1 /* $NetBSD: tcp_subr.c,v 1.270 2017/03/03 07:13:06 ozaki-r 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 1.270 2017/03/03 07:13:06 ozaki-r Exp $"); 95 96 #ifdef _KERNEL_OPT 97 #include "opt_inet.h" 98 #include "opt_ipsec.h" 99 #include "opt_tcp_compat_42.h" 100 #include "opt_inet_csum.h" 101 #include "opt_mbuftrace.h" 102 #endif 103 104 #include <sys/param.h> 105 #include <sys/atomic.h> 106 #include <sys/proc.h> 107 #include <sys/systm.h> 108 #include <sys/mbuf.h> 109 #include <sys/once.h> 110 #include <sys/socket.h> 111 #include <sys/socketvar.h> 112 #include <sys/protosw.h> 113 #include <sys/errno.h> 114 #include <sys/kernel.h> 115 #include <sys/pool.h> 116 #include <sys/md5.h> 117 #include <sys/cprng.h> 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_vtw.h> 148 #include <netinet/tcp_private.h> 149 #include <netinet/tcp_congctl.h> 150 #include <netinet/tcpip.h> 151 152 #ifdef IPSEC 153 #include <netipsec/ipsec.h> 154 #include <netipsec/xform.h> 155 #ifdef INET6 156 #include <netipsec/ipsec6.h> 157 #endif 158 #include <netipsec/key.h> 159 #endif /* IPSEC*/ 160 161 162 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 163 u_int32_t tcp_now; /* slow ticks, for RFC 1323 timestamps */ 164 165 percpu_t *tcpstat_percpu; 166 167 /* patchable/settable parameters for tcp */ 168 int tcp_mssdflt = TCP_MSS; 169 int tcp_minmss = TCP_MINMSS; 170 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 171 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 172 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 173 int tcp_do_sack = 1; /* selective acknowledgement */ 174 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 175 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 176 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 177 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 178 #ifndef TCP_INIT_WIN 179 #define TCP_INIT_WIN 4 /* initial slow start window */ 180 #endif 181 #ifndef TCP_INIT_WIN_LOCAL 182 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 183 #endif 184 /* 185 * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460. 186 * This is to simulate current behavior for iw == 4 187 */ 188 int tcp_init_win_max[] = { 189 1 * 1460, 190 1 * 1460, 191 2 * 1460, 192 2 * 1460, 193 3 * 1460, 194 5 * 1460, 195 6 * 1460, 196 7 * 1460, 197 8 * 1460, 198 9 * 1460, 199 10 * 1460 200 }; 201 int tcp_init_win = TCP_INIT_WIN; 202 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 203 int tcp_mss_ifmtu = 0; 204 #ifdef TCP_COMPAT_42 205 int tcp_compat_42 = 1; 206 #else 207 int tcp_compat_42 = 0; 208 #endif 209 int tcp_rst_ppslim = 100; /* 100pps */ 210 int tcp_ackdrop_ppslim = 100; /* 100pps */ 211 int tcp_do_loopback_cksum = 0; 212 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 213 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 214 int tcp_sack_tp_maxholes = 32; 215 int tcp_sack_globalmaxholes = 1024; 216 int tcp_sack_globalholes = 0; 217 int tcp_ecn_maxretries = 1; 218 int tcp_msl_enable = 1; /* enable TIME_WAIT truncation */ 219 int tcp_msl_loop = PR_SLOWHZ; /* MSL for loopback */ 220 int tcp_msl_local = 5 * PR_SLOWHZ; /* MSL for 'local' */ 221 int tcp_msl_remote = TCPTV_MSL; /* MSL otherwise */ 222 int tcp_msl_remote_threshold = TCPTV_SRTTDFLT; /* RTT threshold */ 223 int tcp_rttlocal = 0; /* Use RTT to decide who's 'local' */ 224 225 int tcp4_vtw_enable = 0; /* 1 to enable */ 226 int tcp6_vtw_enable = 0; /* 1 to enable */ 227 int tcp_vtw_was_enabled = 0; 228 int tcp_vtw_entries = 1 << 4; /* 16 vestigial TIME_WAIT entries */ 229 230 /* tcb hash */ 231 #ifndef TCBHASHSIZE 232 #define TCBHASHSIZE 128 233 #endif 234 int tcbhashsize = TCBHASHSIZE; 235 236 /* syn hash parameters */ 237 #define TCP_SYN_HASH_SIZE 293 238 #define TCP_SYN_BUCKET_SIZE 35 239 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 240 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 241 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 242 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 243 244 int tcp_freeq(struct tcpcb *); 245 static int tcp_iss_secret_init(void); 246 247 #ifdef INET 248 static void tcp_mtudisc_callback(struct in_addr); 249 #endif 250 251 #ifdef INET6 252 void tcp6_mtudisc(struct in6pcb *, int); 253 #endif 254 255 static struct pool tcpcb_pool; 256 257 static int tcp_drainwanted; 258 259 #ifdef TCP_CSUM_COUNTERS 260 #include <sys/device.h> 261 262 #if defined(INET) 263 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp", "hwcsum bad"); 265 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 266 NULL, "tcp", "hwcsum ok"); 267 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 268 NULL, "tcp", "hwcsum data"); 269 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 270 NULL, "tcp", "swcsum"); 271 272 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 273 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 274 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 275 EVCNT_ATTACH_STATIC(tcp_swcsum); 276 #endif /* defined(INET) */ 277 278 #if defined(INET6) 279 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 NULL, "tcp6", "hwcsum bad"); 281 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 NULL, "tcp6", "hwcsum ok"); 283 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp6", "hwcsum data"); 285 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp6", "swcsum"); 287 288 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 289 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 290 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 291 EVCNT_ATTACH_STATIC(tcp6_swcsum); 292 #endif /* defined(INET6) */ 293 #endif /* TCP_CSUM_COUNTERS */ 294 295 296 #ifdef TCP_OUTPUT_COUNTERS 297 #include <sys/device.h> 298 299 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 300 NULL, "tcp", "output big header"); 301 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 302 NULL, "tcp", "output predict hit"); 303 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 304 NULL, "tcp", "output predict miss"); 305 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 306 NULL, "tcp", "output copy small"); 307 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 308 NULL, "tcp", "output copy big"); 309 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 310 NULL, "tcp", "output reference big"); 311 312 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 313 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 314 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 315 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 316 EVCNT_ATTACH_STATIC(tcp_output_copybig); 317 EVCNT_ATTACH_STATIC(tcp_output_refbig); 318 319 #endif /* TCP_OUTPUT_COUNTERS */ 320 321 #ifdef TCP_REASS_COUNTERS 322 #include <sys/device.h> 323 324 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 NULL, "tcp_reass", "calls"); 326 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "insert into empty queue"); 328 struct evcnt tcp_reass_iteration[8] = { 329 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 331 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 332 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 333 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 335 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 336 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 337 }; 338 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 339 &tcp_reass_, "tcp_reass", "prepend to first"); 340 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 341 &tcp_reass_, "tcp_reass", "prepend"); 342 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 343 &tcp_reass_, "tcp_reass", "insert"); 344 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 345 &tcp_reass_, "tcp_reass", "insert at tail"); 346 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 347 &tcp_reass_, "tcp_reass", "append"); 348 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 349 &tcp_reass_, "tcp_reass", "append to tail fragment"); 350 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 351 &tcp_reass_, "tcp_reass", "overlap at end"); 352 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 353 &tcp_reass_, "tcp_reass", "overlap at start"); 354 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 355 &tcp_reass_, "tcp_reass", "duplicate segment"); 356 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 357 &tcp_reass_, "tcp_reass", "duplicate fragment"); 358 359 EVCNT_ATTACH_STATIC(tcp_reass_); 360 EVCNT_ATTACH_STATIC(tcp_reass_empty); 361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 363 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 364 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 365 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 366 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 367 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 368 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 369 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 370 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 371 EVCNT_ATTACH_STATIC(tcp_reass_insert); 372 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 373 EVCNT_ATTACH_STATIC(tcp_reass_append); 374 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 375 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 376 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 377 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 378 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 379 380 #endif /* TCP_REASS_COUNTERS */ 381 382 #ifdef MBUFTRACE 383 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 384 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 385 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 386 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 387 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 388 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 389 #endif 390 391 callout_t tcp_slowtimo_ch; 392 393 static int 394 do_tcpinit(void) 395 { 396 397 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 398 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 399 NULL, IPL_SOFTNET); 400 401 tcp_usrreq_init(); 402 403 /* Initialize timer state. */ 404 tcp_timer_init(); 405 406 /* Initialize the compressed state engine. */ 407 syn_cache_init(); 408 409 /* Initialize the congestion control algorithms. */ 410 tcp_congctl_init(); 411 412 /* Initialize the TCPCB template. */ 413 tcp_tcpcb_template(); 414 415 /* Initialize reassembly queue */ 416 tcpipqent_init(); 417 418 /* SACK */ 419 tcp_sack_init(); 420 421 MOWNER_ATTACH(&tcp_tx_mowner); 422 MOWNER_ATTACH(&tcp_rx_mowner); 423 MOWNER_ATTACH(&tcp_reass_mowner); 424 MOWNER_ATTACH(&tcp_sock_mowner); 425 MOWNER_ATTACH(&tcp_sock_tx_mowner); 426 MOWNER_ATTACH(&tcp_sock_rx_mowner); 427 MOWNER_ATTACH(&tcp_mowner); 428 429 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS); 430 431 vtw_earlyinit(); 432 433 callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE); 434 callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL); 435 436 return 0; 437 } 438 439 void 440 tcp_init_common(unsigned basehlen) 441 { 442 static ONCE_DECL(dotcpinit); 443 unsigned hlen = basehlen + sizeof(struct tcphdr); 444 unsigned oldhlen; 445 446 if (max_linkhdr + hlen > MHLEN) 447 panic("tcp_init"); 448 while ((oldhlen = max_protohdr) < hlen) 449 atomic_cas_uint(&max_protohdr, oldhlen, hlen); 450 451 RUN_ONCE(&dotcpinit, do_tcpinit); 452 } 453 454 /* 455 * Tcp initialization 456 */ 457 void 458 tcp_init(void) 459 { 460 461 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 462 463 tcp_init_common(sizeof(struct ip)); 464 } 465 466 /* 467 * Create template to be used to send tcp packets on a connection. 468 * Call after host entry created, allocates an mbuf and fills 469 * in a skeletal tcp/ip header, minimizing the amount of work 470 * necessary when the connection is used. 471 */ 472 struct mbuf * 473 tcp_template(struct tcpcb *tp) 474 { 475 struct inpcb *inp = tp->t_inpcb; 476 #ifdef INET6 477 struct in6pcb *in6p = tp->t_in6pcb; 478 #endif 479 struct tcphdr *n; 480 struct mbuf *m; 481 int hlen; 482 483 switch (tp->t_family) { 484 case AF_INET: 485 hlen = sizeof(struct ip); 486 if (inp) 487 break; 488 #ifdef INET6 489 if (in6p) { 490 /* mapped addr case */ 491 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 492 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 493 break; 494 } 495 #endif 496 return NULL; /*EINVAL*/ 497 #ifdef INET6 498 case AF_INET6: 499 hlen = sizeof(struct ip6_hdr); 500 if (in6p) { 501 /* more sainty check? */ 502 break; 503 } 504 return NULL; /*EINVAL*/ 505 #endif 506 default: 507 hlen = 0; /*pacify gcc*/ 508 return NULL; /*EAFNOSUPPORT*/ 509 } 510 #ifdef DIAGNOSTIC 511 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 512 panic("mclbytes too small for t_template"); 513 #endif 514 m = tp->t_template; 515 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 516 ; 517 else { 518 if (m) 519 m_freem(m); 520 m = tp->t_template = NULL; 521 MGETHDR(m, M_DONTWAIT, MT_HEADER); 522 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 523 MCLGET(m, M_DONTWAIT); 524 if ((m->m_flags & M_EXT) == 0) { 525 m_free(m); 526 m = NULL; 527 } 528 } 529 if (m == NULL) 530 return NULL; 531 MCLAIM(m, &tcp_mowner); 532 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 533 } 534 535 memset(mtod(m, void *), 0, m->m_len); 536 537 n = (struct tcphdr *)(mtod(m, char *) + hlen); 538 539 switch (tp->t_family) { 540 case AF_INET: 541 { 542 struct ipovly *ipov; 543 mtod(m, struct ip *)->ip_v = 4; 544 mtod(m, struct ip *)->ip_hl = hlen >> 2; 545 ipov = mtod(m, struct ipovly *); 546 ipov->ih_pr = IPPROTO_TCP; 547 ipov->ih_len = htons(sizeof(struct tcphdr)); 548 if (inp) { 549 ipov->ih_src = inp->inp_laddr; 550 ipov->ih_dst = inp->inp_faddr; 551 } 552 #ifdef INET6 553 else if (in6p) { 554 /* mapped addr case */ 555 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 556 sizeof(ipov->ih_src)); 557 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 558 sizeof(ipov->ih_dst)); 559 } 560 #endif 561 /* 562 * Compute the pseudo-header portion of the checksum 563 * now. We incrementally add in the TCP option and 564 * payload lengths later, and then compute the TCP 565 * checksum right before the packet is sent off onto 566 * the wire. 567 */ 568 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 569 ipov->ih_dst.s_addr, 570 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 571 break; 572 } 573 #ifdef INET6 574 case AF_INET6: 575 { 576 struct ip6_hdr *ip6; 577 mtod(m, struct ip *)->ip_v = 6; 578 ip6 = mtod(m, struct ip6_hdr *); 579 ip6->ip6_nxt = IPPROTO_TCP; 580 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 581 ip6->ip6_src = in6p->in6p_laddr; 582 ip6->ip6_dst = in6p->in6p_faddr; 583 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 584 if (ip6_auto_flowlabel) { 585 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 586 ip6->ip6_flow |= 587 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 588 } 589 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 590 ip6->ip6_vfc |= IPV6_VERSION; 591 592 /* 593 * Compute the pseudo-header portion of the checksum 594 * now. We incrementally add in the TCP option and 595 * payload lengths later, and then compute the TCP 596 * checksum right before the packet is sent off onto 597 * the wire. 598 */ 599 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 600 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 601 htonl(IPPROTO_TCP)); 602 break; 603 } 604 #endif 605 } 606 if (inp) { 607 n->th_sport = inp->inp_lport; 608 n->th_dport = inp->inp_fport; 609 } 610 #ifdef INET6 611 else if (in6p) { 612 n->th_sport = in6p->in6p_lport; 613 n->th_dport = in6p->in6p_fport; 614 } 615 #endif 616 n->th_seq = 0; 617 n->th_ack = 0; 618 n->th_x2 = 0; 619 n->th_off = 5; 620 n->th_flags = 0; 621 n->th_win = 0; 622 n->th_urp = 0; 623 return (m); 624 } 625 626 /* 627 * Send a single message to the TCP at address specified by 628 * the given TCP/IP header. If m == 0, then we make a copy 629 * of the tcpiphdr at ti and send directly to the addressed host. 630 * This is used to force keep alive messages out using the TCP 631 * template for a connection tp->t_template. If flags are given 632 * then we send a message back to the TCP which originated the 633 * segment ti, and discard the mbuf containing it and any other 634 * attached mbufs. 635 * 636 * In any case the ack and sequence number of the transmitted 637 * segment are as specified by the parameters. 638 */ 639 int 640 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m, 641 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 642 { 643 struct route *ro; 644 int error, tlen, win = 0; 645 int hlen; 646 struct ip *ip; 647 #ifdef INET6 648 struct ip6_hdr *ip6; 649 #endif 650 int family; /* family on packet, not inpcb/in6pcb! */ 651 struct tcphdr *th; 652 653 if (tp != NULL && (flags & TH_RST) == 0) { 654 #ifdef DIAGNOSTIC 655 if (tp->t_inpcb && tp->t_in6pcb) 656 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 657 #endif 658 #ifdef INET 659 if (tp->t_inpcb) 660 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 661 #endif 662 #ifdef INET6 663 if (tp->t_in6pcb) 664 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 665 #endif 666 } 667 668 th = NULL; /* Quell uninitialized warning */ 669 ip = NULL; 670 #ifdef INET6 671 ip6 = NULL; 672 #endif 673 if (m == 0) { 674 if (!mtemplate) 675 return EINVAL; 676 677 /* get family information from template */ 678 switch (mtod(mtemplate, struct ip *)->ip_v) { 679 case 4: 680 family = AF_INET; 681 hlen = sizeof(struct ip); 682 break; 683 #ifdef INET6 684 case 6: 685 family = AF_INET6; 686 hlen = sizeof(struct ip6_hdr); 687 break; 688 #endif 689 default: 690 return EAFNOSUPPORT; 691 } 692 693 MGETHDR(m, M_DONTWAIT, MT_HEADER); 694 if (m) { 695 MCLAIM(m, &tcp_tx_mowner); 696 MCLGET(m, M_DONTWAIT); 697 if ((m->m_flags & M_EXT) == 0) { 698 m_free(m); 699 m = NULL; 700 } 701 } 702 if (m == NULL) 703 return (ENOBUFS); 704 705 if (tcp_compat_42) 706 tlen = 1; 707 else 708 tlen = 0; 709 710 m->m_data += max_linkhdr; 711 bcopy(mtod(mtemplate, void *), mtod(m, void *), 712 mtemplate->m_len); 713 switch (family) { 714 case AF_INET: 715 ip = mtod(m, struct ip *); 716 th = (struct tcphdr *)(ip + 1); 717 break; 718 #ifdef INET6 719 case AF_INET6: 720 ip6 = mtod(m, struct ip6_hdr *); 721 th = (struct tcphdr *)(ip6 + 1); 722 break; 723 #endif 724 #if 0 725 default: 726 /* noone will visit here */ 727 m_freem(m); 728 return EAFNOSUPPORT; 729 #endif 730 } 731 flags = TH_ACK; 732 } else { 733 734 if ((m->m_flags & M_PKTHDR) == 0) { 735 #if 0 736 printf("non PKTHDR to tcp_respond\n"); 737 #endif 738 m_freem(m); 739 return EINVAL; 740 } 741 #ifdef DIAGNOSTIC 742 if (!th0) 743 panic("th0 == NULL in tcp_respond"); 744 #endif 745 746 /* get family information from m */ 747 switch (mtod(m, struct ip *)->ip_v) { 748 case 4: 749 family = AF_INET; 750 hlen = sizeof(struct ip); 751 ip = mtod(m, struct ip *); 752 break; 753 #ifdef INET6 754 case 6: 755 family = AF_INET6; 756 hlen = sizeof(struct ip6_hdr); 757 ip6 = mtod(m, struct ip6_hdr *); 758 break; 759 #endif 760 default: 761 m_freem(m); 762 return EAFNOSUPPORT; 763 } 764 /* clear h/w csum flags inherited from rx packet */ 765 m->m_pkthdr.csum_flags = 0; 766 767 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 768 tlen = sizeof(*th0); 769 else 770 tlen = th0->th_off << 2; 771 772 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 773 mtod(m, char *) + hlen == (char *)th0) { 774 m->m_len = hlen + tlen; 775 m_freem(m->m_next); 776 m->m_next = NULL; 777 } else { 778 struct mbuf *n; 779 780 #ifdef DIAGNOSTIC 781 if (max_linkhdr + hlen + tlen > MCLBYTES) { 782 m_freem(m); 783 return EMSGSIZE; 784 } 785 #endif 786 MGETHDR(n, M_DONTWAIT, MT_HEADER); 787 if (n && max_linkhdr + hlen + tlen > MHLEN) { 788 MCLGET(n, M_DONTWAIT); 789 if ((n->m_flags & M_EXT) == 0) { 790 m_freem(n); 791 n = NULL; 792 } 793 } 794 if (!n) { 795 m_freem(m); 796 return ENOBUFS; 797 } 798 799 MCLAIM(n, &tcp_tx_mowner); 800 n->m_data += max_linkhdr; 801 n->m_len = hlen + tlen; 802 m_copyback(n, 0, hlen, mtod(m, void *)); 803 m_copyback(n, hlen, tlen, (void *)th0); 804 805 m_freem(m); 806 m = n; 807 n = NULL; 808 } 809 810 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 811 switch (family) { 812 case AF_INET: 813 ip = mtod(m, struct ip *); 814 th = (struct tcphdr *)(ip + 1); 815 ip->ip_p = IPPROTO_TCP; 816 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 817 ip->ip_p = IPPROTO_TCP; 818 break; 819 #ifdef INET6 820 case AF_INET6: 821 ip6 = mtod(m, struct ip6_hdr *); 822 th = (struct tcphdr *)(ip6 + 1); 823 ip6->ip6_nxt = IPPROTO_TCP; 824 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 825 ip6->ip6_nxt = IPPROTO_TCP; 826 break; 827 #endif 828 #if 0 829 default: 830 /* noone will visit here */ 831 m_freem(m); 832 return EAFNOSUPPORT; 833 #endif 834 } 835 xchg(th->th_dport, th->th_sport, u_int16_t); 836 #undef xchg 837 tlen = 0; /*be friendly with the following code*/ 838 } 839 th->th_seq = htonl(seq); 840 th->th_ack = htonl(ack); 841 th->th_x2 = 0; 842 if ((flags & TH_SYN) == 0) { 843 if (tp) 844 win >>= tp->rcv_scale; 845 if (win > TCP_MAXWIN) 846 win = TCP_MAXWIN; 847 th->th_win = htons((u_int16_t)win); 848 th->th_off = sizeof (struct tcphdr) >> 2; 849 tlen += sizeof(*th); 850 } else 851 tlen += th->th_off << 2; 852 m->m_len = hlen + tlen; 853 m->m_pkthdr.len = hlen + tlen; 854 m_reset_rcvif(m); 855 th->th_flags = flags; 856 th->th_urp = 0; 857 858 switch (family) { 859 #ifdef INET 860 case AF_INET: 861 { 862 struct ipovly *ipov = (struct ipovly *)ip; 863 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1); 864 ipov->ih_len = htons((u_int16_t)tlen); 865 866 th->th_sum = 0; 867 th->th_sum = in_cksum(m, hlen + tlen); 868 ip->ip_len = htons(hlen + tlen); 869 ip->ip_ttl = ip_defttl; 870 break; 871 } 872 #endif 873 #ifdef INET6 874 case AF_INET6: 875 { 876 th->th_sum = 0; 877 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 878 tlen); 879 ip6->ip6_plen = htons(tlen); 880 if (tp && tp->t_in6pcb) 881 ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb); 882 else 883 ip6->ip6_hlim = ip6_defhlim; 884 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 885 if (ip6_auto_flowlabel) { 886 ip6->ip6_flow |= 887 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 888 } 889 break; 890 } 891 #endif 892 } 893 894 if (tp != NULL && tp->t_inpcb != NULL) { 895 ro = &tp->t_inpcb->inp_route; 896 #ifdef DIAGNOSTIC 897 if (family != AF_INET) 898 panic("tcp_respond: address family mismatch"); 899 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 900 panic("tcp_respond: ip_dst %x != inp_faddr %x", 901 ntohl(ip->ip_dst.s_addr), 902 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 903 } 904 #endif 905 } 906 #ifdef INET6 907 else if (tp != NULL && tp->t_in6pcb != NULL) { 908 ro = (struct route *)&tp->t_in6pcb->in6p_route; 909 #ifdef DIAGNOSTIC 910 if (family == AF_INET) { 911 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 912 panic("tcp_respond: not mapped addr"); 913 if (memcmp(&ip->ip_dst, 914 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 915 sizeof(ip->ip_dst)) != 0) { 916 panic("tcp_respond: ip_dst != in6p_faddr"); 917 } 918 } else if (family == AF_INET6) { 919 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 920 &tp->t_in6pcb->in6p_faddr)) 921 panic("tcp_respond: ip6_dst != in6p_faddr"); 922 } else 923 panic("tcp_respond: address family mismatch"); 924 #endif 925 } 926 #endif 927 else 928 ro = NULL; 929 930 switch (family) { 931 #ifdef INET 932 case AF_INET: 933 error = ip_output(m, NULL, ro, 934 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, 935 tp ? tp->t_inpcb : NULL); 936 break; 937 #endif 938 #ifdef INET6 939 case AF_INET6: 940 error = ip6_output(m, NULL, ro, 0, NULL, 941 tp ? tp->t_in6pcb : NULL, NULL); 942 break; 943 #endif 944 default: 945 error = EAFNOSUPPORT; 946 break; 947 } 948 949 return (error); 950 } 951 952 /* 953 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 954 * a bunch of members individually, we maintain this template for the 955 * static and mostly-static components of the TCPCB, and copy it into 956 * the new TCPCB instead. 957 */ 958 static struct tcpcb tcpcb_template = { 959 .t_srtt = TCPTV_SRTTBASE, 960 .t_rttmin = TCPTV_MIN, 961 962 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 963 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 964 .snd_numholes = 0, 965 .snd_cubic_wmax = 0, 966 .snd_cubic_wmax_last = 0, 967 .snd_cubic_ctime = 0, 968 969 .t_partialacks = -1, 970 .t_bytes_acked = 0, 971 .t_sndrexmitpack = 0, 972 .t_rcvoopack = 0, 973 .t_sndzerowin = 0, 974 }; 975 976 /* 977 * Updates the TCPCB template whenever a parameter that would affect 978 * the template is changed. 979 */ 980 void 981 tcp_tcpcb_template(void) 982 { 983 struct tcpcb *tp = &tcpcb_template; 984 int flags; 985 986 tp->t_peermss = tcp_mssdflt; 987 tp->t_ourmss = tcp_mssdflt; 988 tp->t_segsz = tcp_mssdflt; 989 990 flags = 0; 991 if (tcp_do_rfc1323 && tcp_do_win_scale) 992 flags |= TF_REQ_SCALE; 993 if (tcp_do_rfc1323 && tcp_do_timestamps) 994 flags |= TF_REQ_TSTMP; 995 tp->t_flags = flags; 996 997 /* 998 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 999 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 1000 * reasonable initial retransmit time. 1001 */ 1002 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 1003 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1004 TCPTV_MIN, TCPTV_REXMTMAX); 1005 1006 /* Keep Alive */ 1007 tp->t_keepinit = tcp_keepinit; 1008 tp->t_keepidle = tcp_keepidle; 1009 tp->t_keepintvl = tcp_keepintvl; 1010 tp->t_keepcnt = tcp_keepcnt; 1011 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 1012 1013 /* MSL */ 1014 tp->t_msl = TCPTV_MSL; 1015 } 1016 1017 /* 1018 * Create a new TCP control block, making an 1019 * empty reassembly queue and hooking it to the argument 1020 * protocol control block. 1021 */ 1022 /* family selects inpcb, or in6pcb */ 1023 struct tcpcb * 1024 tcp_newtcpcb(int family, void *aux) 1025 { 1026 struct tcpcb *tp; 1027 int i; 1028 1029 /* XXX Consider using a pool_cache for speed. */ 1030 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 1031 if (tp == NULL) 1032 return (NULL); 1033 memcpy(tp, &tcpcb_template, sizeof(*tp)); 1034 TAILQ_INIT(&tp->segq); 1035 TAILQ_INIT(&tp->timeq); 1036 tp->t_family = family; /* may be overridden later on */ 1037 TAILQ_INIT(&tp->snd_holes); 1038 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1039 1040 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1041 for (i = 0; i < TCPT_NTIMERS; i++) { 1042 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 1043 TCP_TIMER_INIT(tp, i); 1044 } 1045 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 1046 1047 switch (family) { 1048 case AF_INET: 1049 { 1050 struct inpcb *inp = (struct inpcb *)aux; 1051 1052 inp->inp_ip.ip_ttl = ip_defttl; 1053 inp->inp_ppcb = (void *)tp; 1054 1055 tp->t_inpcb = inp; 1056 tp->t_mtudisc = ip_mtudisc; 1057 break; 1058 } 1059 #ifdef INET6 1060 case AF_INET6: 1061 { 1062 struct in6pcb *in6p = (struct in6pcb *)aux; 1063 1064 in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p); 1065 in6p->in6p_ppcb = (void *)tp; 1066 1067 tp->t_in6pcb = in6p; 1068 /* for IPv6, always try to run path MTU discovery */ 1069 tp->t_mtudisc = 1; 1070 break; 1071 } 1072 #endif /* INET6 */ 1073 default: 1074 for (i = 0; i < TCPT_NTIMERS; i++) 1075 callout_destroy(&tp->t_timer[i]); 1076 callout_destroy(&tp->t_delack_ch); 1077 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1078 return (NULL); 1079 } 1080 1081 /* 1082 * Initialize our timebase. When we send timestamps, we take 1083 * the delta from tcp_now -- this means each connection always 1084 * gets a timebase of 1, which makes it, among other things, 1085 * more difficult to determine how long a system has been up, 1086 * and thus how many TCP sequence increments have occurred. 1087 * 1088 * We start with 1, because 0 doesn't work with linux, which 1089 * considers timestamp 0 in a SYN packet as a bug and disables 1090 * timestamps. 1091 */ 1092 tp->ts_timebase = tcp_now - 1; 1093 1094 tcp_congctl_select(tp, tcp_congctl_global_name); 1095 1096 return (tp); 1097 } 1098 1099 /* 1100 * Drop a TCP connection, reporting 1101 * the specified error. If connection is synchronized, 1102 * then send a RST to peer. 1103 */ 1104 struct tcpcb * 1105 tcp_drop(struct tcpcb *tp, int errno) 1106 { 1107 struct socket *so = NULL; 1108 1109 #ifdef DIAGNOSTIC 1110 if (tp->t_inpcb && tp->t_in6pcb) 1111 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1112 #endif 1113 #ifdef INET 1114 if (tp->t_inpcb) 1115 so = tp->t_inpcb->inp_socket; 1116 #endif 1117 #ifdef INET6 1118 if (tp->t_in6pcb) 1119 so = tp->t_in6pcb->in6p_socket; 1120 #endif 1121 if (!so) 1122 return NULL; 1123 1124 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1125 tp->t_state = TCPS_CLOSED; 1126 (void) tcp_output(tp); 1127 TCP_STATINC(TCP_STAT_DROPS); 1128 } else 1129 TCP_STATINC(TCP_STAT_CONNDROPS); 1130 if (errno == ETIMEDOUT && tp->t_softerror) 1131 errno = tp->t_softerror; 1132 so->so_error = errno; 1133 return (tcp_close(tp)); 1134 } 1135 1136 /* 1137 * Close a TCP control block: 1138 * discard all space held by the tcp 1139 * discard internet protocol block 1140 * wake up any sleepers 1141 */ 1142 struct tcpcb * 1143 tcp_close(struct tcpcb *tp) 1144 { 1145 struct inpcb *inp; 1146 #ifdef INET6 1147 struct in6pcb *in6p; 1148 #endif 1149 struct socket *so; 1150 #ifdef RTV_RTT 1151 struct rtentry *rt = NULL; 1152 #endif 1153 struct route *ro; 1154 int j; 1155 1156 inp = tp->t_inpcb; 1157 #ifdef INET6 1158 in6p = tp->t_in6pcb; 1159 #endif 1160 so = NULL; 1161 ro = NULL; 1162 if (inp) { 1163 so = inp->inp_socket; 1164 ro = &inp->inp_route; 1165 } 1166 #ifdef INET6 1167 else if (in6p) { 1168 so = in6p->in6p_socket; 1169 ro = (struct route *)&in6p->in6p_route; 1170 } 1171 #endif 1172 1173 #ifdef RTV_RTT 1174 /* 1175 * If we sent enough data to get some meaningful characteristics, 1176 * save them in the routing entry. 'Enough' is arbitrarily 1177 * defined as the sendpipesize (default 4K) * 16. This would 1178 * give us 16 rtt samples assuming we only get one sample per 1179 * window (the usual case on a long haul net). 16 samples is 1180 * enough for the srtt filter to converge to within 5% of the correct 1181 * value; fewer samples and we could save a very bogus rtt. 1182 * 1183 * Don't update the default route's characteristics and don't 1184 * update anything that the user "locked". 1185 */ 1186 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1187 ro && (rt = rtcache_validate(ro)) != NULL && 1188 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1189 u_long i = 0; 1190 1191 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1192 i = tp->t_srtt * 1193 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1194 if (rt->rt_rmx.rmx_rtt && i) 1195 /* 1196 * filter this update to half the old & half 1197 * the new values, converting scale. 1198 * See route.h and tcp_var.h for a 1199 * description of the scaling constants. 1200 */ 1201 rt->rt_rmx.rmx_rtt = 1202 (rt->rt_rmx.rmx_rtt + i) / 2; 1203 else 1204 rt->rt_rmx.rmx_rtt = i; 1205 } 1206 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1207 i = tp->t_rttvar * 1208 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1209 if (rt->rt_rmx.rmx_rttvar && i) 1210 rt->rt_rmx.rmx_rttvar = 1211 (rt->rt_rmx.rmx_rttvar + i) / 2; 1212 else 1213 rt->rt_rmx.rmx_rttvar = i; 1214 } 1215 /* 1216 * update the pipelimit (ssthresh) if it has been updated 1217 * already or if a pipesize was specified & the threshhold 1218 * got below half the pipesize. I.e., wait for bad news 1219 * before we start updating, then update on both good 1220 * and bad news. 1221 */ 1222 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1223 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1224 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1225 /* 1226 * convert the limit from user data bytes to 1227 * packets then to packet data bytes. 1228 */ 1229 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1230 if (i < 2) 1231 i = 2; 1232 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1233 if (rt->rt_rmx.rmx_ssthresh) 1234 rt->rt_rmx.rmx_ssthresh = 1235 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1236 else 1237 rt->rt_rmx.rmx_ssthresh = i; 1238 } 1239 } 1240 rtcache_unref(rt, ro); 1241 #endif /* RTV_RTT */ 1242 /* free the reassembly queue, if any */ 1243 TCP_REASS_LOCK(tp); 1244 (void) tcp_freeq(tp); 1245 TCP_REASS_UNLOCK(tp); 1246 1247 /* free the SACK holes list. */ 1248 tcp_free_sackholes(tp); 1249 tcp_congctl_release(tp); 1250 syn_cache_cleanup(tp); 1251 1252 if (tp->t_template) { 1253 m_free(tp->t_template); 1254 tp->t_template = NULL; 1255 } 1256 1257 /* 1258 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1259 * the timers can also drop the lock. We need to prevent access 1260 * to the tcpcb as it's half torn down. Flag the pcb as dead 1261 * (prevents access by timers) and only then detach it. 1262 */ 1263 tp->t_flags |= TF_DEAD; 1264 if (inp) { 1265 inp->inp_ppcb = 0; 1266 soisdisconnected(so); 1267 in_pcbdetach(inp); 1268 } 1269 #ifdef INET6 1270 else if (in6p) { 1271 in6p->in6p_ppcb = 0; 1272 soisdisconnected(so); 1273 in6_pcbdetach(in6p); 1274 } 1275 #endif 1276 /* 1277 * pcb is no longer visble elsewhere, so we can safely release 1278 * the lock in callout_halt() if needed. 1279 */ 1280 TCP_STATINC(TCP_STAT_CLOSED); 1281 for (j = 0; j < TCPT_NTIMERS; j++) { 1282 callout_halt(&tp->t_timer[j], softnet_lock); 1283 callout_destroy(&tp->t_timer[j]); 1284 } 1285 callout_halt(&tp->t_delack_ch, softnet_lock); 1286 callout_destroy(&tp->t_delack_ch); 1287 pool_put(&tcpcb_pool, tp); 1288 1289 return NULL; 1290 } 1291 1292 int 1293 tcp_freeq(struct tcpcb *tp) 1294 { 1295 struct ipqent *qe; 1296 int rv = 0; 1297 #ifdef TCPREASS_DEBUG 1298 int i = 0; 1299 #endif 1300 1301 TCP_REASS_LOCK_CHECK(tp); 1302 1303 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1304 #ifdef TCPREASS_DEBUG 1305 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1306 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1307 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1308 #endif 1309 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1310 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1311 m_freem(qe->ipqe_m); 1312 tcpipqent_free(qe); 1313 rv = 1; 1314 } 1315 tp->t_segqlen = 0; 1316 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1317 return (rv); 1318 } 1319 1320 void 1321 tcp_fasttimo(void) 1322 { 1323 if (tcp_drainwanted) { 1324 tcp_drain(); 1325 tcp_drainwanted = 0; 1326 } 1327 } 1328 1329 void 1330 tcp_drainstub(void) 1331 { 1332 tcp_drainwanted = 1; 1333 } 1334 1335 /* 1336 * Protocol drain routine. Called when memory is in short supply. 1337 * Called from pr_fasttimo thus a callout context. 1338 */ 1339 void 1340 tcp_drain(void) 1341 { 1342 struct inpcb_hdr *inph; 1343 struct tcpcb *tp; 1344 1345 mutex_enter(softnet_lock); 1346 KERNEL_LOCK(1, NULL); 1347 1348 /* 1349 * Free the sequence queue of all TCP connections. 1350 */ 1351 TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1352 switch (inph->inph_af) { 1353 case AF_INET: 1354 tp = intotcpcb((struct inpcb *)inph); 1355 break; 1356 #ifdef INET6 1357 case AF_INET6: 1358 tp = in6totcpcb((struct in6pcb *)inph); 1359 break; 1360 #endif 1361 default: 1362 tp = NULL; 1363 break; 1364 } 1365 if (tp != NULL) { 1366 /* 1367 * We may be called from a device's interrupt 1368 * context. If the tcpcb is already busy, 1369 * just bail out now. 1370 */ 1371 if (tcp_reass_lock_try(tp) == 0) 1372 continue; 1373 if (tcp_freeq(tp)) 1374 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1375 TCP_REASS_UNLOCK(tp); 1376 } 1377 } 1378 1379 KERNEL_UNLOCK_ONE(NULL); 1380 mutex_exit(softnet_lock); 1381 } 1382 1383 /* 1384 * Notify a tcp user of an asynchronous error; 1385 * store error as soft error, but wake up user 1386 * (for now, won't do anything until can select for soft error). 1387 */ 1388 void 1389 tcp_notify(struct inpcb *inp, int error) 1390 { 1391 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1392 struct socket *so = inp->inp_socket; 1393 1394 /* 1395 * Ignore some errors if we are hooked up. 1396 * If connection hasn't completed, has retransmitted several times, 1397 * and receives a second error, give up now. This is better 1398 * than waiting a long time to establish a connection that 1399 * can never complete. 1400 */ 1401 if (tp->t_state == TCPS_ESTABLISHED && 1402 (error == EHOSTUNREACH || error == ENETUNREACH || 1403 error == EHOSTDOWN)) { 1404 return; 1405 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1406 tp->t_rxtshift > 3 && tp->t_softerror) 1407 so->so_error = error; 1408 else 1409 tp->t_softerror = error; 1410 cv_broadcast(&so->so_cv); 1411 sorwakeup(so); 1412 sowwakeup(so); 1413 } 1414 1415 #ifdef INET6 1416 void 1417 tcp6_notify(struct in6pcb *in6p, int error) 1418 { 1419 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1420 struct socket *so = in6p->in6p_socket; 1421 1422 /* 1423 * Ignore some errors if we are hooked up. 1424 * If connection hasn't completed, has retransmitted several times, 1425 * and receives a second error, give up now. This is better 1426 * than waiting a long time to establish a connection that 1427 * can never complete. 1428 */ 1429 if (tp->t_state == TCPS_ESTABLISHED && 1430 (error == EHOSTUNREACH || error == ENETUNREACH || 1431 error == EHOSTDOWN)) { 1432 return; 1433 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1434 tp->t_rxtshift > 3 && tp->t_softerror) 1435 so->so_error = error; 1436 else 1437 tp->t_softerror = error; 1438 cv_broadcast(&so->so_cv); 1439 sorwakeup(so); 1440 sowwakeup(so); 1441 } 1442 #endif 1443 1444 #ifdef INET6 1445 void * 1446 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1447 { 1448 struct tcphdr th; 1449 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1450 int nmatch; 1451 struct ip6_hdr *ip6; 1452 const struct sockaddr_in6 *sa6_src = NULL; 1453 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1454 struct mbuf *m; 1455 int off; 1456 1457 if (sa->sa_family != AF_INET6 || 1458 sa->sa_len != sizeof(struct sockaddr_in6)) 1459 return NULL; 1460 if ((unsigned)cmd >= PRC_NCMDS) 1461 return NULL; 1462 else if (cmd == PRC_QUENCH) { 1463 /* 1464 * Don't honor ICMP Source Quench messages meant for 1465 * TCP connections. 1466 */ 1467 return NULL; 1468 } else if (PRC_IS_REDIRECT(cmd)) 1469 notify = in6_rtchange, d = NULL; 1470 else if (cmd == PRC_MSGSIZE) 1471 ; /* special code is present, see below */ 1472 else if (cmd == PRC_HOSTDEAD) 1473 d = NULL; 1474 else if (inet6ctlerrmap[cmd] == 0) 1475 return NULL; 1476 1477 /* if the parameter is from icmp6, decode it. */ 1478 if (d != NULL) { 1479 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1480 m = ip6cp->ip6c_m; 1481 ip6 = ip6cp->ip6c_ip6; 1482 off = ip6cp->ip6c_off; 1483 sa6_src = ip6cp->ip6c_src; 1484 } else { 1485 m = NULL; 1486 ip6 = NULL; 1487 sa6_src = &sa6_any; 1488 off = 0; 1489 } 1490 1491 if (ip6) { 1492 /* 1493 * XXX: We assume that when ip6 is non NULL, 1494 * M and OFF are valid. 1495 */ 1496 1497 /* check if we can safely examine src and dst ports */ 1498 if (m->m_pkthdr.len < off + sizeof(th)) { 1499 if (cmd == PRC_MSGSIZE) 1500 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1501 return NULL; 1502 } 1503 1504 memset(&th, 0, sizeof(th)); 1505 m_copydata(m, off, sizeof(th), (void *)&th); 1506 1507 if (cmd == PRC_MSGSIZE) { 1508 int valid = 0; 1509 1510 /* 1511 * Check to see if we have a valid TCP connection 1512 * corresponding to the address in the ICMPv6 message 1513 * payload. 1514 */ 1515 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1516 th.th_dport, 1517 (const struct in6_addr *)&sa6_src->sin6_addr, 1518 th.th_sport, 0, 0)) 1519 valid++; 1520 1521 /* 1522 * Depending on the value of "valid" and routing table 1523 * size (mtudisc_{hi,lo}wat), we will: 1524 * - recalcurate the new MTU and create the 1525 * corresponding routing entry, or 1526 * - ignore the MTU change notification. 1527 */ 1528 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1529 1530 /* 1531 * no need to call in6_pcbnotify, it should have been 1532 * called via callback if necessary 1533 */ 1534 return NULL; 1535 } 1536 1537 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1538 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1539 if (nmatch == 0 && syn_cache_count && 1540 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1541 inet6ctlerrmap[cmd] == ENETUNREACH || 1542 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1543 syn_cache_unreach((const struct sockaddr *)sa6_src, 1544 sa, &th); 1545 } else { 1546 (void) in6_pcbnotify(&tcbtable, sa, 0, 1547 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1548 } 1549 1550 return NULL; 1551 } 1552 #endif 1553 1554 #ifdef INET 1555 /* assumes that ip header and tcp header are contiguous on mbuf */ 1556 void * 1557 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1558 { 1559 struct ip *ip = v; 1560 struct tcphdr *th; 1561 struct icmp *icp; 1562 extern const int inetctlerrmap[]; 1563 void (*notify)(struct inpcb *, int) = tcp_notify; 1564 int errno; 1565 int nmatch; 1566 struct tcpcb *tp; 1567 u_int mtu; 1568 tcp_seq seq; 1569 struct inpcb *inp; 1570 #ifdef INET6 1571 struct in6pcb *in6p; 1572 struct in6_addr src6, dst6; 1573 #endif 1574 1575 if (sa->sa_family != AF_INET || 1576 sa->sa_len != sizeof(struct sockaddr_in)) 1577 return NULL; 1578 if ((unsigned)cmd >= PRC_NCMDS) 1579 return NULL; 1580 errno = inetctlerrmap[cmd]; 1581 if (cmd == PRC_QUENCH) 1582 /* 1583 * Don't honor ICMP Source Quench messages meant for 1584 * TCP connections. 1585 */ 1586 return NULL; 1587 else if (PRC_IS_REDIRECT(cmd)) 1588 notify = in_rtchange, ip = 0; 1589 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1590 /* 1591 * Check to see if we have a valid TCP connection 1592 * corresponding to the address in the ICMP message 1593 * payload. 1594 * 1595 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1596 */ 1597 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1598 #ifdef INET6 1599 in6_in_2_v4mapin6(&ip->ip_src, &src6); 1600 in6_in_2_v4mapin6(&ip->ip_dst, &dst6); 1601 #endif 1602 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1603 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL) 1604 #ifdef INET6 1605 in6p = NULL; 1606 #else 1607 ; 1608 #endif 1609 #ifdef INET6 1610 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1611 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL) 1612 ; 1613 #endif 1614 else 1615 return NULL; 1616 1617 /* 1618 * Now that we've validated that we are actually communicating 1619 * with the host indicated in the ICMP message, locate the 1620 * ICMP header, recalculate the new MTU, and create the 1621 * corresponding routing entry. 1622 */ 1623 icp = (struct icmp *)((char *)ip - 1624 offsetof(struct icmp, icmp_ip)); 1625 if (inp) { 1626 if ((tp = intotcpcb(inp)) == NULL) 1627 return NULL; 1628 } 1629 #ifdef INET6 1630 else if (in6p) { 1631 if ((tp = in6totcpcb(in6p)) == NULL) 1632 return NULL; 1633 } 1634 #endif 1635 else 1636 return NULL; 1637 seq = ntohl(th->th_seq); 1638 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1639 return NULL; 1640 /* 1641 * If the ICMP message advertises a Next-Hop MTU 1642 * equal or larger than the maximum packet size we have 1643 * ever sent, drop the message. 1644 */ 1645 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1646 if (mtu >= tp->t_pmtud_mtu_sent) 1647 return NULL; 1648 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1649 /* 1650 * Calculate new MTU, and create corresponding 1651 * route (traditional PMTUD). 1652 */ 1653 tp->t_flags &= ~TF_PMTUD_PEND; 1654 icmp_mtudisc(icp, ip->ip_dst); 1655 } else { 1656 /* 1657 * Record the information got in the ICMP 1658 * message; act on it later. 1659 * If we had already recorded an ICMP message, 1660 * replace the old one only if the new message 1661 * refers to an older TCP segment 1662 */ 1663 if (tp->t_flags & TF_PMTUD_PEND) { 1664 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1665 return NULL; 1666 } else 1667 tp->t_flags |= TF_PMTUD_PEND; 1668 tp->t_pmtud_th_seq = seq; 1669 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1670 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1671 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1672 } 1673 return NULL; 1674 } else if (cmd == PRC_HOSTDEAD) 1675 ip = 0; 1676 else if (errno == 0) 1677 return NULL; 1678 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1679 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1680 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1681 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1682 if (nmatch == 0 && syn_cache_count && 1683 (inetctlerrmap[cmd] == EHOSTUNREACH || 1684 inetctlerrmap[cmd] == ENETUNREACH || 1685 inetctlerrmap[cmd] == EHOSTDOWN)) { 1686 struct sockaddr_in sin; 1687 memset(&sin, 0, sizeof(sin)); 1688 sin.sin_len = sizeof(sin); 1689 sin.sin_family = AF_INET; 1690 sin.sin_port = th->th_sport; 1691 sin.sin_addr = ip->ip_src; 1692 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1693 } 1694 1695 /* XXX mapped address case */ 1696 } else 1697 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1698 notify); 1699 return NULL; 1700 } 1701 1702 /* 1703 * When a source quench is received, we are being notified of congestion. 1704 * Close the congestion window down to the Loss Window (one segment). 1705 * We will gradually open it again as we proceed. 1706 */ 1707 void 1708 tcp_quench(struct inpcb *inp, int errno) 1709 { 1710 struct tcpcb *tp = intotcpcb(inp); 1711 1712 if (tp) { 1713 tp->snd_cwnd = tp->t_segsz; 1714 tp->t_bytes_acked = 0; 1715 } 1716 } 1717 #endif 1718 1719 #ifdef INET6 1720 void 1721 tcp6_quench(struct in6pcb *in6p, int errno) 1722 { 1723 struct tcpcb *tp = in6totcpcb(in6p); 1724 1725 if (tp) { 1726 tp->snd_cwnd = tp->t_segsz; 1727 tp->t_bytes_acked = 0; 1728 } 1729 } 1730 #endif 1731 1732 #ifdef INET 1733 /* 1734 * Path MTU Discovery handlers. 1735 */ 1736 void 1737 tcp_mtudisc_callback(struct in_addr faddr) 1738 { 1739 #ifdef INET6 1740 struct in6_addr in6; 1741 #endif 1742 1743 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1744 #ifdef INET6 1745 in6_in_2_v4mapin6(&faddr, &in6); 1746 tcp6_mtudisc_callback(&in6); 1747 #endif 1748 } 1749 1750 /* 1751 * On receipt of path MTU corrections, flush old route and replace it 1752 * with the new one. Retransmit all unacknowledged packets, to ensure 1753 * that all packets will be received. 1754 */ 1755 void 1756 tcp_mtudisc(struct inpcb *inp, int errno) 1757 { 1758 struct tcpcb *tp = intotcpcb(inp); 1759 struct rtentry *rt; 1760 1761 if (tp == NULL) 1762 return; 1763 1764 rt = in_pcbrtentry(inp); 1765 if (rt != NULL) { 1766 /* 1767 * If this was not a host route, remove and realloc. 1768 */ 1769 if ((rt->rt_flags & RTF_HOST) == 0) { 1770 in_pcbrtentry_unref(rt, inp); 1771 in_rtchange(inp, errno); 1772 if ((rt = in_pcbrtentry(inp)) == NULL) 1773 return; 1774 } 1775 1776 /* 1777 * Slow start out of the error condition. We 1778 * use the MTU because we know it's smaller 1779 * than the previously transmitted segment. 1780 * 1781 * Note: This is more conservative than the 1782 * suggestion in draft-floyd-incr-init-win-03. 1783 */ 1784 if (rt->rt_rmx.rmx_mtu != 0) 1785 tp->snd_cwnd = 1786 TCP_INITIAL_WINDOW(tcp_init_win, 1787 rt->rt_rmx.rmx_mtu); 1788 in_pcbrtentry_unref(rt, inp); 1789 } 1790 1791 /* 1792 * Resend unacknowledged packets. 1793 */ 1794 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1795 tcp_output(tp); 1796 } 1797 #endif /* INET */ 1798 1799 #ifdef INET6 1800 /* 1801 * Path MTU Discovery handlers. 1802 */ 1803 void 1804 tcp6_mtudisc_callback(struct in6_addr *faddr) 1805 { 1806 struct sockaddr_in6 sin6; 1807 1808 memset(&sin6, 0, sizeof(sin6)); 1809 sin6.sin6_family = AF_INET6; 1810 sin6.sin6_len = sizeof(struct sockaddr_in6); 1811 sin6.sin6_addr = *faddr; 1812 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1813 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1814 } 1815 1816 void 1817 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1818 { 1819 struct tcpcb *tp = in6totcpcb(in6p); 1820 struct rtentry *rt; 1821 1822 if (tp == NULL) 1823 return; 1824 1825 rt = in6_pcbrtentry(in6p); 1826 if (rt != NULL) { 1827 /* 1828 * If this was not a host route, remove and realloc. 1829 */ 1830 if ((rt->rt_flags & RTF_HOST) == 0) { 1831 in6_pcbrtentry_unref(rt, in6p); 1832 in6_rtchange(in6p, errno); 1833 rt = in6_pcbrtentry(in6p); 1834 if (rt == NULL) 1835 return; 1836 } 1837 1838 /* 1839 * Slow start out of the error condition. We 1840 * use the MTU because we know it's smaller 1841 * than the previously transmitted segment. 1842 * 1843 * Note: This is more conservative than the 1844 * suggestion in draft-floyd-incr-init-win-03. 1845 */ 1846 if (rt->rt_rmx.rmx_mtu != 0) { 1847 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1848 rt->rt_rmx.rmx_mtu); 1849 } 1850 in6_pcbrtentry_unref(rt, in6p); 1851 } 1852 1853 /* 1854 * Resend unacknowledged packets. 1855 */ 1856 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1857 tcp_output(tp); 1858 } 1859 #endif /* INET6 */ 1860 1861 /* 1862 * Compute the MSS to advertise to the peer. Called only during 1863 * the 3-way handshake. If we are the server (peer initiated 1864 * connection), we are called with a pointer to the interface 1865 * on which the SYN packet arrived. If we are the client (we 1866 * initiated connection), we are called with a pointer to the 1867 * interface out which this connection should go. 1868 * 1869 * NOTE: Do not subtract IP option/extension header size nor IPsec 1870 * header size from MSS advertisement. MSS option must hold the maximum 1871 * segment size we can accept, so it must always be: 1872 * max(if mtu) - ip header - tcp header 1873 */ 1874 u_long 1875 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1876 { 1877 extern u_long in_maxmtu; 1878 u_long mss = 0; 1879 u_long hdrsiz; 1880 1881 /* 1882 * In order to avoid defeating path MTU discovery on the peer, 1883 * we advertise the max MTU of all attached networks as our MSS, 1884 * per RFC 1191, section 3.1. 1885 * 1886 * We provide the option to advertise just the MTU of 1887 * the interface on which we hope this connection will 1888 * be receiving. If we are responding to a SYN, we 1889 * will have a pretty good idea about this, but when 1890 * initiating a connection there is a bit more doubt. 1891 * 1892 * We also need to ensure that loopback has a large enough 1893 * MSS, as the loopback MTU is never included in in_maxmtu. 1894 */ 1895 1896 if (ifp != NULL) 1897 switch (af) { 1898 case AF_INET: 1899 mss = ifp->if_mtu; 1900 break; 1901 #ifdef INET6 1902 case AF_INET6: 1903 mss = IN6_LINKMTU(ifp); 1904 break; 1905 #endif 1906 } 1907 1908 if (tcp_mss_ifmtu == 0) 1909 switch (af) { 1910 case AF_INET: 1911 mss = max(in_maxmtu, mss); 1912 break; 1913 #ifdef INET6 1914 case AF_INET6: 1915 mss = max(in6_maxmtu, mss); 1916 break; 1917 #endif 1918 } 1919 1920 switch (af) { 1921 case AF_INET: 1922 hdrsiz = sizeof(struct ip); 1923 break; 1924 #ifdef INET6 1925 case AF_INET6: 1926 hdrsiz = sizeof(struct ip6_hdr); 1927 break; 1928 #endif 1929 default: 1930 hdrsiz = 0; 1931 break; 1932 } 1933 hdrsiz += sizeof(struct tcphdr); 1934 if (mss > hdrsiz) 1935 mss -= hdrsiz; 1936 1937 mss = max(tcp_mssdflt, mss); 1938 return (mss); 1939 } 1940 1941 /* 1942 * Set connection variables based on the peer's advertised MSS. 1943 * We are passed the TCPCB for the actual connection. If we 1944 * are the server, we are called by the compressed state engine 1945 * when the 3-way handshake is complete. If we are the client, 1946 * we are called when we receive the SYN,ACK from the server. 1947 * 1948 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1949 * before this routine is called! 1950 */ 1951 void 1952 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1953 { 1954 struct socket *so; 1955 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1956 struct rtentry *rt; 1957 #endif 1958 u_long bufsize; 1959 int mss; 1960 1961 #ifdef DIAGNOSTIC 1962 if (tp->t_inpcb && tp->t_in6pcb) 1963 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1964 #endif 1965 so = NULL; 1966 rt = NULL; 1967 #ifdef INET 1968 if (tp->t_inpcb) { 1969 so = tp->t_inpcb->inp_socket; 1970 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1971 rt = in_pcbrtentry(tp->t_inpcb); 1972 #endif 1973 } 1974 #endif 1975 #ifdef INET6 1976 if (tp->t_in6pcb) { 1977 so = tp->t_in6pcb->in6p_socket; 1978 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1979 rt = in6_pcbrtentry(tp->t_in6pcb); 1980 #endif 1981 } 1982 #endif 1983 1984 /* 1985 * As per RFC1122, use the default MSS value, unless they 1986 * sent us an offer. Do not accept offers less than 256 bytes. 1987 */ 1988 mss = tcp_mssdflt; 1989 if (offer) 1990 mss = offer; 1991 mss = max(mss, 256); /* sanity */ 1992 tp->t_peermss = mss; 1993 mss -= tcp_optlen(tp); 1994 #ifdef INET 1995 if (tp->t_inpcb) 1996 mss -= ip_optlen(tp->t_inpcb); 1997 #endif 1998 #ifdef INET6 1999 if (tp->t_in6pcb) 2000 mss -= ip6_optlen(tp->t_in6pcb); 2001 #endif 2002 2003 /* 2004 * If there's a pipesize, change the socket buffer to that size. 2005 * Make the socket buffer an integral number of MSS units. If 2006 * the MSS is larger than the socket buffer, artificially decrease 2007 * the MSS. 2008 */ 2009 #ifdef RTV_SPIPE 2010 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 2011 bufsize = rt->rt_rmx.rmx_sendpipe; 2012 else 2013 #endif 2014 { 2015 KASSERT(so != NULL); 2016 bufsize = so->so_snd.sb_hiwat; 2017 } 2018 if (bufsize < mss) 2019 mss = bufsize; 2020 else { 2021 bufsize = roundup(bufsize, mss); 2022 if (bufsize > sb_max) 2023 bufsize = sb_max; 2024 (void) sbreserve(&so->so_snd, bufsize, so); 2025 } 2026 tp->t_segsz = mss; 2027 2028 #ifdef RTV_SSTHRESH 2029 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 2030 /* 2031 * There's some sort of gateway or interface buffer 2032 * limit on the path. Use this to set the slow 2033 * start threshold, but set the threshold to no less 2034 * than 2 * MSS. 2035 */ 2036 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2037 } 2038 #endif 2039 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 2040 #ifdef INET 2041 if (tp->t_inpcb) 2042 in_pcbrtentry_unref(rt, tp->t_inpcb); 2043 #endif 2044 #ifdef INET6 2045 if (tp->t_in6pcb) 2046 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2047 #endif 2048 #endif 2049 } 2050 2051 /* 2052 * Processing necessary when a TCP connection is established. 2053 */ 2054 void 2055 tcp_established(struct tcpcb *tp) 2056 { 2057 struct socket *so; 2058 #ifdef RTV_RPIPE 2059 struct rtentry *rt; 2060 #endif 2061 u_long bufsize; 2062 2063 #ifdef DIAGNOSTIC 2064 if (tp->t_inpcb && tp->t_in6pcb) 2065 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2066 #endif 2067 so = NULL; 2068 rt = NULL; 2069 #ifdef INET 2070 /* This is a while() to reduce the dreadful stairstepping below */ 2071 while (tp->t_inpcb) { 2072 so = tp->t_inpcb->inp_socket; 2073 #if defined(RTV_RPIPE) 2074 rt = in_pcbrtentry(tp->t_inpcb); 2075 #endif 2076 if (__predict_true(tcp_msl_enable)) { 2077 if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) { 2078 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2079 break; 2080 } 2081 2082 if (__predict_false(tcp_rttlocal)) { 2083 /* This may be adjusted by tcp_input */ 2084 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2085 break; 2086 } 2087 if (in_localaddr(tp->t_inpcb->inp_faddr)) { 2088 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2089 break; 2090 } 2091 } 2092 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2093 break; 2094 } 2095 #endif 2096 #ifdef INET6 2097 /* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */ 2098 while (!tp->t_inpcb && tp->t_in6pcb) { 2099 so = tp->t_in6pcb->in6p_socket; 2100 #if defined(RTV_RPIPE) 2101 rt = in6_pcbrtentry(tp->t_in6pcb); 2102 #endif 2103 if (__predict_true(tcp_msl_enable)) { 2104 extern const struct in6_addr in6addr_loopback; 2105 2106 if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr, 2107 &in6addr_loopback)) { 2108 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2109 break; 2110 } 2111 2112 if (__predict_false(tcp_rttlocal)) { 2113 /* This may be adjusted by tcp_input */ 2114 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2115 break; 2116 } 2117 if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) { 2118 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2119 break; 2120 } 2121 } 2122 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2123 break; 2124 } 2125 #endif 2126 2127 tp->t_state = TCPS_ESTABLISHED; 2128 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2129 2130 #ifdef RTV_RPIPE 2131 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2132 bufsize = rt->rt_rmx.rmx_recvpipe; 2133 else 2134 #endif 2135 { 2136 KASSERT(so != NULL); 2137 bufsize = so->so_rcv.sb_hiwat; 2138 } 2139 if (bufsize > tp->t_ourmss) { 2140 bufsize = roundup(bufsize, tp->t_ourmss); 2141 if (bufsize > sb_max) 2142 bufsize = sb_max; 2143 (void) sbreserve(&so->so_rcv, bufsize, so); 2144 } 2145 #ifdef RTV_RPIPE 2146 #ifdef INET 2147 if (tp->t_inpcb) 2148 in_pcbrtentry_unref(rt, tp->t_inpcb); 2149 #endif 2150 #ifdef INET6 2151 if (tp->t_in6pcb) 2152 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2153 #endif 2154 #endif 2155 } 2156 2157 /* 2158 * Check if there's an initial rtt or rttvar. Convert from the 2159 * route-table units to scaled multiples of the slow timeout timer. 2160 * Called only during the 3-way handshake. 2161 */ 2162 void 2163 tcp_rmx_rtt(struct tcpcb *tp) 2164 { 2165 #ifdef RTV_RTT 2166 struct rtentry *rt = NULL; 2167 int rtt; 2168 2169 #ifdef DIAGNOSTIC 2170 if (tp->t_inpcb && tp->t_in6pcb) 2171 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2172 #endif 2173 #ifdef INET 2174 if (tp->t_inpcb) 2175 rt = in_pcbrtentry(tp->t_inpcb); 2176 #endif 2177 #ifdef INET6 2178 if (tp->t_in6pcb) 2179 rt = in6_pcbrtentry(tp->t_in6pcb); 2180 #endif 2181 if (rt == NULL) 2182 return; 2183 2184 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2185 /* 2186 * XXX The lock bit for MTU indicates that the value 2187 * is also a minimum value; this is subject to time. 2188 */ 2189 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2190 TCPT_RANGESET(tp->t_rttmin, 2191 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2192 TCPTV_MIN, TCPTV_REXMTMAX); 2193 tp->t_srtt = rtt / 2194 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2195 if (rt->rt_rmx.rmx_rttvar) { 2196 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2197 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2198 (TCP_RTTVAR_SHIFT + 2)); 2199 } else { 2200 /* Default variation is +- 1 rtt */ 2201 tp->t_rttvar = 2202 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2203 } 2204 TCPT_RANGESET(tp->t_rxtcur, 2205 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2206 tp->t_rttmin, TCPTV_REXMTMAX); 2207 } 2208 #ifdef INET 2209 if (tp->t_inpcb) 2210 in_pcbrtentry_unref(rt, tp->t_inpcb); 2211 #endif 2212 #ifdef INET6 2213 if (tp->t_in6pcb) 2214 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2215 #endif 2216 #endif 2217 } 2218 2219 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2220 2221 /* 2222 * Get a new sequence value given a tcp control block 2223 */ 2224 tcp_seq 2225 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2226 { 2227 2228 #ifdef INET 2229 if (tp->t_inpcb != NULL) { 2230 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2231 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2232 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2233 addin)); 2234 } 2235 #endif 2236 #ifdef INET6 2237 if (tp->t_in6pcb != NULL) { 2238 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2239 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2240 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2241 addin)); 2242 } 2243 #endif 2244 /* Not possible. */ 2245 panic("tcp_new_iss"); 2246 } 2247 2248 static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2249 2250 /* 2251 * Initialize RFC 1948 ISS Secret 2252 */ 2253 static int 2254 tcp_iss_secret_init(void) 2255 { 2256 cprng_strong(kern_cprng, 2257 tcp_iss_secret, sizeof(tcp_iss_secret), 0); 2258 2259 return 0; 2260 } 2261 2262 /* 2263 * This routine actually generates a new TCP initial sequence number. 2264 */ 2265 tcp_seq 2266 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2267 size_t addrsz, tcp_seq addin) 2268 { 2269 tcp_seq tcp_iss; 2270 2271 if (tcp_do_rfc1948) { 2272 MD5_CTX ctx; 2273 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2274 static ONCE_DECL(tcp_iss_secret_control); 2275 2276 /* 2277 * If we haven't been here before, initialize our cryptographic 2278 * hash secret. 2279 */ 2280 RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init); 2281 2282 /* 2283 * Compute the base value of the ISS. It is a hash 2284 * of (saddr, sport, daddr, dport, secret). 2285 */ 2286 MD5Init(&ctx); 2287 2288 MD5Update(&ctx, (u_char *) laddr, addrsz); 2289 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2290 2291 MD5Update(&ctx, (u_char *) faddr, addrsz); 2292 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2293 2294 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2295 2296 MD5Final(hash, &ctx); 2297 2298 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2299 2300 /* 2301 * Now increment our "timer", and add it in to 2302 * the computed value. 2303 * 2304 * XXX Use `addin'? 2305 * XXX TCP_ISSINCR too large to use? 2306 */ 2307 tcp_iss_seq += TCP_ISSINCR; 2308 #ifdef TCPISS_DEBUG 2309 printf("ISS hash 0x%08x, ", tcp_iss); 2310 #endif 2311 tcp_iss += tcp_iss_seq + addin; 2312 #ifdef TCPISS_DEBUG 2313 printf("new ISS 0x%08x\n", tcp_iss); 2314 #endif 2315 } else { 2316 /* 2317 * Randomize. 2318 */ 2319 tcp_iss = cprng_fast32(); 2320 2321 /* 2322 * If we were asked to add some amount to a known value, 2323 * we will take a random value obtained above, mask off 2324 * the upper bits, and add in the known value. We also 2325 * add in a constant to ensure that we are at least a 2326 * certain distance from the original value. 2327 * 2328 * This is used when an old connection is in timed wait 2329 * and we have a new one coming in, for instance. 2330 */ 2331 if (addin != 0) { 2332 #ifdef TCPISS_DEBUG 2333 printf("Random %08x, ", tcp_iss); 2334 #endif 2335 tcp_iss &= TCP_ISS_RANDOM_MASK; 2336 tcp_iss += addin + TCP_ISSINCR; 2337 #ifdef TCPISS_DEBUG 2338 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2339 #endif 2340 } else { 2341 tcp_iss &= TCP_ISS_RANDOM_MASK; 2342 tcp_iss += tcp_iss_seq; 2343 tcp_iss_seq += TCP_ISSINCR; 2344 #ifdef TCPISS_DEBUG 2345 printf("ISS %08x\n", tcp_iss); 2346 #endif 2347 } 2348 } 2349 2350 if (tcp_compat_42) { 2351 /* 2352 * Limit it to the positive range for really old TCP 2353 * implementations. 2354 * Just AND off the top bit instead of checking if 2355 * is set first - saves a branch 50% of the time. 2356 */ 2357 tcp_iss &= 0x7fffffff; /* XXX */ 2358 } 2359 2360 return (tcp_iss); 2361 } 2362 2363 #if defined(IPSEC) 2364 /* compute ESP/AH header size for TCP, including outer IP header. */ 2365 size_t 2366 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2367 { 2368 struct inpcb *inp; 2369 size_t hdrsiz; 2370 2371 /* XXX mapped addr case (tp->t_in6pcb) */ 2372 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2373 return 0; 2374 switch (tp->t_family) { 2375 case AF_INET: 2376 /* XXX: should use currect direction. */ 2377 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2378 break; 2379 default: 2380 hdrsiz = 0; 2381 break; 2382 } 2383 2384 return hdrsiz; 2385 } 2386 2387 #ifdef INET6 2388 size_t 2389 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2390 { 2391 struct in6pcb *in6p; 2392 size_t hdrsiz; 2393 2394 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2395 return 0; 2396 switch (tp->t_family) { 2397 case AF_INET6: 2398 /* XXX: should use currect direction. */ 2399 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2400 break; 2401 case AF_INET: 2402 /* mapped address case - tricky */ 2403 default: 2404 hdrsiz = 0; 2405 break; 2406 } 2407 2408 return hdrsiz; 2409 } 2410 #endif 2411 #endif /*IPSEC*/ 2412 2413 /* 2414 * Determine the length of the TCP options for this connection. 2415 * 2416 * XXX: What do we do for SACK, when we add that? Just reserve 2417 * all of the space? Otherwise we can't exactly be incrementing 2418 * cwnd by an amount that varies depending on the amount we last 2419 * had to SACK! 2420 */ 2421 2422 u_int 2423 tcp_optlen(struct tcpcb *tp) 2424 { 2425 u_int optlen; 2426 2427 optlen = 0; 2428 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2429 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2430 optlen += TCPOLEN_TSTAMP_APPA; 2431 2432 #ifdef TCP_SIGNATURE 2433 if (tp->t_flags & TF_SIGNATURE) 2434 optlen += TCPOLEN_SIGLEN; 2435 #endif /* TCP_SIGNATURE */ 2436 2437 return optlen; 2438 } 2439 2440 u_int 2441 tcp_hdrsz(struct tcpcb *tp) 2442 { 2443 u_int hlen; 2444 2445 switch (tp->t_family) { 2446 #ifdef INET6 2447 case AF_INET6: 2448 hlen = sizeof(struct ip6_hdr); 2449 break; 2450 #endif 2451 case AF_INET: 2452 hlen = sizeof(struct ip); 2453 break; 2454 default: 2455 hlen = 0; 2456 break; 2457 } 2458 hlen += sizeof(struct tcphdr); 2459 2460 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2461 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2462 hlen += TCPOLEN_TSTAMP_APPA; 2463 #ifdef TCP_SIGNATURE 2464 if (tp->t_flags & TF_SIGNATURE) 2465 hlen += TCPOLEN_SIGLEN; 2466 #endif 2467 return hlen; 2468 } 2469 2470 void 2471 tcp_statinc(u_int stat) 2472 { 2473 2474 KASSERT(stat < TCP_NSTATS); 2475 TCP_STATINC(stat); 2476 } 2477 2478 void 2479 tcp_statadd(u_int stat, uint64_t val) 2480 { 2481 2482 KASSERT(stat < TCP_NSTATS); 2483 TCP_STATADD(stat, val); 2484 } 2485