1 /* $NetBSD: tcp_subr.c,v 1.276 2018/03/29 18:54:48 maxv 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 1.276 2018/03/29 18:54:48 maxv 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 #include <netinet/tcpip.h> 147 148 #ifdef IPSEC 149 #include <netipsec/ipsec.h> 150 #include <netipsec/xform.h> 151 #ifdef INET6 152 #include <netipsec/ipsec6.h> 153 #endif 154 #include <netipsec/key.h> 155 #endif 156 157 158 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */ 159 u_int32_t tcp_now; /* slow ticks, for RFC 1323 timestamps */ 160 161 percpu_t *tcpstat_percpu; 162 163 /* patchable/settable parameters for tcp */ 164 int tcp_mssdflt = TCP_MSS; 165 int tcp_minmss = TCP_MINMSS; 166 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 167 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */ 168 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */ 169 int tcp_do_sack = 1; /* selective acknowledgement */ 170 int tcp_do_win_scale = 1; /* RFC1323 window scaling */ 171 int tcp_do_timestamps = 1; /* RFC1323 timestamps */ 172 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */ 173 int tcp_do_ecn = 0; /* Explicit Congestion Notification */ 174 #ifndef TCP_INIT_WIN 175 #define TCP_INIT_WIN 4 /* initial slow start window */ 176 #endif 177 #ifndef TCP_INIT_WIN_LOCAL 178 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */ 179 #endif 180 /* 181 * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460. 182 * This is to simulate current behavior for iw == 4 183 */ 184 int tcp_init_win_max[] = { 185 1 * 1460, 186 1 * 1460, 187 2 * 1460, 188 2 * 1460, 189 3 * 1460, 190 5 * 1460, 191 6 * 1460, 192 7 * 1460, 193 8 * 1460, 194 9 * 1460, 195 10 * 1460 196 }; 197 int tcp_init_win = TCP_INIT_WIN; 198 int tcp_init_win_local = TCP_INIT_WIN_LOCAL; 199 int tcp_mss_ifmtu = 0; 200 int tcp_rst_ppslim = 100; /* 100pps */ 201 int tcp_ackdrop_ppslim = 100; /* 100pps */ 202 int tcp_do_loopback_cksum = 0; 203 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */ 204 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */ 205 int tcp_sack_tp_maxholes = 32; 206 int tcp_sack_globalmaxholes = 1024; 207 int tcp_sack_globalholes = 0; 208 int tcp_ecn_maxretries = 1; 209 int tcp_msl_enable = 1; /* enable TIME_WAIT truncation */ 210 int tcp_msl_loop = PR_SLOWHZ; /* MSL for loopback */ 211 int tcp_msl_local = 5 * PR_SLOWHZ; /* MSL for 'local' */ 212 int tcp_msl_remote = TCPTV_MSL; /* MSL otherwise */ 213 int tcp_msl_remote_threshold = TCPTV_SRTTDFLT; /* RTT threshold */ 214 int tcp_rttlocal = 0; /* Use RTT to decide who's 'local' */ 215 216 int tcp4_vtw_enable = 0; /* 1 to enable */ 217 int tcp6_vtw_enable = 0; /* 1 to enable */ 218 int tcp_vtw_was_enabled = 0; 219 int tcp_vtw_entries = 1 << 4; /* 16 vestigial TIME_WAIT entries */ 220 221 /* tcb hash */ 222 #ifndef TCBHASHSIZE 223 #define TCBHASHSIZE 128 224 #endif 225 int tcbhashsize = TCBHASHSIZE; 226 227 /* syn hash parameters */ 228 #define TCP_SYN_HASH_SIZE 293 229 #define TCP_SYN_BUCKET_SIZE 35 230 int tcp_syn_cache_size = TCP_SYN_HASH_SIZE; 231 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 232 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 233 struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE]; 234 235 int tcp_freeq(struct tcpcb *); 236 static int tcp_iss_secret_init(void); 237 238 static void tcp_mtudisc_callback(struct in_addr); 239 240 #ifdef INET6 241 static void tcp6_mtudisc(struct in6pcb *, int); 242 #endif 243 244 static struct pool tcpcb_pool; 245 246 static int tcp_drainwanted; 247 248 #ifdef TCP_CSUM_COUNTERS 249 #include <sys/device.h> 250 251 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 252 NULL, "tcp", "hwcsum bad"); 253 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 254 NULL, "tcp", "hwcsum ok"); 255 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 256 NULL, "tcp", "hwcsum data"); 257 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 258 NULL, "tcp", "swcsum"); 259 260 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad); 261 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok); 262 EVCNT_ATTACH_STATIC(tcp_hwcsum_data); 263 EVCNT_ATTACH_STATIC(tcp_swcsum); 264 265 #if defined(INET6) 266 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 267 NULL, "tcp6", "hwcsum bad"); 268 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 269 NULL, "tcp6", "hwcsum ok"); 270 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 271 NULL, "tcp6", "hwcsum data"); 272 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 273 NULL, "tcp6", "swcsum"); 274 275 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad); 276 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok); 277 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data); 278 EVCNT_ATTACH_STATIC(tcp6_swcsum); 279 #endif /* defined(INET6) */ 280 #endif /* TCP_CSUM_COUNTERS */ 281 282 283 #ifdef TCP_OUTPUT_COUNTERS 284 #include <sys/device.h> 285 286 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 287 NULL, "tcp", "output big header"); 288 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 289 NULL, "tcp", "output predict hit"); 290 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 291 NULL, "tcp", "output predict miss"); 292 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 293 NULL, "tcp", "output copy small"); 294 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 295 NULL, "tcp", "output copy big"); 296 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 297 NULL, "tcp", "output reference big"); 298 299 EVCNT_ATTACH_STATIC(tcp_output_bigheader); 300 EVCNT_ATTACH_STATIC(tcp_output_predict_hit); 301 EVCNT_ATTACH_STATIC(tcp_output_predict_miss); 302 EVCNT_ATTACH_STATIC(tcp_output_copysmall); 303 EVCNT_ATTACH_STATIC(tcp_output_copybig); 304 EVCNT_ATTACH_STATIC(tcp_output_refbig); 305 306 #endif /* TCP_OUTPUT_COUNTERS */ 307 308 #ifdef TCP_REASS_COUNTERS 309 #include <sys/device.h> 310 311 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 312 NULL, "tcp_reass", "calls"); 313 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 314 &tcp_reass_, "tcp_reass", "insert into empty queue"); 315 struct evcnt tcp_reass_iteration[8] = { 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"), 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"), 318 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"), 319 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"), 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"), 321 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"), 322 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"), 323 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"), 324 }; 325 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 326 &tcp_reass_, "tcp_reass", "prepend to first"); 327 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 328 &tcp_reass_, "tcp_reass", "prepend"); 329 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 330 &tcp_reass_, "tcp_reass", "insert"); 331 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 332 &tcp_reass_, "tcp_reass", "insert at tail"); 333 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 334 &tcp_reass_, "tcp_reass", "append"); 335 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 336 &tcp_reass_, "tcp_reass", "append to tail fragment"); 337 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 338 &tcp_reass_, "tcp_reass", "overlap at end"); 339 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 340 &tcp_reass_, "tcp_reass", "overlap at start"); 341 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 342 &tcp_reass_, "tcp_reass", "duplicate segment"); 343 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 344 &tcp_reass_, "tcp_reass", "duplicate fragment"); 345 346 EVCNT_ATTACH_STATIC(tcp_reass_); 347 EVCNT_ATTACH_STATIC(tcp_reass_empty); 348 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0); 349 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1); 350 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2); 351 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3); 352 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4); 353 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5); 354 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6); 355 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7); 356 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst); 357 EVCNT_ATTACH_STATIC(tcp_reass_prepend); 358 EVCNT_ATTACH_STATIC(tcp_reass_insert); 359 EVCNT_ATTACH_STATIC(tcp_reass_inserttail); 360 EVCNT_ATTACH_STATIC(tcp_reass_append); 361 EVCNT_ATTACH_STATIC(tcp_reass_appendtail); 362 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail); 363 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront); 364 EVCNT_ATTACH_STATIC(tcp_reass_segdup); 365 EVCNT_ATTACH_STATIC(tcp_reass_fragdup); 366 367 #endif /* TCP_REASS_COUNTERS */ 368 369 #ifdef MBUFTRACE 370 struct mowner tcp_mowner = MOWNER_INIT("tcp", ""); 371 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx"); 372 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx"); 373 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock"); 374 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx"); 375 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx"); 376 #endif 377 378 static int 379 do_tcpinit(void) 380 { 381 382 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize); 383 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", 384 NULL, IPL_SOFTNET); 385 386 tcp_usrreq_init(); 387 388 /* Initialize timer state. */ 389 tcp_timer_init(); 390 391 /* Initialize the compressed state engine. */ 392 syn_cache_init(); 393 394 /* Initialize the congestion control algorithms. */ 395 tcp_congctl_init(); 396 397 /* Initialize the TCPCB template. */ 398 tcp_tcpcb_template(); 399 400 /* Initialize reassembly queue */ 401 tcpipqent_init(); 402 403 /* SACK */ 404 tcp_sack_init(); 405 406 MOWNER_ATTACH(&tcp_tx_mowner); 407 MOWNER_ATTACH(&tcp_rx_mowner); 408 MOWNER_ATTACH(&tcp_reass_mowner); 409 MOWNER_ATTACH(&tcp_sock_mowner); 410 MOWNER_ATTACH(&tcp_sock_tx_mowner); 411 MOWNER_ATTACH(&tcp_sock_rx_mowner); 412 MOWNER_ATTACH(&tcp_mowner); 413 414 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS); 415 416 vtw_earlyinit(); 417 418 tcp_slowtimo_init(); 419 420 return 0; 421 } 422 423 void 424 tcp_init_common(unsigned basehlen) 425 { 426 static ONCE_DECL(dotcpinit); 427 unsigned hlen = basehlen + sizeof(struct tcphdr); 428 unsigned oldhlen; 429 430 if (max_linkhdr + hlen > MHLEN) 431 panic("tcp_init"); 432 while ((oldhlen = max_protohdr) < hlen) 433 atomic_cas_uint(&max_protohdr, oldhlen, hlen); 434 435 RUN_ONCE(&dotcpinit, do_tcpinit); 436 } 437 438 /* 439 * Tcp initialization 440 */ 441 void 442 tcp_init(void) 443 { 444 445 icmp_mtudisc_callback_register(tcp_mtudisc_callback); 446 447 tcp_init_common(sizeof(struct ip)); 448 } 449 450 /* 451 * Create template to be used to send tcp packets on a connection. 452 * Call after host entry created, allocates an mbuf and fills 453 * in a skeletal tcp/ip header, minimizing the amount of work 454 * necessary when the connection is used. 455 */ 456 struct mbuf * 457 tcp_template(struct tcpcb *tp) 458 { 459 struct inpcb *inp = tp->t_inpcb; 460 #ifdef INET6 461 struct in6pcb *in6p = tp->t_in6pcb; 462 #endif 463 struct tcphdr *n; 464 struct mbuf *m; 465 int hlen; 466 467 switch (tp->t_family) { 468 case AF_INET: 469 hlen = sizeof(struct ip); 470 if (inp) 471 break; 472 #ifdef INET6 473 if (in6p) { 474 /* mapped addr case */ 475 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr) 476 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr)) 477 break; 478 } 479 #endif 480 return NULL; /*EINVAL*/ 481 #ifdef INET6 482 case AF_INET6: 483 hlen = sizeof(struct ip6_hdr); 484 if (in6p) { 485 /* more sainty check? */ 486 break; 487 } 488 return NULL; /*EINVAL*/ 489 #endif 490 default: 491 return NULL; /*EAFNOSUPPORT*/ 492 } 493 494 KASSERT(hlen + sizeof(struct tcphdr) <= MCLBYTES); 495 496 m = tp->t_template; 497 if (m && m->m_len == hlen + sizeof(struct tcphdr)) { 498 ; 499 } else { 500 if (m) 501 m_freem(m); 502 m = tp->t_template = NULL; 503 MGETHDR(m, M_DONTWAIT, MT_HEADER); 504 if (m && hlen + sizeof(struct tcphdr) > MHLEN) { 505 MCLGET(m, M_DONTWAIT); 506 if ((m->m_flags & M_EXT) == 0) { 507 m_free(m); 508 m = NULL; 509 } 510 } 511 if (m == NULL) 512 return NULL; 513 MCLAIM(m, &tcp_mowner); 514 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr); 515 } 516 517 memset(mtod(m, void *), 0, m->m_len); 518 519 n = (struct tcphdr *)(mtod(m, char *) + hlen); 520 521 switch (tp->t_family) { 522 case AF_INET: 523 { 524 struct ipovly *ipov; 525 mtod(m, struct ip *)->ip_v = 4; 526 mtod(m, struct ip *)->ip_hl = hlen >> 2; 527 ipov = mtod(m, struct ipovly *); 528 ipov->ih_pr = IPPROTO_TCP; 529 ipov->ih_len = htons(sizeof(struct tcphdr)); 530 if (inp) { 531 ipov->ih_src = inp->inp_laddr; 532 ipov->ih_dst = inp->inp_faddr; 533 } 534 #ifdef INET6 535 else if (in6p) { 536 /* mapped addr case */ 537 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src, 538 sizeof(ipov->ih_src)); 539 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst, 540 sizeof(ipov->ih_dst)); 541 } 542 #endif 543 544 /* 545 * Compute the pseudo-header portion of the checksum 546 * now. We incrementally add in the TCP option and 547 * payload lengths later, and then compute the TCP 548 * checksum right before the packet is sent off onto 549 * the wire. 550 */ 551 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr, 552 ipov->ih_dst.s_addr, 553 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 554 break; 555 } 556 #ifdef INET6 557 case AF_INET6: 558 { 559 struct ip6_hdr *ip6; 560 mtod(m, struct ip *)->ip_v = 6; 561 ip6 = mtod(m, struct ip6_hdr *); 562 ip6->ip6_nxt = IPPROTO_TCP; 563 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 564 ip6->ip6_src = in6p->in6p_laddr; 565 ip6->ip6_dst = in6p->in6p_faddr; 566 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK; 567 if (ip6_auto_flowlabel) { 568 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 569 ip6->ip6_flow |= 570 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 571 } 572 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 573 ip6->ip6_vfc |= IPV6_VERSION; 574 575 /* 576 * Compute the pseudo-header portion of the checksum 577 * now. We incrementally add in the TCP option and 578 * payload lengths later, and then compute the TCP 579 * checksum right before the packet is sent off onto 580 * the wire. 581 */ 582 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr, 583 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)), 584 htonl(IPPROTO_TCP)); 585 break; 586 } 587 #endif 588 } 589 590 if (inp) { 591 n->th_sport = inp->inp_lport; 592 n->th_dport = inp->inp_fport; 593 } 594 #ifdef INET6 595 else if (in6p) { 596 n->th_sport = in6p->in6p_lport; 597 n->th_dport = in6p->in6p_fport; 598 } 599 #endif 600 601 n->th_seq = 0; 602 n->th_ack = 0; 603 n->th_x2 = 0; 604 n->th_off = 5; 605 n->th_flags = 0; 606 n->th_win = 0; 607 n->th_urp = 0; 608 return m; 609 } 610 611 /* 612 * Send a single message to the TCP at address specified by 613 * the given TCP/IP header. If m == 0, then we make a copy 614 * of the tcpiphdr at ti and send directly to the addressed host. 615 * This is used to force keep alive messages out using the TCP 616 * template for a connection tp->t_template. If flags are given 617 * then we send a message back to the TCP which originated the 618 * segment ti, and discard the mbuf containing it and any other 619 * attached mbufs. 620 * 621 * In any case the ack and sequence number of the transmitted 622 * segment are as specified by the parameters. 623 */ 624 int 625 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m, 626 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags) 627 { 628 struct route *ro; 629 int error, tlen, win = 0; 630 int hlen; 631 struct ip *ip; 632 #ifdef INET6 633 struct ip6_hdr *ip6; 634 #endif 635 int family; /* family on packet, not inpcb/in6pcb! */ 636 struct tcphdr *th; 637 638 if (tp != NULL && (flags & TH_RST) == 0) { 639 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 640 641 if (tp->t_inpcb) 642 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 643 #ifdef INET6 644 if (tp->t_in6pcb) 645 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv); 646 #endif 647 } 648 649 th = NULL; /* Quell uninitialized warning */ 650 ip = NULL; 651 #ifdef INET6 652 ip6 = NULL; 653 #endif 654 if (m == NULL) { 655 if (!mtemplate) 656 return EINVAL; 657 658 /* get family information from template */ 659 switch (mtod(mtemplate, struct ip *)->ip_v) { 660 case 4: 661 family = AF_INET; 662 hlen = sizeof(struct ip); 663 break; 664 #ifdef INET6 665 case 6: 666 family = AF_INET6; 667 hlen = sizeof(struct ip6_hdr); 668 break; 669 #endif 670 default: 671 return EAFNOSUPPORT; 672 } 673 674 MGETHDR(m, M_DONTWAIT, MT_HEADER); 675 if (m) { 676 MCLAIM(m, &tcp_tx_mowner); 677 MCLGET(m, M_DONTWAIT); 678 if ((m->m_flags & M_EXT) == 0) { 679 m_free(m); 680 m = NULL; 681 } 682 } 683 if (m == NULL) 684 return ENOBUFS; 685 686 tlen = 0; 687 688 m->m_data += max_linkhdr; 689 bcopy(mtod(mtemplate, void *), mtod(m, void *), 690 mtemplate->m_len); 691 switch (family) { 692 case AF_INET: 693 ip = mtod(m, struct ip *); 694 th = (struct tcphdr *)(ip + 1); 695 break; 696 #ifdef INET6 697 case AF_INET6: 698 ip6 = mtod(m, struct ip6_hdr *); 699 th = (struct tcphdr *)(ip6 + 1); 700 break; 701 #endif 702 } 703 flags = TH_ACK; 704 } else { 705 if ((m->m_flags & M_PKTHDR) == 0) { 706 m_freem(m); 707 return EINVAL; 708 } 709 KASSERT(th0 != NULL); 710 711 /* get family information from m */ 712 switch (mtod(m, struct ip *)->ip_v) { 713 case 4: 714 family = AF_INET; 715 hlen = sizeof(struct ip); 716 ip = mtod(m, struct ip *); 717 break; 718 #ifdef INET6 719 case 6: 720 family = AF_INET6; 721 hlen = sizeof(struct ip6_hdr); 722 ip6 = mtod(m, struct ip6_hdr *); 723 break; 724 #endif 725 default: 726 m_freem(m); 727 return EAFNOSUPPORT; 728 } 729 /* clear h/w csum flags inherited from rx packet */ 730 m->m_pkthdr.csum_flags = 0; 731 732 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2)) 733 tlen = sizeof(*th0); 734 else 735 tlen = th0->th_off << 2; 736 737 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 && 738 mtod(m, char *) + hlen == (char *)th0) { 739 m->m_len = hlen + tlen; 740 m_freem(m->m_next); 741 m->m_next = NULL; 742 } else { 743 struct mbuf *n; 744 745 KASSERT(max_linkhdr + hlen + tlen <= MCLBYTES); 746 747 MGETHDR(n, M_DONTWAIT, MT_HEADER); 748 if (n && max_linkhdr + hlen + tlen > MHLEN) { 749 MCLGET(n, M_DONTWAIT); 750 if ((n->m_flags & M_EXT) == 0) { 751 m_freem(n); 752 n = NULL; 753 } 754 } 755 if (!n) { 756 m_freem(m); 757 return ENOBUFS; 758 } 759 760 MCLAIM(n, &tcp_tx_mowner); 761 n->m_data += max_linkhdr; 762 n->m_len = hlen + tlen; 763 m_copyback(n, 0, hlen, mtod(m, void *)); 764 m_copyback(n, hlen, tlen, (void *)th0); 765 766 m_freem(m); 767 m = n; 768 n = NULL; 769 } 770 771 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 772 switch (family) { 773 case AF_INET: 774 ip = mtod(m, struct ip *); 775 th = (struct tcphdr *)(ip + 1); 776 ip->ip_p = IPPROTO_TCP; 777 xchg(ip->ip_dst, ip->ip_src, struct in_addr); 778 ip->ip_p = IPPROTO_TCP; 779 break; 780 #ifdef INET6 781 case AF_INET6: 782 ip6 = mtod(m, struct ip6_hdr *); 783 th = (struct tcphdr *)(ip6 + 1); 784 ip6->ip6_nxt = IPPROTO_TCP; 785 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 786 ip6->ip6_nxt = IPPROTO_TCP; 787 break; 788 #endif 789 } 790 xchg(th->th_dport, th->th_sport, u_int16_t); 791 #undef xchg 792 tlen = 0; /*be friendly with the following code*/ 793 } 794 th->th_seq = htonl(seq); 795 th->th_ack = htonl(ack); 796 th->th_x2 = 0; 797 if ((flags & TH_SYN) == 0) { 798 if (tp) 799 win >>= tp->rcv_scale; 800 if (win > TCP_MAXWIN) 801 win = TCP_MAXWIN; 802 th->th_win = htons((u_int16_t)win); 803 th->th_off = sizeof (struct tcphdr) >> 2; 804 tlen += sizeof(*th); 805 } else { 806 tlen += th->th_off << 2; 807 } 808 m->m_len = hlen + tlen; 809 m->m_pkthdr.len = hlen + tlen; 810 m_reset_rcvif(m); 811 th->th_flags = flags; 812 th->th_urp = 0; 813 814 switch (family) { 815 case AF_INET: 816 { 817 struct ipovly *ipov = (struct ipovly *)ip; 818 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1); 819 ipov->ih_len = htons((u_int16_t)tlen); 820 821 th->th_sum = 0; 822 th->th_sum = in_cksum(m, hlen + tlen); 823 ip->ip_len = htons(hlen + tlen); 824 ip->ip_ttl = ip_defttl; 825 break; 826 } 827 #ifdef INET6 828 case AF_INET6: 829 { 830 th->th_sum = 0; 831 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 832 tlen); 833 ip6->ip6_plen = htons(tlen); 834 if (tp && tp->t_in6pcb) 835 ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb); 836 else 837 ip6->ip6_hlim = ip6_defhlim; 838 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK; 839 if (ip6_auto_flowlabel) { 840 ip6->ip6_flow |= 841 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 842 } 843 break; 844 } 845 #endif 846 } 847 848 if (tp != NULL && tp->t_inpcb != NULL) { 849 ro = &tp->t_inpcb->inp_route; 850 KASSERT(family == AF_INET); 851 KASSERT(in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)); 852 } 853 #ifdef INET6 854 else if (tp != NULL && tp->t_in6pcb != NULL) { 855 ro = (struct route *)&tp->t_in6pcb->in6p_route; 856 857 #ifdef DIAGNOSTIC 858 if (family == AF_INET) { 859 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr)) 860 panic("tcp_respond: not mapped addr"); 861 if (memcmp(&ip->ip_dst, 862 &tp->t_in6pcb->in6p_faddr.s6_addr32[3], 863 sizeof(ip->ip_dst)) != 0) { 864 panic("tcp_respond: ip_dst != in6p_faddr"); 865 } 866 } else if (family == AF_INET6) { 867 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 868 &tp->t_in6pcb->in6p_faddr)) 869 panic("tcp_respond: ip6_dst != in6p_faddr"); 870 } else 871 panic("tcp_respond: address family mismatch"); 872 #endif 873 } 874 #endif 875 else 876 ro = NULL; 877 878 switch (family) { 879 case AF_INET: 880 error = ip_output(m, NULL, ro, 881 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, 882 tp ? tp->t_inpcb : NULL); 883 break; 884 #ifdef INET6 885 case AF_INET6: 886 error = ip6_output(m, NULL, ro, 0, NULL, 887 tp ? tp->t_in6pcb : NULL, NULL); 888 break; 889 #endif 890 default: 891 error = EAFNOSUPPORT; 892 break; 893 } 894 895 return error; 896 } 897 898 /* 899 * Template TCPCB. Rather than zeroing a new TCPCB and initializing 900 * a bunch of members individually, we maintain this template for the 901 * static and mostly-static components of the TCPCB, and copy it into 902 * the new TCPCB instead. 903 */ 904 static struct tcpcb tcpcb_template = { 905 .t_srtt = TCPTV_SRTTBASE, 906 .t_rttmin = TCPTV_MIN, 907 908 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT, 909 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT, 910 .snd_numholes = 0, 911 .snd_cubic_wmax = 0, 912 .snd_cubic_wmax_last = 0, 913 .snd_cubic_ctime = 0, 914 915 .t_partialacks = -1, 916 .t_bytes_acked = 0, 917 .t_sndrexmitpack = 0, 918 .t_rcvoopack = 0, 919 .t_sndzerowin = 0, 920 }; 921 922 /* 923 * Updates the TCPCB template whenever a parameter that would affect 924 * the template is changed. 925 */ 926 void 927 tcp_tcpcb_template(void) 928 { 929 struct tcpcb *tp = &tcpcb_template; 930 int flags; 931 932 tp->t_peermss = tcp_mssdflt; 933 tp->t_ourmss = tcp_mssdflt; 934 tp->t_segsz = tcp_mssdflt; 935 936 flags = 0; 937 if (tcp_do_rfc1323 && tcp_do_win_scale) 938 flags |= TF_REQ_SCALE; 939 if (tcp_do_rfc1323 && tcp_do_timestamps) 940 flags |= TF_REQ_TSTMP; 941 tp->t_flags = flags; 942 943 /* 944 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 945 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives 946 * reasonable initial retransmit time. 947 */ 948 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1); 949 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 950 TCPTV_MIN, TCPTV_REXMTMAX); 951 952 /* Keep Alive */ 953 tp->t_keepinit = tcp_keepinit; 954 tp->t_keepidle = tcp_keepidle; 955 tp->t_keepintvl = tcp_keepintvl; 956 tp->t_keepcnt = tcp_keepcnt; 957 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 958 959 /* MSL */ 960 tp->t_msl = TCPTV_MSL; 961 } 962 963 /* 964 * Create a new TCP control block, making an 965 * empty reassembly queue and hooking it to the argument 966 * protocol control block. 967 */ 968 /* family selects inpcb, or in6pcb */ 969 struct tcpcb * 970 tcp_newtcpcb(int family, void *aux) 971 { 972 struct tcpcb *tp; 973 int i; 974 975 /* XXX Consider using a pool_cache for speed. */ 976 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 977 if (tp == NULL) 978 return NULL; 979 memcpy(tp, &tcpcb_template, sizeof(*tp)); 980 TAILQ_INIT(&tp->segq); 981 TAILQ_INIT(&tp->timeq); 982 tp->t_family = family; /* may be overridden later on */ 983 TAILQ_INIT(&tp->snd_holes); 984 LIST_INIT(&tp->t_sc); /* XXX can template this */ 985 986 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 987 for (i = 0; i < TCPT_NTIMERS; i++) { 988 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 989 TCP_TIMER_INIT(tp, i); 990 } 991 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 992 993 switch (family) { 994 case AF_INET: 995 { 996 struct inpcb *inp = (struct inpcb *)aux; 997 998 inp->inp_ip.ip_ttl = ip_defttl; 999 inp->inp_ppcb = (void *)tp; 1000 1001 tp->t_inpcb = inp; 1002 tp->t_mtudisc = ip_mtudisc; 1003 break; 1004 } 1005 #ifdef INET6 1006 case AF_INET6: 1007 { 1008 struct in6pcb *in6p = (struct in6pcb *)aux; 1009 1010 in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p); 1011 in6p->in6p_ppcb = (void *)tp; 1012 1013 tp->t_in6pcb = in6p; 1014 /* for IPv6, always try to run path MTU discovery */ 1015 tp->t_mtudisc = 1; 1016 break; 1017 } 1018 #endif /* INET6 */ 1019 default: 1020 for (i = 0; i < TCPT_NTIMERS; i++) 1021 callout_destroy(&tp->t_timer[i]); 1022 callout_destroy(&tp->t_delack_ch); 1023 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1024 return NULL; 1025 } 1026 1027 /* 1028 * Initialize our timebase. When we send timestamps, we take 1029 * the delta from tcp_now -- this means each connection always 1030 * gets a timebase of 1, which makes it, among other things, 1031 * more difficult to determine how long a system has been up, 1032 * and thus how many TCP sequence increments have occurred. 1033 * 1034 * We start with 1, because 0 doesn't work with linux, which 1035 * considers timestamp 0 in a SYN packet as a bug and disables 1036 * timestamps. 1037 */ 1038 tp->ts_timebase = tcp_now - 1; 1039 1040 tcp_congctl_select(tp, tcp_congctl_global_name); 1041 1042 return tp; 1043 } 1044 1045 /* 1046 * Drop a TCP connection, reporting 1047 * the specified error. If connection is synchronized, 1048 * then send a RST to peer. 1049 */ 1050 struct tcpcb * 1051 tcp_drop(struct tcpcb *tp, int errno) 1052 { 1053 struct socket *so = NULL; 1054 1055 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1056 1057 if (tp->t_inpcb) 1058 so = tp->t_inpcb->inp_socket; 1059 #ifdef INET6 1060 if (tp->t_in6pcb) 1061 so = tp->t_in6pcb->in6p_socket; 1062 #endif 1063 if (!so) 1064 return NULL; 1065 1066 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1067 tp->t_state = TCPS_CLOSED; 1068 (void) tcp_output(tp); 1069 TCP_STATINC(TCP_STAT_DROPS); 1070 } else 1071 TCP_STATINC(TCP_STAT_CONNDROPS); 1072 if (errno == ETIMEDOUT && tp->t_softerror) 1073 errno = tp->t_softerror; 1074 so->so_error = errno; 1075 return (tcp_close(tp)); 1076 } 1077 1078 /* 1079 * Close a TCP control block: 1080 * discard all space held by the tcp 1081 * discard internet protocol block 1082 * wake up any sleepers 1083 */ 1084 struct tcpcb * 1085 tcp_close(struct tcpcb *tp) 1086 { 1087 struct inpcb *inp; 1088 #ifdef INET6 1089 struct in6pcb *in6p; 1090 #endif 1091 struct socket *so; 1092 #ifdef RTV_RTT 1093 struct rtentry *rt = NULL; 1094 #endif 1095 struct route *ro; 1096 int j; 1097 1098 inp = tp->t_inpcb; 1099 #ifdef INET6 1100 in6p = tp->t_in6pcb; 1101 #endif 1102 so = NULL; 1103 ro = NULL; 1104 if (inp) { 1105 so = inp->inp_socket; 1106 ro = &inp->inp_route; 1107 } 1108 #ifdef INET6 1109 else if (in6p) { 1110 so = in6p->in6p_socket; 1111 ro = (struct route *)&in6p->in6p_route; 1112 } 1113 #endif 1114 1115 #ifdef RTV_RTT 1116 /* 1117 * If we sent enough data to get some meaningful characteristics, 1118 * save them in the routing entry. 'Enough' is arbitrarily 1119 * defined as the sendpipesize (default 4K) * 16. This would 1120 * give us 16 rtt samples assuming we only get one sample per 1121 * window (the usual case on a long haul net). 16 samples is 1122 * enough for the srtt filter to converge to within 5% of the correct 1123 * value; fewer samples and we could save a very bogus rtt. 1124 * 1125 * Don't update the default route's characteristics and don't 1126 * update anything that the user "locked". 1127 */ 1128 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1129 ro && (rt = rtcache_validate(ro)) != NULL && 1130 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1131 u_long i = 0; 1132 1133 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1134 i = tp->t_srtt * 1135 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1136 if (rt->rt_rmx.rmx_rtt && i) 1137 /* 1138 * filter this update to half the old & half 1139 * the new values, converting scale. 1140 * See route.h and tcp_var.h for a 1141 * description of the scaling constants. 1142 */ 1143 rt->rt_rmx.rmx_rtt = 1144 (rt->rt_rmx.rmx_rtt + i) / 2; 1145 else 1146 rt->rt_rmx.rmx_rtt = i; 1147 } 1148 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1149 i = tp->t_rttvar * 1150 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1151 if (rt->rt_rmx.rmx_rttvar && i) 1152 rt->rt_rmx.rmx_rttvar = 1153 (rt->rt_rmx.rmx_rttvar + i) / 2; 1154 else 1155 rt->rt_rmx.rmx_rttvar = i; 1156 } 1157 /* 1158 * update the pipelimit (ssthresh) if it has been updated 1159 * already or if a pipesize was specified & the threshhold 1160 * got below half the pipesize. I.e., wait for bad news 1161 * before we start updating, then update on both good 1162 * and bad news. 1163 */ 1164 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1165 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1166 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1167 /* 1168 * convert the limit from user data bytes to 1169 * packets then to packet data bytes. 1170 */ 1171 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1172 if (i < 2) 1173 i = 2; 1174 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1175 if (rt->rt_rmx.rmx_ssthresh) 1176 rt->rt_rmx.rmx_ssthresh = 1177 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1178 else 1179 rt->rt_rmx.rmx_ssthresh = i; 1180 } 1181 } 1182 rtcache_unref(rt, ro); 1183 #endif /* RTV_RTT */ 1184 /* free the reassembly queue, if any */ 1185 TCP_REASS_LOCK(tp); 1186 (void) tcp_freeq(tp); 1187 TCP_REASS_UNLOCK(tp); 1188 1189 /* free the SACK holes list. */ 1190 tcp_free_sackholes(tp); 1191 tcp_congctl_release(tp); 1192 syn_cache_cleanup(tp); 1193 1194 if (tp->t_template) { 1195 m_free(tp->t_template); 1196 tp->t_template = NULL; 1197 } 1198 1199 /* 1200 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1201 * the timers can also drop the lock. We need to prevent access 1202 * to the tcpcb as it's half torn down. Flag the pcb as dead 1203 * (prevents access by timers) and only then detach it. 1204 */ 1205 tp->t_flags |= TF_DEAD; 1206 if (inp) { 1207 inp->inp_ppcb = 0; 1208 soisdisconnected(so); 1209 in_pcbdetach(inp); 1210 } 1211 #ifdef INET6 1212 else if (in6p) { 1213 in6p->in6p_ppcb = 0; 1214 soisdisconnected(so); 1215 in6_pcbdetach(in6p); 1216 } 1217 #endif 1218 /* 1219 * pcb is no longer visble elsewhere, so we can safely release 1220 * the lock in callout_halt() if needed. 1221 */ 1222 TCP_STATINC(TCP_STAT_CLOSED); 1223 for (j = 0; j < TCPT_NTIMERS; j++) { 1224 callout_halt(&tp->t_timer[j], softnet_lock); 1225 callout_destroy(&tp->t_timer[j]); 1226 } 1227 callout_halt(&tp->t_delack_ch, softnet_lock); 1228 callout_destroy(&tp->t_delack_ch); 1229 pool_put(&tcpcb_pool, tp); 1230 1231 return NULL; 1232 } 1233 1234 int 1235 tcp_freeq(struct tcpcb *tp) 1236 { 1237 struct ipqent *qe; 1238 int rv = 0; 1239 1240 TCP_REASS_LOCK_CHECK(tp); 1241 1242 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1243 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1244 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1245 m_freem(qe->ipqe_m); 1246 tcpipqent_free(qe); 1247 rv = 1; 1248 } 1249 tp->t_segqlen = 0; 1250 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1251 return (rv); 1252 } 1253 1254 void 1255 tcp_fasttimo(void) 1256 { 1257 if (tcp_drainwanted) { 1258 tcp_drain(); 1259 tcp_drainwanted = 0; 1260 } 1261 } 1262 1263 void 1264 tcp_drainstub(void) 1265 { 1266 tcp_drainwanted = 1; 1267 } 1268 1269 /* 1270 * Protocol drain routine. Called when memory is in short supply. 1271 * Called from pr_fasttimo thus a callout context. 1272 */ 1273 void 1274 tcp_drain(void) 1275 { 1276 struct inpcb_hdr *inph; 1277 struct tcpcb *tp; 1278 1279 mutex_enter(softnet_lock); 1280 KERNEL_LOCK(1, NULL); 1281 1282 /* 1283 * Free the sequence queue of all TCP connections. 1284 */ 1285 TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1286 switch (inph->inph_af) { 1287 case AF_INET: 1288 tp = intotcpcb((struct inpcb *)inph); 1289 break; 1290 #ifdef INET6 1291 case AF_INET6: 1292 tp = in6totcpcb((struct in6pcb *)inph); 1293 break; 1294 #endif 1295 default: 1296 tp = NULL; 1297 break; 1298 } 1299 if (tp != NULL) { 1300 /* 1301 * We may be called from a device's interrupt 1302 * context. If the tcpcb is already busy, 1303 * just bail out now. 1304 */ 1305 if (tcp_reass_lock_try(tp) == 0) 1306 continue; 1307 if (tcp_freeq(tp)) 1308 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1309 TCP_REASS_UNLOCK(tp); 1310 } 1311 } 1312 1313 KERNEL_UNLOCK_ONE(NULL); 1314 mutex_exit(softnet_lock); 1315 } 1316 1317 /* 1318 * Notify a tcp user of an asynchronous error; 1319 * store error as soft error, but wake up user 1320 * (for now, won't do anything until can select for soft error). 1321 */ 1322 void 1323 tcp_notify(struct inpcb *inp, int error) 1324 { 1325 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1326 struct socket *so = inp->inp_socket; 1327 1328 /* 1329 * Ignore some errors if we are hooked up. 1330 * If connection hasn't completed, has retransmitted several times, 1331 * and receives a second error, give up now. This is better 1332 * than waiting a long time to establish a connection that 1333 * can never complete. 1334 */ 1335 if (tp->t_state == TCPS_ESTABLISHED && 1336 (error == EHOSTUNREACH || error == ENETUNREACH || 1337 error == EHOSTDOWN)) { 1338 return; 1339 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1340 tp->t_rxtshift > 3 && tp->t_softerror) 1341 so->so_error = error; 1342 else 1343 tp->t_softerror = error; 1344 cv_broadcast(&so->so_cv); 1345 sorwakeup(so); 1346 sowwakeup(so); 1347 } 1348 1349 #ifdef INET6 1350 void 1351 tcp6_notify(struct in6pcb *in6p, int error) 1352 { 1353 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1354 struct socket *so = in6p->in6p_socket; 1355 1356 /* 1357 * Ignore some errors if we are hooked up. 1358 * If connection hasn't completed, has retransmitted several times, 1359 * and receives a second error, give up now. This is better 1360 * than waiting a long time to establish a connection that 1361 * can never complete. 1362 */ 1363 if (tp->t_state == TCPS_ESTABLISHED && 1364 (error == EHOSTUNREACH || error == ENETUNREACH || 1365 error == EHOSTDOWN)) { 1366 return; 1367 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1368 tp->t_rxtshift > 3 && tp->t_softerror) 1369 so->so_error = error; 1370 else 1371 tp->t_softerror = error; 1372 cv_broadcast(&so->so_cv); 1373 sorwakeup(so); 1374 sowwakeup(so); 1375 } 1376 #endif 1377 1378 #ifdef INET6 1379 void * 1380 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1381 { 1382 struct tcphdr th; 1383 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1384 int nmatch; 1385 struct ip6_hdr *ip6; 1386 const struct sockaddr_in6 *sa6_src = NULL; 1387 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1388 struct mbuf *m; 1389 int off; 1390 1391 if (sa->sa_family != AF_INET6 || 1392 sa->sa_len != sizeof(struct sockaddr_in6)) 1393 return NULL; 1394 if ((unsigned)cmd >= PRC_NCMDS) 1395 return NULL; 1396 else if (cmd == PRC_QUENCH) { 1397 /* 1398 * Don't honor ICMP Source Quench messages meant for 1399 * TCP connections. 1400 */ 1401 return NULL; 1402 } else if (PRC_IS_REDIRECT(cmd)) 1403 notify = in6_rtchange, d = NULL; 1404 else if (cmd == PRC_MSGSIZE) 1405 ; /* special code is present, see below */ 1406 else if (cmd == PRC_HOSTDEAD) 1407 d = NULL; 1408 else if (inet6ctlerrmap[cmd] == 0) 1409 return NULL; 1410 1411 /* if the parameter is from icmp6, decode it. */ 1412 if (d != NULL) { 1413 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1414 m = ip6cp->ip6c_m; 1415 ip6 = ip6cp->ip6c_ip6; 1416 off = ip6cp->ip6c_off; 1417 sa6_src = ip6cp->ip6c_src; 1418 } else { 1419 m = NULL; 1420 ip6 = NULL; 1421 sa6_src = &sa6_any; 1422 off = 0; 1423 } 1424 1425 if (ip6) { 1426 /* 1427 * XXX: We assume that when ip6 is non NULL, 1428 * M and OFF are valid. 1429 */ 1430 1431 /* check if we can safely examine src and dst ports */ 1432 if (m->m_pkthdr.len < off + sizeof(th)) { 1433 if (cmd == PRC_MSGSIZE) 1434 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1435 return NULL; 1436 } 1437 1438 memset(&th, 0, sizeof(th)); 1439 m_copydata(m, off, sizeof(th), (void *)&th); 1440 1441 if (cmd == PRC_MSGSIZE) { 1442 int valid = 0; 1443 1444 /* 1445 * Check to see if we have a valid TCP connection 1446 * corresponding to the address in the ICMPv6 message 1447 * payload. 1448 */ 1449 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1450 th.th_dport, 1451 (const struct in6_addr *)&sa6_src->sin6_addr, 1452 th.th_sport, 0, 0)) 1453 valid++; 1454 1455 /* 1456 * Depending on the value of "valid" and routing table 1457 * size (mtudisc_{hi,lo}wat), we will: 1458 * - recalcurate the new MTU and create the 1459 * corresponding routing entry, or 1460 * - ignore the MTU change notification. 1461 */ 1462 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1463 1464 /* 1465 * no need to call in6_pcbnotify, it should have been 1466 * called via callback if necessary 1467 */ 1468 return NULL; 1469 } 1470 1471 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1472 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1473 if (nmatch == 0 && syn_cache_count && 1474 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1475 inet6ctlerrmap[cmd] == ENETUNREACH || 1476 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1477 syn_cache_unreach((const struct sockaddr *)sa6_src, 1478 sa, &th); 1479 } else { 1480 (void) in6_pcbnotify(&tcbtable, sa, 0, 1481 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1482 } 1483 1484 return NULL; 1485 } 1486 #endif 1487 1488 /* assumes that ip header and tcp header are contiguous on mbuf */ 1489 void * 1490 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1491 { 1492 struct ip *ip = v; 1493 struct tcphdr *th; 1494 struct icmp *icp; 1495 extern const int inetctlerrmap[]; 1496 void (*notify)(struct inpcb *, int) = tcp_notify; 1497 int errno; 1498 int nmatch; 1499 struct tcpcb *tp; 1500 u_int mtu; 1501 tcp_seq seq; 1502 struct inpcb *inp; 1503 #ifdef INET6 1504 struct in6pcb *in6p; 1505 struct in6_addr src6, dst6; 1506 #endif 1507 1508 if (sa->sa_family != AF_INET || 1509 sa->sa_len != sizeof(struct sockaddr_in)) 1510 return NULL; 1511 if ((unsigned)cmd >= PRC_NCMDS) 1512 return NULL; 1513 errno = inetctlerrmap[cmd]; 1514 if (cmd == PRC_QUENCH) 1515 /* 1516 * Don't honor ICMP Source Quench messages meant for 1517 * TCP connections. 1518 */ 1519 return NULL; 1520 else if (PRC_IS_REDIRECT(cmd)) 1521 notify = in_rtchange, ip = 0; 1522 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1523 /* 1524 * Check to see if we have a valid TCP connection 1525 * corresponding to the address in the ICMP message 1526 * payload. 1527 * 1528 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1529 */ 1530 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1531 #ifdef INET6 1532 in6_in_2_v4mapin6(&ip->ip_src, &src6); 1533 in6_in_2_v4mapin6(&ip->ip_dst, &dst6); 1534 #endif 1535 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1536 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL) 1537 #ifdef INET6 1538 in6p = NULL; 1539 #else 1540 ; 1541 #endif 1542 #ifdef INET6 1543 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1544 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL) 1545 ; 1546 #endif 1547 else 1548 return NULL; 1549 1550 /* 1551 * Now that we've validated that we are actually communicating 1552 * with the host indicated in the ICMP message, locate the 1553 * ICMP header, recalculate the new MTU, and create the 1554 * corresponding routing entry. 1555 */ 1556 icp = (struct icmp *)((char *)ip - 1557 offsetof(struct icmp, icmp_ip)); 1558 if (inp) { 1559 if ((tp = intotcpcb(inp)) == NULL) 1560 return NULL; 1561 } 1562 #ifdef INET6 1563 else if (in6p) { 1564 if ((tp = in6totcpcb(in6p)) == NULL) 1565 return NULL; 1566 } 1567 #endif 1568 else 1569 return NULL; 1570 seq = ntohl(th->th_seq); 1571 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1572 return NULL; 1573 /* 1574 * If the ICMP message advertises a Next-Hop MTU 1575 * equal or larger than the maximum packet size we have 1576 * ever sent, drop the message. 1577 */ 1578 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1579 if (mtu >= tp->t_pmtud_mtu_sent) 1580 return NULL; 1581 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1582 /* 1583 * Calculate new MTU, and create corresponding 1584 * route (traditional PMTUD). 1585 */ 1586 tp->t_flags &= ~TF_PMTUD_PEND; 1587 icmp_mtudisc(icp, ip->ip_dst); 1588 } else { 1589 /* 1590 * Record the information got in the ICMP 1591 * message; act on it later. 1592 * If we had already recorded an ICMP message, 1593 * replace the old one only if the new message 1594 * refers to an older TCP segment 1595 */ 1596 if (tp->t_flags & TF_PMTUD_PEND) { 1597 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1598 return NULL; 1599 } else 1600 tp->t_flags |= TF_PMTUD_PEND; 1601 tp->t_pmtud_th_seq = seq; 1602 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1603 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1604 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1605 } 1606 return NULL; 1607 } else if (cmd == PRC_HOSTDEAD) 1608 ip = 0; 1609 else if (errno == 0) 1610 return NULL; 1611 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1612 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1613 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1614 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1615 if (nmatch == 0 && syn_cache_count && 1616 (inetctlerrmap[cmd] == EHOSTUNREACH || 1617 inetctlerrmap[cmd] == ENETUNREACH || 1618 inetctlerrmap[cmd] == EHOSTDOWN)) { 1619 struct sockaddr_in sin; 1620 memset(&sin, 0, sizeof(sin)); 1621 sin.sin_len = sizeof(sin); 1622 sin.sin_family = AF_INET; 1623 sin.sin_port = th->th_sport; 1624 sin.sin_addr = ip->ip_src; 1625 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1626 } 1627 1628 /* XXX mapped address case */ 1629 } else 1630 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1631 notify); 1632 return NULL; 1633 } 1634 1635 /* 1636 * When a source quench is received, we are being notified of congestion. 1637 * Close the congestion window down to the Loss Window (one segment). 1638 * We will gradually open it again as we proceed. 1639 */ 1640 void 1641 tcp_quench(struct inpcb *inp, int errno) 1642 { 1643 struct tcpcb *tp = intotcpcb(inp); 1644 1645 if (tp) { 1646 tp->snd_cwnd = tp->t_segsz; 1647 tp->t_bytes_acked = 0; 1648 } 1649 } 1650 1651 #ifdef INET6 1652 void 1653 tcp6_quench(struct in6pcb *in6p, int errno) 1654 { 1655 struct tcpcb *tp = in6totcpcb(in6p); 1656 1657 if (tp) { 1658 tp->snd_cwnd = tp->t_segsz; 1659 tp->t_bytes_acked = 0; 1660 } 1661 } 1662 #endif 1663 1664 /* 1665 * Path MTU Discovery handlers. 1666 */ 1667 void 1668 tcp_mtudisc_callback(struct in_addr faddr) 1669 { 1670 #ifdef INET6 1671 struct in6_addr in6; 1672 #endif 1673 1674 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1675 #ifdef INET6 1676 in6_in_2_v4mapin6(&faddr, &in6); 1677 tcp6_mtudisc_callback(&in6); 1678 #endif 1679 } 1680 1681 /* 1682 * On receipt of path MTU corrections, flush old route and replace it 1683 * with the new one. Retransmit all unacknowledged packets, to ensure 1684 * that all packets will be received. 1685 */ 1686 void 1687 tcp_mtudisc(struct inpcb *inp, int errno) 1688 { 1689 struct tcpcb *tp = intotcpcb(inp); 1690 struct rtentry *rt; 1691 1692 if (tp == NULL) 1693 return; 1694 1695 rt = in_pcbrtentry(inp); 1696 if (rt != NULL) { 1697 /* 1698 * If this was not a host route, remove and realloc. 1699 */ 1700 if ((rt->rt_flags & RTF_HOST) == 0) { 1701 in_pcbrtentry_unref(rt, inp); 1702 in_rtchange(inp, errno); 1703 if ((rt = in_pcbrtentry(inp)) == NULL) 1704 return; 1705 } 1706 1707 /* 1708 * Slow start out of the error condition. We 1709 * use the MTU because we know it's smaller 1710 * than the previously transmitted segment. 1711 * 1712 * Note: This is more conservative than the 1713 * suggestion in draft-floyd-incr-init-win-03. 1714 */ 1715 if (rt->rt_rmx.rmx_mtu != 0) 1716 tp->snd_cwnd = 1717 TCP_INITIAL_WINDOW(tcp_init_win, 1718 rt->rt_rmx.rmx_mtu); 1719 in_pcbrtentry_unref(rt, inp); 1720 } 1721 1722 /* 1723 * Resend unacknowledged packets. 1724 */ 1725 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1726 tcp_output(tp); 1727 } 1728 1729 #ifdef INET6 1730 /* 1731 * Path MTU Discovery handlers. 1732 */ 1733 void 1734 tcp6_mtudisc_callback(struct in6_addr *faddr) 1735 { 1736 struct sockaddr_in6 sin6; 1737 1738 memset(&sin6, 0, sizeof(sin6)); 1739 sin6.sin6_family = AF_INET6; 1740 sin6.sin6_len = sizeof(struct sockaddr_in6); 1741 sin6.sin6_addr = *faddr; 1742 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1743 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1744 } 1745 1746 void 1747 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1748 { 1749 struct tcpcb *tp = in6totcpcb(in6p); 1750 struct rtentry *rt; 1751 1752 if (tp == NULL) 1753 return; 1754 1755 rt = in6_pcbrtentry(in6p); 1756 if (rt != NULL) { 1757 /* 1758 * If this was not a host route, remove and realloc. 1759 */ 1760 if ((rt->rt_flags & RTF_HOST) == 0) { 1761 in6_pcbrtentry_unref(rt, in6p); 1762 in6_rtchange(in6p, errno); 1763 rt = in6_pcbrtentry(in6p); 1764 if (rt == NULL) 1765 return; 1766 } 1767 1768 /* 1769 * Slow start out of the error condition. We 1770 * use the MTU because we know it's smaller 1771 * than the previously transmitted segment. 1772 * 1773 * Note: This is more conservative than the 1774 * suggestion in draft-floyd-incr-init-win-03. 1775 */ 1776 if (rt->rt_rmx.rmx_mtu != 0) { 1777 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1778 rt->rt_rmx.rmx_mtu); 1779 } 1780 in6_pcbrtentry_unref(rt, in6p); 1781 } 1782 1783 /* 1784 * Resend unacknowledged packets. 1785 */ 1786 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1787 tcp_output(tp); 1788 } 1789 #endif /* INET6 */ 1790 1791 /* 1792 * Compute the MSS to advertise to the peer. Called only during 1793 * the 3-way handshake. If we are the server (peer initiated 1794 * connection), we are called with a pointer to the interface 1795 * on which the SYN packet arrived. If we are the client (we 1796 * initiated connection), we are called with a pointer to the 1797 * interface out which this connection should go. 1798 * 1799 * NOTE: Do not subtract IP option/extension header size nor IPsec 1800 * header size from MSS advertisement. MSS option must hold the maximum 1801 * segment size we can accept, so it must always be: 1802 * max(if mtu) - ip header - tcp header 1803 */ 1804 u_long 1805 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1806 { 1807 extern u_long in_maxmtu; 1808 u_long mss = 0; 1809 u_long hdrsiz; 1810 1811 /* 1812 * In order to avoid defeating path MTU discovery on the peer, 1813 * we advertise the max MTU of all attached networks as our MSS, 1814 * per RFC 1191, section 3.1. 1815 * 1816 * We provide the option to advertise just the MTU of 1817 * the interface on which we hope this connection will 1818 * be receiving. If we are responding to a SYN, we 1819 * will have a pretty good idea about this, but when 1820 * initiating a connection there is a bit more doubt. 1821 * 1822 * We also need to ensure that loopback has a large enough 1823 * MSS, as the loopback MTU is never included in in_maxmtu. 1824 */ 1825 1826 if (ifp != NULL) 1827 switch (af) { 1828 case AF_INET: 1829 mss = ifp->if_mtu; 1830 break; 1831 #ifdef INET6 1832 case AF_INET6: 1833 mss = IN6_LINKMTU(ifp); 1834 break; 1835 #endif 1836 } 1837 1838 if (tcp_mss_ifmtu == 0) 1839 switch (af) { 1840 case AF_INET: 1841 mss = max(in_maxmtu, mss); 1842 break; 1843 #ifdef INET6 1844 case AF_INET6: 1845 mss = max(in6_maxmtu, mss); 1846 break; 1847 #endif 1848 } 1849 1850 switch (af) { 1851 case AF_INET: 1852 hdrsiz = sizeof(struct ip); 1853 break; 1854 #ifdef INET6 1855 case AF_INET6: 1856 hdrsiz = sizeof(struct ip6_hdr); 1857 break; 1858 #endif 1859 default: 1860 hdrsiz = 0; 1861 break; 1862 } 1863 hdrsiz += sizeof(struct tcphdr); 1864 if (mss > hdrsiz) 1865 mss -= hdrsiz; 1866 1867 mss = max(tcp_mssdflt, mss); 1868 return (mss); 1869 } 1870 1871 /* 1872 * Set connection variables based on the peer's advertised MSS. 1873 * We are passed the TCPCB for the actual connection. If we 1874 * are the server, we are called by the compressed state engine 1875 * when the 3-way handshake is complete. If we are the client, 1876 * we are called when we receive the SYN,ACK from the server. 1877 * 1878 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1879 * before this routine is called! 1880 */ 1881 void 1882 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1883 { 1884 struct socket *so; 1885 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1886 struct rtentry *rt; 1887 #endif 1888 u_long bufsize; 1889 int mss; 1890 1891 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1892 1893 so = NULL; 1894 rt = NULL; 1895 1896 if (tp->t_inpcb) { 1897 so = tp->t_inpcb->inp_socket; 1898 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1899 rt = in_pcbrtentry(tp->t_inpcb); 1900 #endif 1901 } 1902 1903 #ifdef INET6 1904 if (tp->t_in6pcb) { 1905 so = tp->t_in6pcb->in6p_socket; 1906 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1907 rt = in6_pcbrtentry(tp->t_in6pcb); 1908 #endif 1909 } 1910 #endif 1911 1912 /* 1913 * As per RFC1122, use the default MSS value, unless they 1914 * sent us an offer. Do not accept offers less than 256 bytes. 1915 */ 1916 mss = tcp_mssdflt; 1917 if (offer) 1918 mss = offer; 1919 mss = max(mss, 256); /* sanity */ 1920 tp->t_peermss = mss; 1921 mss -= tcp_optlen(tp); 1922 if (tp->t_inpcb) 1923 mss -= ip_optlen(tp->t_inpcb); 1924 #ifdef INET6 1925 if (tp->t_in6pcb) 1926 mss -= ip6_optlen(tp->t_in6pcb); 1927 #endif 1928 1929 /* 1930 * If there's a pipesize, change the socket buffer to that size. 1931 * Make the socket buffer an integral number of MSS units. If 1932 * the MSS is larger than the socket buffer, artificially decrease 1933 * the MSS. 1934 */ 1935 #ifdef RTV_SPIPE 1936 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1937 bufsize = rt->rt_rmx.rmx_sendpipe; 1938 else 1939 #endif 1940 { 1941 KASSERT(so != NULL); 1942 bufsize = so->so_snd.sb_hiwat; 1943 } 1944 if (bufsize < mss) 1945 mss = bufsize; 1946 else { 1947 bufsize = roundup(bufsize, mss); 1948 if (bufsize > sb_max) 1949 bufsize = sb_max; 1950 (void) sbreserve(&so->so_snd, bufsize, so); 1951 } 1952 tp->t_segsz = mss; 1953 1954 #ifdef RTV_SSTHRESH 1955 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1956 /* 1957 * There's some sort of gateway or interface buffer 1958 * limit on the path. Use this to set the slow 1959 * start threshold, but set the threshold to no less 1960 * than 2 * MSS. 1961 */ 1962 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1963 } 1964 #endif 1965 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1966 if (tp->t_inpcb) 1967 in_pcbrtentry_unref(rt, tp->t_inpcb); 1968 #ifdef INET6 1969 if (tp->t_in6pcb) 1970 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 1971 #endif 1972 #endif 1973 } 1974 1975 /* 1976 * Processing necessary when a TCP connection is established. 1977 */ 1978 void 1979 tcp_established(struct tcpcb *tp) 1980 { 1981 struct socket *so; 1982 #ifdef RTV_RPIPE 1983 struct rtentry *rt; 1984 #endif 1985 u_long bufsize; 1986 1987 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1988 1989 so = NULL; 1990 rt = NULL; 1991 1992 /* This is a while() to reduce the dreadful stairstepping below */ 1993 while (tp->t_inpcb) { 1994 so = tp->t_inpcb->inp_socket; 1995 #if defined(RTV_RPIPE) 1996 rt = in_pcbrtentry(tp->t_inpcb); 1997 #endif 1998 if (__predict_true(tcp_msl_enable)) { 1999 if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) { 2000 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2001 break; 2002 } 2003 2004 if (__predict_false(tcp_rttlocal)) { 2005 /* This may be adjusted by tcp_input */ 2006 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2007 break; 2008 } 2009 if (in_localaddr(tp->t_inpcb->inp_faddr)) { 2010 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2011 break; 2012 } 2013 } 2014 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2015 break; 2016 } 2017 2018 #ifdef INET6 2019 /* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */ 2020 while (!tp->t_inpcb && tp->t_in6pcb) { 2021 so = tp->t_in6pcb->in6p_socket; 2022 #if defined(RTV_RPIPE) 2023 rt = in6_pcbrtentry(tp->t_in6pcb); 2024 #endif 2025 if (__predict_true(tcp_msl_enable)) { 2026 extern const struct in6_addr in6addr_loopback; 2027 2028 if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr, 2029 &in6addr_loopback)) { 2030 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 2031 break; 2032 } 2033 2034 if (__predict_false(tcp_rttlocal)) { 2035 /* This may be adjusted by tcp_input */ 2036 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2037 break; 2038 } 2039 if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) { 2040 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2041 break; 2042 } 2043 } 2044 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2045 break; 2046 } 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, tcp_seq addin) 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 addin)); 2147 } 2148 #ifdef INET6 2149 if (tp->t_in6pcb != NULL) { 2150 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2151 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2152 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2153 addin)); 2154 } 2155 #endif 2156 2157 panic("tcp_new_iss: unreachable"); 2158 } 2159 2160 static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2161 2162 /* 2163 * Initialize RFC 1948 ISS Secret 2164 */ 2165 static int 2166 tcp_iss_secret_init(void) 2167 { 2168 cprng_strong(kern_cprng, 2169 tcp_iss_secret, sizeof(tcp_iss_secret), 0); 2170 2171 return 0; 2172 } 2173 2174 /* 2175 * This routine actually generates a new TCP initial sequence number. 2176 */ 2177 tcp_seq 2178 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2179 size_t addrsz, tcp_seq addin) 2180 { 2181 tcp_seq tcp_iss; 2182 2183 if (tcp_do_rfc1948) { 2184 MD5_CTX ctx; 2185 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2186 static ONCE_DECL(tcp_iss_secret_control); 2187 2188 /* 2189 * If we haven't been here before, initialize our cryptographic 2190 * hash secret. 2191 */ 2192 RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init); 2193 2194 /* 2195 * Compute the base value of the ISS. It is a hash 2196 * of (saddr, sport, daddr, dport, secret). 2197 */ 2198 MD5Init(&ctx); 2199 2200 MD5Update(&ctx, (u_char *) laddr, addrsz); 2201 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2202 2203 MD5Update(&ctx, (u_char *) faddr, addrsz); 2204 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2205 2206 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2207 2208 MD5Final(hash, &ctx); 2209 2210 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2211 2212 /* 2213 * Now increment our "timer", and add it in to 2214 * the computed value. 2215 * 2216 * XXX Use `addin'? 2217 * XXX TCP_ISSINCR too large to use? 2218 */ 2219 tcp_iss_seq += TCP_ISSINCR; 2220 #ifdef TCPISS_DEBUG 2221 printf("ISS hash 0x%08x, ", tcp_iss); 2222 #endif 2223 tcp_iss += tcp_iss_seq + addin; 2224 #ifdef TCPISS_DEBUG 2225 printf("new ISS 0x%08x\n", tcp_iss); 2226 #endif 2227 } else { 2228 /* 2229 * Randomize. 2230 */ 2231 tcp_iss = cprng_fast32(); 2232 2233 /* 2234 * If we were asked to add some amount to a known value, 2235 * we will take a random value obtained above, mask off 2236 * the upper bits, and add in the known value. We also 2237 * add in a constant to ensure that we are at least a 2238 * certain distance from the original value. 2239 * 2240 * This is used when an old connection is in timed wait 2241 * and we have a new one coming in, for instance. 2242 */ 2243 if (addin != 0) { 2244 #ifdef TCPISS_DEBUG 2245 printf("Random %08x, ", tcp_iss); 2246 #endif 2247 tcp_iss &= TCP_ISS_RANDOM_MASK; 2248 tcp_iss += addin + TCP_ISSINCR; 2249 #ifdef TCPISS_DEBUG 2250 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2251 #endif 2252 } else { 2253 tcp_iss &= TCP_ISS_RANDOM_MASK; 2254 tcp_iss += tcp_iss_seq; 2255 tcp_iss_seq += TCP_ISSINCR; 2256 #ifdef TCPISS_DEBUG 2257 printf("ISS %08x\n", tcp_iss); 2258 #endif 2259 } 2260 } 2261 2262 return (tcp_iss); 2263 } 2264 2265 #if defined(IPSEC) 2266 /* compute ESP/AH header size for TCP, including outer IP header. */ 2267 size_t 2268 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2269 { 2270 struct inpcb *inp; 2271 size_t hdrsiz; 2272 2273 /* XXX mapped addr case (tp->t_in6pcb) */ 2274 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2275 return 0; 2276 switch (tp->t_family) { 2277 case AF_INET: 2278 /* XXX: should use correct direction. */ 2279 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2280 break; 2281 default: 2282 hdrsiz = 0; 2283 break; 2284 } 2285 2286 return hdrsiz; 2287 } 2288 2289 #ifdef INET6 2290 size_t 2291 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2292 { 2293 struct in6pcb *in6p; 2294 size_t hdrsiz; 2295 2296 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2297 return 0; 2298 switch (tp->t_family) { 2299 case AF_INET6: 2300 /* XXX: should use correct direction. */ 2301 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2302 break; 2303 case AF_INET: 2304 /* mapped address case - tricky */ 2305 default: 2306 hdrsiz = 0; 2307 break; 2308 } 2309 2310 return hdrsiz; 2311 } 2312 #endif 2313 #endif /*IPSEC*/ 2314 2315 /* 2316 * Determine the length of the TCP options for this connection. 2317 * 2318 * XXX: What do we do for SACK, when we add that? Just reserve 2319 * all of the space? Otherwise we can't exactly be incrementing 2320 * cwnd by an amount that varies depending on the amount we last 2321 * had to SACK! 2322 */ 2323 2324 u_int 2325 tcp_optlen(struct tcpcb *tp) 2326 { 2327 u_int optlen; 2328 2329 optlen = 0; 2330 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2331 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2332 optlen += TCPOLEN_TSTAMP_APPA; 2333 2334 #ifdef TCP_SIGNATURE 2335 if (tp->t_flags & TF_SIGNATURE) 2336 optlen += TCPOLEN_SIGLEN; 2337 #endif 2338 2339 return optlen; 2340 } 2341 2342 u_int 2343 tcp_hdrsz(struct tcpcb *tp) 2344 { 2345 u_int hlen; 2346 2347 switch (tp->t_family) { 2348 #ifdef INET6 2349 case AF_INET6: 2350 hlen = sizeof(struct ip6_hdr); 2351 break; 2352 #endif 2353 case AF_INET: 2354 hlen = sizeof(struct ip); 2355 break; 2356 default: 2357 hlen = 0; 2358 break; 2359 } 2360 hlen += sizeof(struct tcphdr); 2361 2362 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2363 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2364 hlen += TCPOLEN_TSTAMP_APPA; 2365 #ifdef TCP_SIGNATURE 2366 if (tp->t_flags & TF_SIGNATURE) 2367 hlen += TCPOLEN_SIGLEN; 2368 #endif 2369 return hlen; 2370 } 2371 2372 void 2373 tcp_statinc(u_int stat) 2374 { 2375 2376 KASSERT(stat < TCP_NSTATS); 2377 TCP_STATINC(stat); 2378 } 2379 2380 void 2381 tcp_statadd(u_int stat, uint64_t val) 2382 { 2383 2384 KASSERT(stat < TCP_NSTATS); 2385 TCP_STATADD(stat, val); 2386 } 2387