tcp_subr.c revision 1.231
1 /* $NetBSD: tcp_subr.c,v 1.231 2008/05/02 13:40:33 ad Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1997, 1998, 2000, 2001, 2008 The NetBSD Foundation, Inc. 34 * All rights reserved. 35 * 36 * This code is derived from software contributed to The NetBSD Foundation 37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 38 * Facility, NASA Ames Research Center. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 50 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 51 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 52 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 53 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 54 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 55 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 56 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 57 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 58 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 59 * POSSIBILITY OF SUCH DAMAGE. 60 */ 61 62 /* 63 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 64 * The Regents of the University of California. All rights reserved. 65 * 66 * Redistribution and use in source and binary forms, with or without 67 * modification, are permitted provided that the following conditions 68 * are met: 69 * 1. Redistributions of source code must retain the above copyright 70 * notice, this list of conditions and the following disclaimer. 71 * 2. Redistributions in binary form must reproduce the above copyright 72 * notice, this list of conditions and the following disclaimer in the 73 * documentation and/or other materials provided with the distribution. 74 * 3. Neither the name of the University nor the names of its contributors 75 * may be used to endorse or promote products derived from this software 76 * without specific prior written permission. 77 * 78 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 79 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 80 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 81 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 82 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 83 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 84 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 85 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 86 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 87 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 88 * SUCH DAMAGE. 89 * 90 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 91 */ 92 93 #include <sys/cdefs.h> 94 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.231 2008/05/02 13:40:33 ad Exp $"); 95 96 #include "opt_inet.h" 97 #include "opt_ipsec.h" 98 #include "opt_tcp_compat_42.h" 99 #include "opt_inet_csum.h" 100 #include "opt_mbuftrace.h" 101 #include "rnd.h" 102 103 #include <sys/param.h> 104 #include <sys/proc.h> 105 #include <sys/systm.h> 106 #include <sys/malloc.h> 107 #include <sys/mbuf.h> 108 #include <sys/socket.h> 109 #include <sys/socketvar.h> 110 #include <sys/protosw.h> 111 #include <sys/errno.h> 112 #include <sys/kernel.h> 113 #include <sys/pool.h> 114 #if NRND > 0 115 #include <sys/md5.h> 116 #include <sys/rnd.h> 117 #endif 118 119 #include <net/route.h> 120 #include <net/if.h> 121 122 #include <netinet/in.h> 123 #include <netinet/in_systm.h> 124 #include <netinet/ip.h> 125 #include <netinet/in_pcb.h> 126 #include <netinet/ip_var.h> 127 #include <netinet/ip_icmp.h> 128 129 #ifdef INET6 130 #ifndef INET 131 #include <netinet/in.h> 132 #endif 133 #include <netinet/ip6.h> 134 #include <netinet6/in6_pcb.h> 135 #include <netinet6/ip6_var.h> 136 #include <netinet6/in6_var.h> 137 #include <netinet6/ip6protosw.h> 138 #include <netinet/icmp6.h> 139 #include <netinet6/nd6.h> 140 #endif 141 142 #include <netinet/tcp.h> 143 #include <netinet/tcp_fsm.h> 144 #include <netinet/tcp_seq.h> 145 #include <netinet/tcp_timer.h> 146 #include <netinet/tcp_var.h> 147 #include <netinet/tcp_private.h> 148 #include <netinet/tcp_congctl.h> 149 #include <netinet/tcpip.h> 150 151 #ifdef IPSEC 152 #include <netinet6/ipsec.h> 153 #include <netkey/key.h> 154 #endif /*IPSEC*/ 155 156 #ifdef FAST_IPSEC 157 #include <netipsec/ipsec.h> 158 #include <netipsec/xform.h> 159 #ifdef INET6 160 #include <netipsec/ipsec6.h> 161 #endif 162 #include <netipsec/key.h> 163 #endif /* FAST_IPSEC*/ 164 165 166 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 167 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 168 169 percpu_t *tcpstat_percpu; 170 171 /* patchable/settable parameters for tcp */ 172 int tcp_mssdflt = TCP_MSS; 173 int tcp_minmss = TCP_MINMSS; 174 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 175 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 176 #if NRND > 0 177 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 178 #endif 179 int tcp_do_sack = 1; /* selective acknowledgement */ 180 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 181 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 182 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 183 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 184 #ifndef TCP_INIT_WIN 185 #define TCP_INIT_WIN 0 /* initial slow start window */ 186 #endif 187 #ifndef TCP_INIT_WIN_LOCAL 188 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 189 #endif 190 int tcp_init_win = TCP_INIT_WIN; 191 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 192 int tcp_mss_ifmtu = 0; 193 #ifdef TCP_COMPAT_42 194 int tcp_compat_42 = 1; 195 #else 196 int tcp_compat_42 = 0; 197 #endif 198 int tcp_rst_ppslim = 100; /* 100pps */ 199 int tcp_ackdrop_ppslim = 100; /* 100pps */ 200 int tcp_do_loopback_cksum = 0; 201 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 202 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 203 int tcp_sack_tp_maxholes = 32; 204 int tcp_sack_globalmaxholes = 1024; 205 int tcp_sack_globalholes = 0; 206 int tcp_ecn_maxretries = 1; 207 208 /* tcb hash */ 209 #ifndef TCBHASHSIZE 210 #define TCBHASHSIZE 128 211 #endif 212 int tcbhashsize = TCBHASHSIZE; 213 214 /* syn hash parameters */ 215 #define TCP_SYN_HASH_SIZE 293 216 #define TCP_SYN_BUCKET_SIZE 35 217 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 218 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 219 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 220 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 221 222 int tcp_freeq(struct tcpcb *); 223 224 #ifdef INET 225 void tcp_mtudisc_callback(struct in_addr); 226 #endif 227 #ifdef INET6 228 void tcp6_mtudisc_callback(struct in6_addr *); 229 #endif 230 231 #ifdef INET6 232 void tcp6_mtudisc(struct in6pcb *, int); 233 #endif 234 235 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL, 236 IPL_SOFTNET); 237 238 #ifdef TCP_CSUM_COUNTERS 239 #include <sys/device.h> 240 241 #if defined(INET) 242 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 243 NULL, "tcp", "hwcsum bad"); 244 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 245 NULL, "tcp", "hwcsum ok"); 246 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 247 NULL, "tcp", "hwcsum data"); 248 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 249 NULL, "tcp", "swcsum"); 250 251 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 252 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 253 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 254 EVCNT_ATTACH_STATIC(tcp_swcsum); 255 #endif /* defined(INET) */ 256 257 #if defined(INET6) 258 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 259 NULL, "tcp6", "hwcsum bad"); 260 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 261 NULL, "tcp6", "hwcsum ok"); 262 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 263 NULL, "tcp6", "hwcsum data"); 264 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 265 NULL, "tcp6", "swcsum"); 266 267 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 268 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 269 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 270 EVCNT_ATTACH_STATIC(tcp6_swcsum); 271 #endif /* defined(INET6) */ 272 #endif /* TCP_CSUM_COUNTERS */ 273 274 275 #ifdef TCP_OUTPUT_COUNTERS 276 #include <sys/device.h> 277 278 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 279 NULL, "tcp", "output big header"); 280 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 281 NULL, "tcp", "output predict hit"); 282 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 NULL, "tcp", "output predict miss"); 284 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 285 NULL, "tcp", "output copy small"); 286 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 287 NULL, "tcp", "output copy big"); 288 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 289 NULL, "tcp", "output reference big"); 290 291 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 292 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 293 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 294 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 295 EVCNT_ATTACH_STATIC(tcp_output_copybig); 296 EVCNT_ATTACH_STATIC(tcp_output_refbig); 297 298 #endif /* TCP_OUTPUT_COUNTERS */ 299 300 #ifdef TCP_REASS_COUNTERS 301 #include <sys/device.h> 302 303 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 304 NULL, "tcp_reass", "calls"); 305 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 306 &tcp_reass_, "tcp_reass", "insert into empty queue"); 307 struct evcnt tcp_reass_iteration[8] = { 308 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 316 }; 317 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 318 &tcp_reass_, "tcp_reass", "prepend to first"); 319 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 320 &tcp_reass_, "tcp_reass", "prepend"); 321 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 322 &tcp_reass_, "tcp_reass", "insert"); 323 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 324 &tcp_reass_, "tcp_reass", "insert at tail"); 325 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 326 &tcp_reass_, "tcp_reass", "append"); 327 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 328 &tcp_reass_, "tcp_reass", "append to tail fragment"); 329 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 330 &tcp_reass_, "tcp_reass", "overlap at end"); 331 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 332 &tcp_reass_, "tcp_reass", "overlap at start"); 333 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 334 &tcp_reass_, "tcp_reass", "duplicate segment"); 335 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 336 &tcp_reass_, "tcp_reass", "duplicate fragment"); 337 338 EVCNT_ATTACH_STATIC(tcp_reass_); 339 EVCNT_ATTACH_STATIC(tcp_reass_empty); 340 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 341 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 342 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 348 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 349 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 350 EVCNT_ATTACH_STATIC(tcp_reass_insert); 351 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 352 EVCNT_ATTACH_STATIC(tcp_reass_append); 353 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 354 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 355 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 356 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 357 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 358 359 #endif /* TCP_REASS_COUNTERS */ 360 361 #ifdef MBUFTRACE 362 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 363 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 364 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 365 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 366 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 367 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 368 #endif 369 370 /* 371 * Tcp initialization 372 */ 373 void 374 tcp_init(void) 375 { 376 int hlen; 377 378 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 379 380 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 381 #ifdef INET6 382 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 383 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 384 #endif 385 if (max_protohdr < hlen) 386 max_protohdr = hlen; 387 if (max_linkhdr + hlen > MHLEN) 388 panic("tcp_init"); 389 390 #ifdef INET 391 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 392 #endif 393 #ifdef INET6 394 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 395 #endif 396 397 /* Initialize timer state. */ 398 tcp_timer_init(); 399 400 /* Initialize the compressed state engine. */ 401 syn_cache_init(); 402 403 /* Initialize the congestion control algorithms. */ 404 tcp_congctl_init(); 405 406 /* Initialize the TCPCB template. */ 407 tcp_tcpcb_template(); 408 409 MOWNER_ATTACH(&tcp_tx_mowner); 410 MOWNER_ATTACH(&tcp_rx_mowner); 411 MOWNER_ATTACH(&tcp_reass_mowner); 412 MOWNER_ATTACH(&tcp_sock_mowner); 413 MOWNER_ATTACH(&tcp_sock_tx_mowner); 414 MOWNER_ATTACH(&tcp_sock_rx_mowner); 415 MOWNER_ATTACH(&tcp_mowner); 416 417 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS); 418 } 419 420 /* 421 * Create template to be used to send tcp packets on a connection. 422 * Call after host entry created, allocates an mbuf and fills 423 * in a skeletal tcp/ip header, minimizing the amount of work 424 * necessary when the connection is used. 425 */ 426 struct mbuf * 427 tcp_template(struct tcpcb *tp) 428 { 429 struct inpcb *inp = tp->t_inpcb; 430 #ifdef INET6 431 struct in6pcb *in6p = tp->t_in6pcb; 432 #endif 433 struct tcphdr *n; 434 struct mbuf *m; 435 int hlen; 436 437 switch (tp->t_family) { 438 case AF_INET: 439 hlen = sizeof(struct ip); 440 if (inp) 441 break; 442 #ifdef INET6 443 if (in6p) { 444 /* mapped addr case */ 445 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 446 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 447 break; 448 } 449 #endif 450 return NULL; /*EINVAL*/ 451 #ifdef INET6 452 case AF_INET6: 453 hlen = sizeof(struct ip6_hdr); 454 if (in6p) { 455 /* more sainty check? */ 456 break; 457 } 458 return NULL; /*EINVAL*/ 459 #endif 460 default: 461 hlen = 0; /*pacify gcc*/ 462 return NULL; /*EAFNOSUPPORT*/ 463 } 464 #ifdef DIAGNOSTIC 465 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 466 panic("mclbytes too small for t_template"); 467 #endif 468 m = tp->t_template; 469 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 470 ; 471 else { 472 if (m) 473 m_freem(m); 474 m = tp->t_template = NULL; 475 MGETHDR(m, M_DONTWAIT, MT_HEADER); 476 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 477 MCLGET(m, M_DONTWAIT); 478 if ((m->m_flags & M_EXT) == 0) { 479 m_free(m); 480 m = NULL; 481 } 482 } 483 if (m == NULL) 484 return NULL; 485 MCLAIM(m, &tcp_mowner); 486 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 487 } 488 489 bzero(mtod(m, void *), m->m_len); 490 491 n = (struct tcphdr *)(mtod(m, char *) + hlen); 492 493 switch (tp->t_family) { 494 case AF_INET: 495 { 496 struct ipovly *ipov; 497 mtod(m, struct ip *)->ip_v = 4; 498 mtod(m, struct ip *)->ip_hl = hlen >> 2; 499 ipov = mtod(m, struct ipovly *); 500 ipov->ih_pr = IPPROTO_TCP; 501 ipov->ih_len = htons(sizeof(struct tcphdr)); 502 if (inp) { 503 ipov->ih_src = inp->inp_laddr; 504 ipov->ih_dst = inp->inp_faddr; 505 } 506 #ifdef INET6 507 else if (in6p) { 508 /* mapped addr case */ 509 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 510 sizeof(ipov->ih_src)); 511 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 512 sizeof(ipov->ih_dst)); 513 } 514 #endif 515 /* 516 * Compute the pseudo-header portion of the checksum 517 * now. We incrementally add in the TCP option and 518 * payload lengths later, and then compute the TCP 519 * checksum right before the packet is sent off onto 520 * the wire. 521 */ 522 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 523 ipov->ih_dst.s_addr, 524 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 525 break; 526 } 527 #ifdef INET6 528 case AF_INET6: 529 { 530 struct ip6_hdr *ip6; 531 mtod(m, struct ip *)->ip_v = 6; 532 ip6 = mtod(m, struct ip6_hdr *); 533 ip6->ip6_nxt = IPPROTO_TCP; 534 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 535 ip6->ip6_src = in6p->in6p_laddr; 536 ip6->ip6_dst = in6p->in6p_faddr; 537 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 538 if (ip6_auto_flowlabel) { 539 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 540 ip6->ip6_flow |= 541 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 542 } 543 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 544 ip6->ip6_vfc |= IPV6_VERSION; 545 546 /* 547 * Compute the pseudo-header portion of the checksum 548 * now. We incrementally add in the TCP option and 549 * payload lengths later, and then compute the TCP 550 * checksum right before the packet is sent off onto 551 * the wire. 552 */ 553 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 554 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 555 htonl(IPPROTO_TCP)); 556 break; 557 } 558 #endif 559 } 560 if (inp) { 561 n->th_sport = inp->inp_lport; 562 n->th_dport = inp->inp_fport; 563 } 564 #ifdef INET6 565 else if (in6p) { 566 n->th_sport = in6p->in6p_lport; 567 n->th_dport = in6p->in6p_fport; 568 } 569 #endif 570 n->th_seq = 0; 571 n->th_ack = 0; 572 n->th_x2 = 0; 573 n->th_off = 5; 574 n->th_flags = 0; 575 n->th_win = 0; 576 n->th_urp = 0; 577 return (m); 578 } 579 580 /* 581 * Send a single message to the TCP at address specified by 582 * the given TCP/IP header. If m == 0, then we make a copy 583 * of the tcpiphdr at ti and send directly to the addressed host. 584 * This is used to force keep alive messages out using the TCP 585 * template for a connection tp->t_template. If flags are given 586 * then we send a message back to the TCP which originated the 587 * segment ti, and discard the mbuf containing it and any other 588 * attached mbufs. 589 * 590 * In any case the ack and sequence number of the transmitted 591 * segment are as specified by the parameters. 592 */ 593 int 594 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 595 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 596 { 597 #ifdef INET6 598 struct rtentry *rt; 599 #endif 600 struct route *ro; 601 int error, tlen, win = 0; 602 int hlen; 603 struct ip *ip; 604 #ifdef INET6 605 struct ip6_hdr *ip6; 606 #endif 607 int family; /* family on packet, not inpcb/in6pcb! */ 608 struct tcphdr *th; 609 struct socket *so; 610 611 if (tp != NULL && (flags & TH_RST) == 0) { 612 #ifdef DIAGNOSTIC 613 if (tp->t_inpcb && tp->t_in6pcb) 614 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 615 #endif 616 #ifdef INET 617 if (tp->t_inpcb) 618 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 619 #endif 620 #ifdef INET6 621 if (tp->t_in6pcb) 622 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 623 #endif 624 } 625 626 th = NULL; /* Quell uninitialized warning */ 627 ip = NULL; 628 #ifdef INET6 629 ip6 = NULL; 630 #endif 631 if (m == 0) { 632 if (!template) 633 return EINVAL; 634 635 /* get family information from template */ 636 switch (mtod(template, struct ip *)->ip_v) { 637 case 4: 638 family = AF_INET; 639 hlen = sizeof(struct ip); 640 break; 641 #ifdef INET6 642 case 6: 643 family = AF_INET6; 644 hlen = sizeof(struct ip6_hdr); 645 break; 646 #endif 647 default: 648 return EAFNOSUPPORT; 649 } 650 651 MGETHDR(m, M_DONTWAIT, MT_HEADER); 652 if (m) { 653 MCLAIM(m, &tcp_tx_mowner); 654 MCLGET(m, M_DONTWAIT); 655 if ((m->m_flags & M_EXT) == 0) { 656 m_free(m); 657 m = NULL; 658 } 659 } 660 if (m == NULL) 661 return (ENOBUFS); 662 663 if (tcp_compat_42) 664 tlen = 1; 665 else 666 tlen = 0; 667 668 m->m_data += max_linkhdr; 669 bcopy(mtod(template, void *), mtod(m, void *), 670 template->m_len); 671 switch (family) { 672 case AF_INET: 673 ip = mtod(m, struct ip *); 674 th = (struct tcphdr *)(ip + 1); 675 break; 676 #ifdef INET6 677 case AF_INET6: 678 ip6 = mtod(m, struct ip6_hdr *); 679 th = (struct tcphdr *)(ip6 + 1); 680 break; 681 #endif 682 #if 0 683 default: 684 /* noone will visit here */ 685 m_freem(m); 686 return EAFNOSUPPORT; 687 #endif 688 } 689 flags = TH_ACK; 690 } else { 691 692 if ((m->m_flags & M_PKTHDR) == 0) { 693 #if 0 694 printf("non PKTHDR to tcp_respond\n"); 695 #endif 696 m_freem(m); 697 return EINVAL; 698 } 699 #ifdef DIAGNOSTIC 700 if (!th0) 701 panic("th0 == NULL in tcp_respond"); 702 #endif 703 704 /* get family information from m */ 705 switch (mtod(m, struct ip *)->ip_v) { 706 case 4: 707 family = AF_INET; 708 hlen = sizeof(struct ip); 709 ip = mtod(m, struct ip *); 710 break; 711 #ifdef INET6 712 case 6: 713 family = AF_INET6; 714 hlen = sizeof(struct ip6_hdr); 715 ip6 = mtod(m, struct ip6_hdr *); 716 break; 717 #endif 718 default: 719 m_freem(m); 720 return EAFNOSUPPORT; 721 } 722 /* clear h/w csum flags inherited from rx packet */ 723 m->m_pkthdr.csum_flags = 0; 724 725 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 726 tlen = sizeof(*th0); 727 else 728 tlen = th0->th_off << 2; 729 730 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 731 mtod(m, char *) + hlen == (char *)th0) { 732 m->m_len = hlen + tlen; 733 m_freem(m->m_next); 734 m->m_next = NULL; 735 } else { 736 struct mbuf *n; 737 738 #ifdef DIAGNOSTIC 739 if (max_linkhdr + hlen + tlen > MCLBYTES) { 740 m_freem(m); 741 return EMSGSIZE; 742 } 743 #endif 744 MGETHDR(n, M_DONTWAIT, MT_HEADER); 745 if (n && max_linkhdr + hlen + tlen > MHLEN) { 746 MCLGET(n, M_DONTWAIT); 747 if ((n->m_flags & M_EXT) == 0) { 748 m_freem(n); 749 n = NULL; 750 } 751 } 752 if (!n) { 753 m_freem(m); 754 return ENOBUFS; 755 } 756 757 MCLAIM(n, &tcp_tx_mowner); 758 n->m_data += max_linkhdr; 759 n->m_len = hlen + tlen; 760 m_copyback(n, 0, hlen, mtod(m, void *)); 761 m_copyback(n, hlen, tlen, (void *)th0); 762 763 m_freem(m); 764 m = n; 765 n = NULL; 766 } 767 768 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 769 switch (family) { 770 case AF_INET: 771 ip = mtod(m, struct ip *); 772 th = (struct tcphdr *)(ip + 1); 773 ip->ip_p = IPPROTO_TCP; 774 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 775 ip->ip_p = IPPROTO_TCP; 776 break; 777 #ifdef INET6 778 case AF_INET6: 779 ip6 = mtod(m, struct ip6_hdr *); 780 th = (struct tcphdr *)(ip6 + 1); 781 ip6->ip6_nxt = IPPROTO_TCP; 782 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 783 ip6->ip6_nxt = IPPROTO_TCP; 784 break; 785 #endif 786 #if 0 787 default: 788 /* noone will visit here */ 789 m_freem(m); 790 return EAFNOSUPPORT; 791 #endif 792 } 793 xchg(th->th_dport, th->th_sport, u_int16_t); 794 #undef xchg 795 tlen = 0; /*be friendly with the following code*/ 796 } 797 th->th_seq = htonl(seq); 798 th->th_ack = htonl(ack); 799 th->th_x2 = 0; 800 if ((flags & TH_SYN) == 0) { 801 if (tp) 802 win >>= tp->rcv_scale; 803 if (win > TCP_MAXWIN) 804 win = TCP_MAXWIN; 805 th->th_win = htons((u_int16_t)win); 806 th->th_off = sizeof (struct tcphdr) >> 2; 807 tlen += sizeof(*th); 808 } else 809 tlen += th->th_off << 2; 810 m->m_len = hlen + tlen; 811 m->m_pkthdr.len = hlen + tlen; 812 m->m_pkthdr.rcvif = (struct ifnet *) 0; 813 th->th_flags = flags; 814 th->th_urp = 0; 815 816 switch (family) { 817 #ifdef INET 818 case AF_INET: 819 { 820 struct ipovly *ipov = (struct ipovly *)ip; 821 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 822 ipov->ih_len = htons((u_int16_t)tlen); 823 824 th->th_sum = 0; 825 th->th_sum = in_cksum(m, hlen + tlen); 826 ip->ip_len = htons(hlen + tlen); 827 ip->ip_ttl = ip_defttl; 828 break; 829 } 830 #endif 831 #ifdef INET6 832 case AF_INET6: 833 { 834 th->th_sum = 0; 835 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 836 tlen); 837 ip6->ip6_plen = htons(tlen); 838 if (tp && tp->t_in6pcb) { 839 struct ifnet *oifp; 840 ro = &tp->t_in6pcb->in6p_route; 841 oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 842 : NULL; 843 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 844 } else 845 ip6->ip6_hlim = ip6_defhlim; 846 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 847 if (ip6_auto_flowlabel) { 848 ip6->ip6_flow |= 849 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 850 } 851 break; 852 } 853 #endif 854 } 855 856 if (tp && tp->t_inpcb) 857 so = tp->t_inpcb->inp_socket; 858 #ifdef INET6 859 else if (tp && tp->t_in6pcb) 860 so = tp->t_in6pcb->in6p_socket; 861 #endif 862 else 863 so = NULL; 864 865 if (tp != NULL && tp->t_inpcb != NULL) { 866 ro = &tp->t_inpcb->inp_route; 867 #ifdef DIAGNOSTIC 868 if (family != AF_INET) 869 panic("tcp_respond: address family mismatch"); 870 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 871 panic("tcp_respond: ip_dst %x != inp_faddr %x", 872 ntohl(ip->ip_dst.s_addr), 873 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 874 } 875 #endif 876 } 877 #ifdef INET6 878 else if (tp != NULL && tp->t_in6pcb != NULL) { 879 ro = (struct route *)&tp->t_in6pcb->in6p_route; 880 #ifdef DIAGNOSTIC 881 if (family == AF_INET) { 882 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 883 panic("tcp_respond: not mapped addr"); 884 if (bcmp(&ip->ip_dst, 885 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 886 sizeof(ip->ip_dst)) != 0) { 887 panic("tcp_respond: ip_dst != in6p_faddr"); 888 } 889 } else if (family == AF_INET6) { 890 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 891 &tp->t_in6pcb->in6p_faddr)) 892 panic("tcp_respond: ip6_dst != in6p_faddr"); 893 } else 894 panic("tcp_respond: address family mismatch"); 895 #endif 896 } 897 #endif 898 else 899 ro = NULL; 900 901 switch (family) { 902 #ifdef INET 903 case AF_INET: 904 error = ip_output(m, NULL, ro, 905 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 906 (struct ip_moptions *)0, so); 907 break; 908 #endif 909 #ifdef INET6 910 case AF_INET6: 911 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL); 912 break; 913 #endif 914 default: 915 error = EAFNOSUPPORT; 916 break; 917 } 918 919 return (error); 920 } 921 922 /* 923 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 924 * a bunch of members individually, we maintain this template for the 925 * static and mostly-static components of the TCPCB, and copy it into 926 * the new TCPCB instead. 927 */ 928 static struct tcpcb tcpcb_template = { 929 .t_srtt = TCPTV_SRTTBASE, 930 .t_rttmin = TCPTV_MIN, 931 932 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 933 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 934 .snd_numholes = 0, 935 936 .t_partialacks = -1, 937 .t_bytes_acked = 0, 938 }; 939 940 /* 941 * Updates the TCPCB template whenever a parameter that would affect 942 * the template is changed. 943 */ 944 void 945 tcp_tcpcb_template(void) 946 { 947 struct tcpcb *tp = &tcpcb_template; 948 int flags; 949 950 tp->t_peermss = tcp_mssdflt; 951 tp->t_ourmss = tcp_mssdflt; 952 tp->t_segsz = tcp_mssdflt; 953 954 flags = 0; 955 if (tcp_do_rfc1323 && tcp_do_win_scale) 956 flags |= TF_REQ_SCALE; 957 if (tcp_do_rfc1323 && tcp_do_timestamps) 958 flags |= TF_REQ_TSTMP; 959 tp->t_flags = flags; 960 961 /* 962 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 963 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 964 * reasonable initial retransmit time. 965 */ 966 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 967 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 968 TCPTV_MIN, TCPTV_REXMTMAX); 969 970 /* Keep Alive */ 971 tp->t_keepinit = tcp_keepinit; 972 tp->t_keepidle = tcp_keepidle; 973 tp->t_keepintvl = tcp_keepintvl; 974 tp->t_keepcnt = tcp_keepcnt; 975 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 976 } 977 978 /* 979 * Create a new TCP control block, making an 980 * empty reassembly queue and hooking it to the argument 981 * protocol control block. 982 */ 983 /* family selects inpcb, or in6pcb */ 984 struct tcpcb * 985 tcp_newtcpcb(int family, void *aux) 986 { 987 #ifdef INET6 988 struct rtentry *rt; 989 #endif 990 struct tcpcb *tp; 991 int i; 992 993 /* XXX Consider using a pool_cache for speed. */ 994 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 995 if (tp == NULL) 996 return (NULL); 997 memcpy(tp, &tcpcb_template, sizeof(*tp)); 998 TAILQ_INIT(&tp->segq); 999 TAILQ_INIT(&tp->timeq); 1000 tp->t_family = family; /* may be overridden later on */ 1001 TAILQ_INIT(&tp->snd_holes); 1002 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1003 1004 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1005 for (i = 0; i < TCPT_NTIMERS; i++) { 1006 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 1007 TCP_TIMER_INIT(tp, i); 1008 } 1009 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 1010 1011 switch (family) { 1012 case AF_INET: 1013 { 1014 struct inpcb *inp = (struct inpcb *)aux; 1015 1016 inp->inp_ip.ip_ttl = ip_defttl; 1017 inp->inp_ppcb = (void *)tp; 1018 1019 tp->t_inpcb = inp; 1020 tp->t_mtudisc = ip_mtudisc; 1021 break; 1022 } 1023 #ifdef INET6 1024 case AF_INET6: 1025 { 1026 struct in6pcb *in6p = (struct in6pcb *)aux; 1027 1028 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1029 (rt = rtcache_validate(&in6p->in6p_route)) != NULL 1030 ? rt->rt_ifp 1031 : NULL); 1032 in6p->in6p_ppcb = (void *)tp; 1033 1034 tp->t_in6pcb = in6p; 1035 /* for IPv6, always try to run path MTU discovery */ 1036 tp->t_mtudisc = 1; 1037 break; 1038 } 1039 #endif /* INET6 */ 1040 default: 1041 for (i = 0; i < TCPT_NTIMERS; i++) 1042 callout_destroy(&tp->t_timer[i]); 1043 callout_destroy(&tp->t_delack_ch); 1044 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1045 return (NULL); 1046 } 1047 1048 /* 1049 * Initialize our timebase. When we send timestamps, we take 1050 * the delta from tcp_now -- this means each connection always 1051 * gets a timebase of 1, which makes it, among other things, 1052 * more difficult to determine how long a system has been up, 1053 * and thus how many TCP sequence increments have occurred. 1054 * 1055 * We start with 1, because 0 doesn't work with linux, which 1056 * considers timestamp 0 in a SYN packet as a bug and disables 1057 * timestamps. 1058 */ 1059 tp->ts_timebase = tcp_now - 1; 1060 1061 tcp_congctl_select(tp, tcp_congctl_global_name); 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 TCP_STATINC(TCP_STAT_DROPS); 1095 } else 1096 TCP_STATINC(TCP_STAT_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 * Close a TCP control block: 1105 * discard all space held by the tcp 1106 * discard internet protocol block 1107 * wake up any sleepers 1108 */ 1109 struct tcpcb * 1110 tcp_close(struct tcpcb *tp) 1111 { 1112 struct inpcb *inp; 1113 #ifdef INET6 1114 struct in6pcb *in6p; 1115 #endif 1116 struct socket *so; 1117 #ifdef RTV_RTT 1118 struct rtentry *rt; 1119 #endif 1120 struct route *ro; 1121 int j; 1122 1123 inp = tp->t_inpcb; 1124 #ifdef INET6 1125 in6p = tp->t_in6pcb; 1126 #endif 1127 so = NULL; 1128 ro = NULL; 1129 if (inp) { 1130 so = inp->inp_socket; 1131 ro = &inp->inp_route; 1132 } 1133 #ifdef INET6 1134 else if (in6p) { 1135 so = in6p->in6p_socket; 1136 ro = (struct route *)&in6p->in6p_route; 1137 } 1138 #endif 1139 1140 #ifdef RTV_RTT 1141 /* 1142 * If we sent enough data to get some meaningful characteristics, 1143 * save them in the routing entry. 'Enough' is arbitrarily 1144 * defined as the sendpipesize (default 4K) * 16. This would 1145 * give us 16 rtt samples assuming we only get one sample per 1146 * window (the usual case on a long haul net). 16 samples is 1147 * enough for the srtt filter to converge to within 5% of the correct 1148 * value; fewer samples and we could save a very bogus rtt. 1149 * 1150 * Don't update the default route's characteristics and don't 1151 * update anything that the user "locked". 1152 */ 1153 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1154 ro && (rt = rtcache_validate(ro)) != NULL && 1155 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1156 u_long i = 0; 1157 1158 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1159 i = tp->t_srtt * 1160 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1161 if (rt->rt_rmx.rmx_rtt && i) 1162 /* 1163 * filter this update to half the old & half 1164 * the new values, converting scale. 1165 * See route.h and tcp_var.h for a 1166 * description of the scaling constants. 1167 */ 1168 rt->rt_rmx.rmx_rtt = 1169 (rt->rt_rmx.rmx_rtt + i) / 2; 1170 else 1171 rt->rt_rmx.rmx_rtt = i; 1172 } 1173 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1174 i = tp->t_rttvar * 1175 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1176 if (rt->rt_rmx.rmx_rttvar && i) 1177 rt->rt_rmx.rmx_rttvar = 1178 (rt->rt_rmx.rmx_rttvar + i) / 2; 1179 else 1180 rt->rt_rmx.rmx_rttvar = i; 1181 } 1182 /* 1183 * update the pipelimit (ssthresh) if it has been updated 1184 * already or if a pipesize was specified & the threshhold 1185 * got below half the pipesize. I.e., wait for bad news 1186 * before we start updating, then update on both good 1187 * and bad news. 1188 */ 1189 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1190 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1191 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1192 /* 1193 * convert the limit from user data bytes to 1194 * packets then to packet data bytes. 1195 */ 1196 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1197 if (i < 2) 1198 i = 2; 1199 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1200 if (rt->rt_rmx.rmx_ssthresh) 1201 rt->rt_rmx.rmx_ssthresh = 1202 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1203 else 1204 rt->rt_rmx.rmx_ssthresh = i; 1205 } 1206 } 1207 #endif /* RTV_RTT */ 1208 /* free the reassembly queue, if any */ 1209 TCP_REASS_LOCK(tp); 1210 (void) tcp_freeq(tp); 1211 TCP_REASS_UNLOCK(tp); 1212 1213 /* free the SACK holes list. */ 1214 tcp_free_sackholes(tp); 1215 1216 tcp_congctl_release(tp); 1217 1218 tcp_canceltimers(tp); 1219 TCP_CLEAR_DELACK(tp); 1220 syn_cache_cleanup(tp); 1221 1222 if (tp->t_template) { 1223 m_free(tp->t_template); 1224 tp->t_template = NULL; 1225 } 1226 tp->t_flags |= TF_DEAD; 1227 for (j = 0; j < TCPT_NTIMERS; j++) { 1228 callout_halt(&tp->t_timer[j], softnet_lock); 1229 callout_destroy(&tp->t_timer[j]); 1230 } 1231 callout_halt(&tp->t_delack_ch, softnet_lock); 1232 callout_destroy(&tp->t_delack_ch); 1233 pool_put(&tcpcb_pool, tp); 1234 1235 if (inp) { 1236 inp->inp_ppcb = 0; 1237 soisdisconnected(so); 1238 in_pcbdetach(inp); 1239 } 1240 #ifdef INET6 1241 else if (in6p) { 1242 in6p->in6p_ppcb = 0; 1243 soisdisconnected(so); 1244 in6_pcbdetach(in6p); 1245 } 1246 #endif 1247 TCP_STATINC(TCP_STAT_CLOSED); 1248 return ((struct tcpcb *)0); 1249 } 1250 1251 int 1252 tcp_freeq(struct tcpcb *tp) 1253 { 1254 struct ipqent *qe; 1255 int rv = 0; 1256 #ifdef TCPREASS_DEBUG 1257 int i = 0; 1258 #endif 1259 1260 TCP_REASS_LOCK_CHECK(tp); 1261 1262 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1263 #ifdef TCPREASS_DEBUG 1264 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1265 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1266 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1267 #endif 1268 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1269 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1270 m_freem(qe->ipqe_m); 1271 tcpipqent_free(qe); 1272 rv = 1; 1273 } 1274 tp->t_segqlen = 0; 1275 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1276 return (rv); 1277 } 1278 1279 /* 1280 * Protocol drain routine. Called when memory is in short supply. 1281 * Don't acquire softnet_lock as can be called from hardware 1282 * interrupt handler. 1283 */ 1284 void 1285 tcp_drain(void) 1286 { 1287 struct inpcb_hdr *inph; 1288 struct tcpcb *tp; 1289 1290 KERNEL_LOCK(1, NULL); 1291 1292 /* 1293 * Free the sequence queue of all TCP connections. 1294 */ 1295 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1296 switch (inph->inph_af) { 1297 case AF_INET: 1298 tp = intotcpcb((struct inpcb *)inph); 1299 break; 1300 #ifdef INET6 1301 case AF_INET6: 1302 tp = in6totcpcb((struct in6pcb *)inph); 1303 break; 1304 #endif 1305 default: 1306 tp = NULL; 1307 break; 1308 } 1309 if (tp != NULL) { 1310 /* 1311 * We may be called from a device's interrupt 1312 * context. If the tcpcb is already busy, 1313 * just bail out now. 1314 */ 1315 if (tcp_reass_lock_try(tp) == 0) 1316 continue; 1317 if (tcp_freeq(tp)) 1318 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1319 TCP_REASS_UNLOCK(tp); 1320 } 1321 } 1322 1323 KERNEL_UNLOCK_ONE(NULL); 1324 } 1325 1326 /* 1327 * Notify a tcp user of an asynchronous error; 1328 * store error as soft error, but wake up user 1329 * (for now, won't do anything until can select for soft error). 1330 */ 1331 void 1332 tcp_notify(struct inpcb *inp, int error) 1333 { 1334 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1335 struct socket *so = inp->inp_socket; 1336 1337 /* 1338 * Ignore some errors if we are hooked up. 1339 * If connection hasn't completed, has retransmitted several times, 1340 * and receives a second error, give up now. This is better 1341 * than waiting a long time to establish a connection that 1342 * can never complete. 1343 */ 1344 if (tp->t_state == TCPS_ESTABLISHED && 1345 (error == EHOSTUNREACH || error == ENETUNREACH || 1346 error == EHOSTDOWN)) { 1347 return; 1348 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1349 tp->t_rxtshift > 3 && tp->t_softerror) 1350 so->so_error = error; 1351 else 1352 tp->t_softerror = error; 1353 cv_broadcast(&so->so_cv); 1354 sorwakeup(so); 1355 sowwakeup(so); 1356 } 1357 1358 #ifdef INET6 1359 void 1360 tcp6_notify(struct in6pcb *in6p, int error) 1361 { 1362 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1363 struct socket *so = in6p->in6p_socket; 1364 1365 /* 1366 * Ignore some errors if we are hooked up. 1367 * If connection hasn't completed, has retransmitted several times, 1368 * and receives a second error, give up now. This is better 1369 * than waiting a long time to establish a connection that 1370 * can never complete. 1371 */ 1372 if (tp->t_state == TCPS_ESTABLISHED && 1373 (error == EHOSTUNREACH || error == ENETUNREACH || 1374 error == EHOSTDOWN)) { 1375 return; 1376 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1377 tp->t_rxtshift > 3 && tp->t_softerror) 1378 so->so_error = error; 1379 else 1380 tp->t_softerror = error; 1381 cv_broadcast(&so->so_cv); 1382 sorwakeup(so); 1383 sowwakeup(so); 1384 } 1385 #endif 1386 1387 #ifdef INET6 1388 void * 1389 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1390 { 1391 struct tcphdr th; 1392 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1393 int nmatch; 1394 struct ip6_hdr *ip6; 1395 const struct sockaddr_in6 *sa6_src = NULL; 1396 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1397 struct mbuf *m; 1398 int off; 1399 1400 if (sa->sa_family != AF_INET6 || 1401 sa->sa_len != sizeof(struct sockaddr_in6)) 1402 return NULL; 1403 if ((unsigned)cmd >= PRC_NCMDS) 1404 return NULL; 1405 else if (cmd == PRC_QUENCH) { 1406 /* 1407 * Don't honor ICMP Source Quench messages meant for 1408 * TCP connections. 1409 */ 1410 return NULL; 1411 } else if (PRC_IS_REDIRECT(cmd)) 1412 notify = in6_rtchange, d = NULL; 1413 else if (cmd == PRC_MSGSIZE) 1414 ; /* special code is present, see below */ 1415 else if (cmd == PRC_HOSTDEAD) 1416 d = NULL; 1417 else if (inet6ctlerrmap[cmd] == 0) 1418 return NULL; 1419 1420 /* if the parameter is from icmp6, decode it. */ 1421 if (d != NULL) { 1422 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1423 m = ip6cp->ip6c_m; 1424 ip6 = ip6cp->ip6c_ip6; 1425 off = ip6cp->ip6c_off; 1426 sa6_src = ip6cp->ip6c_src; 1427 } else { 1428 m = NULL; 1429 ip6 = NULL; 1430 sa6_src = &sa6_any; 1431 off = 0; 1432 } 1433 1434 if (ip6) { 1435 /* 1436 * XXX: We assume that when ip6 is non NULL, 1437 * M and OFF are valid. 1438 */ 1439 1440 /* check if we can safely examine src and dst ports */ 1441 if (m->m_pkthdr.len < off + sizeof(th)) { 1442 if (cmd == PRC_MSGSIZE) 1443 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1444 return NULL; 1445 } 1446 1447 bzero(&th, sizeof(th)); 1448 m_copydata(m, off, sizeof(th), (void *)&th); 1449 1450 if (cmd == PRC_MSGSIZE) { 1451 int valid = 0; 1452 1453 /* 1454 * Check to see if we have a valid TCP connection 1455 * corresponding to the address in the ICMPv6 message 1456 * payload. 1457 */ 1458 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1459 th.th_dport, 1460 (const struct in6_addr *)&sa6_src->sin6_addr, 1461 th.th_sport, 0)) 1462 valid++; 1463 1464 /* 1465 * Depending on the value of "valid" and routing table 1466 * size (mtudisc_{hi,lo}wat), we will: 1467 * - recalcurate the new MTU and create the 1468 * corresponding routing entry, or 1469 * - ignore the MTU change notification. 1470 */ 1471 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1472 1473 /* 1474 * no need to call in6_pcbnotify, it should have been 1475 * called via callback if necessary 1476 */ 1477 return NULL; 1478 } 1479 1480 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1481 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1482 if (nmatch == 0 && syn_cache_count && 1483 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1484 inet6ctlerrmap[cmd] == ENETUNREACH || 1485 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1486 syn_cache_unreach((const struct sockaddr *)sa6_src, 1487 sa, &th); 1488 } else { 1489 (void) in6_pcbnotify(&tcbtable, sa, 0, 1490 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1491 } 1492 1493 return NULL; 1494 } 1495 #endif 1496 1497 #ifdef INET 1498 /* assumes that ip header and tcp header are contiguous on mbuf */ 1499 void * 1500 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1501 { 1502 struct ip *ip = v; 1503 struct tcphdr *th; 1504 struct icmp *icp; 1505 extern const int inetctlerrmap[]; 1506 void (*notify)(struct inpcb *, int) = tcp_notify; 1507 int errno; 1508 int nmatch; 1509 struct tcpcb *tp; 1510 u_int mtu; 1511 tcp_seq seq; 1512 struct inpcb *inp; 1513 #ifdef INET6 1514 struct in6pcb *in6p; 1515 struct in6_addr src6, dst6; 1516 #endif 1517 1518 if (sa->sa_family != AF_INET || 1519 sa->sa_len != sizeof(struct sockaddr_in)) 1520 return NULL; 1521 if ((unsigned)cmd >= PRC_NCMDS) 1522 return NULL; 1523 errno = inetctlerrmap[cmd]; 1524 if (cmd == PRC_QUENCH) 1525 /* 1526 * Don't honor ICMP Source Quench messages meant for 1527 * TCP connections. 1528 */ 1529 return NULL; 1530 else if (PRC_IS_REDIRECT(cmd)) 1531 notify = in_rtchange, ip = 0; 1532 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1533 /* 1534 * Check to see if we have a valid TCP connection 1535 * corresponding to the address in the ICMP message 1536 * payload. 1537 * 1538 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1539 */ 1540 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1541 #ifdef INET6 1542 memset(&src6, 0, sizeof(src6)); 1543 memset(&dst6, 0, sizeof(dst6)); 1544 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1545 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1546 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1547 #endif 1548 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1549 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1550 #ifdef INET6 1551 in6p = NULL; 1552 #else 1553 ; 1554 #endif 1555 #ifdef INET6 1556 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1557 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1558 ; 1559 #endif 1560 else 1561 return NULL; 1562 1563 /* 1564 * Now that we've validated that we are actually communicating 1565 * with the host indicated in the ICMP message, locate the 1566 * ICMP header, recalculate the new MTU, and create the 1567 * corresponding routing entry. 1568 */ 1569 icp = (struct icmp *)((char *)ip - 1570 offsetof(struct icmp, icmp_ip)); 1571 if (inp) { 1572 if ((tp = intotcpcb(inp)) == NULL) 1573 return NULL; 1574 } 1575 #ifdef INET6 1576 else if (in6p) { 1577 if ((tp = in6totcpcb(in6p)) == NULL) 1578 return NULL; 1579 } 1580 #endif 1581 else 1582 return NULL; 1583 seq = ntohl(th->th_seq); 1584 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1585 return NULL; 1586 /* 1587 * If the ICMP message advertises a Next-Hop MTU 1588 * equal or larger than the maximum packet size we have 1589 * ever sent, drop the message. 1590 */ 1591 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1592 if (mtu >= tp->t_pmtud_mtu_sent) 1593 return NULL; 1594 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1595 /* 1596 * Calculate new MTU, and create corresponding 1597 * route (traditional PMTUD). 1598 */ 1599 tp->t_flags &= ~TF_PMTUD_PEND; 1600 icmp_mtudisc(icp, ip->ip_dst); 1601 } else { 1602 /* 1603 * Record the information got in the ICMP 1604 * message; act on it later. 1605 * If we had already recorded an ICMP message, 1606 * replace the old one only if the new message 1607 * refers to an older TCP segment 1608 */ 1609 if (tp->t_flags & TF_PMTUD_PEND) { 1610 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1611 return NULL; 1612 } else 1613 tp->t_flags |= TF_PMTUD_PEND; 1614 tp->t_pmtud_th_seq = seq; 1615 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1616 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1617 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1618 } 1619 return NULL; 1620 } else if (cmd == PRC_HOSTDEAD) 1621 ip = 0; 1622 else if (errno == 0) 1623 return NULL; 1624 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1625 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1626 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1627 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1628 if (nmatch == 0 && syn_cache_count && 1629 (inetctlerrmap[cmd] == EHOSTUNREACH || 1630 inetctlerrmap[cmd] == ENETUNREACH || 1631 inetctlerrmap[cmd] == EHOSTDOWN)) { 1632 struct sockaddr_in sin; 1633 bzero(&sin, sizeof(sin)); 1634 sin.sin_len = sizeof(sin); 1635 sin.sin_family = AF_INET; 1636 sin.sin_port = th->th_sport; 1637 sin.sin_addr = ip->ip_src; 1638 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1639 } 1640 1641 /* XXX mapped address case */ 1642 } else 1643 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1644 notify); 1645 return NULL; 1646 } 1647 1648 /* 1649 * When a source quench is received, we are being notified of congestion. 1650 * Close the congestion window down to the Loss Window (one segment). 1651 * We will gradually open it again as we proceed. 1652 */ 1653 void 1654 tcp_quench(struct inpcb *inp, int errno) 1655 { 1656 struct tcpcb *tp = intotcpcb(inp); 1657 1658 if (tp) { 1659 tp->snd_cwnd = tp->t_segsz; 1660 tp->t_bytes_acked = 0; 1661 } 1662 } 1663 #endif 1664 1665 #ifdef INET6 1666 void 1667 tcp6_quench(struct in6pcb *in6p, int errno) 1668 { 1669 struct tcpcb *tp = in6totcpcb(in6p); 1670 1671 if (tp) { 1672 tp->snd_cwnd = tp->t_segsz; 1673 tp->t_bytes_acked = 0; 1674 } 1675 } 1676 #endif 1677 1678 #ifdef INET 1679 /* 1680 * Path MTU Discovery handlers. 1681 */ 1682 void 1683 tcp_mtudisc_callback(struct in_addr faddr) 1684 { 1685 #ifdef INET6 1686 struct in6_addr in6; 1687 #endif 1688 1689 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1690 #ifdef INET6 1691 memset(&in6, 0, sizeof(in6)); 1692 in6.s6_addr16[5] = 0xffff; 1693 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1694 tcp6_mtudisc_callback(&in6); 1695 #endif 1696 } 1697 1698 /* 1699 * On receipt of path MTU corrections, flush old route and replace it 1700 * with the new one. Retransmit all unacknowledged packets, to ensure 1701 * that all packets will be received. 1702 */ 1703 void 1704 tcp_mtudisc(struct inpcb *inp, int errno) 1705 { 1706 struct tcpcb *tp = intotcpcb(inp); 1707 struct rtentry *rt = in_pcbrtentry(inp); 1708 1709 if (tp != 0) { 1710 if (rt != 0) { 1711 /* 1712 * If this was not a host route, remove and realloc. 1713 */ 1714 if ((rt->rt_flags & RTF_HOST) == 0) { 1715 in_rtchange(inp, errno); 1716 if ((rt = in_pcbrtentry(inp)) == 0) 1717 return; 1718 } 1719 1720 /* 1721 * Slow start out of the error condition. We 1722 * use the MTU because we know it's smaller 1723 * than the previously transmitted segment. 1724 * 1725 * Note: This is more conservative than the 1726 * suggestion in draft-floyd-incr-init-win-03. 1727 */ 1728 if (rt->rt_rmx.rmx_mtu != 0) 1729 tp->snd_cwnd = 1730 TCP_INITIAL_WINDOW(tcp_init_win, 1731 rt->rt_rmx.rmx_mtu); 1732 } 1733 1734 /* 1735 * Resend unacknowledged packets. 1736 */ 1737 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1738 tcp_output(tp); 1739 } 1740 } 1741 #endif 1742 1743 #ifdef INET6 1744 /* 1745 * Path MTU Discovery handlers. 1746 */ 1747 void 1748 tcp6_mtudisc_callback(struct in6_addr *faddr) 1749 { 1750 struct sockaddr_in6 sin6; 1751 1752 bzero(&sin6, sizeof(sin6)); 1753 sin6.sin6_family = AF_INET6; 1754 sin6.sin6_len = sizeof(struct sockaddr_in6); 1755 sin6.sin6_addr = *faddr; 1756 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1757 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1758 } 1759 1760 void 1761 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1762 { 1763 struct tcpcb *tp = in6totcpcb(in6p); 1764 struct rtentry *rt = in6_pcbrtentry(in6p); 1765 1766 if (tp != 0) { 1767 if (rt != 0) { 1768 /* 1769 * If this was not a host route, remove and realloc. 1770 */ 1771 if ((rt->rt_flags & RTF_HOST) == 0) { 1772 in6_rtchange(in6p, errno); 1773 if ((rt = in6_pcbrtentry(in6p)) == 0) 1774 return; 1775 } 1776 1777 /* 1778 * Slow start out of the error condition. We 1779 * use the MTU because we know it's smaller 1780 * than the previously transmitted segment. 1781 * 1782 * Note: This is more conservative than the 1783 * suggestion in draft-floyd-incr-init-win-03. 1784 */ 1785 if (rt->rt_rmx.rmx_mtu != 0) 1786 tp->snd_cwnd = 1787 TCP_INITIAL_WINDOW(tcp_init_win, 1788 rt->rt_rmx.rmx_mtu); 1789 } 1790 1791 /* 1792 * Resend unacknowledged packets. 1793 */ 1794 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1795 tcp_output(tp); 1796 } 1797 } 1798 #endif /* INET6 */ 1799 1800 /* 1801 * Compute the MSS to advertise to the peer. Called only during 1802 * the 3-way handshake. If we are the server (peer initiated 1803 * connection), we are called with a pointer to the interface 1804 * on which the SYN packet arrived. If we are the client (we 1805 * initiated connection), we are called with a pointer to the 1806 * interface out which this connection should go. 1807 * 1808 * NOTE: Do not subtract IP option/extension header size nor IPsec 1809 * header size from MSS advertisement. MSS option must hold the maximum 1810 * segment size we can accept, so it must always be: 1811 * max(if mtu) - ip header - tcp header 1812 */ 1813 u_long 1814 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1815 { 1816 extern u_long in_maxmtu; 1817 u_long mss = 0; 1818 u_long hdrsiz; 1819 1820 /* 1821 * In order to avoid defeating path MTU discovery on the peer, 1822 * we advertise the max MTU of all attached networks as our MSS, 1823 * per RFC 1191, section 3.1. 1824 * 1825 * We provide the option to advertise just the MTU of 1826 * the interface on which we hope this connection will 1827 * be receiving. If we are responding to a SYN, we 1828 * will have a pretty good idea about this, but when 1829 * initiating a connection there is a bit more doubt. 1830 * 1831 * We also need to ensure that loopback has a large enough 1832 * MSS, as the loopback MTU is never included in in_maxmtu. 1833 */ 1834 1835 if (ifp != NULL) 1836 switch (af) { 1837 case AF_INET: 1838 mss = ifp->if_mtu; 1839 break; 1840 #ifdef INET6 1841 case AF_INET6: 1842 mss = IN6_LINKMTU(ifp); 1843 break; 1844 #endif 1845 } 1846 1847 if (tcp_mss_ifmtu == 0) 1848 switch (af) { 1849 case AF_INET: 1850 mss = max(in_maxmtu, mss); 1851 break; 1852 #ifdef INET6 1853 case AF_INET6: 1854 mss = max(in6_maxmtu, mss); 1855 break; 1856 #endif 1857 } 1858 1859 switch (af) { 1860 case AF_INET: 1861 hdrsiz = sizeof(struct ip); 1862 break; 1863 #ifdef INET6 1864 case AF_INET6: 1865 hdrsiz = sizeof(struct ip6_hdr); 1866 break; 1867 #endif 1868 default: 1869 hdrsiz = 0; 1870 break; 1871 } 1872 hdrsiz += sizeof(struct tcphdr); 1873 if (mss > hdrsiz) 1874 mss -= hdrsiz; 1875 1876 mss = max(tcp_mssdflt, mss); 1877 return (mss); 1878 } 1879 1880 /* 1881 * Set connection variables based on the peer's advertised MSS. 1882 * We are passed the TCPCB for the actual connection. If we 1883 * are the server, we are called by the compressed state engine 1884 * when the 3-way handshake is complete. If we are the client, 1885 * we are called when we receive the SYN,ACK from the server. 1886 * 1887 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1888 * before this routine is called! 1889 */ 1890 void 1891 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1892 { 1893 struct socket *so; 1894 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1895 struct rtentry *rt; 1896 #endif 1897 u_long bufsize; 1898 int mss; 1899 1900 #ifdef DIAGNOSTIC 1901 if (tp->t_inpcb && tp->t_in6pcb) 1902 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1903 #endif 1904 so = NULL; 1905 rt = NULL; 1906 #ifdef INET 1907 if (tp->t_inpcb) { 1908 so = tp->t_inpcb->inp_socket; 1909 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1910 rt = in_pcbrtentry(tp->t_inpcb); 1911 #endif 1912 } 1913 #endif 1914 #ifdef INET6 1915 if (tp->t_in6pcb) { 1916 so = tp->t_in6pcb->in6p_socket; 1917 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1918 rt = in6_pcbrtentry(tp->t_in6pcb); 1919 #endif 1920 } 1921 #endif 1922 1923 /* 1924 * As per RFC1122, use the default MSS value, unless they 1925 * sent us an offer. Do not accept offers less than 256 bytes. 1926 */ 1927 mss = tcp_mssdflt; 1928 if (offer) 1929 mss = offer; 1930 mss = max(mss, 256); /* sanity */ 1931 tp->t_peermss = mss; 1932 mss -= tcp_optlen(tp); 1933 #ifdef INET 1934 if (tp->t_inpcb) 1935 mss -= ip_optlen(tp->t_inpcb); 1936 #endif 1937 #ifdef INET6 1938 if (tp->t_in6pcb) 1939 mss -= ip6_optlen(tp->t_in6pcb); 1940 #endif 1941 1942 /* 1943 * If there's a pipesize, change the socket buffer to that size. 1944 * Make the socket buffer an integral number of MSS units. If 1945 * the MSS is larger than the socket buffer, artificially decrease 1946 * the MSS. 1947 */ 1948 #ifdef RTV_SPIPE 1949 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1950 bufsize = rt->rt_rmx.rmx_sendpipe; 1951 else 1952 #endif 1953 { 1954 KASSERT(so != NULL); 1955 bufsize = so->so_snd.sb_hiwat; 1956 } 1957 if (bufsize < mss) 1958 mss = bufsize; 1959 else { 1960 bufsize = roundup(bufsize, mss); 1961 if (bufsize > sb_max) 1962 bufsize = sb_max; 1963 (void) sbreserve(&so->so_snd, bufsize, so); 1964 } 1965 tp->t_segsz = mss; 1966 1967 #ifdef RTV_SSTHRESH 1968 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1969 /* 1970 * There's some sort of gateway or interface buffer 1971 * limit on the path. Use this to set the slow 1972 * start threshold, but set the threshold to no less 1973 * than 2 * MSS. 1974 */ 1975 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1976 } 1977 #endif 1978 } 1979 1980 /* 1981 * Processing necessary when a TCP connection is established. 1982 */ 1983 void 1984 tcp_established(struct tcpcb *tp) 1985 { 1986 struct socket *so; 1987 #ifdef RTV_RPIPE 1988 struct rtentry *rt; 1989 #endif 1990 u_long bufsize; 1991 1992 #ifdef DIAGNOSTIC 1993 if (tp->t_inpcb && tp->t_in6pcb) 1994 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 1995 #endif 1996 so = NULL; 1997 rt = NULL; 1998 #ifdef INET 1999 if (tp->t_inpcb) { 2000 so = tp->t_inpcb->inp_socket; 2001 #if defined(RTV_RPIPE) 2002 rt = in_pcbrtentry(tp->t_inpcb); 2003 #endif 2004 } 2005 #endif 2006 #ifdef INET6 2007 if (tp->t_in6pcb) { 2008 so = tp->t_in6pcb->in6p_socket; 2009 #if defined(RTV_RPIPE) 2010 rt = in6_pcbrtentry(tp->t_in6pcb); 2011 #endif 2012 } 2013 #endif 2014 2015 tp->t_state = TCPS_ESTABLISHED; 2016 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2017 2018 #ifdef RTV_RPIPE 2019 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2020 bufsize = rt->rt_rmx.rmx_recvpipe; 2021 else 2022 #endif 2023 { 2024 KASSERT(so != NULL); 2025 bufsize = so->so_rcv.sb_hiwat; 2026 } 2027 if (bufsize > tp->t_ourmss) { 2028 bufsize = roundup(bufsize, tp->t_ourmss); 2029 if (bufsize > sb_max) 2030 bufsize = sb_max; 2031 (void) sbreserve(&so->so_rcv, bufsize, so); 2032 } 2033 } 2034 2035 /* 2036 * Check if there's an initial rtt or rttvar. Convert from the 2037 * route-table units to scaled multiples of the slow timeout timer. 2038 * Called only during the 3-way handshake. 2039 */ 2040 void 2041 tcp_rmx_rtt(struct tcpcb *tp) 2042 { 2043 #ifdef RTV_RTT 2044 struct rtentry *rt = NULL; 2045 int rtt; 2046 2047 #ifdef DIAGNOSTIC 2048 if (tp->t_inpcb && tp->t_in6pcb) 2049 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2050 #endif 2051 #ifdef INET 2052 if (tp->t_inpcb) 2053 rt = in_pcbrtentry(tp->t_inpcb); 2054 #endif 2055 #ifdef INET6 2056 if (tp->t_in6pcb) 2057 rt = in6_pcbrtentry(tp->t_in6pcb); 2058 #endif 2059 if (rt == NULL) 2060 return; 2061 2062 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2063 /* 2064 * XXX The lock bit for MTU indicates that the value 2065 * is also a minimum value; this is subject to time. 2066 */ 2067 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2068 TCPT_RANGESET(tp->t_rttmin, 2069 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2070 TCPTV_MIN, TCPTV_REXMTMAX); 2071 tp->t_srtt = rtt / 2072 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2073 if (rt->rt_rmx.rmx_rttvar) { 2074 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2075 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2076 (TCP_RTTVAR_SHIFT + 2)); 2077 } else { 2078 /* Default variation is +- 1 rtt */ 2079 tp->t_rttvar = 2080 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2081 } 2082 TCPT_RANGESET(tp->t_rxtcur, 2083 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2084 tp->t_rttmin, TCPTV_REXMTMAX); 2085 } 2086 #endif 2087 } 2088 2089 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2090 #if NRND > 0 2091 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2092 #endif 2093 2094 /* 2095 * Get a new sequence value given a tcp control block 2096 */ 2097 tcp_seq 2098 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2099 { 2100 2101 #ifdef INET 2102 if (tp->t_inpcb != NULL) { 2103 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2104 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2105 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2106 addin)); 2107 } 2108 #endif 2109 #ifdef INET6 2110 if (tp->t_in6pcb != NULL) { 2111 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2112 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2113 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2114 addin)); 2115 } 2116 #endif 2117 /* Not possible. */ 2118 panic("tcp_new_iss"); 2119 } 2120 2121 /* 2122 * This routine actually generates a new TCP initial sequence number. 2123 */ 2124 tcp_seq 2125 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2126 size_t addrsz, tcp_seq addin) 2127 { 2128 tcp_seq tcp_iss; 2129 2130 #if NRND > 0 2131 static bool tcp_iss_gotten_secret; 2132 2133 /* 2134 * If we haven't been here before, initialize our cryptographic 2135 * hash secret. 2136 */ 2137 if (tcp_iss_gotten_secret == false) { 2138 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2139 RND_EXTRACT_ANY); 2140 tcp_iss_gotten_secret = true; 2141 } 2142 2143 if (tcp_do_rfc1948) { 2144 MD5_CTX ctx; 2145 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2146 2147 /* 2148 * Compute the base value of the ISS. It is a hash 2149 * of (saddr, sport, daddr, dport, secret). 2150 */ 2151 MD5Init(&ctx); 2152 2153 MD5Update(&ctx, (u_char *) laddr, addrsz); 2154 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2155 2156 MD5Update(&ctx, (u_char *) faddr, addrsz); 2157 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2158 2159 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2160 2161 MD5Final(hash, &ctx); 2162 2163 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2164 2165 /* 2166 * Now increment our "timer", and add it in to 2167 * the computed value. 2168 * 2169 * XXX Use `addin'? 2170 * XXX TCP_ISSINCR too large to use? 2171 */ 2172 tcp_iss_seq += TCP_ISSINCR; 2173 #ifdef TCPISS_DEBUG 2174 printf("ISS hash 0x%08x, ", tcp_iss); 2175 #endif 2176 tcp_iss += tcp_iss_seq + addin; 2177 #ifdef TCPISS_DEBUG 2178 printf("new ISS 0x%08x\n", tcp_iss); 2179 #endif 2180 } else 2181 #endif /* NRND > 0 */ 2182 { 2183 /* 2184 * Randomize. 2185 */ 2186 #if NRND > 0 2187 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2188 #else 2189 tcp_iss = arc4random(); 2190 #endif 2191 2192 /* 2193 * If we were asked to add some amount to a known value, 2194 * we will take a random value obtained above, mask off 2195 * the upper bits, and add in the known value. We also 2196 * add in a constant to ensure that we are at least a 2197 * certain distance from the original value. 2198 * 2199 * This is used when an old connection is in timed wait 2200 * and we have a new one coming in, for instance. 2201 */ 2202 if (addin != 0) { 2203 #ifdef TCPISS_DEBUG 2204 printf("Random %08x, ", tcp_iss); 2205 #endif 2206 tcp_iss &= TCP_ISS_RANDOM_MASK; 2207 tcp_iss += addin + TCP_ISSINCR; 2208 #ifdef TCPISS_DEBUG 2209 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2210 #endif 2211 } else { 2212 tcp_iss &= TCP_ISS_RANDOM_MASK; 2213 tcp_iss += tcp_iss_seq; 2214 tcp_iss_seq += TCP_ISSINCR; 2215 #ifdef TCPISS_DEBUG 2216 printf("ISS %08x\n", tcp_iss); 2217 #endif 2218 } 2219 } 2220 2221 if (tcp_compat_42) { 2222 /* 2223 * Limit it to the positive range for really old TCP 2224 * implementations. 2225 * Just AND off the top bit instead of checking if 2226 * is set first - saves a branch 50% of the time. 2227 */ 2228 tcp_iss &= 0x7fffffff; /* XXX */ 2229 } 2230 2231 return (tcp_iss); 2232 } 2233 2234 #if defined(IPSEC) || defined(FAST_IPSEC) 2235 /* compute ESP/AH header size for TCP, including outer IP header. */ 2236 size_t 2237 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2238 { 2239 struct inpcb *inp; 2240 size_t hdrsiz; 2241 2242 /* XXX mapped addr case (tp->t_in6pcb) */ 2243 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2244 return 0; 2245 switch (tp->t_family) { 2246 case AF_INET: 2247 /* XXX: should use currect direction. */ 2248 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2249 break; 2250 default: 2251 hdrsiz = 0; 2252 break; 2253 } 2254 2255 return hdrsiz; 2256 } 2257 2258 #ifdef INET6 2259 size_t 2260 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2261 { 2262 struct in6pcb *in6p; 2263 size_t hdrsiz; 2264 2265 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2266 return 0; 2267 switch (tp->t_family) { 2268 case AF_INET6: 2269 /* XXX: should use currect direction. */ 2270 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2271 break; 2272 case AF_INET: 2273 /* mapped address case - tricky */ 2274 default: 2275 hdrsiz = 0; 2276 break; 2277 } 2278 2279 return hdrsiz; 2280 } 2281 #endif 2282 #endif /*IPSEC*/ 2283 2284 /* 2285 * Determine the length of the TCP options for this connection. 2286 * 2287 * XXX: What do we do for SACK, when we add that? Just reserve 2288 * all of the space? Otherwise we can't exactly be incrementing 2289 * cwnd by an amount that varies depending on the amount we last 2290 * had to SACK! 2291 */ 2292 2293 u_int 2294 tcp_optlen(struct tcpcb *tp) 2295 { 2296 u_int optlen; 2297 2298 optlen = 0; 2299 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2300 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2301 optlen += TCPOLEN_TSTAMP_APPA; 2302 2303 #ifdef TCP_SIGNATURE 2304 if (tp->t_flags & TF_SIGNATURE) 2305 optlen += TCPOLEN_SIGNATURE + 2; 2306 #endif /* TCP_SIGNATURE */ 2307 2308 return optlen; 2309 } 2310 2311 u_int 2312 tcp_hdrsz(struct tcpcb *tp) 2313 { 2314 u_int hlen; 2315 2316 switch (tp->t_family) { 2317 #ifdef INET6 2318 case AF_INET6: 2319 hlen = sizeof(struct ip6_hdr); 2320 break; 2321 #endif 2322 case AF_INET: 2323 hlen = sizeof(struct ip); 2324 break; 2325 default: 2326 hlen = 0; 2327 break; 2328 } 2329 hlen += sizeof(struct tcphdr); 2330 2331 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2332 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2333 hlen += TCPOLEN_TSTAMP_APPA; 2334 #ifdef TCP_SIGNATURE 2335 if (tp->t_flags & TF_SIGNATURE) 2336 hlen += TCPOLEN_SIGLEN; 2337 #endif 2338 return hlen; 2339 } 2340 2341 void 2342 tcp_statinc(u_int stat) 2343 { 2344 2345 KASSERT(stat < TCP_NSTATS); 2346 TCP_STATINC(stat); 2347 } 2348 2349 void 2350 tcp_statadd(u_int stat, uint64_t val) 2351 { 2352 2353 KASSERT(stat < TCP_NSTATS); 2354 TCP_STATADD(stat, val); 2355 } 2356
Indexes created Tue Sep 23 13:09:54 GMT 2025