1 /* $NetBSD: tcp_subr.c,v 1.281 2018/09/03 16:29:36 riastradh 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 1.281 2018/09/03 16:29:36 riastradh 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 = tcp_keepinit; 952 tp->t_keepidle = tcp_keepidle; 953 tp->t_keepintvl = tcp_keepintvl; 954 tp->t_keepcnt = tcp_keepcnt; 955 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl; 956 957 /* MSL */ 958 tp->t_msl = TCPTV_MSL; 959 } 960 961 /* 962 * Create a new TCP control block, making an 963 * empty reassembly queue and hooking it to the argument 964 * protocol control block. 965 */ 966 /* family selects inpcb, or in6pcb */ 967 struct tcpcb * 968 tcp_newtcpcb(int family, void *aux) 969 { 970 struct tcpcb *tp; 971 int i; 972 973 /* XXX Consider using a pool_cache for speed. */ 974 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */ 975 if (tp == NULL) 976 return NULL; 977 memcpy(tp, &tcpcb_template, sizeof(*tp)); 978 TAILQ_INIT(&tp->segq); 979 TAILQ_INIT(&tp->timeq); 980 tp->t_family = family; /* may be overridden later on */ 981 TAILQ_INIT(&tp->snd_holes); 982 LIST_INIT(&tp->t_sc); /* XXX can template this */ 983 984 /* Don't sweat this loop; hopefully the compiler will unroll it. */ 985 for (i = 0; i < TCPT_NTIMERS; i++) { 986 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE); 987 TCP_TIMER_INIT(tp, i); 988 } 989 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE); 990 991 switch (family) { 992 case AF_INET: 993 { 994 struct inpcb *inp = (struct inpcb *)aux; 995 996 inp->inp_ip.ip_ttl = ip_defttl; 997 inp->inp_ppcb = (void *)tp; 998 999 tp->t_inpcb = inp; 1000 tp->t_mtudisc = ip_mtudisc; 1001 break; 1002 } 1003 #ifdef INET6 1004 case AF_INET6: 1005 { 1006 struct in6pcb *in6p = (struct in6pcb *)aux; 1007 1008 in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p); 1009 in6p->in6p_ppcb = (void *)tp; 1010 1011 tp->t_in6pcb = in6p; 1012 /* for IPv6, always try to run path MTU discovery */ 1013 tp->t_mtudisc = 1; 1014 break; 1015 } 1016 #endif /* INET6 */ 1017 default: 1018 for (i = 0; i < TCPT_NTIMERS; i++) 1019 callout_destroy(&tp->t_timer[i]); 1020 callout_destroy(&tp->t_delack_ch); 1021 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */ 1022 return NULL; 1023 } 1024 1025 /* 1026 * Initialize our timebase. When we send timestamps, we take 1027 * the delta from tcp_now -- this means each connection always 1028 * gets a timebase of 1, which makes it, among other things, 1029 * more difficult to determine how long a system has been up, 1030 * and thus how many TCP sequence increments have occurred. 1031 * 1032 * We start with 1, because 0 doesn't work with linux, which 1033 * considers timestamp 0 in a SYN packet as a bug and disables 1034 * timestamps. 1035 */ 1036 tp->ts_timebase = tcp_now - 1; 1037 1038 tcp_congctl_select(tp, tcp_congctl_global_name); 1039 1040 return tp; 1041 } 1042 1043 /* 1044 * Drop a TCP connection, reporting 1045 * the specified error. If connection is synchronized, 1046 * then send a RST to peer. 1047 */ 1048 struct tcpcb * 1049 tcp_drop(struct tcpcb *tp, int errno) 1050 { 1051 struct socket *so = NULL; 1052 1053 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1054 1055 if (tp->t_inpcb) 1056 so = tp->t_inpcb->inp_socket; 1057 #ifdef INET6 1058 if (tp->t_in6pcb) 1059 so = tp->t_in6pcb->in6p_socket; 1060 #endif 1061 if (!so) 1062 return NULL; 1063 1064 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1065 tp->t_state = TCPS_CLOSED; 1066 (void) tcp_output(tp); 1067 TCP_STATINC(TCP_STAT_DROPS); 1068 } else 1069 TCP_STATINC(TCP_STAT_CONNDROPS); 1070 if (errno == ETIMEDOUT && tp->t_softerror) 1071 errno = tp->t_softerror; 1072 so->so_error = errno; 1073 return (tcp_close(tp)); 1074 } 1075 1076 /* 1077 * Close a TCP control block: 1078 * discard all space held by the tcp 1079 * discard internet protocol block 1080 * wake up any sleepers 1081 */ 1082 struct tcpcb * 1083 tcp_close(struct tcpcb *tp) 1084 { 1085 struct inpcb *inp; 1086 #ifdef INET6 1087 struct in6pcb *in6p; 1088 #endif 1089 struct socket *so; 1090 #ifdef RTV_RTT 1091 struct rtentry *rt = NULL; 1092 #endif 1093 struct route *ro; 1094 int j; 1095 1096 inp = tp->t_inpcb; 1097 #ifdef INET6 1098 in6p = tp->t_in6pcb; 1099 #endif 1100 so = NULL; 1101 ro = NULL; 1102 if (inp) { 1103 so = inp->inp_socket; 1104 ro = &inp->inp_route; 1105 } 1106 #ifdef INET6 1107 else if (in6p) { 1108 so = in6p->in6p_socket; 1109 ro = (struct route *)&in6p->in6p_route; 1110 } 1111 #endif 1112 1113 #ifdef RTV_RTT 1114 /* 1115 * If we sent enough data to get some meaningful characteristics, 1116 * save them in the routing entry. 'Enough' is arbitrarily 1117 * defined as the sendpipesize (default 4K) * 16. This would 1118 * give us 16 rtt samples assuming we only get one sample per 1119 * window (the usual case on a long haul net). 16 samples is 1120 * enough for the srtt filter to converge to within 5% of the correct 1121 * value; fewer samples and we could save a very bogus rtt. 1122 * 1123 * Don't update the default route's characteristics and don't 1124 * update anything that the user "locked". 1125 */ 1126 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) && 1127 ro && (rt = rtcache_validate(ro)) != NULL && 1128 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) { 1129 u_long i = 0; 1130 1131 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 1132 i = tp->t_srtt * 1133 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 1134 if (rt->rt_rmx.rmx_rtt && i) 1135 /* 1136 * filter this update to half the old & half 1137 * the new values, converting scale. 1138 * See route.h and tcp_var.h for a 1139 * description of the scaling constants. 1140 */ 1141 rt->rt_rmx.rmx_rtt = 1142 (rt->rt_rmx.rmx_rtt + i) / 2; 1143 else 1144 rt->rt_rmx.rmx_rtt = i; 1145 } 1146 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 1147 i = tp->t_rttvar * 1148 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2)); 1149 if (rt->rt_rmx.rmx_rttvar && i) 1150 rt->rt_rmx.rmx_rttvar = 1151 (rt->rt_rmx.rmx_rttvar + i) / 2; 1152 else 1153 rt->rt_rmx.rmx_rttvar = i; 1154 } 1155 /* 1156 * update the pipelimit (ssthresh) if it has been updated 1157 * already or if a pipesize was specified & the threshhold 1158 * got below half the pipesize. I.e., wait for bad news 1159 * before we start updating, then update on both good 1160 * and bad news. 1161 */ 1162 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 1163 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) || 1164 i < (rt->rt_rmx.rmx_sendpipe / 2)) { 1165 /* 1166 * convert the limit from user data bytes to 1167 * packets then to packet data bytes. 1168 */ 1169 i = (i + tp->t_segsz / 2) / tp->t_segsz; 1170 if (i < 2) 1171 i = 2; 1172 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr)); 1173 if (rt->rt_rmx.rmx_ssthresh) 1174 rt->rt_rmx.rmx_ssthresh = 1175 (rt->rt_rmx.rmx_ssthresh + i) / 2; 1176 else 1177 rt->rt_rmx.rmx_ssthresh = i; 1178 } 1179 } 1180 rtcache_unref(rt, ro); 1181 #endif /* RTV_RTT */ 1182 /* free the reassembly queue, if any */ 1183 TCP_REASS_LOCK(tp); 1184 (void) tcp_freeq(tp); 1185 TCP_REASS_UNLOCK(tp); 1186 1187 /* free the SACK holes list. */ 1188 tcp_free_sackholes(tp); 1189 tcp_congctl_release(tp); 1190 syn_cache_cleanup(tp); 1191 1192 if (tp->t_template) { 1193 m_free(tp->t_template); 1194 tp->t_template = NULL; 1195 } 1196 1197 /* 1198 * Detaching the pcb will unlock the socket/tcpcb, and stopping 1199 * the timers can also drop the lock. We need to prevent access 1200 * to the tcpcb as it's half torn down. Flag the pcb as dead 1201 * (prevents access by timers) and only then detach it. 1202 */ 1203 tp->t_flags |= TF_DEAD; 1204 if (inp) { 1205 inp->inp_ppcb = 0; 1206 soisdisconnected(so); 1207 in_pcbdetach(inp); 1208 } 1209 #ifdef INET6 1210 else if (in6p) { 1211 in6p->in6p_ppcb = 0; 1212 soisdisconnected(so); 1213 in6_pcbdetach(in6p); 1214 } 1215 #endif 1216 /* 1217 * pcb is no longer visble elsewhere, so we can safely release 1218 * the lock in callout_halt() if needed. 1219 */ 1220 TCP_STATINC(TCP_STAT_CLOSED); 1221 for (j = 0; j < TCPT_NTIMERS; j++) { 1222 callout_halt(&tp->t_timer[j], softnet_lock); 1223 callout_destroy(&tp->t_timer[j]); 1224 } 1225 callout_halt(&tp->t_delack_ch, softnet_lock); 1226 callout_destroy(&tp->t_delack_ch); 1227 pool_put(&tcpcb_pool, tp); 1228 1229 return NULL; 1230 } 1231 1232 int 1233 tcp_freeq(struct tcpcb *tp) 1234 { 1235 struct ipqent *qe; 1236 int rv = 0; 1237 1238 TCP_REASS_LOCK_CHECK(tp); 1239 1240 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) { 1241 TAILQ_REMOVE(&tp->segq, qe, ipqe_q); 1242 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq); 1243 m_freem(qe->ipqe_m); 1244 tcpipqent_free(qe); 1245 rv = 1; 1246 } 1247 tp->t_segqlen = 0; 1248 KASSERT(TAILQ_EMPTY(&tp->timeq)); 1249 return (rv); 1250 } 1251 1252 void 1253 tcp_fasttimo(void) 1254 { 1255 if (tcp_drainwanted) { 1256 tcp_drain(); 1257 tcp_drainwanted = 0; 1258 } 1259 } 1260 1261 void 1262 tcp_drainstub(void) 1263 { 1264 tcp_drainwanted = 1; 1265 } 1266 1267 /* 1268 * Protocol drain routine. Called when memory is in short supply. 1269 * Called from pr_fasttimo thus a callout context. 1270 */ 1271 void 1272 tcp_drain(void) 1273 { 1274 struct inpcb_hdr *inph; 1275 struct tcpcb *tp; 1276 1277 mutex_enter(softnet_lock); 1278 KERNEL_LOCK(1, NULL); 1279 1280 /* 1281 * Free the sequence queue of all TCP connections. 1282 */ 1283 TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) { 1284 switch (inph->inph_af) { 1285 case AF_INET: 1286 tp = intotcpcb((struct inpcb *)inph); 1287 break; 1288 #ifdef INET6 1289 case AF_INET6: 1290 tp = in6totcpcb((struct in6pcb *)inph); 1291 break; 1292 #endif 1293 default: 1294 tp = NULL; 1295 break; 1296 } 1297 if (tp != NULL) { 1298 /* 1299 * We may be called from a device's interrupt 1300 * context. If the tcpcb is already busy, 1301 * just bail out now. 1302 */ 1303 if (tcp_reass_lock_try(tp) == 0) 1304 continue; 1305 if (tcp_freeq(tp)) 1306 TCP_STATINC(TCP_STAT_CONNSDRAINED); 1307 TCP_REASS_UNLOCK(tp); 1308 } 1309 } 1310 1311 KERNEL_UNLOCK_ONE(NULL); 1312 mutex_exit(softnet_lock); 1313 } 1314 1315 /* 1316 * Notify a tcp user of an asynchronous error; 1317 * store error as soft error, but wake up user 1318 * (for now, won't do anything until can select for soft error). 1319 */ 1320 void 1321 tcp_notify(struct inpcb *inp, int error) 1322 { 1323 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 1324 struct socket *so = inp->inp_socket; 1325 1326 /* 1327 * Ignore some errors if we are hooked up. 1328 * If connection hasn't completed, has retransmitted several times, 1329 * and receives a second error, give up now. This is better 1330 * than waiting a long time to establish a connection that 1331 * can never complete. 1332 */ 1333 if (tp->t_state == TCPS_ESTABLISHED && 1334 (error == EHOSTUNREACH || error == ENETUNREACH || 1335 error == EHOSTDOWN)) { 1336 return; 1337 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1338 tp->t_rxtshift > 3 && tp->t_softerror) 1339 so->so_error = error; 1340 else 1341 tp->t_softerror = error; 1342 cv_broadcast(&so->so_cv); 1343 sorwakeup(so); 1344 sowwakeup(so); 1345 } 1346 1347 #ifdef INET6 1348 void 1349 tcp6_notify(struct in6pcb *in6p, int error) 1350 { 1351 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb; 1352 struct socket *so = in6p->in6p_socket; 1353 1354 /* 1355 * Ignore some errors if we are hooked up. 1356 * If connection hasn't completed, has retransmitted several times, 1357 * and receives a second error, give up now. This is better 1358 * than waiting a long time to establish a connection that 1359 * can never complete. 1360 */ 1361 if (tp->t_state == TCPS_ESTABLISHED && 1362 (error == EHOSTUNREACH || error == ENETUNREACH || 1363 error == EHOSTDOWN)) { 1364 return; 1365 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 && 1366 tp->t_rxtshift > 3 && tp->t_softerror) 1367 so->so_error = error; 1368 else 1369 tp->t_softerror = error; 1370 cv_broadcast(&so->so_cv); 1371 sorwakeup(so); 1372 sowwakeup(so); 1373 } 1374 #endif 1375 1376 #ifdef INET6 1377 void * 1378 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d) 1379 { 1380 struct tcphdr th; 1381 void (*notify)(struct in6pcb *, int) = tcp6_notify; 1382 int nmatch; 1383 struct ip6_hdr *ip6; 1384 const struct sockaddr_in6 *sa6_src = NULL; 1385 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa; 1386 struct mbuf *m; 1387 int off; 1388 1389 if (sa->sa_family != AF_INET6 || 1390 sa->sa_len != sizeof(struct sockaddr_in6)) 1391 return NULL; 1392 if ((unsigned)cmd >= PRC_NCMDS) 1393 return NULL; 1394 else if (cmd == PRC_QUENCH) { 1395 /* 1396 * Don't honor ICMP Source Quench messages meant for 1397 * TCP connections. 1398 */ 1399 return NULL; 1400 } else if (PRC_IS_REDIRECT(cmd)) 1401 notify = in6_rtchange, d = NULL; 1402 else if (cmd == PRC_MSGSIZE) 1403 ; /* special code is present, see below */ 1404 else if (cmd == PRC_HOSTDEAD) 1405 d = NULL; 1406 else if (inet6ctlerrmap[cmd] == 0) 1407 return NULL; 1408 1409 /* if the parameter is from icmp6, decode it. */ 1410 if (d != NULL) { 1411 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d; 1412 m = ip6cp->ip6c_m; 1413 ip6 = ip6cp->ip6c_ip6; 1414 off = ip6cp->ip6c_off; 1415 sa6_src = ip6cp->ip6c_src; 1416 } else { 1417 m = NULL; 1418 ip6 = NULL; 1419 sa6_src = &sa6_any; 1420 off = 0; 1421 } 1422 1423 if (ip6) { 1424 /* check if we can safely examine src and dst ports */ 1425 if (m->m_pkthdr.len < off + sizeof(th)) { 1426 if (cmd == PRC_MSGSIZE) 1427 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0); 1428 return NULL; 1429 } 1430 1431 memset(&th, 0, sizeof(th)); 1432 m_copydata(m, off, sizeof(th), (void *)&th); 1433 1434 if (cmd == PRC_MSGSIZE) { 1435 int valid = 0; 1436 1437 /* 1438 * Check to see if we have a valid TCP connection 1439 * corresponding to the address in the ICMPv6 message 1440 * payload. 1441 */ 1442 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr, 1443 th.th_dport, 1444 (const struct in6_addr *)&sa6_src->sin6_addr, 1445 th.th_sport, 0, 0)) 1446 valid++; 1447 1448 /* 1449 * Depending on the value of "valid" and routing table 1450 * size (mtudisc_{hi,lo}wat), we will: 1451 * - recalcurate the new MTU and create the 1452 * corresponding routing entry, or 1453 * - ignore the MTU change notification. 1454 */ 1455 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid); 1456 1457 /* 1458 * no need to call in6_pcbnotify, it should have been 1459 * called via callback if necessary 1460 */ 1461 return NULL; 1462 } 1463 1464 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport, 1465 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify); 1466 if (nmatch == 0 && syn_cache_count && 1467 (inet6ctlerrmap[cmd] == EHOSTUNREACH || 1468 inet6ctlerrmap[cmd] == ENETUNREACH || 1469 inet6ctlerrmap[cmd] == EHOSTDOWN)) 1470 syn_cache_unreach((const struct sockaddr *)sa6_src, 1471 sa, &th); 1472 } else { 1473 (void) in6_pcbnotify(&tcbtable, sa, 0, 1474 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); 1475 } 1476 1477 return NULL; 1478 } 1479 #endif 1480 1481 /* assumes that ip header and tcp header are contiguous on mbuf */ 1482 void * 1483 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v) 1484 { 1485 struct ip *ip = v; 1486 struct tcphdr *th; 1487 struct icmp *icp; 1488 extern const int inetctlerrmap[]; 1489 void (*notify)(struct inpcb *, int) = tcp_notify; 1490 int errno; 1491 int nmatch; 1492 struct tcpcb *tp; 1493 u_int mtu; 1494 tcp_seq seq; 1495 struct inpcb *inp; 1496 #ifdef INET6 1497 struct in6pcb *in6p; 1498 struct in6_addr src6, dst6; 1499 #endif 1500 1501 if (sa->sa_family != AF_INET || 1502 sa->sa_len != sizeof(struct sockaddr_in)) 1503 return NULL; 1504 if ((unsigned)cmd >= PRC_NCMDS) 1505 return NULL; 1506 errno = inetctlerrmap[cmd]; 1507 if (cmd == PRC_QUENCH) 1508 /* 1509 * Don't honor ICMP Source Quench messages meant for 1510 * TCP connections. 1511 */ 1512 return NULL; 1513 else if (PRC_IS_REDIRECT(cmd)) 1514 notify = in_rtchange, ip = 0; 1515 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) { 1516 /* 1517 * Check to see if we have a valid TCP connection 1518 * corresponding to the address in the ICMP message 1519 * payload. 1520 * 1521 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN. 1522 */ 1523 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1524 #ifdef INET6 1525 in6_in_2_v4mapin6(&ip->ip_src, &src6); 1526 in6_in_2_v4mapin6(&ip->ip_dst, &dst6); 1527 #endif 1528 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst, 1529 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL) 1530 #ifdef INET6 1531 in6p = NULL; 1532 #else 1533 ; 1534 #endif 1535 #ifdef INET6 1536 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6, 1537 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL) 1538 ; 1539 #endif 1540 else 1541 return NULL; 1542 1543 /* 1544 * Now that we've validated that we are actually communicating 1545 * with the host indicated in the ICMP message, locate the 1546 * ICMP header, recalculate the new MTU, and create the 1547 * corresponding routing entry. 1548 */ 1549 icp = (struct icmp *)((char *)ip - 1550 offsetof(struct icmp, icmp_ip)); 1551 if (inp) { 1552 if ((tp = intotcpcb(inp)) == NULL) 1553 return NULL; 1554 } 1555 #ifdef INET6 1556 else if (in6p) { 1557 if ((tp = in6totcpcb(in6p)) == NULL) 1558 return NULL; 1559 } 1560 #endif 1561 else 1562 return NULL; 1563 seq = ntohl(th->th_seq); 1564 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max)) 1565 return NULL; 1566 /* 1567 * If the ICMP message advertises a Next-Hop MTU 1568 * equal or larger than the maximum packet size we have 1569 * ever sent, drop the message. 1570 */ 1571 mtu = (u_int)ntohs(icp->icmp_nextmtu); 1572 if (mtu >= tp->t_pmtud_mtu_sent) 1573 return NULL; 1574 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) { 1575 /* 1576 * Calculate new MTU, and create corresponding 1577 * route (traditional PMTUD). 1578 */ 1579 tp->t_flags &= ~TF_PMTUD_PEND; 1580 icmp_mtudisc(icp, ip->ip_dst); 1581 } else { 1582 /* 1583 * Record the information got in the ICMP 1584 * message; act on it later. 1585 * If we had already recorded an ICMP message, 1586 * replace the old one only if the new message 1587 * refers to an older TCP segment 1588 */ 1589 if (tp->t_flags & TF_PMTUD_PEND) { 1590 if (SEQ_LT(tp->t_pmtud_th_seq, seq)) 1591 return NULL; 1592 } else 1593 tp->t_flags |= TF_PMTUD_PEND; 1594 tp->t_pmtud_th_seq = seq; 1595 tp->t_pmtud_nextmtu = icp->icmp_nextmtu; 1596 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len; 1597 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl; 1598 } 1599 return NULL; 1600 } else if (cmd == PRC_HOSTDEAD) 1601 ip = 0; 1602 else if (errno == 0) 1603 return NULL; 1604 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) { 1605 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2)); 1606 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr, 1607 th->th_dport, ip->ip_src, th->th_sport, errno, notify); 1608 if (nmatch == 0 && syn_cache_count && 1609 (inetctlerrmap[cmd] == EHOSTUNREACH || 1610 inetctlerrmap[cmd] == ENETUNREACH || 1611 inetctlerrmap[cmd] == EHOSTDOWN)) { 1612 struct sockaddr_in sin; 1613 memset(&sin, 0, sizeof(sin)); 1614 sin.sin_len = sizeof(sin); 1615 sin.sin_family = AF_INET; 1616 sin.sin_port = th->th_sport; 1617 sin.sin_addr = ip->ip_src; 1618 syn_cache_unreach((struct sockaddr *)&sin, sa, th); 1619 } 1620 1621 /* XXX mapped address case */ 1622 } else 1623 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno, 1624 notify); 1625 return NULL; 1626 } 1627 1628 /* 1629 * When a source quench is received, we are being notified of congestion. 1630 * Close the congestion window down to the Loss Window (one segment). 1631 * We will gradually open it again as we proceed. 1632 */ 1633 void 1634 tcp_quench(struct inpcb *inp, int errno) 1635 { 1636 struct tcpcb *tp = intotcpcb(inp); 1637 1638 if (tp) { 1639 tp->snd_cwnd = tp->t_segsz; 1640 tp->t_bytes_acked = 0; 1641 } 1642 } 1643 1644 #ifdef INET6 1645 void 1646 tcp6_quench(struct in6pcb *in6p, int errno) 1647 { 1648 struct tcpcb *tp = in6totcpcb(in6p); 1649 1650 if (tp) { 1651 tp->snd_cwnd = tp->t_segsz; 1652 tp->t_bytes_acked = 0; 1653 } 1654 } 1655 #endif 1656 1657 /* 1658 * Path MTU Discovery handlers. 1659 */ 1660 void 1661 tcp_mtudisc_callback(struct in_addr faddr) 1662 { 1663 #ifdef INET6 1664 struct in6_addr in6; 1665 #endif 1666 1667 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc); 1668 #ifdef INET6 1669 in6_in_2_v4mapin6(&faddr, &in6); 1670 tcp6_mtudisc_callback(&in6); 1671 #endif 1672 } 1673 1674 /* 1675 * On receipt of path MTU corrections, flush old route and replace it 1676 * with the new one. Retransmit all unacknowledged packets, to ensure 1677 * that all packets will be received. 1678 */ 1679 void 1680 tcp_mtudisc(struct inpcb *inp, int errno) 1681 { 1682 struct tcpcb *tp = intotcpcb(inp); 1683 struct rtentry *rt; 1684 1685 if (tp == NULL) 1686 return; 1687 1688 rt = in_pcbrtentry(inp); 1689 if (rt != NULL) { 1690 /* 1691 * If this was not a host route, remove and realloc. 1692 */ 1693 if ((rt->rt_flags & RTF_HOST) == 0) { 1694 in_pcbrtentry_unref(rt, inp); 1695 in_rtchange(inp, errno); 1696 if ((rt = in_pcbrtentry(inp)) == NULL) 1697 return; 1698 } 1699 1700 /* 1701 * Slow start out of the error condition. We 1702 * use the MTU because we know it's smaller 1703 * than the previously transmitted segment. 1704 * 1705 * Note: This is more conservative than the 1706 * suggestion in draft-floyd-incr-init-win-03. 1707 */ 1708 if (rt->rt_rmx.rmx_mtu != 0) 1709 tp->snd_cwnd = 1710 TCP_INITIAL_WINDOW(tcp_init_win, 1711 rt->rt_rmx.rmx_mtu); 1712 in_pcbrtentry_unref(rt, inp); 1713 } 1714 1715 /* 1716 * Resend unacknowledged packets. 1717 */ 1718 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1719 tcp_output(tp); 1720 } 1721 1722 #ifdef INET6 1723 /* 1724 * Path MTU Discovery handlers. 1725 */ 1726 void 1727 tcp6_mtudisc_callback(struct in6_addr *faddr) 1728 { 1729 struct sockaddr_in6 sin6; 1730 1731 memset(&sin6, 0, sizeof(sin6)); 1732 sin6.sin6_family = AF_INET6; 1733 sin6.sin6_len = sizeof(struct sockaddr_in6); 1734 sin6.sin6_addr = *faddr; 1735 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0, 1736 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc); 1737 } 1738 1739 void 1740 tcp6_mtudisc(struct in6pcb *in6p, int errno) 1741 { 1742 struct tcpcb *tp = in6totcpcb(in6p); 1743 struct rtentry *rt; 1744 1745 if (tp == NULL) 1746 return; 1747 1748 rt = in6_pcbrtentry(in6p); 1749 if (rt != NULL) { 1750 /* 1751 * If this was not a host route, remove and realloc. 1752 */ 1753 if ((rt->rt_flags & RTF_HOST) == 0) { 1754 in6_pcbrtentry_unref(rt, in6p); 1755 in6_rtchange(in6p, errno); 1756 rt = in6_pcbrtentry(in6p); 1757 if (rt == NULL) 1758 return; 1759 } 1760 1761 /* 1762 * Slow start out of the error condition. We 1763 * use the MTU because we know it's smaller 1764 * than the previously transmitted segment. 1765 * 1766 * Note: This is more conservative than the 1767 * suggestion in draft-floyd-incr-init-win-03. 1768 */ 1769 if (rt->rt_rmx.rmx_mtu != 0) { 1770 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1771 rt->rt_rmx.rmx_mtu); 1772 } 1773 in6_pcbrtentry_unref(rt, in6p); 1774 } 1775 1776 /* 1777 * Resend unacknowledged packets. 1778 */ 1779 tp->snd_nxt = tp->sack_newdata = tp->snd_una; 1780 tcp_output(tp); 1781 } 1782 #endif /* INET6 */ 1783 1784 /* 1785 * Compute the MSS to advertise to the peer. Called only during 1786 * the 3-way handshake. If we are the server (peer initiated 1787 * connection), we are called with a pointer to the interface 1788 * on which the SYN packet arrived. If we are the client (we 1789 * initiated connection), we are called with a pointer to the 1790 * interface out which this connection should go. 1791 * 1792 * NOTE: Do not subtract IP option/extension header size nor IPsec 1793 * header size from MSS advertisement. MSS option must hold the maximum 1794 * segment size we can accept, so it must always be: 1795 * max(if mtu) - ip header - tcp header 1796 */ 1797 u_long 1798 tcp_mss_to_advertise(const struct ifnet *ifp, int af) 1799 { 1800 extern u_long in_maxmtu; 1801 u_long mss = 0; 1802 u_long hdrsiz; 1803 1804 /* 1805 * In order to avoid defeating path MTU discovery on the peer, 1806 * we advertise the max MTU of all attached networks as our MSS, 1807 * per RFC 1191, section 3.1. 1808 * 1809 * We provide the option to advertise just the MTU of 1810 * the interface on which we hope this connection will 1811 * be receiving. If we are responding to a SYN, we 1812 * will have a pretty good idea about this, but when 1813 * initiating a connection there is a bit more doubt. 1814 * 1815 * We also need to ensure that loopback has a large enough 1816 * MSS, as the loopback MTU is never included in in_maxmtu. 1817 */ 1818 1819 if (ifp != NULL) 1820 switch (af) { 1821 case AF_INET: 1822 mss = ifp->if_mtu; 1823 break; 1824 #ifdef INET6 1825 case AF_INET6: 1826 mss = IN6_LINKMTU(ifp); 1827 break; 1828 #endif 1829 } 1830 1831 if (tcp_mss_ifmtu == 0) 1832 switch (af) { 1833 case AF_INET: 1834 mss = uimax(in_maxmtu, mss); 1835 break; 1836 #ifdef INET6 1837 case AF_INET6: 1838 mss = uimax(in6_maxmtu, mss); 1839 break; 1840 #endif 1841 } 1842 1843 switch (af) { 1844 case AF_INET: 1845 hdrsiz = sizeof(struct ip); 1846 break; 1847 #ifdef INET6 1848 case AF_INET6: 1849 hdrsiz = sizeof(struct ip6_hdr); 1850 break; 1851 #endif 1852 default: 1853 hdrsiz = 0; 1854 break; 1855 } 1856 hdrsiz += sizeof(struct tcphdr); 1857 if (mss > hdrsiz) 1858 mss -= hdrsiz; 1859 1860 mss = uimax(tcp_mssdflt, mss); 1861 return (mss); 1862 } 1863 1864 /* 1865 * Set connection variables based on the peer's advertised MSS. 1866 * We are passed the TCPCB for the actual connection. If we 1867 * are the server, we are called by the compressed state engine 1868 * when the 3-way handshake is complete. If we are the client, 1869 * we are called when we receive the SYN,ACK from the server. 1870 * 1871 * NOTE: Our advertised MSS value must be initialized in the TCPCB 1872 * before this routine is called! 1873 */ 1874 void 1875 tcp_mss_from_peer(struct tcpcb *tp, int offer) 1876 { 1877 struct socket *so; 1878 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1879 struct rtentry *rt; 1880 #endif 1881 u_long bufsize; 1882 int mss; 1883 1884 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1885 1886 so = NULL; 1887 rt = NULL; 1888 1889 if (tp->t_inpcb) { 1890 so = tp->t_inpcb->inp_socket; 1891 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1892 rt = in_pcbrtentry(tp->t_inpcb); 1893 #endif 1894 } 1895 1896 #ifdef INET6 1897 if (tp->t_in6pcb) { 1898 so = tp->t_in6pcb->in6p_socket; 1899 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1900 rt = in6_pcbrtentry(tp->t_in6pcb); 1901 #endif 1902 } 1903 #endif 1904 1905 /* 1906 * As per RFC1122, use the default MSS value, unless they 1907 * sent us an offer. Do not accept offers less than 256 bytes. 1908 */ 1909 mss = tcp_mssdflt; 1910 if (offer) 1911 mss = offer; 1912 mss = uimax(mss, 256); /* sanity */ 1913 tp->t_peermss = mss; 1914 mss -= tcp_optlen(tp); 1915 if (tp->t_inpcb) 1916 mss -= ip_optlen(tp->t_inpcb); 1917 #ifdef INET6 1918 if (tp->t_in6pcb) 1919 mss -= ip6_optlen(tp->t_in6pcb); 1920 #endif 1921 /* 1922 * XXX XXX What if mss goes negative or zero? This can happen if a 1923 * socket has large IPv6 options. We crash below. 1924 */ 1925 1926 /* 1927 * If there's a pipesize, change the socket buffer to that size. 1928 * Make the socket buffer an integral number of MSS units. If 1929 * the MSS is larger than the socket buffer, artificially decrease 1930 * the MSS. 1931 */ 1932 #ifdef RTV_SPIPE 1933 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0) 1934 bufsize = rt->rt_rmx.rmx_sendpipe; 1935 else 1936 #endif 1937 { 1938 KASSERT(so != NULL); 1939 bufsize = so->so_snd.sb_hiwat; 1940 } 1941 if (bufsize < mss) 1942 mss = bufsize; 1943 else { 1944 bufsize = roundup(bufsize, mss); 1945 if (bufsize > sb_max) 1946 bufsize = sb_max; 1947 (void) sbreserve(&so->so_snd, bufsize, so); 1948 } 1949 tp->t_segsz = mss; 1950 1951 #ifdef RTV_SSTHRESH 1952 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) { 1953 /* 1954 * There's some sort of gateway or interface buffer 1955 * limit on the path. Use this to set the slow 1956 * start threshold, but set the threshold to no less 1957 * than 2 * MSS. 1958 */ 1959 tp->snd_ssthresh = uimax(2 * mss, rt->rt_rmx.rmx_ssthresh); 1960 } 1961 #endif 1962 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH) 1963 if (tp->t_inpcb) 1964 in_pcbrtentry_unref(rt, tp->t_inpcb); 1965 #ifdef INET6 1966 if (tp->t_in6pcb) 1967 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 1968 #endif 1969 #endif 1970 } 1971 1972 /* 1973 * Processing necessary when a TCP connection is established. 1974 */ 1975 void 1976 tcp_established(struct tcpcb *tp) 1977 { 1978 struct socket *so; 1979 #ifdef RTV_RPIPE 1980 struct rtentry *rt; 1981 #endif 1982 u_long bufsize; 1983 1984 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 1985 1986 so = NULL; 1987 rt = NULL; 1988 1989 /* This is a while() to reduce the dreadful stairstepping below */ 1990 while (tp->t_inpcb) { 1991 so = tp->t_inpcb->inp_socket; 1992 #if defined(RTV_RPIPE) 1993 rt = in_pcbrtentry(tp->t_inpcb); 1994 #endif 1995 if (__predict_true(tcp_msl_enable)) { 1996 if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) { 1997 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2); 1998 break; 1999 } 2000 2001 if (__predict_false(tcp_rttlocal)) { 2002 /* This may be adjusted by tcp_input */ 2003 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2004 break; 2005 } 2006 if (in_localaddr(tp->t_inpcb->inp_faddr)) { 2007 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1); 2008 break; 2009 } 2010 } 2011 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL; 2012 break; 2013 } 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 #endif 2045 2046 tp->t_state = TCPS_ESTABLISHED; 2047 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 2048 2049 #ifdef RTV_RPIPE 2050 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0) 2051 bufsize = rt->rt_rmx.rmx_recvpipe; 2052 else 2053 #endif 2054 { 2055 KASSERT(so != NULL); 2056 bufsize = so->so_rcv.sb_hiwat; 2057 } 2058 if (bufsize > tp->t_ourmss) { 2059 bufsize = roundup(bufsize, tp->t_ourmss); 2060 if (bufsize > sb_max) 2061 bufsize = sb_max; 2062 (void) sbreserve(&so->so_rcv, bufsize, so); 2063 } 2064 #ifdef RTV_RPIPE 2065 if (tp->t_inpcb) 2066 in_pcbrtentry_unref(rt, tp->t_inpcb); 2067 #ifdef INET6 2068 if (tp->t_in6pcb) 2069 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2070 #endif 2071 #endif 2072 } 2073 2074 /* 2075 * Check if there's an initial rtt or rttvar. Convert from the 2076 * route-table units to scaled multiples of the slow timeout timer. 2077 * Called only during the 3-way handshake. 2078 */ 2079 void 2080 tcp_rmx_rtt(struct tcpcb *tp) 2081 { 2082 #ifdef RTV_RTT 2083 struct rtentry *rt = NULL; 2084 int rtt; 2085 2086 KASSERT(!(tp->t_inpcb && tp->t_in6pcb)); 2087 2088 if (tp->t_inpcb) 2089 rt = in_pcbrtentry(tp->t_inpcb); 2090 #ifdef INET6 2091 if (tp->t_in6pcb) 2092 rt = in6_pcbrtentry(tp->t_in6pcb); 2093 #endif 2094 if (rt == NULL) 2095 return; 2096 2097 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2098 /* 2099 * XXX The lock bit for MTU indicates that the value 2100 * is also a minimum value; this is subject to time. 2101 */ 2102 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2103 TCPT_RANGESET(tp->t_rttmin, 2104 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2105 TCPTV_MIN, TCPTV_REXMTMAX); 2106 tp->t_srtt = rtt / 2107 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2)); 2108 if (rt->rt_rmx.rmx_rttvar) { 2109 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2110 ((RTM_RTTUNIT / PR_SLOWHZ) >> 2111 (TCP_RTTVAR_SHIFT + 2)); 2112 } else { 2113 /* Default variation is +- 1 rtt */ 2114 tp->t_rttvar = 2115 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT); 2116 } 2117 TCPT_RANGESET(tp->t_rxtcur, 2118 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2), 2119 tp->t_rttmin, TCPTV_REXMTMAX); 2120 } 2121 if (tp->t_inpcb) 2122 in_pcbrtentry_unref(rt, tp->t_inpcb); 2123 #ifdef INET6 2124 if (tp->t_in6pcb) 2125 in6_pcbrtentry_unref(rt, tp->t_in6pcb); 2126 #endif 2127 #endif 2128 } 2129 2130 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */ 2131 2132 /* 2133 * Get a new sequence value given a tcp control block 2134 */ 2135 tcp_seq 2136 tcp_new_iss(struct tcpcb *tp, tcp_seq addin) 2137 { 2138 2139 if (tp->t_inpcb != NULL) { 2140 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr, 2141 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport, 2142 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr), 2143 addin)); 2144 } 2145 #ifdef INET6 2146 if (tp->t_in6pcb != NULL) { 2147 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr, 2148 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport, 2149 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr), 2150 addin)); 2151 } 2152 #endif 2153 2154 panic("tcp_new_iss: unreachable"); 2155 } 2156 2157 static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */ 2158 2159 /* 2160 * Initialize RFC 1948 ISS Secret 2161 */ 2162 static int 2163 tcp_iss_secret_init(void) 2164 { 2165 cprng_strong(kern_cprng, 2166 tcp_iss_secret, sizeof(tcp_iss_secret), 0); 2167 2168 return 0; 2169 } 2170 2171 /* 2172 * This routine actually generates a new TCP initial sequence number. 2173 */ 2174 tcp_seq 2175 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport, 2176 size_t addrsz, tcp_seq addin) 2177 { 2178 tcp_seq tcp_iss; 2179 2180 if (tcp_do_rfc1948) { 2181 MD5_CTX ctx; 2182 u_int8_t hash[16]; /* XXX MD5 knowledge */ 2183 static ONCE_DECL(tcp_iss_secret_control); 2184 2185 /* 2186 * If we haven't been here before, initialize our cryptographic 2187 * hash secret. 2188 */ 2189 RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init); 2190 2191 /* 2192 * Compute the base value of the ISS. It is a hash 2193 * of (saddr, sport, daddr, dport, secret). 2194 */ 2195 MD5Init(&ctx); 2196 2197 MD5Update(&ctx, (u_char *) laddr, addrsz); 2198 MD5Update(&ctx, (u_char *) &lport, sizeof(lport)); 2199 2200 MD5Update(&ctx, (u_char *) faddr, addrsz); 2201 MD5Update(&ctx, (u_char *) &fport, sizeof(fport)); 2202 2203 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret)); 2204 2205 MD5Final(hash, &ctx); 2206 2207 memcpy(&tcp_iss, hash, sizeof(tcp_iss)); 2208 2209 /* 2210 * Now increment our "timer", and add it in to 2211 * the computed value. 2212 * 2213 * XXX Use `addin'? 2214 * XXX TCP_ISSINCR too large to use? 2215 */ 2216 tcp_iss_seq += TCP_ISSINCR; 2217 #ifdef TCPISS_DEBUG 2218 printf("ISS hash 0x%08x, ", tcp_iss); 2219 #endif 2220 tcp_iss += tcp_iss_seq + addin; 2221 #ifdef TCPISS_DEBUG 2222 printf("new ISS 0x%08x\n", tcp_iss); 2223 #endif 2224 } else { 2225 /* 2226 * Randomize. 2227 */ 2228 tcp_iss = cprng_fast32(); 2229 2230 /* 2231 * If we were asked to add some amount to a known value, 2232 * we will take a random value obtained above, mask off 2233 * the upper bits, and add in the known value. We also 2234 * add in a constant to ensure that we are at least a 2235 * certain distance from the original value. 2236 * 2237 * This is used when an old connection is in timed wait 2238 * and we have a new one coming in, for instance. 2239 */ 2240 if (addin != 0) { 2241 #ifdef TCPISS_DEBUG 2242 printf("Random %08x, ", tcp_iss); 2243 #endif 2244 tcp_iss &= TCP_ISS_RANDOM_MASK; 2245 tcp_iss += addin + TCP_ISSINCR; 2246 #ifdef TCPISS_DEBUG 2247 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss); 2248 #endif 2249 } else { 2250 tcp_iss &= TCP_ISS_RANDOM_MASK; 2251 tcp_iss += tcp_iss_seq; 2252 tcp_iss_seq += TCP_ISSINCR; 2253 #ifdef TCPISS_DEBUG 2254 printf("ISS %08x\n", tcp_iss); 2255 #endif 2256 } 2257 } 2258 2259 return (tcp_iss); 2260 } 2261 2262 #if defined(IPSEC) 2263 /* compute ESP/AH header size for TCP, including outer IP header. */ 2264 size_t 2265 ipsec4_hdrsiz_tcp(struct tcpcb *tp) 2266 { 2267 struct inpcb *inp; 2268 size_t hdrsiz; 2269 2270 /* XXX mapped addr case (tp->t_in6pcb) */ 2271 if (!tp || !tp->t_template || !(inp = tp->t_inpcb)) 2272 return 0; 2273 switch (tp->t_family) { 2274 case AF_INET: 2275 /* XXX: should use correct direction. */ 2276 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp); 2277 break; 2278 default: 2279 hdrsiz = 0; 2280 break; 2281 } 2282 2283 return hdrsiz; 2284 } 2285 2286 #ifdef INET6 2287 size_t 2288 ipsec6_hdrsiz_tcp(struct tcpcb *tp) 2289 { 2290 struct in6pcb *in6p; 2291 size_t hdrsiz; 2292 2293 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb)) 2294 return 0; 2295 switch (tp->t_family) { 2296 case AF_INET6: 2297 /* XXX: should use correct direction. */ 2298 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p); 2299 break; 2300 case AF_INET: 2301 /* mapped address case - tricky */ 2302 default: 2303 hdrsiz = 0; 2304 break; 2305 } 2306 2307 return hdrsiz; 2308 } 2309 #endif 2310 #endif /*IPSEC*/ 2311 2312 /* 2313 * Determine the length of the TCP options for this connection. 2314 * 2315 * XXX: What do we do for SACK, when we add that? Just reserve 2316 * all of the space? Otherwise we can't exactly be incrementing 2317 * cwnd by an amount that varies depending on the amount we last 2318 * had to SACK! 2319 */ 2320 2321 u_int 2322 tcp_optlen(struct tcpcb *tp) 2323 { 2324 u_int optlen; 2325 2326 optlen = 0; 2327 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 2328 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 2329 optlen += TCPOLEN_TSTAMP_APPA; 2330 2331 #ifdef TCP_SIGNATURE 2332 if (tp->t_flags & TF_SIGNATURE) 2333 optlen += TCPOLEN_SIGLEN; 2334 #endif 2335 2336 return optlen; 2337 } 2338 2339 u_int 2340 tcp_hdrsz(struct tcpcb *tp) 2341 { 2342 u_int hlen; 2343 2344 switch (tp->t_family) { 2345 #ifdef INET6 2346 case AF_INET6: 2347 hlen = sizeof(struct ip6_hdr); 2348 break; 2349 #endif 2350 case AF_INET: 2351 hlen = sizeof(struct ip); 2352 break; 2353 default: 2354 hlen = 0; 2355 break; 2356 } 2357 hlen += sizeof(struct tcphdr); 2358 2359 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2360 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2361 hlen += TCPOLEN_TSTAMP_APPA; 2362 #ifdef TCP_SIGNATURE 2363 if (tp->t_flags & TF_SIGNATURE) 2364 hlen += TCPOLEN_SIGLEN; 2365 #endif 2366 return hlen; 2367 } 2368 2369 void 2370 tcp_statinc(u_int stat) 2371 { 2372 2373 KASSERT(stat < TCP_NSTATS); 2374 TCP_STATINC(stat); 2375 } 2376 2377 void 2378 tcp_statadd(u_int stat, uint64_t val) 2379 { 2380 2381 KASSERT(stat < TCP_NSTATS); 2382 TCP_STATADD(stat, val); 2383 } 2384