tcp_subr.c revision 1.194
1 /* $NetBSD: tcp_subr.c,v 1.194 2005/08/10 13:05:16 yamt 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 1.194 2005/08/10 13:05:16 yamt 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/tcpip.h> 155 156 #ifdef IPSEC 157 #include <netinet6/ipsec.h> 158 #include <netkey/key.h> 159 #endif /*IPSEC*/ 160 161 #ifdef FAST_IPSEC 162 #include <netipsec/ipsec.h> 163 #include <netipsec/xform.h> 164 #ifdef INET6 165 #include <netipsec/ipsec6.h> 166 #endif 167 #include <netipsec/key.h> 168 #endif /* FAST_IPSEC*/ 169 170 171 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 172 struct tcpstat tcpstat; /* tcp statistics */ 173 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 174 175 /* patchable/settable parameters for tcp */ 176 int tcp_mssdflt = TCP_MSS; 177 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 178 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 179 #if NRND > 0 180 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 181 #endif 182 int tcp_do_sack = 1; /* selective acknowledgement */ 183 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 184 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 185 int tcp_do_newreno = 1; /* Use the New Reno algorithms */ 186 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 187 #ifndef TCP_INIT_WIN 188 #define TCP_INIT_WIN 0 /* initial slow start window */ 189 #endif 190 #ifndef TCP_INIT_WIN_LOCAL 191 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 192 #endif 193 int tcp_init_win = TCP_INIT_WIN; 194 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 195 int tcp_mss_ifmtu = 0; 196 #ifdef TCP_COMPAT_42 197 int tcp_compat_42 = 1; 198 #else 199 int tcp_compat_42 = 0; 200 #endif 201 int tcp_rst_ppslim = 100; /* 100pps */ 202 int tcp_ackdrop_ppslim = 100; /* 100pps */ 203 int tcp_sack_tp_maxholes = 32; 204 int tcp_sack_globalmaxholes = 1024; 205 int tcp_sack_globalholes = 0; 206 207 208 /* tcb hash */ 209 #ifndef TCBHASHSIZE 210 #define TCBHASHSIZE 128 211 #endif 212 int tcbhashsize = TCBHASHSIZE; 213 214 /* syn hash parameters */ 215 #define TCP_SYN_HASH_SIZE 293 216 #define TCP_SYN_BUCKET_SIZE 35 217 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 218 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 219 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 220 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 221 222 int tcp_freeq(struct tcpcb *); 223 224 #ifdef INET 225 void tcp_mtudisc_callback(struct in_addr); 226 #endif 227 #ifdef INET6 228 void tcp6_mtudisc_callback(struct in6_addr *); 229 #endif 230 231 #ifdef INET6 232 void tcp6_mtudisc(struct in6pcb *, int); 233 #endif 234 235 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL); 236 237 #ifdef TCP_CSUM_COUNTERS 238 #include <sys/device.h> 239 240 #if defined(INET) 241 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "hwcsum bad"); 243 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "hwcsum ok"); 245 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 246 NULL, "tcp", "hwcsum data"); 247 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "swcsum"); 249 250 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 251 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 252 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 253 EVCNT_ATTACH_STATIC(tcp_swcsum); 254 #endif /* defined(INET) */ 255 256 #if defined(INET6) 257 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 258 NULL, "tcp6", "hwcsum bad"); 259 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp6", "hwcsum ok"); 261 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 NULL, "tcp6", "hwcsum data"); 263 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "swcsum"); 265 266 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 267 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 268 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 269 EVCNT_ATTACH_STATIC(tcp6_swcsum); 270 #endif /* defined(INET6) */ 271 #endif /* TCP_CSUM_COUNTERS */ 272 273 274 #ifdef TCP_OUTPUT_COUNTERS 275 #include <sys/device.h> 276 277 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 278 NULL, "tcp", "output big header"); 279 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 NULL, "tcp", "output predict hit"); 281 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 NULL, "tcp", "output predict miss"); 283 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output copy small"); 285 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output copy big"); 287 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output reference big"); 289 290 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 291 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 292 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 293 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 294 EVCNT_ATTACH_STATIC(tcp_output_copybig); 295 EVCNT_ATTACH_STATIC(tcp_output_refbig); 296 297 #endif /* TCP_OUTPUT_COUNTERS */ 298 299 #ifdef TCP_REASS_COUNTERS 300 #include <sys/device.h> 301 302 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 303 NULL, "tcp_reass", "calls"); 304 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 305 &tcp_reass_, "tcp_reass", "insert into empty queue"); 306 struct evcnt tcp_reass_iteration[8] = { 307 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 308 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 315 }; 316 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 317 &tcp_reass_, "tcp_reass", "prepend to first"); 318 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 319 &tcp_reass_, "tcp_reass", "prepend"); 320 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 321 &tcp_reass_, "tcp_reass", "insert"); 322 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "insert at tail"); 324 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "append"); 326 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "append to tail fragment"); 328 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "overlap at end"); 330 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "overlap at start"); 332 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "duplicate segment"); 334 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "duplicate fragment"); 336 337 EVCNT_ATTACH_STATIC(tcp_reass_); 338 EVCNT_ATTACH_STATIC(tcp_reass_empty); 339 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 340 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 341 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 342 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 347 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 348 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 349 EVCNT_ATTACH_STATIC(tcp_reass_insert); 350 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 351 EVCNT_ATTACH_STATIC(tcp_reass_append); 352 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 353 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 354 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 355 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 356 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 357 358 #endif /* TCP_REASS_COUNTERS */ 359 360 #ifdef MBUFTRACE 361 struct mowner tcp_mowner = { "tcp" }; 362 struct mowner tcp_rx_mowner = { "tcp", "rx" }; 363 struct mowner tcp_tx_mowner = { "tcp", "tx" }; 364 #endif 365 366 /* 367 * Tcp initialization 368 */ 369 void 370 tcp_init(void) 371 { 372 int hlen; 373 374 /* Initialize the TCPCB template. */ 375 tcp_tcpcb_template(); 376 377 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 378 379 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 380 #ifdef INET6 381 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 382 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 383 #endif 384 if (max_protohdr < hlen) 385 max_protohdr = hlen; 386 if (max_linkhdr + hlen > MHLEN) 387 panic("tcp_init"); 388 389 #ifdef INET 390 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 391 #endif 392 #ifdef INET6 393 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 394 #endif 395 396 /* Initialize timer state. */ 397 tcp_timer_init(); 398 399 /* Initialize the compressed state engine. */ 400 syn_cache_init(); 401 402 MOWNER_ATTACH(&tcp_tx_mowner); 403 MOWNER_ATTACH(&tcp_rx_mowner); 404 MOWNER_ATTACH(&tcp_mowner); 405 } 406 407 /* 408 * Create template to be used to send tcp packets on a connection. 409 * Call after host entry created, allocates an mbuf and fills 410 * in a skeletal tcp/ip header, minimizing the amount of work 411 * necessary when the connection is used. 412 */ 413 struct mbuf * 414 tcp_template(struct tcpcb *tp) 415 { 416 struct inpcb *inp = tp->t_inpcb; 417 #ifdef INET6 418 struct in6pcb *in6p = tp->t_in6pcb; 419 #endif 420 struct tcphdr *n; 421 struct mbuf *m; 422 int hlen; 423 424 switch (tp->t_family) { 425 case AF_INET: 426 hlen = sizeof(struct ip); 427 if (inp) 428 break; 429 #ifdef INET6 430 if (in6p) { 431 /* mapped addr case */ 432 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 433 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 434 break; 435 } 436 #endif 437 return NULL; /*EINVAL*/ 438 #ifdef INET6 439 case AF_INET6: 440 hlen = sizeof(struct ip6_hdr); 441 if (in6p) { 442 /* more sainty check? */ 443 break; 444 } 445 return NULL; /*EINVAL*/ 446 #endif 447 default: 448 hlen = 0; /*pacify gcc*/ 449 return NULL; /*EAFNOSUPPORT*/ 450 } 451 #ifdef DIAGNOSTIC 452 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 453 panic("mclbytes too small for t_template"); 454 #endif 455 m = tp->t_template; 456 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 457 ; 458 else { 459 if (m) 460 m_freem(m); 461 m = tp->t_template = NULL; 462 MGETHDR(m, M_DONTWAIT, MT_HEADER); 463 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 464 MCLGET(m, M_DONTWAIT); 465 if ((m->m_flags & M_EXT) == 0) { 466 m_free(m); 467 m = NULL; 468 } 469 } 470 if (m == NULL) 471 return NULL; 472 MCLAIM(m, &tcp_mowner); 473 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 474 } 475 476 bzero(mtod(m, caddr_t), m->m_len); 477 478 n = (struct tcphdr *)(mtod(m, caddr_t) + hlen); 479 480 switch (tp->t_family) { 481 case AF_INET: 482 { 483 struct ipovly *ipov; 484 mtod(m, struct ip *)->ip_v = 4; 485 mtod(m, struct ip *)->ip_hl = hlen >> 2; 486 ipov = mtod(m, struct ipovly *); 487 ipov->ih_pr = IPPROTO_TCP; 488 ipov->ih_len = htons(sizeof(struct tcphdr)); 489 if (inp) { 490 ipov->ih_src = inp->inp_laddr; 491 ipov->ih_dst = inp->inp_faddr; 492 } 493 #ifdef INET6 494 else if (in6p) { 495 /* mapped addr case */ 496 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 497 sizeof(ipov->ih_src)); 498 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 499 sizeof(ipov->ih_dst)); 500 } 501 #endif 502 /* 503 * Compute the pseudo-header portion of the checksum 504 * now. We incrementally add in the TCP option and 505 * payload lengths later, and then compute the TCP 506 * checksum right before the packet is sent off onto 507 * the wire. 508 */ 509 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 510 ipov->ih_dst.s_addr, 511 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 512 break; 513 } 514 #ifdef INET6 515 case AF_INET6: 516 { 517 struct ip6_hdr *ip6; 518 mtod(m, struct ip *)->ip_v = 6; 519 ip6 = mtod(m, struct ip6_hdr *); 520 ip6->ip6_nxt = IPPROTO_TCP; 521 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 522 ip6->ip6_src = in6p->in6p_laddr; 523 ip6->ip6_dst = in6p->in6p_faddr; 524 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 525 if (ip6_auto_flowlabel) { 526 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 527 ip6->ip6_flow |= 528 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 529 } 530 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 531 ip6->ip6_vfc |= IPV6_VERSION; 532 533 /* 534 * Compute the pseudo-header portion of the checksum 535 * now. We incrementally add in the TCP option and 536 * payload lengths later, and then compute the TCP 537 * checksum right before the packet is sent off onto 538 * the wire. 539 */ 540 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 541 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 542 htonl(IPPROTO_TCP)); 543 break; 544 } 545 #endif 546 } 547 if (inp) { 548 n->th_sport = inp->inp_lport; 549 n->th_dport = inp->inp_fport; 550 } 551 #ifdef INET6 552 else if (in6p) { 553 n->th_sport = in6p->in6p_lport; 554 n->th_dport = in6p->in6p_fport; 555 } 556 #endif 557 n->th_seq = 0; 558 n->th_ack = 0; 559 n->th_x2 = 0; 560 n->th_off = 5; 561 n->th_flags = 0; 562 n->th_win = 0; 563 n->th_urp = 0; 564 return (m); 565 } 566 567 /* 568 * Send a single message to the TCP at address specified by 569 * the given TCP/IP header. If m == 0, then we make a copy 570 * of the tcpiphdr at ti and send directly to the addressed host. 571 * This is used to force keep alive messages out using the TCP 572 * template for a connection tp->t_template. If flags are given 573 * then we send a message back to the TCP which originated the 574 * segment ti, and discard the mbuf containing it and any other 575 * attached mbufs. 576 * 577 * In any case the ack and sequence number of the transmitted 578 * segment are as specified by the parameters. 579 */ 580 int 581 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 582 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 583 { 584 struct route *ro; 585 int error, tlen, win = 0; 586 int hlen; 587 struct ip *ip; 588 #ifdef INET6 589 struct ip6_hdr *ip6; 590 #endif 591 int family; /* family on packet, not inpcb/in6pcb! */ 592 struct tcphdr *th; 593 struct socket *so; 594 595 if (tp != NULL && (flags & TH_RST) == 0) { 596 #ifdef DIAGNOSTIC 597 if (tp->t_inpcb && tp->t_in6pcb) 598 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 599 #endif 600 #ifdef INET 601 if (tp->t_inpcb) 602 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 603 #endif 604 #ifdef INET6 605 if (tp->t_in6pcb) 606 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 607 #endif 608 } 609 610 th = NULL; /* Quell uninitialized warning */ 611 ip = NULL; 612 #ifdef INET6 613 ip6 = NULL; 614 #endif 615 if (m == 0) { 616 if (!template) 617 return EINVAL; 618 619 /* get family information from template */ 620 switch (mtod(template, struct ip *)->ip_v) { 621 case 4: 622 family = AF_INET; 623 hlen = sizeof(struct ip); 624 break; 625 #ifdef INET6 626 case 6: 627 family = AF_INET6; 628 hlen = sizeof(struct ip6_hdr); 629 break; 630 #endif 631 default: 632 return EAFNOSUPPORT; 633 } 634 635 MGETHDR(m, M_DONTWAIT, MT_HEADER); 636 if (m) { 637 MCLAIM(m, &tcp_tx_mowner); 638 MCLGET(m, M_DONTWAIT); 639 if ((m->m_flags & M_EXT) == 0) { 640 m_free(m); 641 m = NULL; 642 } 643 } 644 if (m == NULL) 645 return (ENOBUFS); 646 647 if (tcp_compat_42) 648 tlen = 1; 649 else 650 tlen = 0; 651 652 m->m_data += max_linkhdr; 653 bcopy(mtod(template, caddr_t), mtod(m, caddr_t), 654 template->m_len); 655 switch (family) { 656 case AF_INET: 657 ip = mtod(m, struct ip *); 658 th = (struct tcphdr *)(ip + 1); 659 break; 660 #ifdef INET6 661 case AF_INET6: 662 ip6 = mtod(m, struct ip6_hdr *); 663 th = (struct tcphdr *)(ip6 + 1); 664 break; 665 #endif 666 #if 0 667 default: 668 /* noone will visit here */ 669 m_freem(m); 670 return EAFNOSUPPORT; 671 #endif 672 } 673 flags = TH_ACK; 674 } else { 675 676 if ((m->m_flags & M_PKTHDR) == 0) { 677 #if 0 678 printf("non PKTHDR to tcp_respond\n"); 679 #endif 680 m_freem(m); 681 return EINVAL; 682 } 683 #ifdef DIAGNOSTIC 684 if (!th0) 685 panic("th0 == NULL in tcp_respond"); 686 #endif 687 688 /* get family information from m */ 689 switch (mtod(m, struct ip *)->ip_v) { 690 case 4: 691 family = AF_INET; 692 hlen = sizeof(struct ip); 693 ip = mtod(m, struct ip *); 694 break; 695 #ifdef INET6 696 case 6: 697 family = AF_INET6; 698 hlen = sizeof(struct ip6_hdr); 699 ip6 = mtod(m, struct ip6_hdr *); 700 break; 701 #endif 702 default: 703 m_freem(m); 704 return EAFNOSUPPORT; 705 } 706 /* clear h/w csum flags inherited from rx packet */ 707 m->m_pkthdr.csum_flags = 0; 708 709 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 710 tlen = sizeof(*th0); 711 else 712 tlen = th0->th_off << 2; 713 714 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 715 mtod(m, caddr_t) + hlen == (caddr_t)th0) { 716 m->m_len = hlen + tlen; 717 m_freem(m->m_next); 718 m->m_next = NULL; 719 } else { 720 struct mbuf *n; 721 722 #ifdef DIAGNOSTIC 723 if (max_linkhdr + hlen + tlen > MCLBYTES) { 724 m_freem(m); 725 return EMSGSIZE; 726 } 727 #endif 728 MGETHDR(n, M_DONTWAIT, MT_HEADER); 729 if (n && max_linkhdr + hlen + tlen > MHLEN) { 730 MCLGET(n, M_DONTWAIT); 731 if ((n->m_flags & M_EXT) == 0) { 732 m_freem(n); 733 n = NULL; 734 } 735 } 736 if (!n) { 737 m_freem(m); 738 return ENOBUFS; 739 } 740 741 MCLAIM(n, &tcp_tx_mowner); 742 n->m_data += max_linkhdr; 743 n->m_len = hlen + tlen; 744 m_copyback(n, 0, hlen, mtod(m, caddr_t)); 745 m_copyback(n, hlen, tlen, (caddr_t)th0); 746 747 m_freem(m); 748 m = n; 749 n = NULL; 750 } 751 752 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 753 switch (family) { 754 case AF_INET: 755 ip = mtod(m, struct ip *); 756 th = (struct tcphdr *)(ip + 1); 757 ip->ip_p = IPPROTO_TCP; 758 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 759 ip->ip_p = IPPROTO_TCP; 760 break; 761 #ifdef INET6 762 case AF_INET6: 763 ip6 = mtod(m, struct ip6_hdr *); 764 th = (struct tcphdr *)(ip6 + 1); 765 ip6->ip6_nxt = IPPROTO_TCP; 766 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 767 ip6->ip6_nxt = IPPROTO_TCP; 768 break; 769 #endif 770 #if 0 771 default: 772 /* noone will visit here */ 773 m_freem(m); 774 return EAFNOSUPPORT; 775 #endif 776 } 777 xchg(th->th_dport, th->th_sport, u_int16_t); 778 #undef xchg 779 tlen = 0; /*be friendly with the following code*/ 780 } 781 th->th_seq = htonl(seq); 782 th->th_ack = htonl(ack); 783 th->th_x2 = 0; 784 if ((flags & TH_SYN) == 0) { 785 if (tp) 786 win >>= tp->rcv_scale; 787 if (win > TCP_MAXWIN) 788 win = TCP_MAXWIN; 789 th->th_win = htons((u_int16_t)win); 790 th->th_off = sizeof (struct tcphdr) >> 2; 791 tlen += sizeof(*th); 792 } else 793 tlen += th->th_off << 2; 794 m->m_len = hlen + tlen; 795 m->m_pkthdr.len = hlen + tlen; 796 m->m_pkthdr.rcvif = (struct ifnet *) 0; 797 th->th_flags = flags; 798 th->th_urp = 0; 799 800 switch (family) { 801 #ifdef INET 802 case AF_INET: 803 { 804 struct ipovly *ipov = (struct ipovly *)ip; 805 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 806 ipov->ih_len = htons((u_int16_t)tlen); 807 808 th->th_sum = 0; 809 th->th_sum = in_cksum(m, hlen + tlen); 810 ip->ip_len = htons(hlen + tlen); 811 ip->ip_ttl = ip_defttl; 812 break; 813 } 814 #endif 815 #ifdef INET6 816 case AF_INET6: 817 { 818 th->th_sum = 0; 819 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 820 tlen); 821 ip6->ip6_plen = htons(tlen); 822 if (tp && tp->t_in6pcb) { 823 struct ifnet *oifp; 824 ro = (struct route *)&tp->t_in6pcb->in6p_route; 825 oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL; 826 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 827 } else 828 ip6->ip6_hlim = ip6_defhlim; 829 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 830 if (ip6_auto_flowlabel) { 831 ip6->ip6_flow |= 832 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 833 } 834 break; 835 } 836 #endif 837 } 838 839 if (tp && tp->t_inpcb) 840 so = tp->t_inpcb->inp_socket; 841 #ifdef INET6 842 else if (tp && tp->t_in6pcb) 843 so = tp->t_in6pcb->in6p_socket; 844 #endif 845 else 846 so = NULL; 847 848 if (tp != NULL && tp->t_inpcb != NULL) { 849 ro = &tp->t_inpcb->inp_route; 850 #ifdef DIAGNOSTIC 851 if (family != AF_INET) 852 panic("tcp_respond: address family mismatch"); 853 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 854 panic("tcp_respond: ip_dst %x != inp_faddr %x", 855 ntohl(ip->ip_dst.s_addr), 856 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 857 } 858 #endif 859 } 860 #ifdef INET6 861 else if (tp != NULL && tp->t_in6pcb != NULL) { 862 ro = (struct route *)&tp->t_in6pcb->in6p_route; 863 #ifdef DIAGNOSTIC 864 if (family == AF_INET) { 865 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 866 panic("tcp_respond: not mapped addr"); 867 if (bcmp(&ip->ip_dst, 868 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 869 sizeof(ip->ip_dst)) != 0) { 870 panic("tcp_respond: ip_dst != in6p_faddr"); 871 } 872 } else if (family == AF_INET6) { 873 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 874 &tp->t_in6pcb->in6p_faddr)) 875 panic("tcp_respond: ip6_dst != in6p_faddr"); 876 } else 877 panic("tcp_respond: address family mismatch"); 878 #endif 879 } 880 #endif 881 else 882 ro = NULL; 883 884 switch (family) { 885 #ifdef INET 886 case AF_INET: 887 error = ip_output(m, NULL, ro, 888 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 889 (struct ip_moptions *)0, so); 890 break; 891 #endif 892 #ifdef INET6 893 case AF_INET6: 894 error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, 895 (struct ip6_moptions *)0, so, NULL); 896 break; 897 #endif 898 default: 899 error = EAFNOSUPPORT; 900 break; 901 } 902 903 return (error); 904 } 905 906 /* 907 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 908 * a bunch of members individually, we maintain this template for the 909 * static and mostly-static components of the TCPCB, and copy it into 910 * the new TCPCB instead. 911 */ 912 static struct tcpcb tcpcb_template = { 913 /* 914 * If TCP_NTIMERS ever changes, we'll need to update this 915 * initializer. 916 */ 917 .t_timer = { 918 CALLOUT_INITIALIZER, 919 CALLOUT_INITIALIZER, 920 CALLOUT_INITIALIZER, 921 CALLOUT_INITIALIZER, 922 }, 923 .t_delack_ch = CALLOUT_INITIALIZER, 924 925 .t_srtt = TCPTV_SRTTBASE, 926 .t_rttmin = TCPTV_MIN, 927 928 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 929 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 930 .snd_numholes = 0, 931 932 .t_partialacks = -1, 933 }; 934 935 /* 936 * Updates the TCPCB template whenever a parameter that would affect 937 * the template is changed. 938 */ 939 void 940 tcp_tcpcb_template(void) 941 { 942 struct tcpcb *tp = &tcpcb_template; 943 int flags; 944 945 tp->t_peermss = tcp_mssdflt; 946 tp->t_ourmss = tcp_mssdflt; 947 tp->t_segsz = tcp_mssdflt; 948 949 flags = 0; 950 if (tcp_do_rfc1323 && tcp_do_win_scale) 951 flags |= TF_REQ_SCALE; 952 if (tcp_do_rfc1323 && tcp_do_timestamps) 953 flags |= TF_REQ_TSTMP; 954 tp->t_flags = flags; 955 956 /* 957 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 958 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 959 * reasonable initial retransmit time. 960 */ 961 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 962 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 963 TCPTV_MIN, TCPTV_REXMTMAX); 964 } 965 966 /* 967 * Create a new TCP control block, making an 968 * empty reassembly queue and hooking it to the argument 969 * protocol control block. 970 */ 971 /* family selects inpcb, or in6pcb */ 972 struct tcpcb * 973 tcp_newtcpcb(int family, void *aux) 974 { 975 struct tcpcb *tp; 976 int i; 977 978 /* XXX Consider using a pool_cache for speed. */ 979 tp = pool_get(&tcpcb_pool, PR_NOWAIT); 980 if (tp == NULL) 981 return (NULL); 982 memcpy(tp, &tcpcb_template, sizeof(*tp)); 983 TAILQ_INIT(&tp->segq); 984 TAILQ_INIT(&tp->timeq); 985 tp->t_family = family; /* may be overridden later on */ 986 TAILQ_INIT(&tp->snd_holes); 987 LIST_INIT(&tp->t_sc); /* XXX can template this */ 988 989 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 990 for (i = 0; i < TCPT_NTIMERS; i++) 991 TCP_TIMER_INIT(tp, i); 992 993 switch (family) { 994 case AF_INET: 995 { 996 struct inpcb *inp = (struct inpcb *)aux; 997 998 inp->inp_ip.ip_ttl = ip_defttl; 999 inp->inp_ppcb = (caddr_t)tp; 1000 1001 tp->t_inpcb = inp; 1002 tp->t_mtudisc = ip_mtudisc; 1003 break; 1004 } 1005 #ifdef INET6 1006 case AF_INET6: 1007 { 1008 struct in6pcb *in6p = (struct in6pcb *)aux; 1009 1010 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1011 in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp 1012 : NULL); 1013 in6p->in6p_ppcb = (caddr_t)tp; 1014 1015 tp->t_in6pcb = in6p; 1016 /* for IPv6, always try to run path MTU discovery */ 1017 tp->t_mtudisc = 1; 1018 break; 1019 } 1020 #endif /* INET6 */ 1021 default: 1022 pool_put(&tcpcb_pool, tp); 1023 return (NULL); 1024 } 1025 1026 /* 1027 * Initialize our timebase. When we send timestamps, we take 1028 * the delta from tcp_now -- this means each connection always 1029 * gets a timebase of 0, which makes it, among other things, 1030 * more difficult to determine how long a system has been up, 1031 * and thus how many TCP sequence increments have occurred. 1032 */ 1033 tp->ts_timebase = tcp_now; 1034 1035 return (tp); 1036 } 1037 1038 /* 1039 * Drop a TCP connection, reporting 1040 * the specified error. If connection is synchronized, 1041 * then send a RST to peer. 1042 */ 1043 struct tcpcb * 1044 tcp_drop(struct tcpcb *tp, int errno) 1045 { 1046 struct socket *so = NULL; 1047 1048 #ifdef DIAGNOSTIC 1049 if (tp->t_inpcb && tp->t_in6pcb) 1050 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1051 #endif 1052 #ifdef INET 1053 if (tp->t_inpcb) 1054 so = tp->t_inpcb->inp_socket; 1055 #endif 1056 #ifdef INET6 1057 if (tp->t_in6pcb) 1058 so = tp->t_in6pcb->in6p_socket; 1059 #endif 1060 if (!so) 1061 return NULL; 1062 1063 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1064 tp->t_state = TCPS_CLOSED; 1065 (void) tcp_output(tp); 1066 tcpstat.tcps_drops++; 1067 } else 1068 tcpstat.tcps_conndrops++; 1069 if (errno == ETIMEDOUT && tp->t_softerror) 1070 errno = tp->t_softerror; 1071 so->so_error = errno; 1072 return (tcp_close(tp)); 1073 } 1074 1075 /* 1076 * Return whether this tcpcb is marked as dead, indicating 1077 * to the calling timer function that no further action should 1078 * be taken, as we are about to release this tcpcb. The release 1079 * of the storage will be done if this is the last timer running. 1080 * 1081 * This should be called from the callout handler function after 1082 * callout_ack() is done, so that the number of invoking timer 1083 * functions is 0. 1084 */ 1085 int 1086 tcp_isdead(struct tcpcb *tp) 1087 { 1088 int dead = (tp->t_flags & TF_DEAD); 1089 1090 if (__predict_false(dead)) { 1091 if (tcp_timers_invoking(tp) > 0) 1092 /* not quite there yet -- count separately? */ 1093 return dead; 1094 tcpstat.tcps_delayed_free++; 1095 pool_put(&tcpcb_pool, tp); 1096 } 1097 return dead; 1098 } 1099 1100 /* 1101 * Close a TCP control block: 1102 * discard all space held by the tcp 1103 * discard internet protocol block 1104 * wake up any sleepers 1105 */ 1106 struct tcpcb * 1107 tcp_close(struct tcpcb *tp) 1108 { 1109 struct inpcb *inp; 1110 #ifdef INET6 1111 struct in6pcb *in6p; 1112 #endif 1113 struct socket *so; 1114 #ifdef RTV_RTT 1115 struct rtentry *rt; 1116 #endif 1117 struct route *ro; 1118 1119 inp = tp->t_inpcb; 1120 #ifdef INET6 1121 in6p = tp->t_in6pcb; 1122 #endif 1123 so = NULL; 1124 ro = NULL; 1125 if (inp) { 1126 so = inp->inp_socket; 1127 ro = &inp->inp_route; 1128 } 1129 #ifdef INET6 1130 else if (in6p) { 1131 so = in6p->in6p_socket; 1132 ro = (struct route *)&in6p->in6p_route; 1133 } 1134 #endif 1135 1136 #ifdef RTV_RTT 1137 /* 1138 * If we sent enough data to get some meaningful characteristics, 1139 * save them in the routing entry. 'Enough' is arbitrarily 1140 * defined as the sendpipesize (default 4K) * 16. This would 1141 * give us 16 rtt samples assuming we only get one sample per 1142 * window (the usual case on a long haul net). 16 samples is 1143 * enough for the srtt filter to converge to within 5% of the correct 1144 * value; fewer samples and we could save a very bogus rtt. 1145 * 1146 * Don't update the default route's characteristics and don't 1147 * update anything that the user "locked". 1148 */ 1149 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1150 ro && (rt = ro->ro_rt) && 1151 !in_nullhost(satosin(rt_key(rt))->sin_addr)) { 1152 u_long i = 0; 1153 1154 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1155 i = tp->t_srtt * 1156 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1157 if (rt->rt_rmx.rmx_rtt && i) 1158 /* 1159 * filter this update to half the old & half 1160 * the new values, converting scale. 1161 * See route.h and tcp_var.h for a 1162 * description of the scaling constants. 1163 */ 1164 rt->rt_rmx.rmx_rtt = 1165 (rt->rt_rmx.rmx_rtt + i) / 2; 1166 else 1167 rt->rt_rmx.rmx_rtt = i; 1168 } 1169 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1170 i = tp->t_rttvar * 1171 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1172 if (rt->rt_rmx.rmx_rttvar && i) 1173 rt->rt_rmx.rmx_rttvar = 1174 (rt->rt_rmx.rmx_rttvar + i) / 2; 1175 else 1176 rt->rt_rmx.rmx_rttvar = i; 1177 } 1178 /* 1179 * update the pipelimit (ssthresh) if it has been updated 1180 * already or if a pipesize was specified & the threshhold 1181 * got below half the pipesize. I.e., wait for bad news 1182 * before we start updating, then update on both good 1183 * and bad news. 1184 */ 1185 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1186 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1187 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1188 /* 1189 * convert the limit from user data bytes to 1190 * packets then to packet data bytes. 1191 */ 1192 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1193 if (i < 2) 1194 i = 2; 1195 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1196 if (rt->rt_rmx.rmx_ssthresh) 1197 rt->rt_rmx.rmx_ssthresh = 1198 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1199 else 1200 rt->rt_rmx.rmx_ssthresh = i; 1201 } 1202 } 1203 #endif /* RTV_RTT */ 1204 /* free the reassembly queue, if any */ 1205 TCP_REASS_LOCK(tp); 1206 (void) tcp_freeq(tp); 1207 TCP_REASS_UNLOCK(tp); 1208 1209 /* free the SACK holes list. */ 1210 tcp_free_sackholes(tp); 1211 1212 tcp_canceltimers(tp); 1213 TCP_CLEAR_DELACK(tp); 1214 syn_cache_cleanup(tp); 1215 1216 if (tp->t_template) { 1217 m_free(tp->t_template); 1218 tp->t_template = NULL; 1219 } 1220 if (tcp_timers_invoking(tp)) 1221 tp->t_flags |= TF_DEAD; 1222 else 1223 pool_put(&tcpcb_pool, tp); 1224 1225 if (inp) { 1226 inp->inp_ppcb = 0; 1227 soisdisconnected(so); 1228 in_pcbdetach(inp); 1229 } 1230 #ifdef INET6 1231 else if (in6p) { 1232 in6p->in6p_ppcb = 0; 1233 soisdisconnected(so); 1234 in6_pcbdetach(in6p); 1235 } 1236 #endif 1237 tcpstat.tcps_closed++; 1238 return ((struct tcpcb *)0); 1239 } 1240 1241 int 1242 tcp_freeq(tp) 1243 struct tcpcb *tp; 1244 { 1245 struct ipqent *qe; 1246 int rv = 0; 1247 #ifdef TCPREASS_DEBUG 1248 int i = 0; 1249 #endif 1250 1251 TCP_REASS_LOCK_CHECK(tp); 1252 1253 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1254 #ifdef TCPREASS_DEBUG 1255 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1256 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1257 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1258 #endif 1259 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1260 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1261 m_freem(qe->ipqe_m); 1262 tcpipqent_free(qe); 1263 rv = 1; 1264 } 1265 tp->t_segqlen = 0; 1266 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1267 return (rv); 1268 } 1269 1270 /* 1271 * Protocol drain routine. Called when memory is in short supply. 1272 */ 1273 void 1274 tcp_drain(void) 1275 { 1276 struct inpcb_hdr *inph; 1277 struct tcpcb *tp; 1278 1279 /* 1280 * Free the sequence queue of all TCP connections. 1281 */ 1282 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1283 switch (inph->inph_af) { 1284 case AF_INET: 1285 tp = intotcpcb((struct inpcb *)inph); 1286 break; 1287 #ifdef INET6 1288 case AF_INET6: 1289 tp = in6totcpcb((struct in6pcb *)inph); 1290 break; 1291 #endif 1292 default: 1293 tp = NULL; 1294 break; 1295 } 1296 if (tp != NULL) { 1297 /* 1298 * We may be called from a device's interrupt 1299 * context. If the tcpcb is already busy, 1300 * just bail out now. 1301 */ 1302 if (tcp_reass_lock_try(tp) == 0) 1303 continue; 1304 if (tcp_freeq(tp)) 1305 tcpstat.tcps_connsdrained++; 1306 TCP_REASS_UNLOCK(tp); 1307 } 1308 } 1309 } 1310 1311 /* 1312 * Notify a tcp user of an asynchronous error; 1313 * store error as soft error, but wake up user 1314 * (for now, won't do anything until can select for soft error). 1315 */ 1316 void 1317 tcp_notify(struct inpcb *inp, int error) 1318 { 1319 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1320 struct socket *so = inp->inp_socket; 1321 1322 /* 1323 * Ignore some errors if we are hooked up. 1324 * If connection hasn't completed, has retransmitted several times, 1325 * and receives a second error, give up now. This is better 1326 * than waiting a long time to establish a connection that 1327 * can never complete. 1328 */ 1329 if (tp->t_state == TCPS_ESTABLISHED && 1330 (error == EHOSTUNREACH || error == ENETUNREACH || 1331 error == EHOSTDOWN)) { 1332 return; 1333 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1334 tp->t_rxtshift > 3 && tp->t_softerror) 1335 so->so_error = error; 1336 else 1337 tp->t_softerror = error; 1338 wakeup((caddr_t) &so->so_timeo); 1339 sorwakeup(so); 1340 sowwakeup(so); 1341 } 1342 1343 #ifdef INET6 1344 void 1345 tcp6_notify(struct in6pcb *in6p, int error) 1346 { 1347 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1348 struct socket *so = in6p->in6p_socket; 1349 1350 /* 1351 * Ignore some errors if we are hooked up. 1352 * If connection hasn't completed, has retransmitted several times, 1353 * and receives a second error, give up now. This is better 1354 * than waiting a long time to establish a connection that 1355 * can never complete. 1356 */ 1357 if (tp->t_state == TCPS_ESTABLISHED && 1358 (error == EHOSTUNREACH || error == ENETUNREACH || 1359 error == EHOSTDOWN)) { 1360 return; 1361 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1362 tp->t_rxtshift > 3 && tp->t_softerror) 1363 so->so_error = error; 1364 else 1365 tp->t_softerror = error; 1366 wakeup((caddr_t) &so->so_timeo); 1367 sorwakeup(so); 1368 sowwakeup(so); 1369 } 1370 #endif 1371 1372 #ifdef INET6 1373 void 1374 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1375 { 1376 struct tcphdr th; 1377 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1378 int nmatch; 1379 struct ip6_hdr *ip6; 1380 const struct sockaddr_in6 *sa6_src = NULL; 1381 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; 1382 struct mbuf *m; 1383 int off; 1384 1385 if (sa->sa_family != AF_INET6 || 1386 sa->sa_len != sizeof(struct sockaddr_in6)) 1387 return; 1388 if ((unsigned)cmd >= PRC_NCMDS) 1389 return; 1390 else if (cmd == PRC_QUENCH) { 1391 /* 1392 * Don't honor ICMP Source Quench messages meant for 1393 * TCP connections. 1394 */ 1395 return; 1396 } else if (PRC_IS_REDIRECT(cmd)) 1397 notify = in6_rtchange, d = NULL; 1398 else if (cmd == PRC_MSGSIZE) 1399 ; /* special code is present, see below */ 1400 else if (cmd == PRC_HOSTDEAD) 1401 d = NULL; 1402 else if (inet6ctlerrmap[cmd] == 0) 1403 return; 1404 1405 /* if the parameter is from icmp6, decode it. */ 1406 if (d != NULL) { 1407 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1408 m = ip6cp->ip6c_m; 1409 ip6 = ip6cp->ip6c_ip6; 1410 off = ip6cp->ip6c_off; 1411 sa6_src = ip6cp->ip6c_src; 1412 } else { 1413 m = NULL; 1414 ip6 = NULL; 1415 sa6_src = &sa6_any; 1416 off = 0; 1417 } 1418 1419 if (ip6) { 1420 /* 1421 * XXX: We assume that when ip6 is non NULL, 1422 * M and OFF are valid. 1423 */ 1424 1425 /* check if we can safely examine src and dst ports */ 1426 if (m->m_pkthdr.len < off + sizeof(th)) { 1427 if (cmd == PRC_MSGSIZE) 1428 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1429 return; 1430 } 1431 1432 bzero(&th, sizeof(th)); 1433 m_copydata(m, off, sizeof(th), (caddr_t)&th); 1434 1435 if (cmd == PRC_MSGSIZE) { 1436 int valid = 0; 1437 1438 /* 1439 * Check to see if we have a valid TCP connection 1440 * corresponding to the address in the ICMPv6 message 1441 * payload. 1442 */ 1443 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1444 th.th_dport, (const struct in6_addr *)&sa6_src->sin6_addr, 1445 th.th_sport, 0)) 1446 valid++; 1447 1448 /* 1449 * Depending on the value of "valid" and routing table 1450 * size (mtudisc_{hi,lo}wat), we will: 1451 * - recalcurate the new MTU and create the 1452 * corresponding routing entry, or 1453 * - ignore the MTU change notification. 1454 */ 1455 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1456 1457 /* 1458 * no need to call in6_pcbnotify, it should have been 1459 * called via callback if necessary 1460 */ 1461 return; 1462 } 1463 1464 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1465 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1466 if (nmatch == 0 && syn_cache_count && 1467 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1468 inet6ctlerrmap[cmd] == ENETUNREACH || 1469 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1470 syn_cache_unreach((const struct sockaddr *)sa6_src, 1471 sa, &th); 1472 } else { 1473 (void) in6_pcbnotify(&tcbtable, sa, 0, 1474 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1475 } 1476 } 1477 #endif 1478 1479 #ifdef INET 1480 /* assumes that ip header and tcp header are contiguous on mbuf */ 1481 void * 1482 tcp_ctlinput(int cmd, struct sockaddr *sa, void *v) 1483 { 1484 struct ip *ip = v; 1485 struct tcphdr *th; 1486 struct icmp *icp; 1487 extern const int inetctlerrmap[]; 1488 void (*notify)(struct inpcb *, int) = tcp_notify; 1489 int errno; 1490 int nmatch; 1491 struct tcpcb *tp; 1492 u_int mtu; 1493 tcp_seq seq; 1494 struct inpcb *inp; 1495 #ifdef INET6 1496 struct in6pcb *in6p; 1497 struct in6_addr src6, dst6; 1498 #endif 1499 1500 if (sa->sa_family != AF_INET || 1501 sa->sa_len != sizeof(struct sockaddr_in)) 1502 return NULL; 1503 if ((unsigned)cmd >= PRC_NCMDS) 1504 return NULL; 1505 errno = inetctlerrmap[cmd]; 1506 if (cmd == PRC_QUENCH) 1507 /* 1508 * Don't honor ICMP Source Quench messages meant for 1509 * TCP connections. 1510 */ 1511 return NULL; 1512 else if (PRC_IS_REDIRECT(cmd)) 1513 notify = in_rtchange, ip = 0; 1514 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1515 /* 1516 * Check to see if we have a valid TCP connection 1517 * corresponding to the address in the ICMP message 1518 * payload. 1519 * 1520 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1521 */ 1522 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1523 #ifdef INET6 1524 memset(&src6, 0, sizeof(src6)); 1525 memset(&dst6, 0, sizeof(dst6)); 1526 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1527 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1528 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1529 #endif 1530 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1531 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1532 #ifdef INET6 1533 in6p = NULL; 1534 #else 1535 ; 1536 #endif 1537 #ifdef INET6 1538 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1539 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1540 ; 1541 #endif 1542 else 1543 return NULL; 1544 1545 /* 1546 * Now that we've validated that we are actually communicating 1547 * with the host indicated in the ICMP message, locate the 1548 * ICMP header, recalculate the new MTU, and create the 1549 * corresponding routing entry. 1550 */ 1551 icp = (struct icmp *)((caddr_t)ip - 1552 offsetof(struct icmp, icmp_ip)); 1553 if (inp) { 1554 if ((tp = intotcpcb(inp)) == NULL) 1555 return NULL; 1556 } 1557 #ifdef INET6 1558 else if (in6p) { 1559 if ((tp = in6totcpcb(in6p)) == NULL) 1560 return NULL; 1561 } 1562 #endif 1563 else 1564 return NULL; 1565 seq = ntohl(th->th_seq); 1566 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1567 return NULL; 1568 /* 1569 * If the ICMP message advertises a Next-Hop MTU 1570 * equal or larger than the maximum packet size we have 1571 * ever sent, drop the message. 1572 */ 1573 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1574 if (mtu >= tp->t_pmtud_mtu_sent) 1575 return NULL; 1576 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1577 /* 1578 * Calculate new MTU, and create corresponding 1579 * route (traditional PMTUD). 1580 */ 1581 tp->t_flags &= ~TF_PMTUD_PEND; 1582 icmp_mtudisc(icp, ip->ip_dst); 1583 } else { 1584 /* 1585 * Record the information got in the ICMP 1586 * message; act on it later. 1587 * If we had already recorded an ICMP message, 1588 * replace the old one only if the new message 1589 * refers to an older TCP segment 1590 */ 1591 if (tp->t_flags & TF_PMTUD_PEND) { 1592 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1593 return NULL; 1594 } else 1595 tp->t_flags |= TF_PMTUD_PEND; 1596 tp->t_pmtud_th_seq = seq; 1597 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1598 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1599 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1600 } 1601 return NULL; 1602 } else if (cmd == PRC_HOSTDEAD) 1603 ip = 0; 1604 else if (errno == 0) 1605 return NULL; 1606 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1607 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1608 nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, 1609 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1610 if (nmatch == 0 && syn_cache_count && 1611 (inetctlerrmap[cmd] == EHOSTUNREACH || 1612 inetctlerrmap[cmd] == ENETUNREACH || 1613 inetctlerrmap[cmd] == EHOSTDOWN)) { 1614 struct sockaddr_in sin; 1615 bzero(&sin, sizeof(sin)); 1616 sin.sin_len = sizeof(sin); 1617 sin.sin_family = AF_INET; 1618 sin.sin_port = th->th_sport; 1619 sin.sin_addr = ip->ip_src; 1620 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1621 } 1622 1623 /* XXX mapped address case */ 1624 } else 1625 in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno, 1626 notify); 1627 return NULL; 1628 } 1629 1630 /* 1631 * When a source quench is received, we are being notified of congestion. 1632 * Close the congestion window down to the Loss Window (one segment). 1633 * We will gradually open it again as we proceed. 1634 */ 1635 void 1636 tcp_quench(struct inpcb *inp, int errno) 1637 { 1638 struct tcpcb *tp = intotcpcb(inp); 1639 1640 if (tp) 1641 tp->snd_cwnd = tp->t_segsz; 1642 } 1643 #endif 1644 1645 #ifdef INET6 1646 void 1647 tcp6_quench(struct in6pcb *in6p, int errno) 1648 { 1649 struct tcpcb *tp = in6totcpcb(in6p); 1650 1651 if (tp) 1652 tp->snd_cwnd = tp->t_segsz; 1653 } 1654 #endif 1655 1656 #ifdef INET 1657 /* 1658 * Path MTU Discovery handlers. 1659 */ 1660 void 1661 tcp_mtudisc_callback(struct in_addr faddr) 1662 { 1663 #ifdef INET6 1664 struct in6_addr in6; 1665 #endif 1666 1667 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1668 #ifdef INET6 1669 memset(&in6, 0, sizeof(in6)); 1670 in6.s6_addr16[5] = 0xffff; 1671 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1672 tcp6_mtudisc_callback(&in6); 1673 #endif 1674 } 1675 1676 /* 1677 * On receipt of path MTU corrections, flush old route and replace it 1678 * with the new one. Retransmit all unacknowledged packets, to ensure 1679 * that all packets will be received. 1680 */ 1681 void 1682 tcp_mtudisc(struct inpcb *inp, int errno) 1683 { 1684 struct tcpcb *tp = intotcpcb(inp); 1685 struct rtentry *rt = in_pcbrtentry(inp); 1686 1687 if (tp != 0) { 1688 if (rt != 0) { 1689 /* 1690 * If this was not a host route, remove and realloc. 1691 */ 1692 if ((rt->rt_flags & RTF_HOST) == 0) { 1693 in_rtchange(inp, errno); 1694 if ((rt = in_pcbrtentry(inp)) == 0) 1695 return; 1696 } 1697 1698 /* 1699 * Slow start out of the error condition. We 1700 * use the MTU because we know it's smaller 1701 * than the previously transmitted segment. 1702 * 1703 * Note: This is more conservative than the 1704 * suggestion in draft-floyd-incr-init-win-03. 1705 */ 1706 if (rt->rt_rmx.rmx_mtu != 0) 1707 tp->snd_cwnd = 1708 TCP_INITIAL_WINDOW(tcp_init_win, 1709 rt->rt_rmx.rmx_mtu); 1710 } 1711 1712 /* 1713 * Resend unacknowledged packets. 1714 */ 1715 tp->snd_nxt = tp->snd_una; 1716 tcp_output(tp); 1717 } 1718 } 1719 #endif 1720 1721 #ifdef INET6 1722 /* 1723 * Path MTU Discovery handlers. 1724 */ 1725 void 1726 tcp6_mtudisc_callback(struct in6_addr *faddr) 1727 { 1728 struct sockaddr_in6 sin6; 1729 1730 bzero(&sin6, sizeof(sin6)); 1731 sin6.sin6_family = AF_INET6; 1732 sin6.sin6_len = sizeof(struct sockaddr_in6); 1733 sin6.sin6_addr = *faddr; 1734 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1735 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1736 } 1737 1738 void 1739 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1740 { 1741 struct tcpcb *tp = in6totcpcb(in6p); 1742 struct rtentry *rt = in6_pcbrtentry(in6p); 1743 1744 if (tp != 0) { 1745 if (rt != 0) { 1746 /* 1747 * If this was not a host route, remove and realloc. 1748 */ 1749 if ((rt->rt_flags & RTF_HOST) == 0) { 1750 in6_rtchange(in6p, errno); 1751 if ((rt = in6_pcbrtentry(in6p)) == 0) 1752 return; 1753 } 1754 1755 /* 1756 * Slow start out of the error condition. We 1757 * use the MTU because we know it's smaller 1758 * than the previously transmitted segment. 1759 * 1760 * Note: This is more conservative than the 1761 * suggestion in draft-floyd-incr-init-win-03. 1762 */ 1763 if (rt->rt_rmx.rmx_mtu != 0) 1764 tp->snd_cwnd = 1765 TCP_INITIAL_WINDOW(tcp_init_win, 1766 rt->rt_rmx.rmx_mtu); 1767 } 1768 1769 /* 1770 * Resend unacknowledged packets. 1771 */ 1772 tp->snd_nxt = tp->snd_una; 1773 tcp_output(tp); 1774 } 1775 } 1776 #endif /* INET6 */ 1777 1778 /* 1779 * Compute the MSS to advertise to the peer. Called only during 1780 * the 3-way handshake. If we are the server (peer initiated 1781 * connection), we are called with a pointer to the interface 1782 * on which the SYN packet arrived. If we are the client (we 1783 * initiated connection), we are called with a pointer to the 1784 * interface out which this connection should go. 1785 * 1786 * NOTE: Do not subtract IP option/extension header size nor IPsec 1787 * header size from MSS advertisement. MSS option must hold the maximum 1788 * segment size we can accept, so it must always be: 1789 * max(if mtu) - ip header - tcp header 1790 */ 1791 u_long 1792 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1793 { 1794 extern u_long in_maxmtu; 1795 u_long mss = 0; 1796 u_long hdrsiz; 1797 1798 /* 1799 * In order to avoid defeating path MTU discovery on the peer, 1800 * we advertise the max MTU of all attached networks as our MSS, 1801 * per RFC 1191, section 3.1. 1802 * 1803 * We provide the option to advertise just the MTU of 1804 * the interface on which we hope this connection will 1805 * be receiving. If we are responding to a SYN, we 1806 * will have a pretty good idea about this, but when 1807 * initiating a connection there is a bit more doubt. 1808 * 1809 * We also need to ensure that loopback has a large enough 1810 * MSS, as the loopback MTU is never included in in_maxmtu. 1811 */ 1812 1813 if (ifp != NULL) 1814 switch (af) { 1815 case AF_INET: 1816 mss = ifp->if_mtu; 1817 break; 1818 #ifdef INET6 1819 case AF_INET6: 1820 mss = IN6_LINKMTU(ifp); 1821 break; 1822 #endif 1823 } 1824 1825 if (tcp_mss_ifmtu == 0) 1826 switch (af) { 1827 case AF_INET: 1828 mss = max(in_maxmtu, mss); 1829 break; 1830 #ifdef INET6 1831 case AF_INET6: 1832 mss = max(in6_maxmtu, mss); 1833 break; 1834 #endif 1835 } 1836 1837 switch (af) { 1838 case AF_INET: 1839 hdrsiz = sizeof(struct ip); 1840 break; 1841 #ifdef INET6 1842 case AF_INET6: 1843 hdrsiz = sizeof(struct ip6_hdr); 1844 break; 1845 #endif 1846 default: 1847 hdrsiz = 0; 1848 break; 1849 } 1850 hdrsiz += sizeof(struct tcphdr); 1851 if (mss > hdrsiz) 1852 mss -= hdrsiz; 1853 1854 mss = max(tcp_mssdflt, mss); 1855 return (mss); 1856 } 1857 1858 /* 1859 * Set connection variables based on the peer's advertised MSS. 1860 * We are passed the TCPCB for the actual connection. If we 1861 * are the server, we are called by the compressed state engine 1862 * when the 3-way handshake is complete. If we are the client, 1863 * we are called when we receive the SYN,ACK from the server. 1864 * 1865 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1866 * before this routine is called! 1867 */ 1868 void 1869 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1870 { 1871 struct socket *so; 1872 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1873 struct rtentry *rt; 1874 #endif 1875 u_long bufsize; 1876 int mss; 1877 1878 #ifdef DIAGNOSTIC 1879 if (tp->t_inpcb && tp->t_in6pcb) 1880 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1881 #endif 1882 so = NULL; 1883 rt = NULL; 1884 #ifdef INET 1885 if (tp->t_inpcb) { 1886 so = tp->t_inpcb->inp_socket; 1887 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1888 rt = in_pcbrtentry(tp->t_inpcb); 1889 #endif 1890 } 1891 #endif 1892 #ifdef INET6 1893 if (tp->t_in6pcb) { 1894 so = tp->t_in6pcb->in6p_socket; 1895 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1896 rt = in6_pcbrtentry(tp->t_in6pcb); 1897 #endif 1898 } 1899 #endif 1900 1901 /* 1902 * As per RFC1122, use the default MSS value, unless they 1903 * sent us an offer. Do not accept offers less than 256 bytes. 1904 */ 1905 mss = tcp_mssdflt; 1906 if (offer) 1907 mss = offer; 1908 mss = max(mss, 256); /* sanity */ 1909 tp->t_peermss = mss; 1910 mss -= tcp_optlen(tp); 1911 #ifdef INET 1912 if (tp->t_inpcb) 1913 mss -= ip_optlen(tp->t_inpcb); 1914 #endif 1915 #ifdef INET6 1916 if (tp->t_in6pcb) 1917 mss -= ip6_optlen(tp->t_in6pcb); 1918 #endif 1919 1920 /* 1921 * If there's a pipesize, change the socket buffer to that size. 1922 * Make the socket buffer an integral number of MSS units. If 1923 * the MSS is larger than the socket buffer, artificially decrease 1924 * the MSS. 1925 */ 1926 #ifdef RTV_SPIPE 1927 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1928 bufsize = rt->rt_rmx.rmx_sendpipe; 1929 else 1930 #endif 1931 bufsize = so->so_snd.sb_hiwat; 1932 if (bufsize < mss) 1933 mss = bufsize; 1934 else { 1935 bufsize = roundup(bufsize, mss); 1936 if (bufsize > sb_max) 1937 bufsize = sb_max; 1938 (void) sbreserve(&so->so_snd, bufsize, so); 1939 } 1940 tp->t_segsz = mss; 1941 1942 #ifdef RTV_SSTHRESH 1943 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1944 /* 1945 * There's some sort of gateway or interface buffer 1946 * limit on the path. Use this to set the slow 1947 * start threshold, but set the threshold to no less 1948 * than 2 * MSS. 1949 */ 1950 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1951 } 1952 #endif 1953 } 1954 1955 /* 1956 * Processing necessary when a TCP connection is established. 1957 */ 1958 void 1959 tcp_established(struct tcpcb *tp) 1960 { 1961 struct socket *so; 1962 #ifdef RTV_RPIPE 1963 struct rtentry *rt; 1964 #endif 1965 u_long bufsize; 1966 1967 #ifdef DIAGNOSTIC 1968 if (tp->t_inpcb && tp->t_in6pcb) 1969 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1970 #endif 1971 so = NULL; 1972 rt = NULL; 1973 #ifdef INET 1974 if (tp->t_inpcb) { 1975 so = tp->t_inpcb->inp_socket; 1976 #if defined(RTV_RPIPE) 1977 rt = in_pcbrtentry(tp->t_inpcb); 1978 #endif 1979 } 1980 #endif 1981 #ifdef INET6 1982 if (tp->t_in6pcb) { 1983 so = tp->t_in6pcb->in6p_socket; 1984 #if defined(RTV_RPIPE) 1985 rt = in6_pcbrtentry(tp->t_in6pcb); 1986 #endif 1987 } 1988 #endif 1989 1990 tp->t_state = TCPS_ESTABLISHED; 1991 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1992 1993 #ifdef RTV_RPIPE 1994 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 1995 bufsize = rt->rt_rmx.rmx_recvpipe; 1996 else 1997 #endif 1998 bufsize = so->so_rcv.sb_hiwat; 1999 if (bufsize > tp->t_ourmss) { 2000 bufsize = roundup(bufsize, tp->t_ourmss); 2001 if (bufsize > sb_max) 2002 bufsize = sb_max; 2003 (void) sbreserve(&so->so_rcv, bufsize, so); 2004 } 2005 } 2006 2007 /* 2008 * Check if there's an initial rtt or rttvar. Convert from the 2009 * route-table units to scaled multiples of the slow timeout timer. 2010 * Called only during the 3-way handshake. 2011 */ 2012 void 2013 tcp_rmx_rtt(struct tcpcb *tp) 2014 { 2015 #ifdef RTV_RTT 2016 struct rtentry *rt = NULL; 2017 int rtt; 2018 2019 #ifdef DIAGNOSTIC 2020 if (tp->t_inpcb && tp->t_in6pcb) 2021 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2022 #endif 2023 #ifdef INET 2024 if (tp->t_inpcb) 2025 rt = in_pcbrtentry(tp->t_inpcb); 2026 #endif 2027 #ifdef INET6 2028 if (tp->t_in6pcb) 2029 rt = in6_pcbrtentry(tp->t_in6pcb); 2030 #endif 2031 if (rt == NULL) 2032 return; 2033 2034 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2035 /* 2036 * XXX The lock bit for MTU indicates that the value 2037 * is also a minimum value; this is subject to time. 2038 */ 2039 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2040 TCPT_RANGESET(tp->t_rttmin, 2041 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2042 TCPTV_MIN, TCPTV_REXMTMAX); 2043 tp->t_srtt = rtt / 2044 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2045 if (rt->rt_rmx.rmx_rttvar) { 2046 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2047 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2048 (TCP_RTTVAR_SHIFT + 2)); 2049 } else { 2050 /* Default variation is +- 1 rtt */ 2051 tp->t_rttvar = 2052 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2053 } 2054 TCPT_RANGESET(tp->t_rxtcur, 2055 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2056 tp->t_rttmin, TCPTV_REXMTMAX); 2057 } 2058 #endif 2059 } 2060 2061 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2062 #if NRND > 0 2063 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2064 #endif 2065 2066 /* 2067 * Get a new sequence value given a tcp control block 2068 */ 2069 tcp_seq 2070 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2071 { 2072 2073 #ifdef INET 2074 if (tp->t_inpcb != NULL) { 2075 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2076 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2077 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2078 addin)); 2079 } 2080 #endif 2081 #ifdef INET6 2082 if (tp->t_in6pcb != NULL) { 2083 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2084 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2085 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2086 addin)); 2087 } 2088 #endif 2089 /* Not possible. */ 2090 panic("tcp_new_iss"); 2091 } 2092 2093 /* 2094 * This routine actually generates a new TCP initial sequence number. 2095 */ 2096 tcp_seq 2097 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2098 size_t addrsz, tcp_seq addin) 2099 { 2100 tcp_seq tcp_iss; 2101 2102 #if NRND > 0 2103 static int beenhere; 2104 2105 /* 2106 * If we haven't been here before, initialize our cryptographic 2107 * hash secret. 2108 */ 2109 if (beenhere == 0) { 2110 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2111 RND_EXTRACT_ANY); 2112 beenhere = 1; 2113 } 2114 2115 if (tcp_do_rfc1948) { 2116 MD5_CTX ctx; 2117 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2118 2119 /* 2120 * Compute the base value of the ISS. It is a hash 2121 * of (saddr, sport, daddr, dport, secret). 2122 */ 2123 MD5Init(&ctx); 2124 2125 MD5Update(&ctx, (u_char *) laddr, addrsz); 2126 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2127 2128 MD5Update(&ctx, (u_char *) faddr, addrsz); 2129 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2130 2131 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2132 2133 MD5Final(hash, &ctx); 2134 2135 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2136 2137 /* 2138 * Now increment our "timer", and add it in to 2139 * the computed value. 2140 * 2141 * XXX Use `addin'? 2142 * XXX TCP_ISSINCR too large to use? 2143 */ 2144 tcp_iss_seq += TCP_ISSINCR; 2145 #ifdef TCPISS_DEBUG 2146 printf("ISS hash 0x%08x, ", tcp_iss); 2147 #endif 2148 tcp_iss += tcp_iss_seq + addin; 2149 #ifdef TCPISS_DEBUG 2150 printf("new ISS 0x%08x\n", tcp_iss); 2151 #endif 2152 } else 2153 #endif /* NRND > 0 */ 2154 { 2155 /* 2156 * Randomize. 2157 */ 2158 #if NRND > 0 2159 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2160 #else 2161 tcp_iss = arc4random(); 2162 #endif 2163 2164 /* 2165 * If we were asked to add some amount to a known value, 2166 * we will take a random value obtained above, mask off 2167 * the upper bits, and add in the known value. We also 2168 * add in a constant to ensure that we are at least a 2169 * certain distance from the original value. 2170 * 2171 * This is used when an old connection is in timed wait 2172 * and we have a new one coming in, for instance. 2173 */ 2174 if (addin != 0) { 2175 #ifdef TCPISS_DEBUG 2176 printf("Random %08x, ", tcp_iss); 2177 #endif 2178 tcp_iss &= TCP_ISS_RANDOM_MASK; 2179 tcp_iss += addin + TCP_ISSINCR; 2180 #ifdef TCPISS_DEBUG 2181 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2182 #endif 2183 } else { 2184 tcp_iss &= TCP_ISS_RANDOM_MASK; 2185 tcp_iss += tcp_iss_seq; 2186 tcp_iss_seq += TCP_ISSINCR; 2187 #ifdef TCPISS_DEBUG 2188 printf("ISS %08x\n", tcp_iss); 2189 #endif 2190 } 2191 } 2192 2193 if (tcp_compat_42) { 2194 /* 2195 * Limit it to the positive range for really old TCP 2196 * implementations. 2197 * Just AND off the top bit instead of checking if 2198 * is set first - saves a branch 50% of the time. 2199 */ 2200 tcp_iss &= 0x7fffffff; /* XXX */ 2201 } 2202 2203 return (tcp_iss); 2204 } 2205 2206 #if defined(IPSEC) || defined(FAST_IPSEC) 2207 /* compute ESP/AH header size for TCP, including outer IP header. */ 2208 size_t 2209 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2210 { 2211 struct inpcb *inp; 2212 size_t hdrsiz; 2213 2214 /* XXX mapped addr case (tp->t_in6pcb) */ 2215 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2216 return 0; 2217 switch (tp->t_family) { 2218 case AF_INET: 2219 /* XXX: should use currect direction. */ 2220 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2221 break; 2222 default: 2223 hdrsiz = 0; 2224 break; 2225 } 2226 2227 return hdrsiz; 2228 } 2229 2230 #ifdef INET6 2231 size_t 2232 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2233 { 2234 struct in6pcb *in6p; 2235 size_t hdrsiz; 2236 2237 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2238 return 0; 2239 switch (tp->t_family) { 2240 case AF_INET6: 2241 /* XXX: should use currect direction. */ 2242 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2243 break; 2244 case AF_INET: 2245 /* mapped address case - tricky */ 2246 default: 2247 hdrsiz = 0; 2248 break; 2249 } 2250 2251 return hdrsiz; 2252 } 2253 #endif 2254 #endif /*IPSEC*/ 2255 2256 /* 2257 * Determine the length of the TCP options for this connection. 2258 * 2259 * XXX: What do we do for SACK, when we add that? Just reserve 2260 * all of the space? Otherwise we can't exactly be incrementing 2261 * cwnd by an amount that varies depending on the amount we last 2262 * had to SACK! 2263 */ 2264 2265 u_int 2266 tcp_optlen(struct tcpcb *tp) 2267 { 2268 u_int optlen; 2269 2270 optlen = 0; 2271 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2272 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2273 optlen += TCPOLEN_TSTAMP_APPA; 2274 2275 #ifdef TCP_SIGNATURE 2276 #if defined(INET6) && defined(FAST_IPSEC) 2277 if (tp->t_family == AF_INET) 2278 #endif 2279 if (tp->t_flags & TF_SIGNATURE) 2280 optlen += TCPOLEN_SIGNATURE + 2; 2281 #endif /* TCP_SIGNATURE */ 2282 2283 return optlen; 2284 } 2285 2286 u_int 2287 tcp_hdrsz(struct tcpcb *tp) 2288 { 2289 u_int hlen; 2290 2291 switch (tp->t_family) { 2292 #ifdef INET6 2293 case AF_INET6: 2294 hlen = sizeof(struct ip6_hdr); 2295 break; 2296 #endif 2297 case AF_INET: 2298 hlen = sizeof(struct ip); 2299 break; 2300 default: 2301 hlen = 0; 2302 break; 2303 } 2304 hlen += sizeof(struct tcphdr); 2305 2306 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2307 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2308 hlen += TCPOLEN_TSTAMP_APPA; 2309 #ifdef TCP_SIGNATURE 2310 if (tp->t_flags & TF_SIGNATURE) 2311 hlen += TCPOLEN_SIGLEN; 2312 #endif 2313 return hlen; 2314 } 2315
Indexes created Sun Sep 21 20:09:37 GMT 2025