1 /* $NetBSD: tcp_subr.c,v 1.238 2009/09/16 15:23:05 pooka 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 1.238 2009/09/16 15:23:05 pooka Exp $"); 95 96 #include "opt_inet.h" 97 #include "opt_ipsec.h" 98 #include "opt_tcp_compat_42.h" 99 #include "opt_inet_csum.h" 100 #include "opt_mbuftrace.h" 101 #include "rnd.h" 102 103 #include <sys/param.h> 104 #include <sys/proc.h> 105 #include <sys/systm.h> 106 #include <sys/malloc.h> 107 #include <sys/mbuf.h> 108 #include <sys/socket.h> 109 #include <sys/socketvar.h> 110 #include <sys/protosw.h> 111 #include <sys/errno.h> 112 #include <sys/kernel.h> 113 #include <sys/pool.h> 114 #if NRND > 0 115 #include <sys/md5.h> 116 #include <sys/rnd.h> 117 #endif 118 119 #include <net/route.h> 120 #include <net/if.h> 121 122 #include <netinet/in.h> 123 #include <netinet/in_systm.h> 124 #include <netinet/ip.h> 125 #include <netinet/in_pcb.h> 126 #include <netinet/ip_var.h> 127 #include <netinet/ip_icmp.h> 128 129 #ifdef INET6 130 #ifndef INET 131 #include <netinet/in.h> 132 #endif 133 #include <netinet/ip6.h> 134 #include <netinet6/in6_pcb.h> 135 #include <netinet6/ip6_var.h> 136 #include <netinet6/in6_var.h> 137 #include <netinet6/ip6protosw.h> 138 #include <netinet/icmp6.h> 139 #include <netinet6/nd6.h> 140 #endif 141 142 #include <netinet/tcp.h> 143 #include <netinet/tcp_fsm.h> 144 #include <netinet/tcp_seq.h> 145 #include <netinet/tcp_timer.h> 146 #include <netinet/tcp_var.h> 147 #include <netinet/tcp_private.h> 148 #include <netinet/tcp_congctl.h> 149 #include <netinet/tcpip.h> 150 151 #ifdef IPSEC 152 #include <netinet6/ipsec.h> 153 #include <netkey/key.h> 154 #endif /*IPSEC*/ 155 156 #ifdef FAST_IPSEC 157 #include <netipsec/ipsec.h> 158 #include <netipsec/xform.h> 159 #ifdef INET6 160 #include <netipsec/ipsec6.h> 161 #endif 162 #include <netipsec/key.h> 163 #endif /* FAST_IPSEC*/ 164 165 166 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 167 u_int32_t tcp_now; /* for RFC 1323 timestamps */ 168 169 percpu_t *tcpstat_percpu; 170 171 /* patchable/settable parameters for tcp */ 172 int tcp_mssdflt = TCP_MSS; 173 int tcp_minmss = TCP_MINMSS; 174 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 175 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 176 #if NRND > 0 177 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 178 #endif 179 int tcp_do_sack = 1; /* selective acknowledgement */ 180 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 181 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 182 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 183 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 184 #ifndef TCP_INIT_WIN 185 #define TCP_INIT_WIN 0 /* initial slow start window */ 186 #endif 187 #ifndef TCP_INIT_WIN_LOCAL 188 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 189 #endif 190 int tcp_init_win = TCP_INIT_WIN; 191 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 192 int tcp_mss_ifmtu = 0; 193 #ifdef TCP_COMPAT_42 194 int tcp_compat_42 = 1; 195 #else 196 int tcp_compat_42 = 0; 197 #endif 198 int tcp_rst_ppslim = 100; /* 100pps */ 199 int tcp_ackdrop_ppslim = 100; /* 100pps */ 200 int tcp_do_loopback_cksum = 0; 201 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 202 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 203 int tcp_sack_tp_maxholes = 32; 204 int tcp_sack_globalmaxholes = 1024; 205 int tcp_sack_globalholes = 0; 206 int tcp_ecn_maxretries = 1; 207 208 /* tcb hash */ 209 #ifndef TCBHASHSIZE 210 #define TCBHASHSIZE 128 211 #endif 212 int tcbhashsize = TCBHASHSIZE; 213 214 /* syn hash parameters */ 215 #define TCP_SYN_HASH_SIZE 293 216 #define TCP_SYN_BUCKET_SIZE 35 217 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 218 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 219 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 220 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 221 222 int tcp_freeq(struct tcpcb *); 223 224 #ifdef INET 225 void tcp_mtudisc_callback(struct in_addr); 226 #endif 227 #ifdef INET6 228 void tcp6_mtudisc_callback(struct in6_addr *); 229 #endif 230 231 #ifdef INET6 232 void tcp6_mtudisc(struct in6pcb *, int); 233 #endif 234 235 static struct pool tcpcb_pool; 236 237 #ifdef TCP_CSUM_COUNTERS 238 #include <sys/device.h> 239 240 #if defined(INET) 241 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 242 NULL, "tcp", "hwcsum bad"); 243 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 244 NULL, "tcp", "hwcsum ok"); 245 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 246 NULL, "tcp", "hwcsum data"); 247 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 248 NULL, "tcp", "swcsum"); 249 250 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 251 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 252 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 253 EVCNT_ATTACH_STATIC(tcp_swcsum); 254 #endif /* defined(INET) */ 255 256 #if defined(INET6) 257 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 258 NULL, "tcp6", "hwcsum bad"); 259 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 260 NULL, "tcp6", "hwcsum ok"); 261 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 262 NULL, "tcp6", "hwcsum data"); 263 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 264 NULL, "tcp6", "swcsum"); 265 266 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 267 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 268 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 269 EVCNT_ATTACH_STATIC(tcp6_swcsum); 270 #endif /* defined(INET6) */ 271 #endif /* TCP_CSUM_COUNTERS */ 272 273 274 #ifdef TCP_OUTPUT_COUNTERS 275 #include <sys/device.h> 276 277 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 278 NULL, "tcp", "output big header"); 279 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 280 NULL, "tcp", "output predict hit"); 281 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 282 NULL, "tcp", "output predict miss"); 283 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 284 NULL, "tcp", "output copy small"); 285 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 286 NULL, "tcp", "output copy big"); 287 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 288 NULL, "tcp", "output reference big"); 289 290 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 291 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 292 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 293 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 294 EVCNT_ATTACH_STATIC(tcp_output_copybig); 295 EVCNT_ATTACH_STATIC(tcp_output_refbig); 296 297 #endif /* TCP_OUTPUT_COUNTERS */ 298 299 #ifdef TCP_REASS_COUNTERS 300 #include <sys/device.h> 301 302 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 303 NULL, "tcp_reass", "calls"); 304 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 305 &tcp_reass_, "tcp_reass", "insert into empty queue"); 306 struct evcnt tcp_reass_iteration[8] = { 307 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 308 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 315 }; 316 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 317 &tcp_reass_, "tcp_reass", "prepend to first"); 318 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 319 &tcp_reass_, "tcp_reass", "prepend"); 320 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 321 &tcp_reass_, "tcp_reass", "insert"); 322 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 323 &tcp_reass_, "tcp_reass", "insert at tail"); 324 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 325 &tcp_reass_, "tcp_reass", "append"); 326 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 327 &tcp_reass_, "tcp_reass", "append to tail fragment"); 328 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 329 &tcp_reass_, "tcp_reass", "overlap at end"); 330 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 331 &tcp_reass_, "tcp_reass", "overlap at start"); 332 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 333 &tcp_reass_, "tcp_reass", "duplicate segment"); 334 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 335 &tcp_reass_, "tcp_reass", "duplicate fragment"); 336 337 EVCNT_ATTACH_STATIC(tcp_reass_); 338 EVCNT_ATTACH_STATIC(tcp_reass_empty); 339 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 340 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 341 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 342 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 347 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 348 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 349 EVCNT_ATTACH_STATIC(tcp_reass_insert); 350 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 351 EVCNT_ATTACH_STATIC(tcp_reass_append); 352 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 353 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 354 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 355 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 356 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 357 358 #endif /* TCP_REASS_COUNTERS */ 359 360 #ifdef MBUFTRACE 361 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 362 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 363 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 364 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 365 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 366 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 367 #endif 368 369 /* 370 * Tcp initialization 371 */ 372 void 373 tcp_init(void) 374 { 375 int hlen; 376 377 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 378 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 379 NULL, IPL_SOFTNET); 380 381 hlen = sizeof(struct ip) + sizeof(struct tcphdr); 382 #ifdef INET6 383 if (sizeof(struct ip) < sizeof(struct ip6_hdr)) 384 hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 385 #endif 386 if (max_protohdr < hlen) 387 max_protohdr = hlen; 388 if (max_linkhdr + hlen > MHLEN) 389 panic("tcp_init"); 390 391 #ifdef INET 392 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 393 #endif 394 #ifdef INET6 395 icmp6_mtudisc_callback_register(tcp6_mtudisc_callback); 396 #endif 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 } 428 429 /* 430 * Create template to be used to send tcp packets on a connection. 431 * Call after host entry created, allocates an mbuf and fills 432 * in a skeletal tcp/ip header, minimizing the amount of work 433 * necessary when the connection is used. 434 */ 435 struct mbuf * 436 tcp_template(struct tcpcb *tp) 437 { 438 struct inpcb *inp = tp->t_inpcb; 439 #ifdef INET6 440 struct in6pcb *in6p = tp->t_in6pcb; 441 #endif 442 struct tcphdr *n; 443 struct mbuf *m; 444 int hlen; 445 446 switch (tp->t_family) { 447 case AF_INET: 448 hlen = sizeof(struct ip); 449 if (inp) 450 break; 451 #ifdef INET6 452 if (in6p) { 453 /* mapped addr case */ 454 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 455 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 456 break; 457 } 458 #endif 459 return NULL; /*EINVAL*/ 460 #ifdef INET6 461 case AF_INET6: 462 hlen = sizeof(struct ip6_hdr); 463 if (in6p) { 464 /* more sainty check? */ 465 break; 466 } 467 return NULL; /*EINVAL*/ 468 #endif 469 default: 470 hlen = 0; /*pacify gcc*/ 471 return NULL; /*EAFNOSUPPORT*/ 472 } 473 #ifdef DIAGNOSTIC 474 if (hlen + sizeof(struct tcphdr) > MCLBYTES) 475 panic("mclbytes too small for t_template"); 476 #endif 477 m = tp->t_template; 478 if (m && m->m_len == hlen + sizeof(struct tcphdr)) 479 ; 480 else { 481 if (m) 482 m_freem(m); 483 m = tp->t_template = NULL; 484 MGETHDR(m, M_DONTWAIT, MT_HEADER); 485 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 486 MCLGET(m, M_DONTWAIT); 487 if ((m->m_flags & M_EXT) == 0) { 488 m_free(m); 489 m = NULL; 490 } 491 } 492 if (m == NULL) 493 return NULL; 494 MCLAIM(m, &tcp_mowner); 495 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 496 } 497 498 memset(mtod(m, void *), 0, m->m_len); 499 500 n = (struct tcphdr *)(mtod(m, char *) + hlen); 501 502 switch (tp->t_family) { 503 case AF_INET: 504 { 505 struct ipovly *ipov; 506 mtod(m, struct ip *)->ip_v = 4; 507 mtod(m, struct ip *)->ip_hl = hlen >> 2; 508 ipov = mtod(m, struct ipovly *); 509 ipov->ih_pr = IPPROTO_TCP; 510 ipov->ih_len = htons(sizeof(struct tcphdr)); 511 if (inp) { 512 ipov->ih_src = inp->inp_laddr; 513 ipov->ih_dst = inp->inp_faddr; 514 } 515 #ifdef INET6 516 else if (in6p) { 517 /* mapped addr case */ 518 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 519 sizeof(ipov->ih_src)); 520 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 521 sizeof(ipov->ih_dst)); 522 } 523 #endif 524 /* 525 * Compute the pseudo-header portion of the checksum 526 * now. We incrementally add in the TCP option and 527 * payload lengths later, and then compute the TCP 528 * checksum right before the packet is sent off onto 529 * the wire. 530 */ 531 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 532 ipov->ih_dst.s_addr, 533 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 534 break; 535 } 536 #ifdef INET6 537 case AF_INET6: 538 { 539 struct ip6_hdr *ip6; 540 mtod(m, struct ip *)->ip_v = 6; 541 ip6 = mtod(m, struct ip6_hdr *); 542 ip6->ip6_nxt = IPPROTO_TCP; 543 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 544 ip6->ip6_src = in6p->in6p_laddr; 545 ip6->ip6_dst = in6p->in6p_faddr; 546 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 547 if (ip6_auto_flowlabel) { 548 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 549 ip6->ip6_flow |= 550 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 551 } 552 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 553 ip6->ip6_vfc |= IPV6_VERSION; 554 555 /* 556 * Compute the pseudo-header portion of the checksum 557 * now. We incrementally add in the TCP option and 558 * payload lengths later, and then compute the TCP 559 * checksum right before the packet is sent off onto 560 * the wire. 561 */ 562 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 563 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 564 htonl(IPPROTO_TCP)); 565 break; 566 } 567 #endif 568 } 569 if (inp) { 570 n->th_sport = inp->inp_lport; 571 n->th_dport = inp->inp_fport; 572 } 573 #ifdef INET6 574 else if (in6p) { 575 n->th_sport = in6p->in6p_lport; 576 n->th_dport = in6p->in6p_fport; 577 } 578 #endif 579 n->th_seq = 0; 580 n->th_ack = 0; 581 n->th_x2 = 0; 582 n->th_off = 5; 583 n->th_flags = 0; 584 n->th_win = 0; 585 n->th_urp = 0; 586 return (m); 587 } 588 589 /* 590 * Send a single message to the TCP at address specified by 591 * the given TCP/IP header. If m == 0, then we make a copy 592 * of the tcpiphdr at ti and send directly to the addressed host. 593 * This is used to force keep alive messages out using the TCP 594 * template for a connection tp->t_template. If flags are given 595 * then we send a message back to the TCP which originated the 596 * segment ti, and discard the mbuf containing it and any other 597 * attached mbufs. 598 * 599 * In any case the ack and sequence number of the transmitted 600 * segment are as specified by the parameters. 601 */ 602 int 603 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m, 604 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 605 { 606 #ifdef INET6 607 struct rtentry *rt; 608 #endif 609 struct route *ro; 610 int error, tlen, win = 0; 611 int hlen; 612 struct ip *ip; 613 #ifdef INET6 614 struct ip6_hdr *ip6; 615 #endif 616 int family; /* family on packet, not inpcb/in6pcb! */ 617 struct tcphdr *th; 618 struct socket *so; 619 620 if (tp != NULL && (flags & TH_RST) == 0) { 621 #ifdef DIAGNOSTIC 622 if (tp->t_inpcb && tp->t_in6pcb) 623 panic("tcp_respond: both t_inpcb and t_in6pcb are set"); 624 #endif 625 #ifdef INET 626 if (tp->t_inpcb) 627 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 628 #endif 629 #ifdef INET6 630 if (tp->t_in6pcb) 631 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 632 #endif 633 } 634 635 th = NULL; /* Quell uninitialized warning */ 636 ip = NULL; 637 #ifdef INET6 638 ip6 = NULL; 639 #endif 640 if (m == 0) { 641 if (!template) 642 return EINVAL; 643 644 /* get family information from template */ 645 switch (mtod(template, struct ip *)->ip_v) { 646 case 4: 647 family = AF_INET; 648 hlen = sizeof(struct ip); 649 break; 650 #ifdef INET6 651 case 6: 652 family = AF_INET6; 653 hlen = sizeof(struct ip6_hdr); 654 break; 655 #endif 656 default: 657 return EAFNOSUPPORT; 658 } 659 660 MGETHDR(m, M_DONTWAIT, MT_HEADER); 661 if (m) { 662 MCLAIM(m, &tcp_tx_mowner); 663 MCLGET(m, M_DONTWAIT); 664 if ((m->m_flags & M_EXT) == 0) { 665 m_free(m); 666 m = NULL; 667 } 668 } 669 if (m == NULL) 670 return (ENOBUFS); 671 672 if (tcp_compat_42) 673 tlen = 1; 674 else 675 tlen = 0; 676 677 m->m_data += max_linkhdr; 678 bcopy(mtod(template, void *), mtod(m, void *), 679 template->m_len); 680 switch (family) { 681 case AF_INET: 682 ip = mtod(m, struct ip *); 683 th = (struct tcphdr *)(ip + 1); 684 break; 685 #ifdef INET6 686 case AF_INET6: 687 ip6 = mtod(m, struct ip6_hdr *); 688 th = (struct tcphdr *)(ip6 + 1); 689 break; 690 #endif 691 #if 0 692 default: 693 /* noone will visit here */ 694 m_freem(m); 695 return EAFNOSUPPORT; 696 #endif 697 } 698 flags = TH_ACK; 699 } else { 700 701 if ((m->m_flags & M_PKTHDR) == 0) { 702 #if 0 703 printf("non PKTHDR to tcp_respond\n"); 704 #endif 705 m_freem(m); 706 return EINVAL; 707 } 708 #ifdef DIAGNOSTIC 709 if (!th0) 710 panic("th0 == NULL in tcp_respond"); 711 #endif 712 713 /* get family information from m */ 714 switch (mtod(m, struct ip *)->ip_v) { 715 case 4: 716 family = AF_INET; 717 hlen = sizeof(struct ip); 718 ip = mtod(m, struct ip *); 719 break; 720 #ifdef INET6 721 case 6: 722 family = AF_INET6; 723 hlen = sizeof(struct ip6_hdr); 724 ip6 = mtod(m, struct ip6_hdr *); 725 break; 726 #endif 727 default: 728 m_freem(m); 729 return EAFNOSUPPORT; 730 } 731 /* clear h/w csum flags inherited from rx packet */ 732 m->m_pkthdr.csum_flags = 0; 733 734 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 735 tlen = sizeof(*th0); 736 else 737 tlen = th0->th_off << 2; 738 739 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 740 mtod(m, char *) + hlen == (char *)th0) { 741 m->m_len = hlen + tlen; 742 m_freem(m->m_next); 743 m->m_next = NULL; 744 } else { 745 struct mbuf *n; 746 747 #ifdef DIAGNOSTIC 748 if (max_linkhdr + hlen + tlen > MCLBYTES) { 749 m_freem(m); 750 return EMSGSIZE; 751 } 752 #endif 753 MGETHDR(n, M_DONTWAIT, MT_HEADER); 754 if (n && max_linkhdr + hlen + tlen > MHLEN) { 755 MCLGET(n, M_DONTWAIT); 756 if ((n->m_flags & M_EXT) == 0) { 757 m_freem(n); 758 n = NULL; 759 } 760 } 761 if (!n) { 762 m_freem(m); 763 return ENOBUFS; 764 } 765 766 MCLAIM(n, &tcp_tx_mowner); 767 n->m_data += max_linkhdr; 768 n->m_len = hlen + tlen; 769 m_copyback(n, 0, hlen, mtod(m, void *)); 770 m_copyback(n, hlen, tlen, (void *)th0); 771 772 m_freem(m); 773 m = n; 774 n = NULL; 775 } 776 777 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 778 switch (family) { 779 case AF_INET: 780 ip = mtod(m, struct ip *); 781 th = (struct tcphdr *)(ip + 1); 782 ip->ip_p = IPPROTO_TCP; 783 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 784 ip->ip_p = IPPROTO_TCP; 785 break; 786 #ifdef INET6 787 case AF_INET6: 788 ip6 = mtod(m, struct ip6_hdr *); 789 th = (struct tcphdr *)(ip6 + 1); 790 ip6->ip6_nxt = IPPROTO_TCP; 791 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 792 ip6->ip6_nxt = IPPROTO_TCP; 793 break; 794 #endif 795 #if 0 796 default: 797 /* noone will visit here */ 798 m_freem(m); 799 return EAFNOSUPPORT; 800 #endif 801 } 802 xchg(th->th_dport, th->th_sport, u_int16_t); 803 #undef xchg 804 tlen = 0; /*be friendly with the following code*/ 805 } 806 th->th_seq = htonl(seq); 807 th->th_ack = htonl(ack); 808 th->th_x2 = 0; 809 if ((flags & TH_SYN) == 0) { 810 if (tp) 811 win >>= tp->rcv_scale; 812 if (win > TCP_MAXWIN) 813 win = TCP_MAXWIN; 814 th->th_win = htons((u_int16_t)win); 815 th->th_off = sizeof (struct tcphdr) >> 2; 816 tlen += sizeof(*th); 817 } else 818 tlen += th->th_off << 2; 819 m->m_len = hlen + tlen; 820 m->m_pkthdr.len = hlen + tlen; 821 m->m_pkthdr.rcvif = (struct ifnet *) 0; 822 th->th_flags = flags; 823 th->th_urp = 0; 824 825 switch (family) { 826 #ifdef INET 827 case AF_INET: 828 { 829 struct ipovly *ipov = (struct ipovly *)ip; 830 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1); 831 ipov->ih_len = htons((u_int16_t)tlen); 832 833 th->th_sum = 0; 834 th->th_sum = in_cksum(m, hlen + tlen); 835 ip->ip_len = htons(hlen + tlen); 836 ip->ip_ttl = ip_defttl; 837 break; 838 } 839 #endif 840 #ifdef INET6 841 case AF_INET6: 842 { 843 th->th_sum = 0; 844 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 845 tlen); 846 ip6->ip6_plen = htons(tlen); 847 if (tp && tp->t_in6pcb) { 848 struct ifnet *oifp; 849 ro = &tp->t_in6pcb->in6p_route; 850 oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 851 : NULL; 852 ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp); 853 } else 854 ip6->ip6_hlim = ip6_defhlim; 855 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 856 if (ip6_auto_flowlabel) { 857 ip6->ip6_flow |= 858 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 859 } 860 break; 861 } 862 #endif 863 } 864 865 if (tp && tp->t_inpcb) 866 so = tp->t_inpcb->inp_socket; 867 #ifdef INET6 868 else if (tp && tp->t_in6pcb) 869 so = tp->t_in6pcb->in6p_socket; 870 #endif 871 else 872 so = NULL; 873 874 if (tp != NULL && tp->t_inpcb != NULL) { 875 ro = &tp->t_inpcb->inp_route; 876 #ifdef DIAGNOSTIC 877 if (family != AF_INET) 878 panic("tcp_respond: address family mismatch"); 879 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) { 880 panic("tcp_respond: ip_dst %x != inp_faddr %x", 881 ntohl(ip->ip_dst.s_addr), 882 ntohl(tp->t_inpcb->inp_faddr.s_addr)); 883 } 884 #endif 885 } 886 #ifdef INET6 887 else if (tp != NULL && tp->t_in6pcb != NULL) { 888 ro = (struct route *)&tp->t_in6pcb->in6p_route; 889 #ifdef DIAGNOSTIC 890 if (family == AF_INET) { 891 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 892 panic("tcp_respond: not mapped addr"); 893 if (memcmp(&ip->ip_dst, 894 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 895 sizeof(ip->ip_dst)) != 0) { 896 panic("tcp_respond: ip_dst != in6p_faddr"); 897 } 898 } else if (family == AF_INET6) { 899 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 900 &tp->t_in6pcb->in6p_faddr)) 901 panic("tcp_respond: ip6_dst != in6p_faddr"); 902 } else 903 panic("tcp_respond: address family mismatch"); 904 #endif 905 } 906 #endif 907 else 908 ro = NULL; 909 910 switch (family) { 911 #ifdef INET 912 case AF_INET: 913 error = ip_output(m, NULL, ro, 914 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), 915 (struct ip_moptions *)0, so); 916 break; 917 #endif 918 #ifdef INET6 919 case AF_INET6: 920 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL); 921 break; 922 #endif 923 default: 924 error = EAFNOSUPPORT; 925 break; 926 } 927 928 return (error); 929 } 930 931 /* 932 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 933 * a bunch of members individually, we maintain this template for the 934 * static and mostly-static components of the TCPCB, and copy it into 935 * the new TCPCB instead. 936 */ 937 static struct tcpcb tcpcb_template = { 938 .t_srtt = TCPTV_SRTTBASE, 939 .t_rttmin = TCPTV_MIN, 940 941 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 942 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 943 .snd_numholes = 0, 944 945 .t_partialacks = -1, 946 .t_bytes_acked = 0, 947 }; 948 949 /* 950 * Updates the TCPCB template whenever a parameter that would affect 951 * the template is changed. 952 */ 953 void 954 tcp_tcpcb_template(void) 955 { 956 struct tcpcb *tp = &tcpcb_template; 957 int flags; 958 959 tp->t_peermss = tcp_mssdflt; 960 tp->t_ourmss = tcp_mssdflt; 961 tp->t_segsz = tcp_mssdflt; 962 963 flags = 0; 964 if (tcp_do_rfc1323 && tcp_do_win_scale) 965 flags |= TF_REQ_SCALE; 966 if (tcp_do_rfc1323 && tcp_do_timestamps) 967 flags |= TF_REQ_TSTMP; 968 tp->t_flags = flags; 969 970 /* 971 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 972 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 973 * reasonable initial retransmit time. 974 */ 975 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 976 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 977 TCPTV_MIN, TCPTV_REXMTMAX); 978 979 /* Keep Alive */ 980 tp->t_keepinit = tcp_keepinit; 981 tp->t_keepidle = tcp_keepidle; 982 tp->t_keepintvl = tcp_keepintvl; 983 tp->t_keepcnt = tcp_keepcnt; 984 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 985 } 986 987 /* 988 * Create a new TCP control block, making an 989 * empty reassembly queue and hooking it to the argument 990 * protocol control block. 991 */ 992 /* family selects inpcb, or in6pcb */ 993 struct tcpcb * 994 tcp_newtcpcb(int family, void *aux) 995 { 996 #ifdef INET6 997 struct rtentry *rt; 998 #endif 999 struct tcpcb *tp; 1000 int i; 1001 1002 /* XXX Consider using a pool_cache for speed. */ 1003 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 1004 if (tp == NULL) 1005 return (NULL); 1006 memcpy(tp, &tcpcb_template, sizeof(*tp)); 1007 TAILQ_INIT(&tp->segq); 1008 TAILQ_INIT(&tp->timeq); 1009 tp->t_family = family; /* may be overridden later on */ 1010 TAILQ_INIT(&tp->snd_holes); 1011 LIST_INIT(&tp->t_sc); /* XXX can template this */ 1012 1013 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 1014 for (i = 0; i < TCPT_NTIMERS; i++) { 1015 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 1016 TCP_TIMER_INIT(tp, i); 1017 } 1018 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 1019 1020 switch (family) { 1021 case AF_INET: 1022 { 1023 struct inpcb *inp = (struct inpcb *)aux; 1024 1025 inp->inp_ip.ip_ttl = ip_defttl; 1026 inp->inp_ppcb = (void *)tp; 1027 1028 tp->t_inpcb = inp; 1029 tp->t_mtudisc = ip_mtudisc; 1030 break; 1031 } 1032 #ifdef INET6 1033 case AF_INET6: 1034 { 1035 struct in6pcb *in6p = (struct in6pcb *)aux; 1036 1037 in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p, 1038 (rt = rtcache_validate(&in6p->in6p_route)) != NULL 1039 ? rt->rt_ifp 1040 : NULL); 1041 in6p->in6p_ppcb = (void *)tp; 1042 1043 tp->t_in6pcb = in6p; 1044 /* for IPv6, always try to run path MTU discovery */ 1045 tp->t_mtudisc = 1; 1046 break; 1047 } 1048 #endif /* INET6 */ 1049 default: 1050 for (i = 0; i < TCPT_NTIMERS; i++) 1051 callout_destroy(&tp->t_timer[i]); 1052 callout_destroy(&tp->t_delack_ch); 1053 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1054 return (NULL); 1055 } 1056 1057 /* 1058 * Initialize our timebase. When we send timestamps, we take 1059 * the delta from tcp_now -- this means each connection always 1060 * gets a timebase of 1, which makes it, among other things, 1061 * more difficult to determine how long a system has been up, 1062 * and thus how many TCP sequence increments have occurred. 1063 * 1064 * We start with 1, because 0 doesn't work with linux, which 1065 * considers timestamp 0 in a SYN packet as a bug and disables 1066 * timestamps. 1067 */ 1068 tp->ts_timebase = tcp_now - 1; 1069 1070 tcp_congctl_select(tp, tcp_congctl_global_name); 1071 1072 return (tp); 1073 } 1074 1075 /* 1076 * Drop a TCP connection, reporting 1077 * the specified error. If connection is synchronized, 1078 * then send a RST to peer. 1079 */ 1080 struct tcpcb * 1081 tcp_drop(struct tcpcb *tp, int errno) 1082 { 1083 struct socket *so = NULL; 1084 1085 #ifdef DIAGNOSTIC 1086 if (tp->t_inpcb && tp->t_in6pcb) 1087 panic("tcp_drop: both t_inpcb and t_in6pcb are set"); 1088 #endif 1089 #ifdef INET 1090 if (tp->t_inpcb) 1091 so = tp->t_inpcb->inp_socket; 1092 #endif 1093 #ifdef INET6 1094 if (tp->t_in6pcb) 1095 so = tp->t_in6pcb->in6p_socket; 1096 #endif 1097 if (!so) 1098 return NULL; 1099 1100 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1101 tp->t_state = TCPS_CLOSED; 1102 (void) tcp_output(tp); 1103 TCP_STATINC(TCP_STAT_DROPS); 1104 } else 1105 TCP_STATINC(TCP_STAT_CONNDROPS); 1106 if (errno == ETIMEDOUT && tp->t_softerror) 1107 errno = tp->t_softerror; 1108 so->so_error = errno; 1109 return (tcp_close(tp)); 1110 } 1111 1112 /* 1113 * Close a TCP control block: 1114 * discard all space held by the tcp 1115 * discard internet protocol block 1116 * wake up any sleepers 1117 */ 1118 struct tcpcb * 1119 tcp_close(struct tcpcb *tp) 1120 { 1121 struct inpcb *inp; 1122 #ifdef INET6 1123 struct in6pcb *in6p; 1124 #endif 1125 struct socket *so; 1126 #ifdef RTV_RTT 1127 struct rtentry *rt; 1128 #endif 1129 struct route *ro; 1130 int j; 1131 1132 inp = tp->t_inpcb; 1133 #ifdef INET6 1134 in6p = tp->t_in6pcb; 1135 #endif 1136 so = NULL; 1137 ro = NULL; 1138 if (inp) { 1139 so = inp->inp_socket; 1140 ro = &inp->inp_route; 1141 } 1142 #ifdef INET6 1143 else if (in6p) { 1144 so = in6p->in6p_socket; 1145 ro = (struct route *)&in6p->in6p_route; 1146 } 1147 #endif 1148 1149 #ifdef RTV_RTT 1150 /* 1151 * If we sent enough data to get some meaningful characteristics, 1152 * save them in the routing entry. 'Enough' is arbitrarily 1153 * defined as the sendpipesize (default 4K) * 16. This would 1154 * give us 16 rtt samples assuming we only get one sample per 1155 * window (the usual case on a long haul net). 16 samples is 1156 * enough for the srtt filter to converge to within 5% of the correct 1157 * value; fewer samples and we could save a very bogus rtt. 1158 * 1159 * Don't update the default route's characteristics and don't 1160 * update anything that the user "locked". 1161 */ 1162 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1163 ro && (rt = rtcache_validate(ro)) != NULL && 1164 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1165 u_long i = 0; 1166 1167 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1168 i = tp->t_srtt * 1169 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1170 if (rt->rt_rmx.rmx_rtt && i) 1171 /* 1172 * filter this update to half the old & half 1173 * the new values, converting scale. 1174 * See route.h and tcp_var.h for a 1175 * description of the scaling constants. 1176 */ 1177 rt->rt_rmx.rmx_rtt = 1178 (rt->rt_rmx.rmx_rtt + i) / 2; 1179 else 1180 rt->rt_rmx.rmx_rtt = i; 1181 } 1182 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1183 i = tp->t_rttvar * 1184 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1185 if (rt->rt_rmx.rmx_rttvar && i) 1186 rt->rt_rmx.rmx_rttvar = 1187 (rt->rt_rmx.rmx_rttvar + i) / 2; 1188 else 1189 rt->rt_rmx.rmx_rttvar = i; 1190 } 1191 /* 1192 * update the pipelimit (ssthresh) if it has been updated 1193 * already or if a pipesize was specified & the threshhold 1194 * got below half the pipesize. I.e., wait for bad news 1195 * before we start updating, then update on both good 1196 * and bad news. 1197 */ 1198 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1199 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1200 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1201 /* 1202 * convert the limit from user data bytes to 1203 * packets then to packet data bytes. 1204 */ 1205 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1206 if (i < 2) 1207 i = 2; 1208 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1209 if (rt->rt_rmx.rmx_ssthresh) 1210 rt->rt_rmx.rmx_ssthresh = 1211 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1212 else 1213 rt->rt_rmx.rmx_ssthresh = i; 1214 } 1215 } 1216 #endif /* RTV_RTT */ 1217 /* free the reassembly queue, if any */ 1218 TCP_REASS_LOCK(tp); 1219 (void) tcp_freeq(tp); 1220 TCP_REASS_UNLOCK(tp); 1221 1222 /* free the SACK holes list. */ 1223 tcp_free_sackholes(tp); 1224 tcp_congctl_release(tp); 1225 syn_cache_cleanup(tp); 1226 1227 if (tp->t_template) { 1228 m_free(tp->t_template); 1229 tp->t_template = NULL; 1230 } 1231 1232 /* 1233 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1234 * the timers can also drop the lock. We need to prevent access 1235 * to the tcpcb as it's half torn down. Flag the pcb as dead 1236 * (prevents access by timers) and only then detach it. 1237 */ 1238 tp->t_flags |= TF_DEAD; 1239 if (inp) { 1240 inp->inp_ppcb = 0; 1241 soisdisconnected(so); 1242 in_pcbdetach(inp); 1243 } 1244 #ifdef INET6 1245 else if (in6p) { 1246 in6p->in6p_ppcb = 0; 1247 soisdisconnected(so); 1248 in6_pcbdetach(in6p); 1249 } 1250 #endif 1251 /* 1252 * pcb is no longer visble elsewhere, so we can safely release 1253 * the lock in callout_halt() if needed. 1254 */ 1255 TCP_STATINC(TCP_STAT_CLOSED); 1256 for (j = 0; j < TCPT_NTIMERS; j++) { 1257 callout_halt(&tp->t_timer[j], softnet_lock); 1258 callout_destroy(&tp->t_timer[j]); 1259 } 1260 callout_halt(&tp->t_delack_ch, softnet_lock); 1261 callout_destroy(&tp->t_delack_ch); 1262 pool_put(&tcpcb_pool, tp); 1263 1264 return ((struct tcpcb *)0); 1265 } 1266 1267 int 1268 tcp_freeq(struct tcpcb *tp) 1269 { 1270 struct ipqent *qe; 1271 int rv = 0; 1272 #ifdef TCPREASS_DEBUG 1273 int i = 0; 1274 #endif 1275 1276 TCP_REASS_LOCK_CHECK(tp); 1277 1278 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1279 #ifdef TCPREASS_DEBUG 1280 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n", 1281 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len, 1282 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST)); 1283 #endif 1284 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1285 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1286 m_freem(qe->ipqe_m); 1287 tcpipqent_free(qe); 1288 rv = 1; 1289 } 1290 tp->t_segqlen = 0; 1291 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1292 return (rv); 1293 } 1294 1295 /* 1296 * Protocol drain routine. Called when memory is in short supply. 1297 * Don't acquire softnet_lock as can be called from hardware 1298 * interrupt handler. 1299 */ 1300 void 1301 tcp_drain(void) 1302 { 1303 struct inpcb_hdr *inph; 1304 struct tcpcb *tp; 1305 1306 KERNEL_LOCK(1, NULL); 1307 1308 /* 1309 * Free the sequence queue of all TCP connections. 1310 */ 1311 CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1312 switch (inph->inph_af) { 1313 case AF_INET: 1314 tp = intotcpcb((struct inpcb *)inph); 1315 break; 1316 #ifdef INET6 1317 case AF_INET6: 1318 tp = in6totcpcb((struct in6pcb *)inph); 1319 break; 1320 #endif 1321 default: 1322 tp = NULL; 1323 break; 1324 } 1325 if (tp != NULL) { 1326 /* 1327 * We may be called from a device's interrupt 1328 * context. If the tcpcb is already busy, 1329 * just bail out now. 1330 */ 1331 if (tcp_reass_lock_try(tp) == 0) 1332 continue; 1333 if (tcp_freeq(tp)) 1334 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1335 TCP_REASS_UNLOCK(tp); 1336 } 1337 } 1338 1339 KERNEL_UNLOCK_ONE(NULL); 1340 } 1341 1342 /* 1343 * Notify a tcp user of an asynchronous error; 1344 * store error as soft error, but wake up user 1345 * (for now, won't do anything until can select for soft error). 1346 */ 1347 void 1348 tcp_notify(struct inpcb *inp, int error) 1349 { 1350 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1351 struct socket *so = inp->inp_socket; 1352 1353 /* 1354 * Ignore some errors if we are hooked up. 1355 * If connection hasn't completed, has retransmitted several times, 1356 * and receives a second error, give up now. This is better 1357 * than waiting a long time to establish a connection that 1358 * can never complete. 1359 */ 1360 if (tp->t_state == TCPS_ESTABLISHED && 1361 (error == EHOSTUNREACH || error == ENETUNREACH || 1362 error == EHOSTDOWN)) { 1363 return; 1364 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1365 tp->t_rxtshift > 3 && tp->t_softerror) 1366 so->so_error = error; 1367 else 1368 tp->t_softerror = error; 1369 cv_broadcast(&so->so_cv); 1370 sorwakeup(so); 1371 sowwakeup(so); 1372 } 1373 1374 #ifdef INET6 1375 void 1376 tcp6_notify(struct in6pcb *in6p, int error) 1377 { 1378 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1379 struct socket *so = in6p->in6p_socket; 1380 1381 /* 1382 * Ignore some errors if we are hooked up. 1383 * If connection hasn't completed, has retransmitted several times, 1384 * and receives a second error, give up now. This is better 1385 * than waiting a long time to establish a connection that 1386 * can never complete. 1387 */ 1388 if (tp->t_state == TCPS_ESTABLISHED && 1389 (error == EHOSTUNREACH || error == ENETUNREACH || 1390 error == EHOSTDOWN)) { 1391 return; 1392 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1393 tp->t_rxtshift > 3 && tp->t_softerror) 1394 so->so_error = error; 1395 else 1396 tp->t_softerror = error; 1397 cv_broadcast(&so->so_cv); 1398 sorwakeup(so); 1399 sowwakeup(so); 1400 } 1401 #endif 1402 1403 #ifdef INET6 1404 void * 1405 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1406 { 1407 struct tcphdr th; 1408 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1409 int nmatch; 1410 struct ip6_hdr *ip6; 1411 const struct sockaddr_in6 *sa6_src = NULL; 1412 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1413 struct mbuf *m; 1414 int off; 1415 1416 if (sa->sa_family != AF_INET6 || 1417 sa->sa_len != sizeof(struct sockaddr_in6)) 1418 return NULL; 1419 if ((unsigned)cmd >= PRC_NCMDS) 1420 return NULL; 1421 else if (cmd == PRC_QUENCH) { 1422 /* 1423 * Don't honor ICMP Source Quench messages meant for 1424 * TCP connections. 1425 */ 1426 return NULL; 1427 } else if (PRC_IS_REDIRECT(cmd)) 1428 notify = in6_rtchange, d = NULL; 1429 else if (cmd == PRC_MSGSIZE) 1430 ; /* special code is present, see below */ 1431 else if (cmd == PRC_HOSTDEAD) 1432 d = NULL; 1433 else if (inet6ctlerrmap[cmd] == 0) 1434 return NULL; 1435 1436 /* if the parameter is from icmp6, decode it. */ 1437 if (d != NULL) { 1438 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1439 m = ip6cp->ip6c_m; 1440 ip6 = ip6cp->ip6c_ip6; 1441 off = ip6cp->ip6c_off; 1442 sa6_src = ip6cp->ip6c_src; 1443 } else { 1444 m = NULL; 1445 ip6 = NULL; 1446 sa6_src = &sa6_any; 1447 off = 0; 1448 } 1449 1450 if (ip6) { 1451 /* 1452 * XXX: We assume that when ip6 is non NULL, 1453 * M and OFF are valid. 1454 */ 1455 1456 /* check if we can safely examine src and dst ports */ 1457 if (m->m_pkthdr.len < off + sizeof(th)) { 1458 if (cmd == PRC_MSGSIZE) 1459 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1460 return NULL; 1461 } 1462 1463 memset(&th, 0, sizeof(th)); 1464 m_copydata(m, off, sizeof(th), (void *)&th); 1465 1466 if (cmd == PRC_MSGSIZE) { 1467 int valid = 0; 1468 1469 /* 1470 * Check to see if we have a valid TCP connection 1471 * corresponding to the address in the ICMPv6 message 1472 * payload. 1473 */ 1474 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1475 th.th_dport, 1476 (const struct in6_addr *)&sa6_src->sin6_addr, 1477 th.th_sport, 0)) 1478 valid++; 1479 1480 /* 1481 * Depending on the value of "valid" and routing table 1482 * size (mtudisc_{hi,lo}wat), we will: 1483 * - recalcurate the new MTU and create the 1484 * corresponding routing entry, or 1485 * - ignore the MTU change notification. 1486 */ 1487 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1488 1489 /* 1490 * no need to call in6_pcbnotify, it should have been 1491 * called via callback if necessary 1492 */ 1493 return NULL; 1494 } 1495 1496 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1497 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1498 if (nmatch == 0 && syn_cache_count && 1499 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1500 inet6ctlerrmap[cmd] == ENETUNREACH || 1501 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1502 syn_cache_unreach((const struct sockaddr *)sa6_src, 1503 sa, &th); 1504 } else { 1505 (void) in6_pcbnotify(&tcbtable, sa, 0, 1506 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1507 } 1508 1509 return NULL; 1510 } 1511 #endif 1512 1513 #ifdef INET 1514 /* assumes that ip header and tcp header are contiguous on mbuf */ 1515 void * 1516 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1517 { 1518 struct ip *ip = v; 1519 struct tcphdr *th; 1520 struct icmp *icp; 1521 extern const int inetctlerrmap[]; 1522 void (*notify)(struct inpcb *, int) = tcp_notify; 1523 int errno; 1524 int nmatch; 1525 struct tcpcb *tp; 1526 u_int mtu; 1527 tcp_seq seq; 1528 struct inpcb *inp; 1529 #ifdef INET6 1530 struct in6pcb *in6p; 1531 struct in6_addr src6, dst6; 1532 #endif 1533 1534 if (sa->sa_family != AF_INET || 1535 sa->sa_len != sizeof(struct sockaddr_in)) 1536 return NULL; 1537 if ((unsigned)cmd >= PRC_NCMDS) 1538 return NULL; 1539 errno = inetctlerrmap[cmd]; 1540 if (cmd == PRC_QUENCH) 1541 /* 1542 * Don't honor ICMP Source Quench messages meant for 1543 * TCP connections. 1544 */ 1545 return NULL; 1546 else if (PRC_IS_REDIRECT(cmd)) 1547 notify = in_rtchange, ip = 0; 1548 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1549 /* 1550 * Check to see if we have a valid TCP connection 1551 * corresponding to the address in the ICMP message 1552 * payload. 1553 * 1554 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1555 */ 1556 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1557 #ifdef INET6 1558 memset(&src6, 0, sizeof(src6)); 1559 memset(&dst6, 0, sizeof(dst6)); 1560 src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff; 1561 memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr)); 1562 memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr)); 1563 #endif 1564 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1565 th->th_dport, ip->ip_src, th->th_sport)) != NULL) 1566 #ifdef INET6 1567 in6p = NULL; 1568 #else 1569 ; 1570 #endif 1571 #ifdef INET6 1572 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1573 th->th_dport, &src6, th->th_sport, 0)) != NULL) 1574 ; 1575 #endif 1576 else 1577 return NULL; 1578 1579 /* 1580 * Now that we've validated that we are actually communicating 1581 * with the host indicated in the ICMP message, locate the 1582 * ICMP header, recalculate the new MTU, and create the 1583 * corresponding routing entry. 1584 */ 1585 icp = (struct icmp *)((char *)ip - 1586 offsetof(struct icmp, icmp_ip)); 1587 if (inp) { 1588 if ((tp = intotcpcb(inp)) == NULL) 1589 return NULL; 1590 } 1591 #ifdef INET6 1592 else if (in6p) { 1593 if ((tp = in6totcpcb(in6p)) == NULL) 1594 return NULL; 1595 } 1596 #endif 1597 else 1598 return NULL; 1599 seq = ntohl(th->th_seq); 1600 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1601 return NULL; 1602 /* 1603 * If the ICMP message advertises a Next-Hop MTU 1604 * equal or larger than the maximum packet size we have 1605 * ever sent, drop the message. 1606 */ 1607 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1608 if (mtu >= tp->t_pmtud_mtu_sent) 1609 return NULL; 1610 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1611 /* 1612 * Calculate new MTU, and create corresponding 1613 * route (traditional PMTUD). 1614 */ 1615 tp->t_flags &= ~TF_PMTUD_PEND; 1616 icmp_mtudisc(icp, ip->ip_dst); 1617 } else { 1618 /* 1619 * Record the information got in the ICMP 1620 * message; act on it later. 1621 * If we had already recorded an ICMP message, 1622 * replace the old one only if the new message 1623 * refers to an older TCP segment 1624 */ 1625 if (tp->t_flags & TF_PMTUD_PEND) { 1626 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1627 return NULL; 1628 } else 1629 tp->t_flags |= TF_PMTUD_PEND; 1630 tp->t_pmtud_th_seq = seq; 1631 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1632 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1633 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1634 } 1635 return NULL; 1636 } else if (cmd == PRC_HOSTDEAD) 1637 ip = 0; 1638 else if (errno == 0) 1639 return NULL; 1640 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1641 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1642 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1643 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1644 if (nmatch == 0 && syn_cache_count && 1645 (inetctlerrmap[cmd] == EHOSTUNREACH || 1646 inetctlerrmap[cmd] == ENETUNREACH || 1647 inetctlerrmap[cmd] == EHOSTDOWN)) { 1648 struct sockaddr_in sin; 1649 memset(&sin, 0, sizeof(sin)); 1650 sin.sin_len = sizeof(sin); 1651 sin.sin_family = AF_INET; 1652 sin.sin_port = th->th_sport; 1653 sin.sin_addr = ip->ip_src; 1654 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1655 } 1656 1657 /* XXX mapped address case */ 1658 } else 1659 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1660 notify); 1661 return NULL; 1662 } 1663 1664 /* 1665 * When a source quench is received, we are being notified of congestion. 1666 * Close the congestion window down to the Loss Window (one segment). 1667 * We will gradually open it again as we proceed. 1668 */ 1669 void 1670 tcp_quench(struct inpcb *inp, int errno) 1671 { 1672 struct tcpcb *tp = intotcpcb(inp); 1673 1674 if (tp) { 1675 tp->snd_cwnd = tp->t_segsz; 1676 tp->t_bytes_acked = 0; 1677 } 1678 } 1679 #endif 1680 1681 #ifdef INET6 1682 void 1683 tcp6_quench(struct in6pcb *in6p, int errno) 1684 { 1685 struct tcpcb *tp = in6totcpcb(in6p); 1686 1687 if (tp) { 1688 tp->snd_cwnd = tp->t_segsz; 1689 tp->t_bytes_acked = 0; 1690 } 1691 } 1692 #endif 1693 1694 #ifdef INET 1695 /* 1696 * Path MTU Discovery handlers. 1697 */ 1698 void 1699 tcp_mtudisc_callback(struct in_addr faddr) 1700 { 1701 #ifdef INET6 1702 struct in6_addr in6; 1703 #endif 1704 1705 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1706 #ifdef INET6 1707 memset(&in6, 0, sizeof(in6)); 1708 in6.s6_addr16[5] = 0xffff; 1709 memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr)); 1710 tcp6_mtudisc_callback(&in6); 1711 #endif 1712 } 1713 1714 /* 1715 * On receipt of path MTU corrections, flush old route and replace it 1716 * with the new one. Retransmit all unacknowledged packets, to ensure 1717 * that all packets will be received. 1718 */ 1719 void 1720 tcp_mtudisc(struct inpcb *inp, int errno) 1721 { 1722 struct tcpcb *tp = intotcpcb(inp); 1723 struct rtentry *rt = in_pcbrtentry(inp); 1724 1725 if (tp != 0) { 1726 if (rt != 0) { 1727 /* 1728 * If this was not a host route, remove and realloc. 1729 */ 1730 if ((rt->rt_flags & RTF_HOST) == 0) { 1731 in_rtchange(inp, errno); 1732 if ((rt = in_pcbrtentry(inp)) == 0) 1733 return; 1734 } 1735 1736 /* 1737 * Slow start out of the error condition. We 1738 * use the MTU because we know it's smaller 1739 * than the previously transmitted segment. 1740 * 1741 * Note: This is more conservative than the 1742 * suggestion in draft-floyd-incr-init-win-03. 1743 */ 1744 if (rt->rt_rmx.rmx_mtu != 0) 1745 tp->snd_cwnd = 1746 TCP_INITIAL_WINDOW(tcp_init_win, 1747 rt->rt_rmx.rmx_mtu); 1748 } 1749 1750 /* 1751 * Resend unacknowledged packets. 1752 */ 1753 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1754 tcp_output(tp); 1755 } 1756 } 1757 #endif 1758 1759 #ifdef INET6 1760 /* 1761 * Path MTU Discovery handlers. 1762 */ 1763 void 1764 tcp6_mtudisc_callback(struct in6_addr *faddr) 1765 { 1766 struct sockaddr_in6 sin6; 1767 1768 memset(&sin6, 0, sizeof(sin6)); 1769 sin6.sin6_family = AF_INET6; 1770 sin6.sin6_len = sizeof(struct sockaddr_in6); 1771 sin6.sin6_addr = *faddr; 1772 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1773 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1774 } 1775 1776 void 1777 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1778 { 1779 struct tcpcb *tp = in6totcpcb(in6p); 1780 struct rtentry *rt = in6_pcbrtentry(in6p); 1781 1782 if (tp != 0) { 1783 if (rt != 0) { 1784 /* 1785 * If this was not a host route, remove and realloc. 1786 */ 1787 if ((rt->rt_flags & RTF_HOST) == 0) { 1788 in6_rtchange(in6p, errno); 1789 if ((rt = in6_pcbrtentry(in6p)) == 0) 1790 return; 1791 } 1792 1793 /* 1794 * Slow start out of the error condition. We 1795 * use the MTU because we know it's smaller 1796 * than the previously transmitted segment. 1797 * 1798 * Note: This is more conservative than the 1799 * suggestion in draft-floyd-incr-init-win-03. 1800 */ 1801 if (rt->rt_rmx.rmx_mtu != 0) 1802 tp->snd_cwnd = 1803 TCP_INITIAL_WINDOW(tcp_init_win, 1804 rt->rt_rmx.rmx_mtu); 1805 } 1806 1807 /* 1808 * Resend unacknowledged packets. 1809 */ 1810 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1811 tcp_output(tp); 1812 } 1813 } 1814 #endif /* INET6 */ 1815 1816 /* 1817 * Compute the MSS to advertise to the peer. Called only during 1818 * the 3-way handshake. If we are the server (peer initiated 1819 * connection), we are called with a pointer to the interface 1820 * on which the SYN packet arrived. If we are the client (we 1821 * initiated connection), we are called with a pointer to the 1822 * interface out which this connection should go. 1823 * 1824 * NOTE: Do not subtract IP option/extension header size nor IPsec 1825 * header size from MSS advertisement. MSS option must hold the maximum 1826 * segment size we can accept, so it must always be: 1827 * max(if mtu) - ip header - tcp header 1828 */ 1829 u_long 1830 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1831 { 1832 extern u_long in_maxmtu; 1833 u_long mss = 0; 1834 u_long hdrsiz; 1835 1836 /* 1837 * In order to avoid defeating path MTU discovery on the peer, 1838 * we advertise the max MTU of all attached networks as our MSS, 1839 * per RFC 1191, section 3.1. 1840 * 1841 * We provide the option to advertise just the MTU of 1842 * the interface on which we hope this connection will 1843 * be receiving. If we are responding to a SYN, we 1844 * will have a pretty good idea about this, but when 1845 * initiating a connection there is a bit more doubt. 1846 * 1847 * We also need to ensure that loopback has a large enough 1848 * MSS, as the loopback MTU is never included in in_maxmtu. 1849 */ 1850 1851 if (ifp != NULL) 1852 switch (af) { 1853 case AF_INET: 1854 mss = ifp->if_mtu; 1855 break; 1856 #ifdef INET6 1857 case AF_INET6: 1858 mss = IN6_LINKMTU(ifp); 1859 break; 1860 #endif 1861 } 1862 1863 if (tcp_mss_ifmtu == 0) 1864 switch (af) { 1865 case AF_INET: 1866 mss = max(in_maxmtu, mss); 1867 break; 1868 #ifdef INET6 1869 case AF_INET6: 1870 mss = max(in6_maxmtu, mss); 1871 break; 1872 #endif 1873 } 1874 1875 switch (af) { 1876 case AF_INET: 1877 hdrsiz = sizeof(struct ip); 1878 break; 1879 #ifdef INET6 1880 case AF_INET6: 1881 hdrsiz = sizeof(struct ip6_hdr); 1882 break; 1883 #endif 1884 default: 1885 hdrsiz = 0; 1886 break; 1887 } 1888 hdrsiz += sizeof(struct tcphdr); 1889 if (mss > hdrsiz) 1890 mss -= hdrsiz; 1891 1892 mss = max(tcp_mssdflt, mss); 1893 return (mss); 1894 } 1895 1896 /* 1897 * Set connection variables based on the peer's advertised MSS. 1898 * We are passed the TCPCB for the actual connection. If we 1899 * are the server, we are called by the compressed state engine 1900 * when the 3-way handshake is complete. If we are the client, 1901 * we are called when we receive the SYN,ACK from the server. 1902 * 1903 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1904 * before this routine is called! 1905 */ 1906 void 1907 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1908 { 1909 struct socket *so; 1910 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1911 struct rtentry *rt; 1912 #endif 1913 u_long bufsize; 1914 int mss; 1915 1916 #ifdef DIAGNOSTIC 1917 if (tp->t_inpcb && tp->t_in6pcb) 1918 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set"); 1919 #endif 1920 so = NULL; 1921 rt = NULL; 1922 #ifdef INET 1923 if (tp->t_inpcb) { 1924 so = tp->t_inpcb->inp_socket; 1925 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1926 rt = in_pcbrtentry(tp->t_inpcb); 1927 #endif 1928 } 1929 #endif 1930 #ifdef INET6 1931 if (tp->t_in6pcb) { 1932 so = tp->t_in6pcb->in6p_socket; 1933 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1934 rt = in6_pcbrtentry(tp->t_in6pcb); 1935 #endif 1936 } 1937 #endif 1938 1939 /* 1940 * As per RFC1122, use the default MSS value, unless they 1941 * sent us an offer. Do not accept offers less than 256 bytes. 1942 */ 1943 mss = tcp_mssdflt; 1944 if (offer) 1945 mss = offer; 1946 mss = max(mss, 256); /* sanity */ 1947 tp->t_peermss = mss; 1948 mss -= tcp_optlen(tp); 1949 #ifdef INET 1950 if (tp->t_inpcb) 1951 mss -= ip_optlen(tp->t_inpcb); 1952 #endif 1953 #ifdef INET6 1954 if (tp->t_in6pcb) 1955 mss -= ip6_optlen(tp->t_in6pcb); 1956 #endif 1957 1958 /* 1959 * If there's a pipesize, change the socket buffer to that size. 1960 * Make the socket buffer an integral number of MSS units. If 1961 * the MSS is larger than the socket buffer, artificially decrease 1962 * the MSS. 1963 */ 1964 #ifdef RTV_SPIPE 1965 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1966 bufsize = rt->rt_rmx.rmx_sendpipe; 1967 else 1968 #endif 1969 { 1970 KASSERT(so != NULL); 1971 bufsize = so->so_snd.sb_hiwat; 1972 } 1973 if (bufsize < mss) 1974 mss = bufsize; 1975 else { 1976 bufsize = roundup(bufsize, mss); 1977 if (bufsize > sb_max) 1978 bufsize = sb_max; 1979 (void) sbreserve(&so->so_snd, bufsize, so); 1980 } 1981 tp->t_segsz = mss; 1982 1983 #ifdef RTV_SSTHRESH 1984 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1985 /* 1986 * There's some sort of gateway or interface buffer 1987 * limit on the path. Use this to set the slow 1988 * start threshold, but set the threshold to no less 1989 * than 2 * MSS. 1990 */ 1991 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1992 } 1993 #endif 1994 } 1995 1996 /* 1997 * Processing necessary when a TCP connection is established. 1998 */ 1999 void 2000 tcp_established(struct tcpcb *tp) 2001 { 2002 struct socket *so; 2003 #ifdef RTV_RPIPE 2004 struct rtentry *rt; 2005 #endif 2006 u_long bufsize; 2007 2008 #ifdef DIAGNOSTIC 2009 if (tp->t_inpcb && tp->t_in6pcb) 2010 panic("tcp_established: both t_inpcb and t_in6pcb are set"); 2011 #endif 2012 so = NULL; 2013 rt = NULL; 2014 #ifdef INET 2015 if (tp->t_inpcb) { 2016 so = tp->t_inpcb->inp_socket; 2017 #if defined(RTV_RPIPE) 2018 rt = in_pcbrtentry(tp->t_inpcb); 2019 #endif 2020 } 2021 #endif 2022 #ifdef INET6 2023 if (tp->t_in6pcb) { 2024 so = tp->t_in6pcb->in6p_socket; 2025 #if defined(RTV_RPIPE) 2026 rt = in6_pcbrtentry(tp->t_in6pcb); 2027 #endif 2028 } 2029 #endif 2030 2031 tp->t_state = TCPS_ESTABLISHED; 2032 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2033 2034 #ifdef RTV_RPIPE 2035 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2036 bufsize = rt->rt_rmx.rmx_recvpipe; 2037 else 2038 #endif 2039 { 2040 KASSERT(so != NULL); 2041 bufsize = so->so_rcv.sb_hiwat; 2042 } 2043 if (bufsize > tp->t_ourmss) { 2044 bufsize = roundup(bufsize, tp->t_ourmss); 2045 if (bufsize > sb_max) 2046 bufsize = sb_max; 2047 (void) sbreserve(&so->so_rcv, bufsize, so); 2048 } 2049 } 2050 2051 /* 2052 * Check if there's an initial rtt or rttvar. Convert from the 2053 * route-table units to scaled multiples of the slow timeout timer. 2054 * Called only during the 3-way handshake. 2055 */ 2056 void 2057 tcp_rmx_rtt(struct tcpcb *tp) 2058 { 2059 #ifdef RTV_RTT 2060 struct rtentry *rt = NULL; 2061 int rtt; 2062 2063 #ifdef DIAGNOSTIC 2064 if (tp->t_inpcb && tp->t_in6pcb) 2065 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set"); 2066 #endif 2067 #ifdef INET 2068 if (tp->t_inpcb) 2069 rt = in_pcbrtentry(tp->t_inpcb); 2070 #endif 2071 #ifdef INET6 2072 if (tp->t_in6pcb) 2073 rt = in6_pcbrtentry(tp->t_in6pcb); 2074 #endif 2075 if (rt == NULL) 2076 return; 2077 2078 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2079 /* 2080 * XXX The lock bit for MTU indicates that the value 2081 * is also a minimum value; this is subject to time. 2082 */ 2083 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2084 TCPT_RANGESET(tp->t_rttmin, 2085 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2086 TCPTV_MIN, TCPTV_REXMTMAX); 2087 tp->t_srtt = rtt / 2088 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2089 if (rt->rt_rmx.rmx_rttvar) { 2090 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2091 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2092 (TCP_RTTVAR_SHIFT + 2)); 2093 } else { 2094 /* Default variation is +- 1 rtt */ 2095 tp->t_rttvar = 2096 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2097 } 2098 TCPT_RANGESET(tp->t_rxtcur, 2099 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2100 tp->t_rttmin, TCPTV_REXMTMAX); 2101 } 2102 #endif 2103 } 2104 2105 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2106 #if NRND > 0 2107 u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2108 #endif 2109 2110 /* 2111 * Get a new sequence value given a tcp control block 2112 */ 2113 tcp_seq 2114 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2115 { 2116 2117 #ifdef INET 2118 if (tp->t_inpcb != NULL) { 2119 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2120 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2121 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2122 addin)); 2123 } 2124 #endif 2125 #ifdef INET6 2126 if (tp->t_in6pcb != NULL) { 2127 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2128 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2129 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2130 addin)); 2131 } 2132 #endif 2133 /* Not possible. */ 2134 panic("tcp_new_iss"); 2135 } 2136 2137 /* 2138 * This routine actually generates a new TCP initial sequence number. 2139 */ 2140 tcp_seq 2141 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2142 size_t addrsz, tcp_seq addin) 2143 { 2144 tcp_seq tcp_iss; 2145 2146 #if NRND > 0 2147 static bool tcp_iss_gotten_secret; 2148 2149 /* 2150 * If we haven't been here before, initialize our cryptographic 2151 * hash secret. 2152 */ 2153 if (tcp_iss_gotten_secret == false) { 2154 rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret), 2155 RND_EXTRACT_ANY); 2156 tcp_iss_gotten_secret = true; 2157 } 2158 2159 if (tcp_do_rfc1948) { 2160 MD5_CTX ctx; 2161 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2162 2163 /* 2164 * Compute the base value of the ISS. It is a hash 2165 * of (saddr, sport, daddr, dport, secret). 2166 */ 2167 MD5Init(&ctx); 2168 2169 MD5Update(&ctx, (u_char *) laddr, addrsz); 2170 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2171 2172 MD5Update(&ctx, (u_char *) faddr, addrsz); 2173 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2174 2175 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2176 2177 MD5Final(hash, &ctx); 2178 2179 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2180 2181 /* 2182 * Now increment our "timer", and add it in to 2183 * the computed value. 2184 * 2185 * XXX Use `addin'? 2186 * XXX TCP_ISSINCR too large to use? 2187 */ 2188 tcp_iss_seq += TCP_ISSINCR; 2189 #ifdef TCPISS_DEBUG 2190 printf("ISS hash 0x%08x, ", tcp_iss); 2191 #endif 2192 tcp_iss += tcp_iss_seq + addin; 2193 #ifdef TCPISS_DEBUG 2194 printf("new ISS 0x%08x\n", tcp_iss); 2195 #endif 2196 } else 2197 #endif /* NRND > 0 */ 2198 { 2199 /* 2200 * Randomize. 2201 */ 2202 #if NRND > 0 2203 rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY); 2204 #else 2205 tcp_iss = arc4random(); 2206 #endif 2207 2208 /* 2209 * If we were asked to add some amount to a known value, 2210 * we will take a random value obtained above, mask off 2211 * the upper bits, and add in the known value. We also 2212 * add in a constant to ensure that we are at least a 2213 * certain distance from the original value. 2214 * 2215 * This is used when an old connection is in timed wait 2216 * and we have a new one coming in, for instance. 2217 */ 2218 if (addin != 0) { 2219 #ifdef TCPISS_DEBUG 2220 printf("Random %08x, ", tcp_iss); 2221 #endif 2222 tcp_iss &= TCP_ISS_RANDOM_MASK; 2223 tcp_iss += addin + TCP_ISSINCR; 2224 #ifdef TCPISS_DEBUG 2225 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2226 #endif 2227 } else { 2228 tcp_iss &= TCP_ISS_RANDOM_MASK; 2229 tcp_iss += tcp_iss_seq; 2230 tcp_iss_seq += TCP_ISSINCR; 2231 #ifdef TCPISS_DEBUG 2232 printf("ISS %08x\n", tcp_iss); 2233 #endif 2234 } 2235 } 2236 2237 if (tcp_compat_42) { 2238 /* 2239 * Limit it to the positive range for really old TCP 2240 * implementations. 2241 * Just AND off the top bit instead of checking if 2242 * is set first - saves a branch 50% of the time. 2243 */ 2244 tcp_iss &= 0x7fffffff; /* XXX */ 2245 } 2246 2247 return (tcp_iss); 2248 } 2249 2250 #if defined(IPSEC) || defined(FAST_IPSEC) 2251 /* compute ESP/AH header size for TCP, including outer IP header. */ 2252 size_t 2253 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2254 { 2255 struct inpcb *inp; 2256 size_t hdrsiz; 2257 2258 /* XXX mapped addr case (tp->t_in6pcb) */ 2259 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2260 return 0; 2261 switch (tp->t_family) { 2262 case AF_INET: 2263 /* XXX: should use currect direction. */ 2264 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2265 break; 2266 default: 2267 hdrsiz = 0; 2268 break; 2269 } 2270 2271 return hdrsiz; 2272 } 2273 2274 #ifdef INET6 2275 size_t 2276 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2277 { 2278 struct in6pcb *in6p; 2279 size_t hdrsiz; 2280 2281 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2282 return 0; 2283 switch (tp->t_family) { 2284 case AF_INET6: 2285 /* XXX: should use currect direction. */ 2286 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2287 break; 2288 case AF_INET: 2289 /* mapped address case - tricky */ 2290 default: 2291 hdrsiz = 0; 2292 break; 2293 } 2294 2295 return hdrsiz; 2296 } 2297 #endif 2298 #endif /*IPSEC*/ 2299 2300 /* 2301 * Determine the length of the TCP options for this connection. 2302 * 2303 * XXX: What do we do for SACK, when we add that? Just reserve 2304 * all of the space? Otherwise we can't exactly be incrementing 2305 * cwnd by an amount that varies depending on the amount we last 2306 * had to SACK! 2307 */ 2308 2309 u_int 2310 tcp_optlen(struct tcpcb *tp) 2311 { 2312 u_int optlen; 2313 2314 optlen = 0; 2315 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2316 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2317 optlen += TCPOLEN_TSTAMP_APPA; 2318 2319 #ifdef TCP_SIGNATURE 2320 if (tp->t_flags & TF_SIGNATURE) 2321 optlen += TCPOLEN_SIGNATURE + 2; 2322 #endif /* TCP_SIGNATURE */ 2323 2324 return optlen; 2325 } 2326 2327 u_int 2328 tcp_hdrsz(struct tcpcb *tp) 2329 { 2330 u_int hlen; 2331 2332 switch (tp->t_family) { 2333 #ifdef INET6 2334 case AF_INET6: 2335 hlen = sizeof(struct ip6_hdr); 2336 break; 2337 #endif 2338 case AF_INET: 2339 hlen = sizeof(struct ip); 2340 break; 2341 default: 2342 hlen = 0; 2343 break; 2344 } 2345 hlen += sizeof(struct tcphdr); 2346 2347 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2348 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2349 hlen += TCPOLEN_TSTAMP_APPA; 2350 #ifdef TCP_SIGNATURE 2351 if (tp->t_flags & TF_SIGNATURE) 2352 hlen += TCPOLEN_SIGLEN; 2353 #endif 2354 return hlen; 2355 } 2356 2357 void 2358 tcp_statinc(u_int stat) 2359 { 2360 2361 KASSERT(stat < TCP_NSTATS); 2362 TCP_STATINC(stat); 2363 } 2364 2365 void 2366 tcp_statadd(u_int stat, uint64_t val) 2367 { 2368 2369 KASSERT(stat < TCP_NSTATS); 2370 TCP_STATADD(stat, val); 2371 } 2372