define TCBHASHSIZE 128 216 #endif 217 int tcbhashsize = TCBHASHSIZE; 218 219 /* syn hash parameters */ 220 #define TCP_SYN_HASH_SIZE 293 221 #define TCP_SYN_BUCKET_SIZE 35 222 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 223 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 224 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 225 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 226 227 int tcp_freeq(struct tcpcb *); 228 229 #ifdef INET 230 void tcp_mtudisc_callback(struct in_addr); 231 #endif 232 #ifdef INET6 233 void tcp6_mtudisc_callback(struct in6_addr *); 234 #endif 235 236 #ifdef INET6 237 void tcp6_mtudisc(struct in6pcb *, int); 238 #endif 239 240 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL, 241 IPL_SOFTNET); 242 243 #ifdef TCP_CSUM_COUNTERS 244 #include <sys/device.h> 245 246 #if defined(INET) 247 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "hwcsum bad"); 249 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 250 NULL, "tcp", "hwcsum ok"); 251 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 252 NULL, "tcp", "hwcsum data"); 253 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 254 NULL, "tcp", "swcsum"); 255 256 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 257 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 258 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 259 EVCNT_ATTACH_STATIC(tcp_swcsum); 260 #endif /* defined(INET) */ 261 262 #if defined(INET6) 263 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "hwcsum bad"); 265 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 NULL, "tcp6", "hwcsum ok"); 267 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 268 NULL, "tcp6", "hwcsum data"); 269 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 270 NULL, "tcp6", "swcsum"); 271 272 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 273 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 274 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 275 EVCNT_ATTACH_STATIC(tcp6_swcsum); 276 #endif /* defined(INET6) */ 277 #endif /* TCP_CSUM_COUNTERS */ 278 279 280 #ifdef TCP_OUTPUT_COUNTERS 281 #include <sys/device.h> 282 283 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output big header"); 285 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output predict hit"); 287 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output predict miss"); 289 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 290 NULL, "tcp", "output copy small"); 291 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 292 NULL, "tcp", "output copy big"); 293 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 294 NULL, "tcp", "output reference big"); 295 296 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 297 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 298 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 299 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 300 EVCNT_ATTACH_STATIC(tcp_output_copybig); 301 EVCNT_ATTACH_STATIC(tcp_output_refbig); 302 303 #endif /* TCP_OUTPUT_COUNTERS */ 304 305 #ifdef TCP_REASS_COUNTERS 306 #include <sys/device.h> 307 308 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 309 NULL, "tcp_reass", "calls"); 310 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 311 &tcp_reass_, "tcp_reass", "insert into empty queue"); 312 struct evcnt tcp_reass_iteration[8] = { 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 318 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 319 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 321 }; 322 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "prepend to first"); 324 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "prepend"); 326 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "insert"); 328 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "insert at tail"); 330 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "append"); 332 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "append to tail fragment"); 334 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "overlap at end"); 336 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 337 &tcp_reass_, "tcp_reass", "overlap at start"); 338 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 339 &tcp_reass_, "tcp_reass", "duplicate segment"); 340 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 341 &tcp_reass_, "tcp_reass", "duplicate fragment"); 342 343 EVCNT_ATTACH_STATIC(tcp_reass_); 344 EVCNT_ATTACH_STATIC(tcp_reass_empty); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 349 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 350 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 351 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 352 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 353 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 354 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 355 EVCNT_ATTACH_STATIC(tcp_reass_insert); 356 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 357 EVCNT_ATTACH_STATIC(tcp_reass_append); 358 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 359 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 360 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 361 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 362 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 363 364 #endif /* TCP_REASS_COUNTERS */ 365 366 #ifdef MBUFTRACE 367 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 368 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 369 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 370 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 371 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 372 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 373 #endif 374 375 /* 376 * Tcp initialization 377 */ 378 void 379 tcp_init(void) 380 { 381 int hlen; 382 383 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 384 385 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 386 #ifdef INET6 387 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 388 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 389 #endif 390 if (max_protohdr < hlen) 391 max_protohdr = hlen; 392 if (max_linkhdr + hlen > MHLEN) 393 panic("tcp_init"); 394 395 #ifdef INET 396 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 397 #endif 398 #ifdef INET6 399 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 400 #endif 401 402 /* Initialize timer state. */ 403 tcp_timer_init(); 404 405 /* Initialize the compressed state engine. */ 406 syn_cache_init(); 407 408 /* Initialize the congestion control algorithms. */ 409 tcp_congctl_init(); 410 411 /* Initialize the TCPCB template. */ 412 tcp_tcpcb_template(); 413 414 MOWNER_ATTACH(&tcp_tx_mowner); 415 MOWNER_ATTACH(&tcp_rx_mowner); 416 MOWNER_ATTACH(&tcp_reass_mowner); 417 MOWNER_ATTACH(&tcp_sock_mowner); 418 MOWNER_ATTACH(&tcp_sock_tx_mowner); 419 MOWNER_ATTACH(&tcp_sock_rx_mowner); 420 MOWNER_ATTACH(&tcp_mowner); 421 } 422 423 /* 424 * Create template to be used to send tcp packets on a connection. 425 * Call after host entry created, allocates an mbuf and fills 426 * in a skeletal tcp/ip header, minimizing the amount of work 427 * necessary when the connection is used. 428 */ 429 struct mbuf * 430 tcp_template(struct tcpcb *tp) 431 { 432 struct inpcb *inp = tp->t_inpcb; 433 #ifdef INET6 434 struct in6pcb *in6p = tp->t_in6pcb; 435 #endif 436 struct tcphdr *n; 437 struct mbuf *m; 438 int hlen; 439 440 switch (tp->t_family) { 441 case AF_INET: 442 hlen = sizeof(struct ip); 443 if (inp) 444 break; 445 #ifdef INET6 446 if (in6p) { 447 /* mapped addr case */ 448 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 449 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 450 break; 451 } 452 #endif 453 return NULL; /*EINVAL*/ 454 #ifdef INET6 455 case AF_INET6: 456 hlen = sizeof(struct ip6_hdr); 457 if (in6p) { 458 /* more sainty check? */ 459 break; 460 } 461 return NULL; /*EINVAL*/ 462 #endif 463 default: 464 hlen = 0; /*pacify gcc*/ 465 return NULL; /*EAFNOSUPPORT*/ 466 } 467 #ifdef DIAGNOSTIC 468 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 469 panic("mclbytes too small for t_template"); 470 #endif 471 m = tp->t_template; 472 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 473 ; 474 else { 475 if (m) 476 m_freem(m); 477 m = tp->t_template = NULL; 478 MGETHDR(m, M_DONTWAIT, MT_HEADER); 479 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 480 MCLGET(m, M_DONTWAIT); 481 if ((m->m_flags & M_EXT) == 0) { 482 m_free(m); 483 m = NULL; 484 } 485 } 486 if (m == NULL) 487 return NULL; 488 MCLAIM(m, &tcp_mowner); 489 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 490 } 491 492 bzero(mtod(m, void *), m->m_len); 493 494 n = (struct tcphdr *)(mtod(m, char *) + hlen); 495 496 switch (tp->t_family) { 497 case AF_INET: 498 { 499 struct ipovly *ipov; 500 mtod(m, struct ip *)->ip_v = 4; 501 mtod(m, struct ip *)->ip_hl = hlen >> 2; 502 ipov = mtod(m, struct ipovly *); 503 ipov->ih_pr = IPPROTO_TCP; 504 ipov->ih_len = htons(sizeof(struct tcphdr)); 505 if (inp) { 506 ipov->ih_src = inp->inp_laddr; 507 ipov->ih_dst = inp->inp_faddr; 508 } 509 #ifdef INET6 510 else if (in6p) { 511 /* mapped addr case */ 512 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 513 sizeof(ipov->ih_src)); 514 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 515 sizeof(ipov->ih_dst)); 516 } 517 #endif 518 /* 519 * Compute the pseudo-header portion of the checksum 520 * now. We incrementally add in the TCP option and 521 * payload lengths later, and then compute the TCP 522 * checksum right before the packet is sent off onto 523 * the wire. 524 */ 525 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 526 ipov->ih_dst.s_addr, 527 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 528 break; 529 } 530 #ifdef INET6 531 case AF_INET6: 532 { 533 struct ip6_hdr *ip6; 534 mtod(m, struct ip *)->ip_v = 6; 535 ip6 = mtod(m, struct ip6_hdr *); 536 ip6->ip6_nxt = IPPROTO_TCP; 537 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 538 ip6->ip6_src = in6p->in6p_laddr; 539 ip6->ip6_dst = in6p->in6p_faddr; 540 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 541 if (ip6_auto_flowlabel) { 542 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 543 ip6->ip6_flow |= 544 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 545 } 546 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 547 ip6->ip6_vfc |= IPV6_VERSION; 548 549 /* 550 * Compute the pseudo-header portion of the checksum 551 * now. We incrementally add in the TCP option and 552 * payload lengths later, and then compute the TCP 553 * checksum right before the packet is sent off onto 554 * the wire. 555 */ 556 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 557 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 558 htonl(IPPROTO_TCP)); 559 break; 560 } 561 #endif 562 } 563 if (inp) { 564 n->th_sport = inp->inp_lport; 565 n->th_dport = inp->inp_fport; 566 } 567 #ifdef INET6 568 else if (in6p) { 569 n->th_sport = in6p->in6p_lport; 570 n->th_dport = in6p->in6p_fport; 571 } 572 #endif 573 n->th_seq = 0; 574 n->th_ack = 0; 575 n->th_x2 = 0; 576 n->th_off = 5; 577 n->th_flags = 0; 578 n->th_win = 0; 579 n->th_urp = 0; 580 return (m); 581 } 582 583 /* 584 * Send a single message to the TCP at address specified by 585 * the given TCP/IP header. If m == 0, then we make a copy 586 * of the tcpiphdr at ti and send directly to the addressed host. 587 * This is used to force keep alive messages out using the TCP 588 * template for a connection tp->t_template. If flags are given 589 * then we send a message back to the TCP which originated the 590 * segment ti, and discard the mbuf containing it and any other 591 * attached mbufs. 592 * 593 * In any case the ack and sequence number of the transmitted 594 * segment are as specified by the parameters. 595 */ 596 int 597 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 598 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 599 { 600 #ifdef INET6 601 struct rtentry *rt; 602 #endif 603 struct route *ro; 604 int error, tlen, win = 0; 605 int hlen; 606 struct ip *ip; 607 #ifdef INET6 608 struct ip6_hdr *ip6; 609 #endif 610 int family; /* family on packet, not inpcb/in6pcb! */ 611 struct tcphdr *th; 612 struct socket *so; 613 614 if (tp != NULL && (flags & TH_RST) == 0) { 615 #ifdef DIAGNOSTIC 616 if (tp->t_inpcb && tp->t_in6pcb) 617 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 618 #endif 619 #ifdef INET 620 if (tp->t_inpcb) 621 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 622 #endif 623 #ifdef INET6 624 if (tp->t_in6pcb) 625 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 626 #endif 627 } 628 629 th = NULL; /* Quell uninitialized warning */ 630 ip = NULL; 631 #ifdef INET6 632 ip6 = NULL; 633 #endif 634 if (m == 0) { 635 if (!template) 636 return EINVAL; 637 638 /* get family information from template */ 639 switch (mtod(template, struct ip *)->ip_v) { 640 case 4: 641 family = AF_INET; 642 hlen = sizeof(struct ip); 643 break; 644 #ifdef INET6 645 case 6: 646 family = AF_INET6; 647 hlen = sizeof(struct ip6_hdr); 648 break; 649 #endif 650 default: 651 return EAFNOSUPPORT; 652 } 653 654 MGETHDR(m, M_DONTWAIT, MT_HEADER); 655 if (m) { 656 MCLAIM(m, &tcp_tx_mowner); 657 MCLGET(m, M_DONTWAIT); 658 if ((m->m_flags & M_EXT) == 0) { 659 m_free(m); 660 m = NULL; 661 } 662 } 663 if (m == NULL) 664 return (ENOBUFS); 665 666 if (tcp_compat_42) 667 tlen = 1; 668 else 669 tlen = 0; 670 671 m->m_data += max_linkhdr; 672 bcopy(mtod(template, void *), mtod(m, void *), 673 template->m_len); 674 switch (family) { 675 case AF_INET: 676 ip = mtod(m, struct ip *); 677 th = (struct tcphdr *)(ip + 1); 678 break; 679 #ifdef INET6 680 case AF_INET6: 681 ip6 = mtod(m, struct ip6_hdr *); 682 th = (struct tcphdr *)(ip6 + 1); 683 break; 684 #endif 685 #if 0 686 default: 687 /* noone will visit here */ 688 m_freem(m); 689 return EAFNOSUPPORT; 690 #endif 691 } 692 flags = TH_ACK; 693 } else { 694 695 if ((m->m_flags & M_PKTHDR) == 0) { 696 #if 0 697 printf("non PKTHDR to tcp_respond\n"); 698 #endif 699 m_freem(m); 700 return EINVAL; 701 } 702 #ifdef DIAGNOSTIC 703 if (!th0) 704 panic("th0 == NULL in tcp_respond"); 705 #endif 706 707 /* get family information from m */ 708 switch (mtod(m, struct ip *)->ip_v) { 709 case 4: 710 family = AF_INET; 711 hlen = sizeof(struct ip); 712 ip = mtod(m, struct ip *); 713 break; 714 #ifdef INET6 715 case 6: 716 family = AF_INET6; 717 hlen = sizeof(struct ip6_hdr); 718 ip6 = mtod(m, struct ip6_hdr *); 719 break; 720 #endif 721 default: 722 m_freem(m); 723 return EAFNOSUPPORT; 724 } 725 /* clear h/w csum flags inherited from rx packet */ 726 m->m_pkthdr.csum_flags = 0; 727 728 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 729 tlen = sizeof(*th0); 730 else 731 tlen = th0->th_off << 2; 732 733 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 734 mtod(m, char *) + hlen == (char *)th0) { 735 m->m_len = hlen + tlen; 736 m_freem(m->m_next); 737 m->m_next = NULL; 738 } else { 739 struct mbuf *n; 740 741 #ifdef DIAGNOSTIC 742 if (max_linkhdr + hlen + tlen > MCLBYTES) { 743 m_freem(m); 744 return EMSGSIZE; 745 } 746 #endif 747 MGETHDR(n, M_DONTWAIT, MT_HEADER); 748 if (n && max_linkhdr + hlen + tlen > MHLEN) { 749 MCLGET(n, M_DONTWAIT); 750 if ((n->m_flags & M_EXT) == 0) { 751 m_freem(n); 752 n = NULL; 753 } 754 } 755 if (!n) { 756 m_freem(m); 757 return ENOBUFS; 758 } 759 760 MCLAIM(n, &tcp_tx_mowner); 761 n->m_data += max_linkhdr; 762 n->m_len = hlen + tlen; 763 m_copyback(n, 0, hlen, mtod(m, void *)); 764 m_copyback(n, hlen, tlen, (void *)th0); 765 766 m_freem(m); 767 m = n; 768 n = NULL; 769 } 770 771 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 772 switch (family) { 773 case AF_INET: 774 ip = mtod(m, struct ip *); 775 th = (struct tcphdr *)(ip + 1); 776 ip->ip_p = IPPROTO_TCP; 777 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 778 ip->ip_p = IPPROTO_TCP; 779 break; 780 #ifdef INET6 781 case AF_INET6: 782 ip6 = mtod(m, struct ip6_hdr *); 783 th = (struct tcphdr *)(ip6 + 1); 784 ip6->ip6_nxt = IPPROTO_TCP; 785 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 786 ip6->ip6_nxt = IPPROTO_TCP; 787 break; 788 #endif 789 #if 0 790 default: 791 /* noone will visit here */ 792 m_freem(m); 793 return EAFNOSUPPORT; 794 #endif 795 } 796 xchg(th->th_dport, th->th_sport, u_int16_t); 797 #undef xchg 798 tlen = 0; /*be friendly with the following code*/ 799 } 800 th->th_seq = htonl(seq); 801 th->th_ack = htonl(ack); 802 th->th_x2 = 0; 803 if ((flags & TH_SYN) == 0) { 804 if (tp) 805 win >>= tp->rcv_scale; 806 if (win > TCP_MAXWIN) 807 win = TCP_MAXWIN; 808 th->th_win = htons((u_int16_t)win); 809 th->th_off = sizeof (struct tcphdr) >> 2; 810 tlen += sizeof(*th); 811 } else 812 tlen += th->th_off << 2; 813 m->m_len = hlen + tlen; 814 m->m_pkthdr.len = hlen + tlen; 815 m->m_pkthdr.rcvif = (struct ifnet *) 0; 816 th->th_flags = flags; 817 th->th_urp = 0; 818 819 switch (family) { 820 #ifdef INET 821 case AF_INET: 822 { 823 struct ipovly *ipov = (struct ipovly *)ip; 824 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 825 ipov->ih_len = htons((u_int16_t)tlen); 826 827 th->th_sum = 0; 828 th->th_sum = in_cksum(m, hlen + tlen); 829 ip->ip_len = htons(hlen + tlen); 830 ip->ip_ttl = ip_defttl; 831 break; 832 } 833 #endif 834 #ifdef INET6 835 case AF_INET6: 836 { 837 th->th_sum = 0; 838 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 839 tlen); 840 ip6->ip6_plen = htons(tlen); 841 if (tp && tp->t_in6pcb) { 842 struct ifnet *oifp; 843 ro = &tp->t_in6pcb->in6p_route; 844 oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 845 : NULL; 846 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 847 } else 848 ip6->ip6_hlim = ip6_defhlim; 849 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 850 if (ip6_auto_flowlabel) { 851 ip6->ip6_flow |= 852 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 853 } 854 break; 855 } 856 #endif 857 } 858 859 if (tp && tp->t_inpcb) 860 so = tp->t_inpcb->inp_socket; 861 #ifdef INET6 862 else if (tp && tp->t_in6pcb) 863 so = tp->t_in6pcb->in6p_socket; 864 #endif 865 else 866 so = NULL; 867 868 if (tp != NULL && tp->t_inpcb != NULL) { 869 ro = &tp->t_inpcb->inp_route; 870 #ifdef DIAGNOSTIC 871 if (family != AF_INET) 872 panic("tcp_respond: address family mismatch"); 873 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 874 panic("tcp_respond: ip_dst %x != inp_faddr %x", 875 ntohl(ip->ip_dst.s_addr), 876 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 877 } 878 #endif 879 } 880 #ifdef INET6 881 else if (tp != NULL && tp->t_in6pcb != NULL) { 882 ro = (struct route *)&tp->t_in6pcb->in6p_route; 883 #ifdef DIAGNOSTIC 884 if (family == AF_INET) { 885 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 886 panic("tcp_respond: not mapped addr"); 887 if (bcmp(&ip->ip_dst, 888 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 889 sizeof(ip->ip_dst)) != 0) { 890 panic("tcp_respond: ip_dst != in6p_faddr"); 891 } 892 } else if (family == AF_INET6) { 893 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 894 &tp->t_in6pcb->in6p_faddr)) 895 panic("tcp_respond: ip6_dst != in6p_faddr"); 896 } else 897 panic("tcp_respond: address family mismatch"); 898 #endif 899 } 900 #endif 901 else 902 ro = NULL; 903 904 switch (family) { 905 #ifdef INET 906 case AF_INET: 907 error = ip_output(m, NULL, ro, 908 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 909 (struct ip_moptions *)0, so); 910 break; 911 #endif 912 #ifdef INET6 913 case AF_INET6: 914 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL); 915 break; 916 #endif 917 default: 918 error = EAFNOSUPPORT; 919 break; 920 } 921 922 return (error); 923 } 924 925 /* 926 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 927 * a bunch of members individually, we maintain this template for the 928 * static and mostly-static components of the TCPCB, and copy it into 929 * the new TCPCB instead. 930 */ 931 static struct tcpcb tcpcb_template = { 932 .t_srtt = TCPTV_SRTTBASE, 933 .t_rttmin = TCPTV_MIN, 934 935 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 936 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 937 .snd_numholes = 0, 938 939 .t_partialacks = -1, 940 .t_bytes_acked = 0, 941 }; 942 943 /* 944 * Updates the TCPCB template whenever a parameter that would affect 945 * the template is changed. 946 */ 947 void 948 tcp_tcpcb_template(void) 949 { 950 struct tcpcb *tp = &tcpcb_template; 951 int flags; 952 953 tp->t_peermss = tcp_mssdflt; 954 tp->t_ourmss = tcp_mssdflt; 955 tp->t_segsz = tcp_mssdflt; 956 957 flags = 0; 958 if (tcp_do_rfc1323 && tcp_do_win_scale) 959 flags |= TF_REQ_SCALE; 960 if (tcp_do_rfc1323 && tcp_do_timestamps) 961 flags |= TF_REQ_TSTMP; 962 tp->t_flags = flags; 963 964 /* 965 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 966 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 967 * reasonable initial retransmit time. 968 */ 969 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 970 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 971 TCPTV_MIN, TCPTV_REXMTMAX); 972 973 /* Keep Alive */ 974 tp->t_keepinit = tcp_keepinit; 975 tp->t_keepidle = tcp_keepidle; 976 tp->t_keepintvl = tcp_keepintvl; 977 tp->t_keepcnt = tcp_keepcnt; 978 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 979 } 980 981 /* 982 * Create a new TCP control block, making an 983 * empty reassembly queue and hooking it to the argument 984 * protocol control block. 985 */ 986 /* family selects inpcb, or in6pcb */ 987 struct tcpcb * 988 tcp_newtcpcb(int family, void *aux) 989 { 990 #ifdef INET6 991 struct rtentry *rt; 992 #endif 993 struct tcpcb *tp; 994 int i; 995 996 /* XXX Consider using a pool_cache for speed. */ 997 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 998 if (tp == NULL) 999 return (NULL); 1000 memcpy(tp, &tcpcb_template, sizeof(*tp)); 1001 TAILQ_INIT(&tp->segq); 1002 TAILQ_INIT(&tp->timeq); 1003 tp->t_family = family; /* may be overridden later on */ 1004 TAILQ_INIT(&tp->snd_holes); 1005 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1006 1007 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1008 for (i = 0; i < TCPT_NTIMERS; i++) { 1009 callout_init(&tp->t_timer[i], 0); 1010 TCP_TIMER_INIT(tp, i); 1011 } 1012 callout_init(&tp->t_delack_ch, 0); 1013 1014 switch (family) { 1015 case AF_INET: 1016 { 1017 struct inpcb *inp = (struct inpcb *)aux; 1018 1019 inp->inp_ip.ip_ttl = ip_defttl; 1020 inp->inp_ppcb = (void *)tp; 1021 1022 tp->t_inpcb = inp; 1023 tp->t_mtudisc = ip_mtudisc; 1024 break; 1025 } 1026 #ifdef INET6 1027 case AF_INET6: 1028 { 1029 struct in6pcb *in6p = (struct in6pcb *)aux; 1030 1031 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1032 (rt = rtcache_validate(&in6p->in6p_route)) != NULL 1033 ? rt->rt_ifp 1034 : NULL); 1035 in6p->in6p_ppcb = (void *)tp; 1036 1037 tp->t_in6pcb = in6p; 1038 /* for IPv6, always try to run path MTU discovery */ 1039 tp->t_mtudisc = 1; 1040 break; 1041 } 1042 #endif /* INET6 */ 1043 default: 1044 for (i = 0; i < TCPT_NTIMERS; i++) 1045 callout_destroy(&tp->t_timer[i]); 1046 callout_destroy(&tp->t_delack_ch); 1047 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1048 return (NULL); 1049 } 1050 1051 /* 1052 * Initialize our timebase. When we send timestamps, we take 1053 * the delta from tcp_now -- this means each connection always 1054 * gets a timebase of 0, which makes it, among other things, 1055 * more difficult to determine how long a system has been up, 1056 * and thus how many TCP sequence increments have occurred. 1057 */ 1058 tp->ts_timebase = tcp_now; 1059 1060 tp->t_congctl = tcp_congctl_global; 1061 tp->t_congctl->refcnt++; 1062 1063 return (tp); 1064 } 1065 1066 /* 1067 * Drop a TCP connection, reporting 1068 * the specified error. If connection is synchronized, 1069 * then send a RST to peer. 1070 */ 1071 struct tcpcb * 1072 tcp_drop(struct tcpcb *tp, int errno) 1073 { 1074 struct socket *so = NULL; 1075 1076 #ifdef DIAGNOSTIC 1077 if (tp->t_inpcb && tp->t_in6pcb) 1078 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1079 #endif 1080 #ifdef INET 1081 if (tp->t_inpcb) 1082 so = tp->t_inpcb->inp_socket; 1083 #endif 1084 #ifdef INET6 1085 if (tp->t_in6pcb) 1086 so = tp->t_in6pcb->in6p_socket; 1087 #endif 1088 if (!so) 1089 return NULL; 1090 1091 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1092 tp->t_state = TCPS_CLOSED; 1093 (void) tcp_output(tp); 1094 tcpstat.tcps_drops++; 1095 } else 1096 tcpstat.tcps_conndrops++; 1097 if (errno == ETIMEDOUT && tp->t_softerror) 1098 errno = tp->t_softerror; 1099 so->so_error = errno; 1100 return (tcp_close(tp)); 1101 } 1102 1103 /* 1104 * Return whether this tcpcb is marked as dead, indicating 1105 * to the calling timer function that no further action should 1106 * be taken, as we are about to release this tcpcb. The release 1107 * of the storage will be done if this is the last timer running. 1108 * 1109 * This should be called from the callout handler function after 1110 * callout_ack() is done, so that the number of invoking timer 1111 * functions is 0. 1112 */ 1113 int 1114 tcp_isdead(struct tcpcb *tp) 1115 { 1116 int i, dead = (tp->t_flags & TF_DEAD); 1117 1118 if (__predict_false(dead)) { 1119 if (tcp_timers_invoking(tp) > 0) 1120 /* not quite there yet -- count separately? */ 1121 return dead; 1122 tcpstat.tcps_delayed_free++; 1123 for (i = 0; i < TCPT_NTIMERS; i++) 1124 callout_destroy(&tp->t_timer[i]); 1125 callout_destroy(&tp->t_delack_ch); 1126 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_timer.c */ 1127 } 1128 return dead; 1129 } 1130 1131 /* 1132 * Close a TCP control block: 1133 * discard all space held by the tcp 1134 * discard internet protocol block 1135 * wake up any sleepers 1136 */ 1137 struct tcpcb * 1138 tcp_close(struct tcpcb *tp) 1139 { 1140 struct inpcb *inp; 1141 #ifdef INET6 1142 struct in6pcb *in6p; 1143 #endif 1144 struct socket *so; 1145 #ifdef RTV_RTT 1146 struct rtentry *rt; 1147 #endif 1148 struct route *ro; 1149 int j; 1150 1151 inp = tp->t_inpcb; 1152 #ifdef INET6 1153 in6p = tp->t_in6pcb; 1154 #endif 1155 so = NULL; 1156 ro = NULL; 1157 if (inp) { 1158 so = inp->inp_socket; 1159 ro = &inp->inp_route; 1160 } 1161 #ifdef INET6 1162 else if (in6p) { 1163 so = in6p->in6p_socket; 1164 ro = (struct route *)&in6p->in6p_route; 1165 } 1166 #endif 1167 1168 #ifdef RTV_RTT 1169 /* 1170 * If we sent enough data to get some meaningful characteristics, 1171 * save them in the routing entry. 'Enough' is arbitrarily 1172 * defined as the sendpipesize (default 4K) * 16. This would 1173 * give us 16 rtt samples assuming we only get one sample per 1174 * window (the usual case on a long haul net). 16 samples is 1175 * enough for the srtt filter to converge to within 5% of the correct 1176 * value; fewer samples and we could save a very bogus rtt. 1177 * 1178 * Don't update the default route's characteristics and don't 1179 * update anything that the user "locked". 1180 */ 1181 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1182 ro && (rt = rtcache_validate(ro)) != NULL && 1183 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1184 u_long i = 0; 1185 1186 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1187 i = tp->t_srtt * 1188 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1189 if (rt->rt_rmx.rmx_rtt && i) 1190 /* 1191 * filter this update to half the old & half 1192 * the new values, converting scale. 1193 * See route.h and tcp_var.h for a 1194 * description of the scaling constants. 1195 */ 1196 rt->rt_rmx.rmx_rtt = 1197 (rt->rt_rmx.rmx_rtt + i) / 2; 1198 else 1199 rt->rt_rmx.rmx_rtt = i; 1200 } 1201 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1202 i = tp->t_rttvar * 1203 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1204 if (rt->rt_rmx.rmx_rttvar && i) 1205 rt->rt_rmx.rmx_rttvar = 1206 (rt->rt_rmx.rmx_rttvar + i) / 2; 1207 else 1208 rt->rt_rmx.rmx_rttvar = i; 1209 } 1210 /* 1211 * update the pipelimit (ssthresh) if it has been updated 1212 * already or if a pipesize was specified & the threshhold 1213 * got below half the pipesize. I.e., wait for bad news 1214 * before we start updating, then update on both good 1215 * and bad news. 1216 */ 1217 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1218 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1219 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1220 /* 1221 * convert the limit from user data bytes to 1222 * packets then to packet data bytes. 1223 */ 1224 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1225 if (i < 2) 1226 i = 2; 1227 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1228 if (rt->rt_rmx.rmx_ssthresh) 1229 rt->rt_rmx.rmx_ssthresh = 1230 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1231 else 1232 rt->rt_rmx.rmx_ssthresh = i; 1233 } 1234 } 1235 #endif /* RTV_RTT */ 1236 /* free the reassembly queue, if any */ 1237 TCP_REASS_LOCK(tp); 1238 (void) tcp_freeq(tp); 1239 TCP_REASS_UNLOCK(tp); 1240 1241 /* free the SACK holes list. */ 1242 tcp_free_sackholes(tp); 1243 1244 tp->t_congctl->refcnt--; 1245 1246 tcp_canceltimers(tp); 1247 TCP_CLEAR_DELACK(tp); 1248 syn_cache_cleanup(tp); 1249 1250 if (tp->t_template) { 1251 m_free(tp->t_template); 1252 tp->t_template = NULL; 1253 } 1254 if (tcp_timers_invoking(tp)) 1255 tp->t_flags |= TF_DEAD; 1256 else { 1257 for (j = 0; j < TCPT_NTIMERS; j++) 1258 callout_destroy(&tp->t_timer[j]); 1259 callout_destroy(&tp->t_delack_ch); 1260 pool_put(&tcpcb_pool, tp); 1261 } 1262 1263 if (inp) { 1264 inp->inp_ppcb = 0; 1265 soisdisconnected(so); 1266 in_pcbdetach(inp); 1267 } 1268 #ifdef INET6 1269 else if (in6p) { 1270 in6p->in6p_ppcb = 0; 1271 soisdisconnected(so); 1272 in6_pcbdetach(in6p); 1273 } 1274 #endif 1275 tcpstat.tcps_closed++; 1276 return ((struct tcpcb *)0); 1277 } 1278 1279 int 1280 tcp_freeq(tp) 1281 struct tcpcb *tp; 1282 { 1283 struct ipqent *qe; 1284 int rv = 0; 1285 #ifdef TCPREASS_DEBUG 1286 int i = 0; 1287 #endif 1288 1289 TCP_REASS_LOCK_CHECK(tp); 1290 1291 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1292 #ifdef TCPREASS_DEBUG 1293 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1294 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1295 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1296 #endif 1297 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1298 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1299 m_freem(qe->ipqe_m); 1300 tcpipqent_free(qe); 1301 rv = 1; 1302 } 1303 tp->t_segqlen = 0; 1304 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1305 return (rv); 1306 } 1307 1308 /* 1309 * Protocol drain routine. Called when memory is in short supply. 1310 */ 1311 void 1312 tcp_drain(void) 1313 { 1314 struct inpcb_hdr *inph; 1315 struct tcpcb *tp; 1316 1317 /* 1318 * Free the sequence queue of all TCP connections. 1319 */ 1320 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1321 switch (inph->inph_af) { 1322 case AF_INET: 1323 tp = intotcpcb((struct inpcb *)inph); 1324 break; 1325 #ifdef INET6 1326 case AF_INET6: 1327 tp = in6totcpcb((struct in6pcb *)inph); 1328 break; 1329 #endif 1330 default: 1331 tp = NULL; 1332 break; 1333 } 1334 if (tp != NULL) { 1335 /* 1336 * We may be called from a device's interrupt 1337 * context. If the tcpcb is already busy, 1338 * just bail out now. 1339 */ 1340 if (tcp_reass_lock_try(tp) == 0) 1341 continue; 1342 if (tcp_freeq(tp)) 1343 tcpstat.tcps_connsdrained++; 1344 TCP_REASS_UNLOCK(tp); 1345 } 1346 } 1347 } 1348 1349 /* 1350 * Notify a tcp user of an asynchronous error; 1351 * store error as soft error, but wake up user 1352 * (for now, won't do anything until can select for soft error). 1353 */ 1354 void 1355 tcp_notify(struct inpcb *inp, int error) 1356 { 1357 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1358 struct socket *so = inp->inp_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((void *) &so->so_timeo); 1377 sorwakeup(so); 1378 sowwakeup(so); 1379 } 1380 1381 #ifdef INET6 1382 void 1383 tcp6_notify(struct in6pcb *in6p, int error) 1384 { 1385 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1386 struct socket *so = in6p->in6p_socket; 1387 1388 /* 1389 * Ignore some errors if we are hooked up. 1390 * If connection hasn't completed, has retransmitted several times, 1391 * and receives a second error, give up now. This is better 1392 * than waiting a long time to establish a connection that 1393 * can never complete. 1394 */ 1395 if (tp->t_state == TCPS_ESTABLISHED && 1396 (error == EHOSTUNREACH || error == ENETUNREACH || 1397 error == EHOSTDOWN)) { 1398 return; 1399 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1400 tp->t_rxtshift > 3 && tp->t_softerror) 1401 so->so_error = error; 1402 else 1403 tp->t_softerror = error; 1404 wakeup((void *) &so->so_timeo); 1405 sorwakeup(so); 1406 sowwakeup(so); 1407 } 1408 #endif 1409 1410 #ifdef INET6 1411 void 1412 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1413 { 1414 struct tcphdr th; 1415 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1416 int nmatch; 1417 struct ip6_hdr *ip6; 1418 const struct sockaddr_in6 *sa6_src = NULL; 1419 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1420 struct mbuf *m; 1421 int off; 1422 1423 if (sa->sa_family != AF_INET6 || 1424 sa->sa_len != sizeof(struct sockaddr_in6)) 1425 return; 1426 if ((unsigned)cmd >= PRC_NCMDS) 1427 return; 1428 else if (cmd == PRC_QUENCH) { 1429 /* 1430 * Don't honor ICMP Source Quench messages meant for 1431 * TCP connections. 1432 */ 1433 return; 1434 } else if (PRC_IS_REDIRECT(cmd)) 1435 notify = in6_rtchange, d = NULL; 1436 else if (cmd == PRC_MSGSIZE) 1437 ; /* special code is present, see below */ 1438 else if (cmd == PRC_HOSTDEAD) 1439 d = NULL; 1440 else if (inet6ctlerrmap[cmd] == 0) 1441 return; 1442 1443 /* if the parameter is from icmp6, decode it. */ 1444 if (d != NULL) { 1445 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1446 m = ip6cp->ip6c_m; 1447 ip6 = ip6cp->ip6c_ip6; 1448 off = ip6cp->ip6c_off; 1449 sa6_src = ip6cp->ip6c_src; 1450 } else { 1451 m = NULL; 1452 ip6 = NULL; 1453 sa6_src = &sa6_any; 1454 off = 0; 1455 } 1456 1457 if (ip6) { 1458 /* 1459 * XXX: We assume that when ip6 is non NULL, 1460 * M and OFF are valid. 1461 */ 1462 1463 /* check if we can safely examine src and dst ports */ 1464 if (m->m_pkthdr.len < off + sizeof(th)) { 1465 if (cmd == PRC_MSGSIZE) 1466 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1467 return; 1468 } 1469 1470 bzero(&th, sizeof(th)); 1471 m_copydata(m, off, sizeof(th), (void *)&th); 1472 1473 if (cmd == PRC_MSGSIZE) { 1474 int valid = 0; 1475 1476 /* 1477 * Check to see if we have a valid TCP connection 1478 * corresponding to the address in the ICMPv6 message 1479 * payload. 1480 */ 1481 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1482 th.th_dport, 1483 (const struct in6_addr *)&sa6_src->sin6_addr, 1484 th.th_sport, 0)) 1485 valid++; 1486 1487 /* 1488 * Depending on the value of "valid" and routing table 1489 * size (mtudisc_{hi,lo}wat), we will: 1490 * - recalcurate the new MTU and create the 1491 * corresponding routing entry, or 1492 * - ignore the MTU change notification. 1493 */ 1494 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1495 1496 /* 1497 * no need to call in6_pcbnotify, it should have been 1498 * called via callback if necessary 1499 */ 1500 return; 1501 } 1502 1503 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1504 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1505 if (nmatch == 0 && syn_cache_count && 1506 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1507 inet6ctlerrmap[cmd] == ENETUNREACH || 1508 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1509 syn_cache_unreach((const struct sockaddr *)sa6_src, 1510 sa, &th); 1511 } else { 1512 (void) in6_pcbnotify(&tcbtable, sa, 0, 1513 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1514 } 1515 } 1516 #endif 1517 1518 #ifdef INET 1519 /* assumes that ip header and tcp header are contiguous on mbuf */ 1520 void * 1521 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1522 { 1523 struct ip *ip = v; 1524 struct tcphdr *th; 1525 struct icmp *icp; 1526 extern const int inetctlerrmap[]; 1527 void (*notify)(struct inpcb *, int) = tcp_notify; 1528 int errno; 1529 int nmatch; 1530 struct tcpcb *tp; 1531 u_int mtu; 1532 tcp_seq seq; 1533 struct inpcb *inp; 1534 #ifdef INET6 1535 struct in6pcb *in6p; 1536 struct in6_addr src6, dst6; 1537 #endif 1538 1539 if (sa->sa_family != AF_INET || 1540 sa->sa_len != sizeof(struct sockaddr_in)) 1541 return NULL; 1542 if ((unsigned)cmd >= PRC_NCMDS) 1543 return NULL; 1544 errno = inetctlerrmap[cmd]; 1545 if (cmd == PRC_QUENCH) 1546 /* 1547 * Don't honor ICMP Source Quench messages meant for 1548 * TCP connections. 1549 */ 1550 return NULL; 1551 else if (PRC_IS_REDIRECT(cmd)) 1552 notify = in_rtchange, ip = 0; 1553 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1554 /* 1555 * Check to see if we have a valid TCP connection 1556 * corresponding to the address in the ICMP message 1557 * payload. 1558 * 1559 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1560 */ 1561 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1562 #ifdef INET6 1563 memset(&src6, 0, sizeof(src6)); 1564 memset(&dst6, 0, sizeof(dst6)); 1565 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1566 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1567 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1568 #endif 1569 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1570 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1571 #ifdef INET6 1572 in6p = NULL; 1573 #else 1574 ; 1575 #endif 1576 #ifdef INET6 1577 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1578 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1579 ; 1580 #endif 1581 else 1582 return NULL; 1583 1584 /* 1585 * Now that we've validated that we are actually communicating 1586 * with the host indicated in the ICMP message, locate the 1587 * ICMP header, recalculate the new MTU, and create the 1588 * corresponding routing entry. 1589 */ 1590 icp = (struct icmp *)((char *)ip - 1591 offsetof(struct icmp, icmp_ip)); 1592 if (inp) { 1593 if ((tp = intotcpcb(inp)) == NULL) 1594 return NULL; 1595 } 1596 #ifdef INET6 1597 else if (in6p) { 1598 if ((tp = in6totcpcb(in6p)) == NULL) 1599 return NULL; 1600 } 1601 #endif 1602 else 1603 return NULL; 1604 seq = ntohl(th->th_seq); 1605 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1606 return NULL; 1607 /* 1608 * If the ICMP message advertises a Next-Hop MTU 1609 * equal or larger than the maximum packet size we have 1610 * ever sent, drop the message. 1611 */ 1612 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1613 if (mtu >= tp->t_pmtud_mtu_sent) 1614 return NULL; 1615 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1616 /* 1617 * Calculate new MTU, and create corresponding 1618 * route (traditional PMTUD). 1619 */ 1620 tp->t_flags &= ~TF_PMTUD_PEND; 1621 icmp_mtudisc(icp, ip->ip_dst); 1622 } else { 1623 /* 1624 * Record the information got in the ICMP 1625 * message; act on it later. 1626 * If we had already recorded an ICMP message, 1627 * replace the old one only if the new message 1628 * refers to an older TCP segment 1629 */ 1630 if (tp->t_flags & TF_PMTUD_PEND) { 1631 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1632 return NULL; 1633 } else 1634 tp->t_flags |= TF_PMTUD_PEND; 1635 tp->t_pmtud_th_seq = seq; 1636 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1637 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1638 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1639 } 1640 return NULL; 1641 } else if (cmd == PRC_HOSTDEAD) 1642 ip = 0; 1643 else if (errno == 0) 1644 return NULL; 1645 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1646 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1647 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1648 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1649 if (nmatch == 0 && syn_cache_count && 1650 (inetctlerrmap[cmd] == EHOSTUNREACH || 1651 inetctlerrmap[cmd] == ENETUNREACH || 1652 inetctlerrmap[cmd] == EHOSTDOWN)) { 1653 struct sockaddr_in sin; 1654 bzero(&sin, sizeof(sin)); 1655 sin.sin_len = sizeof(sin); 1656 sin.sin_family = AF_INET; 1657 sin.sin_port = th->th_sport; 1658 sin.sin_addr = ip->ip_src; 1659 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1660 } 1661 1662 /* XXX mapped address case */ 1663 } else 1664 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1665 notify); 1666 return NULL; 1667 } 1668 1669 /* 1670 * When a source quench is received, we are being notified of congestion. 1671 * Close the congestion window down to the Loss Window (one segment). 1672 * We will gradually open it again as we proceed. 1673 */ 1674 void 1675 tcp_quench(struct inpcb *inp, int errno) 1676 { 1677 struct tcpcb *tp = intotcpcb(inp); 1678 1679 if (tp) { 1680 tp->snd_cwnd = tp->t_segsz; 1681 tp->t_bytes_acked = 0; 1682 } 1683 } 1684 #endif 1685 1686 #ifdef INET6 1687 void 1688 tcp6_quench(struct in6pcb *in6p, int errno) 1689 { 1690 struct tcpcb *tp = in6totcpcb(in6p); 1691 1692 if (tp) { 1693 tp->snd_cwnd = tp->t_segsz; 1694 tp->t_bytes_acked = 0; 1695 } 1696 } 1697 #endif 1698 1699 #ifdef INET 1700 /* 1701 * Path MTU Discovery handlers. 1702 */ 1703 void 1704 tcp_mtudisc_callback(struct in_addr faddr) 1705 { 1706 #ifdef INET6 1707 struct in6_addr in6; 1708 #endif 1709 1710 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1711 #ifdef INET6 1712 memset(&in6, 0, sizeof(in6)); 1713 in6.s6_addr16[5] = 0xffff; 1714 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1715 tcp6_mtudisc_callback(&in6); 1716 #endif 1717 } 1718 1719 /* 1720 * On receipt of path MTU corrections, flush old route and replace it 1721 * with the new one. Retransmit all unacknowledged packets, to ensure 1722 * that all packets will be received. 1723 */ 1724 void 1725 tcp_mtudisc(struct inpcb *inp, int errno) 1726 { 1727 struct tcpcb *tp = intotcpcb(inp); 1728 struct rtentry *rt = in_pcbrtentry(inp); 1729 1730 if (tp != 0) { 1731 if (rt != 0) { 1732 /* 1733 * If this was not a host route, remove and realloc. 1734 */ 1735 if ((rt->rt_flags & RTF_HOST) == 0) { 1736 in_rtchange(inp, errno); 1737 if ((rt = in_pcbrtentry(inp)) == 0) 1738 return; 1739 } 1740 1741 /* 1742 * Slow start out of the error condition. We 1743 * use the MTU because we know it's smaller 1744 * than the previously transmitted segment. 1745 * 1746 * Note: This is more conservative than the 1747 * suggestion in draft-floyd-incr-init-win-03. 1748 */ 1749 if (rt->rt_rmx.rmx_mtu != 0) 1750 tp->snd_cwnd = 1751 TCP_INITIAL_WINDOW(tcp_init_win, 1752 rt->rt_rmx.rmx_mtu); 1753 } 1754 1755 /* 1756 * Resend unacknowledged packets. 1757 */ 1758 tp->snd_nxt = tp->snd_una; 1759 tcp_output(tp); 1760 } 1761 } 1762 #endif 1763 1764 #ifdef INET6 1765 /* 1766 * Path MTU Discovery handlers. 1767 */ 1768 void 1769 tcp6_mtudisc_callback(struct in6_addr *faddr) 1770 { 1771 struct sockaddr_in6 sin6; 1772 1773 bzero(&sin6, sizeof(sin6)); 1774 sin6.sin6_family = AF_INET6; 1775 sin6.sin6_len = sizeof(struct sockaddr_in6); 1776 sin6.sin6_addr = *faddr; 1777 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1778 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1779 } 1780 1781 void 1782 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1783 { 1784 struct tcpcb *tp = in6totcpcb(in6p); 1785 struct rtentry *rt = in6_pcbrtentry(in6p); 1786 1787 if (tp != 0) { 1788 if (rt != 0) { 1789 /* 1790 * If this was not a host route, remove and realloc. 1791 */ 1792 if ((rt->rt_flags & RTF_HOST) == 0) { 1793 in6_rtchange(in6p, errno); 1794 if ((rt = in6_pcbrtentry(in6p)) == 0) 1795 return; 1796 } 1797 1798 /* 1799 * Slow start out of the error condition. We 1800 * use the MTU because we know it's smaller 1801 * than the previously transmitted segment. 1802 * 1803 * Note: This is more conservative than the 1804 * suggestion in draft-floyd-incr-init-win-03. 1805 */ 1806 if (rt->rt_rmx.rmx_mtu != 0) 1807 tp->snd_cwnd = 1808 TCP_INITIAL_WINDOW(tcp_init_win, 1809 rt->rt_rmx.rmx_mtu); 1810 } 1811 1812 /* 1813 * Resend unacknowledged packets. 1814 */ 1815 tp->snd_nxt = tp->snd_una; 1816 tcp_output(tp); 1817 } 1818 } 1819 #endif /* INET6 */ 1820 1821 /* 1822 * Compute the MSS to advertise to the peer. Called only during 1823 * the 3-way handshake. If we are the server (peer initiated 1824 * connection), we are called with a pointer to the interface 1825 * on which the SYN packet arrived. If we are the client (we 1826 * initiated connection), we are called with a pointer to the 1827 * interface out which this connection should go. 1828 * 1829 * NOTE: Do not subtract IP option/extension header size nor IPsec 1830 * header size from MSS advertisement. MSS option must hold the maximum 1831 * segment size we can accept, so it must always be: 1832 * max(if mtu) - ip header - tcp header 1833 */ 1834 u_long 1835 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1836 { 1837 extern u_long in_maxmtu; 1838 u_long mss = 0; 1839 u_long hdrsiz; 1840 1841 /* 1842 * In order to avoid defeating path MTU discovery on the peer, 1843 * we advertise the max MTU of all attached networks as our MSS, 1844 * per RFC 1191, section 3.1. 1845 * 1846 * We provide the option to advertise just the MTU of 1847 * the interface on which we hope this connection will 1848 * be receiving. If we are responding to a SYN, we 1849 * will have a pretty good idea about this, but when 1850 * initiating a connection there is a bit more doubt. 1851 * 1852 * We also need to ensure that loopback has a large enough 1853 * MSS, as the loopback MTU is never included in in_maxmtu. 1854 */ 1855 1856 if (ifp != NULL) 1857 switch (af) { 1858 case AF_INET: 1859 mss = ifp->if_mtu; 1860 break; 1861 #ifdef INET6 1862 case AF_INET6: 1863 mss = IN6_LINKMTU(ifp); 1864 break; 1865 #endif 1866 } 1867 1868 if (tcp_mss_ifmtu == 0) 1869 switch (af) { 1870 case AF_INET: 1871 mss = max(in_maxmtu, mss); 1872 break; 1873 #ifdef INET6 1874 case AF_INET6: 1875 mss = max(in6_maxmtu, mss); 1876 break; 1877 #endif 1878 } 1879 1880 switch (af) { 1881 case AF_INET: 1882 hdrsiz = sizeof(struct ip); 1883 break; 1884 #ifdef INET6 1885 case AF_INET6: 1886 hdrsiz = sizeof(struct ip6_hdr); 1887 break; 1888 #endif 1889 default: 1890 hdrsiz = 0; 1891 break; 1892 } 1893 hdrsiz += sizeof(struct tcphdr); 1894 if (mss > hdrsiz) 1895 mss -= hdrsiz; 1896 1897 mss = max(tcp_mssdflt, mss); 1898 return (mss); 1899 } 1900 1901 /* 1902 * Set connection variables based on the peer's advertised MSS. 1903 * We are passed the TCPCB for the actual connection. If we 1904 * are the server, we are called by the compressed state engine 1905 * when the 3-way handshake is complete. If we are the client, 1906 * we are called when we receive the SYN,ACK from the server. 1907 * 1908 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1909 * before this routine is called! 1910 */ 1911 void 1912 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1913 { 1914 struct socket *so; 1915 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1916 struct rtentry *rt; 1917 #endif 1918 u_long bufsize; 1919 int mss; 1920 1921 #ifdef DIAGNOSTIC 1922 if (tp->t_inpcb && tp->t_in6pcb) 1923 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1924 #endif 1925 so = NULL; 1926 rt = NULL; 1927 #ifdef INET 1928 if (tp->t_inpcb) { 1929 so = tp->t_inpcb->inp_socket; 1930 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1931 rt = in_pcbrtentry(tp->t_inpcb); 1932 #endif 1933 } 1934 #endif 1935 #ifdef INET6 1936 if (tp->t_in6pcb) { 1937 so = tp->t_in6pcb->in6p_socket; 1938 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1939 rt = in6_pcbrtentry(tp->t_in6pcb); 1940 #endif 1941 } 1942 #endif 1943 1944 /* 1945 * As per RFC1122, use the default MSS value, unless they 1946 * sent us an offer. Do not accept offers less than 256 bytes. 1947 */ 1948 mss = tcp_mssdflt; 1949 if (offer) 1950 mss = offer; 1951 mss = max(mss, 256); /* sanity */ 1952 tp->t_peermss = mss; 1953 mss -= tcp_optlen(tp); 1954 #ifdef INET 1955 if (tp->t_inpcb) 1956 mss -= ip_optlen(tp->t_inpcb); 1957 #endif 1958 #ifdef INET6 1959 if (tp->t_in6pcb) 1960 mss -= ip6_optlen(tp->t_in6pcb); 1961 #endif 1962 1963 /* 1964 * If there's a pipesize, change the socket buffer to that size. 1965 * Make the socket buffer an integral number of MSS units. If 1966 * the MSS is larger than the socket buffer, artificially decrease 1967 * the MSS. 1968 */ 1969 #ifdef RTV_SPIPE 1970 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1971 bufsize = rt->rt_rmx.rmx_sendpipe; 1972 else 1973 #endif 1974 { 1975 KASSERT(so != NULL); 1976 bufsize = so->so_snd.sb_hiwat; 1977 } 1978 if (bufsize < mss) 1979 mss = bufsize; 1980 else { 1981 bufsize = roundup(bufsize, mss); 1982 if (bufsize > sb_max) 1983 bufsize = sb_max; 1984 (void) sbreserve(&so->so_snd, bufsize, so); 1985 } 1986 tp->t_segsz = mss; 1987 1988 #ifdef RTV_SSTHRESH 1989 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1990 /* 1991 * There's some sort of gateway or interface buffer 1992 * limit on the path. Use this to set the slow 1993 * start threshold, but set the threshold to no less 1994 * than 2 * MSS. 1995 */ 1996 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1997 } 1998 #endif 1999 } 2000 2001 /* 2002 * Processing necessary when a TCP connection is established. 2003 */ 2004 void 2005 tcp_established(struct tcpcb *tp) 2006 { 2007 struct socket *so; 2008 #ifdef RTV_RPIPE 2009 struct rtentry *rt; 2010 #endif 2011 u_long bufsize; 2012 2013 #ifdef DIAGNOSTIC 2014 if (tp->t_inpcb && tp->t_in6pcb) 2015 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2016 #endif 2017 so = NULL; 2018 rt = NULL; 2019 #ifdef INET 2020 if (tp->t_inpcb) { 2021 so = tp->t_inpcb->inp_socket; 2022 #if defined(RTV_RPIPE) 2023 rt = in_pcbrtentry(tp->t_inpcb); 2024 #endif 2025 } 2026 #endif 2027 #ifdef INET6 2028 if (tp->t_in6pcb) { 2029 so = tp->t_in6pcb->in6p_socket; 2030 #if defined(RTV_RPIPE) 2031 rt = in6_pcbrtentry(tp->t_in6pcb); 2032 #endif 2033 } 2034 #endif 2035 2036 tp->t_state = TCPS_ESTABLISHED; 2037 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2038 2039 #ifdef RTV_RPIPE 2040 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2041 bufsize = rt->rt_rmx.rmx_recvpipe; 2042 else 2043 #endif 2044 { 2045 KASSERT(so != NULL); 2046 bufsize = so->so_rcv.sb_hiwat; 2047 } 2048 if (bufsize > tp->t_ourmss) { 2049 bufsize = roundup(bufsize, tp->t_ourmss); 2050 if (bufsize > sb_max) 2051 bufsize = sb_max; 2052 (void) sbreserve(&so->so_rcv, bufsize, so); 2053 } 2054 } 2055 2056 /* 2057 * Check if there's an initial rtt or rttvar. Convert from the 2058 * route-table units to scaled multiples of the slow timeout timer. 2059 * Called only during the 3-way handshake. 2060 */ 2061 void 2062 tcp_rmx_rtt(struct tcpcb *tp) 2063 { 2064 #ifdef RTV_RTT 2065 struct rtentry *rt = NULL; 2066 int rtt; 2067 2068 #ifdef DIAGNOSTIC 2069 if (tp->t_inpcb && tp->t_in6pcb) 2070 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2071 #endif 2072 #ifdef INET 2073 if (tp->t_inpcb) 2074 rt = in_pcbrtentry(tp->t_inpcb); 2075 #endif 2076 #ifdef INET6 2077 if (tp->t_in6pcb) 2078 rt = in6_pcbrtentry(tp->t_in6pcb); 2079 #endif 2080 if (rt == NULL) 2081 return; 2082 2083 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2084 /* 2085 * XXX The lock bit for MTU indicates that the value 2086 * is also a minimum value; this is subject to time. 2087 */ 2088 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2089 TCPT_RANGESET(tp->t_rttmin, 2090 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2091 TCPTV_MIN, TCPTV_REXMTMAX); 2092 tp->t_srtt = rtt / 2093 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2094 if (rt->rt_rmx.rmx_rttvar) { 2095 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2096 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2097 (TCP_RTTVAR_SHIFT + 2)); 2098 } else { 2099 /* Default variation is +- 1 rtt */ 2100 tp->t_rttvar = 2101 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2102 } 2103 TCPT_RANGESET(tp->t_rxtcur, 2104 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2105 tp->t_rttmin, TCPTV_REXMTMAX); 2106 } 2107 #endif 2108 } 2109 2110 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2111 #if NRND > 0 2112 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2113 #endif 2114 2115 /* 2116 * Get a new sequence value given a tcp control block 2117 */ 2118 tcp_seq 2119 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2120 { 2121 2122 #ifdef INET 2123 if (tp->t_inpcb != NULL) { 2124 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2125 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2126 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2127 addin)); 2128 } 2129 #endif 2130 #ifdef INET6 2131 if (tp->t_in6pcb != NULL) { 2132 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2133 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2134 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2135 addin)); 2136 } 2137 #endif 2138 /* Not possible. */ 2139 panic("tcp_new_iss"); 2140 } 2141 2142 /* 2143 * This routine actually generates a new TCP initial sequence number. 2144 */ 2145 tcp_seq 2146 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2147 size_t addrsz, tcp_seq addin) 2148 { 2149 tcp_seq tcp_iss; 2150 2151 #if NRND > 0 2152 static int beenhere; 2153 2154 /* 2155 * If we haven't been here before, initialize our cryptographic 2156 * hash secret. 2157 */ 2158 if (beenhere == 0) { 2159 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2160 RND_EXTRACT_ANY); 2161 beenhere = 1; 2162 } 2163 2164 if (tcp_do_rfc1948) { 2165 MD5_CTX ctx; 2166 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2167 2168 /* 2169 * Compute the base value of the ISS. It is a hash 2170 * of (saddr, sport, daddr, dport, secret). 2171 */ 2172 MD5Init(&ctx); 2173 2174 MD5Update(&ctx, (u_char *) laddr, addrsz); 2175 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2176 2177 MD5Update(&ctx, (u_char *) faddr, addrsz); 2178 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2179 2180 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2181 2182 MD5Final(hash, &ctx); 2183 2184 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2185 2186 /* 2187 * Now increment our "timer", and add it in to 2188 * the computed value. 2189 * 2190 * XXX Use `addin'? 2191 * XXX TCP_ISSINCR too large to use? 2192 */ 2193 tcp_iss_seq += TCP_ISSINCR; 2194 #ifdef TCPISS_DEBUG 2195 printf("ISS hash 0x%08x, ", tcp_iss); 2196 #endif 2197 tcp_iss += tcp_iss_seq + addin; 2198 #ifdef TCPISS_DEBUG 2199 printf("new ISS 0x%08x\n", tcp_iss); 2200 #endif 2201 } else 2202 #endif /* NRND > 0 */ 2203 { 2204 /* 2205 * Randomize. 2206 */ 2207 #if NRND > 0 2208 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2209 #else 2210 tcp_iss = arc4random(); 2211 #endif 2212 2213 /* 2214 * If we were asked to add some amount to a known value, 2215 * we will take a random value obtained above, mask off 2216 * the upper bits, and add in the known value. We also 2217 * add in a constant to ensure that we are at least a 2218 * certain distance from the original value. 2219 * 2220 * This is used when an old connection is in timed wait 2221 * and we have a new one coming in, for instance. 2222 */ 2223 if (addin != 0) { 2224 #ifdef TCPISS_DEBUG 2225 printf("Random %08x, ", tcp_iss); 2226 #endif 2227 tcp_iss &= TCP_ISS_RANDOM_MASK; 2228 tcp_iss += addin + TCP_ISSINCR; 2229 #ifdef TCPISS_DEBUG 2230 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2231 #endif 2232 } else { 2233 tcp_iss &= TCP_ISS_RANDOM_MASK; 2234 tcp_iss += tcp_iss_seq; 2235 tcp_iss_seq += TCP_ISSINCR; 2236 #ifdef TCPISS_DEBUG 2237 printf("ISS %08x\n", tcp_iss); 2238 #endif 2239 } 2240 } 2241 2242 if (tcp_compat_42) { 2243 /* 2244 * Limit it to the positive range for really old TCP 2245 * implementations. 2246 * Just AND off the top bit instead of checking if 2247 * is set first - saves a branch 50% of the time. 2248 */ 2249 tcp_iss &= 0x7fffffff; /* XXX */ 2250 } 2251 2252 return (tcp_iss); 2253 } 2254 2255 #if defined(IPSEC) || defined(FAST_IPSEC) 2256 /* compute ESP/AH header size for TCP, including outer IP header. */ 2257 size_t 2258 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2259 { 2260 struct inpcb *inp; 2261 size_t hdrsiz; 2262 2263 /* XXX mapped addr case (tp->t_in6pcb) */ 2264 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2265 return 0; 2266 switch (tp->t_family) { 2267 case AF_INET: 2268 /* XXX: should use currect direction. */ 2269 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2270 break; 2271 default: 2272 hdrsiz = 0; 2273 break; 2274 } 2275 2276 return hdrsiz; 2277 } 2278 2279 #ifdef INET6 2280 size_t 2281 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2282 { 2283 struct in6pcb *in6p; 2284 size_t hdrsiz; 2285 2286 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2287 return 0; 2288 switch (tp->t_family) { 2289 case AF_INET6: 2290 /* XXX: should use currect direction. */ 2291 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2292 break; 2293 case AF_INET: 2294 /* mapped address case - tricky */ 2295 default: 2296 hdrsiz = 0; 2297 break; 2298 } 2299 2300 return hdrsiz; 2301 } 2302 #endif 2303 #endif /*IPSEC*/ 2304 2305 /* 2306 * Determine the length of the TCP options for this connection. 2307 * 2308 * XXX: What do we do for SACK, when we add that? Just reserve 2309 * all of the space? Otherwise we can't exactly be incrementing 2310 * cwnd by an amount that varies depending on the amount we last 2311 * had to SACK! 2312 */ 2313 2314 u_int 2315 tcp_optlen(struct tcpcb *tp) 2316 { 2317 u_int optlen; 2318 2319 optlen = 0; 2320 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2321 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2322 optlen += TCPOLEN_TSTAMP_APPA; 2323 2324 #ifdef TCP_SIGNATURE 2325 if (tp->t_flags & TF_SIGNATURE) 2326 optlen += TCPOLEN_SIGNATURE + 2; 2327 #endif /* TCP_SIGNATURE */ 2328 2329 return optlen; 2330 } 2331 2332 u_int 2333 tcp_hdrsz(struct tcpcb *tp) 2334 { 2335 u_int hlen; 2336 2337 switch (tp->t_family) { 2338 #ifdef INET6 2339 case AF_INET6: 2340 hlen = sizeof(struct ip6_hdr); 2341 break; 2342 #endif 2343 case AF_INET: 2344 hlen = sizeof(struct ip); 2345 break; 2346 default: 2347 hlen = 0; 2348 break; 2349 } 2350 hlen += sizeof(struct tcphdr); 2351 2352 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2353 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2354 hlen += TCPOLEN_TSTAMP_APPA; 2355 #ifdef TCP_SIGNATURE 2356 if (tp->t_flags & TF_SIGNATURE) 2357 hlen += TCPOLEN_SIGLEN; 2358 #endif 2359 return hlen; 2360 } 2361
Indexes created Sun Sep 21 15:09:59 GMT 2025