1 /* $NetBSD: tcp_input.c,v 1.180 2003/08/22 22:49:34 itojun 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 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001 The NetBSD Foundation, Inc. 74 * All rights reserved. 75 * 76 * This code is derived from software contributed to The NetBSD Foundation 77 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 78 * Facility, NASA Ames Research Center. 79 * 80 * Redistribution and use in source and binary forms, with or without 81 * modification, are permitted provided that the following conditions 82 * are met: 83 * 1. Redistributions of source code must retain the above copyright 84 * notice, this list of conditions and the following disclaimer. 85 * 2. Redistributions in binary form must reproduce the above copyright 86 * notice, this list of conditions and the following disclaimer in the 87 * documentation and/or other materials provided with the distribution. 88 * 3. All advertising materials mentioning features or use of this software 89 * must display the following acknowledgement: 90 * This product includes software developed by the NetBSD 91 * Foundation, Inc. and its contributors. 92 * 4. Neither the name of The NetBSD Foundation nor the names of its 93 * contributors may be used to endorse or promote products derived 94 * from this software without specific prior written permission. 95 * 96 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 97 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 98 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 99 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 100 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 101 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 102 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 103 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 104 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 105 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 106 * POSSIBILITY OF SUCH DAMAGE. 107 */ 108 109 /* 110 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 111 * The Regents of the University of California. All rights reserved. 112 * 113 * Redistribution and use in source and binary forms, with or without 114 * modification, are permitted provided that the following conditions 115 * are met: 116 * 1. Redistributions of source code must retain the above copyright 117 * notice, this list of conditions and the following disclaimer. 118 * 2. Redistributions in binary form must reproduce the above copyright 119 * notice, this list of conditions and the following disclaimer in the 120 * documentation and/or other materials provided with the distribution. 121 * 3. Neither the name of the University nor the names of its contributors 122 * may be used to endorse or promote products derived from this software 123 * without specific prior written permission. 124 * 125 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 126 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 127 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 128 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 129 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 130 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 131 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 132 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 133 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 134 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 135 * SUCH DAMAGE. 136 * 137 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 138 */ 139 140 /* 141 * TODO list for SYN cache stuff: 142 * 143 * Find room for a "state" field, which is needed to keep a 144 * compressed state for TIME_WAIT TCBs. It's been noted already 145 * that this is fairly important for very high-volume web and 146 * mail servers, which use a large number of short-lived 147 * connections. 148 */ 149 150 #include <sys/cdefs.h> 151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.180 2003/08/22 22:49:34 itojun Exp $"); 152 153 #include "opt_inet.h" 154 #include "opt_ipsec.h" 155 #include "opt_inet_csum.h" 156 #include "opt_tcp_debug.h" 157 158 #include <sys/param.h> 159 #include <sys/systm.h> 160 #include <sys/malloc.h> 161 #include <sys/mbuf.h> 162 #include <sys/protosw.h> 163 #include <sys/socket.h> 164 #include <sys/socketvar.h> 165 #include <sys/errno.h> 166 #include <sys/syslog.h> 167 #include <sys/pool.h> 168 #include <sys/domain.h> 169 #include <sys/kernel.h> 170 171 #include <net/if.h> 172 #include <net/route.h> 173 #include <net/if_types.h> 174 175 #include <netinet/in.h> 176 #include <netinet/in_systm.h> 177 #include <netinet/ip.h> 178 #include <netinet/in_pcb.h> 179 #include <netinet/in_var.h> 180 #include <netinet/ip_var.h> 181 182 #ifdef INET6 183 #ifndef INET 184 #include <netinet/in.h> 185 #endif 186 #include <netinet/ip6.h> 187 #include <netinet6/ip6_var.h> 188 #include <netinet6/in6_pcb.h> 189 #include <netinet6/ip6_var.h> 190 #include <netinet6/in6_var.h> 191 #include <netinet/icmp6.h> 192 #include <netinet6/nd6.h> 193 #endif 194 195 #ifndef INET6 196 /* always need ip6.h for IP6_EXTHDR_GET */ 197 #include <netinet/ip6.h> 198 #endif 199 200 #include <netinet/tcp.h> 201 #include <netinet/tcp_fsm.h> 202 #include <netinet/tcp_seq.h> 203 #include <netinet/tcp_timer.h> 204 #include <netinet/tcp_var.h> 205 #include <netinet/tcpip.h> 206 #include <netinet/tcp_debug.h> 207 208 #include <machine/stdarg.h> 209 210 #ifdef IPSEC 211 #include <netinet6/ipsec.h> 212 #include <netkey/key.h> 213 #endif /*IPSEC*/ 214 #ifdef INET6 215 #include "faith.h" 216 #if defined(NFAITH) && NFAITH > 0 217 #include <net/if_faith.h> 218 #endif 219 #endif /* IPSEC */ 220 221 #ifdef FAST_IPSEC 222 #include <netipsec/ipsec.h> 223 #include <netipsec/key.h> 224 #endif /* FAST_IPSEC*/ 225 226 227 int tcprexmtthresh = 3; 228 int tcp_log_refused; 229 230 static int tcp_rst_ppslim_count = 0; 231 static struct timeval tcp_rst_ppslim_last; 232 233 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 234 235 /* for modulo comparisons of timestamps */ 236 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 237 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 238 239 /* 240 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 241 */ 242 #ifdef INET6 243 #define ND6_HINT(tp) \ 244 do { \ 245 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 246 && tp->t_in6pcb->in6p_route.ro_rt) { \ 247 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 248 } \ 249 } while (/*CONSTCOND*/ 0) 250 #else 251 #define ND6_HINT(tp) 252 #endif 253 254 /* 255 * Macro to compute ACK transmission behavior. Delay the ACK unless 256 * we have already delayed an ACK (must send an ACK every two segments). 257 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 258 * option is enabled. 259 */ 260 #define TCP_SETUP_ACK(tp, th) \ 261 do { \ 262 if ((tp)->t_flags & TF_DELACK || \ 263 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 264 tp->t_flags |= TF_ACKNOW; \ 265 else \ 266 TCP_SET_DELACK(tp); \ 267 } while (/*CONSTCOND*/ 0) 268 269 /* 270 * Convert TCP protocol fields to host order for easier processing. 271 */ 272 #define TCP_FIELDS_TO_HOST(th) \ 273 do { \ 274 NTOHL((th)->th_seq); \ 275 NTOHL((th)->th_ack); \ 276 NTOHS((th)->th_win); \ 277 NTOHS((th)->th_urp); \ 278 } while (/*CONSTCOND*/ 0) 279 280 /* 281 * ... and reverse the above. 282 */ 283 #define TCP_FIELDS_TO_NET(th) \ 284 do { \ 285 HTONL((th)->th_seq); \ 286 HTONL((th)->th_ack); \ 287 HTONS((th)->th_win); \ 288 HTONS((th)->th_urp); \ 289 } while (/*CONSTCOND*/ 0) 290 291 #ifdef TCP_CSUM_COUNTERS 292 #include <sys/device.h> 293 294 extern struct evcnt tcp_hwcsum_ok; 295 extern struct evcnt tcp_hwcsum_bad; 296 extern struct evcnt tcp_hwcsum_data; 297 extern struct evcnt tcp_swcsum; 298 299 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 300 301 #else 302 303 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 304 305 #endif /* TCP_CSUM_COUNTERS */ 306 307 #ifdef TCP_REASS_COUNTERS 308 #include <sys/device.h> 309 310 extern struct evcnt tcp_reass_; 311 extern struct evcnt tcp_reass_empty; 312 extern struct evcnt tcp_reass_iteration[8]; 313 extern struct evcnt tcp_reass_prependfirst; 314 extern struct evcnt tcp_reass_prepend; 315 extern struct evcnt tcp_reass_insert; 316 extern struct evcnt tcp_reass_inserttail; 317 extern struct evcnt tcp_reass_append; 318 extern struct evcnt tcp_reass_appendtail; 319 extern struct evcnt tcp_reass_overlaptail; 320 extern struct evcnt tcp_reass_overlapfront; 321 extern struct evcnt tcp_reass_segdup; 322 extern struct evcnt tcp_reass_fragdup; 323 324 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 325 326 #else 327 328 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 329 330 #endif /* TCP_REASS_COUNTERS */ 331 332 #ifdef INET 333 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 334 #endif 335 #ifdef INET6 336 static void tcp6_log_refused 337 __P((const struct ip6_hdr *, const struct tcphdr *)); 338 #endif 339 340 int 341 tcp_reass(tp, th, m, tlen) 342 struct tcpcb *tp; 343 struct tcphdr *th; 344 struct mbuf *m; 345 int *tlen; 346 { 347 struct ipqent *p, *q, *nq, *tiqe = NULL; 348 struct socket *so = NULL; 349 int pkt_flags; 350 tcp_seq pkt_seq; 351 unsigned pkt_len; 352 u_long rcvpartdupbyte = 0; 353 u_long rcvoobyte; 354 #ifdef TCP_REASS_COUNTERS 355 u_int count = 0; 356 #endif 357 358 if (tp->t_inpcb) 359 so = tp->t_inpcb->inp_socket; 360 #ifdef INET6 361 else if (tp->t_in6pcb) 362 so = tp->t_in6pcb->in6p_socket; 363 #endif 364 365 TCP_REASS_LOCK_CHECK(tp); 366 367 /* 368 * Call with th==0 after become established to 369 * force pre-ESTABLISHED data up to user socket. 370 */ 371 if (th == 0) 372 goto present; 373 374 rcvoobyte = *tlen; 375 /* 376 * Copy these to local variables because the tcpiphdr 377 * gets munged while we are collapsing mbufs. 378 */ 379 pkt_seq = th->th_seq; 380 pkt_len = *tlen; 381 pkt_flags = th->th_flags; 382 383 TCP_REASS_COUNTER_INCR(&tcp_reass_); 384 385 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 386 /* 387 * When we miss a packet, the vast majority of time we get 388 * packets that follow it in order. So optimize for that. 389 */ 390 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 391 p->ipqe_len += pkt_len; 392 p->ipqe_flags |= pkt_flags; 393 m_cat(p->ipqe_m, m); 394 tiqe = p; 395 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 396 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 397 goto skip_replacement; 398 } 399 /* 400 * While we're here, if the pkt is completely beyond 401 * anything we have, just insert it at the tail. 402 */ 403 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 404 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 405 goto insert_it; 406 } 407 } 408 409 q = TAILQ_FIRST(&tp->segq); 410 411 if (q != NULL) { 412 /* 413 * If this segment immediately precedes the first out-of-order 414 * block, simply slap the segment in front of it and (mostly) 415 * skip the complicated logic. 416 */ 417 if (pkt_seq + pkt_len == q->ipqe_seq) { 418 q->ipqe_seq = pkt_seq; 419 q->ipqe_len += pkt_len; 420 q->ipqe_flags |= pkt_flags; 421 m_cat(m, q->ipqe_m); 422 q->ipqe_m = m; 423 tiqe = q; 424 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 425 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 426 goto skip_replacement; 427 } 428 } else { 429 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 430 } 431 432 /* 433 * Find a segment which begins after this one does. 434 */ 435 for (p = NULL; q != NULL; q = nq) { 436 nq = TAILQ_NEXT(q, ipqe_q); 437 #ifdef TCP_REASS_COUNTERS 438 count++; 439 #endif 440 /* 441 * If the received segment is just right after this 442 * fragment, merge the two together and then check 443 * for further overlaps. 444 */ 445 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 446 #ifdef TCPREASS_DEBUG 447 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 448 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 449 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 450 #endif 451 pkt_len += q->ipqe_len; 452 pkt_flags |= q->ipqe_flags; 453 pkt_seq = q->ipqe_seq; 454 m_cat(q->ipqe_m, m); 455 m = q->ipqe_m; 456 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 457 goto free_ipqe; 458 } 459 /* 460 * If the received segment is completely past this 461 * fragment, we need to go the next fragment. 462 */ 463 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 464 p = q; 465 continue; 466 } 467 /* 468 * If the fragment is past the received segment, 469 * it (or any following) can't be concatenated. 470 */ 471 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 472 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 473 break; 474 } 475 476 /* 477 * We've received all the data in this segment before. 478 * mark it as a duplicate and return. 479 */ 480 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 481 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 482 tcpstat.tcps_rcvduppack++; 483 tcpstat.tcps_rcvdupbyte += pkt_len; 484 m_freem(m); 485 if (tiqe != NULL) 486 pool_put(&ipqent_pool, tiqe); 487 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 488 return (0); 489 } 490 /* 491 * Received segment completely overlaps this fragment 492 * so we drop the fragment (this keeps the temporal 493 * ordering of segments correct). 494 */ 495 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 496 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 497 rcvpartdupbyte += q->ipqe_len; 498 m_freem(q->ipqe_m); 499 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 500 goto free_ipqe; 501 } 502 /* 503 * RX'ed segment extends past the end of the 504 * fragment. Drop the overlapping bytes. Then 505 * merge the fragment and segment then treat as 506 * a longer received packet. 507 */ 508 if (SEQ_LT(q->ipqe_seq, pkt_seq) 509 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 510 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 511 #ifdef TCPREASS_DEBUG 512 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 513 tp, overlap, 514 pkt_seq, pkt_seq + pkt_len, pkt_len); 515 #endif 516 m_adj(m, overlap); 517 rcvpartdupbyte += overlap; 518 m_cat(q->ipqe_m, m); 519 m = q->ipqe_m; 520 pkt_seq = q->ipqe_seq; 521 pkt_len += q->ipqe_len - overlap; 522 rcvoobyte -= overlap; 523 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 524 goto free_ipqe; 525 } 526 /* 527 * RX'ed segment extends past the front of the 528 * fragment. Drop the overlapping bytes on the 529 * received packet. The packet will then be 530 * contatentated with this fragment a bit later. 531 */ 532 if (SEQ_GT(q->ipqe_seq, pkt_seq) 533 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 534 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 535 #ifdef TCPREASS_DEBUG 536 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 537 tp, overlap, 538 pkt_seq, pkt_seq + pkt_len, pkt_len); 539 #endif 540 m_adj(m, -overlap); 541 pkt_len -= overlap; 542 rcvpartdupbyte += overlap; 543 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 544 rcvoobyte -= overlap; 545 } 546 /* 547 * If the received segment immediates precedes this 548 * fragment then tack the fragment onto this segment 549 * and reinsert the data. 550 */ 551 if (q->ipqe_seq == pkt_seq + pkt_len) { 552 #ifdef TCPREASS_DEBUG 553 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 554 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 555 pkt_seq, pkt_seq + pkt_len, pkt_len); 556 #endif 557 pkt_len += q->ipqe_len; 558 pkt_flags |= q->ipqe_flags; 559 m_cat(m, q->ipqe_m); 560 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 561 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 562 if (tiqe == NULL) { 563 tiqe = q; 564 } else { 565 pool_put(&ipqent_pool, q); 566 } 567 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 568 break; 569 } 570 /* 571 * If the fragment is before the segment, remember it. 572 * When this loop is terminated, p will contain the 573 * pointer to fragment that is right before the received 574 * segment. 575 */ 576 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 577 p = q; 578 579 continue; 580 581 /* 582 * This is a common operation. It also will allow 583 * to save doing a malloc/free in most instances. 584 */ 585 free_ipqe: 586 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 587 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 588 if (tiqe == NULL) { 589 tiqe = q; 590 } else { 591 pool_put(&ipqent_pool, q); 592 } 593 } 594 595 #ifdef TCP_REASS_COUNTERS 596 if (count > 7) 597 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 598 else if (count > 0) 599 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 600 #endif 601 602 insert_it: 603 604 /* 605 * Allocate a new queue entry since the received segment did not 606 * collapse onto any other out-of-order block; thus we are allocating 607 * a new block. If it had collapsed, tiqe would not be NULL and 608 * we would be reusing it. 609 * XXX If we can't, just drop the packet. XXX 610 */ 611 if (tiqe == NULL) { 612 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 613 if (tiqe == NULL) { 614 tcpstat.tcps_rcvmemdrop++; 615 m_freem(m); 616 return (0); 617 } 618 } 619 620 /* 621 * Update the counters. 622 */ 623 tcpstat.tcps_rcvoopack++; 624 tcpstat.tcps_rcvoobyte += rcvoobyte; 625 if (rcvpartdupbyte) { 626 tcpstat.tcps_rcvpartduppack++; 627 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 628 } 629 630 /* 631 * Insert the new fragment queue entry into both queues. 632 */ 633 tiqe->ipqe_m = m; 634 tiqe->ipqe_seq = pkt_seq; 635 tiqe->ipqe_len = pkt_len; 636 tiqe->ipqe_flags = pkt_flags; 637 if (p == NULL) { 638 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 639 #ifdef TCPREASS_DEBUG 640 if (tiqe->ipqe_seq != tp->rcv_nxt) 641 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 642 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 643 #endif 644 } else { 645 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 646 #ifdef TCPREASS_DEBUG 647 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 648 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 649 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 650 #endif 651 } 652 653 skip_replacement: 654 655 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 656 657 present: 658 /* 659 * Present data to user, advancing rcv_nxt through 660 * completed sequence space. 661 */ 662 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 663 return (0); 664 q = TAILQ_FIRST(&tp->segq); 665 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 666 return (0); 667 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 668 return (0); 669 670 tp->rcv_nxt += q->ipqe_len; 671 pkt_flags = q->ipqe_flags & TH_FIN; 672 ND6_HINT(tp); 673 674 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 675 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 676 if (so->so_state & SS_CANTRCVMORE) 677 m_freem(q->ipqe_m); 678 else 679 sbappendstream(&so->so_rcv, q->ipqe_m); 680 pool_put(&ipqent_pool, q); 681 sorwakeup(so); 682 return (pkt_flags); 683 } 684 685 #ifdef INET6 686 int 687 tcp6_input(mp, offp, proto) 688 struct mbuf **mp; 689 int *offp, proto; 690 { 691 struct mbuf *m = *mp; 692 693 /* 694 * draft-itojun-ipv6-tcp-to-anycast 695 * better place to put this in? 696 */ 697 if (m->m_flags & M_ANYCAST6) { 698 struct ip6_hdr *ip6; 699 if (m->m_len < sizeof(struct ip6_hdr)) { 700 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 701 tcpstat.tcps_rcvshort++; 702 return IPPROTO_DONE; 703 } 704 } 705 ip6 = mtod(m, struct ip6_hdr *); 706 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 707 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 708 return IPPROTO_DONE; 709 } 710 711 tcp_input(m, *offp, proto); 712 return IPPROTO_DONE; 713 } 714 #endif 715 716 #ifdef INET 717 static void 718 tcp4_log_refused(ip, th) 719 const struct ip *ip; 720 const struct tcphdr *th; 721 { 722 char src[4*sizeof "123"]; 723 char dst[4*sizeof "123"]; 724 725 if (ip) { 726 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 727 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 728 } 729 else { 730 strlcpy(src, "(unknown)", sizeof(src)); 731 strlcpy(dst, "(unknown)", sizeof(dst)); 732 } 733 log(LOG_INFO, 734 "Connection attempt to TCP %s:%d from %s:%d\n", 735 dst, ntohs(th->th_dport), 736 src, ntohs(th->th_sport)); 737 } 738 #endif 739 740 #ifdef INET6 741 static void 742 tcp6_log_refused(ip6, th) 743 const struct ip6_hdr *ip6; 744 const struct tcphdr *th; 745 { 746 char src[INET6_ADDRSTRLEN]; 747 char dst[INET6_ADDRSTRLEN]; 748 749 if (ip6) { 750 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 751 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 752 } 753 else { 754 strlcpy(src, "(unknown v6)", sizeof(src)); 755 strlcpy(dst, "(unknown v6)", sizeof(dst)); 756 } 757 log(LOG_INFO, 758 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 759 dst, ntohs(th->th_dport), 760 src, ntohs(th->th_sport)); 761 } 762 #endif 763 764 /* 765 * TCP input routine, follows pages 65-76 of the 766 * protocol specification dated September, 1981 very closely. 767 */ 768 void 769 #if __STDC__ 770 tcp_input(struct mbuf *m, ...) 771 #else 772 tcp_input(m, va_alist) 773 struct mbuf *m; 774 #endif 775 { 776 struct tcphdr *th; 777 struct ip *ip; 778 struct inpcb *inp; 779 #ifdef INET6 780 struct ip6_hdr *ip6; 781 struct in6pcb *in6p; 782 #endif 783 u_int8_t *optp = NULL; 784 int optlen = 0; 785 int len, tlen, toff, hdroptlen = 0; 786 struct tcpcb *tp = 0; 787 int tiflags; 788 struct socket *so = NULL; 789 int todrop, acked, ourfinisacked, needoutput = 0; 790 #ifdef TCP_DEBUG 791 short ostate = 0; 792 #endif 793 int iss = 0; 794 u_long tiwin; 795 struct tcp_opt_info opti; 796 int off, iphlen; 797 va_list ap; 798 int af; /* af on the wire */ 799 struct mbuf *tcp_saveti = NULL; 800 801 MCLAIM(m, &tcp_rx_mowner); 802 va_start(ap, m); 803 toff = va_arg(ap, int); 804 (void)va_arg(ap, int); /* ignore value, advance ap */ 805 va_end(ap); 806 807 tcpstat.tcps_rcvtotal++; 808 809 bzero(&opti, sizeof(opti)); 810 opti.ts_present = 0; 811 opti.maxseg = 0; 812 813 /* 814 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 815 * 816 * TCP is, by definition, unicast, so we reject all 817 * multicast outright. 818 * 819 * Note, there are additional src/dst address checks in 820 * the AF-specific code below. 821 */ 822 if (m->m_flags & (M_BCAST|M_MCAST)) { 823 /* XXX stat */ 824 goto drop; 825 } 826 #ifdef INET6 827 if (m->m_flags & M_ANYCAST6) { 828 /* XXX stat */ 829 goto drop; 830 } 831 #endif 832 833 /* 834 * Get IP and TCP header together in first mbuf. 835 * Note: IP leaves IP header in first mbuf. 836 */ 837 ip = mtod(m, struct ip *); 838 #ifdef INET6 839 ip6 = NULL; 840 #endif 841 switch (ip->ip_v) { 842 #ifdef INET 843 case 4: 844 af = AF_INET; 845 iphlen = sizeof(struct ip); 846 ip = mtod(m, struct ip *); 847 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 848 sizeof(struct tcphdr)); 849 if (th == NULL) { 850 tcpstat.tcps_rcvshort++; 851 return; 852 } 853 /* We do the checksum after PCB lookup... */ 854 len = ntohs(ip->ip_len); 855 tlen = len - toff; 856 break; 857 #endif 858 #ifdef INET6 859 case 6: 860 ip = NULL; 861 iphlen = sizeof(struct ip6_hdr); 862 af = AF_INET6; 863 ip6 = mtod(m, struct ip6_hdr *); 864 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 865 sizeof(struct tcphdr)); 866 if (th == NULL) { 867 tcpstat.tcps_rcvshort++; 868 return; 869 } 870 871 /* Be proactive about malicious use of IPv4 mapped address */ 872 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 873 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 874 /* XXX stat */ 875 goto drop; 876 } 877 878 /* 879 * Be proactive about unspecified IPv6 address in source. 880 * As we use all-zero to indicate unbounded/unconnected pcb, 881 * unspecified IPv6 address can be used to confuse us. 882 * 883 * Note that packets with unspecified IPv6 destination is 884 * already dropped in ip6_input. 885 */ 886 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 887 /* XXX stat */ 888 goto drop; 889 } 890 891 /* 892 * Make sure destination address is not multicast. 893 * Source address checked in ip6_input(). 894 */ 895 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 896 /* XXX stat */ 897 goto drop; 898 } 899 900 /* We do the checksum after PCB lookup... */ 901 len = m->m_pkthdr.len; 902 tlen = len - toff; 903 break; 904 #endif 905 default: 906 m_freem(m); 907 return; 908 } 909 910 KASSERT(TCP_HDR_ALIGNED_P(th)); 911 912 /* 913 * Check that TCP offset makes sense, 914 * pull out TCP options and adjust length. XXX 915 */ 916 off = th->th_off << 2; 917 if (off < sizeof (struct tcphdr) || off > tlen) { 918 tcpstat.tcps_rcvbadoff++; 919 goto drop; 920 } 921 tlen -= off; 922 923 /* 924 * tcp_input() has been modified to use tlen to mean the TCP data 925 * length throughout the function. Other functions can use 926 * m->m_pkthdr.len as the basis for calculating the TCP data length. 927 * rja 928 */ 929 930 if (off > sizeof (struct tcphdr)) { 931 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 932 if (th == NULL) { 933 tcpstat.tcps_rcvshort++; 934 return; 935 } 936 /* 937 * NOTE: ip/ip6 will not be affected by m_pulldown() 938 * (as they're before toff) and we don't need to update those. 939 */ 940 KASSERT(TCP_HDR_ALIGNED_P(th)); 941 optlen = off - sizeof (struct tcphdr); 942 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 943 /* 944 * Do quick retrieval of timestamp options ("options 945 * prediction?"). If timestamp is the only option and it's 946 * formatted as recommended in RFC 1323 appendix A, we 947 * quickly get the values now and not bother calling 948 * tcp_dooptions(), etc. 949 */ 950 if ((optlen == TCPOLEN_TSTAMP_APPA || 951 (optlen > TCPOLEN_TSTAMP_APPA && 952 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 953 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 954 (th->th_flags & TH_SYN) == 0) { 955 opti.ts_present = 1; 956 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 957 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 958 optp = NULL; /* we've parsed the options */ 959 } 960 } 961 tiflags = th->th_flags; 962 963 /* 964 * Locate pcb for segment. 965 */ 966 findpcb: 967 inp = NULL; 968 #ifdef INET6 969 in6p = NULL; 970 #endif 971 switch (af) { 972 #ifdef INET 973 case AF_INET: 974 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 975 ip->ip_dst, th->th_dport); 976 if (inp == 0) { 977 ++tcpstat.tcps_pcbhashmiss; 978 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 979 } 980 #ifdef INET6 981 if (inp == 0) { 982 struct in6_addr s, d; 983 984 /* mapped addr case */ 985 bzero(&s, sizeof(s)); 986 s.s6_addr16[5] = htons(0xffff); 987 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 988 bzero(&d, sizeof(d)); 989 d.s6_addr16[5] = htons(0xffff); 990 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 991 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 992 &d, th->th_dport, 0); 993 if (in6p == 0) { 994 ++tcpstat.tcps_pcbhashmiss; 995 in6p = in6_pcblookup_bind(&tcb6, &d, 996 th->th_dport, 0); 997 } 998 } 999 #endif 1000 #ifndef INET6 1001 if (inp == 0) 1002 #else 1003 if (inp == 0 && in6p == 0) 1004 #endif 1005 { 1006 ++tcpstat.tcps_noport; 1007 if (tcp_log_refused && 1008 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1009 tcp4_log_refused(ip, th); 1010 } 1011 TCP_FIELDS_TO_HOST(th); 1012 goto dropwithreset_ratelim; 1013 } 1014 #if defined(IPSEC) || defined(FAST_IPSEC) 1015 if (inp && ipsec4_in_reject(m, inp)) { 1016 ipsecstat.in_polvio++; 1017 goto drop; 1018 } 1019 #ifdef INET6 1020 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1021 ipsecstat.in_polvio++; 1022 goto drop; 1023 } 1024 #endif 1025 #endif /*IPSEC*/ 1026 break; 1027 #endif /*INET*/ 1028 #ifdef INET6 1029 case AF_INET6: 1030 { 1031 int faith; 1032 1033 #if defined(NFAITH) && NFAITH > 0 1034 faith = faithprefix(&ip6->ip6_dst); 1035 #else 1036 faith = 0; 1037 #endif 1038 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 1039 &ip6->ip6_dst, th->th_dport, faith); 1040 if (in6p == NULL) { 1041 ++tcpstat.tcps_pcbhashmiss; 1042 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 1043 th->th_dport, faith); 1044 } 1045 if (in6p == NULL) { 1046 ++tcpstat.tcps_noport; 1047 if (tcp_log_refused && 1048 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1049 tcp6_log_refused(ip6, th); 1050 } 1051 TCP_FIELDS_TO_HOST(th); 1052 goto dropwithreset_ratelim; 1053 } 1054 #if defined(IPSEC) || defined(FAST_IPSEC) 1055 if (ipsec6_in_reject(m, in6p)) { 1056 ipsec6stat.in_polvio++; 1057 goto drop; 1058 } 1059 #endif /*IPSEC*/ 1060 break; 1061 } 1062 #endif 1063 } 1064 1065 /* 1066 * If the state is CLOSED (i.e., TCB does not exist) then 1067 * all data in the incoming segment is discarded. 1068 * If the TCB exists but is in CLOSED state, it is embryonic, 1069 * but should either do a listen or a connect soon. 1070 */ 1071 tp = NULL; 1072 so = NULL; 1073 if (inp) { 1074 tp = intotcpcb(inp); 1075 so = inp->inp_socket; 1076 } 1077 #ifdef INET6 1078 else if (in6p) { 1079 tp = in6totcpcb(in6p); 1080 so = in6p->in6p_socket; 1081 } 1082 #endif 1083 if (tp == 0) { 1084 TCP_FIELDS_TO_HOST(th); 1085 goto dropwithreset_ratelim; 1086 } 1087 if (tp->t_state == TCPS_CLOSED) 1088 goto drop; 1089 1090 /* 1091 * Checksum extended TCP header and data. 1092 */ 1093 switch (af) { 1094 #ifdef INET 1095 case AF_INET: 1096 switch (m->m_pkthdr.csum_flags & 1097 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1098 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1099 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1100 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1101 goto badcsum; 1102 1103 case M_CSUM_TCPv4|M_CSUM_DATA: { 1104 u_int32_t hw_csum = m->m_pkthdr.csum_data; 1105 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1106 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 1107 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 1108 ip->ip_dst.s_addr, 1109 htonl(hw_csum + ntohs(ip->ip_len) + 1110 IPPROTO_UDP)); 1111 } 1112 if ((hw_csum ^ 0xffff) != 0) 1113 goto badcsum; 1114 break; 1115 } 1116 1117 case M_CSUM_TCPv4: 1118 /* Checksum was okay. */ 1119 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1120 break; 1121 1122 default: 1123 /* Must compute it ourselves. */ 1124 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1125 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1126 goto badcsum; 1127 break; 1128 } 1129 break; 1130 #endif /* INET4 */ 1131 1132 #ifdef INET6 1133 case AF_INET6: 1134 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1135 goto badcsum; 1136 break; 1137 #endif /* INET6 */ 1138 } 1139 1140 TCP_FIELDS_TO_HOST(th); 1141 1142 /* Unscale the window into a 32-bit value. */ 1143 if ((tiflags & TH_SYN) == 0) 1144 tiwin = th->th_win << tp->snd_scale; 1145 else 1146 tiwin = th->th_win; 1147 1148 #ifdef INET6 1149 /* save packet options if user wanted */ 1150 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1151 if (in6p->in6p_options) { 1152 m_freem(in6p->in6p_options); 1153 in6p->in6p_options = 0; 1154 } 1155 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1156 } 1157 #endif 1158 1159 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1160 union syn_cache_sa src; 1161 union syn_cache_sa dst; 1162 1163 bzero(&src, sizeof(src)); 1164 bzero(&dst, sizeof(dst)); 1165 switch (af) { 1166 #ifdef INET 1167 case AF_INET: 1168 src.sin.sin_len = sizeof(struct sockaddr_in); 1169 src.sin.sin_family = AF_INET; 1170 src.sin.sin_addr = ip->ip_src; 1171 src.sin.sin_port = th->th_sport; 1172 1173 dst.sin.sin_len = sizeof(struct sockaddr_in); 1174 dst.sin.sin_family = AF_INET; 1175 dst.sin.sin_addr = ip->ip_dst; 1176 dst.sin.sin_port = th->th_dport; 1177 break; 1178 #endif 1179 #ifdef INET6 1180 case AF_INET6: 1181 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1182 src.sin6.sin6_family = AF_INET6; 1183 src.sin6.sin6_addr = ip6->ip6_src; 1184 src.sin6.sin6_port = th->th_sport; 1185 1186 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1187 dst.sin6.sin6_family = AF_INET6; 1188 dst.sin6.sin6_addr = ip6->ip6_dst; 1189 dst.sin6.sin6_port = th->th_dport; 1190 break; 1191 #endif /* INET6 */ 1192 default: 1193 goto badsyn; /*sanity*/ 1194 } 1195 1196 if (so->so_options & SO_DEBUG) { 1197 #ifdef TCP_DEBUG 1198 ostate = tp->t_state; 1199 #endif 1200 1201 tcp_saveti = NULL; 1202 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1203 goto nosave; 1204 1205 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1206 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1207 if (!tcp_saveti) 1208 goto nosave; 1209 } else { 1210 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1211 if (!tcp_saveti) 1212 goto nosave; 1213 MCLAIM(m, &tcp_mowner); 1214 tcp_saveti->m_len = iphlen; 1215 m_copydata(m, 0, iphlen, 1216 mtod(tcp_saveti, caddr_t)); 1217 } 1218 1219 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1220 m_freem(tcp_saveti); 1221 tcp_saveti = NULL; 1222 } else { 1223 tcp_saveti->m_len += sizeof(struct tcphdr); 1224 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1225 sizeof(struct tcphdr)); 1226 } 1227 nosave:; 1228 } 1229 if (so->so_options & SO_ACCEPTCONN) { 1230 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1231 if (tiflags & TH_RST) { 1232 syn_cache_reset(&src.sa, &dst.sa, th); 1233 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1234 (TH_ACK|TH_SYN)) { 1235 /* 1236 * Received a SYN,ACK. This should 1237 * never happen while we are in 1238 * LISTEN. Send an RST. 1239 */ 1240 goto badsyn; 1241 } else if (tiflags & TH_ACK) { 1242 so = syn_cache_get(&src.sa, &dst.sa, 1243 th, toff, tlen, so, m); 1244 if (so == NULL) { 1245 /* 1246 * We don't have a SYN for 1247 * this ACK; send an RST. 1248 */ 1249 goto badsyn; 1250 } else if (so == 1251 (struct socket *)(-1)) { 1252 /* 1253 * We were unable to create 1254 * the connection. If the 1255 * 3-way handshake was 1256 * completed, and RST has 1257 * been sent to the peer. 1258 * Since the mbuf might be 1259 * in use for the reply, 1260 * do not free it. 1261 */ 1262 m = NULL; 1263 } else { 1264 /* 1265 * We have created a 1266 * full-blown connection. 1267 */ 1268 tp = NULL; 1269 inp = NULL; 1270 #ifdef INET6 1271 in6p = NULL; 1272 #endif 1273 switch (so->so_proto->pr_domain->dom_family) { 1274 #ifdef INET 1275 case AF_INET: 1276 inp = sotoinpcb(so); 1277 tp = intotcpcb(inp); 1278 break; 1279 #endif 1280 #ifdef INET6 1281 case AF_INET6: 1282 in6p = sotoin6pcb(so); 1283 tp = in6totcpcb(in6p); 1284 break; 1285 #endif 1286 } 1287 if (tp == NULL) 1288 goto badsyn; /*XXX*/ 1289 tiwin <<= tp->snd_scale; 1290 goto after_listen; 1291 } 1292 } else { 1293 /* 1294 * None of RST, SYN or ACK was set. 1295 * This is an invalid packet for a 1296 * TCB in LISTEN state. Send a RST. 1297 */ 1298 goto badsyn; 1299 } 1300 } else { 1301 /* 1302 * Received a SYN. 1303 */ 1304 1305 #ifdef INET6 1306 /* 1307 * If deprecated address is forbidden, we do 1308 * not accept SYN to deprecated interface 1309 * address to prevent any new inbound 1310 * connection from getting established. 1311 * When we do not accept SYN, we send a TCP 1312 * RST, with deprecated source address (instead 1313 * of dropping it). We compromise it as it is 1314 * much better for peer to send a RST, and 1315 * RST will be the final packet for the 1316 * exchange. 1317 * 1318 * If we do not forbid deprecated addresses, we 1319 * accept the SYN packet. RFC2462 does not 1320 * suggest dropping SYN in this case. 1321 * If we decipher RFC2462 5.5.4, it says like 1322 * this: 1323 * 1. use of deprecated addr with existing 1324 * communication is okay - "SHOULD continue 1325 * to be used" 1326 * 2. use of it with new communication: 1327 * (2a) "SHOULD NOT be used if alternate 1328 * address with sufficient scope is 1329 * available" 1330 * (2b) nothing mentioned otherwise. 1331 * Here we fall into (2b) case as we have no 1332 * choice in our source address selection - we 1333 * must obey the peer. 1334 * 1335 * The wording in RFC2462 is confusing, and 1336 * there are multiple description text for 1337 * deprecated address handling - worse, they 1338 * are not exactly the same. I believe 5.5.4 1339 * is the best one, so we follow 5.5.4. 1340 */ 1341 if (af == AF_INET6 && !ip6_use_deprecated) { 1342 struct in6_ifaddr *ia6; 1343 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1344 &ip6->ip6_dst)) && 1345 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1346 tp = NULL; 1347 goto dropwithreset; 1348 } 1349 } 1350 #endif 1351 1352 /* 1353 * LISTEN socket received a SYN 1354 * from itself? This can't possibly 1355 * be valid; drop the packet. 1356 */ 1357 if (th->th_sport == th->th_dport) { 1358 int i; 1359 1360 switch (af) { 1361 #ifdef INET 1362 case AF_INET: 1363 i = in_hosteq(ip->ip_src, ip->ip_dst); 1364 break; 1365 #endif 1366 #ifdef INET6 1367 case AF_INET6: 1368 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1369 break; 1370 #endif 1371 default: 1372 i = 1; 1373 } 1374 if (i) { 1375 tcpstat.tcps_badsyn++; 1376 goto drop; 1377 } 1378 } 1379 1380 /* 1381 * SYN looks ok; create compressed TCP 1382 * state for it. 1383 */ 1384 if (so->so_qlen <= so->so_qlimit && 1385 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1386 so, m, optp, optlen, &opti)) 1387 m = NULL; 1388 } 1389 goto drop; 1390 } 1391 } 1392 1393 after_listen: 1394 #ifdef DIAGNOSTIC 1395 /* 1396 * Should not happen now that all embryonic connections 1397 * are handled with compressed state. 1398 */ 1399 if (tp->t_state == TCPS_LISTEN) 1400 panic("tcp_input: TCPS_LISTEN"); 1401 #endif 1402 1403 /* 1404 * Segment received on connection. 1405 * Reset idle time and keep-alive timer. 1406 */ 1407 tp->t_rcvtime = tcp_now; 1408 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1409 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1410 1411 /* 1412 * Process options. 1413 */ 1414 if (optp) 1415 tcp_dooptions(tp, optp, optlen, th, &opti); 1416 1417 /* 1418 * Header prediction: check for the two common cases 1419 * of a uni-directional data xfer. If the packet has 1420 * no control flags, is in-sequence, the window didn't 1421 * change and we're not retransmitting, it's a 1422 * candidate. If the length is zero and the ack moved 1423 * forward, we're the sender side of the xfer. Just 1424 * free the data acked & wake any higher level process 1425 * that was blocked waiting for space. If the length 1426 * is non-zero and the ack didn't move, we're the 1427 * receiver side. If we're getting packets in-order 1428 * (the reassembly queue is empty), add the data to 1429 * the socket buffer and note that we need a delayed ack. 1430 */ 1431 if (tp->t_state == TCPS_ESTABLISHED && 1432 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1433 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1434 th->th_seq == tp->rcv_nxt && 1435 tiwin && tiwin == tp->snd_wnd && 1436 tp->snd_nxt == tp->snd_max) { 1437 1438 /* 1439 * If last ACK falls within this segment's sequence numbers, 1440 * record the timestamp. 1441 */ 1442 if (opti.ts_present && 1443 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1444 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1445 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1446 tp->ts_recent = opti.ts_val; 1447 } 1448 1449 if (tlen == 0) { 1450 if (SEQ_GT(th->th_ack, tp->snd_una) && 1451 SEQ_LEQ(th->th_ack, tp->snd_max) && 1452 tp->snd_cwnd >= tp->snd_wnd && 1453 tp->t_dupacks < tcprexmtthresh) { 1454 int slen; 1455 1456 /* 1457 * this is a pure ack for outstanding data. 1458 */ 1459 ++tcpstat.tcps_predack; 1460 if (opti.ts_present && opti.ts_ecr) 1461 tcp_xmit_timer(tp, 1462 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1463 else if (tp->t_rtttime && 1464 SEQ_GT(th->th_ack, tp->t_rtseq)) 1465 tcp_xmit_timer(tp, 1466 tcp_now - tp->t_rtttime); 1467 acked = th->th_ack - tp->snd_una; 1468 tcpstat.tcps_rcvackpack++; 1469 tcpstat.tcps_rcvackbyte += acked; 1470 ND6_HINT(tp); 1471 1472 slen = tp->t_lastm->m_len; 1473 sbdrop(&so->so_snd, acked); 1474 if (so->so_snd.sb_cc != 0) { 1475 tp->t_lastoff -= acked; 1476 if (tp->t_lastm->m_len < slen) 1477 tp->t_inoff -= 1478 (slen - tp->t_lastm->m_len); 1479 } 1480 1481 /* 1482 * We want snd_recover to track snd_una to 1483 * avoid sequence wraparound problems for 1484 * very large transfers. 1485 */ 1486 tp->snd_una = tp->snd_recover = th->th_ack; 1487 m_freem(m); 1488 1489 /* 1490 * If all outstanding data are acked, stop 1491 * retransmit timer, otherwise restart timer 1492 * using current (possibly backed-off) value. 1493 * If process is waiting for space, 1494 * wakeup/selwakeup/signal. If data 1495 * are ready to send, let tcp_output 1496 * decide between more output or persist. 1497 */ 1498 if (tp->snd_una == tp->snd_max) 1499 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1500 else if (TCP_TIMER_ISARMED(tp, 1501 TCPT_PERSIST) == 0) 1502 TCP_TIMER_ARM(tp, TCPT_REXMT, 1503 tp->t_rxtcur); 1504 1505 sowwakeup(so); 1506 if (so->so_snd.sb_cc) 1507 (void) tcp_output(tp); 1508 if (tcp_saveti) 1509 m_freem(tcp_saveti); 1510 return; 1511 } 1512 } else if (th->th_ack == tp->snd_una && 1513 TAILQ_FIRST(&tp->segq) == NULL && 1514 tlen <= sbspace(&so->so_rcv)) { 1515 /* 1516 * this is a pure, in-sequence data packet 1517 * with nothing on the reassembly queue and 1518 * we have enough buffer space to take it. 1519 */ 1520 ++tcpstat.tcps_preddat; 1521 tp->rcv_nxt += tlen; 1522 tcpstat.tcps_rcvpack++; 1523 tcpstat.tcps_rcvbyte += tlen; 1524 ND6_HINT(tp); 1525 /* 1526 * Drop TCP, IP headers and TCP options then add data 1527 * to socket buffer. 1528 */ 1529 if (so->so_state & SS_CANTRCVMORE) 1530 m_freem(m); 1531 else { 1532 m_adj(m, toff + off); 1533 sbappendstream(&so->so_rcv, m); 1534 } 1535 sorwakeup(so); 1536 TCP_SETUP_ACK(tp, th); 1537 if (tp->t_flags & TF_ACKNOW) 1538 (void) tcp_output(tp); 1539 if (tcp_saveti) 1540 m_freem(tcp_saveti); 1541 return; 1542 } 1543 } 1544 1545 /* 1546 * Compute mbuf offset to TCP data segment. 1547 */ 1548 hdroptlen = toff + off; 1549 1550 /* 1551 * Calculate amount of space in receive window, 1552 * and then do TCP input processing. 1553 * Receive window is amount of space in rcv queue, 1554 * but not less than advertised window. 1555 */ 1556 { int win; 1557 1558 win = sbspace(&so->so_rcv); 1559 if (win < 0) 1560 win = 0; 1561 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1562 } 1563 1564 switch (tp->t_state) { 1565 case TCPS_LISTEN: 1566 /* 1567 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1568 */ 1569 if (m->m_flags & (M_BCAST|M_MCAST)) 1570 goto drop; 1571 switch (af) { 1572 #ifdef INET6 1573 case AF_INET6: 1574 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1575 goto drop; 1576 break; 1577 #endif /* INET6 */ 1578 case AF_INET: 1579 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1580 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1581 goto drop; 1582 break; 1583 } 1584 break; 1585 1586 /* 1587 * If the state is SYN_SENT: 1588 * if seg contains an ACK, but not for our SYN, drop the input. 1589 * if seg contains a RST, then drop the connection. 1590 * if seg does not contain SYN, then drop it. 1591 * Otherwise this is an acceptable SYN segment 1592 * initialize tp->rcv_nxt and tp->irs 1593 * if seg contains ack then advance tp->snd_una 1594 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1595 * arrange for segment to be acked (eventually) 1596 * continue processing rest of data/controls, beginning with URG 1597 */ 1598 case TCPS_SYN_SENT: 1599 if ((tiflags & TH_ACK) && 1600 (SEQ_LEQ(th->th_ack, tp->iss) || 1601 SEQ_GT(th->th_ack, tp->snd_max))) 1602 goto dropwithreset; 1603 if (tiflags & TH_RST) { 1604 if (tiflags & TH_ACK) 1605 tp = tcp_drop(tp, ECONNREFUSED); 1606 goto drop; 1607 } 1608 if ((tiflags & TH_SYN) == 0) 1609 goto drop; 1610 if (tiflags & TH_ACK) { 1611 tp->snd_una = tp->snd_recover = th->th_ack; 1612 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1613 tp->snd_nxt = tp->snd_una; 1614 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1615 } 1616 tp->irs = th->th_seq; 1617 tcp_rcvseqinit(tp); 1618 tp->t_flags |= TF_ACKNOW; 1619 tcp_mss_from_peer(tp, opti.maxseg); 1620 1621 /* 1622 * Initialize the initial congestion window. If we 1623 * had to retransmit the SYN, we must initialize cwnd 1624 * to 1 segment (i.e. the Loss Window). 1625 */ 1626 if (tp->t_flags & TF_SYN_REXMT) 1627 tp->snd_cwnd = tp->t_peermss; 1628 else { 1629 int ss = tcp_init_win; 1630 #ifdef INET 1631 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1632 ss = tcp_init_win_local; 1633 #endif 1634 #ifdef INET6 1635 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1636 ss = tcp_init_win_local; 1637 #endif 1638 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1639 } 1640 1641 tcp_rmx_rtt(tp); 1642 if (tiflags & TH_ACK) { 1643 tcpstat.tcps_connects++; 1644 soisconnected(so); 1645 tcp_established(tp); 1646 /* Do window scaling on this connection? */ 1647 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1648 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1649 tp->snd_scale = tp->requested_s_scale; 1650 tp->rcv_scale = tp->request_r_scale; 1651 } 1652 TCP_REASS_LOCK(tp); 1653 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1654 TCP_REASS_UNLOCK(tp); 1655 /* 1656 * if we didn't have to retransmit the SYN, 1657 * use its rtt as our initial srtt & rtt var. 1658 */ 1659 if (tp->t_rtttime) 1660 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1661 } else 1662 tp->t_state = TCPS_SYN_RECEIVED; 1663 1664 /* 1665 * Advance th->th_seq to correspond to first data byte. 1666 * If data, trim to stay within window, 1667 * dropping FIN if necessary. 1668 */ 1669 th->th_seq++; 1670 if (tlen > tp->rcv_wnd) { 1671 todrop = tlen - tp->rcv_wnd; 1672 m_adj(m, -todrop); 1673 tlen = tp->rcv_wnd; 1674 tiflags &= ~TH_FIN; 1675 tcpstat.tcps_rcvpackafterwin++; 1676 tcpstat.tcps_rcvbyteafterwin += todrop; 1677 } 1678 tp->snd_wl1 = th->th_seq - 1; 1679 tp->rcv_up = th->th_seq; 1680 goto step6; 1681 1682 /* 1683 * If the state is SYN_RECEIVED: 1684 * If seg contains an ACK, but not for our SYN, drop the input 1685 * and generate an RST. See page 36, rfc793 1686 */ 1687 case TCPS_SYN_RECEIVED: 1688 if ((tiflags & TH_ACK) && 1689 (SEQ_LEQ(th->th_ack, tp->iss) || 1690 SEQ_GT(th->th_ack, tp->snd_max))) 1691 goto dropwithreset; 1692 break; 1693 } 1694 1695 /* 1696 * States other than LISTEN or SYN_SENT. 1697 * First check timestamp, if present. 1698 * Then check that at least some bytes of segment are within 1699 * receive window. If segment begins before rcv_nxt, 1700 * drop leading data (and SYN); if nothing left, just ack. 1701 * 1702 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1703 * and it's less than ts_recent, drop it. 1704 */ 1705 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1706 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1707 1708 /* Check to see if ts_recent is over 24 days old. */ 1709 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1710 TCP_PAWS_IDLE) { 1711 /* 1712 * Invalidate ts_recent. If this segment updates 1713 * ts_recent, the age will be reset later and ts_recent 1714 * will get a valid value. If it does not, setting 1715 * ts_recent to zero will at least satisfy the 1716 * requirement that zero be placed in the timestamp 1717 * echo reply when ts_recent isn't valid. The 1718 * age isn't reset until we get a valid ts_recent 1719 * because we don't want out-of-order segments to be 1720 * dropped when ts_recent is old. 1721 */ 1722 tp->ts_recent = 0; 1723 } else { 1724 tcpstat.tcps_rcvduppack++; 1725 tcpstat.tcps_rcvdupbyte += tlen; 1726 tcpstat.tcps_pawsdrop++; 1727 goto dropafterack; 1728 } 1729 } 1730 1731 todrop = tp->rcv_nxt - th->th_seq; 1732 if (todrop > 0) { 1733 if (tiflags & TH_SYN) { 1734 tiflags &= ~TH_SYN; 1735 th->th_seq++; 1736 if (th->th_urp > 1) 1737 th->th_urp--; 1738 else { 1739 tiflags &= ~TH_URG; 1740 th->th_urp = 0; 1741 } 1742 todrop--; 1743 } 1744 if (todrop > tlen || 1745 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1746 /* 1747 * Any valid FIN must be to the left of the window. 1748 * At this point the FIN must be a duplicate or 1749 * out of sequence; drop it. 1750 */ 1751 tiflags &= ~TH_FIN; 1752 /* 1753 * Send an ACK to resynchronize and drop any data. 1754 * But keep on processing for RST or ACK. 1755 */ 1756 tp->t_flags |= TF_ACKNOW; 1757 todrop = tlen; 1758 tcpstat.tcps_rcvdupbyte += todrop; 1759 tcpstat.tcps_rcvduppack++; 1760 } else { 1761 tcpstat.tcps_rcvpartduppack++; 1762 tcpstat.tcps_rcvpartdupbyte += todrop; 1763 } 1764 hdroptlen += todrop; /*drop from head afterwards*/ 1765 th->th_seq += todrop; 1766 tlen -= todrop; 1767 if (th->th_urp > todrop) 1768 th->th_urp -= todrop; 1769 else { 1770 tiflags &= ~TH_URG; 1771 th->th_urp = 0; 1772 } 1773 } 1774 1775 /* 1776 * If new data are received on a connection after the 1777 * user processes are gone, then RST the other end. 1778 */ 1779 if ((so->so_state & SS_NOFDREF) && 1780 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1781 tp = tcp_close(tp); 1782 tcpstat.tcps_rcvafterclose++; 1783 goto dropwithreset; 1784 } 1785 1786 /* 1787 * If segment ends after window, drop trailing data 1788 * (and PUSH and FIN); if nothing left, just ACK. 1789 */ 1790 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1791 if (todrop > 0) { 1792 tcpstat.tcps_rcvpackafterwin++; 1793 if (todrop >= tlen) { 1794 tcpstat.tcps_rcvbyteafterwin += tlen; 1795 /* 1796 * If a new connection request is received 1797 * while in TIME_WAIT, drop the old connection 1798 * and start over if the sequence numbers 1799 * are above the previous ones. 1800 * 1801 * NOTE: We will checksum the packet again, and 1802 * so we need to put the header fields back into 1803 * network order! 1804 * XXX This kind of sucks, but we don't expect 1805 * XXX this to happen very often, so maybe it 1806 * XXX doesn't matter so much. 1807 */ 1808 if (tiflags & TH_SYN && 1809 tp->t_state == TCPS_TIME_WAIT && 1810 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1811 iss = tcp_new_iss(tp, tp->snd_nxt); 1812 tp = tcp_close(tp); 1813 TCP_FIELDS_TO_NET(th); 1814 goto findpcb; 1815 } 1816 /* 1817 * If window is closed can only take segments at 1818 * window edge, and have to drop data and PUSH from 1819 * incoming segments. Continue processing, but 1820 * remember to ack. Otherwise, drop segment 1821 * and ack. 1822 */ 1823 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1824 tp->t_flags |= TF_ACKNOW; 1825 tcpstat.tcps_rcvwinprobe++; 1826 } else 1827 goto dropafterack; 1828 } else 1829 tcpstat.tcps_rcvbyteafterwin += todrop; 1830 m_adj(m, -todrop); 1831 tlen -= todrop; 1832 tiflags &= ~(TH_PUSH|TH_FIN); 1833 } 1834 1835 /* 1836 * If last ACK falls within this segment's sequence numbers, 1837 * and the timestamp is newer, record it. 1838 */ 1839 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1840 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1841 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1842 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1843 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1844 tp->ts_recent = opti.ts_val; 1845 } 1846 1847 /* 1848 * If the RST bit is set examine the state: 1849 * SYN_RECEIVED STATE: 1850 * If passive open, return to LISTEN state. 1851 * If active open, inform user that connection was refused. 1852 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1853 * Inform user that connection was reset, and close tcb. 1854 * CLOSING, LAST_ACK, TIME_WAIT STATES 1855 * Close the tcb. 1856 */ 1857 if (tiflags&TH_RST) switch (tp->t_state) { 1858 1859 case TCPS_SYN_RECEIVED: 1860 so->so_error = ECONNREFUSED; 1861 goto close; 1862 1863 case TCPS_ESTABLISHED: 1864 case TCPS_FIN_WAIT_1: 1865 case TCPS_FIN_WAIT_2: 1866 case TCPS_CLOSE_WAIT: 1867 so->so_error = ECONNRESET; 1868 close: 1869 tp->t_state = TCPS_CLOSED; 1870 tcpstat.tcps_drops++; 1871 tp = tcp_close(tp); 1872 goto drop; 1873 1874 case TCPS_CLOSING: 1875 case TCPS_LAST_ACK: 1876 case TCPS_TIME_WAIT: 1877 tp = tcp_close(tp); 1878 goto drop; 1879 } 1880 1881 /* 1882 * If a SYN is in the window, then this is an 1883 * error and we send an RST and drop the connection. 1884 */ 1885 if (tiflags & TH_SYN) { 1886 tp = tcp_drop(tp, ECONNRESET); 1887 goto dropwithreset; 1888 } 1889 1890 /* 1891 * If the ACK bit is off we drop the segment and return. 1892 */ 1893 if ((tiflags & TH_ACK) == 0) { 1894 if (tp->t_flags & TF_ACKNOW) 1895 goto dropafterack; 1896 else 1897 goto drop; 1898 } 1899 1900 /* 1901 * Ack processing. 1902 */ 1903 switch (tp->t_state) { 1904 1905 /* 1906 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1907 * ESTABLISHED state and continue processing, otherwise 1908 * send an RST. 1909 */ 1910 case TCPS_SYN_RECEIVED: 1911 if (SEQ_GT(tp->snd_una, th->th_ack) || 1912 SEQ_GT(th->th_ack, tp->snd_max)) 1913 goto dropwithreset; 1914 tcpstat.tcps_connects++; 1915 soisconnected(so); 1916 tcp_established(tp); 1917 /* Do window scaling? */ 1918 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1919 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1920 tp->snd_scale = tp->requested_s_scale; 1921 tp->rcv_scale = tp->request_r_scale; 1922 } 1923 TCP_REASS_LOCK(tp); 1924 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1925 TCP_REASS_UNLOCK(tp); 1926 tp->snd_wl1 = th->th_seq - 1; 1927 /* fall into ... */ 1928 1929 /* 1930 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1931 * ACKs. If the ack is in the range 1932 * tp->snd_una < th->th_ack <= tp->snd_max 1933 * then advance tp->snd_una to th->th_ack and drop 1934 * data from the retransmission queue. If this ACK reflects 1935 * more up to date window information we update our window information. 1936 */ 1937 case TCPS_ESTABLISHED: 1938 case TCPS_FIN_WAIT_1: 1939 case TCPS_FIN_WAIT_2: 1940 case TCPS_CLOSE_WAIT: 1941 case TCPS_CLOSING: 1942 case TCPS_LAST_ACK: 1943 case TCPS_TIME_WAIT: 1944 1945 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1946 if (tlen == 0 && tiwin == tp->snd_wnd) { 1947 tcpstat.tcps_rcvdupack++; 1948 /* 1949 * If we have outstanding data (other than 1950 * a window probe), this is a completely 1951 * duplicate ack (ie, window info didn't 1952 * change), the ack is the biggest we've 1953 * seen and we've seen exactly our rexmt 1954 * threshhold of them, assume a packet 1955 * has been dropped and retransmit it. 1956 * Kludge snd_nxt & the congestion 1957 * window so we send only this one 1958 * packet. 1959 * 1960 * We know we're losing at the current 1961 * window size so do congestion avoidance 1962 * (set ssthresh to half the current window 1963 * and pull our congestion window back to 1964 * the new ssthresh). 1965 * 1966 * Dup acks mean that packets have left the 1967 * network (they're now cached at the receiver) 1968 * so bump cwnd by the amount in the receiver 1969 * to keep a constant cwnd packets in the 1970 * network. 1971 */ 1972 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1973 th->th_ack != tp->snd_una) 1974 tp->t_dupacks = 0; 1975 else if (++tp->t_dupacks == tcprexmtthresh) { 1976 tcp_seq onxt = tp->snd_nxt; 1977 u_int win = 1978 min(tp->snd_wnd, tp->snd_cwnd) / 1979 2 / tp->t_segsz; 1980 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1981 tp->snd_recover)) { 1982 /* 1983 * False fast retransmit after 1984 * timeout. Do not cut window. 1985 */ 1986 tp->snd_cwnd += tp->t_segsz; 1987 tp->t_dupacks = 0; 1988 (void) tcp_output(tp); 1989 goto drop; 1990 } 1991 1992 if (win < 2) 1993 win = 2; 1994 tp->snd_ssthresh = win * tp->t_segsz; 1995 tp->snd_recover = tp->snd_max; 1996 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1997 tp->t_rtttime = 0; 1998 tp->snd_nxt = th->th_ack; 1999 tp->snd_cwnd = tp->t_segsz; 2000 (void) tcp_output(tp); 2001 tp->snd_cwnd = tp->snd_ssthresh + 2002 tp->t_segsz * tp->t_dupacks; 2003 if (SEQ_GT(onxt, tp->snd_nxt)) 2004 tp->snd_nxt = onxt; 2005 goto drop; 2006 } else if (tp->t_dupacks > tcprexmtthresh) { 2007 tp->snd_cwnd += tp->t_segsz; 2008 (void) tcp_output(tp); 2009 goto drop; 2010 } 2011 } else 2012 tp->t_dupacks = 0; 2013 break; 2014 } 2015 /* 2016 * If the congestion window was inflated to account 2017 * for the other side's cached packets, retract it. 2018 */ 2019 if (tcp_do_newreno == 0) { 2020 if (tp->t_dupacks >= tcprexmtthresh && 2021 tp->snd_cwnd > tp->snd_ssthresh) 2022 tp->snd_cwnd = tp->snd_ssthresh; 2023 tp->t_dupacks = 0; 2024 } else if (tp->t_dupacks >= tcprexmtthresh && 2025 tcp_newreno(tp, th) == 0) { 2026 tp->snd_cwnd = tp->snd_ssthresh; 2027 /* 2028 * Window inflation should have left us with approx. 2029 * snd_ssthresh outstanding data. But in case we 2030 * would be inclined to send a burst, better to do 2031 * it via the slow start mechanism. 2032 */ 2033 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 2034 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 2035 + tp->t_segsz; 2036 tp->t_dupacks = 0; 2037 } 2038 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2039 tcpstat.tcps_rcvacktoomuch++; 2040 goto dropafterack; 2041 } 2042 acked = th->th_ack - tp->snd_una; 2043 tcpstat.tcps_rcvackpack++; 2044 tcpstat.tcps_rcvackbyte += acked; 2045 2046 /* 2047 * If we have a timestamp reply, update smoothed 2048 * round trip time. If no timestamp is present but 2049 * transmit timer is running and timed sequence 2050 * number was acked, update smoothed round trip time. 2051 * Since we now have an rtt measurement, cancel the 2052 * timer backoff (cf., Phil Karn's retransmit alg.). 2053 * Recompute the initial retransmit timer. 2054 */ 2055 if (opti.ts_present && opti.ts_ecr) 2056 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2057 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2058 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2059 2060 /* 2061 * If all outstanding data is acked, stop retransmit 2062 * timer and remember to restart (more output or persist). 2063 * If there is more data to be acked, restart retransmit 2064 * timer, using current (possibly backed-off) value. 2065 */ 2066 if (th->th_ack == tp->snd_max) { 2067 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2068 needoutput = 1; 2069 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2070 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2071 /* 2072 * When new data is acked, open the congestion window. 2073 * If the window gives us less than ssthresh packets 2074 * in flight, open exponentially (segsz per packet). 2075 * Otherwise open linearly: segsz per window 2076 * (segsz^2 / cwnd per packet), plus a constant 2077 * fraction of a packet (segsz/8) to help larger windows 2078 * open quickly enough. 2079 */ 2080 { 2081 u_int cw = tp->snd_cwnd; 2082 u_int incr = tp->t_segsz; 2083 2084 if (cw > tp->snd_ssthresh) 2085 incr = incr * incr / cw; 2086 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2087 tp->snd_cwnd = min(cw + incr, 2088 TCP_MAXWIN << tp->snd_scale); 2089 } 2090 ND6_HINT(tp); 2091 if (acked > so->so_snd.sb_cc) { 2092 tp->snd_wnd -= so->so_snd.sb_cc; 2093 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2094 ourfinisacked = 1; 2095 } else { 2096 int slen; 2097 2098 slen = tp->t_lastm->m_len; 2099 sbdrop(&so->so_snd, acked); 2100 tp->snd_wnd -= acked; 2101 if (so->so_snd.sb_cc != 0) { 2102 tp->t_lastoff -= acked; 2103 if (tp->t_lastm->m_len != slen) 2104 tp->t_inoff -= 2105 (slen - tp->t_lastm->m_len); 2106 } 2107 ourfinisacked = 0; 2108 } 2109 sowwakeup(so); 2110 /* 2111 * We want snd_recover to track snd_una to 2112 * avoid sequence wraparound problems for 2113 * very large transfers. 2114 */ 2115 tp->snd_una = tp->snd_recover = th->th_ack; 2116 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2117 tp->snd_nxt = tp->snd_una; 2118 2119 switch (tp->t_state) { 2120 2121 /* 2122 * In FIN_WAIT_1 STATE in addition to the processing 2123 * for the ESTABLISHED state if our FIN is now acknowledged 2124 * then enter FIN_WAIT_2. 2125 */ 2126 case TCPS_FIN_WAIT_1: 2127 if (ourfinisacked) { 2128 /* 2129 * If we can't receive any more 2130 * data, then closing user can proceed. 2131 * Starting the timer is contrary to the 2132 * specification, but if we don't get a FIN 2133 * we'll hang forever. 2134 */ 2135 if (so->so_state & SS_CANTRCVMORE) { 2136 soisdisconnected(so); 2137 if (tcp_maxidle > 0) 2138 TCP_TIMER_ARM(tp, TCPT_2MSL, 2139 tcp_maxidle); 2140 } 2141 tp->t_state = TCPS_FIN_WAIT_2; 2142 } 2143 break; 2144 2145 /* 2146 * In CLOSING STATE in addition to the processing for 2147 * the ESTABLISHED state if the ACK acknowledges our FIN 2148 * then enter the TIME-WAIT state, otherwise ignore 2149 * the segment. 2150 */ 2151 case TCPS_CLOSING: 2152 if (ourfinisacked) { 2153 tp->t_state = TCPS_TIME_WAIT; 2154 tcp_canceltimers(tp); 2155 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2156 soisdisconnected(so); 2157 } 2158 break; 2159 2160 /* 2161 * In LAST_ACK, we may still be waiting for data to drain 2162 * and/or to be acked, as well as for the ack of our FIN. 2163 * If our FIN is now acknowledged, delete the TCB, 2164 * enter the closed state and return. 2165 */ 2166 case TCPS_LAST_ACK: 2167 if (ourfinisacked) { 2168 tp = tcp_close(tp); 2169 goto drop; 2170 } 2171 break; 2172 2173 /* 2174 * In TIME_WAIT state the only thing that should arrive 2175 * is a retransmission of the remote FIN. Acknowledge 2176 * it and restart the finack timer. 2177 */ 2178 case TCPS_TIME_WAIT: 2179 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2180 goto dropafterack; 2181 } 2182 } 2183 2184 step6: 2185 /* 2186 * Update window information. 2187 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2188 */ 2189 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2190 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2191 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2192 /* keep track of pure window updates */ 2193 if (tlen == 0 && 2194 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2195 tcpstat.tcps_rcvwinupd++; 2196 tp->snd_wnd = tiwin; 2197 tp->snd_wl1 = th->th_seq; 2198 tp->snd_wl2 = th->th_ack; 2199 if (tp->snd_wnd > tp->max_sndwnd) 2200 tp->max_sndwnd = tp->snd_wnd; 2201 needoutput = 1; 2202 } 2203 2204 /* 2205 * Process segments with URG. 2206 */ 2207 if ((tiflags & TH_URG) && th->th_urp && 2208 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2209 /* 2210 * This is a kludge, but if we receive and accept 2211 * random urgent pointers, we'll crash in 2212 * soreceive. It's hard to imagine someone 2213 * actually wanting to send this much urgent data. 2214 */ 2215 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2216 th->th_urp = 0; /* XXX */ 2217 tiflags &= ~TH_URG; /* XXX */ 2218 goto dodata; /* XXX */ 2219 } 2220 /* 2221 * If this segment advances the known urgent pointer, 2222 * then mark the data stream. This should not happen 2223 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2224 * a FIN has been received from the remote side. 2225 * In these states we ignore the URG. 2226 * 2227 * According to RFC961 (Assigned Protocols), 2228 * the urgent pointer points to the last octet 2229 * of urgent data. We continue, however, 2230 * to consider it to indicate the first octet 2231 * of data past the urgent section as the original 2232 * spec states (in one of two places). 2233 */ 2234 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2235 tp->rcv_up = th->th_seq + th->th_urp; 2236 so->so_oobmark = so->so_rcv.sb_cc + 2237 (tp->rcv_up - tp->rcv_nxt) - 1; 2238 if (so->so_oobmark == 0) 2239 so->so_state |= SS_RCVATMARK; 2240 sohasoutofband(so); 2241 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2242 } 2243 /* 2244 * Remove out of band data so doesn't get presented to user. 2245 * This can happen independent of advancing the URG pointer, 2246 * but if two URG's are pending at once, some out-of-band 2247 * data may creep in... ick. 2248 */ 2249 if (th->th_urp <= (u_int16_t) tlen 2250 #ifdef SO_OOBINLINE 2251 && (so->so_options & SO_OOBINLINE) == 0 2252 #endif 2253 ) 2254 tcp_pulloutofband(so, th, m, hdroptlen); 2255 } else 2256 /* 2257 * If no out of band data is expected, 2258 * pull receive urgent pointer along 2259 * with the receive window. 2260 */ 2261 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2262 tp->rcv_up = tp->rcv_nxt; 2263 dodata: /* XXX */ 2264 2265 /* 2266 * Process the segment text, merging it into the TCP sequencing queue, 2267 * and arranging for acknowledgement of receipt if necessary. 2268 * This process logically involves adjusting tp->rcv_wnd as data 2269 * is presented to the user (this happens in tcp_usrreq.c, 2270 * case PRU_RCVD). If a FIN has already been received on this 2271 * connection then we just ignore the text. 2272 */ 2273 if ((tlen || (tiflags & TH_FIN)) && 2274 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2275 /* 2276 * Insert segment ti into reassembly queue of tcp with 2277 * control block tp. Return TH_FIN if reassembly now includes 2278 * a segment with FIN. The macro form does the common case 2279 * inline (segment is the next to be received on an 2280 * established connection, and the queue is empty), 2281 * avoiding linkage into and removal from the queue and 2282 * repetition of various conversions. 2283 * Set DELACK for segments received in order, but ack 2284 * immediately when segments are out of order 2285 * (so fast retransmit can work). 2286 */ 2287 /* NOTE: this was TCP_REASS() macro, but used only once */ 2288 TCP_REASS_LOCK(tp); 2289 if (th->th_seq == tp->rcv_nxt && 2290 TAILQ_FIRST(&tp->segq) == NULL && 2291 tp->t_state == TCPS_ESTABLISHED) { 2292 TCP_SETUP_ACK(tp, th); 2293 tp->rcv_nxt += tlen; 2294 tiflags = th->th_flags & TH_FIN; 2295 tcpstat.tcps_rcvpack++; 2296 tcpstat.tcps_rcvbyte += tlen; 2297 ND6_HINT(tp); 2298 if (so->so_state & SS_CANTRCVMORE) 2299 m_freem(m); 2300 else { 2301 m_adj(m, hdroptlen); 2302 sbappendstream(&(so)->so_rcv, m); 2303 } 2304 sorwakeup(so); 2305 } else { 2306 m_adj(m, hdroptlen); 2307 tiflags = tcp_reass(tp, th, m, &tlen); 2308 tp->t_flags |= TF_ACKNOW; 2309 } 2310 TCP_REASS_UNLOCK(tp); 2311 2312 /* 2313 * Note the amount of data that peer has sent into 2314 * our window, in order to estimate the sender's 2315 * buffer size. 2316 */ 2317 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2318 } else { 2319 m_freem(m); 2320 m = NULL; 2321 tiflags &= ~TH_FIN; 2322 } 2323 2324 /* 2325 * If FIN is received ACK the FIN and let the user know 2326 * that the connection is closing. Ignore a FIN received before 2327 * the connection is fully established. 2328 */ 2329 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2330 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2331 socantrcvmore(so); 2332 tp->t_flags |= TF_ACKNOW; 2333 tp->rcv_nxt++; 2334 } 2335 switch (tp->t_state) { 2336 2337 /* 2338 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2339 */ 2340 case TCPS_ESTABLISHED: 2341 tp->t_state = TCPS_CLOSE_WAIT; 2342 break; 2343 2344 /* 2345 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2346 * enter the CLOSING state. 2347 */ 2348 case TCPS_FIN_WAIT_1: 2349 tp->t_state = TCPS_CLOSING; 2350 break; 2351 2352 /* 2353 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2354 * starting the time-wait timer, turning off the other 2355 * standard timers. 2356 */ 2357 case TCPS_FIN_WAIT_2: 2358 tp->t_state = TCPS_TIME_WAIT; 2359 tcp_canceltimers(tp); 2360 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2361 soisdisconnected(so); 2362 break; 2363 2364 /* 2365 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2366 */ 2367 case TCPS_TIME_WAIT: 2368 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2369 break; 2370 } 2371 } 2372 #ifdef TCP_DEBUG 2373 if (so->so_options & SO_DEBUG) 2374 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2375 #endif 2376 2377 /* 2378 * Return any desired output. 2379 */ 2380 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2381 (void) tcp_output(tp); 2382 if (tcp_saveti) 2383 m_freem(tcp_saveti); 2384 return; 2385 2386 badsyn: 2387 /* 2388 * Received a bad SYN. Increment counters and dropwithreset. 2389 */ 2390 tcpstat.tcps_badsyn++; 2391 tp = NULL; 2392 goto dropwithreset; 2393 2394 dropafterack: 2395 /* 2396 * Generate an ACK dropping incoming segment if it occupies 2397 * sequence space, where the ACK reflects our state. 2398 */ 2399 if (tiflags & TH_RST) 2400 goto drop; 2401 m_freem(m); 2402 tp->t_flags |= TF_ACKNOW; 2403 (void) tcp_output(tp); 2404 if (tcp_saveti) 2405 m_freem(tcp_saveti); 2406 return; 2407 2408 dropwithreset_ratelim: 2409 /* 2410 * We may want to rate-limit RSTs in certain situations, 2411 * particularly if we are sending an RST in response to 2412 * an attempt to connect to or otherwise communicate with 2413 * a port for which we have no socket. 2414 */ 2415 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2416 tcp_rst_ppslim) == 0) { 2417 /* XXX stat */ 2418 goto drop; 2419 } 2420 /* ...fall into dropwithreset... */ 2421 2422 dropwithreset: 2423 /* 2424 * Generate a RST, dropping incoming segment. 2425 * Make ACK acceptable to originator of segment. 2426 */ 2427 if (tiflags & TH_RST) 2428 goto drop; 2429 2430 switch (af) { 2431 #ifdef INET6 2432 case AF_INET6: 2433 /* For following calls to tcp_respond */ 2434 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2435 goto drop; 2436 break; 2437 #endif /* INET6 */ 2438 case AF_INET: 2439 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2440 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2441 goto drop; 2442 } 2443 2444 if (tiflags & TH_ACK) 2445 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2446 else { 2447 if (tiflags & TH_SYN) 2448 tlen++; 2449 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2450 TH_RST|TH_ACK); 2451 } 2452 if (tcp_saveti) 2453 m_freem(tcp_saveti); 2454 return; 2455 2456 badcsum: 2457 tcpstat.tcps_rcvbadsum++; 2458 drop: 2459 /* 2460 * Drop space held by incoming segment and return. 2461 */ 2462 if (tp) { 2463 if (tp->t_inpcb) 2464 so = tp->t_inpcb->inp_socket; 2465 #ifdef INET6 2466 else if (tp->t_in6pcb) 2467 so = tp->t_in6pcb->in6p_socket; 2468 #endif 2469 else 2470 so = NULL; 2471 #ifdef TCP_DEBUG 2472 if (so && (so->so_options & SO_DEBUG) != 0) 2473 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2474 #endif 2475 } 2476 if (tcp_saveti) 2477 m_freem(tcp_saveti); 2478 m_freem(m); 2479 return; 2480 } 2481 2482 void 2483 tcp_dooptions(tp, cp, cnt, th, oi) 2484 struct tcpcb *tp; 2485 u_char *cp; 2486 int cnt; 2487 struct tcphdr *th; 2488 struct tcp_opt_info *oi; 2489 { 2490 u_int16_t mss; 2491 int opt, optlen; 2492 2493 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2494 opt = cp[0]; 2495 if (opt == TCPOPT_EOL) 2496 break; 2497 if (opt == TCPOPT_NOP) 2498 optlen = 1; 2499 else { 2500 if (cnt < 2) 2501 break; 2502 optlen = cp[1]; 2503 if (optlen < 2 || optlen > cnt) 2504 break; 2505 } 2506 switch (opt) { 2507 2508 default: 2509 continue; 2510 2511 case TCPOPT_MAXSEG: 2512 if (optlen != TCPOLEN_MAXSEG) 2513 continue; 2514 if (!(th->th_flags & TH_SYN)) 2515 continue; 2516 bcopy(cp + 2, &mss, sizeof(mss)); 2517 oi->maxseg = ntohs(mss); 2518 break; 2519 2520 case TCPOPT_WINDOW: 2521 if (optlen != TCPOLEN_WINDOW) 2522 continue; 2523 if (!(th->th_flags & TH_SYN)) 2524 continue; 2525 tp->t_flags |= TF_RCVD_SCALE; 2526 tp->requested_s_scale = cp[2]; 2527 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2528 #if 0 /*XXX*/ 2529 char *p; 2530 2531 if (ip) 2532 p = ntohl(ip->ip_src); 2533 #ifdef INET6 2534 else if (ip6) 2535 p = ip6_sprintf(&ip6->ip6_src); 2536 #endif 2537 else 2538 p = "(unknown)"; 2539 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2540 "assuming %d\n", 2541 tp->requested_s_scale, p, 2542 TCP_MAX_WINSHIFT); 2543 #else 2544 log(LOG_ERR, "TCP: invalid wscale %d, " 2545 "assuming %d\n", 2546 tp->requested_s_scale, 2547 TCP_MAX_WINSHIFT); 2548 #endif 2549 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2550 } 2551 break; 2552 2553 case TCPOPT_TIMESTAMP: 2554 if (optlen != TCPOLEN_TIMESTAMP) 2555 continue; 2556 oi->ts_present = 1; 2557 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2558 NTOHL(oi->ts_val); 2559 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2560 NTOHL(oi->ts_ecr); 2561 2562 /* 2563 * A timestamp received in a SYN makes 2564 * it ok to send timestamp requests and replies. 2565 */ 2566 if (th->th_flags & TH_SYN) { 2567 tp->t_flags |= TF_RCVD_TSTMP; 2568 tp->ts_recent = oi->ts_val; 2569 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2570 } 2571 break; 2572 case TCPOPT_SACK_PERMITTED: 2573 if (optlen != TCPOLEN_SACK_PERMITTED) 2574 continue; 2575 if (!(th->th_flags & TH_SYN)) 2576 continue; 2577 tp->t_flags &= ~TF_CANT_TXSACK; 2578 break; 2579 2580 case TCPOPT_SACK: 2581 if (tp->t_flags & TF_IGNR_RXSACK) 2582 continue; 2583 if (optlen % 8 != 2 || optlen < 10) 2584 continue; 2585 cp += 2; 2586 optlen -= 2; 2587 for (; optlen > 0; cp -= 8, optlen -= 8) { 2588 tcp_seq lwe, rwe; 2589 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2590 NTOHL(lwe); 2591 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2592 NTOHL(rwe); 2593 /* tcp_mark_sacked(tp, lwe, rwe); */ 2594 } 2595 break; 2596 } 2597 } 2598 } 2599 2600 /* 2601 * Pull out of band byte out of a segment so 2602 * it doesn't appear in the user's data queue. 2603 * It is still reflected in the segment length for 2604 * sequencing purposes. 2605 */ 2606 void 2607 tcp_pulloutofband(so, th, m, off) 2608 struct socket *so; 2609 struct tcphdr *th; 2610 struct mbuf *m; 2611 int off; 2612 { 2613 int cnt = off + th->th_urp - 1; 2614 2615 while (cnt >= 0) { 2616 if (m->m_len > cnt) { 2617 char *cp = mtod(m, caddr_t) + cnt; 2618 struct tcpcb *tp = sototcpcb(so); 2619 2620 tp->t_iobc = *cp; 2621 tp->t_oobflags |= TCPOOB_HAVEDATA; 2622 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2623 m->m_len--; 2624 return; 2625 } 2626 cnt -= m->m_len; 2627 m = m->m_next; 2628 if (m == 0) 2629 break; 2630 } 2631 panic("tcp_pulloutofband"); 2632 } 2633 2634 /* 2635 * Collect new round-trip time estimate 2636 * and update averages and current timeout. 2637 */ 2638 void 2639 tcp_xmit_timer(tp, rtt) 2640 struct tcpcb *tp; 2641 uint32_t rtt; 2642 { 2643 int32_t delta; 2644 2645 tcpstat.tcps_rttupdated++; 2646 if (tp->t_srtt != 0) { 2647 /* 2648 * srtt is stored as fixed point with 3 bits after the 2649 * binary point (i.e., scaled by 8). The following magic 2650 * is equivalent to the smoothing algorithm in rfc793 with 2651 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2652 * point). Adjust rtt to origin 0. 2653 */ 2654 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2655 if ((tp->t_srtt += delta) <= 0) 2656 tp->t_srtt = 1 << 2; 2657 /* 2658 * We accumulate a smoothed rtt variance (actually, a 2659 * smoothed mean difference), then set the retransmit 2660 * timer to smoothed rtt + 4 times the smoothed variance. 2661 * rttvar is stored as fixed point with 2 bits after the 2662 * binary point (scaled by 4). The following is 2663 * equivalent to rfc793 smoothing with an alpha of .75 2664 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2665 * rfc793's wired-in beta. 2666 */ 2667 if (delta < 0) 2668 delta = -delta; 2669 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2670 if ((tp->t_rttvar += delta) <= 0) 2671 tp->t_rttvar = 1 << 2; 2672 } else { 2673 /* 2674 * No rtt measurement yet - use the unsmoothed rtt. 2675 * Set the variance to half the rtt (so our first 2676 * retransmit happens at 3*rtt). 2677 */ 2678 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2679 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2680 } 2681 tp->t_rtttime = 0; 2682 tp->t_rxtshift = 0; 2683 2684 /* 2685 * the retransmit should happen at rtt + 4 * rttvar. 2686 * Because of the way we do the smoothing, srtt and rttvar 2687 * will each average +1/2 tick of bias. When we compute 2688 * the retransmit timer, we want 1/2 tick of rounding and 2689 * 1 extra tick because of +-1/2 tick uncertainty in the 2690 * firing of the timer. The bias will give us exactly the 2691 * 1.5 tick we need. But, because the bias is 2692 * statistical, we have to test that we don't drop below 2693 * the minimum feasible timer (which is 2 ticks). 2694 */ 2695 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2696 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2697 2698 /* 2699 * We received an ack for a packet that wasn't retransmitted; 2700 * it is probably safe to discard any error indications we've 2701 * received recently. This isn't quite right, but close enough 2702 * for now (a route might have failed after we sent a segment, 2703 * and the return path might not be symmetrical). 2704 */ 2705 tp->t_softerror = 0; 2706 } 2707 2708 /* 2709 * Checks for partial ack. If partial ack arrives, force the retransmission 2710 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2711 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2712 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2713 */ 2714 int 2715 tcp_newreno(tp, th) 2716 struct tcpcb *tp; 2717 struct tcphdr *th; 2718 { 2719 tcp_seq onxt = tp->snd_nxt; 2720 u_long ocwnd = tp->snd_cwnd; 2721 2722 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2723 /* 2724 * snd_una has not yet been updated and the socket's send 2725 * buffer has not yet drained off the ACK'd data, so we 2726 * have to leave snd_una as it was to get the correct data 2727 * offset in tcp_output(). 2728 */ 2729 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2730 tp->t_rtttime = 0; 2731 tp->snd_nxt = th->th_ack; 2732 /* 2733 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2734 * is not yet updated when we're called. 2735 */ 2736 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2737 (void) tcp_output(tp); 2738 tp->snd_cwnd = ocwnd; 2739 if (SEQ_GT(onxt, tp->snd_nxt)) 2740 tp->snd_nxt = onxt; 2741 /* 2742 * Partial window deflation. Relies on fact that tp->snd_una 2743 * not updated yet. 2744 */ 2745 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2746 return 1; 2747 } 2748 return 0; 2749 } 2750 2751 2752 /* 2753 * TCP compressed state engine. Currently used to hold compressed 2754 * state for SYN_RECEIVED. 2755 */ 2756 2757 u_long syn_cache_count; 2758 u_int32_t syn_hash1, syn_hash2; 2759 2760 #define SYN_HASH(sa, sp, dp) \ 2761 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2762 ((u_int32_t)(sp)))^syn_hash2))) 2763 #ifndef INET6 2764 #define SYN_HASHALL(hash, src, dst) \ 2765 do { \ 2766 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2767 ((struct sockaddr_in *)(src))->sin_port, \ 2768 ((struct sockaddr_in *)(dst))->sin_port); \ 2769 } while (/*CONSTCOND*/ 0) 2770 #else 2771 #define SYN_HASH6(sa, sp, dp) \ 2772 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2773 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2774 & 0x7fffffff) 2775 2776 #define SYN_HASHALL(hash, src, dst) \ 2777 do { \ 2778 switch ((src)->sa_family) { \ 2779 case AF_INET: \ 2780 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2781 ((struct sockaddr_in *)(src))->sin_port, \ 2782 ((struct sockaddr_in *)(dst))->sin_port); \ 2783 break; \ 2784 case AF_INET6: \ 2785 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2786 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2787 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2788 break; \ 2789 default: \ 2790 hash = 0; \ 2791 } \ 2792 } while (/*CONSTCOND*/0) 2793 #endif /* INET6 */ 2794 2795 #define SYN_CACHE_RM(sc) \ 2796 do { \ 2797 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 2798 (sc), sc_bucketq); \ 2799 (sc)->sc_tp = NULL; \ 2800 LIST_REMOVE((sc), sc_tpq); \ 2801 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2802 callout_stop(&(sc)->sc_timer); \ 2803 syn_cache_count--; \ 2804 } while (/*CONSTCOND*/0) 2805 2806 #define SYN_CACHE_PUT(sc) \ 2807 do { \ 2808 if ((sc)->sc_ipopts) \ 2809 (void) m_free((sc)->sc_ipopts); \ 2810 if ((sc)->sc_route4.ro_rt != NULL) \ 2811 RTFREE((sc)->sc_route4.ro_rt); \ 2812 if (callout_invoking(&(sc)->sc_timer)) \ 2813 (sc)->sc_flags |= SCF_DEAD; \ 2814 else \ 2815 pool_put(&syn_cache_pool, (sc)); \ 2816 } while (/*CONSTCOND*/0) 2817 2818 struct pool syn_cache_pool; 2819 2820 /* 2821 * We don't estimate RTT with SYNs, so each packet starts with the default 2822 * RTT and each timer step has a fixed timeout value. 2823 */ 2824 #define SYN_CACHE_TIMER_ARM(sc) \ 2825 do { \ 2826 TCPT_RANGESET((sc)->sc_rxtcur, \ 2827 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2828 TCPTV_REXMTMAX); \ 2829 callout_reset(&(sc)->sc_timer, \ 2830 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 2831 } while (/*CONSTCOND*/0) 2832 2833 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2834 2835 void 2836 syn_cache_init() 2837 { 2838 int i; 2839 2840 /* Initialize the hash buckets. */ 2841 for (i = 0; i < tcp_syn_cache_size; i++) 2842 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 2843 2844 /* Initialize the syn cache pool. */ 2845 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2846 "synpl", NULL); 2847 } 2848 2849 void 2850 syn_cache_insert(sc, tp) 2851 struct syn_cache *sc; 2852 struct tcpcb *tp; 2853 { 2854 struct syn_cache_head *scp; 2855 struct syn_cache *sc2; 2856 int s; 2857 2858 /* 2859 * If there are no entries in the hash table, reinitialize 2860 * the hash secrets. 2861 */ 2862 if (syn_cache_count == 0) { 2863 struct timeval tv; 2864 microtime(&tv); 2865 syn_hash1 = arc4random() ^ (u_long)≻ 2866 syn_hash2 = arc4random() ^ tv.tv_usec; 2867 } 2868 2869 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2870 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2871 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2872 2873 /* 2874 * Make sure that we don't overflow the per-bucket 2875 * limit or the total cache size limit. 2876 */ 2877 s = splsoftnet(); 2878 if (scp->sch_length >= tcp_syn_bucket_limit) { 2879 tcpstat.tcps_sc_bucketoverflow++; 2880 /* 2881 * The bucket is full. Toss the oldest element in the 2882 * bucket. This will be the first entry in the bucket. 2883 */ 2884 sc2 = TAILQ_FIRST(&scp->sch_bucket); 2885 #ifdef DIAGNOSTIC 2886 /* 2887 * This should never happen; we should always find an 2888 * entry in our bucket. 2889 */ 2890 if (sc2 == NULL) 2891 panic("syn_cache_insert: bucketoverflow: impossible"); 2892 #endif 2893 SYN_CACHE_RM(sc2); 2894 SYN_CACHE_PUT(sc2); 2895 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2896 struct syn_cache_head *scp2, *sce; 2897 2898 tcpstat.tcps_sc_overflowed++; 2899 /* 2900 * The cache is full. Toss the oldest entry in the 2901 * first non-empty bucket we can find. 2902 * 2903 * XXX We would really like to toss the oldest 2904 * entry in the cache, but we hope that this 2905 * condition doesn't happen very often. 2906 */ 2907 scp2 = scp; 2908 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 2909 sce = &tcp_syn_cache[tcp_syn_cache_size]; 2910 for (++scp2; scp2 != scp; scp2++) { 2911 if (scp2 >= sce) 2912 scp2 = &tcp_syn_cache[0]; 2913 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 2914 break; 2915 } 2916 #ifdef DIAGNOSTIC 2917 /* 2918 * This should never happen; we should always find a 2919 * non-empty bucket. 2920 */ 2921 if (scp2 == scp) 2922 panic("syn_cache_insert: cacheoverflow: " 2923 "impossible"); 2924 #endif 2925 } 2926 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 2927 SYN_CACHE_RM(sc2); 2928 SYN_CACHE_PUT(sc2); 2929 } 2930 2931 /* 2932 * Initialize the entry's timer. 2933 */ 2934 sc->sc_rxttot = 0; 2935 sc->sc_rxtshift = 0; 2936 SYN_CACHE_TIMER_ARM(sc); 2937 2938 /* Link it from tcpcb entry */ 2939 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2940 2941 /* Put it into the bucket. */ 2942 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 2943 scp->sch_length++; 2944 syn_cache_count++; 2945 2946 tcpstat.tcps_sc_added++; 2947 splx(s); 2948 } 2949 2950 /* 2951 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2952 * If we have retransmitted an entry the maximum number of times, expire 2953 * that entry. 2954 */ 2955 void 2956 syn_cache_timer(void *arg) 2957 { 2958 struct syn_cache *sc = arg; 2959 int s; 2960 2961 s = splsoftnet(); 2962 callout_ack(&sc->sc_timer); 2963 2964 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 2965 tcpstat.tcps_sc_delayed_free++; 2966 pool_put(&syn_cache_pool, sc); 2967 splx(s); 2968 return; 2969 } 2970 2971 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 2972 /* Drop it -- too many retransmissions. */ 2973 goto dropit; 2974 } 2975 2976 /* 2977 * Compute the total amount of time this entry has 2978 * been on a queue. If this entry has been on longer 2979 * than the keep alive timer would allow, expire it. 2980 */ 2981 sc->sc_rxttot += sc->sc_rxtcur; 2982 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 2983 goto dropit; 2984 2985 tcpstat.tcps_sc_retransmitted++; 2986 (void) syn_cache_respond(sc, NULL); 2987 2988 /* Advance the timer back-off. */ 2989 sc->sc_rxtshift++; 2990 SYN_CACHE_TIMER_ARM(sc); 2991 2992 splx(s); 2993 return; 2994 2995 dropit: 2996 tcpstat.tcps_sc_timed_out++; 2997 SYN_CACHE_RM(sc); 2998 SYN_CACHE_PUT(sc); 2999 splx(s); 3000 } 3001 3002 /* 3003 * Remove syn cache created by the specified tcb entry, 3004 * because this does not make sense to keep them 3005 * (if there's no tcb entry, syn cache entry will never be used) 3006 */ 3007 void 3008 syn_cache_cleanup(tp) 3009 struct tcpcb *tp; 3010 { 3011 struct syn_cache *sc, *nsc; 3012 int s; 3013 3014 s = splsoftnet(); 3015 3016 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3017 nsc = LIST_NEXT(sc, sc_tpq); 3018 3019 #ifdef DIAGNOSTIC 3020 if (sc->sc_tp != tp) 3021 panic("invalid sc_tp in syn_cache_cleanup"); 3022 #endif 3023 SYN_CACHE_RM(sc); 3024 SYN_CACHE_PUT(sc); 3025 } 3026 /* just for safety */ 3027 LIST_INIT(&tp->t_sc); 3028 3029 splx(s); 3030 } 3031 3032 /* 3033 * Find an entry in the syn cache. 3034 */ 3035 struct syn_cache * 3036 syn_cache_lookup(src, dst, headp) 3037 struct sockaddr *src; 3038 struct sockaddr *dst; 3039 struct syn_cache_head **headp; 3040 { 3041 struct syn_cache *sc; 3042 struct syn_cache_head *scp; 3043 u_int32_t hash; 3044 int s; 3045 3046 SYN_HASHALL(hash, src, dst); 3047 3048 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3049 *headp = scp; 3050 s = splsoftnet(); 3051 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3052 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3053 if (sc->sc_hash != hash) 3054 continue; 3055 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3056 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3057 splx(s); 3058 return (sc); 3059 } 3060 } 3061 splx(s); 3062 return (NULL); 3063 } 3064 3065 /* 3066 * This function gets called when we receive an ACK for a 3067 * socket in the LISTEN state. We look up the connection 3068 * in the syn cache, and if its there, we pull it out of 3069 * the cache and turn it into a full-blown connection in 3070 * the SYN-RECEIVED state. 3071 * 3072 * The return values may not be immediately obvious, and their effects 3073 * can be subtle, so here they are: 3074 * 3075 * NULL SYN was not found in cache; caller should drop the 3076 * packet and send an RST. 3077 * 3078 * -1 We were unable to create the new connection, and are 3079 * aborting it. An ACK,RST is being sent to the peer 3080 * (unless we got screwey sequence numbners; see below), 3081 * because the 3-way handshake has been completed. Caller 3082 * should not free the mbuf, since we may be using it. If 3083 * we are not, we will free it. 3084 * 3085 * Otherwise, the return value is a pointer to the new socket 3086 * associated with the connection. 3087 */ 3088 struct socket * 3089 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3090 struct sockaddr *src; 3091 struct sockaddr *dst; 3092 struct tcphdr *th; 3093 unsigned int hlen, tlen; 3094 struct socket *so; 3095 struct mbuf *m; 3096 { 3097 struct syn_cache *sc; 3098 struct syn_cache_head *scp; 3099 struct inpcb *inp = NULL; 3100 #ifdef INET6 3101 struct in6pcb *in6p = NULL; 3102 #endif 3103 struct tcpcb *tp = 0; 3104 struct mbuf *am; 3105 int s; 3106 struct socket *oso; 3107 3108 s = splsoftnet(); 3109 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3110 splx(s); 3111 return (NULL); 3112 } 3113 3114 /* 3115 * Verify the sequence and ack numbers. Try getting the correct 3116 * response again. 3117 */ 3118 if ((th->th_ack != sc->sc_iss + 1) || 3119 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3120 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3121 (void) syn_cache_respond(sc, m); 3122 splx(s); 3123 return ((struct socket *)(-1)); 3124 } 3125 3126 /* Remove this cache entry */ 3127 SYN_CACHE_RM(sc); 3128 splx(s); 3129 3130 /* 3131 * Ok, create the full blown connection, and set things up 3132 * as they would have been set up if we had created the 3133 * connection when the SYN arrived. If we can't create 3134 * the connection, abort it. 3135 */ 3136 /* 3137 * inp still has the OLD in_pcb stuff, set the 3138 * v6-related flags on the new guy, too. This is 3139 * done particularly for the case where an AF_INET6 3140 * socket is bound only to a port, and a v4 connection 3141 * comes in on that port. 3142 * we also copy the flowinfo from the original pcb 3143 * to the new one. 3144 */ 3145 oso = so; 3146 so = sonewconn(so, SS_ISCONNECTED); 3147 if (so == NULL) 3148 goto resetandabort; 3149 3150 switch (so->so_proto->pr_domain->dom_family) { 3151 #ifdef INET 3152 case AF_INET: 3153 inp = sotoinpcb(so); 3154 break; 3155 #endif 3156 #ifdef INET6 3157 case AF_INET6: 3158 in6p = sotoin6pcb(so); 3159 break; 3160 #endif 3161 } 3162 switch (src->sa_family) { 3163 #ifdef INET 3164 case AF_INET: 3165 if (inp) { 3166 struct in_ifaddr *ia; 3167 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3168 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3169 inp->inp_options = ip_srcroute(); 3170 INADDR_TO_IA(inp->inp_laddr, ia); 3171 KASSERT(ia != NULL); 3172 KASSERT(inp->inp_ia == NULL); 3173 inp->inp_ia = ia; 3174 LIST_INSERT_HEAD(&ia->ia_inpcbs, inp, inp_ialink); 3175 IFAREF(&ia->ia_ifa); 3176 in_pcbstate(inp, INP_BOUND); 3177 if (inp->inp_options == NULL) { 3178 inp->inp_options = sc->sc_ipopts; 3179 sc->sc_ipopts = NULL; 3180 } 3181 } 3182 #ifdef INET6 3183 else if (in6p) { 3184 /* IPv4 packet to AF_INET6 socket */ 3185 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3186 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3187 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3188 &in6p->in6p_laddr.s6_addr32[3], 3189 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3190 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3191 in6totcpcb(in6p)->t_family = AF_INET; 3192 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3193 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3194 else 3195 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3196 } 3197 #endif 3198 break; 3199 #endif 3200 #ifdef INET6 3201 case AF_INET6: 3202 if (in6p) { 3203 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3204 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3205 #if 0 3206 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3207 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3208 #endif 3209 } 3210 break; 3211 #endif 3212 } 3213 #ifdef INET6 3214 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3215 struct in6pcb *oin6p = sotoin6pcb(oso); 3216 /* inherit socket options from the listening socket */ 3217 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3218 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3219 m_freem(in6p->in6p_options); 3220 in6p->in6p_options = 0; 3221 } 3222 ip6_savecontrol(in6p, &in6p->in6p_options, 3223 mtod(m, struct ip6_hdr *), m); 3224 } 3225 #endif 3226 3227 #if defined(IPSEC) || defined(FAST_IPSEC) 3228 /* 3229 * we make a copy of policy, instead of sharing the policy, 3230 * for better behavior in terms of SA lookup and dead SA removal. 3231 */ 3232 if (inp) { 3233 /* copy old policy into new socket's */ 3234 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3235 printf("tcp_input: could not copy policy\n"); 3236 } 3237 #ifdef INET6 3238 else if (in6p) { 3239 /* copy old policy into new socket's */ 3240 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3241 in6p->in6p_sp)) 3242 printf("tcp_input: could not copy policy\n"); 3243 } 3244 #endif 3245 #endif 3246 3247 /* 3248 * Give the new socket our cached route reference. 3249 */ 3250 if (inp) 3251 inp->inp_route = sc->sc_route4; /* struct assignment */ 3252 #ifdef INET6 3253 else 3254 in6p->in6p_route = sc->sc_route6; 3255 #endif 3256 sc->sc_route4.ro_rt = NULL; 3257 3258 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3259 if (am == NULL) 3260 goto resetandabort; 3261 MCLAIM(am, &tcp_mowner); 3262 am->m_len = src->sa_len; 3263 bcopy(src, mtod(am, caddr_t), src->sa_len); 3264 if (inp) { 3265 if (in_pcbconnect(inp, am)) { 3266 (void) m_free(am); 3267 goto resetandabort; 3268 } 3269 } 3270 #ifdef INET6 3271 else if (in6p) { 3272 if (src->sa_family == AF_INET) { 3273 /* IPv4 packet to AF_INET6 socket */ 3274 struct sockaddr_in6 *sin6; 3275 sin6 = mtod(am, struct sockaddr_in6 *); 3276 am->m_len = sizeof(*sin6); 3277 bzero(sin6, sizeof(*sin6)); 3278 sin6->sin6_family = AF_INET6; 3279 sin6->sin6_len = sizeof(*sin6); 3280 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3281 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3282 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3283 &sin6->sin6_addr.s6_addr32[3], 3284 sizeof(sin6->sin6_addr.s6_addr32[3])); 3285 } 3286 if (in6_pcbconnect(in6p, am)) { 3287 (void) m_free(am); 3288 goto resetandabort; 3289 } 3290 } 3291 #endif 3292 else { 3293 (void) m_free(am); 3294 goto resetandabort; 3295 } 3296 (void) m_free(am); 3297 3298 if (inp) 3299 tp = intotcpcb(inp); 3300 #ifdef INET6 3301 else if (in6p) 3302 tp = in6totcpcb(in6p); 3303 #endif 3304 else 3305 tp = NULL; 3306 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3307 if (sc->sc_request_r_scale != 15) { 3308 tp->requested_s_scale = sc->sc_requested_s_scale; 3309 tp->request_r_scale = sc->sc_request_r_scale; 3310 tp->snd_scale = sc->sc_requested_s_scale; 3311 tp->rcv_scale = sc->sc_request_r_scale; 3312 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3313 } 3314 if (sc->sc_flags & SCF_TIMESTAMP) 3315 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3316 tp->ts_timebase = sc->sc_timebase; 3317 3318 tp->t_template = tcp_template(tp); 3319 if (tp->t_template == 0) { 3320 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3321 so = NULL; 3322 m_freem(m); 3323 goto abort; 3324 } 3325 3326 tp->iss = sc->sc_iss; 3327 tp->irs = sc->sc_irs; 3328 tcp_sendseqinit(tp); 3329 tcp_rcvseqinit(tp); 3330 tp->t_state = TCPS_SYN_RECEIVED; 3331 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3332 tcpstat.tcps_accepts++; 3333 3334 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3335 tp->t_ourmss = sc->sc_ourmaxseg; 3336 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3337 3338 /* 3339 * Initialize the initial congestion window. If we 3340 * had to retransmit the SYN,ACK, we must initialize cwnd 3341 * to 1 segment (i.e. the Loss Window). 3342 */ 3343 if (sc->sc_rxtshift) 3344 tp->snd_cwnd = tp->t_peermss; 3345 else { 3346 int ss = tcp_init_win; 3347 #ifdef INET 3348 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3349 ss = tcp_init_win_local; 3350 #endif 3351 #ifdef INET6 3352 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3353 ss = tcp_init_win_local; 3354 #endif 3355 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3356 } 3357 3358 tcp_rmx_rtt(tp); 3359 tp->snd_wl1 = sc->sc_irs; 3360 tp->rcv_up = sc->sc_irs + 1; 3361 3362 /* 3363 * This is what whould have happened in tcp_output() when 3364 * the SYN,ACK was sent. 3365 */ 3366 tp->snd_up = tp->snd_una; 3367 tp->snd_max = tp->snd_nxt = tp->iss+1; 3368 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3369 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3370 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3371 tp->last_ack_sent = tp->rcv_nxt; 3372 3373 tcpstat.tcps_sc_completed++; 3374 SYN_CACHE_PUT(sc); 3375 return (so); 3376 3377 resetandabort: 3378 (void) tcp_respond(NULL, m, m, th, 3379 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3380 abort: 3381 if (so != NULL) 3382 (void) soabort(so); 3383 SYN_CACHE_PUT(sc); 3384 tcpstat.tcps_sc_aborted++; 3385 return ((struct socket *)(-1)); 3386 } 3387 3388 /* 3389 * This function is called when we get a RST for a 3390 * non-existent connection, so that we can see if the 3391 * connection is in the syn cache. If it is, zap it. 3392 */ 3393 3394 void 3395 syn_cache_reset(src, dst, th) 3396 struct sockaddr *src; 3397 struct sockaddr *dst; 3398 struct tcphdr *th; 3399 { 3400 struct syn_cache *sc; 3401 struct syn_cache_head *scp; 3402 int s = splsoftnet(); 3403 3404 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3405 splx(s); 3406 return; 3407 } 3408 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3409 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3410 splx(s); 3411 return; 3412 } 3413 SYN_CACHE_RM(sc); 3414 splx(s); 3415 tcpstat.tcps_sc_reset++; 3416 SYN_CACHE_PUT(sc); 3417 } 3418 3419 void 3420 syn_cache_unreach(src, dst, th) 3421 struct sockaddr *src; 3422 struct sockaddr *dst; 3423 struct tcphdr *th; 3424 { 3425 struct syn_cache *sc; 3426 struct syn_cache_head *scp; 3427 int s; 3428 3429 s = splsoftnet(); 3430 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3431 splx(s); 3432 return; 3433 } 3434 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3435 if (ntohl (th->th_seq) != sc->sc_iss) { 3436 splx(s); 3437 return; 3438 } 3439 3440 /* 3441 * If we've rertransmitted 3 times and this is our second error, 3442 * we remove the entry. Otherwise, we allow it to continue on. 3443 * This prevents us from incorrectly nuking an entry during a 3444 * spurious network outage. 3445 * 3446 * See tcp_notify(). 3447 */ 3448 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3449 sc->sc_flags |= SCF_UNREACH; 3450 splx(s); 3451 return; 3452 } 3453 3454 SYN_CACHE_RM(sc); 3455 splx(s); 3456 tcpstat.tcps_sc_unreach++; 3457 SYN_CACHE_PUT(sc); 3458 } 3459 3460 /* 3461 * Given a LISTEN socket and an inbound SYN request, add 3462 * this to the syn cache, and send back a segment: 3463 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3464 * to the source. 3465 * 3466 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3467 * Doing so would require that we hold onto the data and deliver it 3468 * to the application. However, if we are the target of a SYN-flood 3469 * DoS attack, an attacker could send data which would eventually 3470 * consume all available buffer space if it were ACKed. By not ACKing 3471 * the data, we avoid this DoS scenario. 3472 */ 3473 3474 int 3475 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3476 struct sockaddr *src; 3477 struct sockaddr *dst; 3478 struct tcphdr *th; 3479 unsigned int hlen; 3480 struct socket *so; 3481 struct mbuf *m; 3482 u_char *optp; 3483 int optlen; 3484 struct tcp_opt_info *oi; 3485 { 3486 struct tcpcb tb, *tp; 3487 long win; 3488 struct syn_cache *sc; 3489 struct syn_cache_head *scp; 3490 struct mbuf *ipopts; 3491 3492 tp = sototcpcb(so); 3493 3494 /* 3495 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3496 * 3497 * Note this check is performed in tcp_input() very early on. 3498 */ 3499 3500 /* 3501 * Initialize some local state. 3502 */ 3503 win = sbspace(&so->so_rcv); 3504 if (win > TCP_MAXWIN) 3505 win = TCP_MAXWIN; 3506 3507 switch (src->sa_family) { 3508 #ifdef INET 3509 case AF_INET: 3510 /* 3511 * Remember the IP options, if any. 3512 */ 3513 ipopts = ip_srcroute(); 3514 break; 3515 #endif 3516 default: 3517 ipopts = NULL; 3518 } 3519 3520 if (optp) { 3521 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3522 tcp_dooptions(&tb, optp, optlen, th, oi); 3523 } else 3524 tb.t_flags = 0; 3525 3526 /* 3527 * See if we already have an entry for this connection. 3528 * If we do, resend the SYN,ACK. We do not count this 3529 * as a retransmission (XXX though maybe we should). 3530 */ 3531 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3532 tcpstat.tcps_sc_dupesyn++; 3533 if (ipopts) { 3534 /* 3535 * If we were remembering a previous source route, 3536 * forget it and use the new one we've been given. 3537 */ 3538 if (sc->sc_ipopts) 3539 (void) m_free(sc->sc_ipopts); 3540 sc->sc_ipopts = ipopts; 3541 } 3542 sc->sc_timestamp = tb.ts_recent; 3543 if (syn_cache_respond(sc, m) == 0) { 3544 tcpstat.tcps_sndacks++; 3545 tcpstat.tcps_sndtotal++; 3546 } 3547 return (1); 3548 } 3549 3550 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3551 if (sc == NULL) { 3552 if (ipopts) 3553 (void) m_free(ipopts); 3554 return (0); 3555 } 3556 3557 /* 3558 * Fill in the cache, and put the necessary IP and TCP 3559 * options into the reply. 3560 */ 3561 bzero(sc, sizeof(struct syn_cache)); 3562 callout_init(&sc->sc_timer); 3563 bcopy(src, &sc->sc_src, src->sa_len); 3564 bcopy(dst, &sc->sc_dst, dst->sa_len); 3565 sc->sc_flags = 0; 3566 sc->sc_ipopts = ipopts; 3567 sc->sc_irs = th->th_seq; 3568 switch (src->sa_family) { 3569 #ifdef INET 3570 case AF_INET: 3571 { 3572 struct sockaddr_in *srcin = (void *) src; 3573 struct sockaddr_in *dstin = (void *) dst; 3574 3575 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3576 &srcin->sin_addr, dstin->sin_port, 3577 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3578 break; 3579 } 3580 #endif /* INET */ 3581 #ifdef INET6 3582 case AF_INET6: 3583 { 3584 struct sockaddr_in6 *srcin6 = (void *) src; 3585 struct sockaddr_in6 *dstin6 = (void *) dst; 3586 3587 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3588 &srcin6->sin6_addr, dstin6->sin6_port, 3589 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3590 break; 3591 } 3592 #endif /* INET6 */ 3593 } 3594 sc->sc_peermaxseg = oi->maxseg; 3595 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3596 m->m_pkthdr.rcvif : NULL, 3597 sc->sc_src.sa.sa_family); 3598 sc->sc_win = win; 3599 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3600 sc->sc_timestamp = tb.ts_recent; 3601 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3602 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3603 sc->sc_flags |= SCF_TIMESTAMP; 3604 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3605 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3606 sc->sc_requested_s_scale = tb.requested_s_scale; 3607 sc->sc_request_r_scale = 0; 3608 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3609 TCP_MAXWIN << sc->sc_request_r_scale < 3610 so->so_rcv.sb_hiwat) 3611 sc->sc_request_r_scale++; 3612 } else { 3613 sc->sc_requested_s_scale = 15; 3614 sc->sc_request_r_scale = 15; 3615 } 3616 sc->sc_tp = tp; 3617 if (syn_cache_respond(sc, m) == 0) { 3618 syn_cache_insert(sc, tp); 3619 tcpstat.tcps_sndacks++; 3620 tcpstat.tcps_sndtotal++; 3621 } else { 3622 SYN_CACHE_PUT(sc); 3623 tcpstat.tcps_sc_dropped++; 3624 } 3625 return (1); 3626 } 3627 3628 int 3629 syn_cache_respond(sc, m) 3630 struct syn_cache *sc; 3631 struct mbuf *m; 3632 { 3633 struct route *ro; 3634 u_int8_t *optp; 3635 int optlen, error; 3636 u_int16_t tlen; 3637 struct ip *ip = NULL; 3638 #ifdef INET6 3639 struct ip6_hdr *ip6 = NULL; 3640 #endif 3641 struct tcpcb *tp; 3642 struct tcphdr *th; 3643 u_int hlen; 3644 struct socket *so; 3645 3646 switch (sc->sc_src.sa.sa_family) { 3647 case AF_INET: 3648 hlen = sizeof(struct ip); 3649 ro = &sc->sc_route4; 3650 break; 3651 #ifdef INET6 3652 case AF_INET6: 3653 hlen = sizeof(struct ip6_hdr); 3654 ro = (struct route *)&sc->sc_route6; 3655 break; 3656 #endif 3657 default: 3658 if (m) 3659 m_freem(m); 3660 return EAFNOSUPPORT; 3661 } 3662 3663 /* Compute the size of the TCP options. */ 3664 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3665 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3666 3667 tlen = hlen + sizeof(struct tcphdr) + optlen; 3668 3669 /* 3670 * Create the IP+TCP header from scratch. 3671 */ 3672 if (m) 3673 m_freem(m); 3674 #ifdef DIAGNOSTIC 3675 if (max_linkhdr + tlen > MCLBYTES) 3676 return (ENOBUFS); 3677 #endif 3678 MGETHDR(m, M_DONTWAIT, MT_DATA); 3679 if (m && tlen > MHLEN) { 3680 MCLGET(m, M_DONTWAIT); 3681 if ((m->m_flags & M_EXT) == 0) { 3682 m_freem(m); 3683 m = NULL; 3684 } 3685 } 3686 if (m == NULL) 3687 return (ENOBUFS); 3688 MCLAIM(m, &tcp_tx_mowner); 3689 3690 /* Fixup the mbuf. */ 3691 m->m_data += max_linkhdr; 3692 m->m_len = m->m_pkthdr.len = tlen; 3693 if (sc->sc_tp) { 3694 tp = sc->sc_tp; 3695 if (tp->t_inpcb) 3696 so = tp->t_inpcb->inp_socket; 3697 #ifdef INET6 3698 else if (tp->t_in6pcb) 3699 so = tp->t_in6pcb->in6p_socket; 3700 #endif 3701 else 3702 so = NULL; 3703 } else 3704 so = NULL; 3705 m->m_pkthdr.rcvif = NULL; 3706 memset(mtod(m, u_char *), 0, tlen); 3707 3708 switch (sc->sc_src.sa.sa_family) { 3709 case AF_INET: 3710 ip = mtod(m, struct ip *); 3711 ip->ip_dst = sc->sc_src.sin.sin_addr; 3712 ip->ip_src = sc->sc_dst.sin.sin_addr; 3713 ip->ip_p = IPPROTO_TCP; 3714 th = (struct tcphdr *)(ip + 1); 3715 th->th_dport = sc->sc_src.sin.sin_port; 3716 th->th_sport = sc->sc_dst.sin.sin_port; 3717 break; 3718 #ifdef INET6 3719 case AF_INET6: 3720 ip6 = mtod(m, struct ip6_hdr *); 3721 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3722 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3723 ip6->ip6_nxt = IPPROTO_TCP; 3724 /* ip6_plen will be updated in ip6_output() */ 3725 th = (struct tcphdr *)(ip6 + 1); 3726 th->th_dport = sc->sc_src.sin6.sin6_port; 3727 th->th_sport = sc->sc_dst.sin6.sin6_port; 3728 break; 3729 #endif 3730 default: 3731 th = NULL; 3732 } 3733 3734 th->th_seq = htonl(sc->sc_iss); 3735 th->th_ack = htonl(sc->sc_irs + 1); 3736 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3737 th->th_flags = TH_SYN|TH_ACK; 3738 th->th_win = htons(sc->sc_win); 3739 /* th_sum already 0 */ 3740 /* th_urp already 0 */ 3741 3742 /* Tack on the TCP options. */ 3743 optp = (u_int8_t *)(th + 1); 3744 *optp++ = TCPOPT_MAXSEG; 3745 *optp++ = 4; 3746 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3747 *optp++ = sc->sc_ourmaxseg & 0xff; 3748 3749 if (sc->sc_request_r_scale != 15) { 3750 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3751 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3752 sc->sc_request_r_scale); 3753 optp += 4; 3754 } 3755 3756 if (sc->sc_flags & SCF_TIMESTAMP) { 3757 u_int32_t *lp = (u_int32_t *)(optp); 3758 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3759 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3760 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3761 *lp = htonl(sc->sc_timestamp); 3762 optp += TCPOLEN_TSTAMP_APPA; 3763 } 3764 3765 /* Compute the packet's checksum. */ 3766 switch (sc->sc_src.sa.sa_family) { 3767 case AF_INET: 3768 ip->ip_len = htons(tlen - hlen); 3769 th->th_sum = 0; 3770 th->th_sum = in_cksum(m, tlen); 3771 break; 3772 #ifdef INET6 3773 case AF_INET6: 3774 ip6->ip6_plen = htons(tlen - hlen); 3775 th->th_sum = 0; 3776 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3777 break; 3778 #endif 3779 } 3780 3781 /* 3782 * Fill in some straggling IP bits. Note the stack expects 3783 * ip_len to be in host order, for convenience. 3784 */ 3785 switch (sc->sc_src.sa.sa_family) { 3786 #ifdef INET 3787 case AF_INET: 3788 ip->ip_len = htons(tlen); 3789 ip->ip_ttl = ip_defttl; 3790 /* XXX tos? */ 3791 break; 3792 #endif 3793 #ifdef INET6 3794 case AF_INET6: 3795 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3796 ip6->ip6_vfc |= IPV6_VERSION; 3797 ip6->ip6_plen = htons(tlen - hlen); 3798 /* ip6_hlim will be initialized afterwards */ 3799 /* XXX flowlabel? */ 3800 break; 3801 #endif 3802 } 3803 3804 /* XXX use IPsec policy on listening socket, on SYN ACK */ 3805 tp = sc->sc_tp; 3806 3807 switch (sc->sc_src.sa.sa_family) { 3808 #ifdef INET 3809 case AF_INET: 3810 error = ip_output(m, sc->sc_ipopts, ro, 3811 (ip_mtudisc ? IP_MTUDISC : 0), 3812 (struct ip_moptions *)NULL, so); 3813 break; 3814 #endif 3815 #ifdef INET6 3816 case AF_INET6: 3817 ip6->ip6_hlim = in6_selecthlim(NULL, 3818 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3819 3820 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 3821 (struct ip6_moptions *)0, so, NULL); 3822 break; 3823 #endif 3824 default: 3825 error = EAFNOSUPPORT; 3826 break; 3827 } 3828 return (error); 3829 } 3830