tcp_subr.c revision 1.261
1 /* $NetBSD: tcp_subr.c,v 1.261 2015/05/16 10:09:20 kefren 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 1.261 2015/05/16 10:09:20 kefren 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" 101 102 #include <sys/param.h> 103 #include <sys/atomic.h> 104 #include <sys/proc.h> 105 #include <sys/systm.h> 106 #include <sys/mbuf.h> 107 #include <sys/once.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 #include <sys/md5.h> 115 #include <sys/cprng.h> 116 117 #include <net/route.h> 118 #include <net/if.h> 119 120 #include <netinet/in.h> 121 #include <netinet/in_systm.h> 122 #include <netinet/ip.h> 123 #include <netinet/in_pcb.h> 124 #include <netinet/ip_var.h> 125 #include <netinet/ip_icmp.h> 126 127 #ifdef INET6 128 #ifndef INET 129 #include <netinet/in.h> 130 #endif 131 #include <netinet/ip6.h> 132 #include <netinet6/in6_pcb.h> 133 #include <netinet6/ip6_var.h> 134 #include <netinet6/in6_var.h> 135 #include <netinet6/ip6protosw.h> 136 #include <netinet/icmp6.h> 137 #include <netinet6/nd6.h> 138 #endif 139 140 #include <netinet/tcp.h> 141 #include <netinet/tcp_fsm.h> 142 #include <netinet/tcp_seq.h> 143 #include <netinet/tcp_timer.h> 144 #include <netinet/tcp_var.h> 145 #include <netinet/tcp_vtw.h> 146 #include <netinet/tcp_private.h> 147 #include <netinet/tcp_congctl.h> 148 #include <netinet/tcpip.h> 149 150 #ifdef IPSEC 151 #include <netipsec/ipsec.h> 152 #include <netipsec/xform.h> 153 #ifdef INET6 154 #include <netipsec/ipsec6.h> 155 #endif 156 #include <netipsec/key.h> 157 #endif /* IPSEC*/ 158 159 160 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 161 u_int32_t tcp_now; /* slow ticks, for RFC 1323 timestamps */ 162 163 percpu_t *tcpstat_percpu; 164 165 /* patchable/settable parameters for tcp */ 166 int tcp_mssdflt = TCP_MSS; 167 int tcp_minmss = TCP_MINMSS; 168 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 169 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 170 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 171 int tcp_do_sack = 1; /* selective acknowledgement */ 172 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 173 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 174 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 175 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 176 #ifndef TCP_INIT_WIN 177 #define TCP_INIT_WIN 4 /* initial slow start window */ 178 #endif 179 #ifndef TCP_INIT_WIN_LOCAL 180 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 181 #endif 182 /* 183 * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460. 184 * This is to simulate current behavior for iw == 4 185 */ 186 int tcp_init_win_max[] = { 187 1 * 1460, 188 1 * 1460, 189 2 * 1460, 190 2 * 1460, 191 3 * 1460, 192 5 * 1460, 193 6 * 1460, 194 7 * 1460, 195 8 * 1460, 196 9 * 1460, 197 10 * 1460 198 }; 199 int tcp_init_win = TCP_INIT_WIN; 200 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 201 int tcp_mss_ifmtu = 0; 202 #ifdef TCP_COMPAT_42 203 int tcp_compat_42 = 1; 204 #else 205 int tcp_compat_42 = 0; 206 #endif 207 int tcp_rst_ppslim = 100; /* 100pps */ 208 int tcp_ackdrop_ppslim = 100; /* 100pps */ 209 int tcp_do_loopback_cksum = 0; 210 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 211 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 212 int tcp_sack_tp_maxholes = 32; 213 int tcp_sack_globalmaxholes = 1024; 214 int tcp_sack_globalholes = 0; 215 int tcp_ecn_maxretries = 1; 216 int tcp_msl_enable = 1; /* enable TIME_WAIT truncation */ 217 int tcp_msl_loop = PR_SLOWHZ; /* MSL for loopback */ 218 int tcp_msl_local = 5 * PR_SLOWHZ; /* MSL for 'local' */ 219 int tcp_msl_remote = TCPTV_MSL; /* MSL otherwise */ 220 int tcp_msl_remote_threshold = TCPTV_SRTTDFLT; /* RTT threshold */ 221 int tcp_rttlocal = 0; /* Use RTT to decide who's 'local' */ 222 223 int tcp4_vtw_enable = 0; /* 1 to enable */ 224 int tcp6_vtw_enable = 0; /* 1 to enable */ 225 int tcp_vtw_was_enabled = 0; 226 int tcp_vtw_entries = 1 << 4; /* 16 vestigial TIME_WAIT entries */ 227 228 /* tcb hash */ 229 #ifndef TCBHASHSIZE 230 #define TCBHASHSIZE 128 231 #endif 232 int tcbhashsize = TCBHASHSIZE; 233 234 /* syn hash parameters */ 235 #define TCP_SYN_HASH_SIZE 293 236 #define TCP_SYN_BUCKET_SIZE 35 237 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 238 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 239 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 240 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 241 242 int tcp_freeq(struct tcpcb *); 243 244 #ifdef INET 245 static void tcp_mtudisc_callback(struct in_addr); 246 #endif 247 248 #ifdef INET6 249 void tcp6_mtudisc(struct in6pcb *, int); 250 #endif 251 252 static struct pool tcpcb_pool; 253 254 static int tcp_drainwanted; 255 256 #ifdef TCP_CSUM_COUNTERS 257 #include <sys/device.h> 258 259 #if defined(INET) 260 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 261 NULL, "tcp", "hwcsum bad"); 262 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 263 NULL, "tcp", "hwcsum ok"); 264 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 265 NULL, "tcp", "hwcsum data"); 266 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 NULL, "tcp", "swcsum"); 268 269 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 270 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 271 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 272 EVCNT_ATTACH_STATIC(tcp_swcsum); 273 #endif /* defined(INET) */ 274 275 #if defined(INET6) 276 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 277 NULL, "tcp6", "hwcsum bad"); 278 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 279 NULL, "tcp6", "hwcsum ok"); 280 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 281 NULL, "tcp6", "hwcsum data"); 282 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 283 NULL, "tcp6", "swcsum"); 284 285 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 286 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 287 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 288 EVCNT_ATTACH_STATIC(tcp6_swcsum); 289 #endif /* defined(INET6) */ 290 #endif /* TCP_CSUM_COUNTERS */ 291 292 293 #ifdef TCP_OUTPUT_COUNTERS 294 #include <sys/device.h> 295 296 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 297 NULL, "tcp", "output big header"); 298 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 299 NULL, "tcp", "output predict hit"); 300 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 301 NULL, "tcp", "output predict miss"); 302 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 303 NULL, "tcp", "output copy small"); 304 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 305 NULL, "tcp", "output copy big"); 306 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 307 NULL, "tcp", "output reference big"); 308 309 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 310 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 311 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 312 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 313 EVCNT_ATTACH_STATIC(tcp_output_copybig); 314 EVCNT_ATTACH_STATIC(tcp_output_refbig); 315 316 #endif /* TCP_OUTPUT_COUNTERS */ 317 318 #ifdef TCP_REASS_COUNTERS 319 #include <sys/device.h> 320 321 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 322 NULL, "tcp_reass", "calls"); 323 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 324 &tcp_reass_, "tcp_reass", "insert into empty queue"); 325 struct evcnt tcp_reass_iteration[8] = { 326 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 327 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 328 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 329 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 331 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 332 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 333 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 334 }; 335 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 336 &tcp_reass_, "tcp_reass", "prepend to first"); 337 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 338 &tcp_reass_, "tcp_reass", "prepend"); 339 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 340 &tcp_reass_, "tcp_reass", "insert"); 341 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 342 &tcp_reass_, "tcp_reass", "insert at tail"); 343 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 344 &tcp_reass_, "tcp_reass", "append"); 345 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 346 &tcp_reass_, "tcp_reass", "append to tail fragment"); 347 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 348 &tcp_reass_, "tcp_reass", "overlap at end"); 349 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 350 &tcp_reass_, "tcp_reass", "overlap at start"); 351 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 352 &tcp_reass_, "tcp_reass", "duplicate segment"); 353 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 354 &tcp_reass_, "tcp_reass", "duplicate fragment"); 355 356 EVCNT_ATTACH_STATIC(tcp_reass_); 357 EVCNT_ATTACH_STATIC(tcp_reass_empty); 358 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 359 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 360 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 363 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 364 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 365 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 366 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 367 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 368 EVCNT_ATTACH_STATIC(tcp_reass_insert); 369 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 370 EVCNT_ATTACH_STATIC(tcp_reass_append); 371 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 372 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 373 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 374 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 375 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 376 377 #endif /* TCP_REASS_COUNTERS */ 378 379 #ifdef MBUFTRACE 380 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 381 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 382 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 383 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 384 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 385 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 386 #endif 387 388 callout_t tcp_slowtimo_ch; 389 390 static int 391 do_tcpinit(void) 392 { 393 394 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 395 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 396 NULL, IPL_SOFTNET); 397 398 tcp_usrreq_init(); 399 400 /* Initialize timer state. */ 401 tcp_timer_init(); 402 403 /* Initialize the compressed state engine. */ 404 syn_cache_init(); 405 406 /* Initialize the congestion control algorithms. */ 407 tcp_congctl_init(); 408 409 /* Initialize the TCPCB template. */ 410 tcp_tcpcb_template(); 411 412 /* Initialize reassembly queue */ 413 tcpipqent_init(); 414 415 /* SACK */ 416 tcp_sack_init(); 417 418 MOWNER_ATTACH(&tcp_tx_mowner); 419 MOWNER_ATTACH(&tcp_rx_mowner); 420 MOWNER_ATTACH(&tcp_reass_mowner); 421 MOWNER_ATTACH(&tcp_sock_mowner); 422 MOWNER_ATTACH(&tcp_sock_tx_mowner); 423 MOWNER_ATTACH(&tcp_sock_rx_mowner); 424 MOWNER_ATTACH(&tcp_mowner); 425 426 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS); 427 428 vtw_earlyinit(); 429 430 callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE); 431 callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL); 432 433 return 0; 434 } 435 436 void 437 tcp_init_common(unsigned basehlen) 438 { 439 static ONCE_DECL(dotcpinit); 440 unsigned hlen = basehlen + sizeof(struct tcphdr); 441 unsigned oldhlen; 442 443 if (max_linkhdr + hlen > MHLEN) 444 panic("tcp_init"); 445 while ((oldhlen = max_protohdr) < hlen) 446 atomic_cas_uint(&max_protohdr, oldhlen, hlen); 447 448 RUN_ONCE(&dotcpinit, do_tcpinit); 449 } 450 451 /* 452 * Tcp initialization 453 */ 454 void 455 tcp_init(void) 456 { 457 458 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 459 460 tcp_init_common(sizeof(struct ip)); 461 } 462 463 /* 464 * Create template to be used to send tcp packets on a connection. 465 * Call after host entry created, allocates an mbuf and fills 466 * in a skeletal tcp/ip header, minimizing the amount of work 467 * necessary when the connection is used. 468 */ 469 struct mbuf * 470 tcp_template(struct tcpcb *tp) 471 { 472 struct inpcb *inp = tp->t_inpcb; 473 #ifdef INET6 474 struct in6pcb *in6p = tp->t_in6pcb; 475 #endif 476 struct tcphdr *n; 477 struct mbuf *m; 478 int hlen; 479 480 switch (tp->t_family) { 481 case AF_INET: 482 hlen = sizeof(struct ip); 483 if (inp) 484 break; 485 #ifdef INET6 486 if (in6p) { 487 /* mapped addr case */ 488 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 489 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 490 break; 491 } 492 #endif 493 return NULL; /*EINVAL*/ 494 #ifdef INET6 495 case AF_INET6: 496 hlen = sizeof(struct ip6_hdr); 497 if (in6p) { 498 /* more sainty check? */ 499 break; 500 } 501 return NULL; /*EINVAL*/ 502 #endif 503 default: 504 hlen = 0; /*pacify gcc*/ 505 return NULL; /*EAFNOSUPPORT*/ 506 } 507 #ifdef DIAGNOSTIC 508 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 509 panic("mclbytes too small for t_template"); 510 #endif 511 m = tp->t_template; 512 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 513 ; 514 else { 515 if (m) 516 m_freem(m); 517 m = tp->t_template = NULL; 518 MGETHDR(m, M_DONTWAIT, MT_HEADER); 519 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 520 MCLGET(m, M_DONTWAIT); 521 if ((m->m_flags & M_EXT) == 0) { 522 m_free(m); 523 m = NULL; 524 } 525 } 526 if (m == NULL) 527 return NULL; 528 MCLAIM(m, &tcp_mowner); 529 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 530 } 531 532 memset(mtod(m, void *), 0, m->m_len); 533 534 n = (struct tcphdr *)(mtod(m, char *) + hlen); 535 536 switch (tp->t_family) { 537 case AF_INET: 538 { 539 struct ipovly *ipov; 540 mtod(m, struct ip *)->ip_v = 4; 541 mtod(m, struct ip *)->ip_hl = hlen >> 2; 542 ipov = mtod(m, struct ipovly *); 543 ipov->ih_pr = IPPROTO_TCP; 544 ipov->ih_len = htons(sizeof(struct tcphdr)); 545 if (inp) { 546 ipov->ih_src = inp->inp_laddr; 547 ipov->ih_dst = inp->inp_faddr; 548 } 549 #ifdef INET6 550 else if (in6p) { 551 /* mapped addr case */ 552 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 553 sizeof(ipov->ih_src)); 554 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 555 sizeof(ipov->ih_dst)); 556 } 557 #endif 558 /* 559 * Compute the pseudo-header portion of the checksum 560 * now. We incrementally add in the TCP option and 561 * payload lengths later, and then compute the TCP 562 * checksum right before the packet is sent off onto 563 * the wire. 564 */ 565 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 566 ipov->ih_dst.s_addr, 567 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 568 break; 569 } 570 #ifdef INET6 571 case AF_INET6: 572 { 573 struct ip6_hdr *ip6; 574 mtod(m, struct ip *)->ip_v = 6; 575 ip6 = mtod(m, struct ip6_hdr *); 576 ip6->ip6_nxt = IPPROTO_TCP; 577 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 578 ip6->ip6_src = in6p->in6p_laddr; 579 ip6->ip6_dst = in6p->in6p_faddr; 580 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 581 if (ip6_auto_flowlabel) { 582 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 583 ip6->ip6_flow |= 584 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 585 } 586 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 587 ip6->ip6_vfc |= IPV6_VERSION; 588 589 /* 590 * Compute the pseudo-header portion of the checksum 591 * now. We incrementally add in the TCP option and 592 * payload lengths later, and then compute the TCP 593 * checksum right before the packet is sent off onto 594 * the wire. 595 */ 596 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 597 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 598 htonl(IPPROTO_TCP)); 599 break; 600 } 601 #endif 602 } 603 if (inp) { 604 n->th_sport = inp->inp_lport; 605 n->th_dport = inp->inp_fport; 606 } 607 #ifdef INET6 608 else if (in6p) { 609 n->th_sport = in6p->in6p_lport; 610 n->th_dport = in6p->in6p_fport; 611 } 612 #endif 613 n->th_seq = 0; 614 n->th_ack = 0; 615 n->th_x2 = 0; 616 n->th_off = 5; 617 n->th_flags = 0; 618 n->th_win = 0; 619 n->th_urp = 0; 620 return (m); 621 } 622 623 /* 624 * Send a single message to the TCP at address specified by 625 * the given TCP/IP header. If m == 0, then we make a copy 626 * of the tcpiphdr at ti and send directly to the addressed host. 627 * This is used to force keep alive messages out using the TCP 628 * template for a connection tp->t_template. If flags are given 629 * then we send a message back to the TCP which originated the 630 * segment ti, and discard the mbuf containing it and any other 631 * attached mbufs. 632 * 633 * In any case the ack and sequence number of the transmitted 634 * segment are as specified by the parameters. 635 */ 636 int 637 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m, 638 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 639 { 640 struct route *ro; 641 int error, tlen, win = 0; 642 int hlen; 643 struct ip *ip; 644 #ifdef INET6 645 struct ip6_hdr *ip6; 646 #endif 647 int family; /* family on packet, not inpcb/in6pcb! */ 648 struct tcphdr *th; 649 struct socket *so; 650 651 if (tp != NULL && (flags & TH_RST) == 0) { 652 #ifdef DIAGNOSTIC 653 if (tp->t_inpcb && tp->t_in6pcb) 654 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 655 #endif 656 #ifdef INET 657 if (tp->t_inpcb) 658 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 659 #endif 660 #ifdef INET6 661 if (tp->t_in6pcb) 662 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 663 #endif 664 } 665 666 th = NULL; /* Quell uninitialized warning */ 667 ip = NULL; 668 #ifdef INET6 669 ip6 = NULL; 670 #endif 671 if (m == 0) { 672 if (!mtemplate) 673 return EINVAL; 674 675 /* get family information from template */ 676 switch (mtod(mtemplate, struct ip *)->ip_v) { 677 case 4: 678 family = AF_INET; 679 hlen = sizeof(struct ip); 680 break; 681 #ifdef INET6 682 case 6: 683 family = AF_INET6; 684 hlen = sizeof(struct ip6_hdr); 685 break; 686 #endif 687 default: 688 return EAFNOSUPPORT; 689 } 690 691 MGETHDR(m, M_DONTWAIT, MT_HEADER); 692 if (m) { 693 MCLAIM(m, &tcp_tx_mowner); 694 MCLGET(m, M_DONTWAIT); 695 if ((m->m_flags & M_EXT) == 0) { 696 m_free(m); 697 m = NULL; 698 } 699 } 700 if (m == NULL) 701 return (ENOBUFS); 702 703 if (tcp_compat_42) 704 tlen = 1; 705 else 706 tlen = 0; 707 708 m->m_data += max_linkhdr; 709 bcopy(mtod(mtemplate, void *), mtod(m, void *), 710 mtemplate->m_len); 711 switch (family) { 712 case AF_INET: 713 ip = mtod(m, struct ip *); 714 th = (struct tcphdr *)(ip + 1); 715 break; 716 #ifdef INET6 717 case AF_INET6: 718 ip6 = mtod(m, struct ip6_hdr *); 719 th = (struct tcphdr *)(ip6 + 1); 720 break; 721 #endif 722 #if 0 723 default: 724 /* noone will visit here */ 725 m_freem(m); 726 return EAFNOSUPPORT; 727 #endif 728 } 729 flags = TH_ACK; 730 } else { 731 732 if ((m->m_flags & M_PKTHDR) == 0) { 733 #if 0 734 printf("non PKTHDR to tcp_respond\n"); 735 #endif 736 m_freem(m); 737 return EINVAL; 738 } 739 #ifdef DIAGNOSTIC 740 if (!th0) 741 panic("th0 == NULL in tcp_respond"); 742 #endif 743 744 /* get family information from m */ 745 switch (mtod(m, struct ip *)->ip_v) { 746 case 4: 747 family = AF_INET; 748 hlen = sizeof(struct ip); 749 ip = mtod(m, struct ip *); 750 break; 751 #ifdef INET6 752 case 6: 753 family = AF_INET6; 754 hlen = sizeof(struct ip6_hdr); 755 ip6 = mtod(m, struct ip6_hdr *); 756 break; 757 #endif 758 default: 759 m_freem(m); 760 return EAFNOSUPPORT; 761 } 762 /* clear h/w csum flags inherited from rx packet */ 763 m->m_pkthdr.csum_flags = 0; 764 765 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 766 tlen = sizeof(*th0); 767 else 768 tlen = th0->th_off << 2; 769 770 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 771 mtod(m, char *) + hlen == (char *)th0) { 772 m->m_len = hlen + tlen; 773 m_freem(m->m_next); 774 m->m_next = NULL; 775 } else { 776 struct mbuf *n; 777 778 #ifdef DIAGNOSTIC 779 if (max_linkhdr + hlen + tlen > MCLBYTES) { 780 m_freem(m); 781 return EMSGSIZE; 782 } 783 #endif 784 MGETHDR(n, M_DONTWAIT, MT_HEADER); 785 if (n && max_linkhdr + hlen + tlen > MHLEN) { 786 MCLGET(n, M_DONTWAIT); 787 if ((n->m_flags & M_EXT) == 0) { 788 m_freem(n); 789 n = NULL; 790 } 791 } 792 if (!n) { 793 m_freem(m); 794 return ENOBUFS; 795 } 796 797 MCLAIM(n, &tcp_tx_mowner); 798 n->m_data += max_linkhdr; 799 n->m_len = hlen + tlen; 800 m_copyback(n, 0, hlen, mtod(m, void *)); 801 m_copyback(n, hlen, tlen, (void *)th0); 802 803 m_freem(m); 804 m = n; 805 n = NULL; 806 } 807 808 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 809 switch (family) { 810 case AF_INET: 811 ip = mtod(m, struct ip *); 812 th = (struct tcphdr *)(ip + 1); 813 ip->ip_p = IPPROTO_TCP; 814 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 815 ip->ip_p = IPPROTO_TCP; 816 break; 817 #ifdef INET6 818 case AF_INET6: 819 ip6 = mtod(m, struct ip6_hdr *); 820 th = (struct tcphdr *)(ip6 + 1); 821 ip6->ip6_nxt = IPPROTO_TCP; 822 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 823 ip6->ip6_nxt = IPPROTO_TCP; 824 break; 825 #endif 826 #if 0 827 default: 828 /* noone will visit here */ 829 m_freem(m); 830 return EAFNOSUPPORT; 831 #endif 832 } 833 xchg(th->th_dport, th->th_sport, u_int16_t); 834 #undef xchg 835 tlen = 0; /*be friendly with the following code*/ 836 } 837 th->th_seq = htonl(seq); 838 th->th_ack = htonl(ack); 839 th->th_x2 = 0; 840 if ((flags & TH_SYN) == 0) { 841 if (tp) 842 win >>= tp->rcv_scale; 843 if (win > TCP_MAXWIN) 844 win = TCP_MAXWIN; 845 th->th_win = htons((u_int16_t)win); 846 th->th_off = sizeof (struct tcphdr) >> 2; 847 tlen += sizeof(*th); 848 } else 849 tlen += th->th_off << 2; 850 m->m_len = hlen + tlen; 851 m->m_pkthdr.len = hlen + tlen; 852 m->m_pkthdr.rcvif = NULL; 853 th->th_flags = flags; 854 th->th_urp = 0; 855 856 switch (family) { 857 #ifdef INET 858 case AF_INET: 859 { 860 struct ipovly *ipov = (struct ipovly *)ip; 861 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1); 862 ipov->ih_len = htons((u_int16_t)tlen); 863 864 th->th_sum = 0; 865 th->th_sum = in_cksum(m, hlen + tlen); 866 ip->ip_len = htons(hlen + tlen); 867 ip->ip_ttl = ip_defttl; 868 break; 869 } 870 #endif 871 #ifdef INET6 872 case AF_INET6: 873 { 874 th->th_sum = 0; 875 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 876 tlen); 877 ip6->ip6_plen = htons(tlen); 878 if (tp && tp->t_in6pcb) 879 ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb); 880 else 881 ip6->ip6_hlim = ip6_defhlim; 882 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 883 if (ip6_auto_flowlabel) { 884 ip6->ip6_flow |= 885 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 886 } 887 break; 888 } 889 #endif 890 } 891 892 if (tp && tp->t_inpcb) 893 so = tp->t_inpcb->inp_socket; 894 #ifdef INET6 895 else if (tp && tp->t_in6pcb) 896 so = tp->t_in6pcb->in6p_socket; 897 #endif 898 else 899 so = NULL; 900 901 if (tp != NULL && tp->t_inpcb != NULL) { 902 ro = &tp->t_inpcb->inp_route; 903 #ifdef DIAGNOSTIC 904 if (family != AF_INET) 905 panic("tcp_respond: address family mismatch"); 906 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 907 panic("tcp_respond: ip_dst %x != inp_faddr %x", 908 ntohl(ip->ip_dst.s_addr), 909 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 910 } 911 #endif 912 } 913 #ifdef INET6 914 else if (tp != NULL && tp->t_in6pcb != NULL) { 915 ro = (struct route *)&tp->t_in6pcb->in6p_route; 916 #ifdef DIAGNOSTIC 917 if (family == AF_INET) { 918 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 919 panic("tcp_respond: not mapped addr"); 920 if (memcmp(&ip->ip_dst, 921 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 922 sizeof(ip->ip_dst)) != 0) { 923 panic("tcp_respond: ip_dst != in6p_faddr"); 924 } 925 } else if (family == AF_INET6) { 926 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 927 &tp->t_in6pcb->in6p_faddr)) 928 panic("tcp_respond: ip6_dst != in6p_faddr"); 929 } else 930 panic("tcp_respond: address family mismatch"); 931 #endif 932 } 933 #endif 934 else 935 ro = NULL; 936 937 switch (family) { 938 #ifdef INET 939 case AF_INET: 940 error = ip_output(m, NULL, ro, 941 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, so); 942 break; 943 #endif 944 #ifdef INET6 945 case AF_INET6: 946 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL); 947 break; 948 #endif 949 default: 950 error = EAFNOSUPPORT; 951 break; 952 } 953 954 return (error); 955 } 956 957 /* 958 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 959 * a bunch of members individually, we maintain this template for the 960 * static and mostly-static components of the TCPCB, and copy it into 961 * the new TCPCB instead. 962 */ 963 static struct tcpcb tcpcb_template = { 964 .t_srtt = TCPTV_SRTTBASE, 965 .t_rttmin = TCPTV_MIN, 966 967 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 968 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 969 .snd_numholes = 0, 970 .snd_cubic_wmax = 0, 971 .snd_cubic_wmax_last = 0, 972 .snd_cubic_ctime = 0, 973 974 .t_partialacks = -1, 975 .t_bytes_acked = 0, 976 .t_sndrexmitpack = 0, 977 .t_rcvoopack = 0, 978 .t_sndzerowin = 0, 979 }; 980 981 /* 982 * Updates the TCPCB template whenever a parameter that would affect 983 * the template is changed. 984 */ 985 void 986 tcp_tcpcb_template(void) 987 { 988 struct tcpcb *tp = &tcpcb_template; 989 int flags; 990 991 tp->t_peermss = tcp_mssdflt; 992 tp->t_ourmss = tcp_mssdflt; 993 tp->t_segsz = tcp_mssdflt; 994 995 flags = 0; 996 if (tcp_do_rfc1323 && tcp_do_win_scale) 997 flags |= TF_REQ_SCALE; 998 if (tcp_do_rfc1323 && tcp_do_timestamps) 999 flags |= TF_REQ_TSTMP; 1000 tp->t_flags = flags; 1001 1002 /* 1003 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1004 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 1005 * reasonable initial retransmit time. 1006 */ 1007 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 1008 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1009 TCPTV_MIN, TCPTV_REXMTMAX); 1010 1011 /* Keep Alive */ 1012 tp->t_keepinit = tcp_keepinit; 1013 tp->t_keepidle = tcp_keepidle; 1014 tp->t_keepintvl = tcp_keepintvl; 1015 tp->t_keepcnt = tcp_keepcnt; 1016 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 1017 1018 /* MSL */ 1019 tp->t_msl = TCPTV_MSL; 1020 } 1021 1022 /* 1023 * Create a new TCP control block, making an 1024 * empty reassembly queue and hooking it to the argument 1025 * protocol control block. 1026 */ 1027 /* family selects inpcb, or in6pcb */ 1028 struct tcpcb * 1029 tcp_newtcpcb(int family, void *aux) 1030 { 1031 struct tcpcb *tp; 1032 int i; 1033 1034 /* XXX Consider using a pool_cache for speed. */ 1035 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 1036 if (tp == NULL) 1037 return (NULL); 1038 memcpy(tp, &tcpcb_template, sizeof(*tp)); 1039 TAILQ_INIT(&tp->segq); 1040 TAILQ_INIT(&tp->timeq); 1041 tp->t_family = family; /* may be overridden later on */ 1042 TAILQ_INIT(&tp->snd_holes); 1043 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1044 1045 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1046 for (i = 0; i < TCPT_NTIMERS; i++) { 1047 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 1048 TCP_TIMER_INIT(tp, i); 1049 } 1050 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 1051 1052 switch (family) { 1053 case AF_INET: 1054 { 1055 struct inpcb *inp = (struct inpcb *)aux; 1056 1057 inp->inp_ip.ip_ttl = ip_defttl; 1058 inp->inp_ppcb = (void *)tp; 1059 1060 tp->t_inpcb = inp; 1061 tp->t_mtudisc = ip_mtudisc; 1062 break; 1063 } 1064 #ifdef INET6 1065 case AF_INET6: 1066 { 1067 struct in6pcb *in6p = (struct in6pcb *)aux; 1068 1069 in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p); 1070 in6p->in6p_ppcb = (void *)tp; 1071 1072 tp->t_in6pcb = in6p; 1073 /* for IPv6, always try to run path MTU discovery */ 1074 tp->t_mtudisc = 1; 1075 break; 1076 } 1077 #endif /* INET6 */ 1078 default: 1079 for (i = 0; i < TCPT_NTIMERS; i++) 1080 callout_destroy(&tp->t_timer[i]); 1081 callout_destroy(&tp->t_delack_ch); 1082 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1083 return (NULL); 1084 } 1085 1086 /* 1087 * Initialize our timebase. When we send timestamps, we take 1088 * the delta from tcp_now -- this means each connection always 1089 * gets a timebase of 1, which makes it, among other things, 1090 * more difficult to determine how long a system has been up, 1091 * and thus how many TCP sequence increments have occurred. 1092 * 1093 * We start with 1, because 0 doesn't work with linux, which 1094 * considers timestamp 0 in a SYN packet as a bug and disables 1095 * timestamps. 1096 */ 1097 tp->ts_timebase = tcp_now - 1; 1098 1099 tcp_congctl_select(tp, tcp_congctl_global_name); 1100 1101 return (tp); 1102 } 1103 1104 /* 1105 * Drop a TCP connection, reporting 1106 * the specified error. If connection is synchronized, 1107 * then send a RST to peer. 1108 */ 1109 struct tcpcb * 1110 tcp_drop(struct tcpcb *tp, int errno) 1111 { 1112 struct socket *so = NULL; 1113 1114 #ifdef DIAGNOSTIC 1115 if (tp->t_inpcb && tp->t_in6pcb) 1116 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1117 #endif 1118 #ifdef INET 1119 if (tp->t_inpcb) 1120 so = tp->t_inpcb->inp_socket; 1121 #endif 1122 #ifdef INET6 1123 if (tp->t_in6pcb) 1124 so = tp->t_in6pcb->in6p_socket; 1125 #endif 1126 if (!so) 1127 return NULL; 1128 1129 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1130 tp->t_state = TCPS_CLOSED; 1131 (void) tcp_output(tp); 1132 TCP_STATINC(TCP_STAT_DROPS); 1133 } else 1134 TCP_STATINC(TCP_STAT_CONNDROPS); 1135 if (errno == ETIMEDOUT && tp->t_softerror) 1136 errno = tp->t_softerror; 1137 so->so_error = errno; 1138 return (tcp_close(tp)); 1139 } 1140 1141 /* 1142 * Close a TCP control block: 1143 * discard all space held by the tcp 1144 * discard internet protocol block 1145 * wake up any sleepers 1146 */ 1147 struct tcpcb * 1148 tcp_close(struct tcpcb *tp) 1149 { 1150 struct inpcb *inp; 1151 #ifdef INET6 1152 struct in6pcb *in6p; 1153 #endif 1154 struct socket *so; 1155 #ifdef RTV_RTT 1156 struct rtentry *rt; 1157 #endif 1158 struct route *ro; 1159 int j; 1160 1161 inp = tp->t_inpcb; 1162 #ifdef INET6 1163 in6p = tp->t_in6pcb; 1164 #endif 1165 so = NULL; 1166 ro = NULL; 1167 if (inp) { 1168 so = inp->inp_socket; 1169 ro = &inp->inp_route; 1170 } 1171 #ifdef INET6 1172 else if (in6p) { 1173 so = in6p->in6p_socket; 1174 ro = (struct route *)&in6p->in6p_route; 1175 } 1176 #endif 1177 1178 #ifdef RTV_RTT 1179 /* 1180 * If we sent enough data to get some meaningful characteristics, 1181 * save them in the routing entry. 'Enough' is arbitrarily 1182 * defined as the sendpipesize (default 4K) * 16. This would 1183 * give us 16 rtt samples assuming we only get one sample per 1184 * window (the usual case on a long haul net). 16 samples is 1185 * enough for the srtt filter to converge to within 5% of the correct 1186 * value; fewer samples and we could save a very bogus rtt. 1187 * 1188 * Don't update the default route's characteristics and don't 1189 * update anything that the user "locked". 1190 */ 1191 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1192 ro && (rt = rtcache_validate(ro)) != NULL && 1193 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1194 u_long i = 0; 1195 1196 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1197 i = tp->t_srtt * 1198 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1199 if (rt->rt_rmx.rmx_rtt && i) 1200 /* 1201 * filter this update to half the old & half 1202 * the new values, converting scale. 1203 * See route.h and tcp_var.h for a 1204 * description of the scaling constants. 1205 */ 1206 rt->rt_rmx.rmx_rtt = 1207 (rt->rt_rmx.rmx_rtt + i) / 2; 1208 else 1209 rt->rt_rmx.rmx_rtt = i; 1210 } 1211 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1212 i = tp->t_rttvar * 1213 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1214 if (rt->rt_rmx.rmx_rttvar && i) 1215 rt->rt_rmx.rmx_rttvar = 1216 (rt->rt_rmx.rmx_rttvar + i) / 2; 1217 else 1218 rt->rt_rmx.rmx_rttvar = i; 1219 } 1220 /* 1221 * update the pipelimit (ssthresh) if it has been updated 1222 * already or if a pipesize was specified & the threshhold 1223 * got below half the pipesize. I.e., wait for bad news 1224 * before we start updating, then update on both good 1225 * and bad news. 1226 */ 1227 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1228 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1229 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1230 /* 1231 * convert the limit from user data bytes to 1232 * packets then to packet data bytes. 1233 */ 1234 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1235 if (i < 2) 1236 i = 2; 1237 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1238 if (rt->rt_rmx.rmx_ssthresh) 1239 rt->rt_rmx.rmx_ssthresh = 1240 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1241 else 1242 rt->rt_rmx.rmx_ssthresh = i; 1243 } 1244 } 1245 #endif /* RTV_RTT */ 1246 /* free the reassembly queue, if any */ 1247 TCP_REASS_LOCK(tp); 1248 (void) tcp_freeq(tp); 1249 TCP_REASS_UNLOCK(tp); 1250 1251 /* free the SACK holes list. */ 1252 tcp_free_sackholes(tp); 1253 tcp_congctl_release(tp); 1254 syn_cache_cleanup(tp); 1255 1256 if (tp->t_template) { 1257 m_free(tp->t_template); 1258 tp->t_template = NULL; 1259 } 1260 1261 /* 1262 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1263 * the timers can also drop the lock. We need to prevent access 1264 * to the tcpcb as it's half torn down. Flag the pcb as dead 1265 * (prevents access by timers) and only then detach it. 1266 */ 1267 tp->t_flags |= TF_DEAD; 1268 if (inp) { 1269 inp->inp_ppcb = 0; 1270 soisdisconnected(so); 1271 in_pcbdetach(inp); 1272 } 1273 #ifdef INET6 1274 else if (in6p) { 1275 in6p->in6p_ppcb = 0; 1276 soisdisconnected(so); 1277 in6_pcbdetach(in6p); 1278 } 1279 #endif 1280 /* 1281 * pcb is no longer visble elsewhere, so we can safely release 1282 * the lock in callout_halt() if needed. 1283 */ 1284 TCP_STATINC(TCP_STAT_CLOSED); 1285 for (j = 0; j < TCPT_NTIMERS; j++) { 1286 callout_halt(&tp->t_timer[j], softnet_lock); 1287 callout_destroy(&tp->t_timer[j]); 1288 } 1289 callout_halt(&tp->t_delack_ch, softnet_lock); 1290 callout_destroy(&tp->t_delack_ch); 1291 pool_put(&tcpcb_pool, tp); 1292 1293 return NULL; 1294 } 1295 1296 int 1297 tcp_freeq(struct tcpcb *tp) 1298 { 1299 struct ipqent *qe; 1300 int rv = 0; 1301 #ifdef TCPREASS_DEBUG 1302 int i = 0; 1303 #endif 1304 1305 TCP_REASS_LOCK_CHECK(tp); 1306 1307 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1308 #ifdef TCPREASS_DEBUG 1309 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1310 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1311 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1312 #endif 1313 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1314 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1315 m_freem(qe->ipqe_m); 1316 tcpipqent_free(qe); 1317 rv = 1; 1318 } 1319 tp->t_segqlen = 0; 1320 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1321 return (rv); 1322 } 1323 1324 void 1325 tcp_fasttimo(void) 1326 { 1327 if (tcp_drainwanted) { 1328 tcp_drain(); 1329 tcp_drainwanted = 0; 1330 } 1331 } 1332 1333 void 1334 tcp_drainstub(void) 1335 { 1336 tcp_drainwanted = 1; 1337 } 1338 1339 /* 1340 * Protocol drain routine. Called when memory is in short supply. 1341 * Called from pr_fasttimo thus a callout context. 1342 */ 1343 void 1344 tcp_drain(void) 1345 { 1346 struct inpcb_hdr *inph; 1347 struct tcpcb *tp; 1348 1349 mutex_enter(softnet_lock); 1350 KERNEL_LOCK(1, NULL); 1351 1352 /* 1353 * Free the sequence queue of all TCP connections. 1354 */ 1355 TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1356 switch (inph->inph_af) { 1357 case AF_INET: 1358 tp = intotcpcb((struct inpcb *)inph); 1359 break; 1360 #ifdef INET6 1361 case AF_INET6: 1362 tp = in6totcpcb((struct in6pcb *)inph); 1363 break; 1364 #endif 1365 default: 1366 tp = NULL; 1367 break; 1368 } 1369 if (tp != NULL) { 1370 /* 1371 * We may be called from a device's interrupt 1372 * context. If the tcpcb is already busy, 1373 * just bail out now. 1374 */ 1375 if (tcp_reass_lock_try(tp) == 0) 1376 continue; 1377 if (tcp_freeq(tp)) 1378 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1379 TCP_REASS_UNLOCK(tp); 1380 } 1381 } 1382 1383 KERNEL_UNLOCK_ONE(NULL); 1384 mutex_exit(softnet_lock); 1385 } 1386 1387 /* 1388 * Notify a tcp user of an asynchronous error; 1389 * store error as soft error, but wake up user 1390 * (for now, won't do anything until can select for soft error). 1391 */ 1392 void 1393 tcp_notify(struct inpcb *inp, int error) 1394 { 1395 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1396 struct socket *so = inp->inp_socket; 1397 1398 /* 1399 * Ignore some errors if we are hooked up. 1400 * If connection hasn't completed, has retransmitted several times, 1401 * and receives a second error, give up now. This is better 1402 * than waiting a long time to establish a connection that 1403 * can never complete. 1404 */ 1405 if (tp->t_state == TCPS_ESTABLISHED && 1406 (error == EHOSTUNREACH || error == ENETUNREACH || 1407 error == EHOSTDOWN)) { 1408 return; 1409 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1410 tp->t_rxtshift > 3 && tp->t_softerror) 1411 so->so_error = error; 1412 else 1413 tp->t_softerror = error; 1414 cv_broadcast(&so->so_cv); 1415 sorwakeup(so); 1416 sowwakeup(so); 1417 } 1418 1419 #ifdef INET6 1420 void 1421 tcp6_notify(struct in6pcb *in6p, int error) 1422 { 1423 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1424 struct socket *so = in6p->in6p_socket; 1425 1426 /* 1427 * Ignore some errors if we are hooked up. 1428 * If connection hasn't completed, has retransmitted several times, 1429 * and receives a second error, give up now. This is better 1430 * than waiting a long time to establish a connection that 1431 * can never complete. 1432 */ 1433 if (tp->t_state == TCPS_ESTABLISHED && 1434 (error == EHOSTUNREACH || error == ENETUNREACH || 1435 error == EHOSTDOWN)) { 1436 return; 1437 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1438 tp->t_rxtshift > 3 && tp->t_softerror) 1439 so->so_error = error; 1440 else 1441 tp->t_softerror = error; 1442 cv_broadcast(&so->so_cv); 1443 sorwakeup(so); 1444 sowwakeup(so); 1445 } 1446 #endif 1447 1448 #ifdef INET6 1449 void * 1450 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1451 { 1452 struct tcphdr th; 1453 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1454 int nmatch; 1455 struct ip6_hdr *ip6; 1456 const struct sockaddr_in6 *sa6_src = NULL; 1457 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1458 struct mbuf *m; 1459 int off; 1460 1461 if (sa->sa_family != AF_INET6 || 1462 sa->sa_len != sizeof(struct sockaddr_in6)) 1463 return NULL; 1464 if ((unsigned)cmd >= PRC_NCMDS) 1465 return NULL; 1466 else if (cmd == PRC_QUENCH) { 1467 /* 1468 * Don't honor ICMP Source Quench messages meant for 1469 * TCP connections. 1470 */ 1471 return NULL; 1472 } else if (PRC_IS_REDIRECT(cmd)) 1473 notify = in6_rtchange, d = NULL; 1474 else if (cmd == PRC_MSGSIZE) 1475 ; /* special code is present, see below */ 1476 else if (cmd == PRC_HOSTDEAD) 1477 d = NULL; 1478 else if (inet6ctlerrmap[cmd] == 0) 1479 return NULL; 1480 1481 /* if the parameter is from icmp6, decode it. */ 1482 if (d != NULL) { 1483 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1484 m = ip6cp->ip6c_m; 1485 ip6 = ip6cp->ip6c_ip6; 1486 off = ip6cp->ip6c_off; 1487 sa6_src = ip6cp->ip6c_src; 1488 } else { 1489 m = NULL; 1490 ip6 = NULL; 1491 sa6_src = &sa6_any; 1492 off = 0; 1493 } 1494 1495 if (ip6) { 1496 /* 1497 * XXX: We assume that when ip6 is non NULL, 1498 * M and OFF are valid. 1499 */ 1500 1501 /* check if we can safely examine src and dst ports */ 1502 if (m->m_pkthdr.len < off + sizeof(th)) { 1503 if (cmd == PRC_MSGSIZE) 1504 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1505 return NULL; 1506 } 1507 1508 memset(&th, 0, sizeof(th)); 1509 m_copydata(m, off, sizeof(th), (void *)&th); 1510 1511 if (cmd == PRC_MSGSIZE) { 1512 int valid = 0; 1513 1514 /* 1515 * Check to see if we have a valid TCP connection 1516 * corresponding to the address in the ICMPv6 message 1517 * payload. 1518 */ 1519 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1520 th.th_dport, 1521 (const struct in6_addr *)&sa6_src->sin6_addr, 1522 th.th_sport, 0, 0)) 1523 valid++; 1524 1525 /* 1526 * Depending on the value of "valid" and routing table 1527 * size (mtudisc_{hi,lo}wat), we will: 1528 * - recalcurate the new MTU and create the 1529 * corresponding routing entry, or 1530 * - ignore the MTU change notification. 1531 */ 1532 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1533 1534 /* 1535 * no need to call in6_pcbnotify, it should have been 1536 * called via callback if necessary 1537 */ 1538 return NULL; 1539 } 1540 1541 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1542 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1543 if (nmatch == 0 && syn_cache_count && 1544 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1545 inet6ctlerrmap[cmd] == ENETUNREACH || 1546 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1547 syn_cache_unreach((const struct sockaddr *)sa6_src, 1548 sa, &th); 1549 } else { 1550 (void) in6_pcbnotify(&tcbtable, sa, 0, 1551 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1552 } 1553 1554 return NULL; 1555 } 1556 #endif 1557 1558 #ifdef INET 1559 /* assumes that ip header and tcp header are contiguous on mbuf */ 1560 void * 1561 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1562 { 1563 struct ip *ip = v; 1564 struct tcphdr *th; 1565 struct icmp *icp; 1566 extern const int inetctlerrmap[]; 1567 void (*notify)(struct inpcb *, int) = tcp_notify; 1568 int errno; 1569 int nmatch; 1570 struct tcpcb *tp; 1571 u_int mtu; 1572 tcp_seq seq; 1573 struct inpcb *inp; 1574 #ifdef INET6 1575 struct in6pcb *in6p; 1576 struct in6_addr src6, dst6; 1577 #endif 1578 1579 if (sa->sa_family != AF_INET || 1580 sa->sa_len != sizeof(struct sockaddr_in)) 1581 return NULL; 1582 if ((unsigned)cmd >= PRC_NCMDS) 1583 return NULL; 1584 errno = inetctlerrmap[cmd]; 1585 if (cmd == PRC_QUENCH) 1586 /* 1587 * Don't honor ICMP Source Quench messages meant for 1588 * TCP connections. 1589 */ 1590 return NULL; 1591 else if (PRC_IS_REDIRECT(cmd)) 1592 notify = in_rtchange, ip = 0; 1593 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1594 /* 1595 * Check to see if we have a valid TCP connection 1596 * corresponding to the address in the ICMP message 1597 * payload. 1598 * 1599 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1600 */ 1601 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1602 #ifdef INET6 1603 memset(&src6, 0, sizeof(src6)); 1604 memset(&dst6, 0, sizeof(dst6)); 1605 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1606 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1607 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1608 #endif 1609 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1610 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL) 1611 #ifdef INET6 1612 in6p = NULL; 1613 #else 1614 ; 1615 #endif 1616 #ifdef INET6 1617 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1618 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL) 1619 ; 1620 #endif 1621 else 1622 return NULL; 1623 1624 /* 1625 * Now that we've validated that we are actually communicating 1626 * with the host indicated in the ICMP message, locate the 1627 * ICMP header, recalculate the new MTU, and create the 1628 * corresponding routing entry. 1629 */ 1630 icp = (struct icmp *)((char *)ip - 1631 offsetof(struct icmp, icmp_ip)); 1632 if (inp) { 1633 if ((tp = intotcpcb(inp)) == NULL) 1634 return NULL; 1635 } 1636 #ifdef INET6 1637 else if (in6p) { 1638 if ((tp = in6totcpcb(in6p)) == NULL) 1639 return NULL; 1640 } 1641 #endif 1642 else 1643 return NULL; 1644 seq = ntohl(th->th_seq); 1645 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1646 return NULL; 1647 /* 1648 * If the ICMP message advertises a Next-Hop MTU 1649 * equal or larger than the maximum packet size we have 1650 * ever sent, drop the message. 1651 */ 1652 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1653 if (mtu >= tp->t_pmtud_mtu_sent) 1654 return NULL; 1655 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1656 /* 1657 * Calculate new MTU, and create corresponding 1658 * route (traditional PMTUD). 1659 */ 1660 tp->t_flags &= ~TF_PMTUD_PEND; 1661 icmp_mtudisc(icp, ip->ip_dst); 1662 } else { 1663 /* 1664 * Record the information got in the ICMP 1665 * message; act on it later. 1666 * If we had already recorded an ICMP message, 1667 * replace the old one only if the new message 1668 * refers to an older TCP segment 1669 */ 1670 if (tp->t_flags & TF_PMTUD_PEND) { 1671 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1672 return NULL; 1673 } else 1674 tp->t_flags |= TF_PMTUD_PEND; 1675 tp->t_pmtud_th_seq = seq; 1676 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1677 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1678 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1679 } 1680 return NULL; 1681 } else if (cmd == PRC_HOSTDEAD) 1682 ip = 0; 1683 else if (errno == 0) 1684 return NULL; 1685 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1686 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1687 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1688 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1689 if (nmatch == 0 && syn_cache_count && 1690 (inetctlerrmap[cmd] == EHOSTUNREACH || 1691 inetctlerrmap[cmd] == ENETUNREACH || 1692 inetctlerrmap[cmd] == EHOSTDOWN)) { 1693 struct sockaddr_in sin; 1694 memset(&sin, 0, sizeof(sin)); 1695 sin.sin_len = sizeof(sin); 1696 sin.sin_family = AF_INET; 1697 sin.sin_port = th->th_sport; 1698 sin.sin_addr = ip->ip_src; 1699 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1700 } 1701 1702 /* XXX mapped address case */ 1703 } else 1704 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1705 notify); 1706 return NULL; 1707 } 1708 1709 /* 1710 * When a source quench is received, we are being notified of congestion. 1711 * Close the congestion window down to the Loss Window (one segment). 1712 * We will gradually open it again as we proceed. 1713 */ 1714 void 1715 tcp_quench(struct inpcb *inp, int errno) 1716 { 1717 struct tcpcb *tp = intotcpcb(inp); 1718 1719 if (tp) { 1720 tp->snd_cwnd = tp->t_segsz; 1721 tp->t_bytes_acked = 0; 1722 } 1723 } 1724 #endif 1725 1726 #ifdef INET6 1727 void 1728 tcp6_quench(struct in6pcb *in6p, int errno) 1729 { 1730 struct tcpcb *tp = in6totcpcb(in6p); 1731 1732 if (tp) { 1733 tp->snd_cwnd = tp->t_segsz; 1734 tp->t_bytes_acked = 0; 1735 } 1736 } 1737 #endif 1738 1739 #ifdef INET 1740 /* 1741 * Path MTU Discovery handlers. 1742 */ 1743 void 1744 tcp_mtudisc_callback(struct in_addr faddr) 1745 { 1746 #ifdef INET6 1747 struct in6_addr in6; 1748 #endif 1749 1750 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1751 #ifdef INET6 1752 memset(&in6, 0, sizeof(in6)); 1753 in6.s6_addr16[5] = 0xffff; 1754 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1755 tcp6_mtudisc_callback(&in6); 1756 #endif 1757 } 1758 1759 /* 1760 * On receipt of path MTU corrections, flush old route and replace it 1761 * with the new one. Retransmit all unacknowledged packets, to ensure 1762 * that all packets will be received. 1763 */ 1764 void 1765 tcp_mtudisc(struct inpcb *inp, int errno) 1766 { 1767 struct tcpcb *tp = intotcpcb(inp); 1768 struct rtentry *rt = in_pcbrtentry(inp); 1769 1770 if (tp != 0) { 1771 if (rt != 0) { 1772 /* 1773 * If this was not a host route, remove and realloc. 1774 */ 1775 if ((rt->rt_flags & RTF_HOST) == 0) { 1776 in_rtchange(inp, errno); 1777 if ((rt = in_pcbrtentry(inp)) == 0) 1778 return; 1779 } 1780 1781 /* 1782 * Slow start out of the error condition. We 1783 * use the MTU because we know it's smaller 1784 * than the previously transmitted segment. 1785 * 1786 * Note: This is more conservative than the 1787 * suggestion in draft-floyd-incr-init-win-03. 1788 */ 1789 if (rt->rt_rmx.rmx_mtu != 0) 1790 tp->snd_cwnd = 1791 TCP_INITIAL_WINDOW(tcp_init_win, 1792 rt->rt_rmx.rmx_mtu); 1793 } 1794 1795 /* 1796 * Resend unacknowledged packets. 1797 */ 1798 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1799 tcp_output(tp); 1800 } 1801 } 1802 #endif 1803 1804 #ifdef INET6 1805 /* 1806 * Path MTU Discovery handlers. 1807 */ 1808 void 1809 tcp6_mtudisc_callback(struct in6_addr *faddr) 1810 { 1811 struct sockaddr_in6 sin6; 1812 1813 memset(&sin6, 0, sizeof(sin6)); 1814 sin6.sin6_family = AF_INET6; 1815 sin6.sin6_len = sizeof(struct sockaddr_in6); 1816 sin6.sin6_addr = *faddr; 1817 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1818 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1819 } 1820 1821 void 1822 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1823 { 1824 struct tcpcb *tp = in6totcpcb(in6p); 1825 struct rtentry *rt = in6_pcbrtentry(in6p); 1826 1827 if (tp != 0) { 1828 if (rt != 0) { 1829 /* 1830 * If this was not a host route, remove and realloc. 1831 */ 1832 if ((rt->rt_flags & RTF_HOST) == 0) { 1833 in6_rtchange(in6p, errno); 1834 if ((rt = in6_pcbrtentry(in6p)) == 0) 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 = 1848 TCP_INITIAL_WINDOW(tcp_init_win, 1849 rt->rt_rmx.rmx_mtu); 1850 } 1851 1852 /* 1853 * Resend unacknowledged packets. 1854 */ 1855 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1856 tcp_output(tp); 1857 } 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 } 2040 2041 /* 2042 * Processing necessary when a TCP connection is established. 2043 */ 2044 void 2045 tcp_established(struct tcpcb *tp) 2046 { 2047 struct socket *so; 2048 #ifdef RTV_RPIPE 2049 struct rtentry *rt; 2050 #endif 2051 u_long bufsize; 2052 2053 #ifdef DIAGNOSTIC 2054 if (tp->t_inpcb && tp->t_in6pcb) 2055 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2056 #endif 2057 so = NULL; 2058 rt = NULL; 2059 #ifdef INET 2060 /* This is a while() to reduce the dreadful stairstepping below */ 2061 while (tp->t_inpcb) { 2062 so = tp->t_inpcb->inp_socket; 2063 #if defined(RTV_RPIPE) 2064 rt = in_pcbrtentry(tp->t_inpcb); 2065 #endif 2066 if (__predict_true(tcp_msl_enable)) { 2067 if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) { 2068 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2069 break; 2070 } 2071 2072 if (__predict_false(tcp_rttlocal)) { 2073 /* This may be adjusted by tcp_input */ 2074 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2075 break; 2076 } 2077 if (in_localaddr(tp->t_inpcb->inp_faddr)) { 2078 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2079 break; 2080 } 2081 } 2082 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2083 break; 2084 } 2085 #endif 2086 #ifdef INET6 2087 /* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */ 2088 while (!tp->t_inpcb && tp->t_in6pcb) { 2089 so = tp->t_in6pcb->in6p_socket; 2090 #if defined(RTV_RPIPE) 2091 rt = in6_pcbrtentry(tp->t_in6pcb); 2092 #endif 2093 if (__predict_true(tcp_msl_enable)) { 2094 extern const struct in6_addr in6addr_loopback; 2095 2096 if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr, 2097 &in6addr_loopback)) { 2098 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2099 break; 2100 } 2101 2102 if (__predict_false(tcp_rttlocal)) { 2103 /* This may be adjusted by tcp_input */ 2104 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2105 break; 2106 } 2107 if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) { 2108 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2109 break; 2110 } 2111 } 2112 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2113 break; 2114 } 2115 #endif 2116 2117 tp->t_state = TCPS_ESTABLISHED; 2118 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2119 2120 #ifdef RTV_RPIPE 2121 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2122 bufsize = rt->rt_rmx.rmx_recvpipe; 2123 else 2124 #endif 2125 { 2126 KASSERT(so != NULL); 2127 bufsize = so->so_rcv.sb_hiwat; 2128 } 2129 if (bufsize > tp->t_ourmss) { 2130 bufsize = roundup(bufsize, tp->t_ourmss); 2131 if (bufsize > sb_max) 2132 bufsize = sb_max; 2133 (void) sbreserve(&so->so_rcv, bufsize, so); 2134 } 2135 } 2136 2137 /* 2138 * Check if there's an initial rtt or rttvar. Convert from the 2139 * route-table units to scaled multiples of the slow timeout timer. 2140 * Called only during the 3-way handshake. 2141 */ 2142 void 2143 tcp_rmx_rtt(struct tcpcb *tp) 2144 { 2145 #ifdef RTV_RTT 2146 struct rtentry *rt = NULL; 2147 int rtt; 2148 2149 #ifdef DIAGNOSTIC 2150 if (tp->t_inpcb && tp->t_in6pcb) 2151 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2152 #endif 2153 #ifdef INET 2154 if (tp->t_inpcb) 2155 rt = in_pcbrtentry(tp->t_inpcb); 2156 #endif 2157 #ifdef INET6 2158 if (tp->t_in6pcb) 2159 rt = in6_pcbrtentry(tp->t_in6pcb); 2160 #endif 2161 if (rt == NULL) 2162 return; 2163 2164 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2165 /* 2166 * XXX The lock bit for MTU indicates that the value 2167 * is also a minimum value; this is subject to time. 2168 */ 2169 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2170 TCPT_RANGESET(tp->t_rttmin, 2171 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2172 TCPTV_MIN, TCPTV_REXMTMAX); 2173 tp->t_srtt = rtt / 2174 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2175 if (rt->rt_rmx.rmx_rttvar) { 2176 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2177 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2178 (TCP_RTTVAR_SHIFT + 2)); 2179 } else { 2180 /* Default variation is +- 1 rtt */ 2181 tp->t_rttvar = 2182 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2183 } 2184 TCPT_RANGESET(tp->t_rxtcur, 2185 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2186 tp->t_rttmin, TCPTV_REXMTMAX); 2187 } 2188 #endif 2189 } 2190 2191 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2192 2193 /* 2194 * Get a new sequence value given a tcp control block 2195 */ 2196 tcp_seq 2197 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2198 { 2199 2200 #ifdef INET 2201 if (tp->t_inpcb != NULL) { 2202 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2203 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2204 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2205 addin)); 2206 } 2207 #endif 2208 #ifdef INET6 2209 if (tp->t_in6pcb != NULL) { 2210 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2211 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2212 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2213 addin)); 2214 } 2215 #endif 2216 /* Not possible. */ 2217 panic("tcp_new_iss"); 2218 } 2219 2220 /* 2221 * This routine actually generates a new TCP initial sequence number. 2222 */ 2223 tcp_seq 2224 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2225 size_t addrsz, tcp_seq addin) 2226 { 2227 tcp_seq tcp_iss; 2228 2229 /* RFC1948 specifics */ 2230 static bool tcp_iss_gotten_secret; 2231 static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2232 2233 if (tcp_do_rfc1948) { 2234 MD5_CTX ctx; 2235 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2236 2237 /* 2238 * If we haven't been here before, initialize our cryptographic 2239 * hash secret. 2240 */ 2241 if (tcp_iss_gotten_secret == false) { 2242 cprng_strong(kern_cprng, 2243 tcp_iss_secret, sizeof(tcp_iss_secret), 0); 2244 tcp_iss_gotten_secret = true; 2245 } 2246 2247 /* 2248 * Compute the base value of the ISS. It is a hash 2249 * of (saddr, sport, daddr, dport, secret). 2250 */ 2251 MD5Init(&ctx); 2252 2253 MD5Update(&ctx, (u_char *) laddr, addrsz); 2254 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2255 2256 MD5Update(&ctx, (u_char *) faddr, addrsz); 2257 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2258 2259 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2260 2261 MD5Final(hash, &ctx); 2262 2263 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2264 2265 /* 2266 * Now increment our "timer", and add it in to 2267 * the computed value. 2268 * 2269 * XXX Use `addin'? 2270 * XXX TCP_ISSINCR too large to use? 2271 */ 2272 tcp_iss_seq += TCP_ISSINCR; 2273 #ifdef TCPISS_DEBUG 2274 printf("ISS hash 0x%08x, ", tcp_iss); 2275 #endif 2276 tcp_iss += tcp_iss_seq + addin; 2277 #ifdef TCPISS_DEBUG 2278 printf("new ISS 0x%08x\n", tcp_iss); 2279 #endif 2280 } else { 2281 /* 2282 * Randomize. 2283 */ 2284 tcp_iss = cprng_fast32(); 2285 2286 /* 2287 * If we were asked to add some amount to a known value, 2288 * we will take a random value obtained above, mask off 2289 * the upper bits, and add in the known value. We also 2290 * add in a constant to ensure that we are at least a 2291 * certain distance from the original value. 2292 * 2293 * This is used when an old connection is in timed wait 2294 * and we have a new one coming in, for instance. 2295 */ 2296 if (addin != 0) { 2297 #ifdef TCPISS_DEBUG 2298 printf("Random %08x, ", tcp_iss); 2299 #endif 2300 tcp_iss &= TCP_ISS_RANDOM_MASK; 2301 tcp_iss += addin + TCP_ISSINCR; 2302 #ifdef TCPISS_DEBUG 2303 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2304 #endif 2305 } else { 2306 tcp_iss &= TCP_ISS_RANDOM_MASK; 2307 tcp_iss += tcp_iss_seq; 2308 tcp_iss_seq += TCP_ISSINCR; 2309 #ifdef TCPISS_DEBUG 2310 printf("ISS %08x\n", tcp_iss); 2311 #endif 2312 } 2313 } 2314 2315 if (tcp_compat_42) { 2316 /* 2317 * Limit it to the positive range for really old TCP 2318 * implementations. 2319 * Just AND off the top bit instead of checking if 2320 * is set first - saves a branch 50% of the time. 2321 */ 2322 tcp_iss &= 0x7fffffff; /* XXX */ 2323 } 2324 2325 return (tcp_iss); 2326 } 2327 2328 #if defined(IPSEC) 2329 /* compute ESP/AH header size for TCP, including outer IP header. */ 2330 size_t 2331 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2332 { 2333 struct inpcb *inp; 2334 size_t hdrsiz; 2335 2336 /* XXX mapped addr case (tp->t_in6pcb) */ 2337 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2338 return 0; 2339 switch (tp->t_family) { 2340 case AF_INET: 2341 /* XXX: should use currect direction. */ 2342 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2343 break; 2344 default: 2345 hdrsiz = 0; 2346 break; 2347 } 2348 2349 return hdrsiz; 2350 } 2351 2352 #ifdef INET6 2353 size_t 2354 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2355 { 2356 struct in6pcb *in6p; 2357 size_t hdrsiz; 2358 2359 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2360 return 0; 2361 switch (tp->t_family) { 2362 case AF_INET6: 2363 /* XXX: should use currect direction. */ 2364 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2365 break; 2366 case AF_INET: 2367 /* mapped address case - tricky */ 2368 default: 2369 hdrsiz = 0; 2370 break; 2371 } 2372 2373 return hdrsiz; 2374 } 2375 #endif 2376 #endif /*IPSEC*/ 2377 2378 /* 2379 * Determine the length of the TCP options for this connection. 2380 * 2381 * XXX: What do we do for SACK, when we add that? Just reserve 2382 * all of the space? Otherwise we can't exactly be incrementing 2383 * cwnd by an amount that varies depending on the amount we last 2384 * had to SACK! 2385 */ 2386 2387 u_int 2388 tcp_optlen(struct tcpcb *tp) 2389 { 2390 u_int optlen; 2391 2392 optlen = 0; 2393 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2394 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2395 optlen += TCPOLEN_TSTAMP_APPA; 2396 2397 #ifdef TCP_SIGNATURE 2398 if (tp->t_flags & TF_SIGNATURE) 2399 optlen += TCPOLEN_SIGNATURE + 2; 2400 #endif /* TCP_SIGNATURE */ 2401 2402 return optlen; 2403 } 2404 2405 u_int 2406 tcp_hdrsz(struct tcpcb *tp) 2407 { 2408 u_int hlen; 2409 2410 switch (tp->t_family) { 2411 #ifdef INET6 2412 case AF_INET6: 2413 hlen = sizeof(struct ip6_hdr); 2414 break; 2415 #endif 2416 case AF_INET: 2417 hlen = sizeof(struct ip); 2418 break; 2419 default: 2420 hlen = 0; 2421 break; 2422 } 2423 hlen += sizeof(struct tcphdr); 2424 2425 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2426 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2427 hlen += TCPOLEN_TSTAMP_APPA; 2428 #ifdef TCP_SIGNATURE 2429 if (tp->t_flags & TF_SIGNATURE) 2430 hlen += TCPOLEN_SIGLEN; 2431 #endif 2432 return hlen; 2433 } 2434 2435 void 2436 tcp_statinc(u_int stat) 2437 { 2438 2439 KASSERT(stat < TCP_NSTATS); 2440 TCP_STATINC(stat); 2441 } 2442 2443 void 2444 tcp_statadd(u_int stat, uint64_t val) 2445 { 2446 2447 KASSERT(stat < TCP_NSTATS); 2448 TCP_STATADD(stat, val); 2449 } 2450
Indexes created Sun Sep 21 20:09:37 GMT 2025