1 /* $NetBSD: tcp_input.c,v 1.155 2002/09/11 02:41:21 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. All advertising materials mentioning features or use of this software 122 * must display the following acknowledgement: 123 * This product includes software developed by the University of 124 * California, Berkeley and its contributors. 125 * 4. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 /* 145 * TODO list for SYN cache stuff: 146 * 147 * Find room for a "state" field, which is needed to keep a 148 * compressed state for TIME_WAIT TCBs. It's been noted already 149 * that this is fairly important for very high-volume web and 150 * mail servers, which use a large number of short-lived 151 * connections. 152 */ 153 154 #include <sys/cdefs.h> 155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.155 2002/09/11 02:41:21 itojun Exp $"); 156 157 #include "opt_inet.h" 158 #include "opt_ipsec.h" 159 #include "opt_inet_csum.h" 160 #include "opt_tcp_debug.h" 161 162 #include <sys/param.h> 163 #include <sys/systm.h> 164 #include <sys/malloc.h> 165 #include <sys/mbuf.h> 166 #include <sys/protosw.h> 167 #include <sys/socket.h> 168 #include <sys/socketvar.h> 169 #include <sys/errno.h> 170 #include <sys/syslog.h> 171 #include <sys/pool.h> 172 #include <sys/domain.h> 173 #include <sys/kernel.h> 174 175 #include <net/if.h> 176 #include <net/route.h> 177 #include <net/if_types.h> 178 179 #include <netinet/in.h> 180 #include <netinet/in_systm.h> 181 #include <netinet/ip.h> 182 #include <netinet/in_pcb.h> 183 #include <netinet/ip_var.h> 184 185 #ifdef INET6 186 #ifndef INET 187 #include <netinet/in.h> 188 #endif 189 #include <netinet/ip6.h> 190 #include <netinet6/ip6_var.h> 191 #include <netinet6/in6_pcb.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_var.h> 194 #include <netinet/icmp6.h> 195 #include <netinet6/nd6.h> 196 #endif 197 198 #ifdef PULLDOWN_TEST 199 #ifndef INET6 200 /* always need ip6.h for IP6_EXTHDR_GET */ 201 #include <netinet/ip6.h> 202 #endif 203 #endif 204 205 #include <netinet/tcp.h> 206 #include <netinet/tcp_fsm.h> 207 #include <netinet/tcp_seq.h> 208 #include <netinet/tcp_timer.h> 209 #include <netinet/tcp_var.h> 210 #include <netinet/tcpip.h> 211 #include <netinet/tcp_debug.h> 212 213 #include <machine/stdarg.h> 214 215 #ifdef IPSEC 216 #include <netinet6/ipsec.h> 217 #include <netkey/key.h> 218 #endif /*IPSEC*/ 219 #ifdef INET6 220 #include "faith.h" 221 #if defined(NFAITH) && NFAITH > 0 222 #include <net/if_faith.h> 223 #endif 224 #endif 225 226 int tcprexmtthresh = 3; 227 int tcp_log_refused; 228 229 static int tcp_rst_ppslim_count = 0; 230 static struct timeval tcp_rst_ppslim_last; 231 232 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 233 234 /* for modulo comparisons of timestamps */ 235 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 236 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 237 238 /* 239 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 240 */ 241 #ifdef INET6 242 #define ND6_HINT(tp) \ 243 do { \ 244 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 245 && tp->t_in6pcb->in6p_route.ro_rt) { \ 246 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 247 } \ 248 } while (0) 249 #else 250 #define ND6_HINT(tp) 251 #endif 252 253 /* 254 * Macro to compute ACK transmission behavior. Delay the ACK unless 255 * we have already delayed an ACK (must send an ACK every two segments). 256 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 257 * option is enabled. 258 */ 259 #define TCP_SETUP_ACK(tp, th) \ 260 do { \ 261 if ((tp)->t_flags & TF_DELACK || \ 262 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 263 tp->t_flags |= TF_ACKNOW; \ 264 else \ 265 TCP_SET_DELACK(tp); \ 266 } while (0) 267 268 /* 269 * Convert TCP protocol fields to host order for easier processing. 270 */ 271 #define TCP_FIELDS_TO_HOST(th) \ 272 do { \ 273 NTOHL((th)->th_seq); \ 274 NTOHL((th)->th_ack); \ 275 NTOHS((th)->th_win); \ 276 NTOHS((th)->th_urp); \ 277 } while (0) 278 279 /* 280 * ... and reverse the above. 281 */ 282 #define TCP_FIELDS_TO_NET(th) \ 283 do { \ 284 HTONL((th)->th_seq); \ 285 HTONL((th)->th_ack); \ 286 HTONS((th)->th_win); \ 287 HTONS((th)->th_urp); \ 288 } while (0) 289 290 #ifdef TCP_CSUM_COUNTERS 291 #include <sys/device.h> 292 293 extern struct evcnt tcp_hwcsum_ok; 294 extern struct evcnt tcp_hwcsum_bad; 295 extern struct evcnt tcp_hwcsum_data; 296 extern struct evcnt tcp_swcsum; 297 298 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 299 300 #else 301 302 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 303 304 #endif /* TCP_CSUM_COUNTERS */ 305 306 #ifdef TCP_REASS_COUNTERS 307 #include <sys/device.h> 308 309 extern struct evcnt tcp_reass_; 310 extern struct evcnt tcp_reass_empty; 311 extern struct evcnt tcp_reass_iteration[8]; 312 extern struct evcnt tcp_reass_prependfirst; 313 extern struct evcnt tcp_reass_prepend; 314 extern struct evcnt tcp_reass_insert; 315 extern struct evcnt tcp_reass_inserttail; 316 extern struct evcnt tcp_reass_append; 317 extern struct evcnt tcp_reass_appendtail; 318 extern struct evcnt tcp_reass_overlaptail; 319 extern struct evcnt tcp_reass_overlapfront; 320 extern struct evcnt tcp_reass_segdup; 321 extern struct evcnt tcp_reass_fragdup; 322 323 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 324 325 #else 326 327 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 328 329 #endif /* TCP_REASS_COUNTERS */ 330 331 #ifdef INET 332 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 333 #endif 334 #ifdef INET6 335 static void tcp6_log_refused 336 __P((const struct ip6_hdr *, const struct tcphdr *)); 337 #endif 338 339 int 340 tcp_reass(tp, th, m, tlen) 341 struct tcpcb *tp; 342 struct tcphdr *th; 343 struct mbuf *m; 344 int *tlen; 345 { 346 struct ipqent *p, *q, *nq, *tiqe = NULL; 347 struct socket *so = NULL; 348 int pkt_flags; 349 tcp_seq pkt_seq; 350 unsigned pkt_len; 351 u_long rcvpartdupbyte = 0; 352 u_long rcvoobyte; 353 #ifdef TCP_REASS_COUNTERS 354 u_int count = 0; 355 #endif 356 357 if (tp->t_inpcb) 358 so = tp->t_inpcb->inp_socket; 359 #ifdef INET6 360 else if (tp->t_in6pcb) 361 so = tp->t_in6pcb->in6p_socket; 362 #endif 363 364 TCP_REASS_LOCK_CHECK(tp); 365 366 /* 367 * Call with th==0 after become established to 368 * force pre-ESTABLISHED data up to user socket. 369 */ 370 if (th == 0) 371 goto present; 372 373 rcvoobyte = *tlen; 374 /* 375 * Copy these to local variables because the tcpiphdr 376 * gets munged while we are collapsing mbufs. 377 */ 378 pkt_seq = th->th_seq; 379 pkt_len = *tlen; 380 pkt_flags = th->th_flags; 381 382 TCP_REASS_COUNTER_INCR(&tcp_reass_); 383 384 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 385 /* 386 * When we miss a packet, the vast majority of time we get 387 * packets that follow it in order. So optimize for that. 388 */ 389 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 390 p->ipqe_len += pkt_len; 391 p->ipqe_flags |= pkt_flags; 392 m_cat(p->ipqe_m, m); 393 tiqe = p; 394 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 395 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 396 goto skip_replacement; 397 } 398 /* 399 * While we're here, if the pkt is completely beyond 400 * anything we have, just insert it at the tail. 401 */ 402 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 403 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 404 goto insert_it; 405 } 406 } 407 408 q = TAILQ_FIRST(&tp->segq); 409 410 if (q != NULL) { 411 /* 412 * If this segment immediately precedes the first out-of-order 413 * block, simply slap the segment in front of it and (mostly) 414 * skip the complicated logic. 415 */ 416 if (pkt_seq + pkt_len == q->ipqe_seq) { 417 q->ipqe_seq = pkt_seq; 418 q->ipqe_len += pkt_len; 419 q->ipqe_flags |= pkt_flags; 420 m_cat(m, q->ipqe_m); 421 q->ipqe_m = m; 422 tiqe = q; 423 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 424 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 425 goto skip_replacement; 426 } 427 } else { 428 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 429 } 430 431 /* 432 * Find a segment which begins after this one does. 433 */ 434 for (p = NULL; q != NULL; q = nq) { 435 nq = TAILQ_NEXT(q, ipqe_q); 436 #ifdef TCP_REASS_COUNTERS 437 count++; 438 #endif 439 /* 440 * If the received segment is just right after this 441 * fragment, merge the two together and then check 442 * for further overlaps. 443 */ 444 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 445 #ifdef TCPREASS_DEBUG 446 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 447 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 448 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 449 #endif 450 pkt_len += q->ipqe_len; 451 pkt_flags |= q->ipqe_flags; 452 pkt_seq = q->ipqe_seq; 453 m_cat(q->ipqe_m, m); 454 m = q->ipqe_m; 455 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 456 goto free_ipqe; 457 } 458 /* 459 * If the received segment is completely past this 460 * fragment, we need to go the next fragment. 461 */ 462 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 463 p = q; 464 continue; 465 } 466 /* 467 * If the fragment is past the received segment, 468 * it (or any following) can't be concatenated. 469 */ 470 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 471 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 472 break; 473 } 474 475 /* 476 * We've received all the data in this segment before. 477 * mark it as a duplicate and return. 478 */ 479 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 480 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 481 tcpstat.tcps_rcvduppack++; 482 tcpstat.tcps_rcvdupbyte += pkt_len; 483 m_freem(m); 484 if (tiqe != NULL) 485 pool_put(&ipqent_pool, tiqe); 486 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 487 return (0); 488 } 489 /* 490 * Received segment completely overlaps this fragment 491 * so we drop the fragment (this keeps the temporal 492 * ordering of segments correct). 493 */ 494 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 495 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 496 rcvpartdupbyte += q->ipqe_len; 497 m_freem(q->ipqe_m); 498 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 499 goto free_ipqe; 500 } 501 /* 502 * RX'ed segment extends past the end of the 503 * fragment. Drop the overlapping bytes. Then 504 * merge the fragment and segment then treat as 505 * a longer received packet. 506 */ 507 if (SEQ_LT(q->ipqe_seq, pkt_seq) 508 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 509 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 510 #ifdef TCPREASS_DEBUG 511 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 512 tp, overlap, 513 pkt_seq, pkt_seq + pkt_len, pkt_len); 514 #endif 515 m_adj(m, overlap); 516 rcvpartdupbyte += overlap; 517 m_cat(q->ipqe_m, m); 518 m = q->ipqe_m; 519 pkt_seq = q->ipqe_seq; 520 pkt_len += q->ipqe_len - overlap; 521 rcvoobyte -= overlap; 522 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 523 goto free_ipqe; 524 } 525 /* 526 * RX'ed segment extends past the front of the 527 * fragment. Drop the overlapping bytes on the 528 * received packet. The packet will then be 529 * contatentated with this fragment a bit later. 530 */ 531 if (SEQ_GT(q->ipqe_seq, pkt_seq) 532 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 533 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 534 #ifdef TCPREASS_DEBUG 535 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 536 tp, overlap, 537 pkt_seq, pkt_seq + pkt_len, pkt_len); 538 #endif 539 m_adj(m, -overlap); 540 pkt_len -= overlap; 541 rcvpartdupbyte += overlap; 542 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 543 rcvoobyte -= overlap; 544 } 545 /* 546 * If the received segment immediates precedes this 547 * fragment then tack the fragment onto this segment 548 * and reinsert the data. 549 */ 550 if (q->ipqe_seq == pkt_seq + pkt_len) { 551 #ifdef TCPREASS_DEBUG 552 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 553 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 554 pkt_seq, pkt_seq + pkt_len, pkt_len); 555 #endif 556 pkt_len += q->ipqe_len; 557 pkt_flags |= q->ipqe_flags; 558 m_cat(m, q->ipqe_m); 559 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 560 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 561 if (tiqe == NULL) { 562 tiqe = q; 563 } else { 564 pool_put(&ipqent_pool, q); 565 } 566 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 567 break; 568 } 569 /* 570 * If the fragment is before the segment, remember it. 571 * When this loop is terminated, p will contain the 572 * pointer to fragment that is right before the received 573 * segment. 574 */ 575 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 576 p = q; 577 578 continue; 579 580 /* 581 * This is a common operation. It also will allow 582 * to save doing a malloc/free in most instances. 583 */ 584 free_ipqe: 585 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 586 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 587 if (tiqe == NULL) { 588 tiqe = q; 589 } else { 590 pool_put(&ipqent_pool, q); 591 } 592 } 593 594 #ifdef TCP_REASS_COUNTERS 595 if (count > 7) 596 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 597 else if (count > 0) 598 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 599 #endif 600 601 insert_it: 602 603 /* 604 * Allocate a new queue entry since the received segment did not 605 * collapse onto any other out-of-order block; thus we are allocating 606 * a new block. If it had collapsed, tiqe would not be NULL and 607 * we would be reusing it. 608 * XXX If we can't, just drop the packet. XXX 609 */ 610 if (tiqe == NULL) { 611 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 612 if (tiqe == NULL) { 613 tcpstat.tcps_rcvmemdrop++; 614 m_freem(m); 615 return (0); 616 } 617 } 618 619 /* 620 * Update the counters. 621 */ 622 tcpstat.tcps_rcvoopack++; 623 tcpstat.tcps_rcvoobyte += rcvoobyte; 624 if (rcvpartdupbyte) { 625 tcpstat.tcps_rcvpartduppack++; 626 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 627 } 628 629 /* 630 * Insert the new fragment queue entry into both queues. 631 */ 632 tiqe->ipqe_m = m; 633 tiqe->ipqe_seq = pkt_seq; 634 tiqe->ipqe_len = pkt_len; 635 tiqe->ipqe_flags = pkt_flags; 636 if (p == NULL) { 637 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 638 #ifdef TCPREASS_DEBUG 639 if (tiqe->ipqe_seq != tp->rcv_nxt) 640 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 641 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 642 #endif 643 } else { 644 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 645 #ifdef TCPREASS_DEBUG 646 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 647 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 648 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 649 #endif 650 } 651 652 skip_replacement: 653 654 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 655 656 present: 657 /* 658 * Present data to user, advancing rcv_nxt through 659 * completed sequence space. 660 */ 661 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 662 return (0); 663 q = TAILQ_FIRST(&tp->segq); 664 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 665 return (0); 666 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 667 return (0); 668 669 tp->rcv_nxt += q->ipqe_len; 670 pkt_flags = q->ipqe_flags & TH_FIN; 671 ND6_HINT(tp); 672 673 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 674 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 675 if (so->so_state & SS_CANTRCVMORE) 676 m_freem(q->ipqe_m); 677 else 678 sbappendstream(&so->so_rcv, q->ipqe_m); 679 pool_put(&ipqent_pool, q); 680 sorwakeup(so); 681 return (pkt_flags); 682 } 683 684 #ifdef INET6 685 int 686 tcp6_input(mp, offp, proto) 687 struct mbuf **mp; 688 int *offp, proto; 689 { 690 struct mbuf *m = *mp; 691 692 /* 693 * draft-itojun-ipv6-tcp-to-anycast 694 * better place to put this in? 695 */ 696 if (m->m_flags & M_ANYCAST6) { 697 struct ip6_hdr *ip6; 698 if (m->m_len < sizeof(struct ip6_hdr)) { 699 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 700 tcpstat.tcps_rcvshort++; 701 return IPPROTO_DONE; 702 } 703 } 704 ip6 = mtod(m, struct ip6_hdr *); 705 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 706 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 707 return IPPROTO_DONE; 708 } 709 710 tcp_input(m, *offp, proto); 711 return IPPROTO_DONE; 712 } 713 #endif 714 715 #ifdef INET 716 static void 717 tcp4_log_refused(ip, th) 718 const struct ip *ip; 719 const struct tcphdr *th; 720 { 721 char src[4*sizeof "123"]; 722 char dst[4*sizeof "123"]; 723 724 if (ip) { 725 strcpy(src, inet_ntoa(ip->ip_src)); 726 strcpy(dst, inet_ntoa(ip->ip_dst)); 727 } 728 else { 729 strcpy(src, "(unknown)"); 730 strcpy(dst, "(unknown)"); 731 } 732 log(LOG_INFO, 733 "Connection attempt to TCP %s:%d from %s:%d\n", 734 dst, ntohs(th->th_dport), 735 src, ntohs(th->th_sport)); 736 } 737 #endif 738 739 #ifdef INET6 740 static void 741 tcp6_log_refused(ip6, th) 742 const struct ip6_hdr *ip6; 743 const struct tcphdr *th; 744 { 745 char src[INET6_ADDRSTRLEN]; 746 char dst[INET6_ADDRSTRLEN]; 747 748 if (ip6) { 749 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 750 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 751 } 752 else { 753 strcpy(src, "(unknown v6)"); 754 strcpy(dst, "(unknown v6)"); 755 } 756 log(LOG_INFO, 757 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 758 dst, ntohs(th->th_dport), 759 src, ntohs(th->th_sport)); 760 } 761 #endif 762 763 /* 764 * TCP input routine, follows pages 65-76 of the 765 * protocol specification dated September, 1981 very closely. 766 */ 767 void 768 #if __STDC__ 769 tcp_input(struct mbuf *m, ...) 770 #else 771 tcp_input(m, va_alist) 772 struct mbuf *m; 773 #endif 774 { 775 int proto; 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 short ostate = 0; 791 int iss = 0; 792 u_long tiwin; 793 struct tcp_opt_info opti; 794 int off, iphlen; 795 va_list ap; 796 int af; /* af on the wire */ 797 struct mbuf *tcp_saveti = NULL; 798 799 va_start(ap, m); 800 toff = va_arg(ap, int); 801 proto = va_arg(ap, int); 802 va_end(ap); 803 804 tcpstat.tcps_rcvtotal++; 805 806 bzero(&opti, sizeof(opti)); 807 opti.ts_present = 0; 808 opti.maxseg = 0; 809 810 /* 811 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 812 * 813 * TCP is, by definition, unicast, so we reject all 814 * multicast outright. 815 * 816 * Note, there are additional src/dst address checks in 817 * the AF-specific code below. 818 */ 819 if (m->m_flags & (M_BCAST|M_MCAST)) { 820 /* XXX stat */ 821 goto drop; 822 } 823 #ifdef INET6 824 if (m->m_flags & M_ANYCAST6) { 825 /* XXX stat */ 826 goto drop; 827 } 828 #endif 829 830 /* 831 * Get IP and TCP header together in first mbuf. 832 * Note: IP leaves IP header in first mbuf. 833 */ 834 ip = mtod(m, struct ip *); 835 #ifdef INET6 836 ip6 = NULL; 837 #endif 838 switch (ip->ip_v) { 839 #ifdef INET 840 case 4: 841 af = AF_INET; 842 iphlen = sizeof(struct ip); 843 #ifndef PULLDOWN_TEST 844 /* would like to get rid of this... */ 845 if (toff > sizeof (struct ip)) { 846 ip_stripoptions(m, (struct mbuf *)0); 847 toff = sizeof(struct ip); 848 } 849 if (m->m_len < toff + sizeof (struct tcphdr)) { 850 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 851 tcpstat.tcps_rcvshort++; 852 return; 853 } 854 } 855 ip = mtod(m, struct ip *); 856 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 857 #else 858 ip = mtod(m, struct ip *); 859 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 860 sizeof(struct tcphdr)); 861 if (th == NULL) { 862 tcpstat.tcps_rcvshort++; 863 return; 864 } 865 #endif 866 /* We do the checksum after PCB lookup... */ 867 len = ntohs(ip->ip_len); 868 tlen = len - toff; 869 break; 870 #endif 871 #ifdef INET6 872 case 6: 873 ip = NULL; 874 iphlen = sizeof(struct ip6_hdr); 875 af = AF_INET6; 876 #ifndef PULLDOWN_TEST 877 if (m->m_len < toff + sizeof(struct tcphdr)) { 878 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 879 if (m == NULL) { 880 tcpstat.tcps_rcvshort++; 881 return; 882 } 883 } 884 ip6 = mtod(m, struct ip6_hdr *); 885 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 886 #else 887 ip6 = mtod(m, struct ip6_hdr *); 888 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 889 sizeof(struct tcphdr)); 890 if (th == NULL) { 891 tcpstat.tcps_rcvshort++; 892 return; 893 } 894 #endif 895 896 /* Be proactive about malicious use of IPv4 mapped address */ 897 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 898 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 899 /* XXX stat */ 900 goto drop; 901 } 902 903 /* 904 * Be proactive about unspecified IPv6 address in source. 905 * As we use all-zero to indicate unbounded/unconnected pcb, 906 * unspecified IPv6 address can be used to confuse us. 907 * 908 * Note that packets with unspecified IPv6 destination is 909 * already dropped in ip6_input. 910 */ 911 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 912 /* XXX stat */ 913 goto drop; 914 } 915 916 /* 917 * Make sure destination address is not multicast. 918 * Source address checked in ip6_input(). 919 */ 920 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 921 /* XXX stat */ 922 goto drop; 923 } 924 925 /* We do the checksum after PCB lookup... */ 926 len = m->m_pkthdr.len; 927 tlen = len - toff; 928 break; 929 #endif 930 default: 931 m_freem(m); 932 return; 933 } 934 935 KASSERT(TCP_HDR_ALIGNED_P(th)); 936 937 /* 938 * Check that TCP offset makes sense, 939 * pull out TCP options and adjust length. XXX 940 */ 941 off = th->th_off << 2; 942 if (off < sizeof (struct tcphdr) || off > tlen) { 943 tcpstat.tcps_rcvbadoff++; 944 goto drop; 945 } 946 tlen -= off; 947 948 /* 949 * tcp_input() has been modified to use tlen to mean the TCP data 950 * length throughout the function. Other functions can use 951 * m->m_pkthdr.len as the basis for calculating the TCP data length. 952 * rja 953 */ 954 955 if (off > sizeof (struct tcphdr)) { 956 #ifndef PULLDOWN_TEST 957 if (m->m_len < toff + off) { 958 if ((m = m_pullup(m, toff + off)) == 0) { 959 tcpstat.tcps_rcvshort++; 960 return; 961 } 962 switch (af) { 963 #ifdef INET 964 case AF_INET: 965 ip = mtod(m, struct ip *); 966 break; 967 #endif 968 #ifdef INET6 969 case AF_INET6: 970 ip6 = mtod(m, struct ip6_hdr *); 971 break; 972 #endif 973 } 974 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 975 } 976 #else 977 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 978 if (th == NULL) { 979 tcpstat.tcps_rcvshort++; 980 return; 981 } 982 /* 983 * NOTE: ip/ip6 will not be affected by m_pulldown() 984 * (as they're before toff) and we don't need to update those. 985 */ 986 #endif 987 KASSERT(TCP_HDR_ALIGNED_P(th)); 988 optlen = off - sizeof (struct tcphdr); 989 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 990 /* 991 * Do quick retrieval of timestamp options ("options 992 * prediction?"). If timestamp is the only option and it's 993 * formatted as recommended in RFC 1323 appendix A, we 994 * quickly get the values now and not bother calling 995 * tcp_dooptions(), etc. 996 */ 997 if ((optlen == TCPOLEN_TSTAMP_APPA || 998 (optlen > TCPOLEN_TSTAMP_APPA && 999 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1000 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1001 (th->th_flags & TH_SYN) == 0) { 1002 opti.ts_present = 1; 1003 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1004 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1005 optp = NULL; /* we've parsed the options */ 1006 } 1007 } 1008 tiflags = th->th_flags; 1009 1010 /* 1011 * Locate pcb for segment. 1012 */ 1013 findpcb: 1014 inp = NULL; 1015 #ifdef INET6 1016 in6p = NULL; 1017 #endif 1018 switch (af) { 1019 #ifdef INET 1020 case AF_INET: 1021 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1022 ip->ip_dst, th->th_dport); 1023 if (inp == 0) { 1024 ++tcpstat.tcps_pcbhashmiss; 1025 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1026 } 1027 #ifdef INET6 1028 if (inp == 0) { 1029 struct in6_addr s, d; 1030 1031 /* mapped addr case */ 1032 bzero(&s, sizeof(s)); 1033 s.s6_addr16[5] = htons(0xffff); 1034 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1035 bzero(&d, sizeof(d)); 1036 d.s6_addr16[5] = htons(0xffff); 1037 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1038 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 1039 &d, th->th_dport, 0); 1040 if (in6p == 0) { 1041 ++tcpstat.tcps_pcbhashmiss; 1042 in6p = in6_pcblookup_bind(&tcb6, &d, 1043 th->th_dport, 0); 1044 } 1045 } 1046 #endif 1047 #ifndef INET6 1048 if (inp == 0) 1049 #else 1050 if (inp == 0 && in6p == 0) 1051 #endif 1052 { 1053 ++tcpstat.tcps_noport; 1054 if (tcp_log_refused && (tiflags & TH_SYN)) { 1055 tcp4_log_refused(ip, th); 1056 } 1057 TCP_FIELDS_TO_HOST(th); 1058 goto dropwithreset_ratelim; 1059 } 1060 #ifdef IPSEC 1061 if (inp && ipsec4_in_reject(m, inp)) { 1062 ipsecstat.in_polvio++; 1063 goto drop; 1064 } 1065 #ifdef INET6 1066 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1067 ipsecstat.in_polvio++; 1068 goto drop; 1069 } 1070 #endif 1071 #endif /*IPSEC*/ 1072 break; 1073 #endif /*INET*/ 1074 #ifdef INET6 1075 case AF_INET6: 1076 { 1077 int faith; 1078 1079 #if defined(NFAITH) && NFAITH > 0 1080 faith = faithprefix(&ip6->ip6_dst); 1081 #else 1082 faith = 0; 1083 #endif 1084 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 1085 &ip6->ip6_dst, th->th_dport, faith); 1086 if (in6p == NULL) { 1087 ++tcpstat.tcps_pcbhashmiss; 1088 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 1089 th->th_dport, faith); 1090 } 1091 if (in6p == NULL) { 1092 ++tcpstat.tcps_noport; 1093 if (tcp_log_refused && (tiflags & TH_SYN)) { 1094 tcp6_log_refused(ip6, th); 1095 } 1096 TCP_FIELDS_TO_HOST(th); 1097 goto dropwithreset_ratelim; 1098 } 1099 #ifdef IPSEC 1100 if (ipsec6_in_reject(m, in6p)) { 1101 ipsec6stat.in_polvio++; 1102 goto drop; 1103 } 1104 #endif /*IPSEC*/ 1105 break; 1106 } 1107 #endif 1108 } 1109 1110 /* 1111 * If the state is CLOSED (i.e., TCB does not exist) then 1112 * all data in the incoming segment is discarded. 1113 * If the TCB exists but is in CLOSED state, it is embryonic, 1114 * but should either do a listen or a connect soon. 1115 */ 1116 tp = NULL; 1117 so = NULL; 1118 if (inp) { 1119 tp = intotcpcb(inp); 1120 so = inp->inp_socket; 1121 } 1122 #ifdef INET6 1123 else if (in6p) { 1124 tp = in6totcpcb(in6p); 1125 so = in6p->in6p_socket; 1126 } 1127 #endif 1128 if (tp == 0) { 1129 TCP_FIELDS_TO_HOST(th); 1130 goto dropwithreset_ratelim; 1131 } 1132 if (tp->t_state == TCPS_CLOSED) 1133 goto drop; 1134 1135 /* 1136 * Checksum extended TCP header and data. 1137 */ 1138 switch (af) { 1139 #ifdef INET 1140 case AF_INET: 1141 switch (m->m_pkthdr.csum_flags & 1142 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1143 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1144 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1145 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1146 goto badcsum; 1147 1148 case M_CSUM_TCPv4|M_CSUM_DATA: 1149 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1150 if ((m->m_pkthdr.csum_data ^ 0xffff) != 0) 1151 goto badcsum; 1152 break; 1153 1154 case M_CSUM_TCPv4: 1155 /* Checksum was okay. */ 1156 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1157 break; 1158 1159 default: 1160 /* Must compute it ourselves. */ 1161 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1162 #ifndef PULLDOWN_TEST 1163 { 1164 struct ipovly *ipov; 1165 ipov = (struct ipovly *)ip; 1166 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 1167 ipov->ih_len = htons(tlen + off); 1168 1169 if (in_cksum(m, len) != 0) 1170 goto badcsum; 1171 } 1172 #else 1173 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1174 goto badcsum; 1175 #endif /* ! PULLDOWN_TEST */ 1176 break; 1177 } 1178 break; 1179 #endif /* INET4 */ 1180 1181 #ifdef INET6 1182 case AF_INET6: 1183 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1184 goto badcsum; 1185 break; 1186 #endif /* INET6 */ 1187 } 1188 1189 TCP_FIELDS_TO_HOST(th); 1190 1191 /* Unscale the window into a 32-bit value. */ 1192 if ((tiflags & TH_SYN) == 0) 1193 tiwin = th->th_win << tp->snd_scale; 1194 else 1195 tiwin = th->th_win; 1196 1197 #ifdef INET6 1198 /* save packet options if user wanted */ 1199 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1200 if (in6p->in6p_options) { 1201 m_freem(in6p->in6p_options); 1202 in6p->in6p_options = 0; 1203 } 1204 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1205 } 1206 #endif 1207 1208 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1209 union syn_cache_sa src; 1210 union syn_cache_sa dst; 1211 1212 bzero(&src, sizeof(src)); 1213 bzero(&dst, sizeof(dst)); 1214 switch (af) { 1215 #ifdef INET 1216 case AF_INET: 1217 src.sin.sin_len = sizeof(struct sockaddr_in); 1218 src.sin.sin_family = AF_INET; 1219 src.sin.sin_addr = ip->ip_src; 1220 src.sin.sin_port = th->th_sport; 1221 1222 dst.sin.sin_len = sizeof(struct sockaddr_in); 1223 dst.sin.sin_family = AF_INET; 1224 dst.sin.sin_addr = ip->ip_dst; 1225 dst.sin.sin_port = th->th_dport; 1226 break; 1227 #endif 1228 #ifdef INET6 1229 case AF_INET6: 1230 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1231 src.sin6.sin6_family = AF_INET6; 1232 src.sin6.sin6_addr = ip6->ip6_src; 1233 src.sin6.sin6_port = th->th_sport; 1234 1235 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1236 dst.sin6.sin6_family = AF_INET6; 1237 dst.sin6.sin6_addr = ip6->ip6_dst; 1238 dst.sin6.sin6_port = th->th_dport; 1239 break; 1240 #endif /* INET6 */ 1241 default: 1242 goto badsyn; /*sanity*/ 1243 } 1244 1245 if (so->so_options & SO_DEBUG) { 1246 ostate = tp->t_state; 1247 1248 tcp_saveti = NULL; 1249 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1250 goto nosave; 1251 1252 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1253 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1254 if (!tcp_saveti) 1255 goto nosave; 1256 } else { 1257 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1258 if (!tcp_saveti) 1259 goto nosave; 1260 tcp_saveti->m_len = iphlen; 1261 m_copydata(m, 0, iphlen, 1262 mtod(tcp_saveti, caddr_t)); 1263 } 1264 1265 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1266 m_freem(tcp_saveti); 1267 tcp_saveti = NULL; 1268 } else { 1269 tcp_saveti->m_len += sizeof(struct tcphdr); 1270 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1271 sizeof(struct tcphdr)); 1272 } 1273 if (tcp_saveti) { 1274 /* 1275 * need to recover version # field, which was 1276 * overwritten on ip_cksum computation. 1277 */ 1278 struct ip *sip; 1279 sip = mtod(tcp_saveti, struct ip *); 1280 switch (af) { 1281 #ifdef INET 1282 case AF_INET: 1283 sip->ip_v = 4; 1284 break; 1285 #endif 1286 #ifdef INET6 1287 case AF_INET6: 1288 sip->ip_v = 6; 1289 break; 1290 #endif 1291 } 1292 } 1293 nosave:; 1294 } 1295 if (so->so_options & SO_ACCEPTCONN) { 1296 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1297 if (tiflags & TH_RST) { 1298 syn_cache_reset(&src.sa, &dst.sa, th); 1299 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1300 (TH_ACK|TH_SYN)) { 1301 /* 1302 * Received a SYN,ACK. This should 1303 * never happen while we are in 1304 * LISTEN. Send an RST. 1305 */ 1306 goto badsyn; 1307 } else if (tiflags & TH_ACK) { 1308 so = syn_cache_get(&src.sa, &dst.sa, 1309 th, toff, tlen, so, m); 1310 if (so == NULL) { 1311 /* 1312 * We don't have a SYN for 1313 * this ACK; send an RST. 1314 */ 1315 goto badsyn; 1316 } else if (so == 1317 (struct socket *)(-1)) { 1318 /* 1319 * We were unable to create 1320 * the connection. If the 1321 * 3-way handshake was 1322 * completed, and RST has 1323 * been sent to the peer. 1324 * Since the mbuf might be 1325 * in use for the reply, 1326 * do not free it. 1327 */ 1328 m = NULL; 1329 } else { 1330 /* 1331 * We have created a 1332 * full-blown connection. 1333 */ 1334 tp = NULL; 1335 inp = NULL; 1336 #ifdef INET6 1337 in6p = NULL; 1338 #endif 1339 switch (so->so_proto->pr_domain->dom_family) { 1340 #ifdef INET 1341 case AF_INET: 1342 inp = sotoinpcb(so); 1343 tp = intotcpcb(inp); 1344 break; 1345 #endif 1346 #ifdef INET6 1347 case AF_INET6: 1348 in6p = sotoin6pcb(so); 1349 tp = in6totcpcb(in6p); 1350 break; 1351 #endif 1352 } 1353 if (tp == NULL) 1354 goto badsyn; /*XXX*/ 1355 tiwin <<= tp->snd_scale; 1356 goto after_listen; 1357 } 1358 } else { 1359 /* 1360 * None of RST, SYN or ACK was set. 1361 * This is an invalid packet for a 1362 * TCB in LISTEN state. Send a RST. 1363 */ 1364 goto badsyn; 1365 } 1366 } else { 1367 /* 1368 * Received a SYN. 1369 */ 1370 1371 #ifdef INET6 1372 /* 1373 * If deprecated address is forbidden, we do 1374 * not accept SYN to deprecated interface 1375 * address to prevent any new inbound 1376 * connection from getting established. 1377 * When we do not accept SYN, we send a TCP 1378 * RST, with deprecated source address (instead 1379 * of dropping it). We compromise it as it is 1380 * much better for peer to send a RST, and 1381 * RST will be the final packet for the 1382 * exchange. 1383 * 1384 * If we do not forbid deprecated addresses, we 1385 * accept the SYN packet. RFC2462 does not 1386 * suggest dropping SYN in this case. 1387 * If we decipher RFC2462 5.5.4, it says like 1388 * this: 1389 * 1. use of deprecated addr with existing 1390 * communication is okay - "SHOULD continue 1391 * to be used" 1392 * 2. use of it with new communication: 1393 * (2a) "SHOULD NOT be used if alternate 1394 * address with sufficient scope is 1395 * available" 1396 * (2b) nothing mentioned otherwise. 1397 * Here we fall into (2b) case as we have no 1398 * choice in our source address selection - we 1399 * must obey the peer. 1400 * 1401 * The wording in RFC2462 is confusing, and 1402 * there are multiple description text for 1403 * deprecated address handling - worse, they 1404 * are not exactly the same. I believe 5.5.4 1405 * is the best one, so we follow 5.5.4. 1406 */ 1407 if (af == AF_INET6 && !ip6_use_deprecated) { 1408 struct in6_ifaddr *ia6; 1409 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1410 &ip6->ip6_dst)) && 1411 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1412 tp = NULL; 1413 goto dropwithreset; 1414 } 1415 } 1416 #endif 1417 1418 /* 1419 * LISTEN socket received a SYN 1420 * from itself? This can't possibly 1421 * be valid; drop the packet. 1422 */ 1423 if (th->th_sport == th->th_dport) { 1424 int i; 1425 1426 switch (af) { 1427 #ifdef INET 1428 case AF_INET: 1429 i = in_hosteq(ip->ip_src, ip->ip_dst); 1430 break; 1431 #endif 1432 #ifdef INET6 1433 case AF_INET6: 1434 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1435 break; 1436 #endif 1437 default: 1438 i = 1; 1439 } 1440 if (i) { 1441 tcpstat.tcps_badsyn++; 1442 goto drop; 1443 } 1444 } 1445 1446 /* 1447 * SYN looks ok; create compressed TCP 1448 * state for it. 1449 */ 1450 if (so->so_qlen <= so->so_qlimit && 1451 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1452 so, m, optp, optlen, &opti)) 1453 m = NULL; 1454 } 1455 goto drop; 1456 } 1457 } 1458 1459 after_listen: 1460 #ifdef DIAGNOSTIC 1461 /* 1462 * Should not happen now that all embryonic connections 1463 * are handled with compressed state. 1464 */ 1465 if (tp->t_state == TCPS_LISTEN) 1466 panic("tcp_input: TCPS_LISTEN"); 1467 #endif 1468 1469 /* 1470 * Segment received on connection. 1471 * Reset idle time and keep-alive timer. 1472 */ 1473 tp->t_rcvtime = tcp_now; 1474 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1475 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1476 1477 /* 1478 * Process options. 1479 */ 1480 if (optp) 1481 tcp_dooptions(tp, optp, optlen, th, &opti); 1482 1483 /* 1484 * Header prediction: check for the two common cases 1485 * of a uni-directional data xfer. If the packet has 1486 * no control flags, is in-sequence, the window didn't 1487 * change and we're not retransmitting, it's a 1488 * candidate. If the length is zero and the ack moved 1489 * forward, we're the sender side of the xfer. Just 1490 * free the data acked & wake any higher level process 1491 * that was blocked waiting for space. If the length 1492 * is non-zero and the ack didn't move, we're the 1493 * receiver side. If we're getting packets in-order 1494 * (the reassembly queue is empty), add the data to 1495 * the socket buffer and note that we need a delayed ack. 1496 */ 1497 if (tp->t_state == TCPS_ESTABLISHED && 1498 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1499 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1500 th->th_seq == tp->rcv_nxt && 1501 tiwin && tiwin == tp->snd_wnd && 1502 tp->snd_nxt == tp->snd_max) { 1503 1504 /* 1505 * If last ACK falls within this segment's sequence numbers, 1506 * record the timestamp. 1507 */ 1508 if (opti.ts_present && 1509 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1510 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1511 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1512 tp->ts_recent = opti.ts_val; 1513 } 1514 1515 if (tlen == 0) { 1516 if (SEQ_GT(th->th_ack, tp->snd_una) && 1517 SEQ_LEQ(th->th_ack, tp->snd_max) && 1518 tp->snd_cwnd >= tp->snd_wnd && 1519 tp->t_dupacks < tcprexmtthresh) { 1520 /* 1521 * this is a pure ack for outstanding data. 1522 */ 1523 ++tcpstat.tcps_predack; 1524 if (opti.ts_present && opti.ts_ecr) 1525 tcp_xmit_timer(tp, 1526 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1527 else if (tp->t_rtttime && 1528 SEQ_GT(th->th_ack, tp->t_rtseq)) 1529 tcp_xmit_timer(tp, 1530 tcp_now - tp->t_rtttime); 1531 acked = th->th_ack - tp->snd_una; 1532 tcpstat.tcps_rcvackpack++; 1533 tcpstat.tcps_rcvackbyte += acked; 1534 ND6_HINT(tp); 1535 sbdrop(&so->so_snd, acked); 1536 /* 1537 * We want snd_recover to track snd_una to 1538 * avoid sequence wraparound problems for 1539 * very large transfers. 1540 */ 1541 tp->snd_una = tp->snd_recover = th->th_ack; 1542 m_freem(m); 1543 1544 /* 1545 * If all outstanding data are acked, stop 1546 * retransmit timer, otherwise restart timer 1547 * using current (possibly backed-off) value. 1548 * If process is waiting for space, 1549 * wakeup/selwakeup/signal. If data 1550 * are ready to send, let tcp_output 1551 * decide between more output or persist. 1552 */ 1553 if (tp->snd_una == tp->snd_max) 1554 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1555 else if (TCP_TIMER_ISARMED(tp, 1556 TCPT_PERSIST) == 0) 1557 TCP_TIMER_ARM(tp, TCPT_REXMT, 1558 tp->t_rxtcur); 1559 1560 sowwakeup(so); 1561 if (so->so_snd.sb_cc) 1562 (void) tcp_output(tp); 1563 if (tcp_saveti) 1564 m_freem(tcp_saveti); 1565 return; 1566 } 1567 } else if (th->th_ack == tp->snd_una && 1568 TAILQ_FIRST(&tp->segq) == NULL && 1569 tlen <= sbspace(&so->so_rcv)) { 1570 /* 1571 * this is a pure, in-sequence data packet 1572 * with nothing on the reassembly queue and 1573 * we have enough buffer space to take it. 1574 */ 1575 ++tcpstat.tcps_preddat; 1576 tp->rcv_nxt += tlen; 1577 tcpstat.tcps_rcvpack++; 1578 tcpstat.tcps_rcvbyte += tlen; 1579 ND6_HINT(tp); 1580 /* 1581 * Drop TCP, IP headers and TCP options then add data 1582 * to socket buffer. 1583 */ 1584 if (so->so_state & SS_CANTRCVMORE) 1585 m_freem(m); 1586 else { 1587 m_adj(m, toff + off); 1588 sbappendstream(&so->so_rcv, m); 1589 } 1590 sorwakeup(so); 1591 TCP_SETUP_ACK(tp, th); 1592 if (tp->t_flags & TF_ACKNOW) 1593 (void) tcp_output(tp); 1594 if (tcp_saveti) 1595 m_freem(tcp_saveti); 1596 return; 1597 } 1598 } 1599 1600 /* 1601 * Compute mbuf offset to TCP data segment. 1602 */ 1603 hdroptlen = toff + off; 1604 1605 /* 1606 * Calculate amount of space in receive window, 1607 * and then do TCP input processing. 1608 * Receive window is amount of space in rcv queue, 1609 * but not less than advertised window. 1610 */ 1611 { int win; 1612 1613 win = sbspace(&so->so_rcv); 1614 if (win < 0) 1615 win = 0; 1616 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1617 } 1618 1619 switch (tp->t_state) { 1620 case TCPS_LISTEN: 1621 /* 1622 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1623 */ 1624 if (m->m_flags & (M_BCAST|M_MCAST)) 1625 goto drop; 1626 switch (af) { 1627 #ifdef INET6 1628 case AF_INET6: 1629 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1630 goto drop; 1631 break; 1632 #endif /* INET6 */ 1633 case AF_INET: 1634 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1635 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1636 goto drop; 1637 break; 1638 } 1639 break; 1640 1641 /* 1642 * If the state is SYN_SENT: 1643 * if seg contains an ACK, but not for our SYN, drop the input. 1644 * if seg contains a RST, then drop the connection. 1645 * if seg does not contain SYN, then drop it. 1646 * Otherwise this is an acceptable SYN segment 1647 * initialize tp->rcv_nxt and tp->irs 1648 * if seg contains ack then advance tp->snd_una 1649 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1650 * arrange for segment to be acked (eventually) 1651 * continue processing rest of data/controls, beginning with URG 1652 */ 1653 case TCPS_SYN_SENT: 1654 if ((tiflags & TH_ACK) && 1655 (SEQ_LEQ(th->th_ack, tp->iss) || 1656 SEQ_GT(th->th_ack, tp->snd_max))) 1657 goto dropwithreset; 1658 if (tiflags & TH_RST) { 1659 if (tiflags & TH_ACK) 1660 tp = tcp_drop(tp, ECONNREFUSED); 1661 goto drop; 1662 } 1663 if ((tiflags & TH_SYN) == 0) 1664 goto drop; 1665 if (tiflags & TH_ACK) { 1666 tp->snd_una = tp->snd_recover = th->th_ack; 1667 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1668 tp->snd_nxt = tp->snd_una; 1669 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1670 } 1671 tp->irs = th->th_seq; 1672 tcp_rcvseqinit(tp); 1673 tp->t_flags |= TF_ACKNOW; 1674 tcp_mss_from_peer(tp, opti.maxseg); 1675 1676 /* 1677 * Initialize the initial congestion window. If we 1678 * had to retransmit the SYN, we must initialize cwnd 1679 * to 1 segment (i.e. the Loss Window). 1680 */ 1681 if (tp->t_flags & TF_SYN_REXMT) 1682 tp->snd_cwnd = tp->t_peermss; 1683 else 1684 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1685 tp->t_peermss); 1686 1687 tcp_rmx_rtt(tp); 1688 if (tiflags & TH_ACK) { 1689 tcpstat.tcps_connects++; 1690 soisconnected(so); 1691 tcp_established(tp); 1692 /* Do window scaling on this connection? */ 1693 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1694 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1695 tp->snd_scale = tp->requested_s_scale; 1696 tp->rcv_scale = tp->request_r_scale; 1697 } 1698 TCP_REASS_LOCK(tp); 1699 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1700 TCP_REASS_UNLOCK(tp); 1701 /* 1702 * if we didn't have to retransmit the SYN, 1703 * use its rtt as our initial srtt & rtt var. 1704 */ 1705 if (tp->t_rtttime) 1706 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1707 } else 1708 tp->t_state = TCPS_SYN_RECEIVED; 1709 1710 /* 1711 * Advance th->th_seq to correspond to first data byte. 1712 * If data, trim to stay within window, 1713 * dropping FIN if necessary. 1714 */ 1715 th->th_seq++; 1716 if (tlen > tp->rcv_wnd) { 1717 todrop = tlen - tp->rcv_wnd; 1718 m_adj(m, -todrop); 1719 tlen = tp->rcv_wnd; 1720 tiflags &= ~TH_FIN; 1721 tcpstat.tcps_rcvpackafterwin++; 1722 tcpstat.tcps_rcvbyteafterwin += todrop; 1723 } 1724 tp->snd_wl1 = th->th_seq - 1; 1725 tp->rcv_up = th->th_seq; 1726 goto step6; 1727 1728 /* 1729 * If the state is SYN_RECEIVED: 1730 * If seg contains an ACK, but not for our SYN, drop the input 1731 * and generate an RST. See page 36, rfc793 1732 */ 1733 case TCPS_SYN_RECEIVED: 1734 if ((tiflags & TH_ACK) && 1735 (SEQ_LEQ(th->th_ack, tp->iss) || 1736 SEQ_GT(th->th_ack, tp->snd_max))) 1737 goto dropwithreset; 1738 break; 1739 } 1740 1741 /* 1742 * States other than LISTEN or SYN_SENT. 1743 * First check timestamp, if present. 1744 * Then check that at least some bytes of segment are within 1745 * receive window. If segment begins before rcv_nxt, 1746 * drop leading data (and SYN); if nothing left, just ack. 1747 * 1748 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1749 * and it's less than ts_recent, drop it. 1750 */ 1751 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1752 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1753 1754 /* Check to see if ts_recent is over 24 days old. */ 1755 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1756 TCP_PAWS_IDLE) { 1757 /* 1758 * Invalidate ts_recent. If this segment updates 1759 * ts_recent, the age will be reset later and ts_recent 1760 * will get a valid value. If it does not, setting 1761 * ts_recent to zero will at least satisfy the 1762 * requirement that zero be placed in the timestamp 1763 * echo reply when ts_recent isn't valid. The 1764 * age isn't reset until we get a valid ts_recent 1765 * because we don't want out-of-order segments to be 1766 * dropped when ts_recent is old. 1767 */ 1768 tp->ts_recent = 0; 1769 } else { 1770 tcpstat.tcps_rcvduppack++; 1771 tcpstat.tcps_rcvdupbyte += tlen; 1772 tcpstat.tcps_pawsdrop++; 1773 goto dropafterack; 1774 } 1775 } 1776 1777 todrop = tp->rcv_nxt - th->th_seq; 1778 if (todrop > 0) { 1779 if (tiflags & TH_SYN) { 1780 tiflags &= ~TH_SYN; 1781 th->th_seq++; 1782 if (th->th_urp > 1) 1783 th->th_urp--; 1784 else { 1785 tiflags &= ~TH_URG; 1786 th->th_urp = 0; 1787 } 1788 todrop--; 1789 } 1790 if (todrop > tlen || 1791 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1792 /* 1793 * Any valid FIN must be to the left of the window. 1794 * At this point the FIN must be a duplicate or 1795 * out of sequence; drop it. 1796 */ 1797 tiflags &= ~TH_FIN; 1798 /* 1799 * Send an ACK to resynchronize and drop any data. 1800 * But keep on processing for RST or ACK. 1801 */ 1802 tp->t_flags |= TF_ACKNOW; 1803 todrop = tlen; 1804 tcpstat.tcps_rcvdupbyte += todrop; 1805 tcpstat.tcps_rcvduppack++; 1806 } else { 1807 tcpstat.tcps_rcvpartduppack++; 1808 tcpstat.tcps_rcvpartdupbyte += todrop; 1809 } 1810 hdroptlen += todrop; /*drop from head afterwards*/ 1811 th->th_seq += todrop; 1812 tlen -= todrop; 1813 if (th->th_urp > todrop) 1814 th->th_urp -= todrop; 1815 else { 1816 tiflags &= ~TH_URG; 1817 th->th_urp = 0; 1818 } 1819 } 1820 1821 /* 1822 * If new data are received on a connection after the 1823 * user processes are gone, then RST the other end. 1824 */ 1825 if ((so->so_state & SS_NOFDREF) && 1826 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1827 tp = tcp_close(tp); 1828 tcpstat.tcps_rcvafterclose++; 1829 goto dropwithreset; 1830 } 1831 1832 /* 1833 * If segment ends after window, drop trailing data 1834 * (and PUSH and FIN); if nothing left, just ACK. 1835 */ 1836 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1837 if (todrop > 0) { 1838 tcpstat.tcps_rcvpackafterwin++; 1839 if (todrop >= tlen) { 1840 tcpstat.tcps_rcvbyteafterwin += tlen; 1841 /* 1842 * If a new connection request is received 1843 * while in TIME_WAIT, drop the old connection 1844 * and start over if the sequence numbers 1845 * are above the previous ones. 1846 * 1847 * NOTE: We will checksum the packet again, and 1848 * so we need to put the header fields back into 1849 * network order! 1850 * XXX This kind of sucks, but we don't expect 1851 * XXX this to happen very often, so maybe it 1852 * XXX doesn't matter so much. 1853 */ 1854 if (tiflags & TH_SYN && 1855 tp->t_state == TCPS_TIME_WAIT && 1856 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1857 iss = tcp_new_iss(tp, tp->snd_nxt); 1858 tp = tcp_close(tp); 1859 TCP_FIELDS_TO_NET(th); 1860 goto findpcb; 1861 } 1862 /* 1863 * If window is closed can only take segments at 1864 * window edge, and have to drop data and PUSH from 1865 * incoming segments. Continue processing, but 1866 * remember to ack. Otherwise, drop segment 1867 * and ack. 1868 */ 1869 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1870 tp->t_flags |= TF_ACKNOW; 1871 tcpstat.tcps_rcvwinprobe++; 1872 } else 1873 goto dropafterack; 1874 } else 1875 tcpstat.tcps_rcvbyteafterwin += todrop; 1876 m_adj(m, -todrop); 1877 tlen -= todrop; 1878 tiflags &= ~(TH_PUSH|TH_FIN); 1879 } 1880 1881 /* 1882 * If last ACK falls within this segment's sequence numbers, 1883 * and the timestamp is newer, record it. 1884 */ 1885 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1886 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1887 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1888 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1889 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1890 tp->ts_recent = opti.ts_val; 1891 } 1892 1893 /* 1894 * If the RST bit is set examine the state: 1895 * SYN_RECEIVED STATE: 1896 * If passive open, return to LISTEN state. 1897 * If active open, inform user that connection was refused. 1898 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1899 * Inform user that connection was reset, and close tcb. 1900 * CLOSING, LAST_ACK, TIME_WAIT STATES 1901 * Close the tcb. 1902 */ 1903 if (tiflags&TH_RST) switch (tp->t_state) { 1904 1905 case TCPS_SYN_RECEIVED: 1906 so->so_error = ECONNREFUSED; 1907 goto close; 1908 1909 case TCPS_ESTABLISHED: 1910 case TCPS_FIN_WAIT_1: 1911 case TCPS_FIN_WAIT_2: 1912 case TCPS_CLOSE_WAIT: 1913 so->so_error = ECONNRESET; 1914 close: 1915 tp->t_state = TCPS_CLOSED; 1916 tcpstat.tcps_drops++; 1917 tp = tcp_close(tp); 1918 goto drop; 1919 1920 case TCPS_CLOSING: 1921 case TCPS_LAST_ACK: 1922 case TCPS_TIME_WAIT: 1923 tp = tcp_close(tp); 1924 goto drop; 1925 } 1926 1927 /* 1928 * If a SYN is in the window, then this is an 1929 * error and we send an RST and drop the connection. 1930 */ 1931 if (tiflags & TH_SYN) { 1932 tp = tcp_drop(tp, ECONNRESET); 1933 goto dropwithreset; 1934 } 1935 1936 /* 1937 * If the ACK bit is off we drop the segment and return. 1938 */ 1939 if ((tiflags & TH_ACK) == 0) { 1940 if (tp->t_flags & TF_ACKNOW) 1941 goto dropafterack; 1942 else 1943 goto drop; 1944 } 1945 1946 /* 1947 * Ack processing. 1948 */ 1949 switch (tp->t_state) { 1950 1951 /* 1952 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1953 * ESTABLISHED state and continue processing, otherwise 1954 * send an RST. 1955 */ 1956 case TCPS_SYN_RECEIVED: 1957 if (SEQ_GT(tp->snd_una, th->th_ack) || 1958 SEQ_GT(th->th_ack, tp->snd_max)) 1959 goto dropwithreset; 1960 tcpstat.tcps_connects++; 1961 soisconnected(so); 1962 tcp_established(tp); 1963 /* Do window scaling? */ 1964 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1965 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1966 tp->snd_scale = tp->requested_s_scale; 1967 tp->rcv_scale = tp->request_r_scale; 1968 } 1969 TCP_REASS_LOCK(tp); 1970 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1971 TCP_REASS_UNLOCK(tp); 1972 tp->snd_wl1 = th->th_seq - 1; 1973 /* fall into ... */ 1974 1975 /* 1976 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1977 * ACKs. If the ack is in the range 1978 * tp->snd_una < th->th_ack <= tp->snd_max 1979 * then advance tp->snd_una to th->th_ack and drop 1980 * data from the retransmission queue. If this ACK reflects 1981 * more up to date window information we update our window information. 1982 */ 1983 case TCPS_ESTABLISHED: 1984 case TCPS_FIN_WAIT_1: 1985 case TCPS_FIN_WAIT_2: 1986 case TCPS_CLOSE_WAIT: 1987 case TCPS_CLOSING: 1988 case TCPS_LAST_ACK: 1989 case TCPS_TIME_WAIT: 1990 1991 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1992 if (tlen == 0 && tiwin == tp->snd_wnd) { 1993 tcpstat.tcps_rcvdupack++; 1994 /* 1995 * If we have outstanding data (other than 1996 * a window probe), this is a completely 1997 * duplicate ack (ie, window info didn't 1998 * change), the ack is the biggest we've 1999 * seen and we've seen exactly our rexmt 2000 * threshhold of them, assume a packet 2001 * has been dropped and retransmit it. 2002 * Kludge snd_nxt & the congestion 2003 * window so we send only this one 2004 * packet. 2005 * 2006 * We know we're losing at the current 2007 * window size so do congestion avoidance 2008 * (set ssthresh to half the current window 2009 * and pull our congestion window back to 2010 * the new ssthresh). 2011 * 2012 * Dup acks mean that packets have left the 2013 * network (they're now cached at the receiver) 2014 * so bump cwnd by the amount in the receiver 2015 * to keep a constant cwnd packets in the 2016 * network. 2017 */ 2018 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2019 th->th_ack != tp->snd_una) 2020 tp->t_dupacks = 0; 2021 else if (++tp->t_dupacks == tcprexmtthresh) { 2022 tcp_seq onxt = tp->snd_nxt; 2023 u_int win = 2024 min(tp->snd_wnd, tp->snd_cwnd) / 2025 2 / tp->t_segsz; 2026 if (tcp_do_newreno && SEQ_LT(th->th_ack, 2027 tp->snd_recover)) { 2028 /* 2029 * False fast retransmit after 2030 * timeout. Do not cut window. 2031 */ 2032 tp->snd_cwnd += tp->t_segsz; 2033 tp->t_dupacks = 0; 2034 (void) tcp_output(tp); 2035 goto drop; 2036 } 2037 2038 if (win < 2) 2039 win = 2; 2040 tp->snd_ssthresh = win * tp->t_segsz; 2041 tp->snd_recover = tp->snd_max; 2042 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2043 tp->t_rtttime = 0; 2044 tp->snd_nxt = th->th_ack; 2045 tp->snd_cwnd = tp->t_segsz; 2046 (void) tcp_output(tp); 2047 tp->snd_cwnd = tp->snd_ssthresh + 2048 tp->t_segsz * tp->t_dupacks; 2049 if (SEQ_GT(onxt, tp->snd_nxt)) 2050 tp->snd_nxt = onxt; 2051 goto drop; 2052 } else if (tp->t_dupacks > tcprexmtthresh) { 2053 tp->snd_cwnd += tp->t_segsz; 2054 (void) tcp_output(tp); 2055 goto drop; 2056 } 2057 } else 2058 tp->t_dupacks = 0; 2059 break; 2060 } 2061 /* 2062 * If the congestion window was inflated to account 2063 * for the other side's cached packets, retract it. 2064 */ 2065 if (tcp_do_newreno == 0) { 2066 if (tp->t_dupacks >= tcprexmtthresh && 2067 tp->snd_cwnd > tp->snd_ssthresh) 2068 tp->snd_cwnd = tp->snd_ssthresh; 2069 tp->t_dupacks = 0; 2070 } else if (tp->t_dupacks >= tcprexmtthresh && 2071 tcp_newreno(tp, th) == 0) { 2072 tp->snd_cwnd = tp->snd_ssthresh; 2073 /* 2074 * Window inflation should have left us with approx. 2075 * snd_ssthresh outstanding data. But in case we 2076 * would be inclined to send a burst, better to do 2077 * it via the slow start mechanism. 2078 */ 2079 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 2080 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 2081 + tp->t_segsz; 2082 tp->t_dupacks = 0; 2083 } 2084 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2085 tcpstat.tcps_rcvacktoomuch++; 2086 goto dropafterack; 2087 } 2088 acked = th->th_ack - tp->snd_una; 2089 tcpstat.tcps_rcvackpack++; 2090 tcpstat.tcps_rcvackbyte += acked; 2091 2092 /* 2093 * If we have a timestamp reply, update smoothed 2094 * round trip time. If no timestamp is present but 2095 * transmit timer is running and timed sequence 2096 * number was acked, update smoothed round trip time. 2097 * Since we now have an rtt measurement, cancel the 2098 * timer backoff (cf., Phil Karn's retransmit alg.). 2099 * Recompute the initial retransmit timer. 2100 */ 2101 if (opti.ts_present && opti.ts_ecr) 2102 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2103 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2104 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2105 2106 /* 2107 * If all outstanding data is acked, stop retransmit 2108 * timer and remember to restart (more output or persist). 2109 * If there is more data to be acked, restart retransmit 2110 * timer, using current (possibly backed-off) value. 2111 */ 2112 if (th->th_ack == tp->snd_max) { 2113 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2114 needoutput = 1; 2115 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2116 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2117 /* 2118 * When new data is acked, open the congestion window. 2119 * If the window gives us less than ssthresh packets 2120 * in flight, open exponentially (segsz per packet). 2121 * Otherwise open linearly: segsz per window 2122 * (segsz^2 / cwnd per packet), plus a constant 2123 * fraction of a packet (segsz/8) to help larger windows 2124 * open quickly enough. 2125 */ 2126 { 2127 u_int cw = tp->snd_cwnd; 2128 u_int incr = tp->t_segsz; 2129 2130 if (cw > tp->snd_ssthresh) 2131 incr = incr * incr / cw; 2132 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2133 tp->snd_cwnd = min(cw + incr, 2134 TCP_MAXWIN << tp->snd_scale); 2135 } 2136 ND6_HINT(tp); 2137 if (acked > so->so_snd.sb_cc) { 2138 tp->snd_wnd -= so->so_snd.sb_cc; 2139 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2140 ourfinisacked = 1; 2141 } else { 2142 sbdrop(&so->so_snd, acked); 2143 tp->snd_wnd -= acked; 2144 ourfinisacked = 0; 2145 } 2146 sowwakeup(so); 2147 /* 2148 * We want snd_recover to track snd_una to 2149 * avoid sequence wraparound problems for 2150 * very large transfers. 2151 */ 2152 tp->snd_una = tp->snd_recover = th->th_ack; 2153 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2154 tp->snd_nxt = tp->snd_una; 2155 2156 switch (tp->t_state) { 2157 2158 /* 2159 * In FIN_WAIT_1 STATE in addition to the processing 2160 * for the ESTABLISHED state if our FIN is now acknowledged 2161 * then enter FIN_WAIT_2. 2162 */ 2163 case TCPS_FIN_WAIT_1: 2164 if (ourfinisacked) { 2165 /* 2166 * If we can't receive any more 2167 * data, then closing user can proceed. 2168 * Starting the timer is contrary to the 2169 * specification, but if we don't get a FIN 2170 * we'll hang forever. 2171 */ 2172 if (so->so_state & SS_CANTRCVMORE) { 2173 soisdisconnected(so); 2174 if (tcp_maxidle > 0) 2175 TCP_TIMER_ARM(tp, TCPT_2MSL, 2176 tcp_maxidle); 2177 } 2178 tp->t_state = TCPS_FIN_WAIT_2; 2179 } 2180 break; 2181 2182 /* 2183 * In CLOSING STATE in addition to the processing for 2184 * the ESTABLISHED state if the ACK acknowledges our FIN 2185 * then enter the TIME-WAIT state, otherwise ignore 2186 * the segment. 2187 */ 2188 case TCPS_CLOSING: 2189 if (ourfinisacked) { 2190 tp->t_state = TCPS_TIME_WAIT; 2191 tcp_canceltimers(tp); 2192 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2193 soisdisconnected(so); 2194 } 2195 break; 2196 2197 /* 2198 * In LAST_ACK, we may still be waiting for data to drain 2199 * and/or to be acked, as well as for the ack of our FIN. 2200 * If our FIN is now acknowledged, delete the TCB, 2201 * enter the closed state and return. 2202 */ 2203 case TCPS_LAST_ACK: 2204 if (ourfinisacked) { 2205 tp = tcp_close(tp); 2206 goto drop; 2207 } 2208 break; 2209 2210 /* 2211 * In TIME_WAIT state the only thing that should arrive 2212 * is a retransmission of the remote FIN. Acknowledge 2213 * it and restart the finack timer. 2214 */ 2215 case TCPS_TIME_WAIT: 2216 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2217 goto dropafterack; 2218 } 2219 } 2220 2221 step6: 2222 /* 2223 * Update window information. 2224 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2225 */ 2226 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2227 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2228 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2229 /* keep track of pure window updates */ 2230 if (tlen == 0 && 2231 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2232 tcpstat.tcps_rcvwinupd++; 2233 tp->snd_wnd = tiwin; 2234 tp->snd_wl1 = th->th_seq; 2235 tp->snd_wl2 = th->th_ack; 2236 if (tp->snd_wnd > tp->max_sndwnd) 2237 tp->max_sndwnd = tp->snd_wnd; 2238 needoutput = 1; 2239 } 2240 2241 /* 2242 * Process segments with URG. 2243 */ 2244 if ((tiflags & TH_URG) && th->th_urp && 2245 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2246 /* 2247 * This is a kludge, but if we receive and accept 2248 * random urgent pointers, we'll crash in 2249 * soreceive. It's hard to imagine someone 2250 * actually wanting to send this much urgent data. 2251 */ 2252 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2253 th->th_urp = 0; /* XXX */ 2254 tiflags &= ~TH_URG; /* XXX */ 2255 goto dodata; /* XXX */ 2256 } 2257 /* 2258 * If this segment advances the known urgent pointer, 2259 * then mark the data stream. This should not happen 2260 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2261 * a FIN has been received from the remote side. 2262 * In these states we ignore the URG. 2263 * 2264 * According to RFC961 (Assigned Protocols), 2265 * the urgent pointer points to the last octet 2266 * of urgent data. We continue, however, 2267 * to consider it to indicate the first octet 2268 * of data past the urgent section as the original 2269 * spec states (in one of two places). 2270 */ 2271 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2272 tp->rcv_up = th->th_seq + th->th_urp; 2273 so->so_oobmark = so->so_rcv.sb_cc + 2274 (tp->rcv_up - tp->rcv_nxt) - 1; 2275 if (so->so_oobmark == 0) 2276 so->so_state |= SS_RCVATMARK; 2277 sohasoutofband(so); 2278 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2279 } 2280 /* 2281 * Remove out of band data so doesn't get presented to user. 2282 * This can happen independent of advancing the URG pointer, 2283 * but if two URG's are pending at once, some out-of-band 2284 * data may creep in... ick. 2285 */ 2286 if (th->th_urp <= (u_int16_t) tlen 2287 #ifdef SO_OOBINLINE 2288 && (so->so_options & SO_OOBINLINE) == 0 2289 #endif 2290 ) 2291 tcp_pulloutofband(so, th, m, hdroptlen); 2292 } else 2293 /* 2294 * If no out of band data is expected, 2295 * pull receive urgent pointer along 2296 * with the receive window. 2297 */ 2298 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2299 tp->rcv_up = tp->rcv_nxt; 2300 dodata: /* XXX */ 2301 2302 /* 2303 * Process the segment text, merging it into the TCP sequencing queue, 2304 * and arranging for acknowledgement of receipt if necessary. 2305 * This process logically involves adjusting tp->rcv_wnd as data 2306 * is presented to the user (this happens in tcp_usrreq.c, 2307 * case PRU_RCVD). If a FIN has already been received on this 2308 * connection then we just ignore the text. 2309 */ 2310 if ((tlen || (tiflags & TH_FIN)) && 2311 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2312 /* 2313 * Insert segment ti into reassembly queue of tcp with 2314 * control block tp. Return TH_FIN if reassembly now includes 2315 * a segment with FIN. The macro form does the common case 2316 * inline (segment is the next to be received on an 2317 * established connection, and the queue is empty), 2318 * avoiding linkage into and removal from the queue and 2319 * repetition of various conversions. 2320 * Set DELACK for segments received in order, but ack 2321 * immediately when segments are out of order 2322 * (so fast retransmit can work). 2323 */ 2324 /* NOTE: this was TCP_REASS() macro, but used only once */ 2325 TCP_REASS_LOCK(tp); 2326 if (th->th_seq == tp->rcv_nxt && 2327 TAILQ_FIRST(&tp->segq) == NULL && 2328 tp->t_state == TCPS_ESTABLISHED) { 2329 TCP_SETUP_ACK(tp, th); 2330 tp->rcv_nxt += tlen; 2331 tiflags = th->th_flags & TH_FIN; 2332 tcpstat.tcps_rcvpack++; 2333 tcpstat.tcps_rcvbyte += tlen; 2334 ND6_HINT(tp); 2335 if (so->so_state & SS_CANTRCVMORE) 2336 m_freem(m); 2337 else { 2338 m_adj(m, hdroptlen); 2339 sbappendstream(&(so)->so_rcv, m); 2340 } 2341 sorwakeup(so); 2342 } else { 2343 m_adj(m, hdroptlen); 2344 tiflags = tcp_reass(tp, th, m, &tlen); 2345 tp->t_flags |= TF_ACKNOW; 2346 } 2347 TCP_REASS_UNLOCK(tp); 2348 2349 /* 2350 * Note the amount of data that peer has sent into 2351 * our window, in order to estimate the sender's 2352 * buffer size. 2353 */ 2354 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2355 } else { 2356 m_freem(m); 2357 m = NULL; 2358 tiflags &= ~TH_FIN; 2359 } 2360 2361 /* 2362 * If FIN is received ACK the FIN and let the user know 2363 * that the connection is closing. Ignore a FIN received before 2364 * the connection is fully established. 2365 */ 2366 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2367 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2368 socantrcvmore(so); 2369 tp->t_flags |= TF_ACKNOW; 2370 tp->rcv_nxt++; 2371 } 2372 switch (tp->t_state) { 2373 2374 /* 2375 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2376 */ 2377 case TCPS_ESTABLISHED: 2378 tp->t_state = TCPS_CLOSE_WAIT; 2379 break; 2380 2381 /* 2382 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2383 * enter the CLOSING state. 2384 */ 2385 case TCPS_FIN_WAIT_1: 2386 tp->t_state = TCPS_CLOSING; 2387 break; 2388 2389 /* 2390 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2391 * starting the time-wait timer, turning off the other 2392 * standard timers. 2393 */ 2394 case TCPS_FIN_WAIT_2: 2395 tp->t_state = TCPS_TIME_WAIT; 2396 tcp_canceltimers(tp); 2397 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2398 soisdisconnected(so); 2399 break; 2400 2401 /* 2402 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2403 */ 2404 case TCPS_TIME_WAIT: 2405 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2406 break; 2407 } 2408 } 2409 #ifdef TCP_DEBUG 2410 if (so->so_options & SO_DEBUG) 2411 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2412 #endif 2413 2414 /* 2415 * Return any desired output. 2416 */ 2417 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2418 (void) tcp_output(tp); 2419 if (tcp_saveti) 2420 m_freem(tcp_saveti); 2421 return; 2422 2423 badsyn: 2424 /* 2425 * Received a bad SYN. Increment counters and dropwithreset. 2426 */ 2427 tcpstat.tcps_badsyn++; 2428 tp = NULL; 2429 goto dropwithreset; 2430 2431 dropafterack: 2432 /* 2433 * Generate an ACK dropping incoming segment if it occupies 2434 * sequence space, where the ACK reflects our state. 2435 */ 2436 if (tiflags & TH_RST) 2437 goto drop; 2438 m_freem(m); 2439 tp->t_flags |= TF_ACKNOW; 2440 (void) tcp_output(tp); 2441 if (tcp_saveti) 2442 m_freem(tcp_saveti); 2443 return; 2444 2445 dropwithreset_ratelim: 2446 /* 2447 * We may want to rate-limit RSTs in certain situations, 2448 * particularly if we are sending an RST in response to 2449 * an attempt to connect to or otherwise communicate with 2450 * a port for which we have no socket. 2451 */ 2452 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2453 tcp_rst_ppslim) == 0) { 2454 /* XXX stat */ 2455 goto drop; 2456 } 2457 /* ...fall into dropwithreset... */ 2458 2459 dropwithreset: 2460 /* 2461 * Generate a RST, dropping incoming segment. 2462 * Make ACK acceptable to originator of segment. 2463 */ 2464 if (tiflags & TH_RST) 2465 goto drop; 2466 2467 switch (af) { 2468 #ifdef INET6 2469 case AF_INET6: 2470 /* For following calls to tcp_respond */ 2471 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2472 goto drop; 2473 break; 2474 #endif /* INET6 */ 2475 case AF_INET: 2476 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2477 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2478 goto drop; 2479 } 2480 2481 { 2482 /* 2483 * need to recover version # field, which was overwritten on 2484 * ip_cksum computation. 2485 */ 2486 struct ip *sip; 2487 sip = mtod(m, struct ip *); 2488 switch (af) { 2489 #ifdef INET 2490 case AF_INET: 2491 sip->ip_v = 4; 2492 break; 2493 #endif 2494 #ifdef INET6 2495 case AF_INET6: 2496 sip->ip_v = 6; 2497 break; 2498 #endif 2499 } 2500 } 2501 if (tiflags & TH_ACK) 2502 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2503 else { 2504 if (tiflags & TH_SYN) 2505 tlen++; 2506 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2507 TH_RST|TH_ACK); 2508 } 2509 if (tcp_saveti) 2510 m_freem(tcp_saveti); 2511 return; 2512 2513 badcsum: 2514 tcpstat.tcps_rcvbadsum++; 2515 drop: 2516 /* 2517 * Drop space held by incoming segment and return. 2518 */ 2519 if (tp) { 2520 if (tp->t_inpcb) 2521 so = tp->t_inpcb->inp_socket; 2522 #ifdef INET6 2523 else if (tp->t_in6pcb) 2524 so = tp->t_in6pcb->in6p_socket; 2525 #endif 2526 else 2527 so = NULL; 2528 #ifdef TCP_DEBUG 2529 if (so && (so->so_options & SO_DEBUG) != 0) 2530 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2531 #endif 2532 } 2533 if (tcp_saveti) 2534 m_freem(tcp_saveti); 2535 m_freem(m); 2536 return; 2537 } 2538 2539 void 2540 tcp_dooptions(tp, cp, cnt, th, oi) 2541 struct tcpcb *tp; 2542 u_char *cp; 2543 int cnt; 2544 struct tcphdr *th; 2545 struct tcp_opt_info *oi; 2546 { 2547 u_int16_t mss; 2548 int opt, optlen; 2549 2550 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2551 opt = cp[0]; 2552 if (opt == TCPOPT_EOL) 2553 break; 2554 if (opt == TCPOPT_NOP) 2555 optlen = 1; 2556 else { 2557 if (cnt < 2) 2558 break; 2559 optlen = cp[1]; 2560 if (optlen < 2 || optlen > cnt) 2561 break; 2562 } 2563 switch (opt) { 2564 2565 default: 2566 continue; 2567 2568 case TCPOPT_MAXSEG: 2569 if (optlen != TCPOLEN_MAXSEG) 2570 continue; 2571 if (!(th->th_flags & TH_SYN)) 2572 continue; 2573 bcopy(cp + 2, &mss, sizeof(mss)); 2574 oi->maxseg = ntohs(mss); 2575 break; 2576 2577 case TCPOPT_WINDOW: 2578 if (optlen != TCPOLEN_WINDOW) 2579 continue; 2580 if (!(th->th_flags & TH_SYN)) 2581 continue; 2582 tp->t_flags |= TF_RCVD_SCALE; 2583 tp->requested_s_scale = cp[2]; 2584 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2585 #if 0 /*XXX*/ 2586 char *p; 2587 2588 if (ip) 2589 p = ntohl(ip->ip_src); 2590 #ifdef INET6 2591 else if (ip6) 2592 p = ip6_sprintf(&ip6->ip6_src); 2593 #endif 2594 else 2595 p = "(unknown)"; 2596 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2597 "assuming %d\n", 2598 tp->requested_s_scale, p, 2599 TCP_MAX_WINSHIFT); 2600 #else 2601 log(LOG_ERR, "TCP: invalid wscale %d, " 2602 "assuming %d\n", 2603 tp->requested_s_scale, 2604 TCP_MAX_WINSHIFT); 2605 #endif 2606 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2607 } 2608 break; 2609 2610 case TCPOPT_TIMESTAMP: 2611 if (optlen != TCPOLEN_TIMESTAMP) 2612 continue; 2613 oi->ts_present = 1; 2614 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2615 NTOHL(oi->ts_val); 2616 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2617 NTOHL(oi->ts_ecr); 2618 2619 /* 2620 * A timestamp received in a SYN makes 2621 * it ok to send timestamp requests and replies. 2622 */ 2623 if (th->th_flags & TH_SYN) { 2624 tp->t_flags |= TF_RCVD_TSTMP; 2625 tp->ts_recent = oi->ts_val; 2626 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2627 } 2628 break; 2629 case TCPOPT_SACK_PERMITTED: 2630 if (optlen != TCPOLEN_SACK_PERMITTED) 2631 continue; 2632 if (!(th->th_flags & TH_SYN)) 2633 continue; 2634 tp->t_flags &= ~TF_CANT_TXSACK; 2635 break; 2636 2637 case TCPOPT_SACK: 2638 if (tp->t_flags & TF_IGNR_RXSACK) 2639 continue; 2640 if (optlen % 8 != 2 || optlen < 10) 2641 continue; 2642 cp += 2; 2643 optlen -= 2; 2644 for (; optlen > 0; cp -= 8, optlen -= 8) { 2645 tcp_seq lwe, rwe; 2646 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2647 NTOHL(lwe); 2648 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2649 NTOHL(rwe); 2650 /* tcp_mark_sacked(tp, lwe, rwe); */ 2651 } 2652 break; 2653 } 2654 } 2655 } 2656 2657 /* 2658 * Pull out of band byte out of a segment so 2659 * it doesn't appear in the user's data queue. 2660 * It is still reflected in the segment length for 2661 * sequencing purposes. 2662 */ 2663 void 2664 tcp_pulloutofband(so, th, m, off) 2665 struct socket *so; 2666 struct tcphdr *th; 2667 struct mbuf *m; 2668 int off; 2669 { 2670 int cnt = off + th->th_urp - 1; 2671 2672 while (cnt >= 0) { 2673 if (m->m_len > cnt) { 2674 char *cp = mtod(m, caddr_t) + cnt; 2675 struct tcpcb *tp = sototcpcb(so); 2676 2677 tp->t_iobc = *cp; 2678 tp->t_oobflags |= TCPOOB_HAVEDATA; 2679 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2680 m->m_len--; 2681 return; 2682 } 2683 cnt -= m->m_len; 2684 m = m->m_next; 2685 if (m == 0) 2686 break; 2687 } 2688 panic("tcp_pulloutofband"); 2689 } 2690 2691 /* 2692 * Collect new round-trip time estimate 2693 * and update averages and current timeout. 2694 */ 2695 void 2696 tcp_xmit_timer(tp, rtt) 2697 struct tcpcb *tp; 2698 uint32_t rtt; 2699 { 2700 int32_t delta; 2701 2702 tcpstat.tcps_rttupdated++; 2703 if (tp->t_srtt != 0) { 2704 /* 2705 * srtt is stored as fixed point with 3 bits after the 2706 * binary point (i.e., scaled by 8). The following magic 2707 * is equivalent to the smoothing algorithm in rfc793 with 2708 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2709 * point). Adjust rtt to origin 0. 2710 */ 2711 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2712 if ((tp->t_srtt += delta) <= 0) 2713 tp->t_srtt = 1 << 2; 2714 /* 2715 * We accumulate a smoothed rtt variance (actually, a 2716 * smoothed mean difference), then set the retransmit 2717 * timer to smoothed rtt + 4 times the smoothed variance. 2718 * rttvar is stored as fixed point with 2 bits after the 2719 * binary point (scaled by 4). The following is 2720 * equivalent to rfc793 smoothing with an alpha of .75 2721 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2722 * rfc793's wired-in beta. 2723 */ 2724 if (delta < 0) 2725 delta = -delta; 2726 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2727 if ((tp->t_rttvar += delta) <= 0) 2728 tp->t_rttvar = 1 << 2; 2729 } else { 2730 /* 2731 * No rtt measurement yet - use the unsmoothed rtt. 2732 * Set the variance to half the rtt (so our first 2733 * retransmit happens at 3*rtt). 2734 */ 2735 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2736 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2737 } 2738 tp->t_rtttime = 0; 2739 tp->t_rxtshift = 0; 2740 2741 /* 2742 * the retransmit should happen at rtt + 4 * rttvar. 2743 * Because of the way we do the smoothing, srtt and rttvar 2744 * will each average +1/2 tick of bias. When we compute 2745 * the retransmit timer, we want 1/2 tick of rounding and 2746 * 1 extra tick because of +-1/2 tick uncertainty in the 2747 * firing of the timer. The bias will give us exactly the 2748 * 1.5 tick we need. But, because the bias is 2749 * statistical, we have to test that we don't drop below 2750 * the minimum feasible timer (which is 2 ticks). 2751 */ 2752 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2753 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2754 2755 /* 2756 * We received an ack for a packet that wasn't retransmitted; 2757 * it is probably safe to discard any error indications we've 2758 * received recently. This isn't quite right, but close enough 2759 * for now (a route might have failed after we sent a segment, 2760 * and the return path might not be symmetrical). 2761 */ 2762 tp->t_softerror = 0; 2763 } 2764 2765 /* 2766 * Checks for partial ack. If partial ack arrives, force the retransmission 2767 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2768 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2769 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2770 */ 2771 int 2772 tcp_newreno(tp, th) 2773 struct tcpcb *tp; 2774 struct tcphdr *th; 2775 { 2776 tcp_seq onxt = tp->snd_nxt; 2777 u_long ocwnd = tp->snd_cwnd; 2778 2779 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2780 /* 2781 * snd_una has not yet been updated and the socket's send 2782 * buffer has not yet drained off the ACK'd data, so we 2783 * have to leave snd_una as it was to get the correct data 2784 * offset in tcp_output(). 2785 */ 2786 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2787 tp->t_rtttime = 0; 2788 tp->snd_nxt = th->th_ack; 2789 /* 2790 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2791 * is not yet updated when we're called. 2792 */ 2793 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2794 (void) tcp_output(tp); 2795 tp->snd_cwnd = ocwnd; 2796 if (SEQ_GT(onxt, tp->snd_nxt)) 2797 tp->snd_nxt = onxt; 2798 /* 2799 * Partial window deflation. Relies on fact that tp->snd_una 2800 * not updated yet. 2801 */ 2802 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2803 return 1; 2804 } 2805 return 0; 2806 } 2807 2808 2809 /* 2810 * TCP compressed state engine. Currently used to hold compressed 2811 * state for SYN_RECEIVED. 2812 */ 2813 2814 u_long syn_cache_count; 2815 u_int32_t syn_hash1, syn_hash2; 2816 2817 #define SYN_HASH(sa, sp, dp) \ 2818 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2819 ((u_int32_t)(sp)))^syn_hash2))) 2820 #ifndef INET6 2821 #define SYN_HASHALL(hash, src, dst) \ 2822 do { \ 2823 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2824 ((struct sockaddr_in *)(src))->sin_port, \ 2825 ((struct sockaddr_in *)(dst))->sin_port); \ 2826 } while (0) 2827 #else 2828 #define SYN_HASH6(sa, sp, dp) \ 2829 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2830 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2831 & 0x7fffffff) 2832 2833 #define SYN_HASHALL(hash, src, dst) \ 2834 do { \ 2835 switch ((src)->sa_family) { \ 2836 case AF_INET: \ 2837 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2838 ((struct sockaddr_in *)(src))->sin_port, \ 2839 ((struct sockaddr_in *)(dst))->sin_port); \ 2840 break; \ 2841 case AF_INET6: \ 2842 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2843 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2844 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2845 break; \ 2846 default: \ 2847 hash = 0; \ 2848 } \ 2849 } while (/*CONSTCOND*/0) 2850 #endif /* INET6 */ 2851 2852 #define SYN_CACHE_RM(sc) \ 2853 do { \ 2854 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 2855 (sc), sc_bucketq); \ 2856 (sc)->sc_tp = NULL; \ 2857 LIST_REMOVE((sc), sc_tpq); \ 2858 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2859 callout_stop(&(sc)->sc_timer); \ 2860 syn_cache_count--; \ 2861 } while (/*CONSTCOND*/0) 2862 2863 #define SYN_CACHE_PUT(sc) \ 2864 do { \ 2865 if ((sc)->sc_ipopts) \ 2866 (void) m_free((sc)->sc_ipopts); \ 2867 if ((sc)->sc_route4.ro_rt != NULL) \ 2868 RTFREE((sc)->sc_route4.ro_rt); \ 2869 pool_put(&syn_cache_pool, (sc)); \ 2870 } while (/*CONSTCOND*/0) 2871 2872 struct pool syn_cache_pool; 2873 2874 /* 2875 * We don't estimate RTT with SYNs, so each packet starts with the default 2876 * RTT and each timer step has a fixed timeout value. 2877 */ 2878 #define SYN_CACHE_TIMER_ARM(sc) \ 2879 do { \ 2880 TCPT_RANGESET((sc)->sc_rxtcur, \ 2881 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2882 TCPTV_REXMTMAX); \ 2883 callout_reset(&(sc)->sc_timer, \ 2884 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 2885 } while (/*CONSTCOND*/0) 2886 2887 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2888 2889 void 2890 syn_cache_init() 2891 { 2892 int i; 2893 2894 /* Initialize the hash buckets. */ 2895 for (i = 0; i < tcp_syn_cache_size; i++) 2896 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 2897 2898 /* Initialize the syn cache pool. */ 2899 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2900 "synpl", NULL); 2901 } 2902 2903 void 2904 syn_cache_insert(sc, tp) 2905 struct syn_cache *sc; 2906 struct tcpcb *tp; 2907 { 2908 struct syn_cache_head *scp; 2909 struct syn_cache *sc2; 2910 int s; 2911 2912 /* 2913 * If there are no entries in the hash table, reinitialize 2914 * the hash secrets. 2915 */ 2916 if (syn_cache_count == 0) { 2917 struct timeval tv; 2918 microtime(&tv); 2919 syn_hash1 = arc4random() ^ (u_long)≻ 2920 syn_hash2 = arc4random() ^ tv.tv_usec; 2921 } 2922 2923 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2924 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2925 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2926 2927 /* 2928 * Make sure that we don't overflow the per-bucket 2929 * limit or the total cache size limit. 2930 */ 2931 s = splsoftnet(); 2932 if (scp->sch_length >= tcp_syn_bucket_limit) { 2933 tcpstat.tcps_sc_bucketoverflow++; 2934 /* 2935 * The bucket is full. Toss the oldest element in the 2936 * bucket. This will be the first entry in the bucket. 2937 */ 2938 sc2 = TAILQ_FIRST(&scp->sch_bucket); 2939 #ifdef DIAGNOSTIC 2940 /* 2941 * This should never happen; we should always find an 2942 * entry in our bucket. 2943 */ 2944 if (sc2 == NULL) 2945 panic("syn_cache_insert: bucketoverflow: impossible"); 2946 #endif 2947 SYN_CACHE_RM(sc2); 2948 SYN_CACHE_PUT(sc2); 2949 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2950 struct syn_cache_head *scp2, *sce; 2951 2952 tcpstat.tcps_sc_overflowed++; 2953 /* 2954 * The cache is full. Toss the oldest entry in the 2955 * first non-empty bucket we can find. 2956 * 2957 * XXX We would really like to toss the oldest 2958 * entry in the cache, but we hope that this 2959 * condition doesn't happen very often. 2960 */ 2961 scp2 = scp; 2962 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 2963 sce = &tcp_syn_cache[tcp_syn_cache_size]; 2964 for (++scp2; scp2 != scp; scp2++) { 2965 if (scp2 >= sce) 2966 scp2 = &tcp_syn_cache[0]; 2967 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 2968 break; 2969 } 2970 #ifdef DIAGNOSTIC 2971 /* 2972 * This should never happen; we should always find a 2973 * non-empty bucket. 2974 */ 2975 if (scp2 == scp) 2976 panic("syn_cache_insert: cacheoverflow: " 2977 "impossible"); 2978 #endif 2979 } 2980 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 2981 SYN_CACHE_RM(sc2); 2982 SYN_CACHE_PUT(sc2); 2983 } 2984 2985 /* 2986 * Initialize the entry's timer. 2987 */ 2988 sc->sc_rxttot = 0; 2989 sc->sc_rxtshift = 0; 2990 SYN_CACHE_TIMER_ARM(sc); 2991 2992 /* Link it from tcpcb entry */ 2993 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2994 2995 /* Put it into the bucket. */ 2996 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 2997 scp->sch_length++; 2998 syn_cache_count++; 2999 3000 tcpstat.tcps_sc_added++; 3001 splx(s); 3002 } 3003 3004 /* 3005 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3006 * If we have retransmitted an entry the maximum number of times, expire 3007 * that entry. 3008 */ 3009 void 3010 syn_cache_timer(void *arg) 3011 { 3012 struct syn_cache *sc = arg; 3013 int s; 3014 3015 s = splsoftnet(); 3016 3017 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3018 /* Drop it -- too many retransmissions. */ 3019 goto dropit; 3020 } 3021 3022 /* 3023 * Compute the total amount of time this entry has 3024 * been on a queue. If this entry has been on longer 3025 * than the keep alive timer would allow, expire it. 3026 */ 3027 sc->sc_rxttot += sc->sc_rxtcur; 3028 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3029 goto dropit; 3030 3031 tcpstat.tcps_sc_retransmitted++; 3032 (void) syn_cache_respond(sc, NULL); 3033 3034 /* Advance the timer back-off. */ 3035 sc->sc_rxtshift++; 3036 SYN_CACHE_TIMER_ARM(sc); 3037 3038 splx(s); 3039 return; 3040 3041 dropit: 3042 tcpstat.tcps_sc_timed_out++; 3043 SYN_CACHE_RM(sc); 3044 SYN_CACHE_PUT(sc); 3045 splx(s); 3046 } 3047 3048 /* 3049 * Remove syn cache created by the specified tcb entry, 3050 * because this does not make sense to keep them 3051 * (if there's no tcb entry, syn cache entry will never be used) 3052 */ 3053 void 3054 syn_cache_cleanup(tp) 3055 struct tcpcb *tp; 3056 { 3057 struct syn_cache *sc, *nsc; 3058 int s; 3059 3060 s = splsoftnet(); 3061 3062 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3063 nsc = LIST_NEXT(sc, sc_tpq); 3064 3065 #ifdef DIAGNOSTIC 3066 if (sc->sc_tp != tp) 3067 panic("invalid sc_tp in syn_cache_cleanup"); 3068 #endif 3069 SYN_CACHE_RM(sc); 3070 SYN_CACHE_PUT(sc); 3071 } 3072 /* just for safety */ 3073 LIST_INIT(&tp->t_sc); 3074 3075 splx(s); 3076 } 3077 3078 /* 3079 * Find an entry in the syn cache. 3080 */ 3081 struct syn_cache * 3082 syn_cache_lookup(src, dst, headp) 3083 struct sockaddr *src; 3084 struct sockaddr *dst; 3085 struct syn_cache_head **headp; 3086 { 3087 struct syn_cache *sc; 3088 struct syn_cache_head *scp; 3089 u_int32_t hash; 3090 int s; 3091 3092 SYN_HASHALL(hash, src, dst); 3093 3094 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3095 *headp = scp; 3096 s = splsoftnet(); 3097 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3098 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3099 if (sc->sc_hash != hash) 3100 continue; 3101 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3102 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3103 splx(s); 3104 return (sc); 3105 } 3106 } 3107 splx(s); 3108 return (NULL); 3109 } 3110 3111 /* 3112 * This function gets called when we receive an ACK for a 3113 * socket in the LISTEN state. We look up the connection 3114 * in the syn cache, and if its there, we pull it out of 3115 * the cache and turn it into a full-blown connection in 3116 * the SYN-RECEIVED state. 3117 * 3118 * The return values may not be immediately obvious, and their effects 3119 * can be subtle, so here they are: 3120 * 3121 * NULL SYN was not found in cache; caller should drop the 3122 * packet and send an RST. 3123 * 3124 * -1 We were unable to create the new connection, and are 3125 * aborting it. An ACK,RST is being sent to the peer 3126 * (unless we got screwey sequence numbners; see below), 3127 * because the 3-way handshake has been completed. Caller 3128 * should not free the mbuf, since we may be using it. If 3129 * we are not, we will free it. 3130 * 3131 * Otherwise, the return value is a pointer to the new socket 3132 * associated with the connection. 3133 */ 3134 struct socket * 3135 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3136 struct sockaddr *src; 3137 struct sockaddr *dst; 3138 struct tcphdr *th; 3139 unsigned int hlen, tlen; 3140 struct socket *so; 3141 struct mbuf *m; 3142 { 3143 struct syn_cache *sc; 3144 struct syn_cache_head *scp; 3145 struct inpcb *inp = NULL; 3146 #ifdef INET6 3147 struct in6pcb *in6p = NULL; 3148 #endif 3149 struct tcpcb *tp = 0; 3150 struct mbuf *am; 3151 int s; 3152 struct socket *oso; 3153 3154 s = splsoftnet(); 3155 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3156 splx(s); 3157 return (NULL); 3158 } 3159 3160 /* 3161 * Verify the sequence and ack numbers. Try getting the correct 3162 * response again. 3163 */ 3164 if ((th->th_ack != sc->sc_iss + 1) || 3165 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3166 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3167 (void) syn_cache_respond(sc, m); 3168 splx(s); 3169 return ((struct socket *)(-1)); 3170 } 3171 3172 /* Remove this cache entry */ 3173 SYN_CACHE_RM(sc); 3174 splx(s); 3175 3176 /* 3177 * Ok, create the full blown connection, and set things up 3178 * as they would have been set up if we had created the 3179 * connection when the SYN arrived. If we can't create 3180 * the connection, abort it. 3181 */ 3182 /* 3183 * inp still has the OLD in_pcb stuff, set the 3184 * v6-related flags on the new guy, too. This is 3185 * done particularly for the case where an AF_INET6 3186 * socket is bound only to a port, and a v4 connection 3187 * comes in on that port. 3188 * we also copy the flowinfo from the original pcb 3189 * to the new one. 3190 */ 3191 { 3192 struct inpcb *parentinpcb; 3193 3194 parentinpcb = (struct inpcb *)so->so_pcb; 3195 3196 oso = so; 3197 so = sonewconn(so, SS_ISCONNECTED); 3198 if (so == NULL) 3199 goto resetandabort; 3200 3201 switch (so->so_proto->pr_domain->dom_family) { 3202 #ifdef INET 3203 case AF_INET: 3204 inp = sotoinpcb(so); 3205 break; 3206 #endif 3207 #ifdef INET6 3208 case AF_INET6: 3209 in6p = sotoin6pcb(so); 3210 break; 3211 #endif 3212 } 3213 } 3214 switch (src->sa_family) { 3215 #ifdef INET 3216 case AF_INET: 3217 if (inp) { 3218 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3219 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3220 inp->inp_options = ip_srcroute(); 3221 in_pcbstate(inp, INP_BOUND); 3222 if (inp->inp_options == NULL) { 3223 inp->inp_options = sc->sc_ipopts; 3224 sc->sc_ipopts = NULL; 3225 } 3226 } 3227 #ifdef INET6 3228 else if (in6p) { 3229 /* IPv4 packet to AF_INET6 socket */ 3230 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3231 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3232 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3233 &in6p->in6p_laddr.s6_addr32[3], 3234 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3235 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3236 in6totcpcb(in6p)->t_family = AF_INET; 3237 } 3238 #endif 3239 break; 3240 #endif 3241 #ifdef INET6 3242 case AF_INET6: 3243 if (in6p) { 3244 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3245 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3246 #if 0 3247 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3248 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3249 #endif 3250 } 3251 break; 3252 #endif 3253 } 3254 #ifdef INET6 3255 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3256 struct in6pcb *oin6p = sotoin6pcb(oso); 3257 /* inherit socket options from the listening socket */ 3258 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3259 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3260 m_freem(in6p->in6p_options); 3261 in6p->in6p_options = 0; 3262 } 3263 ip6_savecontrol(in6p, &in6p->in6p_options, 3264 mtod(m, struct ip6_hdr *), m); 3265 } 3266 #endif 3267 3268 #ifdef IPSEC 3269 /* 3270 * we make a copy of policy, instead of sharing the policy, 3271 * for better behavior in terms of SA lookup and dead SA removal. 3272 */ 3273 if (inp) { 3274 /* copy old policy into new socket's */ 3275 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3276 printf("tcp_input: could not copy policy\n"); 3277 } 3278 #ifdef INET6 3279 else if (in6p) { 3280 /* copy old policy into new socket's */ 3281 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3282 in6p->in6p_sp)) 3283 printf("tcp_input: could not copy policy\n"); 3284 } 3285 #endif 3286 #endif 3287 3288 /* 3289 * Give the new socket our cached route reference. 3290 */ 3291 if (inp) 3292 inp->inp_route = sc->sc_route4; /* struct assignment */ 3293 #ifdef INET6 3294 else 3295 in6p->in6p_route = sc->sc_route6; 3296 #endif 3297 sc->sc_route4.ro_rt = NULL; 3298 3299 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3300 if (am == NULL) 3301 goto resetandabort; 3302 am->m_len = src->sa_len; 3303 bcopy(src, mtod(am, caddr_t), src->sa_len); 3304 if (inp) { 3305 if (in_pcbconnect(inp, am)) { 3306 (void) m_free(am); 3307 goto resetandabort; 3308 } 3309 } 3310 #ifdef INET6 3311 else if (in6p) { 3312 if (src->sa_family == AF_INET) { 3313 /* IPv4 packet to AF_INET6 socket */ 3314 struct sockaddr_in6 *sin6; 3315 sin6 = mtod(am, struct sockaddr_in6 *); 3316 am->m_len = sizeof(*sin6); 3317 bzero(sin6, sizeof(*sin6)); 3318 sin6->sin6_family = AF_INET6; 3319 sin6->sin6_len = sizeof(*sin6); 3320 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3321 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3322 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3323 &sin6->sin6_addr.s6_addr32[3], 3324 sizeof(sin6->sin6_addr.s6_addr32[3])); 3325 } 3326 if (in6_pcbconnect(in6p, am)) { 3327 (void) m_free(am); 3328 goto resetandabort; 3329 } 3330 } 3331 #endif 3332 else { 3333 (void) m_free(am); 3334 goto resetandabort; 3335 } 3336 (void) m_free(am); 3337 3338 if (inp) 3339 tp = intotcpcb(inp); 3340 #ifdef INET6 3341 else if (in6p) 3342 tp = in6totcpcb(in6p); 3343 #endif 3344 else 3345 tp = NULL; 3346 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3347 if (sc->sc_request_r_scale != 15) { 3348 tp->requested_s_scale = sc->sc_requested_s_scale; 3349 tp->request_r_scale = sc->sc_request_r_scale; 3350 tp->snd_scale = sc->sc_requested_s_scale; 3351 tp->rcv_scale = sc->sc_request_r_scale; 3352 tp->t_flags |= TF_RCVD_SCALE; 3353 } 3354 if (sc->sc_flags & SCF_TIMESTAMP) 3355 tp->t_flags |= TF_RCVD_TSTMP; 3356 tp->ts_timebase = sc->sc_timebase; 3357 3358 tp->t_template = tcp_template(tp); 3359 if (tp->t_template == 0) { 3360 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3361 so = NULL; 3362 m_freem(m); 3363 goto abort; 3364 } 3365 3366 tp->iss = sc->sc_iss; 3367 tp->irs = sc->sc_irs; 3368 tcp_sendseqinit(tp); 3369 tcp_rcvseqinit(tp); 3370 tp->t_state = TCPS_SYN_RECEIVED; 3371 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3372 tcpstat.tcps_accepts++; 3373 3374 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3375 tp->t_ourmss = sc->sc_ourmaxseg; 3376 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3377 3378 /* 3379 * Initialize the initial congestion window. If we 3380 * had to retransmit the SYN,ACK, we must initialize cwnd 3381 * to 1 segment (i.e. the Loss Window). 3382 */ 3383 if (sc->sc_rxtshift) 3384 tp->snd_cwnd = tp->t_peermss; 3385 else 3386 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3387 3388 tcp_rmx_rtt(tp); 3389 tp->snd_wl1 = sc->sc_irs; 3390 tp->rcv_up = sc->sc_irs + 1; 3391 3392 /* 3393 * This is what whould have happened in tcp_ouput() when 3394 * the SYN,ACK was sent. 3395 */ 3396 tp->snd_up = tp->snd_una; 3397 tp->snd_max = tp->snd_nxt = tp->iss+1; 3398 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3399 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3400 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3401 tp->last_ack_sent = tp->rcv_nxt; 3402 3403 tcpstat.tcps_sc_completed++; 3404 SYN_CACHE_PUT(sc); 3405 return (so); 3406 3407 resetandabort: 3408 (void) tcp_respond(NULL, m, m, th, 3409 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3410 abort: 3411 if (so != NULL) 3412 (void) soabort(so); 3413 SYN_CACHE_PUT(sc); 3414 tcpstat.tcps_sc_aborted++; 3415 return ((struct socket *)(-1)); 3416 } 3417 3418 /* 3419 * This function is called when we get a RST for a 3420 * non-existent connection, so that we can see if the 3421 * connection is in the syn cache. If it is, zap it. 3422 */ 3423 3424 void 3425 syn_cache_reset(src, dst, th) 3426 struct sockaddr *src; 3427 struct sockaddr *dst; 3428 struct tcphdr *th; 3429 { 3430 struct syn_cache *sc; 3431 struct syn_cache_head *scp; 3432 int s = splsoftnet(); 3433 3434 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3435 splx(s); 3436 return; 3437 } 3438 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3439 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3440 splx(s); 3441 return; 3442 } 3443 SYN_CACHE_RM(sc); 3444 splx(s); 3445 tcpstat.tcps_sc_reset++; 3446 SYN_CACHE_PUT(sc); 3447 } 3448 3449 void 3450 syn_cache_unreach(src, dst, th) 3451 struct sockaddr *src; 3452 struct sockaddr *dst; 3453 struct tcphdr *th; 3454 { 3455 struct syn_cache *sc; 3456 struct syn_cache_head *scp; 3457 int s; 3458 3459 s = splsoftnet(); 3460 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3461 splx(s); 3462 return; 3463 } 3464 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3465 if (ntohl (th->th_seq) != sc->sc_iss) { 3466 splx(s); 3467 return; 3468 } 3469 3470 /* 3471 * If we've rertransmitted 3 times and this is our second error, 3472 * we remove the entry. Otherwise, we allow it to continue on. 3473 * This prevents us from incorrectly nuking an entry during a 3474 * spurious network outage. 3475 * 3476 * See tcp_notify(). 3477 */ 3478 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3479 sc->sc_flags |= SCF_UNREACH; 3480 splx(s); 3481 return; 3482 } 3483 3484 SYN_CACHE_RM(sc); 3485 splx(s); 3486 tcpstat.tcps_sc_unreach++; 3487 SYN_CACHE_PUT(sc); 3488 } 3489 3490 /* 3491 * Given a LISTEN socket and an inbound SYN request, add 3492 * this to the syn cache, and send back a segment: 3493 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3494 * to the source. 3495 * 3496 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3497 * Doing so would require that we hold onto the data and deliver it 3498 * to the application. However, if we are the target of a SYN-flood 3499 * DoS attack, an attacker could send data which would eventually 3500 * consume all available buffer space if it were ACKed. By not ACKing 3501 * the data, we avoid this DoS scenario. 3502 */ 3503 3504 int 3505 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3506 struct sockaddr *src; 3507 struct sockaddr *dst; 3508 struct tcphdr *th; 3509 unsigned int hlen; 3510 struct socket *so; 3511 struct mbuf *m; 3512 u_char *optp; 3513 int optlen; 3514 struct tcp_opt_info *oi; 3515 { 3516 struct tcpcb tb, *tp; 3517 long win; 3518 struct syn_cache *sc; 3519 struct syn_cache_head *scp; 3520 struct mbuf *ipopts; 3521 3522 tp = sototcpcb(so); 3523 3524 /* 3525 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3526 * 3527 * Note this check is performed in tcp_input() very early on. 3528 */ 3529 3530 /* 3531 * Initialize some local state. 3532 */ 3533 win = sbspace(&so->so_rcv); 3534 if (win > TCP_MAXWIN) 3535 win = TCP_MAXWIN; 3536 3537 switch (src->sa_family) { 3538 #ifdef INET 3539 case AF_INET: 3540 /* 3541 * Remember the IP options, if any. 3542 */ 3543 ipopts = ip_srcroute(); 3544 break; 3545 #endif 3546 default: 3547 ipopts = NULL; 3548 } 3549 3550 if (optp) { 3551 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3552 tcp_dooptions(&tb, optp, optlen, th, oi); 3553 } else 3554 tb.t_flags = 0; 3555 3556 /* 3557 * See if we already have an entry for this connection. 3558 * If we do, resend the SYN,ACK. We do not count this 3559 * as a retransmission (XXX though maybe we should). 3560 */ 3561 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3562 tcpstat.tcps_sc_dupesyn++; 3563 if (ipopts) { 3564 /* 3565 * If we were remembering a previous source route, 3566 * forget it and use the new one we've been given. 3567 */ 3568 if (sc->sc_ipopts) 3569 (void) m_free(sc->sc_ipopts); 3570 sc->sc_ipopts = ipopts; 3571 } 3572 sc->sc_timestamp = tb.ts_recent; 3573 if (syn_cache_respond(sc, m) == 0) { 3574 tcpstat.tcps_sndacks++; 3575 tcpstat.tcps_sndtotal++; 3576 } 3577 return (1); 3578 } 3579 3580 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3581 if (sc == NULL) { 3582 if (ipopts) 3583 (void) m_free(ipopts); 3584 return (0); 3585 } 3586 3587 /* 3588 * Fill in the cache, and put the necessary IP and TCP 3589 * options into the reply. 3590 */ 3591 callout_init(&sc->sc_timer); 3592 bzero(sc, sizeof(struct syn_cache)); 3593 bcopy(src, &sc->sc_src, src->sa_len); 3594 bcopy(dst, &sc->sc_dst, dst->sa_len); 3595 sc->sc_flags = 0; 3596 sc->sc_ipopts = ipopts; 3597 sc->sc_irs = th->th_seq; 3598 switch (src->sa_family) { 3599 #ifdef INET 3600 case AF_INET: 3601 { 3602 struct sockaddr_in *srcin = (void *) src; 3603 struct sockaddr_in *dstin = (void *) dst; 3604 3605 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3606 &srcin->sin_addr, dstin->sin_port, 3607 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3608 break; 3609 } 3610 #endif /* INET */ 3611 #ifdef INET6 3612 case AF_INET6: 3613 { 3614 struct sockaddr_in6 *srcin6 = (void *) src; 3615 struct sockaddr_in6 *dstin6 = (void *) dst; 3616 3617 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3618 &srcin6->sin6_addr, dstin6->sin6_port, 3619 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3620 break; 3621 } 3622 #endif /* INET6 */ 3623 } 3624 sc->sc_peermaxseg = oi->maxseg; 3625 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3626 m->m_pkthdr.rcvif : NULL, 3627 sc->sc_src.sa.sa_family); 3628 sc->sc_win = win; 3629 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3630 sc->sc_timestamp = tb.ts_recent; 3631 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3632 sc->sc_flags |= SCF_TIMESTAMP; 3633 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3634 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3635 sc->sc_requested_s_scale = tb.requested_s_scale; 3636 sc->sc_request_r_scale = 0; 3637 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3638 TCP_MAXWIN << sc->sc_request_r_scale < 3639 so->so_rcv.sb_hiwat) 3640 sc->sc_request_r_scale++; 3641 } else { 3642 sc->sc_requested_s_scale = 15; 3643 sc->sc_request_r_scale = 15; 3644 } 3645 sc->sc_tp = tp; 3646 if (syn_cache_respond(sc, m) == 0) { 3647 syn_cache_insert(sc, tp); 3648 tcpstat.tcps_sndacks++; 3649 tcpstat.tcps_sndtotal++; 3650 } else { 3651 SYN_CACHE_PUT(sc); 3652 tcpstat.tcps_sc_dropped++; 3653 } 3654 return (1); 3655 } 3656 3657 int 3658 syn_cache_respond(sc, m) 3659 struct syn_cache *sc; 3660 struct mbuf *m; 3661 { 3662 struct route *ro; 3663 u_int8_t *optp; 3664 int optlen, error; 3665 u_int16_t tlen; 3666 struct ip *ip = NULL; 3667 #ifdef INET6 3668 struct ip6_hdr *ip6 = NULL; 3669 #endif 3670 struct tcphdr *th; 3671 u_int hlen; 3672 3673 switch (sc->sc_src.sa.sa_family) { 3674 case AF_INET: 3675 hlen = sizeof(struct ip); 3676 ro = &sc->sc_route4; 3677 break; 3678 #ifdef INET6 3679 case AF_INET6: 3680 hlen = sizeof(struct ip6_hdr); 3681 ro = (struct route *)&sc->sc_route6; 3682 break; 3683 #endif 3684 default: 3685 if (m) 3686 m_freem(m); 3687 return EAFNOSUPPORT; 3688 } 3689 3690 /* Compute the size of the TCP options. */ 3691 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3692 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3693 3694 tlen = hlen + sizeof(struct tcphdr) + optlen; 3695 3696 /* 3697 * Create the IP+TCP header from scratch. 3698 */ 3699 if (m) 3700 m_freem(m); 3701 #ifdef DIAGNOSTIC 3702 if (max_linkhdr + tlen > MCLBYTES) 3703 return (ENOBUFS); 3704 #endif 3705 MGETHDR(m, M_DONTWAIT, MT_DATA); 3706 if (m && tlen > MHLEN) { 3707 MCLGET(m, M_DONTWAIT); 3708 if ((m->m_flags & M_EXT) == 0) { 3709 m_freem(m); 3710 m = NULL; 3711 } 3712 } 3713 if (m == NULL) 3714 return (ENOBUFS); 3715 3716 /* Fixup the mbuf. */ 3717 m->m_data += max_linkhdr; 3718 m->m_len = m->m_pkthdr.len = tlen; 3719 #ifdef IPSEC 3720 if (sc->sc_tp) { 3721 struct tcpcb *tp; 3722 struct socket *so; 3723 3724 tp = sc->sc_tp; 3725 if (tp->t_inpcb) 3726 so = tp->t_inpcb->inp_socket; 3727 #ifdef INET6 3728 else if (tp->t_in6pcb) 3729 so = tp->t_in6pcb->in6p_socket; 3730 #endif 3731 else 3732 so = NULL; 3733 /* use IPsec policy on listening socket, on SYN ACK */ 3734 if (ipsec_setsocket(m, so) != 0) { 3735 m_freem(m); 3736 return ENOBUFS; 3737 } 3738 } 3739 #endif 3740 m->m_pkthdr.rcvif = NULL; 3741 memset(mtod(m, u_char *), 0, tlen); 3742 3743 switch (sc->sc_src.sa.sa_family) { 3744 case AF_INET: 3745 ip = mtod(m, struct ip *); 3746 ip->ip_dst = sc->sc_src.sin.sin_addr; 3747 ip->ip_src = sc->sc_dst.sin.sin_addr; 3748 ip->ip_p = IPPROTO_TCP; 3749 th = (struct tcphdr *)(ip + 1); 3750 th->th_dport = sc->sc_src.sin.sin_port; 3751 th->th_sport = sc->sc_dst.sin.sin_port; 3752 break; 3753 #ifdef INET6 3754 case AF_INET6: 3755 ip6 = mtod(m, struct ip6_hdr *); 3756 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3757 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3758 ip6->ip6_nxt = IPPROTO_TCP; 3759 /* ip6_plen will be updated in ip6_output() */ 3760 th = (struct tcphdr *)(ip6 + 1); 3761 th->th_dport = sc->sc_src.sin6.sin6_port; 3762 th->th_sport = sc->sc_dst.sin6.sin6_port; 3763 break; 3764 #endif 3765 default: 3766 th = NULL; 3767 } 3768 3769 th->th_seq = htonl(sc->sc_iss); 3770 th->th_ack = htonl(sc->sc_irs + 1); 3771 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3772 th->th_flags = TH_SYN|TH_ACK; 3773 th->th_win = htons(sc->sc_win); 3774 /* th_sum already 0 */ 3775 /* th_urp already 0 */ 3776 3777 /* Tack on the TCP options. */ 3778 optp = (u_int8_t *)(th + 1); 3779 *optp++ = TCPOPT_MAXSEG; 3780 *optp++ = 4; 3781 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3782 *optp++ = sc->sc_ourmaxseg & 0xff; 3783 3784 if (sc->sc_request_r_scale != 15) { 3785 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3786 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3787 sc->sc_request_r_scale); 3788 optp += 4; 3789 } 3790 3791 if (sc->sc_flags & SCF_TIMESTAMP) { 3792 u_int32_t *lp = (u_int32_t *)(optp); 3793 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3794 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3795 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3796 *lp = htonl(sc->sc_timestamp); 3797 optp += TCPOLEN_TSTAMP_APPA; 3798 } 3799 3800 /* Compute the packet's checksum. */ 3801 switch (sc->sc_src.sa.sa_family) { 3802 case AF_INET: 3803 ip->ip_len = htons(tlen - hlen); 3804 th->th_sum = 0; 3805 th->th_sum = in_cksum(m, tlen); 3806 break; 3807 #ifdef INET6 3808 case AF_INET6: 3809 ip6->ip6_plen = htons(tlen - hlen); 3810 th->th_sum = 0; 3811 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3812 break; 3813 #endif 3814 } 3815 3816 /* 3817 * Fill in some straggling IP bits. Note the stack expects 3818 * ip_len to be in host order, for convenience. 3819 */ 3820 switch (sc->sc_src.sa.sa_family) { 3821 #ifdef INET 3822 case AF_INET: 3823 ip->ip_len = htons(tlen); 3824 ip->ip_ttl = ip_defttl; 3825 /* XXX tos? */ 3826 break; 3827 #endif 3828 #ifdef INET6 3829 case AF_INET6: 3830 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3831 ip6->ip6_vfc |= IPV6_VERSION; 3832 ip6->ip6_plen = htons(tlen - hlen); 3833 /* ip6_hlim will be initialized afterwards */ 3834 /* XXX flowlabel? */ 3835 break; 3836 #endif 3837 } 3838 3839 switch (sc->sc_src.sa.sa_family) { 3840 #ifdef INET 3841 case AF_INET: 3842 error = ip_output(m, sc->sc_ipopts, ro, 3843 (ip_mtudisc ? IP_MTUDISC : 0), 3844 NULL); 3845 break; 3846 #endif 3847 #ifdef INET6 3848 case AF_INET6: 3849 ip6->ip6_hlim = in6_selecthlim(NULL, 3850 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3851 3852 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3853 0, NULL, NULL); 3854 break; 3855 #endif 3856 default: 3857 error = EAFNOSUPPORT; 3858 break; 3859 } 3860 return (error); 3861 } 3862