1 /* $NetBSD: tcp_subr.c,v 1.289 2021/07/31 20:29:37 andvar 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 1.289 2021/07/31 20:29:37 andvar Exp $"); 95 96 #ifdef _KERNEL_OPT 97 #include "opt_inet.h" 98 #include "opt_ipsec.h" 99 #include "opt_inet_csum.h" 100 #include "opt_mbuftrace.h" 101 #endif 102 103 #include <sys/param.h> 104 #include <sys/atomic.h> 105 #include <sys/proc.h> 106 #include <sys/systm.h> 107 #include <sys/mbuf.h> 108 #include <sys/once.h> 109 #include <sys/socket.h> 110 #include <sys/socketvar.h> 111 #include <sys/protosw.h> 112 #include <sys/errno.h> 113 #include <sys/kernel.h> 114 #include <sys/pool.h> 115 #include <sys/md5.h> 116 #include <sys/cprng.h> 117 118 #include <net/route.h> 119 #include <net/if.h> 120 121 #include <netinet/in.h> 122 #include <netinet/in_systm.h> 123 #include <netinet/ip.h> 124 #include <netinet/in_pcb.h> 125 #include <netinet/ip_var.h> 126 #include <netinet/ip_icmp.h> 127 128 #ifdef INET6 129 #include <netinet/ip6.h> 130 #include <netinet6/in6_pcb.h> 131 #include <netinet6/ip6_var.h> 132 #include <netinet6/in6_var.h> 133 #include <netinet6/ip6protosw.h> 134 #include <netinet/icmp6.h> 135 #include <netinet6/nd6.h> 136 #endif 137 138 #include <netinet/tcp.h> 139 #include <netinet/tcp_fsm.h> 140 #include <netinet/tcp_seq.h> 141 #include <netinet/tcp_timer.h> 142 #include <netinet/tcp_var.h> 143 #include <netinet/tcp_vtw.h> 144 #include <netinet/tcp_private.h> 145 #include <netinet/tcp_congctl.h> 146 147 #ifdef IPSEC 148 #include <netipsec/ipsec.h> 149 #ifdef INET6 150 #include <netipsec/ipsec6.h> 151 #endif 152 #include <netipsec/key.h> 153 #endif 154 155 156 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 157 u_int32_t tcp_now; /* slow ticks, for RFC 1323 timestamps */ 158 159 percpu_t *tcpstat_percpu; 160 161 /* patchable/settable parameters for tcp */ 162 int tcp_mssdflt = TCP_MSS; 163 int tcp_minmss = TCP_MINMSS; 164 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 165 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 166 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 167 int tcp_do_sack = 1; /* selective acknowledgement */ 168 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 169 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 170 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 171 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 172 #ifndef TCP_INIT_WIN 173 #define TCP_INIT_WIN 4 /* initial slow start window */ 174 #endif 175 #ifndef TCP_INIT_WIN_LOCAL 176 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 177 #endif 178 /* 179 * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460. 180 * This is to simulate current behavior for iw == 4 181 */ 182 int tcp_init_win_max[] = { 183 1 * 1460, 184 1 * 1460, 185 2 * 1460, 186 2 * 1460, 187 3 * 1460, 188 5 * 1460, 189 6 * 1460, 190 7 * 1460, 191 8 * 1460, 192 9 * 1460, 193 10 * 1460 194 }; 195 int tcp_init_win = TCP_INIT_WIN; 196 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 197 int tcp_mss_ifmtu = 0; 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 int tcp_msl_enable = 1; /* enable TIME_WAIT truncation */ 208 int tcp_msl_loop = PR_SLOWHZ; /* MSL for loopback */ 209 int tcp_msl_local = 5 * PR_SLOWHZ; /* MSL for 'local' */ 210 int tcp_msl_remote = TCPTV_MSL; /* MSL otherwise */ 211 int tcp_msl_remote_threshold = TCPTV_SRTTDFLT; /* RTT threshold */ 212 int tcp_rttlocal = 0; /* Use RTT to decide who's 'local' */ 213 214 int tcp4_vtw_enable = 0; /* 1 to enable */ 215 int tcp6_vtw_enable = 0; /* 1 to enable */ 216 int tcp_vtw_was_enabled = 0; 217 int tcp_vtw_entries = 1 << 4; /* 16 vestigial TIME_WAIT entries */ 218 219 /* tcb hash */ 220 #ifndef TCBHASHSIZE 221 #define TCBHASHSIZE 128 222 #endif 223 int tcbhashsize = TCBHASHSIZE; 224 225 /* syn hash parameters */ 226 #define TCP_SYN_HASH_SIZE 293 227 #define TCP_SYN_BUCKET_SIZE 35 228 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 229 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 230 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 231 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 232 233 int tcp_freeq(struct tcpcb *); 234 static int tcp_iss_secret_init(void); 235 236 static void tcp_mtudisc_callback(struct in_addr); 237 238 #ifdef INET6 239 static void tcp6_mtudisc(struct in6pcb *, int); 240 #endif 241 242 static struct pool tcpcb_pool; 243 244 static int tcp_drainwanted; 245 246 #ifdef TCP_CSUM_COUNTERS 247 #include <sys/device.h> 248 249 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 250 NULL, "tcp", "hwcsum bad"); 251 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 252 NULL, "tcp", "hwcsum ok"); 253 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 254 NULL, "tcp", "hwcsum data"); 255 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 256 NULL, "tcp", "swcsum"); 257 258 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 259 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 260 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 261 EVCNT_ATTACH_STATIC(tcp_swcsum); 262 263 #if defined(INET6) 264 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 265 NULL, "tcp6", "hwcsum bad"); 266 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 NULL, "tcp6", "hwcsum ok"); 268 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 269 NULL, "tcp6", "hwcsum data"); 270 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 271 NULL, "tcp6", "swcsum"); 272 273 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 274 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 275 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 276 EVCNT_ATTACH_STATIC(tcp6_swcsum); 277 #endif /* defined(INET6) */ 278 #endif /* TCP_CSUM_COUNTERS */ 279 280 281 #ifdef TCP_OUTPUT_COUNTERS 282 #include <sys/device.h> 283 284 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 285 NULL, "tcp", "output big header"); 286 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 287 NULL, "tcp", "output predict hit"); 288 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 289 NULL, "tcp", "output predict miss"); 290 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 291 NULL, "tcp", "output copy small"); 292 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 293 NULL, "tcp", "output copy big"); 294 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 295 NULL, "tcp", "output reference big"); 296 297 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 298 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 299 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 300 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 301 EVCNT_ATTACH_STATIC(tcp_output_copybig); 302 EVCNT_ATTACH_STATIC(tcp_output_refbig); 303 304 #endif /* TCP_OUTPUT_COUNTERS */ 305 306 #ifdef TCP_REASS_COUNTERS 307 #include <sys/device.h> 308 309 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 310 NULL, "tcp_reass", "calls"); 311 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 312 &tcp_reass_, "tcp_reass", "insert into empty queue"); 313 struct evcnt tcp_reass_iteration[8] = { 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 318 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 319 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 321 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 322 }; 323 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 324 &tcp_reass_, "tcp_reass", "prepend to first"); 325 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 326 &tcp_reass_, "tcp_reass", "prepend"); 327 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 328 &tcp_reass_, "tcp_reass", "insert"); 329 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 330 &tcp_reass_, "tcp_reass", "insert at tail"); 331 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 332 &tcp_reass_, "tcp_reass", "append"); 333 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 334 &tcp_reass_, "tcp_reass", "append to tail fragment"); 335 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 336 &tcp_reass_, "tcp_reass", "overlap at end"); 337 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 338 &tcp_reass_, "tcp_reass", "overlap at start"); 339 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 340 &tcp_reass_, "tcp_reass", "duplicate segment"); 341 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 342 &tcp_reass_, "tcp_reass", "duplicate fragment"); 343 344 EVCNT_ATTACH_STATIC(tcp_reass_); 345 EVCNT_ATTACH_STATIC(tcp_reass_empty); 346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 347 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 349 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 350 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 351 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 352 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 353 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 354 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 355 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 356 EVCNT_ATTACH_STATIC(tcp_reass_insert); 357 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 358 EVCNT_ATTACH_STATIC(tcp_reass_append); 359 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 360 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 361 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 362 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 363 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 364 365 #endif /* TCP_REASS_COUNTERS */ 366 367 #ifdef MBUFTRACE 368 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 369 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 370 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 371 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 372 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 373 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 374 #endif 375 376 static int 377 do_tcpinit(void) 378 { 379 380 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 381 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 382 NULL, IPL_SOFTNET); 383 384 tcp_usrreq_init(); 385 386 /* Initialize timer state. */ 387 tcp_timer_init(); 388 389 /* Initialize the compressed state engine. */ 390 syn_cache_init(); 391 392 /* Initialize the congestion control algorithms. */ 393 tcp_congctl_init(); 394 395 /* Initialize the TCPCB template. */ 396 tcp_tcpcb_template(); 397 398 /* Initialize reassembly queue */ 399 tcpipqent_init(); 400 401 /* SACK */ 402 tcp_sack_init(); 403 404 MOWNER_ATTACH(&tcp_tx_mowner); 405 MOWNER_ATTACH(&tcp_rx_mowner); 406 MOWNER_ATTACH(&tcp_reass_mowner); 407 MOWNER_ATTACH(&tcp_sock_mowner); 408 MOWNER_ATTACH(&tcp_sock_tx_mowner); 409 MOWNER_ATTACH(&tcp_sock_rx_mowner); 410 MOWNER_ATTACH(&tcp_mowner); 411 412 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS); 413 414 vtw_earlyinit(); 415 416 tcp_slowtimo_init(); 417 418 return 0; 419 } 420 421 void 422 tcp_init_common(unsigned basehlen) 423 { 424 static ONCE_DECL(dotcpinit); 425 unsigned hlen = basehlen + sizeof(struct tcphdr); 426 unsigned oldhlen; 427 428 if (max_linkhdr + hlen > MHLEN) 429 panic("tcp_init"); 430 while ((oldhlen = max_protohdr) < hlen) 431 atomic_cas_uint(&max_protohdr, oldhlen, hlen); 432 433 RUN_ONCE(&dotcpinit, do_tcpinit); 434 } 435 436 /* 437 * Tcp initialization 438 */ 439 void 440 tcp_init(void) 441 { 442 443 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 444 445 tcp_init_common(sizeof(struct ip)); 446 } 447 448 /* 449 * Create template to be used to send tcp packets on a connection. 450 * Call after host entry created, allocates an mbuf and fills 451 * in a skeletal tcp/ip header, minimizing the amount of work 452 * necessary when the connection is used. 453 */ 454 struct mbuf * 455 tcp_template(struct tcpcb *tp) 456 { 457 struct inpcb *inp = tp->t_inpcb; 458 #ifdef INET6 459 struct in6pcb *in6p = tp->t_in6pcb; 460 #endif 461 struct tcphdr *n; 462 struct mbuf *m; 463 int hlen; 464 465 switch (tp->t_family) { 466 case AF_INET: 467 hlen = sizeof(struct ip); 468 if (inp) 469 break; 470 #ifdef INET6 471 if (in6p) { 472 /* mapped addr case */ 473 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 474 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 475 break; 476 } 477 #endif 478 return NULL; /*EINVAL*/ 479 #ifdef INET6 480 case AF_INET6: 481 hlen = sizeof(struct ip6_hdr); 482 if (in6p) { 483 /* more sainty check? */ 484 break; 485 } 486 return NULL; /*EINVAL*/ 487 #endif 488 default: 489 return NULL; /*EAFNOSUPPORT*/ 490 } 491 492 KASSERT(hlen + sizeof(struct tcphdr) <= MCLBYTES); 493 494 m = tp->t_template; 495 if (m && m->m_len == hlen + sizeof(struct tcphdr)) { 496 ; 497 } else { 498 if (m) 499 m_freem(m); 500 m = tp->t_template = NULL; 501 MGETHDR(m, M_DONTWAIT, MT_HEADER); 502 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 503 MCLGET(m, M_DONTWAIT); 504 if ((m->m_flags & M_EXT) == 0) { 505 m_free(m); 506 m = NULL; 507 } 508 } 509 if (m == NULL) 510 return NULL; 511 MCLAIM(m, &tcp_mowner); 512 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 513 } 514 515 memset(mtod(m, void *), 0, m->m_len); 516 517 n = (struct tcphdr *)(mtod(m, char *) + hlen); 518 519 switch (tp->t_family) { 520 case AF_INET: 521 { 522 struct ipovly *ipov; 523 mtod(m, struct ip *)->ip_v = 4; 524 mtod(m, struct ip *)->ip_hl = hlen >> 2; 525 ipov = mtod(m, struct ipovly *); 526 ipov->ih_pr = IPPROTO_TCP; 527 ipov->ih_len = htons(sizeof(struct tcphdr)); 528 if (inp) { 529 ipov->ih_src = inp->inp_laddr; 530 ipov->ih_dst = inp->inp_faddr; 531 } 532 #ifdef INET6 533 else if (in6p) { 534 /* mapped addr case */ 535 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 536 sizeof(ipov->ih_src)); 537 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 538 sizeof(ipov->ih_dst)); 539 } 540 #endif 541 542 /* 543 * Compute the pseudo-header portion of the checksum 544 * now. We incrementally add in the TCP option and 545 * payload lengths later, and then compute the TCP 546 * checksum right before the packet is sent off onto 547 * the wire. 548 */ 549 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 550 ipov->ih_dst.s_addr, 551 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 552 break; 553 } 554 #ifdef INET6 555 case AF_INET6: 556 { 557 struct ip6_hdr *ip6; 558 mtod(m, struct ip *)->ip_v = 6; 559 ip6 = mtod(m, struct ip6_hdr *); 560 ip6->ip6_nxt = IPPROTO_TCP; 561 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 562 ip6->ip6_src = in6p->in6p_laddr; 563 ip6->ip6_dst = in6p->in6p_faddr; 564 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 565 if (ip6_auto_flowlabel) { 566 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 567 ip6->ip6_flow |= 568 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 569 } 570 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 571 ip6->ip6_vfc |= IPV6_VERSION; 572 573 /* 574 * Compute the pseudo-header portion of the checksum 575 * now. We incrementally add in the TCP option and 576 * payload lengths later, and then compute the TCP 577 * checksum right before the packet is sent off onto 578 * the wire. 579 */ 580 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 581 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 582 htonl(IPPROTO_TCP)); 583 break; 584 } 585 #endif 586 } 587 588 if (inp) { 589 n->th_sport = inp->inp_lport; 590 n->th_dport = inp->inp_fport; 591 } 592 #ifdef INET6 593 else if (in6p) { 594 n->th_sport = in6p->in6p_lport; 595 n->th_dport = in6p->in6p_fport; 596 } 597 #endif 598 599 n->th_seq = 0; 600 n->th_ack = 0; 601 n->th_x2 = 0; 602 n->th_off = 5; 603 n->th_flags = 0; 604 n->th_win = 0; 605 n->th_urp = 0; 606 return m; 607 } 608 609 /* 610 * Send a single message to the TCP at address specified by 611 * the given TCP/IP header. If m == 0, then we make a copy 612 * of the tcpiphdr at ti and send directly to the addressed host. 613 * This is used to force keep alive messages out using the TCP 614 * template for a connection tp->t_template. If flags are given 615 * then we send a message back to the TCP which originated the 616 * segment ti, and discard the mbuf containing it and any other 617 * attached mbufs. 618 * 619 * In any case the ack and sequence number of the transmitted 620 * segment are as specified by the parameters. 621 */ 622 int 623 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m, 624 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 625 { 626 struct route *ro; 627 int error, tlen, win = 0; 628 int hlen; 629 struct ip *ip; 630 #ifdef INET6 631 struct ip6_hdr *ip6; 632 #endif 633 int family; /* family on packet, not inpcb/in6pcb! */ 634 struct tcphdr *th; 635 636 if (tp != NULL && (flags & TH_RST) == 0) { 637 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 638 639 if (tp->t_inpcb) 640 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 641 #ifdef INET6 642 if (tp->t_in6pcb) 643 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 644 #endif 645 } 646 647 th = NULL; /* Quell uninitialized warning */ 648 ip = NULL; 649 #ifdef INET6 650 ip6 = NULL; 651 #endif 652 if (m == NULL) { 653 if (!mtemplate) 654 return EINVAL; 655 656 /* get family information from template */ 657 switch (mtod(mtemplate, struct ip *)->ip_v) { 658 case 4: 659 family = AF_INET; 660 hlen = sizeof(struct ip); 661 break; 662 #ifdef INET6 663 case 6: 664 family = AF_INET6; 665 hlen = sizeof(struct ip6_hdr); 666 break; 667 #endif 668 default: 669 return EAFNOSUPPORT; 670 } 671 672 MGETHDR(m, M_DONTWAIT, MT_HEADER); 673 if (m) { 674 MCLAIM(m, &tcp_tx_mowner); 675 MCLGET(m, M_DONTWAIT); 676 if ((m->m_flags & M_EXT) == 0) { 677 m_free(m); 678 m = NULL; 679 } 680 } 681 if (m == NULL) 682 return ENOBUFS; 683 684 tlen = 0; 685 686 m->m_data += max_linkhdr; 687 bcopy(mtod(mtemplate, void *), mtod(m, void *), 688 mtemplate->m_len); 689 switch (family) { 690 case AF_INET: 691 ip = mtod(m, struct ip *); 692 th = (struct tcphdr *)(ip + 1); 693 break; 694 #ifdef INET6 695 case AF_INET6: 696 ip6 = mtod(m, struct ip6_hdr *); 697 th = (struct tcphdr *)(ip6 + 1); 698 break; 699 #endif 700 } 701 flags = TH_ACK; 702 } else { 703 if ((m->m_flags & M_PKTHDR) == 0) { 704 m_freem(m); 705 return EINVAL; 706 } 707 KASSERT(th0 != NULL); 708 709 /* get family information from m */ 710 switch (mtod(m, struct ip *)->ip_v) { 711 case 4: 712 family = AF_INET; 713 hlen = sizeof(struct ip); 714 ip = mtod(m, struct ip *); 715 break; 716 #ifdef INET6 717 case 6: 718 family = AF_INET6; 719 hlen = sizeof(struct ip6_hdr); 720 ip6 = mtod(m, struct ip6_hdr *); 721 break; 722 #endif 723 default: 724 m_freem(m); 725 return EAFNOSUPPORT; 726 } 727 /* clear h/w csum flags inherited from rx packet */ 728 m->m_pkthdr.csum_flags = 0; 729 730 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 731 tlen = sizeof(*th0); 732 else 733 tlen = th0->th_off << 2; 734 735 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 736 mtod(m, char *) + hlen == (char *)th0) { 737 m->m_len = hlen + tlen; 738 m_freem(m->m_next); 739 m->m_next = NULL; 740 } else { 741 struct mbuf *n; 742 743 KASSERT(max_linkhdr + hlen + tlen <= MCLBYTES); 744 745 MGETHDR(n, M_DONTWAIT, MT_HEADER); 746 if (n && max_linkhdr + hlen + tlen > MHLEN) { 747 MCLGET(n, M_DONTWAIT); 748 if ((n->m_flags & M_EXT) == 0) { 749 m_freem(n); 750 n = NULL; 751 } 752 } 753 if (!n) { 754 m_freem(m); 755 return ENOBUFS; 756 } 757 758 MCLAIM(n, &tcp_tx_mowner); 759 n->m_data += max_linkhdr; 760 n->m_len = hlen + tlen; 761 m_copyback(n, 0, hlen, mtod(m, void *)); 762 m_copyback(n, hlen, tlen, (void *)th0); 763 764 m_freem(m); 765 m = n; 766 n = NULL; 767 } 768 769 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 770 switch (family) { 771 case AF_INET: 772 ip = mtod(m, struct ip *); 773 th = (struct tcphdr *)(ip + 1); 774 ip->ip_p = IPPROTO_TCP; 775 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 776 ip->ip_p = IPPROTO_TCP; 777 break; 778 #ifdef INET6 779 case AF_INET6: 780 ip6 = mtod(m, struct ip6_hdr *); 781 th = (struct tcphdr *)(ip6 + 1); 782 ip6->ip6_nxt = IPPROTO_TCP; 783 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 784 ip6->ip6_nxt = IPPROTO_TCP; 785 break; 786 #endif 787 } 788 xchg(th->th_dport, th->th_sport, u_int16_t); 789 #undef xchg 790 tlen = 0; /*be friendly with the following code*/ 791 } 792 th->th_seq = htonl(seq); 793 th->th_ack = htonl(ack); 794 th->th_x2 = 0; 795 if ((flags & TH_SYN) == 0) { 796 if (tp) 797 win >>= tp->rcv_scale; 798 if (win > TCP_MAXWIN) 799 win = TCP_MAXWIN; 800 th->th_win = htons((u_int16_t)win); 801 th->th_off = sizeof (struct tcphdr) >> 2; 802 tlen += sizeof(*th); 803 } else { 804 tlen += th->th_off << 2; 805 } 806 m->m_len = hlen + tlen; 807 m->m_pkthdr.len = hlen + tlen; 808 m_reset_rcvif(m); 809 th->th_flags = flags; 810 th->th_urp = 0; 811 812 switch (family) { 813 case AF_INET: 814 { 815 struct ipovly *ipov = (struct ipovly *)ip; 816 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1); 817 ipov->ih_len = htons((u_int16_t)tlen); 818 819 th->th_sum = 0; 820 th->th_sum = in_cksum(m, hlen + tlen); 821 ip->ip_len = htons(hlen + tlen); 822 ip->ip_ttl = ip_defttl; 823 break; 824 } 825 #ifdef INET6 826 case AF_INET6: 827 { 828 th->th_sum = 0; 829 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 830 tlen); 831 ip6->ip6_plen = htons(tlen); 832 if (tp && tp->t_in6pcb) 833 ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb); 834 else 835 ip6->ip6_hlim = ip6_defhlim; 836 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 837 if (ip6_auto_flowlabel) { 838 ip6->ip6_flow |= 839 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 840 } 841 break; 842 } 843 #endif 844 } 845 846 if (tp != NULL && tp->t_inpcb != NULL) { 847 ro = &tp->t_inpcb->inp_route; 848 KASSERT(family == AF_INET); 849 KASSERT(in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)); 850 } 851 #ifdef INET6 852 else if (tp != NULL && tp->t_in6pcb != NULL) { 853 ro = (struct route *)&tp->t_in6pcb->in6p_route; 854 855 #ifdef DIAGNOSTIC 856 if (family == AF_INET) { 857 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 858 panic("tcp_respond: not mapped addr"); 859 if (memcmp(&ip->ip_dst, 860 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 861 sizeof(ip->ip_dst)) != 0) { 862 panic("tcp_respond: ip_dst != in6p_faddr"); 863 } 864 } else if (family == AF_INET6) { 865 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 866 &tp->t_in6pcb->in6p_faddr)) 867 panic("tcp_respond: ip6_dst != in6p_faddr"); 868 } else 869 panic("tcp_respond: address family mismatch"); 870 #endif 871 } 872 #endif 873 else 874 ro = NULL; 875 876 switch (family) { 877 case AF_INET: 878 error = ip_output(m, NULL, ro, 879 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, 880 tp ? tp->t_inpcb : NULL); 881 break; 882 #ifdef INET6 883 case AF_INET6: 884 error = ip6_output(m, NULL, ro, 0, NULL, 885 tp ? tp->t_in6pcb : NULL, NULL); 886 break; 887 #endif 888 default: 889 error = EAFNOSUPPORT; 890 break; 891 } 892 893 return error; 894 } 895 896 /* 897 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 898 * a bunch of members individually, we maintain this template for the 899 * static and mostly-static components of the TCPCB, and copy it into 900 * the new TCPCB instead. 901 */ 902 static struct tcpcb tcpcb_template = { 903 .t_srtt = TCPTV_SRTTBASE, 904 .t_rttmin = TCPTV_MIN, 905 906 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 907 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 908 .snd_numholes = 0, 909 .snd_cubic_wmax = 0, 910 .snd_cubic_wmax_last = 0, 911 .snd_cubic_ctime = 0, 912 913 .t_partialacks = -1, 914 .t_bytes_acked = 0, 915 .t_sndrexmitpack = 0, 916 .t_rcvoopack = 0, 917 .t_sndzerowin = 0, 918 }; 919 920 /* 921 * Updates the TCPCB template whenever a parameter that would affect 922 * the template is changed. 923 */ 924 void 925 tcp_tcpcb_template(void) 926 { 927 struct tcpcb *tp = &tcpcb_template; 928 int flags; 929 930 tp->t_peermss = tcp_mssdflt; 931 tp->t_ourmss = tcp_mssdflt; 932 tp->t_segsz = tcp_mssdflt; 933 934 flags = 0; 935 if (tcp_do_rfc1323 && tcp_do_win_scale) 936 flags |= TF_REQ_SCALE; 937 if (tcp_do_rfc1323 && tcp_do_timestamps) 938 flags |= TF_REQ_TSTMP; 939 tp->t_flags = flags; 940 941 /* 942 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 943 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 944 * reasonable initial retransmit time. 945 */ 946 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 947 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 948 TCPTV_MIN, TCPTV_REXMTMAX); 949 950 /* Keep Alive */ 951 tp->t_keepinit = MIN(tcp_keepinit, TCP_TIMER_MAXTICKS); 952 tp->t_keepidle = MIN(tcp_keepidle, TCP_TIMER_MAXTICKS); 953 tp->t_keepintvl = MIN(tcp_keepintvl, TCP_TIMER_MAXTICKS); 954 tp->t_keepcnt = MAX(1, MIN(tcp_keepcnt, TCP_TIMER_MAXTICKS)); 955 tp->t_maxidle = tp->t_keepcnt * MIN(tp->t_keepintvl, 956 TCP_TIMER_MAXTICKS/tp->t_keepcnt); 957 958 /* MSL */ 959 tp->t_msl = TCPTV_MSL; 960 } 961 962 /* 963 * Create a new TCP control block, making an 964 * empty reassembly queue and hooking it to the argument 965 * protocol control block. 966 */ 967 /* family selects inpcb, or in6pcb */ 968 struct tcpcb * 969 tcp_newtcpcb(int family, void *aux) 970 { 971 struct tcpcb *tp; 972 int i; 973 974 /* XXX Consider using a pool_cache for speed. */ 975 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 976 if (tp == NULL) 977 return NULL; 978 memcpy(tp, &tcpcb_template, sizeof(*tp)); 979 TAILQ_INIT(&tp->segq); 980 TAILQ_INIT(&tp->timeq); 981 tp->t_family = family; /* may be overridden later on */ 982 TAILQ_INIT(&tp->snd_holes); 983 LIST_INIT(&tp->t_sc); /* XXX can template this */ 984 985 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 986 for (i = 0; i < TCPT_NTIMERS; i++) { 987 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 988 TCP_TIMER_INIT(tp, i); 989 } 990 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 991 992 switch (family) { 993 case AF_INET: 994 { 995 struct inpcb *inp = (struct inpcb *)aux; 996 997 inp->inp_ip.ip_ttl = ip_defttl; 998 inp->inp_ppcb = (void *)tp; 999 1000 tp->t_inpcb = inp; 1001 tp->t_mtudisc = ip_mtudisc; 1002 break; 1003 } 1004 #ifdef INET6 1005 case AF_INET6: 1006 { 1007 struct in6pcb *in6p = (struct in6pcb *)aux; 1008 1009 in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p); 1010 in6p->in6p_ppcb = (void *)tp; 1011 1012 tp->t_in6pcb = in6p; 1013 /* for IPv6, always try to run path MTU discovery */ 1014 tp->t_mtudisc = 1; 1015 break; 1016 } 1017 #endif /* INET6 */ 1018 default: 1019 for (i = 0; i < TCPT_NTIMERS; i++) 1020 callout_destroy(&tp->t_timer[i]); 1021 callout_destroy(&tp->t_delack_ch); 1022 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1023 return NULL; 1024 } 1025 1026 /* 1027 * Initialize our timebase. When we send timestamps, we take 1028 * the delta from tcp_now -- this means each connection always 1029 * gets a timebase of 1, which makes it, among other things, 1030 * more difficult to determine how long a system has been up, 1031 * and thus how many TCP sequence increments have occurred. 1032 * 1033 * We start with 1, because 0 doesn't work with linux, which 1034 * considers timestamp 0 in a SYN packet as a bug and disables 1035 * timestamps. 1036 */ 1037 tp->ts_timebase = tcp_now - 1; 1038 1039 tcp_congctl_select(tp, tcp_congctl_global_name); 1040 1041 return tp; 1042 } 1043 1044 /* 1045 * Drop a TCP connection, reporting 1046 * the specified error. If connection is synchronized, 1047 * then send a RST to peer. 1048 */ 1049 struct tcpcb * 1050 tcp_drop(struct tcpcb *tp, int errno) 1051 { 1052 struct socket *so = NULL; 1053 1054 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1055 1056 if (tp->t_inpcb) 1057 so = tp->t_inpcb->inp_socket; 1058 #ifdef INET6 1059 if (tp->t_in6pcb) 1060 so = tp->t_in6pcb->in6p_socket; 1061 #endif 1062 if (!so) 1063 return NULL; 1064 1065 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1066 tp->t_state = TCPS_CLOSED; 1067 (void) tcp_output(tp); 1068 TCP_STATINC(TCP_STAT_DROPS); 1069 } else 1070 TCP_STATINC(TCP_STAT_CONNDROPS); 1071 if (errno == ETIMEDOUT && tp->t_softerror) 1072 errno = tp->t_softerror; 1073 so->so_error = errno; 1074 return (tcp_close(tp)); 1075 } 1076 1077 /* 1078 * Close a TCP control block: 1079 * discard all space held by the tcp 1080 * discard internet protocol block 1081 * wake up any sleepers 1082 */ 1083 struct tcpcb * 1084 tcp_close(struct tcpcb *tp) 1085 { 1086 struct inpcb *inp; 1087 #ifdef INET6 1088 struct in6pcb *in6p; 1089 #endif 1090 struct socket *so; 1091 #ifdef RTV_RTT 1092 struct rtentry *rt = NULL; 1093 #endif 1094 struct route *ro; 1095 int j; 1096 1097 inp = tp->t_inpcb; 1098 #ifdef INET6 1099 in6p = tp->t_in6pcb; 1100 #endif 1101 so = NULL; 1102 ro = NULL; 1103 if (inp) { 1104 so = inp->inp_socket; 1105 ro = &inp->inp_route; 1106 } 1107 #ifdef INET6 1108 else if (in6p) { 1109 so = in6p->in6p_socket; 1110 ro = (struct route *)&in6p->in6p_route; 1111 } 1112 #endif 1113 1114 #ifdef RTV_RTT 1115 /* 1116 * If we sent enough data to get some meaningful characteristics, 1117 * save them in the routing entry. 'Enough' is arbitrarily 1118 * defined as the sendpipesize (default 4K) * 16. This would 1119 * give us 16 rtt samples assuming we only get one sample per 1120 * window (the usual case on a long haul net). 16 samples is 1121 * enough for the srtt filter to converge to within 5% of the correct 1122 * value; fewer samples and we could save a very bogus rtt. 1123 * 1124 * Don't update the default route's characteristics and don't 1125 * update anything that the user "locked". 1126 */ 1127 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1128 ro && (rt = rtcache_validate(ro)) != NULL && 1129 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1130 u_long i = 0; 1131 1132 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1133 i = tp->t_srtt * 1134 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1135 if (rt->rt_rmx.rmx_rtt && i) 1136 /* 1137 * filter this update to half the old & half 1138 * the new values, converting scale. 1139 * See route.h and tcp_var.h for a 1140 * description of the scaling constants. 1141 */ 1142 rt->rt_rmx.rmx_rtt = 1143 (rt->rt_rmx.rmx_rtt + i) / 2; 1144 else 1145 rt->rt_rmx.rmx_rtt = i; 1146 } 1147 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1148 i = tp->t_rttvar * 1149 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1150 if (rt->rt_rmx.rmx_rttvar && i) 1151 rt->rt_rmx.rmx_rttvar = 1152 (rt->rt_rmx.rmx_rttvar + i) / 2; 1153 else 1154 rt->rt_rmx.rmx_rttvar = i; 1155 } 1156 /* 1157 * update the pipelimit (ssthresh) if it has been updated 1158 * already or if a pipesize was specified & the threshold 1159 * got below half the pipesize. I.e., wait for bad news 1160 * before we start updating, then update on both good 1161 * and bad news. 1162 */ 1163 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1164 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1165 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1166 /* 1167 * convert the limit from user data bytes to 1168 * packets then to packet data bytes. 1169 */ 1170 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1171 if (i < 2) 1172 i = 2; 1173 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1174 if (rt->rt_rmx.rmx_ssthresh) 1175 rt->rt_rmx.rmx_ssthresh = 1176 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1177 else 1178 rt->rt_rmx.rmx_ssthresh = i; 1179 } 1180 } 1181 rtcache_unref(rt, ro); 1182 #endif /* RTV_RTT */ 1183 /* free the reassembly queue, if any */ 1184 TCP_REASS_LOCK(tp); 1185 (void) tcp_freeq(tp); 1186 TCP_REASS_UNLOCK(tp); 1187 1188 /* free the SACK holes list. */ 1189 tcp_free_sackholes(tp); 1190 tcp_congctl_release(tp); 1191 syn_cache_cleanup(tp); 1192 1193 if (tp->t_template) { 1194 m_free(tp->t_template); 1195 tp->t_template = NULL; 1196 } 1197 1198 /* 1199 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1200 * the timers can also drop the lock. We need to prevent access 1201 * to the tcpcb as it's half torn down. Flag the pcb as dead 1202 * (prevents access by timers) and only then detach it. 1203 */ 1204 tp->t_flags |= TF_DEAD; 1205 if (inp) { 1206 inp->inp_ppcb = 0; 1207 soisdisconnected(so); 1208 in_pcbdetach(inp); 1209 } 1210 #ifdef INET6 1211 else if (in6p) { 1212 in6p->in6p_ppcb = 0; 1213 soisdisconnected(so); 1214 in6_pcbdetach(in6p); 1215 } 1216 #endif 1217 /* 1218 * pcb is no longer visble elsewhere, so we can safely release 1219 * the lock in callout_halt() if needed. 1220 */ 1221 TCP_STATINC(TCP_STAT_CLOSED); 1222 for (j = 0; j < TCPT_NTIMERS; j++) { 1223 callout_halt(&tp->t_timer[j], softnet_lock); 1224 callout_destroy(&tp->t_timer[j]); 1225 } 1226 callout_halt(&tp->t_delack_ch, softnet_lock); 1227 callout_destroy(&tp->t_delack_ch); 1228 pool_put(&tcpcb_pool, tp); 1229 1230 return NULL; 1231 } 1232 1233 int 1234 tcp_freeq(struct tcpcb *tp) 1235 { 1236 struct ipqent *qe; 1237 int rv = 0; 1238 1239 TCP_REASS_LOCK_CHECK(tp); 1240 1241 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1242 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1243 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1244 m_freem(qe->ipqe_m); 1245 tcpipqent_free(qe); 1246 rv = 1; 1247 } 1248 tp->t_segqlen = 0; 1249 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1250 return (rv); 1251 } 1252 1253 void 1254 tcp_fasttimo(void) 1255 { 1256 if (tcp_drainwanted) { 1257 tcp_drain(); 1258 tcp_drainwanted = 0; 1259 } 1260 } 1261 1262 void 1263 tcp_drainstub(void) 1264 { 1265 tcp_drainwanted = 1; 1266 } 1267 1268 /* 1269 * Protocol drain routine. Called when memory is in short supply. 1270 * Called from pr_fasttimo thus a callout context. 1271 */ 1272 void 1273 tcp_drain(void) 1274 { 1275 struct inpcb_hdr *inph; 1276 struct tcpcb *tp; 1277 1278 mutex_enter(softnet_lock); 1279 KERNEL_LOCK(1, NULL); 1280 1281 /* 1282 * Free the sequence queue of all TCP connections. 1283 */ 1284 TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1285 switch (inph->inph_af) { 1286 case AF_INET: 1287 tp = intotcpcb((struct inpcb *)inph); 1288 break; 1289 #ifdef INET6 1290 case AF_INET6: 1291 tp = in6totcpcb((struct in6pcb *)inph); 1292 break; 1293 #endif 1294 default: 1295 tp = NULL; 1296 break; 1297 } 1298 if (tp != NULL) { 1299 /* 1300 * We may be called from a device's interrupt 1301 * context. If the tcpcb is already busy, 1302 * just bail out now. 1303 */ 1304 if (tcp_reass_lock_try(tp) == 0) 1305 continue; 1306 if (tcp_freeq(tp)) 1307 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1308 TCP_REASS_UNLOCK(tp); 1309 } 1310 } 1311 1312 KERNEL_UNLOCK_ONE(NULL); 1313 mutex_exit(softnet_lock); 1314 } 1315 1316 /* 1317 * Notify a tcp user of an asynchronous error; 1318 * store error as soft error, but wake up user 1319 * (for now, won't do anything until can select for soft error). 1320 */ 1321 void 1322 tcp_notify(struct inpcb *inp, int error) 1323 { 1324 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1325 struct socket *so = inp->inp_socket; 1326 1327 /* 1328 * Ignore some errors if we are hooked up. 1329 * If connection hasn't completed, has retransmitted several times, 1330 * and receives a second error, give up now. This is better 1331 * than waiting a long time to establish a connection that 1332 * can never complete. 1333 */ 1334 if (tp->t_state == TCPS_ESTABLISHED && 1335 (error == EHOSTUNREACH || error == ENETUNREACH || 1336 error == EHOSTDOWN)) { 1337 return; 1338 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1339 tp->t_rxtshift > 3 && tp->t_softerror) 1340 so->so_error = error; 1341 else 1342 tp->t_softerror = error; 1343 cv_broadcast(&so->so_cv); 1344 sorwakeup(so); 1345 sowwakeup(so); 1346 } 1347 1348 #ifdef INET6 1349 void 1350 tcp6_notify(struct in6pcb *in6p, int error) 1351 { 1352 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1353 struct socket *so = in6p->in6p_socket; 1354 1355 /* 1356 * Ignore some errors if we are hooked up. 1357 * If connection hasn't completed, has retransmitted several times, 1358 * and receives a second error, give up now. This is better 1359 * than waiting a long time to establish a connection that 1360 * can never complete. 1361 */ 1362 if (tp->t_state == TCPS_ESTABLISHED && 1363 (error == EHOSTUNREACH || error == ENETUNREACH || 1364 error == EHOSTDOWN)) { 1365 return; 1366 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1367 tp->t_rxtshift > 3 && tp->t_softerror) 1368 so->so_error = error; 1369 else 1370 tp->t_softerror = error; 1371 cv_broadcast(&so->so_cv); 1372 sorwakeup(so); 1373 sowwakeup(so); 1374 } 1375 #endif 1376 1377 #ifdef INET6 1378 void * 1379 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1380 { 1381 struct tcphdr th; 1382 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1383 int nmatch; 1384 struct ip6_hdr *ip6; 1385 const struct sockaddr_in6 *sa6_src = NULL; 1386 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1387 struct mbuf *m; 1388 int off; 1389 1390 if (sa->sa_family != AF_INET6 || 1391 sa->sa_len != sizeof(struct sockaddr_in6)) 1392 return NULL; 1393 if ((unsigned)cmd >= PRC_NCMDS) 1394 return NULL; 1395 else if (cmd == PRC_QUENCH) { 1396 /* 1397 * Don't honor ICMP Source Quench messages meant for 1398 * TCP connections. 1399 */ 1400 return NULL; 1401 } else if (PRC_IS_REDIRECT(cmd)) 1402 notify = in6_rtchange, d = NULL; 1403 else if (cmd == PRC_MSGSIZE) 1404 ; /* special code is present, see below */ 1405 else if (cmd == PRC_HOSTDEAD) 1406 d = NULL; 1407 else if (inet6ctlerrmap[cmd] == 0) 1408 return NULL; 1409 1410 /* if the parameter is from icmp6, decode it. */ 1411 if (d != NULL) { 1412 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1413 m = ip6cp->ip6c_m; 1414 ip6 = ip6cp->ip6c_ip6; 1415 off = ip6cp->ip6c_off; 1416 sa6_src = ip6cp->ip6c_src; 1417 } else { 1418 m = NULL; 1419 ip6 = NULL; 1420 sa6_src = &sa6_any; 1421 off = 0; 1422 } 1423 1424 if (ip6) { 1425 /* check if we can safely examine src and dst ports */ 1426 if (m->m_pkthdr.len < off + sizeof(th)) { 1427 if (cmd == PRC_MSGSIZE) 1428 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1429 return NULL; 1430 } 1431 1432 memset(&th, 0, sizeof(th)); 1433 m_copydata(m, off, sizeof(th), (void *)&th); 1434 1435 if (cmd == PRC_MSGSIZE) { 1436 int valid = 0; 1437 1438 /* 1439 * Check to see if we have a valid TCP connection 1440 * corresponding to the address in the ICMPv6 message 1441 * payload. 1442 */ 1443 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1444 th.th_dport, 1445 (const struct in6_addr *)&sa6_src->sin6_addr, 1446 th.th_sport, 0, 0)) 1447 valid++; 1448 1449 /* 1450 * Depending on the value of "valid" and routing table 1451 * size (mtudisc_{hi,lo}wat), we will: 1452 * - recalcurate the new MTU and create the 1453 * corresponding routing entry, or 1454 * - ignore the MTU change notification. 1455 */ 1456 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1457 1458 /* 1459 * no need to call in6_pcbnotify, it should have been 1460 * called via callback if necessary 1461 */ 1462 return NULL; 1463 } 1464 1465 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1466 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1467 if (nmatch == 0 && syn_cache_count && 1468 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1469 inet6ctlerrmap[cmd] == ENETUNREACH || 1470 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1471 syn_cache_unreach((const struct sockaddr *)sa6_src, 1472 sa, &th); 1473 } else { 1474 (void) in6_pcbnotify(&tcbtable, sa, 0, 1475 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1476 } 1477 1478 return NULL; 1479 } 1480 #endif 1481 1482 /* assumes that ip header and tcp header are contiguous on mbuf */ 1483 void * 1484 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1485 { 1486 struct ip *ip = v; 1487 struct tcphdr *th; 1488 struct icmp *icp; 1489 extern const int inetctlerrmap[]; 1490 void (*notify)(struct inpcb *, int) = tcp_notify; 1491 int errno; 1492 int nmatch; 1493 struct tcpcb *tp; 1494 u_int mtu; 1495 tcp_seq seq; 1496 struct inpcb *inp; 1497 #ifdef INET6 1498 struct in6pcb *in6p; 1499 struct in6_addr src6, dst6; 1500 #endif 1501 1502 if (sa->sa_family != AF_INET || 1503 sa->sa_len != sizeof(struct sockaddr_in)) 1504 return NULL; 1505 if ((unsigned)cmd >= PRC_NCMDS) 1506 return NULL; 1507 errno = inetctlerrmap[cmd]; 1508 if (cmd == PRC_QUENCH) 1509 /* 1510 * Don't honor ICMP Source Quench messages meant for 1511 * TCP connections. 1512 */ 1513 return NULL; 1514 else if (PRC_IS_REDIRECT(cmd)) 1515 notify = in_rtchange, ip = 0; 1516 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1517 /* 1518 * Check to see if we have a valid TCP connection 1519 * corresponding to the address in the ICMP message 1520 * payload. 1521 * 1522 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1523 */ 1524 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1525 #ifdef INET6 1526 in6_in_2_v4mapin6(&ip->ip_src, &src6); 1527 in6_in_2_v4mapin6(&ip->ip_dst, &dst6); 1528 #endif 1529 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1530 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL) 1531 #ifdef INET6 1532 in6p = NULL; 1533 #else 1534 ; 1535 #endif 1536 #ifdef INET6 1537 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1538 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL) 1539 ; 1540 #endif 1541 else 1542 return NULL; 1543 1544 /* 1545 * Now that we've validated that we are actually communicating 1546 * with the host indicated in the ICMP message, locate the 1547 * ICMP header, recalculate the new MTU, and create the 1548 * corresponding routing entry. 1549 */ 1550 icp = (struct icmp *)((char *)ip - 1551 offsetof(struct icmp, icmp_ip)); 1552 if (inp) { 1553 if ((tp = intotcpcb(inp)) == NULL) 1554 return NULL; 1555 } 1556 #ifdef INET6 1557 else if (in6p) { 1558 if ((tp = in6totcpcb(in6p)) == NULL) 1559 return NULL; 1560 } 1561 #endif 1562 else 1563 return NULL; 1564 seq = ntohl(th->th_seq); 1565 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1566 return NULL; 1567 /* 1568 * If the ICMP message advertises a Next-Hop MTU 1569 * equal or larger than the maximum packet size we have 1570 * ever sent, drop the message. 1571 */ 1572 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1573 if (mtu >= tp->t_pmtud_mtu_sent) 1574 return NULL; 1575 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1576 /* 1577 * Calculate new MTU, and create corresponding 1578 * route (traditional PMTUD). 1579 */ 1580 tp->t_flags &= ~TF_PMTUD_PEND; 1581 icmp_mtudisc(icp, ip->ip_dst); 1582 } else { 1583 /* 1584 * Record the information got in the ICMP 1585 * message; act on it later. 1586 * If we had already recorded an ICMP message, 1587 * replace the old one only if the new message 1588 * refers to an older TCP segment 1589 */ 1590 if (tp->t_flags & TF_PMTUD_PEND) { 1591 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1592 return NULL; 1593 } else 1594 tp->t_flags |= TF_PMTUD_PEND; 1595 tp->t_pmtud_th_seq = seq; 1596 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1597 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1598 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1599 } 1600 return NULL; 1601 } else if (cmd == PRC_HOSTDEAD) 1602 ip = 0; 1603 else if (errno == 0) 1604 return NULL; 1605 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1606 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1607 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1608 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1609 if (nmatch == 0 && syn_cache_count && 1610 (inetctlerrmap[cmd] == EHOSTUNREACH || 1611 inetctlerrmap[cmd] == ENETUNREACH || 1612 inetctlerrmap[cmd] == EHOSTDOWN)) { 1613 struct sockaddr_in sin; 1614 memset(&sin, 0, sizeof(sin)); 1615 sin.sin_len = sizeof(sin); 1616 sin.sin_family = AF_INET; 1617 sin.sin_port = th->th_sport; 1618 sin.sin_addr = ip->ip_src; 1619 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1620 } 1621 1622 /* XXX mapped address case */ 1623 } else 1624 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1625 notify); 1626 return NULL; 1627 } 1628 1629 /* 1630 * When a source quench is received, we are being notified of congestion. 1631 * Close the congestion window down to the Loss Window (one segment). 1632 * We will gradually open it again as we proceed. 1633 */ 1634 void 1635 tcp_quench(struct inpcb *inp) 1636 { 1637 struct tcpcb *tp = intotcpcb(inp); 1638 1639 if (tp) { 1640 tp->snd_cwnd = tp->t_segsz; 1641 tp->t_bytes_acked = 0; 1642 } 1643 } 1644 1645 #ifdef INET6 1646 void 1647 tcp6_quench(struct in6pcb *in6p) 1648 { 1649 struct tcpcb *tp = in6totcpcb(in6p); 1650 1651 if (tp) { 1652 tp->snd_cwnd = tp->t_segsz; 1653 tp->t_bytes_acked = 0; 1654 } 1655 } 1656 #endif 1657 1658 /* 1659 * Path MTU Discovery handlers. 1660 */ 1661 void 1662 tcp_mtudisc_callback(struct in_addr faddr) 1663 { 1664 #ifdef INET6 1665 struct in6_addr in6; 1666 #endif 1667 1668 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1669 #ifdef INET6 1670 in6_in_2_v4mapin6(&faddr, &in6); 1671 tcp6_mtudisc_callback(&in6); 1672 #endif 1673 } 1674 1675 /* 1676 * On receipt of path MTU corrections, flush old route and replace it 1677 * with the new one. Retransmit all unacknowledged packets, to ensure 1678 * that all packets will be received. 1679 */ 1680 void 1681 tcp_mtudisc(struct inpcb *inp, int errno) 1682 { 1683 struct tcpcb *tp = intotcpcb(inp); 1684 struct rtentry *rt; 1685 1686 if (tp == NULL) 1687 return; 1688 1689 rt = in_pcbrtentry(inp); 1690 if (rt != NULL) { 1691 /* 1692 * If this was not a host route, remove and realloc. 1693 */ 1694 if ((rt->rt_flags & RTF_HOST) == 0) { 1695 in_pcbrtentry_unref(rt, inp); 1696 in_rtchange(inp, errno); 1697 if ((rt = in_pcbrtentry(inp)) == NULL) 1698 return; 1699 } 1700 1701 /* 1702 * Slow start out of the error condition. We 1703 * use the MTU because we know it's smaller 1704 * than the previously transmitted segment. 1705 * 1706 * Note: This is more conservative than the 1707 * suggestion in draft-floyd-incr-init-win-03. 1708 */ 1709 if (rt->rt_rmx.rmx_mtu != 0) 1710 tp->snd_cwnd = 1711 TCP_INITIAL_WINDOW(tcp_init_win, 1712 rt->rt_rmx.rmx_mtu); 1713 in_pcbrtentry_unref(rt, inp); 1714 } 1715 1716 /* 1717 * Resend unacknowledged packets. 1718 */ 1719 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1720 tcp_output(tp); 1721 } 1722 1723 #ifdef INET6 1724 /* 1725 * Path MTU Discovery handlers. 1726 */ 1727 void 1728 tcp6_mtudisc_callback(struct in6_addr *faddr) 1729 { 1730 struct sockaddr_in6 sin6; 1731 1732 memset(&sin6, 0, sizeof(sin6)); 1733 sin6.sin6_family = AF_INET6; 1734 sin6.sin6_len = sizeof(struct sockaddr_in6); 1735 sin6.sin6_addr = *faddr; 1736 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1737 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1738 } 1739 1740 void 1741 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1742 { 1743 struct tcpcb *tp = in6totcpcb(in6p); 1744 struct rtentry *rt; 1745 1746 if (tp == NULL) 1747 return; 1748 1749 rt = in6_pcbrtentry(in6p); 1750 if (rt != NULL) { 1751 /* 1752 * If this was not a host route, remove and realloc. 1753 */ 1754 if ((rt->rt_flags & RTF_HOST) == 0) { 1755 in6_pcbrtentry_unref(rt, in6p); 1756 in6_rtchange(in6p, errno); 1757 rt = in6_pcbrtentry(in6p); 1758 if (rt == NULL) 1759 return; 1760 } 1761 1762 /* 1763 * Slow start out of the error condition. We 1764 * use the MTU because we know it's smaller 1765 * than the previously transmitted segment. 1766 * 1767 * Note: This is more conservative than the 1768 * suggestion in draft-floyd-incr-init-win-03. 1769 */ 1770 if (rt->rt_rmx.rmx_mtu != 0) { 1771 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1772 rt->rt_rmx.rmx_mtu); 1773 } 1774 in6_pcbrtentry_unref(rt, in6p); 1775 } 1776 1777 /* 1778 * Resend unacknowledged packets. 1779 */ 1780 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1781 tcp_output(tp); 1782 } 1783 #endif /* INET6 */ 1784 1785 /* 1786 * Compute the MSS to advertise to the peer. Called only during 1787 * the 3-way handshake. If we are the server (peer initiated 1788 * connection), we are called with a pointer to the interface 1789 * on which the SYN packet arrived. If we are the client (we 1790 * initiated connection), we are called with a pointer to the 1791 * interface out which this connection should go. 1792 * 1793 * NOTE: Do not subtract IP option/extension header size nor IPsec 1794 * header size from MSS advertisement. MSS option must hold the maximum 1795 * segment size we can accept, so it must always be: 1796 * max(if mtu) - ip header - tcp header 1797 */ 1798 u_long 1799 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1800 { 1801 extern u_long in_maxmtu; 1802 u_long mss = 0; 1803 u_long hdrsiz; 1804 1805 /* 1806 * In order to avoid defeating path MTU discovery on the peer, 1807 * we advertise the max MTU of all attached networks as our MSS, 1808 * per RFC 1191, section 3.1. 1809 * 1810 * We provide the option to advertise just the MTU of 1811 * the interface on which we hope this connection will 1812 * be receiving. If we are responding to a SYN, we 1813 * will have a pretty good idea about this, but when 1814 * initiating a connection there is a bit more doubt. 1815 * 1816 * We also need to ensure that loopback has a large enough 1817 * MSS, as the loopback MTU is never included in in_maxmtu. 1818 */ 1819 1820 if (ifp != NULL) 1821 switch (af) { 1822 #ifdef INET6 1823 case AF_INET6: /* FALLTHROUGH */ 1824 #endif 1825 case AF_INET: 1826 mss = ifp->if_mtu; 1827 break; 1828 } 1829 1830 if (tcp_mss_ifmtu == 0) 1831 switch (af) { 1832 #ifdef INET6 1833 case AF_INET6: /* FALLTHROUGH */ 1834 #endif 1835 case AF_INET: 1836 mss = uimax(in_maxmtu, mss); 1837 break; 1838 } 1839 1840 switch (af) { 1841 case AF_INET: 1842 hdrsiz = sizeof(struct ip); 1843 break; 1844 #ifdef INET6 1845 case AF_INET6: 1846 hdrsiz = sizeof(struct ip6_hdr); 1847 break; 1848 #endif 1849 default: 1850 hdrsiz = 0; 1851 break; 1852 } 1853 hdrsiz += sizeof(struct tcphdr); 1854 if (mss > hdrsiz) 1855 mss -= hdrsiz; 1856 1857 mss = uimax(tcp_mssdflt, mss); 1858 return (mss); 1859 } 1860 1861 /* 1862 * Set connection variables based on the peer's advertised MSS. 1863 * We are passed the TCPCB for the actual connection. If we 1864 * are the server, we are called by the compressed state engine 1865 * when the 3-way handshake is complete. If we are the client, 1866 * we are called when we receive the SYN,ACK from the server. 1867 * 1868 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1869 * before this routine is called! 1870 */ 1871 void 1872 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1873 { 1874 struct socket *so; 1875 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1876 struct rtentry *rt; 1877 #endif 1878 u_long bufsize; 1879 int mss; 1880 1881 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1882 1883 so = NULL; 1884 rt = NULL; 1885 1886 if (tp->t_inpcb) { 1887 so = tp->t_inpcb->inp_socket; 1888 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1889 rt = in_pcbrtentry(tp->t_inpcb); 1890 #endif 1891 } 1892 1893 #ifdef INET6 1894 if (tp->t_in6pcb) { 1895 so = tp->t_in6pcb->in6p_socket; 1896 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1897 rt = in6_pcbrtentry(tp->t_in6pcb); 1898 #endif 1899 } 1900 #endif 1901 1902 /* 1903 * As per RFC1122, use the default MSS value, unless they 1904 * sent us an offer. Do not accept offers less than 256 bytes. 1905 */ 1906 mss = tcp_mssdflt; 1907 if (offer) 1908 mss = offer; 1909 mss = uimax(mss, 256); /* sanity */ 1910 tp->t_peermss = mss; 1911 mss -= tcp_optlen(tp); 1912 if (tp->t_inpcb) 1913 mss -= ip_optlen(tp->t_inpcb); 1914 #ifdef INET6 1915 if (tp->t_in6pcb) 1916 mss -= ip6_optlen(tp->t_in6pcb); 1917 #endif 1918 /* 1919 * XXX XXX What if mss goes negative or zero? This can happen if a 1920 * socket has large IPv6 options. We crash below. 1921 */ 1922 1923 /* 1924 * If there's a pipesize, change the socket buffer to that size. 1925 * Make the socket buffer an integral number of MSS units. If 1926 * the MSS is larger than the socket buffer, artificially decrease 1927 * the MSS. 1928 */ 1929 #ifdef RTV_SPIPE 1930 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1931 bufsize = rt->rt_rmx.rmx_sendpipe; 1932 else 1933 #endif 1934 { 1935 KASSERT(so != NULL); 1936 bufsize = so->so_snd.sb_hiwat; 1937 } 1938 if (bufsize < mss) 1939 mss = bufsize; 1940 else { 1941 bufsize = roundup(bufsize, mss); 1942 if (bufsize > sb_max) 1943 bufsize = sb_max; 1944 (void) sbreserve(&so->so_snd, bufsize, so); 1945 } 1946 tp->t_segsz = mss; 1947 1948 #ifdef RTV_SSTHRESH 1949 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1950 /* 1951 * There's some sort of gateway or interface buffer 1952 * limit on the path. Use this to set the slow 1953 * start threshold, but set the threshold to no less 1954 * than 2 * MSS. 1955 */ 1956 tp->snd_ssthresh = uimax(2 * mss, rt->rt_rmx.rmx_ssthresh); 1957 } 1958 #endif 1959 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1960 if (tp->t_inpcb) 1961 in_pcbrtentry_unref(rt, tp->t_inpcb); 1962 #ifdef INET6 1963 if (tp->t_in6pcb) 1964 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 1965 #endif 1966 #endif 1967 } 1968 1969 /* 1970 * Processing necessary when a TCP connection is established. 1971 */ 1972 void 1973 tcp_established(struct tcpcb *tp) 1974 { 1975 struct socket *so; 1976 #ifdef RTV_RPIPE 1977 struct rtentry *rt; 1978 #endif 1979 u_long bufsize; 1980 1981 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1982 1983 so = NULL; 1984 rt = NULL; 1985 1986 /* This is a while() to reduce the dreadful stairstepping below */ 1987 while (tp->t_inpcb) { 1988 so = tp->t_inpcb->inp_socket; 1989 #if defined(RTV_RPIPE) 1990 rt = in_pcbrtentry(tp->t_inpcb); 1991 #endif 1992 if (__predict_true(tcp_msl_enable)) { 1993 if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) { 1994 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 1995 break; 1996 } 1997 1998 if (__predict_false(tcp_rttlocal)) { 1999 /* This may be adjusted by tcp_input */ 2000 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2001 break; 2002 } 2003 if (in_localaddr(tp->t_inpcb->inp_faddr)) { 2004 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2005 break; 2006 } 2007 } 2008 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2009 break; 2010 } 2011 2012 /* Clamp to a reasonable range. */ 2013 tp->t_msl = MIN(tp->t_msl, TCP_MAXMSL); 2014 2015 #ifdef INET6 2016 /* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */ 2017 while (!tp->t_inpcb && tp->t_in6pcb) { 2018 so = tp->t_in6pcb->in6p_socket; 2019 #if defined(RTV_RPIPE) 2020 rt = in6_pcbrtentry(tp->t_in6pcb); 2021 #endif 2022 if (__predict_true(tcp_msl_enable)) { 2023 extern const struct in6_addr in6addr_loopback; 2024 2025 if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr, 2026 &in6addr_loopback)) { 2027 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2028 break; 2029 } 2030 2031 if (__predict_false(tcp_rttlocal)) { 2032 /* This may be adjusted by tcp_input */ 2033 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2034 break; 2035 } 2036 if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) { 2037 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2038 break; 2039 } 2040 } 2041 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2042 break; 2043 } 2044 2045 /* Clamp to a reasonable range. */ 2046 tp->t_msl = MIN(tp->t_msl, TCP_MAXMSL); 2047 #endif 2048 2049 tp->t_state = TCPS_ESTABLISHED; 2050 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2051 2052 #ifdef RTV_RPIPE 2053 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2054 bufsize = rt->rt_rmx.rmx_recvpipe; 2055 else 2056 #endif 2057 { 2058 KASSERT(so != NULL); 2059 bufsize = so->so_rcv.sb_hiwat; 2060 } 2061 if (bufsize > tp->t_ourmss) { 2062 bufsize = roundup(bufsize, tp->t_ourmss); 2063 if (bufsize > sb_max) 2064 bufsize = sb_max; 2065 (void) sbreserve(&so->so_rcv, bufsize, so); 2066 } 2067 #ifdef RTV_RPIPE 2068 if (tp->t_inpcb) 2069 in_pcbrtentry_unref(rt, tp->t_inpcb); 2070 #ifdef INET6 2071 if (tp->t_in6pcb) 2072 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2073 #endif 2074 #endif 2075 } 2076 2077 /* 2078 * Check if there's an initial rtt or rttvar. Convert from the 2079 * route-table units to scaled multiples of the slow timeout timer. 2080 * Called only during the 3-way handshake. 2081 */ 2082 void 2083 tcp_rmx_rtt(struct tcpcb *tp) 2084 { 2085 #ifdef RTV_RTT 2086 struct rtentry *rt = NULL; 2087 int rtt; 2088 2089 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 2090 2091 if (tp->t_inpcb) 2092 rt = in_pcbrtentry(tp->t_inpcb); 2093 #ifdef INET6 2094 if (tp->t_in6pcb) 2095 rt = in6_pcbrtentry(tp->t_in6pcb); 2096 #endif 2097 if (rt == NULL) 2098 return; 2099 2100 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2101 /* 2102 * XXX The lock bit for MTU indicates that the value 2103 * is also a minimum value; this is subject to time. 2104 */ 2105 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2106 TCPT_RANGESET(tp->t_rttmin, 2107 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2108 TCPTV_MIN, TCPTV_REXMTMAX); 2109 tp->t_srtt = rtt / 2110 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2111 if (rt->rt_rmx.rmx_rttvar) { 2112 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2113 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2114 (TCP_RTTVAR_SHIFT + 2)); 2115 } else { 2116 /* Default variation is +- 1 rtt */ 2117 tp->t_rttvar = 2118 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2119 } 2120 TCPT_RANGESET(tp->t_rxtcur, 2121 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2122 tp->t_rttmin, TCPTV_REXMTMAX); 2123 } 2124 if (tp->t_inpcb) 2125 in_pcbrtentry_unref(rt, tp->t_inpcb); 2126 #ifdef INET6 2127 if (tp->t_in6pcb) 2128 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2129 #endif 2130 #endif 2131 } 2132 2133 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2134 2135 /* 2136 * Get a new sequence value given a tcp control block 2137 */ 2138 tcp_seq 2139 tcp_new_iss(struct tcpcb *tp) 2140 { 2141 2142 if (tp->t_inpcb != NULL) { 2143 return tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2144 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2145 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr)); 2146 } 2147 #ifdef INET6 2148 if (tp->t_in6pcb != NULL) { 2149 return tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2150 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2151 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr)); 2152 } 2153 #endif 2154 2155 panic("tcp_new_iss: unreachable"); 2156 } 2157 2158 static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2159 2160 /* 2161 * Initialize RFC 1948 ISS Secret 2162 */ 2163 static int 2164 tcp_iss_secret_init(void) 2165 { 2166 cprng_strong(kern_cprng, 2167 tcp_iss_secret, sizeof(tcp_iss_secret), 0); 2168 2169 return 0; 2170 } 2171 2172 /* 2173 * This routine actually generates a new TCP initial sequence number. 2174 */ 2175 tcp_seq 2176 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2177 size_t addrsz) 2178 { 2179 tcp_seq tcp_iss; 2180 2181 if (tcp_do_rfc1948) { 2182 MD5_CTX ctx; 2183 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2184 static ONCE_DECL(tcp_iss_secret_control); 2185 2186 /* 2187 * If we haven't been here before, initialize our cryptographic 2188 * hash secret. 2189 */ 2190 RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init); 2191 2192 /* 2193 * Compute the base value of the ISS. It is a hash 2194 * of (saddr, sport, daddr, dport, secret). 2195 */ 2196 MD5Init(&ctx); 2197 2198 MD5Update(&ctx, (u_char *) laddr, addrsz); 2199 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2200 2201 MD5Update(&ctx, (u_char *) faddr, addrsz); 2202 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2203 2204 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2205 2206 MD5Final(hash, &ctx); 2207 2208 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2209 2210 #ifdef TCPISS_DEBUG 2211 printf("ISS hash 0x%08x, ", tcp_iss); 2212 #endif 2213 } else { 2214 /* 2215 * Randomize. 2216 */ 2217 tcp_iss = cprng_fast32() & TCP_ISS_RANDOM_MASK; 2218 #ifdef TCPISS_DEBUG 2219 printf("ISS random 0x%08x, ", tcp_iss); 2220 #endif 2221 } 2222 2223 /* 2224 * Add the offset in to the computed value. 2225 */ 2226 tcp_iss += tcp_iss_seq; 2227 #ifdef TCPISS_DEBUG 2228 printf("ISS %08x\n", tcp_iss); 2229 #endif 2230 return tcp_iss; 2231 } 2232 2233 #if defined(IPSEC) 2234 /* compute ESP/AH header size for TCP, including outer IP header. */ 2235 size_t 2236 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2237 { 2238 struct inpcb *inp; 2239 size_t hdrsiz; 2240 2241 /* XXX mapped addr case (tp->t_in6pcb) */ 2242 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2243 return 0; 2244 switch (tp->t_family) { 2245 case AF_INET: 2246 /* XXX: should use correct direction. */ 2247 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2248 break; 2249 default: 2250 hdrsiz = 0; 2251 break; 2252 } 2253 2254 return hdrsiz; 2255 } 2256 2257 #ifdef INET6 2258 size_t 2259 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2260 { 2261 struct in6pcb *in6p; 2262 size_t hdrsiz; 2263 2264 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2265 return 0; 2266 switch (tp->t_family) { 2267 case AF_INET6: 2268 /* XXX: should use correct direction. */ 2269 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2270 break; 2271 case AF_INET: 2272 /* mapped address case - tricky */ 2273 default: 2274 hdrsiz = 0; 2275 break; 2276 } 2277 2278 return hdrsiz; 2279 } 2280 #endif 2281 #endif /*IPSEC*/ 2282 2283 /* 2284 * Determine the length of the TCP options for this connection. 2285 * 2286 * XXX: What do we do for SACK, when we add that? Just reserve 2287 * all of the space? Otherwise we can't exactly be incrementing 2288 * cwnd by an amount that varies depending on the amount we last 2289 * had to SACK! 2290 */ 2291 2292 u_int 2293 tcp_optlen(struct tcpcb *tp) 2294 { 2295 u_int optlen; 2296 2297 optlen = 0; 2298 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2299 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2300 optlen += TCPOLEN_TSTAMP_APPA; 2301 2302 #ifdef TCP_SIGNATURE 2303 if (tp->t_flags & TF_SIGNATURE) 2304 optlen += TCPOLEN_SIGLEN; 2305 #endif 2306 2307 return optlen; 2308 } 2309 2310 u_int 2311 tcp_hdrsz(struct tcpcb *tp) 2312 { 2313 u_int hlen; 2314 2315 switch (tp->t_family) { 2316 #ifdef INET6 2317 case AF_INET6: 2318 hlen = sizeof(struct ip6_hdr); 2319 break; 2320 #endif 2321 case AF_INET: 2322 hlen = sizeof(struct ip); 2323 break; 2324 default: 2325 hlen = 0; 2326 break; 2327 } 2328 hlen += sizeof(struct tcphdr); 2329 2330 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2331 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2332 hlen += TCPOLEN_TSTAMP_APPA; 2333 #ifdef TCP_SIGNATURE 2334 if (tp->t_flags & TF_SIGNATURE) 2335 hlen += TCPOLEN_SIGLEN; 2336 #endif 2337 return hlen; 2338 } 2339 2340 void 2341 tcp_statinc(u_int stat) 2342 { 2343 2344 KASSERT(stat < TCP_NSTATS); 2345 TCP_STATINC(stat); 2346 } 2347 2348 void 2349 tcp_statadd(u_int stat, uint64_t val) 2350 { 2351 2352 KASSERT(stat < TCP_NSTATS); 2353 TCP_STATADD(stat, val); 2354 } 2355