1 /* $NetBSD: tcp_input.c,v 1.169 2003/06/15 02:49:33 matt 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 1.169 2003/06/15 02:49:33 matt 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/in_var.h> 184 #include <netinet/ip_var.h> 185 186 #ifdef INET6 187 #ifndef INET 188 #include <netinet/in.h> 189 #endif 190 #include <netinet/ip6.h> 191 #include <netinet6/ip6_var.h> 192 #include <netinet6/in6_pcb.h> 193 #include <netinet6/ip6_var.h> 194 #include <netinet6/in6_var.h> 195 #include <netinet/icmp6.h> 196 #include <netinet6/nd6.h> 197 #endif 198 199 #ifndef INET6 200 /* always need ip6.h for IP6_EXTHDR_GET */ 201 #include <netinet/ip6.h> 202 #endif 203 204 #include <netinet/tcp.h> 205 #include <netinet/tcp_fsm.h> 206 #include <netinet/tcp_seq.h> 207 #include <netinet/tcp_timer.h> 208 #include <netinet/tcp_var.h> 209 #include <netinet/tcpip.h> 210 #include <netinet/tcp_debug.h> 211 212 #include <machine/stdarg.h> 213 214 #ifdef IPSEC 215 #include <netinet6/ipsec.h> 216 #include <netkey/key.h> 217 #endif /*IPSEC*/ 218 #ifdef INET6 219 #include "faith.h" 220 #if defined(NFAITH) && NFAITH > 0 221 #include <net/if_faith.h> 222 #endif 223 #endif 224 225 int tcprexmtthresh = 3; 226 int tcp_log_refused; 227 228 static int tcp_rst_ppslim_count = 0; 229 static struct timeval tcp_rst_ppslim_last; 230 231 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 232 233 /* for modulo comparisons of timestamps */ 234 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 235 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 236 237 /* 238 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 239 */ 240 #ifdef INET6 241 #define ND6_HINT(tp) \ 242 do { \ 243 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 244 && tp->t_in6pcb->in6p_route.ro_rt) { \ 245 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 246 } \ 247 } while (/*CONSTCOND*/ 0) 248 #else 249 #define ND6_HINT(tp) 250 #endif 251 252 /* 253 * Macro to compute ACK transmission behavior. Delay the ACK unless 254 * we have already delayed an ACK (must send an ACK every two segments). 255 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 256 * option is enabled. 257 */ 258 #define TCP_SETUP_ACK(tp, th) \ 259 do { \ 260 if ((tp)->t_flags & TF_DELACK || \ 261 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 262 tp->t_flags |= TF_ACKNOW; \ 263 else \ 264 TCP_SET_DELACK(tp); \ 265 } while (/*CONSTCOND*/ 0) 266 267 /* 268 * Convert TCP protocol fields to host order for easier processing. 269 */ 270 #define TCP_FIELDS_TO_HOST(th) \ 271 do { \ 272 NTOHL((th)->th_seq); \ 273 NTOHL((th)->th_ack); \ 274 NTOHS((th)->th_win); \ 275 NTOHS((th)->th_urp); \ 276 } while (/*CONSTCOND*/ 0) 277 278 /* 279 * ... and reverse the above. 280 */ 281 #define TCP_FIELDS_TO_NET(th) \ 282 do { \ 283 HTONL((th)->th_seq); \ 284 HTONL((th)->th_ack); \ 285 HTONS((th)->th_win); \ 286 HTONS((th)->th_urp); \ 287 } while (/*CONSTCOND*/ 0) 288 289 #ifdef TCP_CSUM_COUNTERS 290 #include <sys/device.h> 291 292 extern struct evcnt tcp_hwcsum_ok; 293 extern struct evcnt tcp_hwcsum_bad; 294 extern struct evcnt tcp_hwcsum_data; 295 extern struct evcnt tcp_swcsum; 296 297 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 298 299 #else 300 301 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 302 303 #endif /* TCP_CSUM_COUNTERS */ 304 305 #ifdef TCP_REASS_COUNTERS 306 #include <sys/device.h> 307 308 extern struct evcnt tcp_reass_; 309 extern struct evcnt tcp_reass_empty; 310 extern struct evcnt tcp_reass_iteration[8]; 311 extern struct evcnt tcp_reass_prependfirst; 312 extern struct evcnt tcp_reass_prepend; 313 extern struct evcnt tcp_reass_insert; 314 extern struct evcnt tcp_reass_inserttail; 315 extern struct evcnt tcp_reass_append; 316 extern struct evcnt tcp_reass_appendtail; 317 extern struct evcnt tcp_reass_overlaptail; 318 extern struct evcnt tcp_reass_overlapfront; 319 extern struct evcnt tcp_reass_segdup; 320 extern struct evcnt tcp_reass_fragdup; 321 322 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 323 324 #else 325 326 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 327 328 #endif /* TCP_REASS_COUNTERS */ 329 330 #ifdef INET 331 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *)); 332 #endif 333 #ifdef INET6 334 static void tcp6_log_refused 335 __P((const struct ip6_hdr *, const struct tcphdr *)); 336 #endif 337 338 int 339 tcp_reass(tp, th, m, tlen) 340 struct tcpcb *tp; 341 struct tcphdr *th; 342 struct mbuf *m; 343 int *tlen; 344 { 345 struct ipqent *p, *q, *nq, *tiqe = NULL; 346 struct socket *so = NULL; 347 int pkt_flags; 348 tcp_seq pkt_seq; 349 unsigned pkt_len; 350 u_long rcvpartdupbyte = 0; 351 u_long rcvoobyte; 352 #ifdef TCP_REASS_COUNTERS 353 u_int count = 0; 354 #endif 355 356 if (tp->t_inpcb) 357 so = tp->t_inpcb->inp_socket; 358 #ifdef INET6 359 else if (tp->t_in6pcb) 360 so = tp->t_in6pcb->in6p_socket; 361 #endif 362 363 TCP_REASS_LOCK_CHECK(tp); 364 365 /* 366 * Call with th==0 after become established to 367 * force pre-ESTABLISHED data up to user socket. 368 */ 369 if (th == 0) 370 goto present; 371 372 rcvoobyte = *tlen; 373 /* 374 * Copy these to local variables because the tcpiphdr 375 * gets munged while we are collapsing mbufs. 376 */ 377 pkt_seq = th->th_seq; 378 pkt_len = *tlen; 379 pkt_flags = th->th_flags; 380 381 TCP_REASS_COUNTER_INCR(&tcp_reass_); 382 383 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 384 /* 385 * When we miss a packet, the vast majority of time we get 386 * packets that follow it in order. So optimize for that. 387 */ 388 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 389 p->ipqe_len += pkt_len; 390 p->ipqe_flags |= pkt_flags; 391 m_cat(p->ipqe_m, m); 392 tiqe = p; 393 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 394 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 395 goto skip_replacement; 396 } 397 /* 398 * While we're here, if the pkt is completely beyond 399 * anything we have, just insert it at the tail. 400 */ 401 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 402 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 403 goto insert_it; 404 } 405 } 406 407 q = TAILQ_FIRST(&tp->segq); 408 409 if (q != NULL) { 410 /* 411 * If this segment immediately precedes the first out-of-order 412 * block, simply slap the segment in front of it and (mostly) 413 * skip the complicated logic. 414 */ 415 if (pkt_seq + pkt_len == q->ipqe_seq) { 416 q->ipqe_seq = pkt_seq; 417 q->ipqe_len += pkt_len; 418 q->ipqe_flags |= pkt_flags; 419 m_cat(m, q->ipqe_m); 420 q->ipqe_m = m; 421 tiqe = q; 422 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 423 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 424 goto skip_replacement; 425 } 426 } else { 427 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 428 } 429 430 /* 431 * Find a segment which begins after this one does. 432 */ 433 for (p = NULL; q != NULL; q = nq) { 434 nq = TAILQ_NEXT(q, ipqe_q); 435 #ifdef TCP_REASS_COUNTERS 436 count++; 437 #endif 438 /* 439 * If the received segment is just right after this 440 * fragment, merge the two together and then check 441 * for further overlaps. 442 */ 443 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 444 #ifdef TCPREASS_DEBUG 445 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 446 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 447 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 448 #endif 449 pkt_len += q->ipqe_len; 450 pkt_flags |= q->ipqe_flags; 451 pkt_seq = q->ipqe_seq; 452 m_cat(q->ipqe_m, m); 453 m = q->ipqe_m; 454 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 455 goto free_ipqe; 456 } 457 /* 458 * If the received segment is completely past this 459 * fragment, we need to go the next fragment. 460 */ 461 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 462 p = q; 463 continue; 464 } 465 /* 466 * If the fragment is past the received segment, 467 * it (or any following) can't be concatenated. 468 */ 469 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 470 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 471 break; 472 } 473 474 /* 475 * We've received all the data in this segment before. 476 * mark it as a duplicate and return. 477 */ 478 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 479 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 480 tcpstat.tcps_rcvduppack++; 481 tcpstat.tcps_rcvdupbyte += pkt_len; 482 m_freem(m); 483 if (tiqe != NULL) 484 pool_put(&ipqent_pool, tiqe); 485 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 486 return (0); 487 } 488 /* 489 * Received segment completely overlaps this fragment 490 * so we drop the fragment (this keeps the temporal 491 * ordering of segments correct). 492 */ 493 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 494 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 495 rcvpartdupbyte += q->ipqe_len; 496 m_freem(q->ipqe_m); 497 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 498 goto free_ipqe; 499 } 500 /* 501 * RX'ed segment extends past the end of the 502 * fragment. Drop the overlapping bytes. Then 503 * merge the fragment and segment then treat as 504 * a longer received packet. 505 */ 506 if (SEQ_LT(q->ipqe_seq, pkt_seq) 507 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 508 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 509 #ifdef TCPREASS_DEBUG 510 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 511 tp, overlap, 512 pkt_seq, pkt_seq + pkt_len, pkt_len); 513 #endif 514 m_adj(m, overlap); 515 rcvpartdupbyte += overlap; 516 m_cat(q->ipqe_m, m); 517 m = q->ipqe_m; 518 pkt_seq = q->ipqe_seq; 519 pkt_len += q->ipqe_len - overlap; 520 rcvoobyte -= overlap; 521 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 522 goto free_ipqe; 523 } 524 /* 525 * RX'ed segment extends past the front of the 526 * fragment. Drop the overlapping bytes on the 527 * received packet. The packet will then be 528 * contatentated with this fragment a bit later. 529 */ 530 if (SEQ_GT(q->ipqe_seq, pkt_seq) 531 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 532 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 533 #ifdef TCPREASS_DEBUG 534 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 535 tp, overlap, 536 pkt_seq, pkt_seq + pkt_len, pkt_len); 537 #endif 538 m_adj(m, -overlap); 539 pkt_len -= overlap; 540 rcvpartdupbyte += overlap; 541 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 542 rcvoobyte -= overlap; 543 } 544 /* 545 * If the received segment immediates precedes this 546 * fragment then tack the fragment onto this segment 547 * and reinsert the data. 548 */ 549 if (q->ipqe_seq == pkt_seq + pkt_len) { 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 552 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 553 pkt_seq, pkt_seq + pkt_len, pkt_len); 554 #endif 555 pkt_len += q->ipqe_len; 556 pkt_flags |= q->ipqe_flags; 557 m_cat(m, q->ipqe_m); 558 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 559 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 560 if (tiqe == NULL) { 561 tiqe = q; 562 } else { 563 pool_put(&ipqent_pool, q); 564 } 565 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 566 break; 567 } 568 /* 569 * If the fragment is before the segment, remember it. 570 * When this loop is terminated, p will contain the 571 * pointer to fragment that is right before the received 572 * segment. 573 */ 574 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 575 p = q; 576 577 continue; 578 579 /* 580 * This is a common operation. It also will allow 581 * to save doing a malloc/free in most instances. 582 */ 583 free_ipqe: 584 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 585 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 586 if (tiqe == NULL) { 587 tiqe = q; 588 } else { 589 pool_put(&ipqent_pool, q); 590 } 591 } 592 593 #ifdef TCP_REASS_COUNTERS 594 if (count > 7) 595 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 596 else if (count > 0) 597 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 598 #endif 599 600 insert_it: 601 602 /* 603 * Allocate a new queue entry since the received segment did not 604 * collapse onto any other out-of-order block; thus we are allocating 605 * a new block. If it had collapsed, tiqe would not be NULL and 606 * we would be reusing it. 607 * XXX If we can't, just drop the packet. XXX 608 */ 609 if (tiqe == NULL) { 610 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 611 if (tiqe == NULL) { 612 tcpstat.tcps_rcvmemdrop++; 613 m_freem(m); 614 return (0); 615 } 616 } 617 618 /* 619 * Update the counters. 620 */ 621 tcpstat.tcps_rcvoopack++; 622 tcpstat.tcps_rcvoobyte += rcvoobyte; 623 if (rcvpartdupbyte) { 624 tcpstat.tcps_rcvpartduppack++; 625 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 626 } 627 628 /* 629 * Insert the new fragment queue entry into both queues. 630 */ 631 tiqe->ipqe_m = m; 632 tiqe->ipqe_seq = pkt_seq; 633 tiqe->ipqe_len = pkt_len; 634 tiqe->ipqe_flags = pkt_flags; 635 if (p == NULL) { 636 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 637 #ifdef TCPREASS_DEBUG 638 if (tiqe->ipqe_seq != tp->rcv_nxt) 639 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 640 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 641 #endif 642 } else { 643 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 644 #ifdef TCPREASS_DEBUG 645 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 646 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 647 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 648 #endif 649 } 650 651 skip_replacement: 652 653 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 654 655 present: 656 /* 657 * Present data to user, advancing rcv_nxt through 658 * completed sequence space. 659 */ 660 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 661 return (0); 662 q = TAILQ_FIRST(&tp->segq); 663 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 664 return (0); 665 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 666 return (0); 667 668 tp->rcv_nxt += q->ipqe_len; 669 pkt_flags = q->ipqe_flags & TH_FIN; 670 ND6_HINT(tp); 671 672 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 673 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 674 if (so->so_state & SS_CANTRCVMORE) 675 m_freem(q->ipqe_m); 676 else 677 sbappendstream(&so->so_rcv, q->ipqe_m); 678 pool_put(&ipqent_pool, q); 679 sorwakeup(so); 680 return (pkt_flags); 681 } 682 683 #ifdef INET6 684 int 685 tcp6_input(mp, offp, proto) 686 struct mbuf **mp; 687 int *offp, proto; 688 { 689 struct mbuf *m = *mp; 690 691 /* 692 * draft-itojun-ipv6-tcp-to-anycast 693 * better place to put this in? 694 */ 695 if (m->m_flags & M_ANYCAST6) { 696 struct ip6_hdr *ip6; 697 if (m->m_len < sizeof(struct ip6_hdr)) { 698 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 699 tcpstat.tcps_rcvshort++; 700 return IPPROTO_DONE; 701 } 702 } 703 ip6 = mtod(m, struct ip6_hdr *); 704 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 705 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 706 return IPPROTO_DONE; 707 } 708 709 tcp_input(m, *offp, proto); 710 return IPPROTO_DONE; 711 } 712 #endif 713 714 #ifdef INET 715 static void 716 tcp4_log_refused(ip, th) 717 const struct ip *ip; 718 const struct tcphdr *th; 719 { 720 char src[4*sizeof "123"]; 721 char dst[4*sizeof "123"]; 722 723 if (ip) { 724 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 725 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 726 } 727 else { 728 strlcpy(src, "(unknown)", sizeof(src)); 729 strlcpy(dst, "(unknown)", sizeof(dst)); 730 } 731 log(LOG_INFO, 732 "Connection attempt to TCP %s:%d from %s:%d\n", 733 dst, ntohs(th->th_dport), 734 src, ntohs(th->th_sport)); 735 } 736 #endif 737 738 #ifdef INET6 739 static void 740 tcp6_log_refused(ip6, th) 741 const struct ip6_hdr *ip6; 742 const struct tcphdr *th; 743 { 744 char src[INET6_ADDRSTRLEN]; 745 char dst[INET6_ADDRSTRLEN]; 746 747 if (ip6) { 748 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 749 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 750 } 751 else { 752 strlcpy(src, "(unknown v6)", sizeof(src)); 753 strlcpy(dst, "(unknown v6)", sizeof(dst)); 754 } 755 log(LOG_INFO, 756 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 757 dst, ntohs(th->th_dport), 758 src, ntohs(th->th_sport)); 759 } 760 #endif 761 762 /* 763 * TCP input routine, follows pages 65-76 of the 764 * protocol specification dated September, 1981 very closely. 765 */ 766 void 767 #if __STDC__ 768 tcp_input(struct mbuf *m, ...) 769 #else 770 tcp_input(m, va_alist) 771 struct mbuf *m; 772 #endif 773 { 774 struct tcphdr *th; 775 struct ip *ip; 776 struct inpcb *inp; 777 #ifdef INET6 778 struct ip6_hdr *ip6; 779 struct in6pcb *in6p; 780 #endif 781 u_int8_t *optp = NULL; 782 int optlen = 0; 783 int len, tlen, toff, hdroptlen = 0; 784 struct tcpcb *tp = 0; 785 int tiflags; 786 struct socket *so = NULL; 787 int todrop, acked, ourfinisacked, needoutput = 0; 788 #ifdef TCP_DEBUG 789 short ostate = 0; 790 #endif 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 MCLAIM(m, &tcp_rx_mowner); 800 va_start(ap, m); 801 toff = va_arg(ap, int); 802 (void)va_arg(ap, int); /* ignore value, advance ap */ 803 va_end(ap); 804 805 tcpstat.tcps_rcvtotal++; 806 807 bzero(&opti, sizeof(opti)); 808 opti.ts_present = 0; 809 opti.maxseg = 0; 810 811 /* 812 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 813 * 814 * TCP is, by definition, unicast, so we reject all 815 * multicast outright. 816 * 817 * Note, there are additional src/dst address checks in 818 * the AF-specific code below. 819 */ 820 if (m->m_flags & (M_BCAST|M_MCAST)) { 821 /* XXX stat */ 822 goto drop; 823 } 824 #ifdef INET6 825 if (m->m_flags & M_ANYCAST6) { 826 /* XXX stat */ 827 goto drop; 828 } 829 #endif 830 831 /* 832 * Get IP and TCP header together in first mbuf. 833 * Note: IP leaves IP header in first mbuf. 834 */ 835 ip = mtod(m, struct ip *); 836 #ifdef INET6 837 ip6 = NULL; 838 #endif 839 switch (ip->ip_v) { 840 #ifdef INET 841 case 4: 842 af = AF_INET; 843 iphlen = sizeof(struct ip); 844 ip = mtod(m, struct ip *); 845 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 846 sizeof(struct tcphdr)); 847 if (th == NULL) { 848 tcpstat.tcps_rcvshort++; 849 return; 850 } 851 /* We do the checksum after PCB lookup... */ 852 len = ntohs(ip->ip_len); 853 tlen = len - toff; 854 break; 855 #endif 856 #ifdef INET6 857 case 6: 858 ip = NULL; 859 iphlen = sizeof(struct ip6_hdr); 860 af = AF_INET6; 861 ip6 = mtod(m, struct ip6_hdr *); 862 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 863 sizeof(struct tcphdr)); 864 if (th == NULL) { 865 tcpstat.tcps_rcvshort++; 866 return; 867 } 868 869 /* Be proactive about malicious use of IPv4 mapped address */ 870 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 871 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 872 /* XXX stat */ 873 goto drop; 874 } 875 876 /* 877 * Be proactive about unspecified IPv6 address in source. 878 * As we use all-zero to indicate unbounded/unconnected pcb, 879 * unspecified IPv6 address can be used to confuse us. 880 * 881 * Note that packets with unspecified IPv6 destination is 882 * already dropped in ip6_input. 883 */ 884 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 885 /* XXX stat */ 886 goto drop; 887 } 888 889 /* 890 * Make sure destination address is not multicast. 891 * Source address checked in ip6_input(). 892 */ 893 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 894 /* XXX stat */ 895 goto drop; 896 } 897 898 /* We do the checksum after PCB lookup... */ 899 len = m->m_pkthdr.len; 900 tlen = len - toff; 901 break; 902 #endif 903 default: 904 m_freem(m); 905 return; 906 } 907 908 KASSERT(TCP_HDR_ALIGNED_P(th)); 909 910 /* 911 * Check that TCP offset makes sense, 912 * pull out TCP options and adjust length. XXX 913 */ 914 off = th->th_off << 2; 915 if (off < sizeof (struct tcphdr) || off > tlen) { 916 tcpstat.tcps_rcvbadoff++; 917 goto drop; 918 } 919 tlen -= off; 920 921 /* 922 * tcp_input() has been modified to use tlen to mean the TCP data 923 * length throughout the function. Other functions can use 924 * m->m_pkthdr.len as the basis for calculating the TCP data length. 925 * rja 926 */ 927 928 if (off > sizeof (struct tcphdr)) { 929 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 930 if (th == NULL) { 931 tcpstat.tcps_rcvshort++; 932 return; 933 } 934 /* 935 * NOTE: ip/ip6 will not be affected by m_pulldown() 936 * (as they're before toff) and we don't need to update those. 937 */ 938 KASSERT(TCP_HDR_ALIGNED_P(th)); 939 optlen = off - sizeof (struct tcphdr); 940 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 941 /* 942 * Do quick retrieval of timestamp options ("options 943 * prediction?"). If timestamp is the only option and it's 944 * formatted as recommended in RFC 1323 appendix A, we 945 * quickly get the values now and not bother calling 946 * tcp_dooptions(), etc. 947 */ 948 if ((optlen == TCPOLEN_TSTAMP_APPA || 949 (optlen > TCPOLEN_TSTAMP_APPA && 950 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 951 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 952 (th->th_flags & TH_SYN) == 0) { 953 opti.ts_present = 1; 954 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 955 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 956 optp = NULL; /* we've parsed the options */ 957 } 958 } 959 tiflags = th->th_flags; 960 961 /* 962 * Locate pcb for segment. 963 */ 964 findpcb: 965 inp = NULL; 966 #ifdef INET6 967 in6p = NULL; 968 #endif 969 switch (af) { 970 #ifdef INET 971 case AF_INET: 972 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 973 ip->ip_dst, th->th_dport); 974 if (inp == 0) { 975 ++tcpstat.tcps_pcbhashmiss; 976 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 977 } 978 #ifdef INET6 979 if (inp == 0) { 980 struct in6_addr s, d; 981 982 /* mapped addr case */ 983 bzero(&s, sizeof(s)); 984 s.s6_addr16[5] = htons(0xffff); 985 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 986 bzero(&d, sizeof(d)); 987 d.s6_addr16[5] = htons(0xffff); 988 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 989 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 990 &d, th->th_dport, 0); 991 if (in6p == 0) { 992 ++tcpstat.tcps_pcbhashmiss; 993 in6p = in6_pcblookup_bind(&tcb6, &d, 994 th->th_dport, 0); 995 } 996 } 997 #endif 998 #ifndef INET6 999 if (inp == 0) 1000 #else 1001 if (inp == 0 && in6p == 0) 1002 #endif 1003 { 1004 ++tcpstat.tcps_noport; 1005 if (tcp_log_refused && 1006 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1007 tcp4_log_refused(ip, th); 1008 } 1009 TCP_FIELDS_TO_HOST(th); 1010 goto dropwithreset_ratelim; 1011 } 1012 #ifdef IPSEC 1013 if (inp && ipsec4_in_reject(m, inp)) { 1014 ipsecstat.in_polvio++; 1015 goto drop; 1016 } 1017 #ifdef INET6 1018 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1019 ipsecstat.in_polvio++; 1020 goto drop; 1021 } 1022 #endif 1023 #endif /*IPSEC*/ 1024 break; 1025 #endif /*INET*/ 1026 #ifdef INET6 1027 case AF_INET6: 1028 { 1029 int faith; 1030 1031 #if defined(NFAITH) && NFAITH > 0 1032 faith = faithprefix(&ip6->ip6_dst); 1033 #else 1034 faith = 0; 1035 #endif 1036 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 1037 &ip6->ip6_dst, th->th_dport, faith); 1038 if (in6p == NULL) { 1039 ++tcpstat.tcps_pcbhashmiss; 1040 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 1041 th->th_dport, faith); 1042 } 1043 if (in6p == NULL) { 1044 ++tcpstat.tcps_noport; 1045 if (tcp_log_refused && 1046 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1047 tcp6_log_refused(ip6, th); 1048 } 1049 TCP_FIELDS_TO_HOST(th); 1050 goto dropwithreset_ratelim; 1051 } 1052 #ifdef IPSEC 1053 if (ipsec6_in_reject(m, in6p)) { 1054 ipsec6stat.in_polvio++; 1055 goto drop; 1056 } 1057 #endif /*IPSEC*/ 1058 break; 1059 } 1060 #endif 1061 } 1062 1063 /* 1064 * If the state is CLOSED (i.e., TCB does not exist) then 1065 * all data in the incoming segment is discarded. 1066 * If the TCB exists but is in CLOSED state, it is embryonic, 1067 * but should either do a listen or a connect soon. 1068 */ 1069 tp = NULL; 1070 so = NULL; 1071 if (inp) { 1072 tp = intotcpcb(inp); 1073 so = inp->inp_socket; 1074 } 1075 #ifdef INET6 1076 else if (in6p) { 1077 tp = in6totcpcb(in6p); 1078 so = in6p->in6p_socket; 1079 } 1080 #endif 1081 if (tp == 0) { 1082 TCP_FIELDS_TO_HOST(th); 1083 goto dropwithreset_ratelim; 1084 } 1085 if (tp->t_state == TCPS_CLOSED) 1086 goto drop; 1087 1088 /* 1089 * Checksum extended TCP header and data. 1090 */ 1091 switch (af) { 1092 #ifdef INET 1093 case AF_INET: 1094 switch (m->m_pkthdr.csum_flags & 1095 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 1096 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 1097 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 1098 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 1099 goto badcsum; 1100 1101 case M_CSUM_TCPv4|M_CSUM_DATA: 1102 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 1103 if ((m->m_pkthdr.csum_data ^ 0xffff) != 0) 1104 goto badcsum; 1105 break; 1106 1107 case M_CSUM_TCPv4: 1108 /* Checksum was okay. */ 1109 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 1110 break; 1111 1112 default: 1113 /* Must compute it ourselves. */ 1114 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 1115 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1116 goto badcsum; 1117 break; 1118 } 1119 break; 1120 #endif /* INET4 */ 1121 1122 #ifdef INET6 1123 case AF_INET6: 1124 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1125 goto badcsum; 1126 break; 1127 #endif /* INET6 */ 1128 } 1129 1130 TCP_FIELDS_TO_HOST(th); 1131 1132 /* Unscale the window into a 32-bit value. */ 1133 if ((tiflags & TH_SYN) == 0) 1134 tiwin = th->th_win << tp->snd_scale; 1135 else 1136 tiwin = th->th_win; 1137 1138 #ifdef INET6 1139 /* save packet options if user wanted */ 1140 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1141 if (in6p->in6p_options) { 1142 m_freem(in6p->in6p_options); 1143 in6p->in6p_options = 0; 1144 } 1145 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1146 } 1147 #endif 1148 1149 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1150 union syn_cache_sa src; 1151 union syn_cache_sa dst; 1152 1153 bzero(&src, sizeof(src)); 1154 bzero(&dst, sizeof(dst)); 1155 switch (af) { 1156 #ifdef INET 1157 case AF_INET: 1158 src.sin.sin_len = sizeof(struct sockaddr_in); 1159 src.sin.sin_family = AF_INET; 1160 src.sin.sin_addr = ip->ip_src; 1161 src.sin.sin_port = th->th_sport; 1162 1163 dst.sin.sin_len = sizeof(struct sockaddr_in); 1164 dst.sin.sin_family = AF_INET; 1165 dst.sin.sin_addr = ip->ip_dst; 1166 dst.sin.sin_port = th->th_dport; 1167 break; 1168 #endif 1169 #ifdef INET6 1170 case AF_INET6: 1171 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1172 src.sin6.sin6_family = AF_INET6; 1173 src.sin6.sin6_addr = ip6->ip6_src; 1174 src.sin6.sin6_port = th->th_sport; 1175 1176 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1177 dst.sin6.sin6_family = AF_INET6; 1178 dst.sin6.sin6_addr = ip6->ip6_dst; 1179 dst.sin6.sin6_port = th->th_dport; 1180 break; 1181 #endif /* INET6 */ 1182 default: 1183 goto badsyn; /*sanity*/ 1184 } 1185 1186 if (so->so_options & SO_DEBUG) { 1187 #ifdef TCP_DEBUG 1188 ostate = tp->t_state; 1189 #endif 1190 1191 tcp_saveti = NULL; 1192 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1193 goto nosave; 1194 1195 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1196 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1197 if (!tcp_saveti) 1198 goto nosave; 1199 } else { 1200 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1201 if (!tcp_saveti) 1202 goto nosave; 1203 MCLAIM(m, &tcp_mowner); 1204 tcp_saveti->m_len = iphlen; 1205 m_copydata(m, 0, iphlen, 1206 mtod(tcp_saveti, caddr_t)); 1207 } 1208 1209 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1210 m_freem(tcp_saveti); 1211 tcp_saveti = NULL; 1212 } else { 1213 tcp_saveti->m_len += sizeof(struct tcphdr); 1214 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1215 sizeof(struct tcphdr)); 1216 } 1217 nosave:; 1218 } 1219 if (so->so_options & SO_ACCEPTCONN) { 1220 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1221 if (tiflags & TH_RST) { 1222 syn_cache_reset(&src.sa, &dst.sa, th); 1223 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1224 (TH_ACK|TH_SYN)) { 1225 /* 1226 * Received a SYN,ACK. This should 1227 * never happen while we are in 1228 * LISTEN. Send an RST. 1229 */ 1230 goto badsyn; 1231 } else if (tiflags & TH_ACK) { 1232 so = syn_cache_get(&src.sa, &dst.sa, 1233 th, toff, tlen, so, m); 1234 if (so == NULL) { 1235 /* 1236 * We don't have a SYN for 1237 * this ACK; send an RST. 1238 */ 1239 goto badsyn; 1240 } else if (so == 1241 (struct socket *)(-1)) { 1242 /* 1243 * We were unable to create 1244 * the connection. If the 1245 * 3-way handshake was 1246 * completed, and RST has 1247 * been sent to the peer. 1248 * Since the mbuf might be 1249 * in use for the reply, 1250 * do not free it. 1251 */ 1252 m = NULL; 1253 } else { 1254 /* 1255 * We have created a 1256 * full-blown connection. 1257 */ 1258 tp = NULL; 1259 inp = NULL; 1260 #ifdef INET6 1261 in6p = NULL; 1262 #endif 1263 switch (so->so_proto->pr_domain->dom_family) { 1264 #ifdef INET 1265 case AF_INET: 1266 inp = sotoinpcb(so); 1267 tp = intotcpcb(inp); 1268 break; 1269 #endif 1270 #ifdef INET6 1271 case AF_INET6: 1272 in6p = sotoin6pcb(so); 1273 tp = in6totcpcb(in6p); 1274 break; 1275 #endif 1276 } 1277 if (tp == NULL) 1278 goto badsyn; /*XXX*/ 1279 tiwin <<= tp->snd_scale; 1280 goto after_listen; 1281 } 1282 } else { 1283 /* 1284 * None of RST, SYN or ACK was set. 1285 * This is an invalid packet for a 1286 * TCB in LISTEN state. Send a RST. 1287 */ 1288 goto badsyn; 1289 } 1290 } else { 1291 /* 1292 * Received a SYN. 1293 */ 1294 1295 #ifdef INET6 1296 /* 1297 * If deprecated address is forbidden, we do 1298 * not accept SYN to deprecated interface 1299 * address to prevent any new inbound 1300 * connection from getting established. 1301 * When we do not accept SYN, we send a TCP 1302 * RST, with deprecated source address (instead 1303 * of dropping it). We compromise it as it is 1304 * much better for peer to send a RST, and 1305 * RST will be the final packet for the 1306 * exchange. 1307 * 1308 * If we do not forbid deprecated addresses, we 1309 * accept the SYN packet. RFC2462 does not 1310 * suggest dropping SYN in this case. 1311 * If we decipher RFC2462 5.5.4, it says like 1312 * this: 1313 * 1. use of deprecated addr with existing 1314 * communication is okay - "SHOULD continue 1315 * to be used" 1316 * 2. use of it with new communication: 1317 * (2a) "SHOULD NOT be used if alternate 1318 * address with sufficient scope is 1319 * available" 1320 * (2b) nothing mentioned otherwise. 1321 * Here we fall into (2b) case as we have no 1322 * choice in our source address selection - we 1323 * must obey the peer. 1324 * 1325 * The wording in RFC2462 is confusing, and 1326 * there are multiple description text for 1327 * deprecated address handling - worse, they 1328 * are not exactly the same. I believe 5.5.4 1329 * is the best one, so we follow 5.5.4. 1330 */ 1331 if (af == AF_INET6 && !ip6_use_deprecated) { 1332 struct in6_ifaddr *ia6; 1333 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1334 &ip6->ip6_dst)) && 1335 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1336 tp = NULL; 1337 goto dropwithreset; 1338 } 1339 } 1340 #endif 1341 1342 /* 1343 * LISTEN socket received a SYN 1344 * from itself? This can't possibly 1345 * be valid; drop the packet. 1346 */ 1347 if (th->th_sport == th->th_dport) { 1348 int i; 1349 1350 switch (af) { 1351 #ifdef INET 1352 case AF_INET: 1353 i = in_hosteq(ip->ip_src, ip->ip_dst); 1354 break; 1355 #endif 1356 #ifdef INET6 1357 case AF_INET6: 1358 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1359 break; 1360 #endif 1361 default: 1362 i = 1; 1363 } 1364 if (i) { 1365 tcpstat.tcps_badsyn++; 1366 goto drop; 1367 } 1368 } 1369 1370 /* 1371 * SYN looks ok; create compressed TCP 1372 * state for it. 1373 */ 1374 if (so->so_qlen <= so->so_qlimit && 1375 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1376 so, m, optp, optlen, &opti)) 1377 m = NULL; 1378 } 1379 goto drop; 1380 } 1381 } 1382 1383 after_listen: 1384 #ifdef DIAGNOSTIC 1385 /* 1386 * Should not happen now that all embryonic connections 1387 * are handled with compressed state. 1388 */ 1389 if (tp->t_state == TCPS_LISTEN) 1390 panic("tcp_input: TCPS_LISTEN"); 1391 #endif 1392 1393 /* 1394 * Segment received on connection. 1395 * Reset idle time and keep-alive timer. 1396 */ 1397 tp->t_rcvtime = tcp_now; 1398 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1399 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1400 1401 /* 1402 * Process options. 1403 */ 1404 if (optp) 1405 tcp_dooptions(tp, optp, optlen, th, &opti); 1406 1407 /* 1408 * Header prediction: check for the two common cases 1409 * of a uni-directional data xfer. If the packet has 1410 * no control flags, is in-sequence, the window didn't 1411 * change and we're not retransmitting, it's a 1412 * candidate. If the length is zero and the ack moved 1413 * forward, we're the sender side of the xfer. Just 1414 * free the data acked & wake any higher level process 1415 * that was blocked waiting for space. If the length 1416 * is non-zero and the ack didn't move, we're the 1417 * receiver side. If we're getting packets in-order 1418 * (the reassembly queue is empty), add the data to 1419 * the socket buffer and note that we need a delayed ack. 1420 */ 1421 if (tp->t_state == TCPS_ESTABLISHED && 1422 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1423 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1424 th->th_seq == tp->rcv_nxt && 1425 tiwin && tiwin == tp->snd_wnd && 1426 tp->snd_nxt == tp->snd_max) { 1427 1428 /* 1429 * If last ACK falls within this segment's sequence numbers, 1430 * record the timestamp. 1431 */ 1432 if (opti.ts_present && 1433 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1434 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1435 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1436 tp->ts_recent = opti.ts_val; 1437 } 1438 1439 if (tlen == 0) { 1440 if (SEQ_GT(th->th_ack, tp->snd_una) && 1441 SEQ_LEQ(th->th_ack, tp->snd_max) && 1442 tp->snd_cwnd >= tp->snd_wnd && 1443 tp->t_dupacks < tcprexmtthresh) { 1444 /* 1445 * this is a pure ack for outstanding data. 1446 */ 1447 ++tcpstat.tcps_predack; 1448 if (opti.ts_present && opti.ts_ecr) 1449 tcp_xmit_timer(tp, 1450 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1451 else if (tp->t_rtttime && 1452 SEQ_GT(th->th_ack, tp->t_rtseq)) 1453 tcp_xmit_timer(tp, 1454 tcp_now - tp->t_rtttime); 1455 acked = th->th_ack - tp->snd_una; 1456 tcpstat.tcps_rcvackpack++; 1457 tcpstat.tcps_rcvackbyte += acked; 1458 ND6_HINT(tp); 1459 sbdrop(&so->so_snd, acked); 1460 /* 1461 * We want snd_recover to track snd_una to 1462 * avoid sequence wraparound problems for 1463 * very large transfers. 1464 */ 1465 tp->snd_una = tp->snd_recover = th->th_ack; 1466 m_freem(m); 1467 1468 /* 1469 * If all outstanding data are acked, stop 1470 * retransmit timer, otherwise restart timer 1471 * using current (possibly backed-off) value. 1472 * If process is waiting for space, 1473 * wakeup/selwakeup/signal. If data 1474 * are ready to send, let tcp_output 1475 * decide between more output or persist. 1476 */ 1477 if (tp->snd_una == tp->snd_max) 1478 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1479 else if (TCP_TIMER_ISARMED(tp, 1480 TCPT_PERSIST) == 0) 1481 TCP_TIMER_ARM(tp, TCPT_REXMT, 1482 tp->t_rxtcur); 1483 1484 sowwakeup(so); 1485 if (so->so_snd.sb_cc) 1486 (void) tcp_output(tp); 1487 if (tcp_saveti) 1488 m_freem(tcp_saveti); 1489 return; 1490 } 1491 } else if (th->th_ack == tp->snd_una && 1492 TAILQ_FIRST(&tp->segq) == NULL && 1493 tlen <= sbspace(&so->so_rcv)) { 1494 /* 1495 * this is a pure, in-sequence data packet 1496 * with nothing on the reassembly queue and 1497 * we have enough buffer space to take it. 1498 */ 1499 ++tcpstat.tcps_preddat; 1500 tp->rcv_nxt += tlen; 1501 tcpstat.tcps_rcvpack++; 1502 tcpstat.tcps_rcvbyte += tlen; 1503 ND6_HINT(tp); 1504 /* 1505 * Drop TCP, IP headers and TCP options then add data 1506 * to socket buffer. 1507 */ 1508 if (so->so_state & SS_CANTRCVMORE) 1509 m_freem(m); 1510 else { 1511 m_adj(m, toff + off); 1512 sbappendstream(&so->so_rcv, m); 1513 } 1514 sorwakeup(so); 1515 TCP_SETUP_ACK(tp, th); 1516 if (tp->t_flags & TF_ACKNOW) 1517 (void) tcp_output(tp); 1518 if (tcp_saveti) 1519 m_freem(tcp_saveti); 1520 return; 1521 } 1522 } 1523 1524 /* 1525 * Compute mbuf offset to TCP data segment. 1526 */ 1527 hdroptlen = toff + off; 1528 1529 /* 1530 * Calculate amount of space in receive window, 1531 * and then do TCP input processing. 1532 * Receive window is amount of space in rcv queue, 1533 * but not less than advertised window. 1534 */ 1535 { int win; 1536 1537 win = sbspace(&so->so_rcv); 1538 if (win < 0) 1539 win = 0; 1540 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1541 } 1542 1543 switch (tp->t_state) { 1544 case TCPS_LISTEN: 1545 /* 1546 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1547 */ 1548 if (m->m_flags & (M_BCAST|M_MCAST)) 1549 goto drop; 1550 switch (af) { 1551 #ifdef INET6 1552 case AF_INET6: 1553 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1554 goto drop; 1555 break; 1556 #endif /* INET6 */ 1557 case AF_INET: 1558 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1559 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1560 goto drop; 1561 break; 1562 } 1563 break; 1564 1565 /* 1566 * If the state is SYN_SENT: 1567 * if seg contains an ACK, but not for our SYN, drop the input. 1568 * if seg contains a RST, then drop the connection. 1569 * if seg does not contain SYN, then drop it. 1570 * Otherwise this is an acceptable SYN segment 1571 * initialize tp->rcv_nxt and tp->irs 1572 * if seg contains ack then advance tp->snd_una 1573 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1574 * arrange for segment to be acked (eventually) 1575 * continue processing rest of data/controls, beginning with URG 1576 */ 1577 case TCPS_SYN_SENT: 1578 if ((tiflags & TH_ACK) && 1579 (SEQ_LEQ(th->th_ack, tp->iss) || 1580 SEQ_GT(th->th_ack, tp->snd_max))) 1581 goto dropwithreset; 1582 if (tiflags & TH_RST) { 1583 if (tiflags & TH_ACK) 1584 tp = tcp_drop(tp, ECONNREFUSED); 1585 goto drop; 1586 } 1587 if ((tiflags & TH_SYN) == 0) 1588 goto drop; 1589 if (tiflags & TH_ACK) { 1590 tp->snd_una = tp->snd_recover = th->th_ack; 1591 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1592 tp->snd_nxt = tp->snd_una; 1593 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1594 } 1595 tp->irs = th->th_seq; 1596 tcp_rcvseqinit(tp); 1597 tp->t_flags |= TF_ACKNOW; 1598 tcp_mss_from_peer(tp, opti.maxseg); 1599 1600 /* 1601 * Initialize the initial congestion window. If we 1602 * had to retransmit the SYN, we must initialize cwnd 1603 * to 1 segment (i.e. the Loss Window). 1604 */ 1605 if (tp->t_flags & TF_SYN_REXMT) 1606 tp->snd_cwnd = tp->t_peermss; 1607 else { 1608 int ss = tcp_init_win; 1609 #ifdef INET 1610 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1611 ss = tcp_init_win_local; 1612 #endif 1613 #ifdef INET6 1614 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1615 ss = tcp_init_win_local; 1616 #endif 1617 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1618 } 1619 1620 tcp_rmx_rtt(tp); 1621 if (tiflags & TH_ACK) { 1622 tcpstat.tcps_connects++; 1623 soisconnected(so); 1624 tcp_established(tp); 1625 /* Do window scaling on this connection? */ 1626 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1627 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1628 tp->snd_scale = tp->requested_s_scale; 1629 tp->rcv_scale = tp->request_r_scale; 1630 } 1631 TCP_REASS_LOCK(tp); 1632 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1633 TCP_REASS_UNLOCK(tp); 1634 /* 1635 * if we didn't have to retransmit the SYN, 1636 * use its rtt as our initial srtt & rtt var. 1637 */ 1638 if (tp->t_rtttime) 1639 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1640 } else 1641 tp->t_state = TCPS_SYN_RECEIVED; 1642 1643 /* 1644 * Advance th->th_seq to correspond to first data byte. 1645 * If data, trim to stay within window, 1646 * dropping FIN if necessary. 1647 */ 1648 th->th_seq++; 1649 if (tlen > tp->rcv_wnd) { 1650 todrop = tlen - tp->rcv_wnd; 1651 m_adj(m, -todrop); 1652 tlen = tp->rcv_wnd; 1653 tiflags &= ~TH_FIN; 1654 tcpstat.tcps_rcvpackafterwin++; 1655 tcpstat.tcps_rcvbyteafterwin += todrop; 1656 } 1657 tp->snd_wl1 = th->th_seq - 1; 1658 tp->rcv_up = th->th_seq; 1659 goto step6; 1660 1661 /* 1662 * If the state is SYN_RECEIVED: 1663 * If seg contains an ACK, but not for our SYN, drop the input 1664 * and generate an RST. See page 36, rfc793 1665 */ 1666 case TCPS_SYN_RECEIVED: 1667 if ((tiflags & TH_ACK) && 1668 (SEQ_LEQ(th->th_ack, tp->iss) || 1669 SEQ_GT(th->th_ack, tp->snd_max))) 1670 goto dropwithreset; 1671 break; 1672 } 1673 1674 /* 1675 * States other than LISTEN or SYN_SENT. 1676 * First check timestamp, if present. 1677 * Then check that at least some bytes of segment are within 1678 * receive window. If segment begins before rcv_nxt, 1679 * drop leading data (and SYN); if nothing left, just ack. 1680 * 1681 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1682 * and it's less than ts_recent, drop it. 1683 */ 1684 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1685 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1686 1687 /* Check to see if ts_recent is over 24 days old. */ 1688 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1689 TCP_PAWS_IDLE) { 1690 /* 1691 * Invalidate ts_recent. If this segment updates 1692 * ts_recent, the age will be reset later and ts_recent 1693 * will get a valid value. If it does not, setting 1694 * ts_recent to zero will at least satisfy the 1695 * requirement that zero be placed in the timestamp 1696 * echo reply when ts_recent isn't valid. The 1697 * age isn't reset until we get a valid ts_recent 1698 * because we don't want out-of-order segments to be 1699 * dropped when ts_recent is old. 1700 */ 1701 tp->ts_recent = 0; 1702 } else { 1703 tcpstat.tcps_rcvduppack++; 1704 tcpstat.tcps_rcvdupbyte += tlen; 1705 tcpstat.tcps_pawsdrop++; 1706 goto dropafterack; 1707 } 1708 } 1709 1710 todrop = tp->rcv_nxt - th->th_seq; 1711 if (todrop > 0) { 1712 if (tiflags & TH_SYN) { 1713 tiflags &= ~TH_SYN; 1714 th->th_seq++; 1715 if (th->th_urp > 1) 1716 th->th_urp--; 1717 else { 1718 tiflags &= ~TH_URG; 1719 th->th_urp = 0; 1720 } 1721 todrop--; 1722 } 1723 if (todrop > tlen || 1724 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1725 /* 1726 * Any valid FIN must be to the left of the window. 1727 * At this point the FIN must be a duplicate or 1728 * out of sequence; drop it. 1729 */ 1730 tiflags &= ~TH_FIN; 1731 /* 1732 * Send an ACK to resynchronize and drop any data. 1733 * But keep on processing for RST or ACK. 1734 */ 1735 tp->t_flags |= TF_ACKNOW; 1736 todrop = tlen; 1737 tcpstat.tcps_rcvdupbyte += todrop; 1738 tcpstat.tcps_rcvduppack++; 1739 } else { 1740 tcpstat.tcps_rcvpartduppack++; 1741 tcpstat.tcps_rcvpartdupbyte += todrop; 1742 } 1743 hdroptlen += todrop; /*drop from head afterwards*/ 1744 th->th_seq += todrop; 1745 tlen -= todrop; 1746 if (th->th_urp > todrop) 1747 th->th_urp -= todrop; 1748 else { 1749 tiflags &= ~TH_URG; 1750 th->th_urp = 0; 1751 } 1752 } 1753 1754 /* 1755 * If new data are received on a connection after the 1756 * user processes are gone, then RST the other end. 1757 */ 1758 if ((so->so_state & SS_NOFDREF) && 1759 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1760 tp = tcp_close(tp); 1761 tcpstat.tcps_rcvafterclose++; 1762 goto dropwithreset; 1763 } 1764 1765 /* 1766 * If segment ends after window, drop trailing data 1767 * (and PUSH and FIN); if nothing left, just ACK. 1768 */ 1769 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1770 if (todrop > 0) { 1771 tcpstat.tcps_rcvpackafterwin++; 1772 if (todrop >= tlen) { 1773 tcpstat.tcps_rcvbyteafterwin += tlen; 1774 /* 1775 * If a new connection request is received 1776 * while in TIME_WAIT, drop the old connection 1777 * and start over if the sequence numbers 1778 * are above the previous ones. 1779 * 1780 * NOTE: We will checksum the packet again, and 1781 * so we need to put the header fields back into 1782 * network order! 1783 * XXX This kind of sucks, but we don't expect 1784 * XXX this to happen very often, so maybe it 1785 * XXX doesn't matter so much. 1786 */ 1787 if (tiflags & TH_SYN && 1788 tp->t_state == TCPS_TIME_WAIT && 1789 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1790 iss = tcp_new_iss(tp, tp->snd_nxt); 1791 tp = tcp_close(tp); 1792 TCP_FIELDS_TO_NET(th); 1793 goto findpcb; 1794 } 1795 /* 1796 * If window is closed can only take segments at 1797 * window edge, and have to drop data and PUSH from 1798 * incoming segments. Continue processing, but 1799 * remember to ack. Otherwise, drop segment 1800 * and ack. 1801 */ 1802 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1803 tp->t_flags |= TF_ACKNOW; 1804 tcpstat.tcps_rcvwinprobe++; 1805 } else 1806 goto dropafterack; 1807 } else 1808 tcpstat.tcps_rcvbyteafterwin += todrop; 1809 m_adj(m, -todrop); 1810 tlen -= todrop; 1811 tiflags &= ~(TH_PUSH|TH_FIN); 1812 } 1813 1814 /* 1815 * If last ACK falls within this segment's sequence numbers, 1816 * and the timestamp is newer, record it. 1817 */ 1818 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1819 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1820 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1821 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1822 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1823 tp->ts_recent = opti.ts_val; 1824 } 1825 1826 /* 1827 * If the RST bit is set examine the state: 1828 * SYN_RECEIVED STATE: 1829 * If passive open, return to LISTEN state. 1830 * If active open, inform user that connection was refused. 1831 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1832 * Inform user that connection was reset, and close tcb. 1833 * CLOSING, LAST_ACK, TIME_WAIT STATES 1834 * Close the tcb. 1835 */ 1836 if (tiflags&TH_RST) switch (tp->t_state) { 1837 1838 case TCPS_SYN_RECEIVED: 1839 so->so_error = ECONNREFUSED; 1840 goto close; 1841 1842 case TCPS_ESTABLISHED: 1843 case TCPS_FIN_WAIT_1: 1844 case TCPS_FIN_WAIT_2: 1845 case TCPS_CLOSE_WAIT: 1846 so->so_error = ECONNRESET; 1847 close: 1848 tp->t_state = TCPS_CLOSED; 1849 tcpstat.tcps_drops++; 1850 tp = tcp_close(tp); 1851 goto drop; 1852 1853 case TCPS_CLOSING: 1854 case TCPS_LAST_ACK: 1855 case TCPS_TIME_WAIT: 1856 tp = tcp_close(tp); 1857 goto drop; 1858 } 1859 1860 /* 1861 * If a SYN is in the window, then this is an 1862 * error and we send an RST and drop the connection. 1863 */ 1864 if (tiflags & TH_SYN) { 1865 tp = tcp_drop(tp, ECONNRESET); 1866 goto dropwithreset; 1867 } 1868 1869 /* 1870 * If the ACK bit is off we drop the segment and return. 1871 */ 1872 if ((tiflags & TH_ACK) == 0) { 1873 if (tp->t_flags & TF_ACKNOW) 1874 goto dropafterack; 1875 else 1876 goto drop; 1877 } 1878 1879 /* 1880 * Ack processing. 1881 */ 1882 switch (tp->t_state) { 1883 1884 /* 1885 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1886 * ESTABLISHED state and continue processing, otherwise 1887 * send an RST. 1888 */ 1889 case TCPS_SYN_RECEIVED: 1890 if (SEQ_GT(tp->snd_una, th->th_ack) || 1891 SEQ_GT(th->th_ack, tp->snd_max)) 1892 goto dropwithreset; 1893 tcpstat.tcps_connects++; 1894 soisconnected(so); 1895 tcp_established(tp); 1896 /* Do window scaling? */ 1897 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1898 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1899 tp->snd_scale = tp->requested_s_scale; 1900 tp->rcv_scale = tp->request_r_scale; 1901 } 1902 TCP_REASS_LOCK(tp); 1903 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1904 TCP_REASS_UNLOCK(tp); 1905 tp->snd_wl1 = th->th_seq - 1; 1906 /* fall into ... */ 1907 1908 /* 1909 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1910 * ACKs. If the ack is in the range 1911 * tp->snd_una < th->th_ack <= tp->snd_max 1912 * then advance tp->snd_una to th->th_ack and drop 1913 * data from the retransmission queue. If this ACK reflects 1914 * more up to date window information we update our window information. 1915 */ 1916 case TCPS_ESTABLISHED: 1917 case TCPS_FIN_WAIT_1: 1918 case TCPS_FIN_WAIT_2: 1919 case TCPS_CLOSE_WAIT: 1920 case TCPS_CLOSING: 1921 case TCPS_LAST_ACK: 1922 case TCPS_TIME_WAIT: 1923 1924 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1925 if (tlen == 0 && tiwin == tp->snd_wnd) { 1926 tcpstat.tcps_rcvdupack++; 1927 /* 1928 * If we have outstanding data (other than 1929 * a window probe), this is a completely 1930 * duplicate ack (ie, window info didn't 1931 * change), the ack is the biggest we've 1932 * seen and we've seen exactly our rexmt 1933 * threshhold of them, assume a packet 1934 * has been dropped and retransmit it. 1935 * Kludge snd_nxt & the congestion 1936 * window so we send only this one 1937 * packet. 1938 * 1939 * We know we're losing at the current 1940 * window size so do congestion avoidance 1941 * (set ssthresh to half the current window 1942 * and pull our congestion window back to 1943 * the new ssthresh). 1944 * 1945 * Dup acks mean that packets have left the 1946 * network (they're now cached at the receiver) 1947 * so bump cwnd by the amount in the receiver 1948 * to keep a constant cwnd packets in the 1949 * network. 1950 */ 1951 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1952 th->th_ack != tp->snd_una) 1953 tp->t_dupacks = 0; 1954 else if (++tp->t_dupacks == tcprexmtthresh) { 1955 tcp_seq onxt = tp->snd_nxt; 1956 u_int win = 1957 min(tp->snd_wnd, tp->snd_cwnd) / 1958 2 / tp->t_segsz; 1959 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1960 tp->snd_recover)) { 1961 /* 1962 * False fast retransmit after 1963 * timeout. Do not cut window. 1964 */ 1965 tp->snd_cwnd += tp->t_segsz; 1966 tp->t_dupacks = 0; 1967 (void) tcp_output(tp); 1968 goto drop; 1969 } 1970 1971 if (win < 2) 1972 win = 2; 1973 tp->snd_ssthresh = win * tp->t_segsz; 1974 tp->snd_recover = tp->snd_max; 1975 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1976 tp->t_rtttime = 0; 1977 tp->snd_nxt = th->th_ack; 1978 tp->snd_cwnd = tp->t_segsz; 1979 (void) tcp_output(tp); 1980 tp->snd_cwnd = tp->snd_ssthresh + 1981 tp->t_segsz * tp->t_dupacks; 1982 if (SEQ_GT(onxt, tp->snd_nxt)) 1983 tp->snd_nxt = onxt; 1984 goto drop; 1985 } else if (tp->t_dupacks > tcprexmtthresh) { 1986 tp->snd_cwnd += tp->t_segsz; 1987 (void) tcp_output(tp); 1988 goto drop; 1989 } 1990 } else 1991 tp->t_dupacks = 0; 1992 break; 1993 } 1994 /* 1995 * If the congestion window was inflated to account 1996 * for the other side's cached packets, retract it. 1997 */ 1998 if (tcp_do_newreno == 0) { 1999 if (tp->t_dupacks >= tcprexmtthresh && 2000 tp->snd_cwnd > tp->snd_ssthresh) 2001 tp->snd_cwnd = tp->snd_ssthresh; 2002 tp->t_dupacks = 0; 2003 } else if (tp->t_dupacks >= tcprexmtthresh && 2004 tcp_newreno(tp, th) == 0) { 2005 tp->snd_cwnd = tp->snd_ssthresh; 2006 /* 2007 * Window inflation should have left us with approx. 2008 * snd_ssthresh outstanding data. But in case we 2009 * would be inclined to send a burst, better to do 2010 * it via the slow start mechanism. 2011 */ 2012 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 2013 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 2014 + tp->t_segsz; 2015 tp->t_dupacks = 0; 2016 } 2017 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2018 tcpstat.tcps_rcvacktoomuch++; 2019 goto dropafterack; 2020 } 2021 acked = th->th_ack - tp->snd_una; 2022 tcpstat.tcps_rcvackpack++; 2023 tcpstat.tcps_rcvackbyte += acked; 2024 2025 /* 2026 * If we have a timestamp reply, update smoothed 2027 * round trip time. If no timestamp is present but 2028 * transmit timer is running and timed sequence 2029 * number was acked, update smoothed round trip time. 2030 * Since we now have an rtt measurement, cancel the 2031 * timer backoff (cf., Phil Karn's retransmit alg.). 2032 * Recompute the initial retransmit timer. 2033 */ 2034 if (opti.ts_present && opti.ts_ecr) 2035 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 2036 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2037 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2038 2039 /* 2040 * If all outstanding data is acked, stop retransmit 2041 * timer and remember to restart (more output or persist). 2042 * If there is more data to be acked, restart retransmit 2043 * timer, using current (possibly backed-off) value. 2044 */ 2045 if (th->th_ack == tp->snd_max) { 2046 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2047 needoutput = 1; 2048 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2049 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2050 /* 2051 * When new data is acked, open the congestion window. 2052 * If the window gives us less than ssthresh packets 2053 * in flight, open exponentially (segsz per packet). 2054 * Otherwise open linearly: segsz per window 2055 * (segsz^2 / cwnd per packet), plus a constant 2056 * fraction of a packet (segsz/8) to help larger windows 2057 * open quickly enough. 2058 */ 2059 { 2060 u_int cw = tp->snd_cwnd; 2061 u_int incr = tp->t_segsz; 2062 2063 if (cw > tp->snd_ssthresh) 2064 incr = incr * incr / cw; 2065 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 2066 tp->snd_cwnd = min(cw + incr, 2067 TCP_MAXWIN << tp->snd_scale); 2068 } 2069 ND6_HINT(tp); 2070 if (acked > so->so_snd.sb_cc) { 2071 tp->snd_wnd -= so->so_snd.sb_cc; 2072 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2073 ourfinisacked = 1; 2074 } else { 2075 sbdrop(&so->so_snd, acked); 2076 tp->snd_wnd -= acked; 2077 ourfinisacked = 0; 2078 } 2079 sowwakeup(so); 2080 /* 2081 * We want snd_recover to track snd_una to 2082 * avoid sequence wraparound problems for 2083 * very large transfers. 2084 */ 2085 tp->snd_una = tp->snd_recover = th->th_ack; 2086 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2087 tp->snd_nxt = tp->snd_una; 2088 2089 switch (tp->t_state) { 2090 2091 /* 2092 * In FIN_WAIT_1 STATE in addition to the processing 2093 * for the ESTABLISHED state if our FIN is now acknowledged 2094 * then enter FIN_WAIT_2. 2095 */ 2096 case TCPS_FIN_WAIT_1: 2097 if (ourfinisacked) { 2098 /* 2099 * If we can't receive any more 2100 * data, then closing user can proceed. 2101 * Starting the timer is contrary to the 2102 * specification, but if we don't get a FIN 2103 * we'll hang forever. 2104 */ 2105 if (so->so_state & SS_CANTRCVMORE) { 2106 soisdisconnected(so); 2107 if (tcp_maxidle > 0) 2108 TCP_TIMER_ARM(tp, TCPT_2MSL, 2109 tcp_maxidle); 2110 } 2111 tp->t_state = TCPS_FIN_WAIT_2; 2112 } 2113 break; 2114 2115 /* 2116 * In CLOSING STATE in addition to the processing for 2117 * the ESTABLISHED state if the ACK acknowledges our FIN 2118 * then enter the TIME-WAIT state, otherwise ignore 2119 * the segment. 2120 */ 2121 case TCPS_CLOSING: 2122 if (ourfinisacked) { 2123 tp->t_state = TCPS_TIME_WAIT; 2124 tcp_canceltimers(tp); 2125 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2126 soisdisconnected(so); 2127 } 2128 break; 2129 2130 /* 2131 * In LAST_ACK, we may still be waiting for data to drain 2132 * and/or to be acked, as well as for the ack of our FIN. 2133 * If our FIN is now acknowledged, delete the TCB, 2134 * enter the closed state and return. 2135 */ 2136 case TCPS_LAST_ACK: 2137 if (ourfinisacked) { 2138 tp = tcp_close(tp); 2139 goto drop; 2140 } 2141 break; 2142 2143 /* 2144 * In TIME_WAIT state the only thing that should arrive 2145 * is a retransmission of the remote FIN. Acknowledge 2146 * it and restart the finack timer. 2147 */ 2148 case TCPS_TIME_WAIT: 2149 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2150 goto dropafterack; 2151 } 2152 } 2153 2154 step6: 2155 /* 2156 * Update window information. 2157 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2158 */ 2159 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2160 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2161 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2162 /* keep track of pure window updates */ 2163 if (tlen == 0 && 2164 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2165 tcpstat.tcps_rcvwinupd++; 2166 tp->snd_wnd = tiwin; 2167 tp->snd_wl1 = th->th_seq; 2168 tp->snd_wl2 = th->th_ack; 2169 if (tp->snd_wnd > tp->max_sndwnd) 2170 tp->max_sndwnd = tp->snd_wnd; 2171 needoutput = 1; 2172 } 2173 2174 /* 2175 * Process segments with URG. 2176 */ 2177 if ((tiflags & TH_URG) && th->th_urp && 2178 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2179 /* 2180 * This is a kludge, but if we receive and accept 2181 * random urgent pointers, we'll crash in 2182 * soreceive. It's hard to imagine someone 2183 * actually wanting to send this much urgent data. 2184 */ 2185 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2186 th->th_urp = 0; /* XXX */ 2187 tiflags &= ~TH_URG; /* XXX */ 2188 goto dodata; /* XXX */ 2189 } 2190 /* 2191 * If this segment advances the known urgent pointer, 2192 * then mark the data stream. This should not happen 2193 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2194 * a FIN has been received from the remote side. 2195 * In these states we ignore the URG. 2196 * 2197 * According to RFC961 (Assigned Protocols), 2198 * the urgent pointer points to the last octet 2199 * of urgent data. We continue, however, 2200 * to consider it to indicate the first octet 2201 * of data past the urgent section as the original 2202 * spec states (in one of two places). 2203 */ 2204 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2205 tp->rcv_up = th->th_seq + th->th_urp; 2206 so->so_oobmark = so->so_rcv.sb_cc + 2207 (tp->rcv_up - tp->rcv_nxt) - 1; 2208 if (so->so_oobmark == 0) 2209 so->so_state |= SS_RCVATMARK; 2210 sohasoutofband(so); 2211 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2212 } 2213 /* 2214 * Remove out of band data so doesn't get presented to user. 2215 * This can happen independent of advancing the URG pointer, 2216 * but if two URG's are pending at once, some out-of-band 2217 * data may creep in... ick. 2218 */ 2219 if (th->th_urp <= (u_int16_t) tlen 2220 #ifdef SO_OOBINLINE 2221 && (so->so_options & SO_OOBINLINE) == 0 2222 #endif 2223 ) 2224 tcp_pulloutofband(so, th, m, hdroptlen); 2225 } else 2226 /* 2227 * If no out of band data is expected, 2228 * pull receive urgent pointer along 2229 * with the receive window. 2230 */ 2231 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2232 tp->rcv_up = tp->rcv_nxt; 2233 dodata: /* XXX */ 2234 2235 /* 2236 * Process the segment text, merging it into the TCP sequencing queue, 2237 * and arranging for acknowledgement of receipt if necessary. 2238 * This process logically involves adjusting tp->rcv_wnd as data 2239 * is presented to the user (this happens in tcp_usrreq.c, 2240 * case PRU_RCVD). If a FIN has already been received on this 2241 * connection then we just ignore the text. 2242 */ 2243 if ((tlen || (tiflags & TH_FIN)) && 2244 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2245 /* 2246 * Insert segment ti into reassembly queue of tcp with 2247 * control block tp. Return TH_FIN if reassembly now includes 2248 * a segment with FIN. The macro form does the common case 2249 * inline (segment is the next to be received on an 2250 * established connection, and the queue is empty), 2251 * avoiding linkage into and removal from the queue and 2252 * repetition of various conversions. 2253 * Set DELACK for segments received in order, but ack 2254 * immediately when segments are out of order 2255 * (so fast retransmit can work). 2256 */ 2257 /* NOTE: this was TCP_REASS() macro, but used only once */ 2258 TCP_REASS_LOCK(tp); 2259 if (th->th_seq == tp->rcv_nxt && 2260 TAILQ_FIRST(&tp->segq) == NULL && 2261 tp->t_state == TCPS_ESTABLISHED) { 2262 TCP_SETUP_ACK(tp, th); 2263 tp->rcv_nxt += tlen; 2264 tiflags = th->th_flags & TH_FIN; 2265 tcpstat.tcps_rcvpack++; 2266 tcpstat.tcps_rcvbyte += tlen; 2267 ND6_HINT(tp); 2268 if (so->so_state & SS_CANTRCVMORE) 2269 m_freem(m); 2270 else { 2271 m_adj(m, hdroptlen); 2272 sbappendstream(&(so)->so_rcv, m); 2273 } 2274 sorwakeup(so); 2275 } else { 2276 m_adj(m, hdroptlen); 2277 tiflags = tcp_reass(tp, th, m, &tlen); 2278 tp->t_flags |= TF_ACKNOW; 2279 } 2280 TCP_REASS_UNLOCK(tp); 2281 2282 /* 2283 * Note the amount of data that peer has sent into 2284 * our window, in order to estimate the sender's 2285 * buffer size. 2286 */ 2287 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2288 } else { 2289 m_freem(m); 2290 m = NULL; 2291 tiflags &= ~TH_FIN; 2292 } 2293 2294 /* 2295 * If FIN is received ACK the FIN and let the user know 2296 * that the connection is closing. Ignore a FIN received before 2297 * the connection is fully established. 2298 */ 2299 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2300 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2301 socantrcvmore(so); 2302 tp->t_flags |= TF_ACKNOW; 2303 tp->rcv_nxt++; 2304 } 2305 switch (tp->t_state) { 2306 2307 /* 2308 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2309 */ 2310 case TCPS_ESTABLISHED: 2311 tp->t_state = TCPS_CLOSE_WAIT; 2312 break; 2313 2314 /* 2315 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2316 * enter the CLOSING state. 2317 */ 2318 case TCPS_FIN_WAIT_1: 2319 tp->t_state = TCPS_CLOSING; 2320 break; 2321 2322 /* 2323 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2324 * starting the time-wait timer, turning off the other 2325 * standard timers. 2326 */ 2327 case TCPS_FIN_WAIT_2: 2328 tp->t_state = TCPS_TIME_WAIT; 2329 tcp_canceltimers(tp); 2330 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2331 soisdisconnected(so); 2332 break; 2333 2334 /* 2335 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2336 */ 2337 case TCPS_TIME_WAIT: 2338 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2339 break; 2340 } 2341 } 2342 #ifdef TCP_DEBUG 2343 if (so->so_options & SO_DEBUG) 2344 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2345 #endif 2346 2347 /* 2348 * Return any desired output. 2349 */ 2350 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2351 (void) tcp_output(tp); 2352 if (tcp_saveti) 2353 m_freem(tcp_saveti); 2354 return; 2355 2356 badsyn: 2357 /* 2358 * Received a bad SYN. Increment counters and dropwithreset. 2359 */ 2360 tcpstat.tcps_badsyn++; 2361 tp = NULL; 2362 goto dropwithreset; 2363 2364 dropafterack: 2365 /* 2366 * Generate an ACK dropping incoming segment if it occupies 2367 * sequence space, where the ACK reflects our state. 2368 */ 2369 if (tiflags & TH_RST) 2370 goto drop; 2371 m_freem(m); 2372 tp->t_flags |= TF_ACKNOW; 2373 (void) tcp_output(tp); 2374 if (tcp_saveti) 2375 m_freem(tcp_saveti); 2376 return; 2377 2378 dropwithreset_ratelim: 2379 /* 2380 * We may want to rate-limit RSTs in certain situations, 2381 * particularly if we are sending an RST in response to 2382 * an attempt to connect to or otherwise communicate with 2383 * a port for which we have no socket. 2384 */ 2385 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2386 tcp_rst_ppslim) == 0) { 2387 /* XXX stat */ 2388 goto drop; 2389 } 2390 /* ...fall into dropwithreset... */ 2391 2392 dropwithreset: 2393 /* 2394 * Generate a RST, dropping incoming segment. 2395 * Make ACK acceptable to originator of segment. 2396 */ 2397 if (tiflags & TH_RST) 2398 goto drop; 2399 2400 switch (af) { 2401 #ifdef INET6 2402 case AF_INET6: 2403 /* For following calls to tcp_respond */ 2404 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2405 goto drop; 2406 break; 2407 #endif /* INET6 */ 2408 case AF_INET: 2409 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2410 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2411 goto drop; 2412 } 2413 2414 if (tiflags & TH_ACK) 2415 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2416 else { 2417 if (tiflags & TH_SYN) 2418 tlen++; 2419 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2420 TH_RST|TH_ACK); 2421 } 2422 if (tcp_saveti) 2423 m_freem(tcp_saveti); 2424 return; 2425 2426 badcsum: 2427 tcpstat.tcps_rcvbadsum++; 2428 drop: 2429 /* 2430 * Drop space held by incoming segment and return. 2431 */ 2432 if (tp) { 2433 if (tp->t_inpcb) 2434 so = tp->t_inpcb->inp_socket; 2435 #ifdef INET6 2436 else if (tp->t_in6pcb) 2437 so = tp->t_in6pcb->in6p_socket; 2438 #endif 2439 else 2440 so = NULL; 2441 #ifdef TCP_DEBUG 2442 if (so && (so->so_options & SO_DEBUG) != 0) 2443 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2444 #endif 2445 } 2446 if (tcp_saveti) 2447 m_freem(tcp_saveti); 2448 m_freem(m); 2449 return; 2450 } 2451 2452 void 2453 tcp_dooptions(tp, cp, cnt, th, oi) 2454 struct tcpcb *tp; 2455 u_char *cp; 2456 int cnt; 2457 struct tcphdr *th; 2458 struct tcp_opt_info *oi; 2459 { 2460 u_int16_t mss; 2461 int opt, optlen; 2462 2463 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2464 opt = cp[0]; 2465 if (opt == TCPOPT_EOL) 2466 break; 2467 if (opt == TCPOPT_NOP) 2468 optlen = 1; 2469 else { 2470 if (cnt < 2) 2471 break; 2472 optlen = cp[1]; 2473 if (optlen < 2 || optlen > cnt) 2474 break; 2475 } 2476 switch (opt) { 2477 2478 default: 2479 continue; 2480 2481 case TCPOPT_MAXSEG: 2482 if (optlen != TCPOLEN_MAXSEG) 2483 continue; 2484 if (!(th->th_flags & TH_SYN)) 2485 continue; 2486 bcopy(cp + 2, &mss, sizeof(mss)); 2487 oi->maxseg = ntohs(mss); 2488 break; 2489 2490 case TCPOPT_WINDOW: 2491 if (optlen != TCPOLEN_WINDOW) 2492 continue; 2493 if (!(th->th_flags & TH_SYN)) 2494 continue; 2495 tp->t_flags |= TF_RCVD_SCALE; 2496 tp->requested_s_scale = cp[2]; 2497 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2498 #if 0 /*XXX*/ 2499 char *p; 2500 2501 if (ip) 2502 p = ntohl(ip->ip_src); 2503 #ifdef INET6 2504 else if (ip6) 2505 p = ip6_sprintf(&ip6->ip6_src); 2506 #endif 2507 else 2508 p = "(unknown)"; 2509 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2510 "assuming %d\n", 2511 tp->requested_s_scale, p, 2512 TCP_MAX_WINSHIFT); 2513 #else 2514 log(LOG_ERR, "TCP: invalid wscale %d, " 2515 "assuming %d\n", 2516 tp->requested_s_scale, 2517 TCP_MAX_WINSHIFT); 2518 #endif 2519 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2520 } 2521 break; 2522 2523 case TCPOPT_TIMESTAMP: 2524 if (optlen != TCPOLEN_TIMESTAMP) 2525 continue; 2526 oi->ts_present = 1; 2527 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2528 NTOHL(oi->ts_val); 2529 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2530 NTOHL(oi->ts_ecr); 2531 2532 /* 2533 * A timestamp received in a SYN makes 2534 * it ok to send timestamp requests and replies. 2535 */ 2536 if (th->th_flags & TH_SYN) { 2537 tp->t_flags |= TF_RCVD_TSTMP; 2538 tp->ts_recent = oi->ts_val; 2539 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2540 } 2541 break; 2542 case TCPOPT_SACK_PERMITTED: 2543 if (optlen != TCPOLEN_SACK_PERMITTED) 2544 continue; 2545 if (!(th->th_flags & TH_SYN)) 2546 continue; 2547 tp->t_flags &= ~TF_CANT_TXSACK; 2548 break; 2549 2550 case TCPOPT_SACK: 2551 if (tp->t_flags & TF_IGNR_RXSACK) 2552 continue; 2553 if (optlen % 8 != 2 || optlen < 10) 2554 continue; 2555 cp += 2; 2556 optlen -= 2; 2557 for (; optlen > 0; cp -= 8, optlen -= 8) { 2558 tcp_seq lwe, rwe; 2559 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2560 NTOHL(lwe); 2561 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2562 NTOHL(rwe); 2563 /* tcp_mark_sacked(tp, lwe, rwe); */ 2564 } 2565 break; 2566 } 2567 } 2568 } 2569 2570 /* 2571 * Pull out of band byte out of a segment so 2572 * it doesn't appear in the user's data queue. 2573 * It is still reflected in the segment length for 2574 * sequencing purposes. 2575 */ 2576 void 2577 tcp_pulloutofband(so, th, m, off) 2578 struct socket *so; 2579 struct tcphdr *th; 2580 struct mbuf *m; 2581 int off; 2582 { 2583 int cnt = off + th->th_urp - 1; 2584 2585 while (cnt >= 0) { 2586 if (m->m_len > cnt) { 2587 char *cp = mtod(m, caddr_t) + cnt; 2588 struct tcpcb *tp = sototcpcb(so); 2589 2590 tp->t_iobc = *cp; 2591 tp->t_oobflags |= TCPOOB_HAVEDATA; 2592 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2593 m->m_len--; 2594 return; 2595 } 2596 cnt -= m->m_len; 2597 m = m->m_next; 2598 if (m == 0) 2599 break; 2600 } 2601 panic("tcp_pulloutofband"); 2602 } 2603 2604 /* 2605 * Collect new round-trip time estimate 2606 * and update averages and current timeout. 2607 */ 2608 void 2609 tcp_xmit_timer(tp, rtt) 2610 struct tcpcb *tp; 2611 uint32_t rtt; 2612 { 2613 int32_t delta; 2614 2615 tcpstat.tcps_rttupdated++; 2616 if (tp->t_srtt != 0) { 2617 /* 2618 * srtt is stored as fixed point with 3 bits after the 2619 * binary point (i.e., scaled by 8). The following magic 2620 * is equivalent to the smoothing algorithm in rfc793 with 2621 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2622 * point). Adjust rtt to origin 0. 2623 */ 2624 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2625 if ((tp->t_srtt += delta) <= 0) 2626 tp->t_srtt = 1 << 2; 2627 /* 2628 * We accumulate a smoothed rtt variance (actually, a 2629 * smoothed mean difference), then set the retransmit 2630 * timer to smoothed rtt + 4 times the smoothed variance. 2631 * rttvar is stored as fixed point with 2 bits after the 2632 * binary point (scaled by 4). The following is 2633 * equivalent to rfc793 smoothing with an alpha of .75 2634 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2635 * rfc793's wired-in beta. 2636 */ 2637 if (delta < 0) 2638 delta = -delta; 2639 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2640 if ((tp->t_rttvar += delta) <= 0) 2641 tp->t_rttvar = 1 << 2; 2642 } else { 2643 /* 2644 * No rtt measurement yet - use the unsmoothed rtt. 2645 * Set the variance to half the rtt (so our first 2646 * retransmit happens at 3*rtt). 2647 */ 2648 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2649 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2650 } 2651 tp->t_rtttime = 0; 2652 tp->t_rxtshift = 0; 2653 2654 /* 2655 * the retransmit should happen at rtt + 4 * rttvar. 2656 * Because of the way we do the smoothing, srtt and rttvar 2657 * will each average +1/2 tick of bias. When we compute 2658 * the retransmit timer, we want 1/2 tick of rounding and 2659 * 1 extra tick because of +-1/2 tick uncertainty in the 2660 * firing of the timer. The bias will give us exactly the 2661 * 1.5 tick we need. But, because the bias is 2662 * statistical, we have to test that we don't drop below 2663 * the minimum feasible timer (which is 2 ticks). 2664 */ 2665 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2666 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2667 2668 /* 2669 * We received an ack for a packet that wasn't retransmitted; 2670 * it is probably safe to discard any error indications we've 2671 * received recently. This isn't quite right, but close enough 2672 * for now (a route might have failed after we sent a segment, 2673 * and the return path might not be symmetrical). 2674 */ 2675 tp->t_softerror = 0; 2676 } 2677 2678 /* 2679 * Checks for partial ack. If partial ack arrives, force the retransmission 2680 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2681 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2682 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2683 */ 2684 int 2685 tcp_newreno(tp, th) 2686 struct tcpcb *tp; 2687 struct tcphdr *th; 2688 { 2689 tcp_seq onxt = tp->snd_nxt; 2690 u_long ocwnd = tp->snd_cwnd; 2691 2692 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2693 /* 2694 * snd_una has not yet been updated and the socket's send 2695 * buffer has not yet drained off the ACK'd data, so we 2696 * have to leave snd_una as it was to get the correct data 2697 * offset in tcp_output(). 2698 */ 2699 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2700 tp->t_rtttime = 0; 2701 tp->snd_nxt = th->th_ack; 2702 /* 2703 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2704 * is not yet updated when we're called. 2705 */ 2706 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2707 (void) tcp_output(tp); 2708 tp->snd_cwnd = ocwnd; 2709 if (SEQ_GT(onxt, tp->snd_nxt)) 2710 tp->snd_nxt = onxt; 2711 /* 2712 * Partial window deflation. Relies on fact that tp->snd_una 2713 * not updated yet. 2714 */ 2715 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2716 return 1; 2717 } 2718 return 0; 2719 } 2720 2721 2722 /* 2723 * TCP compressed state engine. Currently used to hold compressed 2724 * state for SYN_RECEIVED. 2725 */ 2726 2727 u_long syn_cache_count; 2728 u_int32_t syn_hash1, syn_hash2; 2729 2730 #define SYN_HASH(sa, sp, dp) \ 2731 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2732 ((u_int32_t)(sp)))^syn_hash2))) 2733 #ifndef INET6 2734 #define SYN_HASHALL(hash, src, dst) \ 2735 do { \ 2736 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2737 ((struct sockaddr_in *)(src))->sin_port, \ 2738 ((struct sockaddr_in *)(dst))->sin_port); \ 2739 } while (/*CONSTCOND*/ 0) 2740 #else 2741 #define SYN_HASH6(sa, sp, dp) \ 2742 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2743 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2744 & 0x7fffffff) 2745 2746 #define SYN_HASHALL(hash, src, dst) \ 2747 do { \ 2748 switch ((src)->sa_family) { \ 2749 case AF_INET: \ 2750 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2751 ((struct sockaddr_in *)(src))->sin_port, \ 2752 ((struct sockaddr_in *)(dst))->sin_port); \ 2753 break; \ 2754 case AF_INET6: \ 2755 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2756 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2757 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2758 break; \ 2759 default: \ 2760 hash = 0; \ 2761 } \ 2762 } while (/*CONSTCOND*/0) 2763 #endif /* INET6 */ 2764 2765 #define SYN_CACHE_RM(sc) \ 2766 do { \ 2767 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 2768 (sc), sc_bucketq); \ 2769 (sc)->sc_tp = NULL; \ 2770 LIST_REMOVE((sc), sc_tpq); \ 2771 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2772 callout_stop(&(sc)->sc_timer); \ 2773 syn_cache_count--; \ 2774 } while (/*CONSTCOND*/0) 2775 2776 #define SYN_CACHE_PUT(sc) \ 2777 do { \ 2778 if ((sc)->sc_ipopts) \ 2779 (void) m_free((sc)->sc_ipopts); \ 2780 if ((sc)->sc_route4.ro_rt != NULL) \ 2781 RTFREE((sc)->sc_route4.ro_rt); \ 2782 pool_put(&syn_cache_pool, (sc)); \ 2783 } while (/*CONSTCOND*/0) 2784 2785 struct pool syn_cache_pool; 2786 2787 /* 2788 * We don't estimate RTT with SYNs, so each packet starts with the default 2789 * RTT and each timer step has a fixed timeout value. 2790 */ 2791 #define SYN_CACHE_TIMER_ARM(sc) \ 2792 do { \ 2793 TCPT_RANGESET((sc)->sc_rxtcur, \ 2794 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2795 TCPTV_REXMTMAX); \ 2796 callout_reset(&(sc)->sc_timer, \ 2797 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 2798 } while (/*CONSTCOND*/0) 2799 2800 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2801 2802 void 2803 syn_cache_init() 2804 { 2805 int i; 2806 2807 /* Initialize the hash buckets. */ 2808 for (i = 0; i < tcp_syn_cache_size; i++) 2809 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 2810 2811 /* Initialize the syn cache pool. */ 2812 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2813 "synpl", NULL); 2814 } 2815 2816 void 2817 syn_cache_insert(sc, tp) 2818 struct syn_cache *sc; 2819 struct tcpcb *tp; 2820 { 2821 struct syn_cache_head *scp; 2822 struct syn_cache *sc2; 2823 int s; 2824 2825 /* 2826 * If there are no entries in the hash table, reinitialize 2827 * the hash secrets. 2828 */ 2829 if (syn_cache_count == 0) { 2830 struct timeval tv; 2831 microtime(&tv); 2832 syn_hash1 = arc4random() ^ (u_long)≻ 2833 syn_hash2 = arc4random() ^ tv.tv_usec; 2834 } 2835 2836 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2837 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2838 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2839 2840 /* 2841 * Make sure that we don't overflow the per-bucket 2842 * limit or the total cache size limit. 2843 */ 2844 s = splsoftnet(); 2845 if (scp->sch_length >= tcp_syn_bucket_limit) { 2846 tcpstat.tcps_sc_bucketoverflow++; 2847 /* 2848 * The bucket is full. Toss the oldest element in the 2849 * bucket. This will be the first entry in the bucket. 2850 */ 2851 sc2 = TAILQ_FIRST(&scp->sch_bucket); 2852 #ifdef DIAGNOSTIC 2853 /* 2854 * This should never happen; we should always find an 2855 * entry in our bucket. 2856 */ 2857 if (sc2 == NULL) 2858 panic("syn_cache_insert: bucketoverflow: impossible"); 2859 #endif 2860 SYN_CACHE_RM(sc2); 2861 SYN_CACHE_PUT(sc2); 2862 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2863 struct syn_cache_head *scp2, *sce; 2864 2865 tcpstat.tcps_sc_overflowed++; 2866 /* 2867 * The cache is full. Toss the oldest entry in the 2868 * first non-empty bucket we can find. 2869 * 2870 * XXX We would really like to toss the oldest 2871 * entry in the cache, but we hope that this 2872 * condition doesn't happen very often. 2873 */ 2874 scp2 = scp; 2875 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 2876 sce = &tcp_syn_cache[tcp_syn_cache_size]; 2877 for (++scp2; scp2 != scp; scp2++) { 2878 if (scp2 >= sce) 2879 scp2 = &tcp_syn_cache[0]; 2880 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 2881 break; 2882 } 2883 #ifdef DIAGNOSTIC 2884 /* 2885 * This should never happen; we should always find a 2886 * non-empty bucket. 2887 */ 2888 if (scp2 == scp) 2889 panic("syn_cache_insert: cacheoverflow: " 2890 "impossible"); 2891 #endif 2892 } 2893 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 2894 SYN_CACHE_RM(sc2); 2895 SYN_CACHE_PUT(sc2); 2896 } 2897 2898 /* 2899 * Initialize the entry's timer. 2900 */ 2901 sc->sc_rxttot = 0; 2902 sc->sc_rxtshift = 0; 2903 SYN_CACHE_TIMER_ARM(sc); 2904 2905 /* Link it from tcpcb entry */ 2906 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2907 2908 /* Put it into the bucket. */ 2909 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 2910 scp->sch_length++; 2911 syn_cache_count++; 2912 2913 tcpstat.tcps_sc_added++; 2914 splx(s); 2915 } 2916 2917 /* 2918 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2919 * If we have retransmitted an entry the maximum number of times, expire 2920 * that entry. 2921 */ 2922 void 2923 syn_cache_timer(void *arg) 2924 { 2925 struct syn_cache *sc = arg; 2926 int s; 2927 2928 s = splsoftnet(); 2929 2930 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 2931 /* Drop it -- too many retransmissions. */ 2932 goto dropit; 2933 } 2934 2935 /* 2936 * Compute the total amount of time this entry has 2937 * been on a queue. If this entry has been on longer 2938 * than the keep alive timer would allow, expire it. 2939 */ 2940 sc->sc_rxttot += sc->sc_rxtcur; 2941 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 2942 goto dropit; 2943 2944 tcpstat.tcps_sc_retransmitted++; 2945 (void) syn_cache_respond(sc, NULL); 2946 2947 /* Advance the timer back-off. */ 2948 sc->sc_rxtshift++; 2949 SYN_CACHE_TIMER_ARM(sc); 2950 2951 splx(s); 2952 return; 2953 2954 dropit: 2955 tcpstat.tcps_sc_timed_out++; 2956 SYN_CACHE_RM(sc); 2957 SYN_CACHE_PUT(sc); 2958 splx(s); 2959 } 2960 2961 /* 2962 * Remove syn cache created by the specified tcb entry, 2963 * because this does not make sense to keep them 2964 * (if there's no tcb entry, syn cache entry will never be used) 2965 */ 2966 void 2967 syn_cache_cleanup(tp) 2968 struct tcpcb *tp; 2969 { 2970 struct syn_cache *sc, *nsc; 2971 int s; 2972 2973 s = splsoftnet(); 2974 2975 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2976 nsc = LIST_NEXT(sc, sc_tpq); 2977 2978 #ifdef DIAGNOSTIC 2979 if (sc->sc_tp != tp) 2980 panic("invalid sc_tp in syn_cache_cleanup"); 2981 #endif 2982 SYN_CACHE_RM(sc); 2983 SYN_CACHE_PUT(sc); 2984 } 2985 /* just for safety */ 2986 LIST_INIT(&tp->t_sc); 2987 2988 splx(s); 2989 } 2990 2991 /* 2992 * Find an entry in the syn cache. 2993 */ 2994 struct syn_cache * 2995 syn_cache_lookup(src, dst, headp) 2996 struct sockaddr *src; 2997 struct sockaddr *dst; 2998 struct syn_cache_head **headp; 2999 { 3000 struct syn_cache *sc; 3001 struct syn_cache_head *scp; 3002 u_int32_t hash; 3003 int s; 3004 3005 SYN_HASHALL(hash, src, dst); 3006 3007 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3008 *headp = scp; 3009 s = splsoftnet(); 3010 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3011 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3012 if (sc->sc_hash != hash) 3013 continue; 3014 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3015 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3016 splx(s); 3017 return (sc); 3018 } 3019 } 3020 splx(s); 3021 return (NULL); 3022 } 3023 3024 /* 3025 * This function gets called when we receive an ACK for a 3026 * socket in the LISTEN state. We look up the connection 3027 * in the syn cache, and if its there, we pull it out of 3028 * the cache and turn it into a full-blown connection in 3029 * the SYN-RECEIVED state. 3030 * 3031 * The return values may not be immediately obvious, and their effects 3032 * can be subtle, so here they are: 3033 * 3034 * NULL SYN was not found in cache; caller should drop the 3035 * packet and send an RST. 3036 * 3037 * -1 We were unable to create the new connection, and are 3038 * aborting it. An ACK,RST is being sent to the peer 3039 * (unless we got screwey sequence numbners; see below), 3040 * because the 3-way handshake has been completed. Caller 3041 * should not free the mbuf, since we may be using it. If 3042 * we are not, we will free it. 3043 * 3044 * Otherwise, the return value is a pointer to the new socket 3045 * associated with the connection. 3046 */ 3047 struct socket * 3048 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3049 struct sockaddr *src; 3050 struct sockaddr *dst; 3051 struct tcphdr *th; 3052 unsigned int hlen, tlen; 3053 struct socket *so; 3054 struct mbuf *m; 3055 { 3056 struct syn_cache *sc; 3057 struct syn_cache_head *scp; 3058 struct inpcb *inp = NULL; 3059 #ifdef INET6 3060 struct in6pcb *in6p = NULL; 3061 #endif 3062 struct tcpcb *tp = 0; 3063 struct mbuf *am; 3064 int s; 3065 struct socket *oso; 3066 3067 s = splsoftnet(); 3068 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3069 splx(s); 3070 return (NULL); 3071 } 3072 3073 /* 3074 * Verify the sequence and ack numbers. Try getting the correct 3075 * response again. 3076 */ 3077 if ((th->th_ack != sc->sc_iss + 1) || 3078 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3079 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3080 (void) syn_cache_respond(sc, m); 3081 splx(s); 3082 return ((struct socket *)(-1)); 3083 } 3084 3085 /* Remove this cache entry */ 3086 SYN_CACHE_RM(sc); 3087 splx(s); 3088 3089 /* 3090 * Ok, create the full blown connection, and set things up 3091 * as they would have been set up if we had created the 3092 * connection when the SYN arrived. If we can't create 3093 * the connection, abort it. 3094 */ 3095 /* 3096 * inp still has the OLD in_pcb stuff, set the 3097 * v6-related flags on the new guy, too. This is 3098 * done particularly for the case where an AF_INET6 3099 * socket is bound only to a port, and a v4 connection 3100 * comes in on that port. 3101 * we also copy the flowinfo from the original pcb 3102 * to the new one. 3103 */ 3104 oso = so; 3105 so = sonewconn(so, SS_ISCONNECTED); 3106 if (so == NULL) 3107 goto resetandabort; 3108 3109 switch (so->so_proto->pr_domain->dom_family) { 3110 #ifdef INET 3111 case AF_INET: 3112 inp = sotoinpcb(so); 3113 break; 3114 #endif 3115 #ifdef INET6 3116 case AF_INET6: 3117 in6p = sotoin6pcb(so); 3118 break; 3119 #endif 3120 } 3121 switch (src->sa_family) { 3122 #ifdef INET 3123 case AF_INET: 3124 if (inp) { 3125 struct in_ifaddr *ia; 3126 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3127 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3128 inp->inp_options = ip_srcroute(); 3129 INADDR_TO_IA(inp->inp_laddr, ia); 3130 KASSERT(ia != NULL); 3131 KASSERT(inp->inp_ia == NULL); 3132 inp->inp_ia = ia; 3133 LIST_INSERT_HEAD(&ia->ia_inpcbs, inp, inp_ialink); 3134 IFAREF(&ia->ia_ifa); 3135 in_pcbstate(inp, INP_BOUND); 3136 if (inp->inp_options == NULL) { 3137 inp->inp_options = sc->sc_ipopts; 3138 sc->sc_ipopts = NULL; 3139 } 3140 } 3141 #ifdef INET6 3142 else if (in6p) { 3143 /* IPv4 packet to AF_INET6 socket */ 3144 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3145 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3146 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3147 &in6p->in6p_laddr.s6_addr32[3], 3148 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3149 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3150 in6totcpcb(in6p)->t_family = AF_INET; 3151 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3152 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3153 else 3154 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3155 } 3156 #endif 3157 break; 3158 #endif 3159 #ifdef INET6 3160 case AF_INET6: 3161 if (in6p) { 3162 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3163 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3164 #if 0 3165 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3166 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3167 #endif 3168 } 3169 break; 3170 #endif 3171 } 3172 #ifdef INET6 3173 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3174 struct in6pcb *oin6p = sotoin6pcb(oso); 3175 /* inherit socket options from the listening socket */ 3176 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3177 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3178 m_freem(in6p->in6p_options); 3179 in6p->in6p_options = 0; 3180 } 3181 ip6_savecontrol(in6p, &in6p->in6p_options, 3182 mtod(m, struct ip6_hdr *), m); 3183 } 3184 #endif 3185 3186 #ifdef IPSEC 3187 /* 3188 * we make a copy of policy, instead of sharing the policy, 3189 * for better behavior in terms of SA lookup and dead SA removal. 3190 */ 3191 if (inp) { 3192 /* copy old policy into new socket's */ 3193 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3194 printf("tcp_input: could not copy policy\n"); 3195 } 3196 #ifdef INET6 3197 else if (in6p) { 3198 /* copy old policy into new socket's */ 3199 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3200 in6p->in6p_sp)) 3201 printf("tcp_input: could not copy policy\n"); 3202 } 3203 #endif 3204 #endif 3205 3206 /* 3207 * Give the new socket our cached route reference. 3208 */ 3209 if (inp) 3210 inp->inp_route = sc->sc_route4; /* struct assignment */ 3211 #ifdef INET6 3212 else 3213 in6p->in6p_route = sc->sc_route6; 3214 #endif 3215 sc->sc_route4.ro_rt = NULL; 3216 3217 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3218 if (am == NULL) 3219 goto resetandabort; 3220 MCLAIM(am, &tcp_mowner); 3221 am->m_len = src->sa_len; 3222 bcopy(src, mtod(am, caddr_t), src->sa_len); 3223 if (inp) { 3224 if (in_pcbconnect(inp, am)) { 3225 (void) m_free(am); 3226 goto resetandabort; 3227 } 3228 } 3229 #ifdef INET6 3230 else if (in6p) { 3231 if (src->sa_family == AF_INET) { 3232 /* IPv4 packet to AF_INET6 socket */ 3233 struct sockaddr_in6 *sin6; 3234 sin6 = mtod(am, struct sockaddr_in6 *); 3235 am->m_len = sizeof(*sin6); 3236 bzero(sin6, sizeof(*sin6)); 3237 sin6->sin6_family = AF_INET6; 3238 sin6->sin6_len = sizeof(*sin6); 3239 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3240 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3241 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3242 &sin6->sin6_addr.s6_addr32[3], 3243 sizeof(sin6->sin6_addr.s6_addr32[3])); 3244 } 3245 if (in6_pcbconnect(in6p, am)) { 3246 (void) m_free(am); 3247 goto resetandabort; 3248 } 3249 } 3250 #endif 3251 else { 3252 (void) m_free(am); 3253 goto resetandabort; 3254 } 3255 (void) m_free(am); 3256 3257 if (inp) 3258 tp = intotcpcb(inp); 3259 #ifdef INET6 3260 else if (in6p) 3261 tp = in6totcpcb(in6p); 3262 #endif 3263 else 3264 tp = NULL; 3265 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3266 if (sc->sc_request_r_scale != 15) { 3267 tp->requested_s_scale = sc->sc_requested_s_scale; 3268 tp->request_r_scale = sc->sc_request_r_scale; 3269 tp->snd_scale = sc->sc_requested_s_scale; 3270 tp->rcv_scale = sc->sc_request_r_scale; 3271 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3272 } 3273 if (sc->sc_flags & SCF_TIMESTAMP) 3274 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3275 tp->ts_timebase = sc->sc_timebase; 3276 3277 tp->t_template = tcp_template(tp); 3278 if (tp->t_template == 0) { 3279 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3280 so = NULL; 3281 m_freem(m); 3282 goto abort; 3283 } 3284 3285 tp->iss = sc->sc_iss; 3286 tp->irs = sc->sc_irs; 3287 tcp_sendseqinit(tp); 3288 tcp_rcvseqinit(tp); 3289 tp->t_state = TCPS_SYN_RECEIVED; 3290 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3291 tcpstat.tcps_accepts++; 3292 3293 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3294 tp->t_ourmss = sc->sc_ourmaxseg; 3295 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3296 3297 /* 3298 * Initialize the initial congestion window. If we 3299 * had to retransmit the SYN,ACK, we must initialize cwnd 3300 * to 1 segment (i.e. the Loss Window). 3301 */ 3302 if (sc->sc_rxtshift) 3303 tp->snd_cwnd = tp->t_peermss; 3304 else { 3305 int ss = tcp_init_win; 3306 #ifdef INET 3307 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3308 ss = tcp_init_win_local; 3309 #endif 3310 #ifdef INET6 3311 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3312 ss = tcp_init_win_local; 3313 #endif 3314 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3315 } 3316 3317 tcp_rmx_rtt(tp); 3318 tp->snd_wl1 = sc->sc_irs; 3319 tp->rcv_up = sc->sc_irs + 1; 3320 3321 /* 3322 * This is what whould have happened in tcp_output() when 3323 * the SYN,ACK was sent. 3324 */ 3325 tp->snd_up = tp->snd_una; 3326 tp->snd_max = tp->snd_nxt = tp->iss+1; 3327 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3328 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3329 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3330 tp->last_ack_sent = tp->rcv_nxt; 3331 3332 tcpstat.tcps_sc_completed++; 3333 SYN_CACHE_PUT(sc); 3334 return (so); 3335 3336 resetandabort: 3337 (void) tcp_respond(NULL, m, m, th, 3338 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3339 abort: 3340 if (so != NULL) 3341 (void) soabort(so); 3342 SYN_CACHE_PUT(sc); 3343 tcpstat.tcps_sc_aborted++; 3344 return ((struct socket *)(-1)); 3345 } 3346 3347 /* 3348 * This function is called when we get a RST for a 3349 * non-existent connection, so that we can see if the 3350 * connection is in the syn cache. If it is, zap it. 3351 */ 3352 3353 void 3354 syn_cache_reset(src, dst, th) 3355 struct sockaddr *src; 3356 struct sockaddr *dst; 3357 struct tcphdr *th; 3358 { 3359 struct syn_cache *sc; 3360 struct syn_cache_head *scp; 3361 int s = splsoftnet(); 3362 3363 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3364 splx(s); 3365 return; 3366 } 3367 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3368 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3369 splx(s); 3370 return; 3371 } 3372 SYN_CACHE_RM(sc); 3373 splx(s); 3374 tcpstat.tcps_sc_reset++; 3375 SYN_CACHE_PUT(sc); 3376 } 3377 3378 void 3379 syn_cache_unreach(src, dst, th) 3380 struct sockaddr *src; 3381 struct sockaddr *dst; 3382 struct tcphdr *th; 3383 { 3384 struct syn_cache *sc; 3385 struct syn_cache_head *scp; 3386 int s; 3387 3388 s = splsoftnet(); 3389 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3390 splx(s); 3391 return; 3392 } 3393 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3394 if (ntohl (th->th_seq) != sc->sc_iss) { 3395 splx(s); 3396 return; 3397 } 3398 3399 /* 3400 * If we've rertransmitted 3 times and this is our second error, 3401 * we remove the entry. Otherwise, we allow it to continue on. 3402 * This prevents us from incorrectly nuking an entry during a 3403 * spurious network outage. 3404 * 3405 * See tcp_notify(). 3406 */ 3407 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3408 sc->sc_flags |= SCF_UNREACH; 3409 splx(s); 3410 return; 3411 } 3412 3413 SYN_CACHE_RM(sc); 3414 splx(s); 3415 tcpstat.tcps_sc_unreach++; 3416 SYN_CACHE_PUT(sc); 3417 } 3418 3419 /* 3420 * Given a LISTEN socket and an inbound SYN request, add 3421 * this to the syn cache, and send back a segment: 3422 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3423 * to the source. 3424 * 3425 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3426 * Doing so would require that we hold onto the data and deliver it 3427 * to the application. However, if we are the target of a SYN-flood 3428 * DoS attack, an attacker could send data which would eventually 3429 * consume all available buffer space if it were ACKed. By not ACKing 3430 * the data, we avoid this DoS scenario. 3431 */ 3432 3433 int 3434 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3435 struct sockaddr *src; 3436 struct sockaddr *dst; 3437 struct tcphdr *th; 3438 unsigned int hlen; 3439 struct socket *so; 3440 struct mbuf *m; 3441 u_char *optp; 3442 int optlen; 3443 struct tcp_opt_info *oi; 3444 { 3445 struct tcpcb tb, *tp; 3446 long win; 3447 struct syn_cache *sc; 3448 struct syn_cache_head *scp; 3449 struct mbuf *ipopts; 3450 3451 tp = sototcpcb(so); 3452 3453 /* 3454 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3455 * 3456 * Note this check is performed in tcp_input() very early on. 3457 */ 3458 3459 /* 3460 * Initialize some local state. 3461 */ 3462 win = sbspace(&so->so_rcv); 3463 if (win > TCP_MAXWIN) 3464 win = TCP_MAXWIN; 3465 3466 switch (src->sa_family) { 3467 #ifdef INET 3468 case AF_INET: 3469 /* 3470 * Remember the IP options, if any. 3471 */ 3472 ipopts = ip_srcroute(); 3473 break; 3474 #endif 3475 default: 3476 ipopts = NULL; 3477 } 3478 3479 if (optp) { 3480 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3481 tcp_dooptions(&tb, optp, optlen, th, oi); 3482 } else 3483 tb.t_flags = 0; 3484 3485 /* 3486 * See if we already have an entry for this connection. 3487 * If we do, resend the SYN,ACK. We do not count this 3488 * as a retransmission (XXX though maybe we should). 3489 */ 3490 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3491 tcpstat.tcps_sc_dupesyn++; 3492 if (ipopts) { 3493 /* 3494 * If we were remembering a previous source route, 3495 * forget it and use the new one we've been given. 3496 */ 3497 if (sc->sc_ipopts) 3498 (void) m_free(sc->sc_ipopts); 3499 sc->sc_ipopts = ipopts; 3500 } 3501 sc->sc_timestamp = tb.ts_recent; 3502 if (syn_cache_respond(sc, m) == 0) { 3503 tcpstat.tcps_sndacks++; 3504 tcpstat.tcps_sndtotal++; 3505 } 3506 return (1); 3507 } 3508 3509 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3510 if (sc == NULL) { 3511 if (ipopts) 3512 (void) m_free(ipopts); 3513 return (0); 3514 } 3515 3516 /* 3517 * Fill in the cache, and put the necessary IP and TCP 3518 * options into the reply. 3519 */ 3520 bzero(sc, sizeof(struct syn_cache)); 3521 callout_init(&sc->sc_timer); 3522 bcopy(src, &sc->sc_src, src->sa_len); 3523 bcopy(dst, &sc->sc_dst, dst->sa_len); 3524 sc->sc_flags = 0; 3525 sc->sc_ipopts = ipopts; 3526 sc->sc_irs = th->th_seq; 3527 switch (src->sa_family) { 3528 #ifdef INET 3529 case AF_INET: 3530 { 3531 struct sockaddr_in *srcin = (void *) src; 3532 struct sockaddr_in *dstin = (void *) dst; 3533 3534 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3535 &srcin->sin_addr, dstin->sin_port, 3536 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3537 break; 3538 } 3539 #endif /* INET */ 3540 #ifdef INET6 3541 case AF_INET6: 3542 { 3543 struct sockaddr_in6 *srcin6 = (void *) src; 3544 struct sockaddr_in6 *dstin6 = (void *) dst; 3545 3546 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3547 &srcin6->sin6_addr, dstin6->sin6_port, 3548 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3549 break; 3550 } 3551 #endif /* INET6 */ 3552 } 3553 sc->sc_peermaxseg = oi->maxseg; 3554 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3555 m->m_pkthdr.rcvif : NULL, 3556 sc->sc_src.sa.sa_family); 3557 sc->sc_win = win; 3558 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3559 sc->sc_timestamp = tb.ts_recent; 3560 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3561 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 3562 sc->sc_flags |= SCF_TIMESTAMP; 3563 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3564 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3565 sc->sc_requested_s_scale = tb.requested_s_scale; 3566 sc->sc_request_r_scale = 0; 3567 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3568 TCP_MAXWIN << sc->sc_request_r_scale < 3569 so->so_rcv.sb_hiwat) 3570 sc->sc_request_r_scale++; 3571 } else { 3572 sc->sc_requested_s_scale = 15; 3573 sc->sc_request_r_scale = 15; 3574 } 3575 sc->sc_tp = tp; 3576 if (syn_cache_respond(sc, m) == 0) { 3577 syn_cache_insert(sc, tp); 3578 tcpstat.tcps_sndacks++; 3579 tcpstat.tcps_sndtotal++; 3580 } else { 3581 SYN_CACHE_PUT(sc); 3582 tcpstat.tcps_sc_dropped++; 3583 } 3584 return (1); 3585 } 3586 3587 int 3588 syn_cache_respond(sc, m) 3589 struct syn_cache *sc; 3590 struct mbuf *m; 3591 { 3592 struct route *ro; 3593 u_int8_t *optp; 3594 int optlen, error; 3595 u_int16_t tlen; 3596 struct ip *ip = NULL; 3597 #ifdef INET6 3598 struct ip6_hdr *ip6 = NULL; 3599 #endif 3600 struct tcphdr *th; 3601 u_int hlen; 3602 3603 switch (sc->sc_src.sa.sa_family) { 3604 case AF_INET: 3605 hlen = sizeof(struct ip); 3606 ro = &sc->sc_route4; 3607 break; 3608 #ifdef INET6 3609 case AF_INET6: 3610 hlen = sizeof(struct ip6_hdr); 3611 ro = (struct route *)&sc->sc_route6; 3612 break; 3613 #endif 3614 default: 3615 if (m) 3616 m_freem(m); 3617 return EAFNOSUPPORT; 3618 } 3619 3620 /* Compute the size of the TCP options. */ 3621 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3622 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3623 3624 tlen = hlen + sizeof(struct tcphdr) + optlen; 3625 3626 /* 3627 * Create the IP+TCP header from scratch. 3628 */ 3629 if (m) 3630 m_freem(m); 3631 #ifdef DIAGNOSTIC 3632 if (max_linkhdr + tlen > MCLBYTES) 3633 return (ENOBUFS); 3634 #endif 3635 MGETHDR(m, M_DONTWAIT, MT_DATA); 3636 if (m && tlen > MHLEN) { 3637 MCLGET(m, M_DONTWAIT); 3638 if ((m->m_flags & M_EXT) == 0) { 3639 m_freem(m); 3640 m = NULL; 3641 } 3642 } 3643 if (m == NULL) 3644 return (ENOBUFS); 3645 MCLAIM(m, &tcp_tx_mowner); 3646 3647 /* Fixup the mbuf. */ 3648 m->m_data += max_linkhdr; 3649 m->m_len = m->m_pkthdr.len = tlen; 3650 #ifdef IPSEC 3651 if (sc->sc_tp) { 3652 struct tcpcb *tp; 3653 struct socket *so; 3654 3655 tp = sc->sc_tp; 3656 if (tp->t_inpcb) 3657 so = tp->t_inpcb->inp_socket; 3658 #ifdef INET6 3659 else if (tp->t_in6pcb) 3660 so = tp->t_in6pcb->in6p_socket; 3661 #endif 3662 else 3663 so = NULL; 3664 /* use IPsec policy on listening socket, on SYN ACK */ 3665 if (ipsec_setsocket(m, so) != 0) { 3666 m_freem(m); 3667 return ENOBUFS; 3668 } 3669 } 3670 #endif 3671 m->m_pkthdr.rcvif = NULL; 3672 memset(mtod(m, u_char *), 0, tlen); 3673 3674 switch (sc->sc_src.sa.sa_family) { 3675 case AF_INET: 3676 ip = mtod(m, struct ip *); 3677 ip->ip_dst = sc->sc_src.sin.sin_addr; 3678 ip->ip_src = sc->sc_dst.sin.sin_addr; 3679 ip->ip_p = IPPROTO_TCP; 3680 th = (struct tcphdr *)(ip + 1); 3681 th->th_dport = sc->sc_src.sin.sin_port; 3682 th->th_sport = sc->sc_dst.sin.sin_port; 3683 break; 3684 #ifdef INET6 3685 case AF_INET6: 3686 ip6 = mtod(m, struct ip6_hdr *); 3687 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3688 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3689 ip6->ip6_nxt = IPPROTO_TCP; 3690 /* ip6_plen will be updated in ip6_output() */ 3691 th = (struct tcphdr *)(ip6 + 1); 3692 th->th_dport = sc->sc_src.sin6.sin6_port; 3693 th->th_sport = sc->sc_dst.sin6.sin6_port; 3694 break; 3695 #endif 3696 default: 3697 th = NULL; 3698 } 3699 3700 th->th_seq = htonl(sc->sc_iss); 3701 th->th_ack = htonl(sc->sc_irs + 1); 3702 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3703 th->th_flags = TH_SYN|TH_ACK; 3704 th->th_win = htons(sc->sc_win); 3705 /* th_sum already 0 */ 3706 /* th_urp already 0 */ 3707 3708 /* Tack on the TCP options. */ 3709 optp = (u_int8_t *)(th + 1); 3710 *optp++ = TCPOPT_MAXSEG; 3711 *optp++ = 4; 3712 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3713 *optp++ = sc->sc_ourmaxseg & 0xff; 3714 3715 if (sc->sc_request_r_scale != 15) { 3716 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3717 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3718 sc->sc_request_r_scale); 3719 optp += 4; 3720 } 3721 3722 if (sc->sc_flags & SCF_TIMESTAMP) { 3723 u_int32_t *lp = (u_int32_t *)(optp); 3724 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3725 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3726 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3727 *lp = htonl(sc->sc_timestamp); 3728 optp += TCPOLEN_TSTAMP_APPA; 3729 } 3730 3731 /* Compute the packet's checksum. */ 3732 switch (sc->sc_src.sa.sa_family) { 3733 case AF_INET: 3734 ip->ip_len = htons(tlen - hlen); 3735 th->th_sum = 0; 3736 th->th_sum = in_cksum(m, tlen); 3737 break; 3738 #ifdef INET6 3739 case AF_INET6: 3740 ip6->ip6_plen = htons(tlen - hlen); 3741 th->th_sum = 0; 3742 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3743 break; 3744 #endif 3745 } 3746 3747 /* 3748 * Fill in some straggling IP bits. Note the stack expects 3749 * ip_len to be in host order, for convenience. 3750 */ 3751 switch (sc->sc_src.sa.sa_family) { 3752 #ifdef INET 3753 case AF_INET: 3754 ip->ip_len = htons(tlen); 3755 ip->ip_ttl = ip_defttl; 3756 /* XXX tos? */ 3757 break; 3758 #endif 3759 #ifdef INET6 3760 case AF_INET6: 3761 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3762 ip6->ip6_vfc |= IPV6_VERSION; 3763 ip6->ip6_plen = htons(tlen - hlen); 3764 /* ip6_hlim will be initialized afterwards */ 3765 /* XXX flowlabel? */ 3766 break; 3767 #endif 3768 } 3769 3770 switch (sc->sc_src.sa.sa_family) { 3771 #ifdef INET 3772 case AF_INET: 3773 error = ip_output(m, sc->sc_ipopts, ro, 3774 (ip_mtudisc ? IP_MTUDISC : 0), 3775 NULL); 3776 break; 3777 #endif 3778 #ifdef INET6 3779 case AF_INET6: 3780 ip6->ip6_hlim = in6_selecthlim(NULL, 3781 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3782 3783 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3784 0, NULL, NULL); 3785 break; 3786 #endif 3787 default: 3788 error = EAFNOSUPPORT; 3789 break; 3790 } 3791 return (error); 3792 } 3793