1 /* $NetBSD: tcp_input.c,v 1.240 2006/04/15 02:32:22 christos 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, 2005 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 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the NetBSD 93 * Foundation, Inc. and its contributors. 94 * 4. Neither the name of The NetBSD Foundation nor the names of its 95 * contributors may be used to endorse or promote products derived 96 * from this software without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 99 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 100 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 101 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 102 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 103 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 104 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 105 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 106 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 107 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 108 * POSSIBILITY OF SUCH DAMAGE. 109 */ 110 111 /* 112 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 113 * The Regents of the University of California. All rights reserved. 114 * 115 * Redistribution and use in source and binary forms, with or without 116 * modification, are permitted provided that the following conditions 117 * are met: 118 * 1. Redistributions of source code must retain the above copyright 119 * notice, this list of conditions and the following disclaimer. 120 * 2. Redistributions in binary form must reproduce the above copyright 121 * notice, this list of conditions and the following disclaimer in the 122 * documentation and/or other materials provided with the distribution. 123 * 3. Neither the name of the University nor the names of its contributors 124 * may be used to endorse or promote products derived from this software 125 * without specific prior written permission. 126 * 127 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 128 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 129 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 130 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 131 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 132 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 133 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 134 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 135 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 136 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 137 * SUCH DAMAGE. 138 * 139 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 140 */ 141 142 /* 143 * TODO list for SYN cache stuff: 144 * 145 * Find room for a "state" field, which is needed to keep a 146 * compressed state for TIME_WAIT TCBs. It's been noted already 147 * that this is fairly important for very high-volume web and 148 * mail servers, which use a large number of short-lived 149 * connections. 150 */ 151 152 #include <sys/cdefs.h> 153 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.240 2006/04/15 02:32:22 christos Exp $"); 154 155 #include "opt_inet.h" 156 #include "opt_ipsec.h" 157 #include "opt_inet_csum.h" 158 #include "opt_tcp_debug.h" 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 176 #include <net/if.h> 177 #include <net/route.h> 178 #include <net/if_types.h> 179 180 #include <netinet/in.h> 181 #include <netinet/in_systm.h> 182 #include <netinet/ip.h> 183 #include <netinet/in_pcb.h> 184 #include <netinet/in_var.h> 185 #include <netinet/ip_var.h> 186 #include <netinet/in_offload.h> 187 188 #ifdef INET6 189 #ifndef INET 190 #include <netinet/in.h> 191 #endif 192 #include <netinet/ip6.h> 193 #include <netinet6/ip6_var.h> 194 #include <netinet6/in6_pcb.h> 195 #include <netinet6/ip6_var.h> 196 #include <netinet6/in6_var.h> 197 #include <netinet/icmp6.h> 198 #include <netinet6/nd6.h> 199 #ifdef TCP_SIGNATURE 200 #include <netinet6/scope6_var.h> 201 #endif 202 #endif 203 204 #ifndef INET6 205 /* always need ip6.h for IP6_EXTHDR_GET */ 206 #include <netinet/ip6.h> 207 #endif 208 209 #include <netinet/tcp.h> 210 #include <netinet/tcp_fsm.h> 211 #include <netinet/tcp_seq.h> 212 #include <netinet/tcp_timer.h> 213 #include <netinet/tcp_var.h> 214 #include <netinet/tcpip.h> 215 #include <netinet/tcp_debug.h> 216 217 #include <machine/stdarg.h> 218 219 #ifdef IPSEC 220 #include <netinet6/ipsec.h> 221 #include <netkey/key.h> 222 #endif /*IPSEC*/ 223 #ifdef INET6 224 #include "faith.h" 225 #if defined(NFAITH) && NFAITH > 0 226 #include <net/if_faith.h> 227 #endif 228 #endif /* IPSEC */ 229 230 #ifdef FAST_IPSEC 231 #include <netipsec/ipsec.h> 232 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 233 #include <netipsec/key.h> 234 #ifdef INET6 235 #include <netipsec/ipsec6.h> 236 #endif 237 #endif /* FAST_IPSEC*/ 238 239 int tcprexmtthresh = 3; 240 int tcp_log_refused; 241 242 static int tcp_rst_ppslim_count = 0; 243 static struct timeval tcp_rst_ppslim_last; 244 static int tcp_ackdrop_ppslim_count = 0; 245 static struct timeval tcp_ackdrop_ppslim_last; 246 247 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 248 249 /* for modulo comparisons of timestamps */ 250 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 251 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 252 253 /* 254 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 255 */ 256 #ifdef INET6 257 #define ND6_HINT(tp) \ 258 do { \ 259 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 260 tp->t_in6pcb->in6p_route.ro_rt) { \ 261 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 262 } \ 263 } while (/*CONSTCOND*/ 0) 264 #else 265 #define ND6_HINT(tp) 266 #endif 267 268 /* 269 * Macro to compute ACK transmission behavior. Delay the ACK unless 270 * we have already delayed an ACK (must send an ACK every two segments). 271 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 272 * option is enabled. 273 */ 274 #define TCP_SETUP_ACK(tp, th) \ 275 do { \ 276 if ((tp)->t_flags & TF_DELACK || \ 277 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 278 tp->t_flags |= TF_ACKNOW; \ 279 else \ 280 TCP_SET_DELACK(tp); \ 281 } while (/*CONSTCOND*/ 0) 282 283 #define ICMP_CHECK(tp, th, acked) \ 284 do { \ 285 /* \ 286 * If we had a pending ICMP message that \ 287 * refers to data that have just been \ 288 * acknowledged, disregard the recorded ICMP \ 289 * message. \ 290 */ \ 291 if (((tp)->t_flags & TF_PMTUD_PEND) && \ 292 SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \ 293 (tp)->t_flags &= ~TF_PMTUD_PEND; \ 294 \ 295 /* \ 296 * Keep track of the largest chunk of data \ 297 * acknowledged since last PMTU update \ 298 */ \ 299 if ((tp)->t_pmtud_mss_acked < (acked)) \ 300 (tp)->t_pmtud_mss_acked = (acked); \ 301 } while (/*CONSTCOND*/ 0) 302 303 /* 304 * Convert TCP protocol fields to host order for easier processing. 305 */ 306 #define TCP_FIELDS_TO_HOST(th) \ 307 do { \ 308 NTOHL((th)->th_seq); \ 309 NTOHL((th)->th_ack); \ 310 NTOHS((th)->th_win); \ 311 NTOHS((th)->th_urp); \ 312 } while (/*CONSTCOND*/ 0) 313 314 /* 315 * ... and reverse the above. 316 */ 317 #define TCP_FIELDS_TO_NET(th) \ 318 do { \ 319 HTONL((th)->th_seq); \ 320 HTONL((th)->th_ack); \ 321 HTONS((th)->th_win); \ 322 HTONS((th)->th_urp); \ 323 } while (/*CONSTCOND*/ 0) 324 325 #ifdef TCP_CSUM_COUNTERS 326 #include <sys/device.h> 327 328 #if defined(INET) 329 extern struct evcnt tcp_hwcsum_ok; 330 extern struct evcnt tcp_hwcsum_bad; 331 extern struct evcnt tcp_hwcsum_data; 332 extern struct evcnt tcp_swcsum; 333 #endif /* defined(INET) */ 334 #if defined(INET6) 335 extern struct evcnt tcp6_hwcsum_ok; 336 extern struct evcnt tcp6_hwcsum_bad; 337 extern struct evcnt tcp6_hwcsum_data; 338 extern struct evcnt tcp6_swcsum; 339 #endif /* defined(INET6) */ 340 341 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 342 343 #else 344 345 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 346 347 #endif /* TCP_CSUM_COUNTERS */ 348 349 #ifdef TCP_REASS_COUNTERS 350 #include <sys/device.h> 351 352 extern struct evcnt tcp_reass_; 353 extern struct evcnt tcp_reass_empty; 354 extern struct evcnt tcp_reass_iteration[8]; 355 extern struct evcnt tcp_reass_prependfirst; 356 extern struct evcnt tcp_reass_prepend; 357 extern struct evcnt tcp_reass_insert; 358 extern struct evcnt tcp_reass_inserttail; 359 extern struct evcnt tcp_reass_append; 360 extern struct evcnt tcp_reass_appendtail; 361 extern struct evcnt tcp_reass_overlaptail; 362 extern struct evcnt tcp_reass_overlapfront; 363 extern struct evcnt tcp_reass_segdup; 364 extern struct evcnt tcp_reass_fragdup; 365 366 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 367 368 #else 369 370 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 371 372 #endif /* TCP_REASS_COUNTERS */ 373 374 #ifdef INET 375 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 376 #endif 377 #ifdef INET6 378 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 379 #endif 380 381 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 382 383 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 384 385 struct ipqent * 386 tcpipqent_alloc() 387 { 388 struct ipqent *ipqe; 389 int s; 390 391 s = splvm(); 392 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 393 splx(s); 394 395 return ipqe; 396 } 397 398 void 399 tcpipqent_free(struct ipqent *ipqe) 400 { 401 int s; 402 403 s = splvm(); 404 pool_put(&tcpipqent_pool, ipqe); 405 splx(s); 406 } 407 408 int 409 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 410 { 411 struct ipqent *p, *q, *nq, *tiqe = NULL; 412 struct socket *so = NULL; 413 int pkt_flags; 414 tcp_seq pkt_seq; 415 unsigned pkt_len; 416 u_long rcvpartdupbyte = 0; 417 u_long rcvoobyte; 418 #ifdef TCP_REASS_COUNTERS 419 u_int count = 0; 420 #endif 421 422 if (tp->t_inpcb) 423 so = tp->t_inpcb->inp_socket; 424 #ifdef INET6 425 else if (tp->t_in6pcb) 426 so = tp->t_in6pcb->in6p_socket; 427 #endif 428 429 TCP_REASS_LOCK_CHECK(tp); 430 431 /* 432 * Call with th==0 after become established to 433 * force pre-ESTABLISHED data up to user socket. 434 */ 435 if (th == 0) 436 goto present; 437 438 rcvoobyte = *tlen; 439 /* 440 * Copy these to local variables because the tcpiphdr 441 * gets munged while we are collapsing mbufs. 442 */ 443 pkt_seq = th->th_seq; 444 pkt_len = *tlen; 445 pkt_flags = th->th_flags; 446 447 TCP_REASS_COUNTER_INCR(&tcp_reass_); 448 449 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 450 /* 451 * When we miss a packet, the vast majority of time we get 452 * packets that follow it in order. So optimize for that. 453 */ 454 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 455 p->ipqe_len += pkt_len; 456 p->ipqe_flags |= pkt_flags; 457 m_cat(p->ipre_mlast, m); 458 TRAVERSE(p->ipre_mlast); 459 m = NULL; 460 tiqe = p; 461 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 462 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 463 goto skip_replacement; 464 } 465 /* 466 * While we're here, if the pkt is completely beyond 467 * anything we have, just insert it at the tail. 468 */ 469 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 470 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 471 goto insert_it; 472 } 473 } 474 475 q = TAILQ_FIRST(&tp->segq); 476 477 if (q != NULL) { 478 /* 479 * If this segment immediately precedes the first out-of-order 480 * block, simply slap the segment in front of it and (mostly) 481 * skip the complicated logic. 482 */ 483 if (pkt_seq + pkt_len == q->ipqe_seq) { 484 q->ipqe_seq = pkt_seq; 485 q->ipqe_len += pkt_len; 486 q->ipqe_flags |= pkt_flags; 487 m_cat(m, q->ipqe_m); 488 q->ipqe_m = m; 489 q->ipre_mlast = m; /* last mbuf may have changed */ 490 TRAVERSE(q->ipre_mlast); 491 tiqe = q; 492 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 493 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 494 goto skip_replacement; 495 } 496 } else { 497 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 498 } 499 500 /* 501 * Find a segment which begins after this one does. 502 */ 503 for (p = NULL; q != NULL; q = nq) { 504 nq = TAILQ_NEXT(q, ipqe_q); 505 #ifdef TCP_REASS_COUNTERS 506 count++; 507 #endif 508 /* 509 * If the received segment is just right after this 510 * fragment, merge the two together and then check 511 * for further overlaps. 512 */ 513 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 514 #ifdef TCPREASS_DEBUG 515 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 516 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 517 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 518 #endif 519 pkt_len += q->ipqe_len; 520 pkt_flags |= q->ipqe_flags; 521 pkt_seq = q->ipqe_seq; 522 m_cat(q->ipre_mlast, m); 523 TRAVERSE(q->ipre_mlast); 524 m = q->ipqe_m; 525 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 526 goto free_ipqe; 527 } 528 /* 529 * If the received segment is completely past this 530 * fragment, we need to go the next fragment. 531 */ 532 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 533 p = q; 534 continue; 535 } 536 /* 537 * If the fragment is past the received segment, 538 * it (or any following) can't be concatenated. 539 */ 540 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 541 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 542 break; 543 } 544 545 /* 546 * We've received all the data in this segment before. 547 * mark it as a duplicate and return. 548 */ 549 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 550 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 551 tcpstat.tcps_rcvduppack++; 552 tcpstat.tcps_rcvdupbyte += pkt_len; 553 tcp_new_dsack(tp, pkt_seq, pkt_len); 554 m_freem(m); 555 if (tiqe != NULL) { 556 tcpipqent_free(tiqe); 557 } 558 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 559 return (0); 560 } 561 /* 562 * Received segment completely overlaps this fragment 563 * so we drop the fragment (this keeps the temporal 564 * ordering of segments correct). 565 */ 566 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 567 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 568 rcvpartdupbyte += q->ipqe_len; 569 m_freem(q->ipqe_m); 570 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 571 goto free_ipqe; 572 } 573 /* 574 * RX'ed segment extends past the end of the 575 * fragment. Drop the overlapping bytes. Then 576 * merge the fragment and segment then treat as 577 * a longer received packet. 578 */ 579 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 580 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 581 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 582 #ifdef TCPREASS_DEBUG 583 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 584 tp, overlap, 585 pkt_seq, pkt_seq + pkt_len, pkt_len); 586 #endif 587 m_adj(m, overlap); 588 rcvpartdupbyte += overlap; 589 m_cat(q->ipre_mlast, m); 590 TRAVERSE(q->ipre_mlast); 591 m = q->ipqe_m; 592 pkt_seq = q->ipqe_seq; 593 pkt_len += q->ipqe_len - overlap; 594 rcvoobyte -= overlap; 595 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 596 goto free_ipqe; 597 } 598 /* 599 * RX'ed segment extends past the front of the 600 * fragment. Drop the overlapping bytes on the 601 * received packet. The packet will then be 602 * contatentated with this fragment a bit later. 603 */ 604 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 605 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 606 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 607 #ifdef TCPREASS_DEBUG 608 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 609 tp, overlap, 610 pkt_seq, pkt_seq + pkt_len, pkt_len); 611 #endif 612 m_adj(m, -overlap); 613 pkt_len -= overlap; 614 rcvpartdupbyte += overlap; 615 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 616 rcvoobyte -= overlap; 617 } 618 /* 619 * If the received segment immediates precedes this 620 * fragment then tack the fragment onto this segment 621 * and reinsert the data. 622 */ 623 if (q->ipqe_seq == pkt_seq + pkt_len) { 624 #ifdef TCPREASS_DEBUG 625 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 626 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 627 pkt_seq, pkt_seq + pkt_len, pkt_len); 628 #endif 629 pkt_len += q->ipqe_len; 630 pkt_flags |= q->ipqe_flags; 631 m_cat(m, q->ipqe_m); 632 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 633 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 634 tp->t_segqlen--; 635 KASSERT(tp->t_segqlen >= 0); 636 KASSERT(tp->t_segqlen != 0 || 637 (TAILQ_EMPTY(&tp->segq) && 638 TAILQ_EMPTY(&tp->timeq))); 639 if (tiqe == NULL) { 640 tiqe = q; 641 } else { 642 tcpipqent_free(q); 643 } 644 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 645 break; 646 } 647 /* 648 * If the fragment is before the segment, remember it. 649 * When this loop is terminated, p will contain the 650 * pointer to fragment that is right before the received 651 * segment. 652 */ 653 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 654 p = q; 655 656 continue; 657 658 /* 659 * This is a common operation. It also will allow 660 * to save doing a malloc/free in most instances. 661 */ 662 free_ipqe: 663 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 664 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 665 tp->t_segqlen--; 666 KASSERT(tp->t_segqlen >= 0); 667 KASSERT(tp->t_segqlen != 0 || 668 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 669 if (tiqe == NULL) { 670 tiqe = q; 671 } else { 672 tcpipqent_free(q); 673 } 674 } 675 676 #ifdef TCP_REASS_COUNTERS 677 if (count > 7) 678 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 679 else if (count > 0) 680 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 681 #endif 682 683 insert_it: 684 685 /* 686 * Allocate a new queue entry since the received segment did not 687 * collapse onto any other out-of-order block; thus we are allocating 688 * a new block. If it had collapsed, tiqe would not be NULL and 689 * we would be reusing it. 690 * XXX If we can't, just drop the packet. XXX 691 */ 692 if (tiqe == NULL) { 693 tiqe = tcpipqent_alloc(); 694 if (tiqe == NULL) { 695 tcpstat.tcps_rcvmemdrop++; 696 m_freem(m); 697 return (0); 698 } 699 } 700 701 /* 702 * Update the counters. 703 */ 704 tcpstat.tcps_rcvoopack++; 705 tcpstat.tcps_rcvoobyte += rcvoobyte; 706 if (rcvpartdupbyte) { 707 tcpstat.tcps_rcvpartduppack++; 708 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 709 } 710 711 /* 712 * Insert the new fragment queue entry into both queues. 713 */ 714 tiqe->ipqe_m = m; 715 tiqe->ipre_mlast = m; 716 tiqe->ipqe_seq = pkt_seq; 717 tiqe->ipqe_len = pkt_len; 718 tiqe->ipqe_flags = pkt_flags; 719 if (p == NULL) { 720 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 721 #ifdef TCPREASS_DEBUG 722 if (tiqe->ipqe_seq != tp->rcv_nxt) 723 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 724 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 725 #endif 726 } else { 727 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 728 #ifdef TCPREASS_DEBUG 729 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 730 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 731 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 732 #endif 733 } 734 tp->t_segqlen++; 735 736 skip_replacement: 737 738 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 739 740 present: 741 /* 742 * Present data to user, advancing rcv_nxt through 743 * completed sequence space. 744 */ 745 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 746 return (0); 747 q = TAILQ_FIRST(&tp->segq); 748 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 749 return (0); 750 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 751 return (0); 752 753 tp->rcv_nxt += q->ipqe_len; 754 pkt_flags = q->ipqe_flags & TH_FIN; 755 ND6_HINT(tp); 756 757 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 758 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 759 tp->t_segqlen--; 760 KASSERT(tp->t_segqlen >= 0); 761 KASSERT(tp->t_segqlen != 0 || 762 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 763 if (so->so_state & SS_CANTRCVMORE) 764 m_freem(q->ipqe_m); 765 else 766 sbappendstream(&so->so_rcv, q->ipqe_m); 767 tcpipqent_free(q); 768 sorwakeup(so); 769 return (pkt_flags); 770 } 771 772 #ifdef INET6 773 int 774 tcp6_input(struct mbuf **mp, int *offp, int proto) 775 { 776 struct mbuf *m = *mp; 777 778 /* 779 * draft-itojun-ipv6-tcp-to-anycast 780 * better place to put this in? 781 */ 782 if (m->m_flags & M_ANYCAST6) { 783 struct ip6_hdr *ip6; 784 if (m->m_len < sizeof(struct ip6_hdr)) { 785 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 786 tcpstat.tcps_rcvshort++; 787 return IPPROTO_DONE; 788 } 789 } 790 ip6 = mtod(m, struct ip6_hdr *); 791 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 792 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 793 return IPPROTO_DONE; 794 } 795 796 tcp_input(m, *offp, proto); 797 return IPPROTO_DONE; 798 } 799 #endif 800 801 #ifdef INET 802 static void 803 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 804 { 805 char src[4*sizeof "123"]; 806 char dst[4*sizeof "123"]; 807 808 if (ip) { 809 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 810 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 811 } 812 else { 813 strlcpy(src, "(unknown)", sizeof(src)); 814 strlcpy(dst, "(unknown)", sizeof(dst)); 815 } 816 log(LOG_INFO, 817 "Connection attempt to TCP %s:%d from %s:%d\n", 818 dst, ntohs(th->th_dport), 819 src, ntohs(th->th_sport)); 820 } 821 #endif 822 823 #ifdef INET6 824 static void 825 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 826 { 827 char src[INET6_ADDRSTRLEN]; 828 char dst[INET6_ADDRSTRLEN]; 829 830 if (ip6) { 831 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 832 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 833 } 834 else { 835 strlcpy(src, "(unknown v6)", sizeof(src)); 836 strlcpy(dst, "(unknown v6)", sizeof(dst)); 837 } 838 log(LOG_INFO, 839 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 840 dst, ntohs(th->th_dport), 841 src, ntohs(th->th_sport)); 842 } 843 #endif 844 845 /* 846 * Checksum extended TCP header and data. 847 */ 848 int 849 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 850 int off, int tlen) 851 { 852 853 /* 854 * XXX it's better to record and check if this mbuf is 855 * already checked. 856 */ 857 858 switch (af) { 859 #ifdef INET 860 case AF_INET: 861 switch (m->m_pkthdr.csum_flags & 862 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 863 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 864 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 865 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 866 goto badcsum; 867 868 case M_CSUM_TCPv4|M_CSUM_DATA: { 869 u_int32_t hw_csum = m->m_pkthdr.csum_data; 870 871 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 872 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 873 const struct ip *ip = 874 mtod(m, const struct ip *); 875 876 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 877 ip->ip_dst.s_addr, 878 htons(hw_csum + tlen + off + IPPROTO_TCP)); 879 } 880 if ((hw_csum ^ 0xffff) != 0) 881 goto badcsum; 882 break; 883 } 884 885 case M_CSUM_TCPv4: 886 /* Checksum was okay. */ 887 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 888 break; 889 890 default: 891 /* 892 * Must compute it ourselves. Maybe skip checksum 893 * on loopback interfaces. 894 */ 895 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 896 IFF_LOOPBACK) || 897 tcp_do_loopback_cksum)) { 898 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 899 if (in4_cksum(m, IPPROTO_TCP, toff, 900 tlen + off) != 0) 901 goto badcsum; 902 } 903 break; 904 } 905 break; 906 #endif /* INET4 */ 907 908 #ifdef INET6 909 case AF_INET6: 910 switch (m->m_pkthdr.csum_flags & 911 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 912 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 913 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 914 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 915 goto badcsum; 916 917 #if 0 /* notyet */ 918 case M_CSUM_TCPv6|M_CSUM_DATA: 919 #endif 920 921 case M_CSUM_TCPv6: 922 /* Checksum was okay. */ 923 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 924 break; 925 926 default: 927 /* 928 * Must compute it ourselves. Maybe skip checksum 929 * on loopback interfaces. 930 */ 931 if (__predict_true((m->m_flags & M_LOOP) == 0 || 932 tcp_do_loopback_cksum)) { 933 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 934 if (in6_cksum(m, IPPROTO_TCP, toff, 935 tlen + off) != 0) 936 goto badcsum; 937 } 938 } 939 break; 940 #endif /* INET6 */ 941 } 942 943 return 0; 944 945 badcsum: 946 tcpstat.tcps_rcvbadsum++; 947 return -1; 948 } 949 950 /* 951 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 952 */ 953 void 954 tcp_input(struct mbuf *m, ...) 955 { 956 struct tcphdr *th; 957 struct ip *ip; 958 struct inpcb *inp; 959 #ifdef INET6 960 struct ip6_hdr *ip6; 961 struct in6pcb *in6p; 962 #endif 963 u_int8_t *optp = NULL; 964 int optlen = 0; 965 int len, tlen, toff, hdroptlen = 0; 966 struct tcpcb *tp = 0; 967 int tiflags; 968 struct socket *so = NULL; 969 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 970 #ifdef TCP_DEBUG 971 short ostate = 0; 972 #endif 973 int iss = 0; 974 u_long tiwin; 975 struct tcp_opt_info opti; 976 int off, iphlen; 977 va_list ap; 978 int af; /* af on the wire */ 979 struct mbuf *tcp_saveti = NULL; 980 uint32_t ts_rtt; 981 982 MCLAIM(m, &tcp_rx_mowner); 983 va_start(ap, m); 984 toff = va_arg(ap, int); 985 (void)va_arg(ap, int); /* ignore value, advance ap */ 986 va_end(ap); 987 988 tcpstat.tcps_rcvtotal++; 989 990 bzero(&opti, sizeof(opti)); 991 opti.ts_present = 0; 992 opti.maxseg = 0; 993 994 /* 995 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 996 * 997 * TCP is, by definition, unicast, so we reject all 998 * multicast outright. 999 * 1000 * Note, there are additional src/dst address checks in 1001 * the AF-specific code below. 1002 */ 1003 if (m->m_flags & (M_BCAST|M_MCAST)) { 1004 /* XXX stat */ 1005 goto drop; 1006 } 1007 #ifdef INET6 1008 if (m->m_flags & M_ANYCAST6) { 1009 /* XXX stat */ 1010 goto drop; 1011 } 1012 #endif 1013 1014 /* 1015 * Get IP and TCP header. 1016 * Note: IP leaves IP header in first mbuf. 1017 */ 1018 ip = mtod(m, struct ip *); 1019 #ifdef INET6 1020 ip6 = NULL; 1021 #endif 1022 switch (ip->ip_v) { 1023 #ifdef INET 1024 case 4: 1025 af = AF_INET; 1026 iphlen = sizeof(struct ip); 1027 ip = mtod(m, struct ip *); 1028 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1029 sizeof(struct tcphdr)); 1030 if (th == NULL) { 1031 tcpstat.tcps_rcvshort++; 1032 return; 1033 } 1034 /* We do the checksum after PCB lookup... */ 1035 len = ntohs(ip->ip_len); 1036 tlen = len - toff; 1037 break; 1038 #endif 1039 #ifdef INET6 1040 case 6: 1041 ip = NULL; 1042 iphlen = sizeof(struct ip6_hdr); 1043 af = AF_INET6; 1044 ip6 = mtod(m, struct ip6_hdr *); 1045 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1046 sizeof(struct tcphdr)); 1047 if (th == NULL) { 1048 tcpstat.tcps_rcvshort++; 1049 return; 1050 } 1051 1052 /* Be proactive about malicious use of IPv4 mapped address */ 1053 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1054 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1055 /* XXX stat */ 1056 goto drop; 1057 } 1058 1059 /* 1060 * Be proactive about unspecified IPv6 address in source. 1061 * As we use all-zero to indicate unbounded/unconnected pcb, 1062 * unspecified IPv6 address can be used to confuse us. 1063 * 1064 * Note that packets with unspecified IPv6 destination is 1065 * already dropped in ip6_input. 1066 */ 1067 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1068 /* XXX stat */ 1069 goto drop; 1070 } 1071 1072 /* 1073 * Make sure destination address is not multicast. 1074 * Source address checked in ip6_input(). 1075 */ 1076 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1077 /* XXX stat */ 1078 goto drop; 1079 } 1080 1081 /* We do the checksum after PCB lookup... */ 1082 len = m->m_pkthdr.len; 1083 tlen = len - toff; 1084 break; 1085 #endif 1086 default: 1087 m_freem(m); 1088 return; 1089 } 1090 1091 KASSERT(TCP_HDR_ALIGNED_P(th)); 1092 1093 /* 1094 * Check that TCP offset makes sense, 1095 * pull out TCP options and adjust length. XXX 1096 */ 1097 off = th->th_off << 2; 1098 if (off < sizeof (struct tcphdr) || off > tlen) { 1099 tcpstat.tcps_rcvbadoff++; 1100 goto drop; 1101 } 1102 tlen -= off; 1103 1104 /* 1105 * tcp_input() has been modified to use tlen to mean the TCP data 1106 * length throughout the function. Other functions can use 1107 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1108 * rja 1109 */ 1110 1111 if (off > sizeof (struct tcphdr)) { 1112 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1113 if (th == NULL) { 1114 tcpstat.tcps_rcvshort++; 1115 return; 1116 } 1117 /* 1118 * NOTE: ip/ip6 will not be affected by m_pulldown() 1119 * (as they're before toff) and we don't need to update those. 1120 */ 1121 KASSERT(TCP_HDR_ALIGNED_P(th)); 1122 optlen = off - sizeof (struct tcphdr); 1123 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1124 /* 1125 * Do quick retrieval of timestamp options ("options 1126 * prediction?"). If timestamp is the only option and it's 1127 * formatted as recommended in RFC 1323 appendix A, we 1128 * quickly get the values now and not bother calling 1129 * tcp_dooptions(), etc. 1130 */ 1131 if ((optlen == TCPOLEN_TSTAMP_APPA || 1132 (optlen > TCPOLEN_TSTAMP_APPA && 1133 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1134 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1135 (th->th_flags & TH_SYN) == 0) { 1136 opti.ts_present = 1; 1137 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1138 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1139 optp = NULL; /* we've parsed the options */ 1140 } 1141 } 1142 tiflags = th->th_flags; 1143 1144 /* 1145 * Locate pcb for segment. 1146 */ 1147 findpcb: 1148 inp = NULL; 1149 #ifdef INET6 1150 in6p = NULL; 1151 #endif 1152 switch (af) { 1153 #ifdef INET 1154 case AF_INET: 1155 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1156 ip->ip_dst, th->th_dport); 1157 if (inp == 0) { 1158 ++tcpstat.tcps_pcbhashmiss; 1159 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1160 } 1161 #ifdef INET6 1162 if (inp == 0) { 1163 struct in6_addr s, d; 1164 1165 /* mapped addr case */ 1166 bzero(&s, sizeof(s)); 1167 s.s6_addr16[5] = htons(0xffff); 1168 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1169 bzero(&d, sizeof(d)); 1170 d.s6_addr16[5] = htons(0xffff); 1171 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1172 in6p = in6_pcblookup_connect(&tcbtable, &s, 1173 th->th_sport, &d, th->th_dport, 0); 1174 if (in6p == 0) { 1175 ++tcpstat.tcps_pcbhashmiss; 1176 in6p = in6_pcblookup_bind(&tcbtable, &d, 1177 th->th_dport, 0); 1178 } 1179 } 1180 #endif 1181 #ifndef INET6 1182 if (inp == 0) 1183 #else 1184 if (inp == 0 && in6p == 0) 1185 #endif 1186 { 1187 ++tcpstat.tcps_noport; 1188 if (tcp_log_refused && 1189 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1190 tcp4_log_refused(ip, th); 1191 } 1192 TCP_FIELDS_TO_HOST(th); 1193 goto dropwithreset_ratelim; 1194 } 1195 #if defined(IPSEC) || defined(FAST_IPSEC) 1196 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1197 ipsec4_in_reject(m, inp)) { 1198 ipsecstat.in_polvio++; 1199 goto drop; 1200 } 1201 #ifdef INET6 1202 else if (in6p && 1203 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1204 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1205 ipsecstat.in_polvio++; 1206 goto drop; 1207 } 1208 #endif 1209 #endif /*IPSEC*/ 1210 break; 1211 #endif /*INET*/ 1212 #ifdef INET6 1213 case AF_INET6: 1214 { 1215 int faith; 1216 1217 #if defined(NFAITH) && NFAITH > 0 1218 faith = faithprefix(&ip6->ip6_dst); 1219 #else 1220 faith = 0; 1221 #endif 1222 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1223 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1224 if (in6p == NULL) { 1225 ++tcpstat.tcps_pcbhashmiss; 1226 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1227 th->th_dport, faith); 1228 } 1229 if (in6p == NULL) { 1230 ++tcpstat.tcps_noport; 1231 if (tcp_log_refused && 1232 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1233 tcp6_log_refused(ip6, th); 1234 } 1235 TCP_FIELDS_TO_HOST(th); 1236 goto dropwithreset_ratelim; 1237 } 1238 #if defined(IPSEC) || defined(FAST_IPSEC) 1239 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1240 ipsec6_in_reject(m, in6p)) { 1241 ipsec6stat.in_polvio++; 1242 goto drop; 1243 } 1244 #endif /*IPSEC*/ 1245 break; 1246 } 1247 #endif 1248 } 1249 1250 /* 1251 * If the state is CLOSED (i.e., TCB does not exist) then 1252 * all data in the incoming segment is discarded. 1253 * If the TCB exists but is in CLOSED state, it is embryonic, 1254 * but should either do a listen or a connect soon. 1255 */ 1256 tp = NULL; 1257 so = NULL; 1258 if (inp) { 1259 tp = intotcpcb(inp); 1260 so = inp->inp_socket; 1261 } 1262 #ifdef INET6 1263 else if (in6p) { 1264 tp = in6totcpcb(in6p); 1265 so = in6p->in6p_socket; 1266 } 1267 #endif 1268 if (tp == 0) { 1269 TCP_FIELDS_TO_HOST(th); 1270 goto dropwithreset_ratelim; 1271 } 1272 if (tp->t_state == TCPS_CLOSED) 1273 goto drop; 1274 1275 /* 1276 * Checksum extended TCP header and data. 1277 */ 1278 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1279 goto badcsum; 1280 1281 TCP_FIELDS_TO_HOST(th); 1282 1283 /* Unscale the window into a 32-bit value. */ 1284 if ((tiflags & TH_SYN) == 0) 1285 tiwin = th->th_win << tp->snd_scale; 1286 else 1287 tiwin = th->th_win; 1288 1289 #ifdef INET6 1290 /* save packet options if user wanted */ 1291 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1292 if (in6p->in6p_options) { 1293 m_freem(in6p->in6p_options); 1294 in6p->in6p_options = 0; 1295 } 1296 KASSERT(ip6 != NULL); 1297 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1298 } 1299 #endif 1300 1301 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1302 union syn_cache_sa src; 1303 union syn_cache_sa dst; 1304 1305 bzero(&src, sizeof(src)); 1306 bzero(&dst, sizeof(dst)); 1307 switch (af) { 1308 #ifdef INET 1309 case AF_INET: 1310 src.sin.sin_len = sizeof(struct sockaddr_in); 1311 src.sin.sin_family = AF_INET; 1312 src.sin.sin_addr = ip->ip_src; 1313 src.sin.sin_port = th->th_sport; 1314 1315 dst.sin.sin_len = sizeof(struct sockaddr_in); 1316 dst.sin.sin_family = AF_INET; 1317 dst.sin.sin_addr = ip->ip_dst; 1318 dst.sin.sin_port = th->th_dport; 1319 break; 1320 #endif 1321 #ifdef INET6 1322 case AF_INET6: 1323 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1324 src.sin6.sin6_family = AF_INET6; 1325 src.sin6.sin6_addr = ip6->ip6_src; 1326 src.sin6.sin6_port = th->th_sport; 1327 1328 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1329 dst.sin6.sin6_family = AF_INET6; 1330 dst.sin6.sin6_addr = ip6->ip6_dst; 1331 dst.sin6.sin6_port = th->th_dport; 1332 break; 1333 #endif /* INET6 */ 1334 default: 1335 goto badsyn; /*sanity*/ 1336 } 1337 1338 if (so->so_options & SO_DEBUG) { 1339 #ifdef TCP_DEBUG 1340 ostate = tp->t_state; 1341 #endif 1342 1343 tcp_saveti = NULL; 1344 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1345 goto nosave; 1346 1347 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1348 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1349 if (!tcp_saveti) 1350 goto nosave; 1351 } else { 1352 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1353 if (!tcp_saveti) 1354 goto nosave; 1355 MCLAIM(m, &tcp_mowner); 1356 tcp_saveti->m_len = iphlen; 1357 m_copydata(m, 0, iphlen, 1358 mtod(tcp_saveti, caddr_t)); 1359 } 1360 1361 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1362 m_freem(tcp_saveti); 1363 tcp_saveti = NULL; 1364 } else { 1365 tcp_saveti->m_len += sizeof(struct tcphdr); 1366 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1367 sizeof(struct tcphdr)); 1368 } 1369 nosave:; 1370 } 1371 if (so->so_options & SO_ACCEPTCONN) { 1372 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1373 if (tiflags & TH_RST) { 1374 syn_cache_reset(&src.sa, &dst.sa, th); 1375 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1376 (TH_ACK|TH_SYN)) { 1377 /* 1378 * Received a SYN,ACK. This should 1379 * never happen while we are in 1380 * LISTEN. Send an RST. 1381 */ 1382 goto badsyn; 1383 } else if (tiflags & TH_ACK) { 1384 so = syn_cache_get(&src.sa, &dst.sa, 1385 th, toff, tlen, so, m); 1386 if (so == NULL) { 1387 /* 1388 * We don't have a SYN for 1389 * this ACK; send an RST. 1390 */ 1391 goto badsyn; 1392 } else if (so == 1393 (struct socket *)(-1)) { 1394 /* 1395 * We were unable to create 1396 * the connection. If the 1397 * 3-way handshake was 1398 * completed, and RST has 1399 * been sent to the peer. 1400 * Since the mbuf might be 1401 * in use for the reply, 1402 * do not free it. 1403 */ 1404 m = NULL; 1405 } else { 1406 /* 1407 * We have created a 1408 * full-blown connection. 1409 */ 1410 tp = NULL; 1411 inp = NULL; 1412 #ifdef INET6 1413 in6p = NULL; 1414 #endif 1415 switch (so->so_proto->pr_domain->dom_family) { 1416 #ifdef INET 1417 case AF_INET: 1418 inp = sotoinpcb(so); 1419 tp = intotcpcb(inp); 1420 break; 1421 #endif 1422 #ifdef INET6 1423 case AF_INET6: 1424 in6p = sotoin6pcb(so); 1425 tp = in6totcpcb(in6p); 1426 break; 1427 #endif 1428 } 1429 if (tp == NULL) 1430 goto badsyn; /*XXX*/ 1431 tiwin <<= tp->snd_scale; 1432 goto after_listen; 1433 } 1434 } else { 1435 /* 1436 * None of RST, SYN or ACK was set. 1437 * This is an invalid packet for a 1438 * TCB in LISTEN state. Send a RST. 1439 */ 1440 goto badsyn; 1441 } 1442 } else { 1443 /* 1444 * Received a SYN. 1445 */ 1446 1447 #ifdef INET6 1448 /* 1449 * If deprecated address is forbidden, we do 1450 * not accept SYN to deprecated interface 1451 * address to prevent any new inbound 1452 * connection from getting established. 1453 * When we do not accept SYN, we send a TCP 1454 * RST, with deprecated source address (instead 1455 * of dropping it). We compromise it as it is 1456 * much better for peer to send a RST, and 1457 * RST will be the final packet for the 1458 * exchange. 1459 * 1460 * If we do not forbid deprecated addresses, we 1461 * accept the SYN packet. RFC2462 does not 1462 * suggest dropping SYN in this case. 1463 * If we decipher RFC2462 5.5.4, it says like 1464 * this: 1465 * 1. use of deprecated addr with existing 1466 * communication is okay - "SHOULD continue 1467 * to be used" 1468 * 2. use of it with new communication: 1469 * (2a) "SHOULD NOT be used if alternate 1470 * address with sufficient scope is 1471 * available" 1472 * (2b) nothing mentioned otherwise. 1473 * Here we fall into (2b) case as we have no 1474 * choice in our source address selection - we 1475 * must obey the peer. 1476 * 1477 * The wording in RFC2462 is confusing, and 1478 * there are multiple description text for 1479 * deprecated address handling - worse, they 1480 * are not exactly the same. I believe 5.5.4 1481 * is the best one, so we follow 5.5.4. 1482 */ 1483 if (af == AF_INET6 && !ip6_use_deprecated) { 1484 struct in6_ifaddr *ia6; 1485 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1486 &ip6->ip6_dst)) && 1487 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1488 tp = NULL; 1489 goto dropwithreset; 1490 } 1491 } 1492 #endif 1493 1494 #ifdef IPSEC 1495 switch (af) { 1496 #ifdef INET 1497 case AF_INET: 1498 if (ipsec4_in_reject_so(m, so)) { 1499 ipsecstat.in_polvio++; 1500 tp = NULL; 1501 goto dropwithreset; 1502 } 1503 break; 1504 #endif 1505 #ifdef INET6 1506 case AF_INET6: 1507 if (ipsec6_in_reject_so(m, so)) { 1508 ipsec6stat.in_polvio++; 1509 tp = NULL; 1510 goto dropwithreset; 1511 } 1512 break; 1513 #endif 1514 } 1515 #endif 1516 1517 /* 1518 * LISTEN socket received a SYN 1519 * from itself? This can't possibly 1520 * be valid; drop the packet. 1521 */ 1522 if (th->th_sport == th->th_dport) { 1523 int i; 1524 1525 switch (af) { 1526 #ifdef INET 1527 case AF_INET: 1528 i = in_hosteq(ip->ip_src, ip->ip_dst); 1529 break; 1530 #endif 1531 #ifdef INET6 1532 case AF_INET6: 1533 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1534 break; 1535 #endif 1536 default: 1537 i = 1; 1538 } 1539 if (i) { 1540 tcpstat.tcps_badsyn++; 1541 goto drop; 1542 } 1543 } 1544 1545 /* 1546 * SYN looks ok; create compressed TCP 1547 * state for it. 1548 */ 1549 if (so->so_qlen <= so->so_qlimit && 1550 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1551 so, m, optp, optlen, &opti)) 1552 m = NULL; 1553 } 1554 goto drop; 1555 } 1556 } 1557 1558 after_listen: 1559 #ifdef DIAGNOSTIC 1560 /* 1561 * Should not happen now that all embryonic connections 1562 * are handled with compressed state. 1563 */ 1564 if (tp->t_state == TCPS_LISTEN) 1565 panic("tcp_input: TCPS_LISTEN"); 1566 #endif 1567 1568 /* 1569 * Segment received on connection. 1570 * Reset idle time and keep-alive timer. 1571 */ 1572 tp->t_rcvtime = tcp_now; 1573 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1574 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1575 1576 /* 1577 * Process options. 1578 */ 1579 #ifdef TCP_SIGNATURE 1580 if (optp || (tp->t_flags & TF_SIGNATURE)) 1581 #else 1582 if (optp) 1583 #endif 1584 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1585 goto drop; 1586 1587 if (TCP_SACK_ENABLED(tp)) { 1588 tcp_del_sackholes(tp, th); 1589 } 1590 1591 if (opti.ts_present && opti.ts_ecr) { 1592 /* 1593 * Calculate the RTT from the returned time stamp and the 1594 * connection's time base. If the time stamp is later than 1595 * the current time, or is extremely old, fall back to non-1323 1596 * RTT calculation. Since ts_ecr is unsigned, we can test both 1597 * at the same time. 1598 */ 1599 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1600 if (ts_rtt > TCP_PAWS_IDLE) 1601 ts_rtt = 0; 1602 } else { 1603 ts_rtt = 0; 1604 } 1605 1606 /* 1607 * Header prediction: check for the two common cases 1608 * of a uni-directional data xfer. If the packet has 1609 * no control flags, is in-sequence, the window didn't 1610 * change and we're not retransmitting, it's a 1611 * candidate. If the length is zero and the ack moved 1612 * forward, we're the sender side of the xfer. Just 1613 * free the data acked & wake any higher level process 1614 * that was blocked waiting for space. If the length 1615 * is non-zero and the ack didn't move, we're the 1616 * receiver side. If we're getting packets in-order 1617 * (the reassembly queue is empty), add the data to 1618 * the socket buffer and note that we need a delayed ack. 1619 */ 1620 if (tp->t_state == TCPS_ESTABLISHED && 1621 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1622 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1623 th->th_seq == tp->rcv_nxt && 1624 tiwin && tiwin == tp->snd_wnd && 1625 tp->snd_nxt == tp->snd_max) { 1626 1627 /* 1628 * If last ACK falls within this segment's sequence numbers, 1629 * record the timestamp. 1630 * NOTE: 1631 * 1) That the test incorporates suggestions from the latest 1632 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1633 * 2) That updating only on newer timestamps interferes with 1634 * our earlier PAWS tests, so this check should be solely 1635 * predicated on the sequence space of this segment. 1636 * 3) That we modify the segment boundary check to be 1637 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1638 * instead of RFC1323's 1639 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1640 * This modified check allows us to overcome RFC1323's 1641 * limitations as described in Stevens TCP/IP Illustrated 1642 * Vol. 2 p.869. In such cases, we can still calculate the 1643 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1644 */ 1645 if (opti.ts_present && 1646 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1647 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1648 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1649 tp->ts_recent_age = tcp_now; 1650 tp->ts_recent = opti.ts_val; 1651 } 1652 1653 if (tlen == 0) { 1654 /* Ack prediction. */ 1655 if (SEQ_GT(th->th_ack, tp->snd_una) && 1656 SEQ_LEQ(th->th_ack, tp->snd_max) && 1657 tp->snd_cwnd >= tp->snd_wnd && 1658 tp->t_partialacks < 0) { 1659 /* 1660 * this is a pure ack for outstanding data. 1661 */ 1662 ++tcpstat.tcps_predack; 1663 if (ts_rtt) 1664 tcp_xmit_timer(tp, ts_rtt); 1665 else if (tp->t_rtttime && 1666 SEQ_GT(th->th_ack, tp->t_rtseq)) 1667 tcp_xmit_timer(tp, 1668 tcp_now - tp->t_rtttime); 1669 acked = th->th_ack - tp->snd_una; 1670 tcpstat.tcps_rcvackpack++; 1671 tcpstat.tcps_rcvackbyte += acked; 1672 ND6_HINT(tp); 1673 1674 if (acked > (tp->t_lastoff - tp->t_inoff)) 1675 tp->t_lastm = NULL; 1676 sbdrop(&so->so_snd, acked); 1677 tp->t_lastoff -= acked; 1678 1679 ICMP_CHECK(tp, th, acked); 1680 1681 tp->snd_una = th->th_ack; 1682 tp->snd_fack = tp->snd_una; 1683 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1684 tp->snd_high = tp->snd_una; 1685 m_freem(m); 1686 1687 /* 1688 * If all outstanding data are acked, stop 1689 * retransmit timer, otherwise restart timer 1690 * using current (possibly backed-off) value. 1691 * If process is waiting for space, 1692 * wakeup/selwakeup/signal. If data 1693 * are ready to send, let tcp_output 1694 * decide between more output or persist. 1695 */ 1696 if (tp->snd_una == tp->snd_max) 1697 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1698 else if (TCP_TIMER_ISARMED(tp, 1699 TCPT_PERSIST) == 0) 1700 TCP_TIMER_ARM(tp, TCPT_REXMT, 1701 tp->t_rxtcur); 1702 1703 sowwakeup(so); 1704 if (so->so_snd.sb_cc) 1705 (void) tcp_output(tp); 1706 if (tcp_saveti) 1707 m_freem(tcp_saveti); 1708 return; 1709 } 1710 } else if (th->th_ack == tp->snd_una && 1711 TAILQ_FIRST(&tp->segq) == NULL && 1712 tlen <= sbspace(&so->so_rcv)) { 1713 /* 1714 * this is a pure, in-sequence data packet 1715 * with nothing on the reassembly queue and 1716 * we have enough buffer space to take it. 1717 */ 1718 ++tcpstat.tcps_preddat; 1719 tp->rcv_nxt += tlen; 1720 tcpstat.tcps_rcvpack++; 1721 tcpstat.tcps_rcvbyte += tlen; 1722 ND6_HINT(tp); 1723 /* 1724 * Drop TCP, IP headers and TCP options then add data 1725 * to socket buffer. 1726 */ 1727 if (so->so_state & SS_CANTRCVMORE) 1728 m_freem(m); 1729 else { 1730 m_adj(m, toff + off); 1731 sbappendstream(&so->so_rcv, m); 1732 } 1733 sorwakeup(so); 1734 TCP_SETUP_ACK(tp, th); 1735 if (tp->t_flags & TF_ACKNOW) 1736 (void) tcp_output(tp); 1737 if (tcp_saveti) 1738 m_freem(tcp_saveti); 1739 return; 1740 } 1741 } 1742 1743 /* 1744 * Compute mbuf offset to TCP data segment. 1745 */ 1746 hdroptlen = toff + off; 1747 1748 /* 1749 * Calculate amount of space in receive window, 1750 * and then do TCP input processing. 1751 * Receive window is amount of space in rcv queue, 1752 * but not less than advertised window. 1753 */ 1754 { int win; 1755 1756 win = sbspace(&so->so_rcv); 1757 if (win < 0) 1758 win = 0; 1759 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1760 } 1761 1762 switch (tp->t_state) { 1763 case TCPS_LISTEN: 1764 /* 1765 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1766 */ 1767 if (m->m_flags & (M_BCAST|M_MCAST)) 1768 goto drop; 1769 switch (af) { 1770 #ifdef INET6 1771 case AF_INET6: 1772 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1773 goto drop; 1774 break; 1775 #endif /* INET6 */ 1776 case AF_INET: 1777 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1778 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1779 goto drop; 1780 break; 1781 } 1782 break; 1783 1784 /* 1785 * If the state is SYN_SENT: 1786 * if seg contains an ACK, but not for our SYN, drop the input. 1787 * if seg contains a RST, then drop the connection. 1788 * if seg does not contain SYN, then drop it. 1789 * Otherwise this is an acceptable SYN segment 1790 * initialize tp->rcv_nxt and tp->irs 1791 * if seg contains ack then advance tp->snd_una 1792 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1793 * arrange for segment to be acked (eventually) 1794 * continue processing rest of data/controls, beginning with URG 1795 */ 1796 case TCPS_SYN_SENT: 1797 if ((tiflags & TH_ACK) && 1798 (SEQ_LEQ(th->th_ack, tp->iss) || 1799 SEQ_GT(th->th_ack, tp->snd_max))) 1800 goto dropwithreset; 1801 if (tiflags & TH_RST) { 1802 if (tiflags & TH_ACK) 1803 tp = tcp_drop(tp, ECONNREFUSED); 1804 goto drop; 1805 } 1806 if ((tiflags & TH_SYN) == 0) 1807 goto drop; 1808 if (tiflags & TH_ACK) { 1809 tp->snd_una = th->th_ack; 1810 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1811 tp->snd_nxt = tp->snd_una; 1812 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1813 tp->snd_high = tp->snd_una; 1814 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1815 } 1816 tp->irs = th->th_seq; 1817 tcp_rcvseqinit(tp); 1818 tp->t_flags |= TF_ACKNOW; 1819 tcp_mss_from_peer(tp, opti.maxseg); 1820 1821 /* 1822 * Initialize the initial congestion window. If we 1823 * had to retransmit the SYN, we must initialize cwnd 1824 * to 1 segment (i.e. the Loss Window). 1825 */ 1826 if (tp->t_flags & TF_SYN_REXMT) 1827 tp->snd_cwnd = tp->t_peermss; 1828 else { 1829 int ss = tcp_init_win; 1830 #ifdef INET 1831 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1832 ss = tcp_init_win_local; 1833 #endif 1834 #ifdef INET6 1835 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1836 ss = tcp_init_win_local; 1837 #endif 1838 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1839 } 1840 1841 tcp_rmx_rtt(tp); 1842 if (tiflags & TH_ACK) { 1843 tcpstat.tcps_connects++; 1844 soisconnected(so); 1845 tcp_established(tp); 1846 /* Do window scaling on this connection? */ 1847 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1848 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1849 tp->snd_scale = tp->requested_s_scale; 1850 tp->rcv_scale = tp->request_r_scale; 1851 } 1852 TCP_REASS_LOCK(tp); 1853 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1854 TCP_REASS_UNLOCK(tp); 1855 /* 1856 * if we didn't have to retransmit the SYN, 1857 * use its rtt as our initial srtt & rtt var. 1858 */ 1859 if (tp->t_rtttime) 1860 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1861 } else 1862 tp->t_state = TCPS_SYN_RECEIVED; 1863 1864 /* 1865 * Advance th->th_seq to correspond to first data byte. 1866 * If data, trim to stay within window, 1867 * dropping FIN if necessary. 1868 */ 1869 th->th_seq++; 1870 if (tlen > tp->rcv_wnd) { 1871 todrop = tlen - tp->rcv_wnd; 1872 m_adj(m, -todrop); 1873 tlen = tp->rcv_wnd; 1874 tiflags &= ~TH_FIN; 1875 tcpstat.tcps_rcvpackafterwin++; 1876 tcpstat.tcps_rcvbyteafterwin += todrop; 1877 } 1878 tp->snd_wl1 = th->th_seq - 1; 1879 tp->rcv_up = th->th_seq; 1880 goto step6; 1881 1882 /* 1883 * If the state is SYN_RECEIVED: 1884 * If seg contains an ACK, but not for our SYN, drop the input 1885 * and generate an RST. See page 36, rfc793 1886 */ 1887 case TCPS_SYN_RECEIVED: 1888 if ((tiflags & TH_ACK) && 1889 (SEQ_LEQ(th->th_ack, tp->iss) || 1890 SEQ_GT(th->th_ack, tp->snd_max))) 1891 goto dropwithreset; 1892 break; 1893 } 1894 1895 /* 1896 * States other than LISTEN or SYN_SENT. 1897 * First check timestamp, if present. 1898 * Then check that at least some bytes of segment are within 1899 * receive window. If segment begins before rcv_nxt, 1900 * drop leading data (and SYN); if nothing left, just ack. 1901 * 1902 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1903 * and it's less than ts_recent, drop it. 1904 */ 1905 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1906 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1907 1908 /* Check to see if ts_recent is over 24 days old. */ 1909 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1910 /* 1911 * Invalidate ts_recent. If this segment updates 1912 * ts_recent, the age will be reset later and ts_recent 1913 * will get a valid value. If it does not, setting 1914 * ts_recent to zero will at least satisfy the 1915 * requirement that zero be placed in the timestamp 1916 * echo reply when ts_recent isn't valid. The 1917 * age isn't reset until we get a valid ts_recent 1918 * because we don't want out-of-order segments to be 1919 * dropped when ts_recent is old. 1920 */ 1921 tp->ts_recent = 0; 1922 } else { 1923 tcpstat.tcps_rcvduppack++; 1924 tcpstat.tcps_rcvdupbyte += tlen; 1925 tcpstat.tcps_pawsdrop++; 1926 tcp_new_dsack(tp, th->th_seq, tlen); 1927 goto dropafterack; 1928 } 1929 } 1930 1931 todrop = tp->rcv_nxt - th->th_seq; 1932 dupseg = FALSE; 1933 if (todrop > 0) { 1934 if (tiflags & TH_SYN) { 1935 tiflags &= ~TH_SYN; 1936 th->th_seq++; 1937 if (th->th_urp > 1) 1938 th->th_urp--; 1939 else { 1940 tiflags &= ~TH_URG; 1941 th->th_urp = 0; 1942 } 1943 todrop--; 1944 } 1945 if (todrop > tlen || 1946 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1947 /* 1948 * Any valid FIN or RST must be to the left of the 1949 * window. At this point the FIN or RST must be a 1950 * duplicate or out of sequence; drop it. 1951 */ 1952 if (tiflags & TH_RST) 1953 goto drop; 1954 tiflags &= ~(TH_FIN|TH_RST); 1955 /* 1956 * Send an ACK to resynchronize and drop any data. 1957 * But keep on processing for RST or ACK. 1958 */ 1959 tp->t_flags |= TF_ACKNOW; 1960 todrop = tlen; 1961 dupseg = TRUE; 1962 tcpstat.tcps_rcvdupbyte += todrop; 1963 tcpstat.tcps_rcvduppack++; 1964 } else if ((tiflags & TH_RST) && 1965 th->th_seq != tp->last_ack_sent) { 1966 /* 1967 * Test for reset before adjusting the sequence 1968 * number for overlapping data. 1969 */ 1970 goto dropafterack_ratelim; 1971 } else { 1972 tcpstat.tcps_rcvpartduppack++; 1973 tcpstat.tcps_rcvpartdupbyte += todrop; 1974 } 1975 tcp_new_dsack(tp, th->th_seq, todrop); 1976 hdroptlen += todrop; /*drop from head afterwards*/ 1977 th->th_seq += todrop; 1978 tlen -= todrop; 1979 if (th->th_urp > todrop) 1980 th->th_urp -= todrop; 1981 else { 1982 tiflags &= ~TH_URG; 1983 th->th_urp = 0; 1984 } 1985 } 1986 1987 /* 1988 * If new data are received on a connection after the 1989 * user processes are gone, then RST the other end. 1990 */ 1991 if ((so->so_state & SS_NOFDREF) && 1992 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1993 tp = tcp_close(tp); 1994 tcpstat.tcps_rcvafterclose++; 1995 goto dropwithreset; 1996 } 1997 1998 /* 1999 * If segment ends after window, drop trailing data 2000 * (and PUSH and FIN); if nothing left, just ACK. 2001 */ 2002 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2003 if (todrop > 0) { 2004 tcpstat.tcps_rcvpackafterwin++; 2005 if (todrop >= tlen) { 2006 /* 2007 * The segment actually starts after the window. 2008 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2009 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2010 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2011 */ 2012 tcpstat.tcps_rcvbyteafterwin += tlen; 2013 /* 2014 * If a new connection request is received 2015 * while in TIME_WAIT, drop the old connection 2016 * and start over if the sequence numbers 2017 * are above the previous ones. 2018 * 2019 * NOTE: We will checksum the packet again, and 2020 * so we need to put the header fields back into 2021 * network order! 2022 * XXX This kind of sucks, but we don't expect 2023 * XXX this to happen very often, so maybe it 2024 * XXX doesn't matter so much. 2025 */ 2026 if (tiflags & TH_SYN && 2027 tp->t_state == TCPS_TIME_WAIT && 2028 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2029 iss = tcp_new_iss(tp, tp->snd_nxt); 2030 tp = tcp_close(tp); 2031 TCP_FIELDS_TO_NET(th); 2032 goto findpcb; 2033 } 2034 /* 2035 * If window is closed can only take segments at 2036 * window edge, and have to drop data and PUSH from 2037 * incoming segments. Continue processing, but 2038 * remember to ack. Otherwise, drop segment 2039 * and (if not RST) ack. 2040 */ 2041 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2042 tp->t_flags |= TF_ACKNOW; 2043 tcpstat.tcps_rcvwinprobe++; 2044 } else 2045 goto dropafterack; 2046 } else 2047 tcpstat.tcps_rcvbyteafterwin += todrop; 2048 m_adj(m, -todrop); 2049 tlen -= todrop; 2050 tiflags &= ~(TH_PUSH|TH_FIN); 2051 } 2052 2053 /* 2054 * If last ACK falls within this segment's sequence numbers, 2055 * and the timestamp is newer, record it. 2056 */ 2057 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 2058 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2059 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 2060 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2061 tp->ts_recent_age = tcp_now; 2062 tp->ts_recent = opti.ts_val; 2063 } 2064 2065 /* 2066 * If the RST bit is set examine the state: 2067 * SYN_RECEIVED STATE: 2068 * If passive open, return to LISTEN state. 2069 * If active open, inform user that connection was refused. 2070 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2071 * Inform user that connection was reset, and close tcb. 2072 * CLOSING, LAST_ACK, TIME_WAIT STATES 2073 * Close the tcb. 2074 */ 2075 if (tiflags & TH_RST) { 2076 if (th->th_seq != tp->last_ack_sent) 2077 goto dropafterack_ratelim; 2078 2079 switch (tp->t_state) { 2080 case TCPS_SYN_RECEIVED: 2081 so->so_error = ECONNREFUSED; 2082 goto close; 2083 2084 case TCPS_ESTABLISHED: 2085 case TCPS_FIN_WAIT_1: 2086 case TCPS_FIN_WAIT_2: 2087 case TCPS_CLOSE_WAIT: 2088 so->so_error = ECONNRESET; 2089 close: 2090 tp->t_state = TCPS_CLOSED; 2091 tcpstat.tcps_drops++; 2092 tp = tcp_close(tp); 2093 goto drop; 2094 2095 case TCPS_CLOSING: 2096 case TCPS_LAST_ACK: 2097 case TCPS_TIME_WAIT: 2098 tp = tcp_close(tp); 2099 goto drop; 2100 } 2101 } 2102 2103 /* 2104 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2105 * we must be in a synchronized state. RFC791 states (under RST 2106 * generation) that any unacceptable segment (an out-of-order SYN 2107 * qualifies) received in a synchronized state must elicit only an 2108 * empty acknowledgment segment ... and the connection remains in 2109 * the same state. 2110 */ 2111 if (tiflags & TH_SYN) { 2112 if (tp->rcv_nxt == th->th_seq) { 2113 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2114 TH_ACK); 2115 if (tcp_saveti) 2116 m_freem(tcp_saveti); 2117 return; 2118 } 2119 2120 goto dropafterack_ratelim; 2121 } 2122 2123 /* 2124 * If the ACK bit is off we drop the segment and return. 2125 */ 2126 if ((tiflags & TH_ACK) == 0) { 2127 if (tp->t_flags & TF_ACKNOW) 2128 goto dropafterack; 2129 else 2130 goto drop; 2131 } 2132 2133 /* 2134 * Ack processing. 2135 */ 2136 switch (tp->t_state) { 2137 2138 /* 2139 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2140 * ESTABLISHED state and continue processing, otherwise 2141 * send an RST. 2142 */ 2143 case TCPS_SYN_RECEIVED: 2144 if (SEQ_GT(tp->snd_una, th->th_ack) || 2145 SEQ_GT(th->th_ack, tp->snd_max)) 2146 goto dropwithreset; 2147 tcpstat.tcps_connects++; 2148 soisconnected(so); 2149 tcp_established(tp); 2150 /* Do window scaling? */ 2151 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2152 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2153 tp->snd_scale = tp->requested_s_scale; 2154 tp->rcv_scale = tp->request_r_scale; 2155 } 2156 TCP_REASS_LOCK(tp); 2157 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2158 TCP_REASS_UNLOCK(tp); 2159 tp->snd_wl1 = th->th_seq - 1; 2160 /* fall into ... */ 2161 2162 /* 2163 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2164 * ACKs. If the ack is in the range 2165 * tp->snd_una < th->th_ack <= tp->snd_max 2166 * then advance tp->snd_una to th->th_ack and drop 2167 * data from the retransmission queue. If this ACK reflects 2168 * more up to date window information we update our window information. 2169 */ 2170 case TCPS_ESTABLISHED: 2171 case TCPS_FIN_WAIT_1: 2172 case TCPS_FIN_WAIT_2: 2173 case TCPS_CLOSE_WAIT: 2174 case TCPS_CLOSING: 2175 case TCPS_LAST_ACK: 2176 case TCPS_TIME_WAIT: 2177 2178 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2179 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2180 tcpstat.tcps_rcvdupack++; 2181 /* 2182 * If we have outstanding data (other than 2183 * a window probe), this is a completely 2184 * duplicate ack (ie, window info didn't 2185 * change), the ack is the biggest we've 2186 * seen and we've seen exactly our rexmt 2187 * threshhold of them, assume a packet 2188 * has been dropped and retransmit it. 2189 * Kludge snd_nxt & the congestion 2190 * window so we send only this one 2191 * packet. 2192 * 2193 * We know we're losing at the current 2194 * window size so do congestion avoidance 2195 * (set ssthresh to half the current window 2196 * and pull our congestion window back to 2197 * the new ssthresh). 2198 * 2199 * Dup acks mean that packets have left the 2200 * network (they're now cached at the receiver) 2201 * so bump cwnd by the amount in the receiver 2202 * to keep a constant cwnd packets in the 2203 * network. 2204 * 2205 * If we are using TCP/SACK, then enter 2206 * Fast Recovery if the receiver SACKs 2207 * data that is tcprexmtthresh * MSS 2208 * bytes past the last ACKed segment, 2209 * irrespective of the number of DupAcks. 2210 */ 2211 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2212 th->th_ack != tp->snd_una) 2213 tp->t_dupacks = 0; 2214 else if (tp->t_partialacks < 0 && 2215 (++tp->t_dupacks == tcprexmtthresh || 2216 TCP_FACK_FASTRECOV(tp))) { 2217 tcp_seq onxt; 2218 u_int win; 2219 2220 if (tcp_do_newreno && 2221 SEQ_LT(th->th_ack, tp->snd_high)) { 2222 /* 2223 * False fast retransmit after 2224 * timeout. Do not enter fast 2225 * recovery. 2226 */ 2227 tp->t_dupacks = 0; 2228 break; 2229 } 2230 2231 onxt = tp->snd_nxt; 2232 win = min(tp->snd_wnd, tp->snd_cwnd) / 2233 2 / tp->t_segsz; 2234 if (win < 2) 2235 win = 2; 2236 tp->snd_ssthresh = win * tp->t_segsz; 2237 tp->snd_recover = tp->snd_max; 2238 tp->t_partialacks = 0; 2239 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2240 tp->t_rtttime = 0; 2241 if (TCP_SACK_ENABLED(tp)) { 2242 tp->t_dupacks = tcprexmtthresh; 2243 tp->sack_newdata = tp->snd_nxt; 2244 tp->snd_cwnd = tp->t_segsz; 2245 (void) tcp_output(tp); 2246 goto drop; 2247 } 2248 tp->snd_nxt = th->th_ack; 2249 tp->snd_cwnd = tp->t_segsz; 2250 (void) tcp_output(tp); 2251 tp->snd_cwnd = tp->snd_ssthresh + 2252 tp->t_segsz * tp->t_dupacks; 2253 if (SEQ_GT(onxt, tp->snd_nxt)) 2254 tp->snd_nxt = onxt; 2255 goto drop; 2256 } else if (tp->t_dupacks > tcprexmtthresh) { 2257 tp->snd_cwnd += tp->t_segsz; 2258 (void) tcp_output(tp); 2259 goto drop; 2260 } 2261 } else { 2262 /* 2263 * If the ack appears to be very old, only 2264 * allow data that is in-sequence. This 2265 * makes it somewhat more difficult to insert 2266 * forged data by guessing sequence numbers. 2267 * Sent an ack to try to update the send 2268 * sequence number on the other side. 2269 */ 2270 if (tlen && th->th_seq != tp->rcv_nxt && 2271 SEQ_LT(th->th_ack, 2272 tp->snd_una - tp->max_sndwnd)) 2273 goto dropafterack; 2274 } 2275 break; 2276 } 2277 /* 2278 * If the congestion window was inflated to account 2279 * for the other side's cached packets, retract it. 2280 */ 2281 if (TCP_SACK_ENABLED(tp)) 2282 tcp_sack_newack(tp, th); 2283 else if (tcp_do_newreno) 2284 tcp_newreno_newack(tp, th); 2285 else 2286 tcp_reno_newack(tp, th); 2287 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2288 tcpstat.tcps_rcvacktoomuch++; 2289 goto dropafterack; 2290 } 2291 acked = th->th_ack - tp->snd_una; 2292 tcpstat.tcps_rcvackpack++; 2293 tcpstat.tcps_rcvackbyte += acked; 2294 2295 /* 2296 * If we have a timestamp reply, update smoothed 2297 * round trip time. If no timestamp is present but 2298 * transmit timer is running and timed sequence 2299 * number was acked, update smoothed round trip time. 2300 * Since we now have an rtt measurement, cancel the 2301 * timer backoff (cf., Phil Karn's retransmit alg.). 2302 * Recompute the initial retransmit timer. 2303 */ 2304 if (ts_rtt) 2305 tcp_xmit_timer(tp, ts_rtt); 2306 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2307 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2308 2309 /* 2310 * If all outstanding data is acked, stop retransmit 2311 * timer and remember to restart (more output or persist). 2312 * If there is more data to be acked, restart retransmit 2313 * timer, using current (possibly backed-off) value. 2314 */ 2315 if (th->th_ack == tp->snd_max) { 2316 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2317 needoutput = 1; 2318 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2319 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2320 /* 2321 * When new data is acked, open the congestion window. 2322 * If the window gives us less than ssthresh packets 2323 * in flight, open exponentially (segsz per packet). 2324 * Otherwise open linearly: segsz per window 2325 * (segsz^2 / cwnd per packet). 2326 * 2327 * If we are still in fast recovery (meaning we are using 2328 * NewReno and we have only received partial acks), do not 2329 * inflate the window yet. 2330 */ 2331 if (tp->t_partialacks < 0) { 2332 u_int cw = tp->snd_cwnd; 2333 u_int incr = tp->t_segsz; 2334 2335 if (cw >= tp->snd_ssthresh) 2336 incr = incr * incr / cw; 2337 tp->snd_cwnd = min(cw + incr, 2338 TCP_MAXWIN << tp->snd_scale); 2339 } 2340 ND6_HINT(tp); 2341 if (acked > so->so_snd.sb_cc) { 2342 tp->snd_wnd -= so->so_snd.sb_cc; 2343 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2344 ourfinisacked = 1; 2345 } else { 2346 if (acked > (tp->t_lastoff - tp->t_inoff)) 2347 tp->t_lastm = NULL; 2348 sbdrop(&so->so_snd, acked); 2349 tp->t_lastoff -= acked; 2350 tp->snd_wnd -= acked; 2351 ourfinisacked = 0; 2352 } 2353 sowwakeup(so); 2354 2355 ICMP_CHECK(tp, th, acked); 2356 2357 tp->snd_una = th->th_ack; 2358 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2359 tp->snd_fack = tp->snd_una; 2360 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2361 tp->snd_nxt = tp->snd_una; 2362 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2363 tp->snd_high = tp->snd_una; 2364 2365 switch (tp->t_state) { 2366 2367 /* 2368 * In FIN_WAIT_1 STATE in addition to the processing 2369 * for the ESTABLISHED state if our FIN is now acknowledged 2370 * then enter FIN_WAIT_2. 2371 */ 2372 case TCPS_FIN_WAIT_1: 2373 if (ourfinisacked) { 2374 /* 2375 * If we can't receive any more 2376 * data, then closing user can proceed. 2377 * Starting the timer is contrary to the 2378 * specification, but if we don't get a FIN 2379 * we'll hang forever. 2380 */ 2381 if (so->so_state & SS_CANTRCVMORE) { 2382 soisdisconnected(so); 2383 if (tcp_maxidle > 0) 2384 TCP_TIMER_ARM(tp, TCPT_2MSL, 2385 tcp_maxidle); 2386 } 2387 tp->t_state = TCPS_FIN_WAIT_2; 2388 } 2389 break; 2390 2391 /* 2392 * In CLOSING STATE in addition to the processing for 2393 * the ESTABLISHED state if the ACK acknowledges our FIN 2394 * then enter the TIME-WAIT state, otherwise ignore 2395 * the segment. 2396 */ 2397 case TCPS_CLOSING: 2398 if (ourfinisacked) { 2399 tp->t_state = TCPS_TIME_WAIT; 2400 tcp_canceltimers(tp); 2401 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2402 soisdisconnected(so); 2403 } 2404 break; 2405 2406 /* 2407 * In LAST_ACK, we may still be waiting for data to drain 2408 * and/or to be acked, as well as for the ack of our FIN. 2409 * If our FIN is now acknowledged, delete the TCB, 2410 * enter the closed state and return. 2411 */ 2412 case TCPS_LAST_ACK: 2413 if (ourfinisacked) { 2414 tp = tcp_close(tp); 2415 goto drop; 2416 } 2417 break; 2418 2419 /* 2420 * In TIME_WAIT state the only thing that should arrive 2421 * is a retransmission of the remote FIN. Acknowledge 2422 * it and restart the finack timer. 2423 */ 2424 case TCPS_TIME_WAIT: 2425 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2426 goto dropafterack; 2427 } 2428 } 2429 2430 step6: 2431 /* 2432 * Update window information. 2433 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2434 */ 2435 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2436 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2437 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2438 /* keep track of pure window updates */ 2439 if (tlen == 0 && 2440 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2441 tcpstat.tcps_rcvwinupd++; 2442 tp->snd_wnd = tiwin; 2443 tp->snd_wl1 = th->th_seq; 2444 tp->snd_wl2 = th->th_ack; 2445 if (tp->snd_wnd > tp->max_sndwnd) 2446 tp->max_sndwnd = tp->snd_wnd; 2447 needoutput = 1; 2448 } 2449 2450 /* 2451 * Process segments with URG. 2452 */ 2453 if ((tiflags & TH_URG) && th->th_urp && 2454 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2455 /* 2456 * This is a kludge, but if we receive and accept 2457 * random urgent pointers, we'll crash in 2458 * soreceive. It's hard to imagine someone 2459 * actually wanting to send this much urgent data. 2460 */ 2461 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2462 th->th_urp = 0; /* XXX */ 2463 tiflags &= ~TH_URG; /* XXX */ 2464 goto dodata; /* XXX */ 2465 } 2466 /* 2467 * If this segment advances the known urgent pointer, 2468 * then mark the data stream. This should not happen 2469 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2470 * a FIN has been received from the remote side. 2471 * In these states we ignore the URG. 2472 * 2473 * According to RFC961 (Assigned Protocols), 2474 * the urgent pointer points to the last octet 2475 * of urgent data. We continue, however, 2476 * to consider it to indicate the first octet 2477 * of data past the urgent section as the original 2478 * spec states (in one of two places). 2479 */ 2480 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2481 tp->rcv_up = th->th_seq + th->th_urp; 2482 so->so_oobmark = so->so_rcv.sb_cc + 2483 (tp->rcv_up - tp->rcv_nxt) - 1; 2484 if (so->so_oobmark == 0) 2485 so->so_state |= SS_RCVATMARK; 2486 sohasoutofband(so); 2487 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2488 } 2489 /* 2490 * Remove out of band data so doesn't get presented to user. 2491 * This can happen independent of advancing the URG pointer, 2492 * but if two URG's are pending at once, some out-of-band 2493 * data may creep in... ick. 2494 */ 2495 if (th->th_urp <= (u_int16_t) tlen 2496 #ifdef SO_OOBINLINE 2497 && (so->so_options & SO_OOBINLINE) == 0 2498 #endif 2499 ) 2500 tcp_pulloutofband(so, th, m, hdroptlen); 2501 } else 2502 /* 2503 * If no out of band data is expected, 2504 * pull receive urgent pointer along 2505 * with the receive window. 2506 */ 2507 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2508 tp->rcv_up = tp->rcv_nxt; 2509 dodata: /* XXX */ 2510 2511 /* 2512 * Process the segment text, merging it into the TCP sequencing queue, 2513 * and arranging for acknowledgement of receipt if necessary. 2514 * This process logically involves adjusting tp->rcv_wnd as data 2515 * is presented to the user (this happens in tcp_usrreq.c, 2516 * case PRU_RCVD). If a FIN has already been received on this 2517 * connection then we just ignore the text. 2518 */ 2519 if ((tlen || (tiflags & TH_FIN)) && 2520 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2521 /* 2522 * Insert segment ti into reassembly queue of tcp with 2523 * control block tp. Return TH_FIN if reassembly now includes 2524 * a segment with FIN. The macro form does the common case 2525 * inline (segment is the next to be received on an 2526 * established connection, and the queue is empty), 2527 * avoiding linkage into and removal from the queue and 2528 * repetition of various conversions. 2529 * Set DELACK for segments received in order, but ack 2530 * immediately when segments are out of order 2531 * (so fast retransmit can work). 2532 */ 2533 /* NOTE: this was TCP_REASS() macro, but used only once */ 2534 TCP_REASS_LOCK(tp); 2535 if (th->th_seq == tp->rcv_nxt && 2536 TAILQ_FIRST(&tp->segq) == NULL && 2537 tp->t_state == TCPS_ESTABLISHED) { 2538 TCP_SETUP_ACK(tp, th); 2539 tp->rcv_nxt += tlen; 2540 tiflags = th->th_flags & TH_FIN; 2541 tcpstat.tcps_rcvpack++; 2542 tcpstat.tcps_rcvbyte += tlen; 2543 ND6_HINT(tp); 2544 if (so->so_state & SS_CANTRCVMORE) 2545 m_freem(m); 2546 else { 2547 m_adj(m, hdroptlen); 2548 sbappendstream(&(so)->so_rcv, m); 2549 } 2550 sorwakeup(so); 2551 } else { 2552 m_adj(m, hdroptlen); 2553 tiflags = tcp_reass(tp, th, m, &tlen); 2554 tp->t_flags |= TF_ACKNOW; 2555 } 2556 TCP_REASS_UNLOCK(tp); 2557 2558 /* 2559 * Note the amount of data that peer has sent into 2560 * our window, in order to estimate the sender's 2561 * buffer size. 2562 */ 2563 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2564 } else { 2565 m_freem(m); 2566 m = NULL; 2567 tiflags &= ~TH_FIN; 2568 } 2569 2570 /* 2571 * If FIN is received ACK the FIN and let the user know 2572 * that the connection is closing. Ignore a FIN received before 2573 * the connection is fully established. 2574 */ 2575 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2576 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2577 socantrcvmore(so); 2578 tp->t_flags |= TF_ACKNOW; 2579 tp->rcv_nxt++; 2580 } 2581 switch (tp->t_state) { 2582 2583 /* 2584 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2585 */ 2586 case TCPS_ESTABLISHED: 2587 tp->t_state = TCPS_CLOSE_WAIT; 2588 break; 2589 2590 /* 2591 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2592 * enter the CLOSING state. 2593 */ 2594 case TCPS_FIN_WAIT_1: 2595 tp->t_state = TCPS_CLOSING; 2596 break; 2597 2598 /* 2599 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2600 * starting the time-wait timer, turning off the other 2601 * standard timers. 2602 */ 2603 case TCPS_FIN_WAIT_2: 2604 tp->t_state = TCPS_TIME_WAIT; 2605 tcp_canceltimers(tp); 2606 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2607 soisdisconnected(so); 2608 break; 2609 2610 /* 2611 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2612 */ 2613 case TCPS_TIME_WAIT: 2614 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2615 break; 2616 } 2617 } 2618 #ifdef TCP_DEBUG 2619 if (so->so_options & SO_DEBUG) 2620 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2621 #endif 2622 2623 /* 2624 * Return any desired output. 2625 */ 2626 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2627 (void) tcp_output(tp); 2628 } 2629 if (tcp_saveti) 2630 m_freem(tcp_saveti); 2631 return; 2632 2633 badsyn: 2634 /* 2635 * Received a bad SYN. Increment counters and dropwithreset. 2636 */ 2637 tcpstat.tcps_badsyn++; 2638 tp = NULL; 2639 goto dropwithreset; 2640 2641 dropafterack: 2642 /* 2643 * Generate an ACK dropping incoming segment if it occupies 2644 * sequence space, where the ACK reflects our state. 2645 */ 2646 if (tiflags & TH_RST) 2647 goto drop; 2648 goto dropafterack2; 2649 2650 dropafterack_ratelim: 2651 /* 2652 * We may want to rate-limit ACKs against SYN/RST attack. 2653 */ 2654 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2655 tcp_ackdrop_ppslim) == 0) { 2656 /* XXX stat */ 2657 goto drop; 2658 } 2659 /* ...fall into dropafterack2... */ 2660 2661 dropafterack2: 2662 m_freem(m); 2663 tp->t_flags |= TF_ACKNOW; 2664 (void) tcp_output(tp); 2665 if (tcp_saveti) 2666 m_freem(tcp_saveti); 2667 return; 2668 2669 dropwithreset_ratelim: 2670 /* 2671 * We may want to rate-limit RSTs in certain situations, 2672 * particularly if we are sending an RST in response to 2673 * an attempt to connect to or otherwise communicate with 2674 * a port for which we have no socket. 2675 */ 2676 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2677 tcp_rst_ppslim) == 0) { 2678 /* XXX stat */ 2679 goto drop; 2680 } 2681 /* ...fall into dropwithreset... */ 2682 2683 dropwithreset: 2684 /* 2685 * Generate a RST, dropping incoming segment. 2686 * Make ACK acceptable to originator of segment. 2687 */ 2688 if (tiflags & TH_RST) 2689 goto drop; 2690 2691 switch (af) { 2692 #ifdef INET6 2693 case AF_INET6: 2694 /* For following calls to tcp_respond */ 2695 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2696 goto drop; 2697 break; 2698 #endif /* INET6 */ 2699 case AF_INET: 2700 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2701 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2702 goto drop; 2703 } 2704 2705 if (tiflags & TH_ACK) 2706 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2707 else { 2708 if (tiflags & TH_SYN) 2709 tlen++; 2710 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2711 TH_RST|TH_ACK); 2712 } 2713 if (tcp_saveti) 2714 m_freem(tcp_saveti); 2715 return; 2716 2717 badcsum: 2718 drop: 2719 /* 2720 * Drop space held by incoming segment and return. 2721 */ 2722 if (tp) { 2723 if (tp->t_inpcb) 2724 so = tp->t_inpcb->inp_socket; 2725 #ifdef INET6 2726 else if (tp->t_in6pcb) 2727 so = tp->t_in6pcb->in6p_socket; 2728 #endif 2729 else 2730 so = NULL; 2731 #ifdef TCP_DEBUG 2732 if (so && (so->so_options & SO_DEBUG) != 0) 2733 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2734 #endif 2735 } 2736 if (tcp_saveti) 2737 m_freem(tcp_saveti); 2738 m_freem(m); 2739 return; 2740 } 2741 2742 #ifdef TCP_SIGNATURE 2743 int 2744 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2745 { 2746 2747 MD5Update(fstate, (u_char *)data, len); 2748 return (0); 2749 } 2750 2751 struct secasvar * 2752 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2753 { 2754 struct secasvar *sav; 2755 #ifdef FAST_IPSEC 2756 union sockaddr_union dst; 2757 #endif 2758 struct ip *ip; 2759 struct ip6_hdr *ip6; 2760 2761 ip = mtod(m, struct ip *); 2762 switch (ip->ip_v) { 2763 case 4: 2764 ip = mtod(m, struct ip *); 2765 ip6 = NULL; 2766 break; 2767 case 6: 2768 ip = NULL; 2769 ip6 = mtod(m, struct ip6_hdr *); 2770 break; 2771 default: 2772 return (NULL); 2773 } 2774 2775 #ifdef FAST_IPSEC 2776 /* Extract the destination from the IP header in the mbuf. */ 2777 bzero(&dst, sizeof(union sockaddr_union)); 2778 dst.sa.sa_len = sizeof(struct sockaddr_in); 2779 dst.sa.sa_family = AF_INET; 2780 dst.sin.sin_addr = ip->ip_dst; 2781 2782 /* 2783 * Look up an SADB entry which matches the address of the peer. 2784 */ 2785 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2786 #else 2787 if (ip) 2788 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2789 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2790 htonl(TCP_SIG_SPI), 0, 0); 2791 else 2792 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2793 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2794 htonl(TCP_SIG_SPI), 0, 0); 2795 #endif 2796 2797 return (sav); /* freesav must be performed by caller */ 2798 } 2799 2800 int 2801 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2802 struct secasvar *sav, char *sig) 2803 { 2804 MD5_CTX ctx; 2805 struct ip *ip; 2806 struct ipovly *ipovly; 2807 struct ip6_hdr *ip6; 2808 struct ippseudo ippseudo; 2809 struct ip6_hdr_pseudo ip6pseudo; 2810 struct tcphdr th0; 2811 int l, tcphdrlen; 2812 2813 if (sav == NULL) 2814 return (-1); 2815 2816 tcphdrlen = th->th_off * 4; 2817 2818 switch (mtod(m, struct ip *)->ip_v) { 2819 case 4: 2820 ip = mtod(m, struct ip *); 2821 ip6 = NULL; 2822 break; 2823 case 6: 2824 ip = NULL; 2825 ip6 = mtod(m, struct ip6_hdr *); 2826 break; 2827 default: 2828 return (-1); 2829 } 2830 2831 MD5Init(&ctx); 2832 2833 if (ip) { 2834 memset(&ippseudo, 0, sizeof(ippseudo)); 2835 ipovly = (struct ipovly *)ip; 2836 ippseudo.ippseudo_src = ipovly->ih_src; 2837 ippseudo.ippseudo_dst = ipovly->ih_dst; 2838 ippseudo.ippseudo_pad = 0; 2839 ippseudo.ippseudo_p = IPPROTO_TCP; 2840 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2841 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2842 } else { 2843 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2844 ip6pseudo.ip6ph_src = ip6->ip6_src; 2845 in6_clearscope(&ip6pseudo.ip6ph_src); 2846 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2847 in6_clearscope(&ip6pseudo.ip6ph_dst); 2848 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2849 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2850 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2851 } 2852 2853 th0 = *th; 2854 th0.th_sum = 0; 2855 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2856 2857 l = m->m_pkthdr.len - thoff - tcphdrlen; 2858 if (l > 0) 2859 m_apply(m, thoff + tcphdrlen, 2860 m->m_pkthdr.len - thoff - tcphdrlen, 2861 tcp_signature_apply, &ctx); 2862 2863 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2864 MD5Final(sig, &ctx); 2865 2866 return (0); 2867 } 2868 #endif 2869 2870 int 2871 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2872 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2873 { 2874 u_int16_t mss; 2875 int opt, optlen = 0; 2876 #ifdef TCP_SIGNATURE 2877 caddr_t sigp = NULL; 2878 char sigbuf[TCP_SIGLEN]; 2879 struct secasvar *sav = NULL; 2880 #endif 2881 2882 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2883 opt = cp[0]; 2884 if (opt == TCPOPT_EOL) 2885 break; 2886 if (opt == TCPOPT_NOP) 2887 optlen = 1; 2888 else { 2889 if (cnt < 2) 2890 break; 2891 optlen = cp[1]; 2892 if (optlen < 2 || optlen > cnt) 2893 break; 2894 } 2895 switch (opt) { 2896 2897 default: 2898 continue; 2899 2900 case TCPOPT_MAXSEG: 2901 if (optlen != TCPOLEN_MAXSEG) 2902 continue; 2903 if (!(th->th_flags & TH_SYN)) 2904 continue; 2905 if (TCPS_HAVERCVDSYN(tp->t_state)) 2906 continue; 2907 bcopy(cp + 2, &mss, sizeof(mss)); 2908 oi->maxseg = ntohs(mss); 2909 break; 2910 2911 case TCPOPT_WINDOW: 2912 if (optlen != TCPOLEN_WINDOW) 2913 continue; 2914 if (!(th->th_flags & TH_SYN)) 2915 continue; 2916 if (TCPS_HAVERCVDSYN(tp->t_state)) 2917 continue; 2918 tp->t_flags |= TF_RCVD_SCALE; 2919 tp->requested_s_scale = cp[2]; 2920 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2921 #if 0 /*XXX*/ 2922 char *p; 2923 2924 if (ip) 2925 p = ntohl(ip->ip_src); 2926 #ifdef INET6 2927 else if (ip6) 2928 p = ip6_sprintf(&ip6->ip6_src); 2929 #endif 2930 else 2931 p = "(unknown)"; 2932 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2933 "assuming %d\n", 2934 tp->requested_s_scale, p, 2935 TCP_MAX_WINSHIFT); 2936 #else 2937 log(LOG_ERR, "TCP: invalid wscale %d, " 2938 "assuming %d\n", 2939 tp->requested_s_scale, 2940 TCP_MAX_WINSHIFT); 2941 #endif 2942 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2943 } 2944 break; 2945 2946 case TCPOPT_TIMESTAMP: 2947 if (optlen != TCPOLEN_TIMESTAMP) 2948 continue; 2949 oi->ts_present = 1; 2950 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2951 NTOHL(oi->ts_val); 2952 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2953 NTOHL(oi->ts_ecr); 2954 2955 if (!(th->th_flags & TH_SYN)) 2956 continue; 2957 if (TCPS_HAVERCVDSYN(tp->t_state)) 2958 continue; 2959 /* 2960 * A timestamp received in a SYN makes 2961 * it ok to send timestamp requests and replies. 2962 */ 2963 tp->t_flags |= TF_RCVD_TSTMP; 2964 tp->ts_recent = oi->ts_val; 2965 tp->ts_recent_age = tcp_now; 2966 break; 2967 2968 case TCPOPT_SACK_PERMITTED: 2969 if (optlen != TCPOLEN_SACK_PERMITTED) 2970 continue; 2971 if (!(th->th_flags & TH_SYN)) 2972 continue; 2973 if (TCPS_HAVERCVDSYN(tp->t_state)) 2974 continue; 2975 if (tcp_do_sack) { 2976 tp->t_flags |= TF_SACK_PERMIT; 2977 tp->t_flags |= TF_WILL_SACK; 2978 } 2979 break; 2980 2981 case TCPOPT_SACK: 2982 tcp_sack_option(tp, th, cp, optlen); 2983 break; 2984 #ifdef TCP_SIGNATURE 2985 case TCPOPT_SIGNATURE: 2986 if (optlen != TCPOLEN_SIGNATURE) 2987 continue; 2988 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2989 return (-1); 2990 2991 sigp = sigbuf; 2992 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2993 memset(cp + 2, 0, TCP_SIGLEN); 2994 tp->t_flags |= TF_SIGNATURE; 2995 break; 2996 #endif 2997 } 2998 } 2999 3000 #ifdef TCP_SIGNATURE 3001 if (tp->t_flags & TF_SIGNATURE) { 3002 3003 sav = tcp_signature_getsav(m, th); 3004 3005 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3006 return (-1); 3007 } 3008 3009 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3010 if (sav == NULL) 3011 return (-1); 3012 #ifdef FAST_IPSEC 3013 KEY_FREESAV(&sav); 3014 #else 3015 key_freesav(sav); 3016 #endif 3017 return (-1); 3018 } 3019 3020 if (sigp) { 3021 char sig[TCP_SIGLEN]; 3022 3023 TCP_FIELDS_TO_NET(th); 3024 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3025 TCP_FIELDS_TO_HOST(th); 3026 if (sav == NULL) 3027 return (-1); 3028 #ifdef FAST_IPSEC 3029 KEY_FREESAV(&sav); 3030 #else 3031 key_freesav(sav); 3032 #endif 3033 return (-1); 3034 } 3035 TCP_FIELDS_TO_HOST(th); 3036 3037 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3038 tcpstat.tcps_badsig++; 3039 if (sav == NULL) 3040 return (-1); 3041 #ifdef FAST_IPSEC 3042 KEY_FREESAV(&sav); 3043 #else 3044 key_freesav(sav); 3045 #endif 3046 return (-1); 3047 } else 3048 tcpstat.tcps_goodsig++; 3049 3050 key_sa_recordxfer(sav, m); 3051 #ifdef FAST_IPSEC 3052 KEY_FREESAV(&sav); 3053 #else 3054 key_freesav(sav); 3055 #endif 3056 } 3057 #endif 3058 3059 return (0); 3060 } 3061 3062 /* 3063 * Pull out of band byte out of a segment so 3064 * it doesn't appear in the user's data queue. 3065 * It is still reflected in the segment length for 3066 * sequencing purposes. 3067 */ 3068 void 3069 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3070 struct mbuf *m, int off) 3071 { 3072 int cnt = off + th->th_urp - 1; 3073 3074 while (cnt >= 0) { 3075 if (m->m_len > cnt) { 3076 char *cp = mtod(m, caddr_t) + cnt; 3077 struct tcpcb *tp = sototcpcb(so); 3078 3079 tp->t_iobc = *cp; 3080 tp->t_oobflags |= TCPOOB_HAVEDATA; 3081 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3082 m->m_len--; 3083 return; 3084 } 3085 cnt -= m->m_len; 3086 m = m->m_next; 3087 if (m == 0) 3088 break; 3089 } 3090 panic("tcp_pulloutofband"); 3091 } 3092 3093 /* 3094 * Collect new round-trip time estimate 3095 * and update averages and current timeout. 3096 */ 3097 void 3098 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3099 { 3100 int32_t delta; 3101 3102 tcpstat.tcps_rttupdated++; 3103 if (tp->t_srtt != 0) { 3104 /* 3105 * srtt is stored as fixed point with 3 bits after the 3106 * binary point (i.e., scaled by 8). The following magic 3107 * is equivalent to the smoothing algorithm in rfc793 with 3108 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3109 * point). Adjust rtt to origin 0. 3110 */ 3111 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3112 if ((tp->t_srtt += delta) <= 0) 3113 tp->t_srtt = 1 << 2; 3114 /* 3115 * We accumulate a smoothed rtt variance (actually, a 3116 * smoothed mean difference), then set the retransmit 3117 * timer to smoothed rtt + 4 times the smoothed variance. 3118 * rttvar is stored as fixed point with 2 bits after the 3119 * binary point (scaled by 4). The following is 3120 * equivalent to rfc793 smoothing with an alpha of .75 3121 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3122 * rfc793's wired-in beta. 3123 */ 3124 if (delta < 0) 3125 delta = -delta; 3126 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3127 if ((tp->t_rttvar += delta) <= 0) 3128 tp->t_rttvar = 1 << 2; 3129 } else { 3130 /* 3131 * No rtt measurement yet - use the unsmoothed rtt. 3132 * Set the variance to half the rtt (so our first 3133 * retransmit happens at 3*rtt). 3134 */ 3135 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3136 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3137 } 3138 tp->t_rtttime = 0; 3139 tp->t_rxtshift = 0; 3140 3141 /* 3142 * the retransmit should happen at rtt + 4 * rttvar. 3143 * Because of the way we do the smoothing, srtt and rttvar 3144 * will each average +1/2 tick of bias. When we compute 3145 * the retransmit timer, we want 1/2 tick of rounding and 3146 * 1 extra tick because of +-1/2 tick uncertainty in the 3147 * firing of the timer. The bias will give us exactly the 3148 * 1.5 tick we need. But, because the bias is 3149 * statistical, we have to test that we don't drop below 3150 * the minimum feasible timer (which is 2 ticks). 3151 */ 3152 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3153 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3154 3155 /* 3156 * We received an ack for a packet that wasn't retransmitted; 3157 * it is probably safe to discard any error indications we've 3158 * received recently. This isn't quite right, but close enough 3159 * for now (a route might have failed after we sent a segment, 3160 * and the return path might not be symmetrical). 3161 */ 3162 tp->t_softerror = 0; 3163 } 3164 3165 void 3166 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3167 { 3168 if (tp->t_partialacks < 0) { 3169 /* 3170 * We were not in fast recovery. Reset the duplicate ack 3171 * counter. 3172 */ 3173 tp->t_dupacks = 0; 3174 } else { 3175 /* 3176 * Clamp the congestion window to the crossover point and 3177 * exit fast recovery. 3178 */ 3179 if (tp->snd_cwnd > tp->snd_ssthresh) 3180 tp->snd_cwnd = tp->snd_ssthresh; 3181 tp->t_partialacks = -1; 3182 tp->t_dupacks = 0; 3183 } 3184 } 3185 3186 /* 3187 * Implement the NewReno response to a new ack, checking for partial acks in 3188 * fast recovery. 3189 */ 3190 void 3191 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3192 { 3193 if (tp->t_partialacks < 0) { 3194 /* 3195 * We were not in fast recovery. Reset the duplicate ack 3196 * counter. 3197 */ 3198 tp->t_dupacks = 0; 3199 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3200 /* 3201 * This is a partial ack. Retransmit the first unacknowledged 3202 * segment and deflate the congestion window by the amount of 3203 * acknowledged data. Do not exit fast recovery. 3204 */ 3205 tcp_seq onxt = tp->snd_nxt; 3206 u_long ocwnd = tp->snd_cwnd; 3207 3208 /* 3209 * snd_una has not yet been updated and the socket's send 3210 * buffer has not yet drained off the ACK'd data, so we 3211 * have to leave snd_una as it was to get the correct data 3212 * offset in tcp_output(). 3213 */ 3214 if (++tp->t_partialacks == 1) 3215 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3216 tp->t_rtttime = 0; 3217 tp->snd_nxt = th->th_ack; 3218 /* 3219 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3220 * is not yet updated when we're called. 3221 */ 3222 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3223 (void) tcp_output(tp); 3224 tp->snd_cwnd = ocwnd; 3225 if (SEQ_GT(onxt, tp->snd_nxt)) 3226 tp->snd_nxt = onxt; 3227 /* 3228 * Partial window deflation. Relies on fact that tp->snd_una 3229 * not updated yet. 3230 */ 3231 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3232 } else { 3233 /* 3234 * Complete ack. Inflate the congestion window to ssthresh 3235 * and exit fast recovery. 3236 * 3237 * Window inflation should have left us with approx. 3238 * snd_ssthresh outstanding data. But in case we 3239 * would be inclined to send a burst, better to do 3240 * it via the slow start mechanism. 3241 */ 3242 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3243 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3244 + tp->t_segsz; 3245 else 3246 tp->snd_cwnd = tp->snd_ssthresh; 3247 tp->t_partialacks = -1; 3248 tp->t_dupacks = 0; 3249 } 3250 } 3251 3252 3253 /* 3254 * TCP compressed state engine. Currently used to hold compressed 3255 * state for SYN_RECEIVED. 3256 */ 3257 3258 u_long syn_cache_count; 3259 u_int32_t syn_hash1, syn_hash2; 3260 3261 #define SYN_HASH(sa, sp, dp) \ 3262 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3263 ((u_int32_t)(sp)))^syn_hash2))) 3264 #ifndef INET6 3265 #define SYN_HASHALL(hash, src, dst) \ 3266 do { \ 3267 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3268 ((const struct sockaddr_in *)(src))->sin_port, \ 3269 ((const struct sockaddr_in *)(dst))->sin_port); \ 3270 } while (/*CONSTCOND*/ 0) 3271 #else 3272 #define SYN_HASH6(sa, sp, dp) \ 3273 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3274 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3275 & 0x7fffffff) 3276 3277 #define SYN_HASHALL(hash, src, dst) \ 3278 do { \ 3279 switch ((src)->sa_family) { \ 3280 case AF_INET: \ 3281 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3282 ((const struct sockaddr_in *)(src))->sin_port, \ 3283 ((const struct sockaddr_in *)(dst))->sin_port); \ 3284 break; \ 3285 case AF_INET6: \ 3286 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3287 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3288 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3289 break; \ 3290 default: \ 3291 hash = 0; \ 3292 } \ 3293 } while (/*CONSTCOND*/0) 3294 #endif /* INET6 */ 3295 3296 #define SYN_CACHE_RM(sc) \ 3297 do { \ 3298 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3299 (sc), sc_bucketq); \ 3300 (sc)->sc_tp = NULL; \ 3301 LIST_REMOVE((sc), sc_tpq); \ 3302 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3303 callout_stop(&(sc)->sc_timer); \ 3304 syn_cache_count--; \ 3305 } while (/*CONSTCOND*/0) 3306 3307 #define SYN_CACHE_PUT(sc) \ 3308 do { \ 3309 if ((sc)->sc_ipopts) \ 3310 (void) m_free((sc)->sc_ipopts); \ 3311 if ((sc)->sc_route4.ro_rt != NULL) \ 3312 RTFREE((sc)->sc_route4.ro_rt); \ 3313 if (callout_invoking(&(sc)->sc_timer)) \ 3314 (sc)->sc_flags |= SCF_DEAD; \ 3315 else \ 3316 pool_put(&syn_cache_pool, (sc)); \ 3317 } while (/*CONSTCOND*/0) 3318 3319 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3320 3321 /* 3322 * We don't estimate RTT with SYNs, so each packet starts with the default 3323 * RTT and each timer step has a fixed timeout value. 3324 */ 3325 #define SYN_CACHE_TIMER_ARM(sc) \ 3326 do { \ 3327 TCPT_RANGESET((sc)->sc_rxtcur, \ 3328 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3329 TCPTV_REXMTMAX); \ 3330 callout_reset(&(sc)->sc_timer, \ 3331 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3332 } while (/*CONSTCOND*/0) 3333 3334 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3335 3336 void 3337 syn_cache_init(void) 3338 { 3339 int i; 3340 3341 /* Initialize the hash buckets. */ 3342 for (i = 0; i < tcp_syn_cache_size; i++) 3343 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3344 } 3345 3346 void 3347 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3348 { 3349 struct syn_cache_head *scp; 3350 struct syn_cache *sc2; 3351 int s; 3352 3353 /* 3354 * If there are no entries in the hash table, reinitialize 3355 * the hash secrets. 3356 */ 3357 if (syn_cache_count == 0) { 3358 syn_hash1 = arc4random(); 3359 syn_hash2 = arc4random(); 3360 } 3361 3362 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3363 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3364 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3365 3366 /* 3367 * Make sure that we don't overflow the per-bucket 3368 * limit or the total cache size limit. 3369 */ 3370 s = splsoftnet(); 3371 if (scp->sch_length >= tcp_syn_bucket_limit) { 3372 tcpstat.tcps_sc_bucketoverflow++; 3373 /* 3374 * The bucket is full. Toss the oldest element in the 3375 * bucket. This will be the first entry in the bucket. 3376 */ 3377 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3378 #ifdef DIAGNOSTIC 3379 /* 3380 * This should never happen; we should always find an 3381 * entry in our bucket. 3382 */ 3383 if (sc2 == NULL) 3384 panic("syn_cache_insert: bucketoverflow: impossible"); 3385 #endif 3386 SYN_CACHE_RM(sc2); 3387 SYN_CACHE_PUT(sc2); 3388 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3389 struct syn_cache_head *scp2, *sce; 3390 3391 tcpstat.tcps_sc_overflowed++; 3392 /* 3393 * The cache is full. Toss the oldest entry in the 3394 * first non-empty bucket we can find. 3395 * 3396 * XXX We would really like to toss the oldest 3397 * entry in the cache, but we hope that this 3398 * condition doesn't happen very often. 3399 */ 3400 scp2 = scp; 3401 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3402 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3403 for (++scp2; scp2 != scp; scp2++) { 3404 if (scp2 >= sce) 3405 scp2 = &tcp_syn_cache[0]; 3406 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3407 break; 3408 } 3409 #ifdef DIAGNOSTIC 3410 /* 3411 * This should never happen; we should always find a 3412 * non-empty bucket. 3413 */ 3414 if (scp2 == scp) 3415 panic("syn_cache_insert: cacheoverflow: " 3416 "impossible"); 3417 #endif 3418 } 3419 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3420 SYN_CACHE_RM(sc2); 3421 SYN_CACHE_PUT(sc2); 3422 } 3423 3424 /* 3425 * Initialize the entry's timer. 3426 */ 3427 sc->sc_rxttot = 0; 3428 sc->sc_rxtshift = 0; 3429 SYN_CACHE_TIMER_ARM(sc); 3430 3431 /* Link it from tcpcb entry */ 3432 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3433 3434 /* Put it into the bucket. */ 3435 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3436 scp->sch_length++; 3437 syn_cache_count++; 3438 3439 tcpstat.tcps_sc_added++; 3440 splx(s); 3441 } 3442 3443 /* 3444 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3445 * If we have retransmitted an entry the maximum number of times, expire 3446 * that entry. 3447 */ 3448 void 3449 syn_cache_timer(void *arg) 3450 { 3451 struct syn_cache *sc = arg; 3452 int s; 3453 3454 s = splsoftnet(); 3455 callout_ack(&sc->sc_timer); 3456 3457 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3458 tcpstat.tcps_sc_delayed_free++; 3459 pool_put(&syn_cache_pool, sc); 3460 splx(s); 3461 return; 3462 } 3463 3464 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3465 /* Drop it -- too many retransmissions. */ 3466 goto dropit; 3467 } 3468 3469 /* 3470 * Compute the total amount of time this entry has 3471 * been on a queue. If this entry has been on longer 3472 * than the keep alive timer would allow, expire it. 3473 */ 3474 sc->sc_rxttot += sc->sc_rxtcur; 3475 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3476 goto dropit; 3477 3478 tcpstat.tcps_sc_retransmitted++; 3479 (void) syn_cache_respond(sc, NULL); 3480 3481 /* Advance the timer back-off. */ 3482 sc->sc_rxtshift++; 3483 SYN_CACHE_TIMER_ARM(sc); 3484 3485 splx(s); 3486 return; 3487 3488 dropit: 3489 tcpstat.tcps_sc_timed_out++; 3490 SYN_CACHE_RM(sc); 3491 SYN_CACHE_PUT(sc); 3492 splx(s); 3493 } 3494 3495 /* 3496 * Remove syn cache created by the specified tcb entry, 3497 * because this does not make sense to keep them 3498 * (if there's no tcb entry, syn cache entry will never be used) 3499 */ 3500 void 3501 syn_cache_cleanup(struct tcpcb *tp) 3502 { 3503 struct syn_cache *sc, *nsc; 3504 int s; 3505 3506 s = splsoftnet(); 3507 3508 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3509 nsc = LIST_NEXT(sc, sc_tpq); 3510 3511 #ifdef DIAGNOSTIC 3512 if (sc->sc_tp != tp) 3513 panic("invalid sc_tp in syn_cache_cleanup"); 3514 #endif 3515 SYN_CACHE_RM(sc); 3516 SYN_CACHE_PUT(sc); 3517 } 3518 /* just for safety */ 3519 LIST_INIT(&tp->t_sc); 3520 3521 splx(s); 3522 } 3523 3524 /* 3525 * Find an entry in the syn cache. 3526 */ 3527 struct syn_cache * 3528 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3529 struct syn_cache_head **headp) 3530 { 3531 struct syn_cache *sc; 3532 struct syn_cache_head *scp; 3533 u_int32_t hash; 3534 int s; 3535 3536 SYN_HASHALL(hash, src, dst); 3537 3538 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3539 *headp = scp; 3540 s = splsoftnet(); 3541 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3542 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3543 if (sc->sc_hash != hash) 3544 continue; 3545 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3546 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3547 splx(s); 3548 return (sc); 3549 } 3550 } 3551 splx(s); 3552 return (NULL); 3553 } 3554 3555 /* 3556 * This function gets called when we receive an ACK for a 3557 * socket in the LISTEN state. We look up the connection 3558 * in the syn cache, and if its there, we pull it out of 3559 * the cache and turn it into a full-blown connection in 3560 * the SYN-RECEIVED state. 3561 * 3562 * The return values may not be immediately obvious, and their effects 3563 * can be subtle, so here they are: 3564 * 3565 * NULL SYN was not found in cache; caller should drop the 3566 * packet and send an RST. 3567 * 3568 * -1 We were unable to create the new connection, and are 3569 * aborting it. An ACK,RST is being sent to the peer 3570 * (unless we got screwey sequence numbners; see below), 3571 * because the 3-way handshake has been completed. Caller 3572 * should not free the mbuf, since we may be using it. If 3573 * we are not, we will free it. 3574 * 3575 * Otherwise, the return value is a pointer to the new socket 3576 * associated with the connection. 3577 */ 3578 struct socket * 3579 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3580 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3581 struct socket *so, struct mbuf *m) 3582 { 3583 struct syn_cache *sc; 3584 struct syn_cache_head *scp; 3585 struct inpcb *inp = NULL; 3586 #ifdef INET6 3587 struct in6pcb *in6p = NULL; 3588 #endif 3589 struct tcpcb *tp = 0; 3590 struct mbuf *am; 3591 int s; 3592 struct socket *oso; 3593 3594 s = splsoftnet(); 3595 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3596 splx(s); 3597 return (NULL); 3598 } 3599 3600 /* 3601 * Verify the sequence and ack numbers. Try getting the correct 3602 * response again. 3603 */ 3604 if ((th->th_ack != sc->sc_iss + 1) || 3605 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3606 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3607 (void) syn_cache_respond(sc, m); 3608 splx(s); 3609 return ((struct socket *)(-1)); 3610 } 3611 3612 /* Remove this cache entry */ 3613 SYN_CACHE_RM(sc); 3614 splx(s); 3615 3616 /* 3617 * Ok, create the full blown connection, and set things up 3618 * as they would have been set up if we had created the 3619 * connection when the SYN arrived. If we can't create 3620 * the connection, abort it. 3621 */ 3622 /* 3623 * inp still has the OLD in_pcb stuff, set the 3624 * v6-related flags on the new guy, too. This is 3625 * done particularly for the case where an AF_INET6 3626 * socket is bound only to a port, and a v4 connection 3627 * comes in on that port. 3628 * we also copy the flowinfo from the original pcb 3629 * to the new one. 3630 */ 3631 oso = so; 3632 so = sonewconn(so, SS_ISCONNECTED); 3633 if (so == NULL) 3634 goto resetandabort; 3635 3636 switch (so->so_proto->pr_domain->dom_family) { 3637 #ifdef INET 3638 case AF_INET: 3639 inp = sotoinpcb(so); 3640 break; 3641 #endif 3642 #ifdef INET6 3643 case AF_INET6: 3644 in6p = sotoin6pcb(so); 3645 break; 3646 #endif 3647 } 3648 switch (src->sa_family) { 3649 #ifdef INET 3650 case AF_INET: 3651 if (inp) { 3652 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3653 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3654 inp->inp_options = ip_srcroute(); 3655 in_pcbstate(inp, INP_BOUND); 3656 if (inp->inp_options == NULL) { 3657 inp->inp_options = sc->sc_ipopts; 3658 sc->sc_ipopts = NULL; 3659 } 3660 } 3661 #ifdef INET6 3662 else if (in6p) { 3663 /* IPv4 packet to AF_INET6 socket */ 3664 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3665 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3666 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3667 &in6p->in6p_laddr.s6_addr32[3], 3668 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3669 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3670 in6totcpcb(in6p)->t_family = AF_INET; 3671 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3672 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3673 else 3674 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3675 in6_pcbstate(in6p, IN6P_BOUND); 3676 } 3677 #endif 3678 break; 3679 #endif 3680 #ifdef INET6 3681 case AF_INET6: 3682 if (in6p) { 3683 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3684 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3685 in6_pcbstate(in6p, IN6P_BOUND); 3686 } 3687 break; 3688 #endif 3689 } 3690 #ifdef INET6 3691 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3692 struct in6pcb *oin6p = sotoin6pcb(oso); 3693 /* inherit socket options from the listening socket */ 3694 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3695 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3696 m_freem(in6p->in6p_options); 3697 in6p->in6p_options = 0; 3698 } 3699 ip6_savecontrol(in6p, &in6p->in6p_options, 3700 mtod(m, struct ip6_hdr *), m); 3701 } 3702 #endif 3703 3704 #if defined(IPSEC) || defined(FAST_IPSEC) 3705 /* 3706 * we make a copy of policy, instead of sharing the policy, 3707 * for better behavior in terms of SA lookup and dead SA removal. 3708 */ 3709 if (inp) { 3710 /* copy old policy into new socket's */ 3711 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3712 printf("tcp_input: could not copy policy\n"); 3713 } 3714 #ifdef INET6 3715 else if (in6p) { 3716 /* copy old policy into new socket's */ 3717 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3718 in6p->in6p_sp)) 3719 printf("tcp_input: could not copy policy\n"); 3720 } 3721 #endif 3722 #endif 3723 3724 /* 3725 * Give the new socket our cached route reference. 3726 */ 3727 if (inp) 3728 inp->inp_route = sc->sc_route4; /* struct assignment */ 3729 #ifdef INET6 3730 else 3731 in6p->in6p_route = sc->sc_route6; 3732 #endif 3733 sc->sc_route4.ro_rt = NULL; 3734 3735 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3736 if (am == NULL) 3737 goto resetandabort; 3738 MCLAIM(am, &tcp_mowner); 3739 am->m_len = src->sa_len; 3740 bcopy(src, mtod(am, caddr_t), src->sa_len); 3741 if (inp) { 3742 if (in_pcbconnect(inp, am, NULL)) { 3743 (void) m_free(am); 3744 goto resetandabort; 3745 } 3746 } 3747 #ifdef INET6 3748 else if (in6p) { 3749 if (src->sa_family == AF_INET) { 3750 /* IPv4 packet to AF_INET6 socket */ 3751 struct sockaddr_in6 *sin6; 3752 sin6 = mtod(am, struct sockaddr_in6 *); 3753 am->m_len = sizeof(*sin6); 3754 bzero(sin6, sizeof(*sin6)); 3755 sin6->sin6_family = AF_INET6; 3756 sin6->sin6_len = sizeof(*sin6); 3757 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3758 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3759 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3760 &sin6->sin6_addr.s6_addr32[3], 3761 sizeof(sin6->sin6_addr.s6_addr32[3])); 3762 } 3763 if (in6_pcbconnect(in6p, am, NULL)) { 3764 (void) m_free(am); 3765 goto resetandabort; 3766 } 3767 } 3768 #endif 3769 else { 3770 (void) m_free(am); 3771 goto resetandabort; 3772 } 3773 (void) m_free(am); 3774 3775 if (inp) 3776 tp = intotcpcb(inp); 3777 #ifdef INET6 3778 else if (in6p) 3779 tp = in6totcpcb(in6p); 3780 #endif 3781 else 3782 tp = NULL; 3783 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3784 if (sc->sc_request_r_scale != 15) { 3785 tp->requested_s_scale = sc->sc_requested_s_scale; 3786 tp->request_r_scale = sc->sc_request_r_scale; 3787 tp->snd_scale = sc->sc_requested_s_scale; 3788 tp->rcv_scale = sc->sc_request_r_scale; 3789 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3790 } 3791 if (sc->sc_flags & SCF_TIMESTAMP) 3792 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3793 tp->ts_timebase = sc->sc_timebase; 3794 3795 tp->t_template = tcp_template(tp); 3796 if (tp->t_template == 0) { 3797 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3798 so = NULL; 3799 m_freem(m); 3800 goto abort; 3801 } 3802 3803 tp->iss = sc->sc_iss; 3804 tp->irs = sc->sc_irs; 3805 tcp_sendseqinit(tp); 3806 tcp_rcvseqinit(tp); 3807 tp->t_state = TCPS_SYN_RECEIVED; 3808 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3809 tcpstat.tcps_accepts++; 3810 3811 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3812 tp->t_flags |= TF_WILL_SACK; 3813 3814 #ifdef TCP_SIGNATURE 3815 if (sc->sc_flags & SCF_SIGNATURE) 3816 tp->t_flags |= TF_SIGNATURE; 3817 #endif 3818 3819 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3820 tp->t_ourmss = sc->sc_ourmaxseg; 3821 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3822 3823 /* 3824 * Initialize the initial congestion window. If we 3825 * had to retransmit the SYN,ACK, we must initialize cwnd 3826 * to 1 segment (i.e. the Loss Window). 3827 */ 3828 if (sc->sc_rxtshift) 3829 tp->snd_cwnd = tp->t_peermss; 3830 else { 3831 int ss = tcp_init_win; 3832 #ifdef INET 3833 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3834 ss = tcp_init_win_local; 3835 #endif 3836 #ifdef INET6 3837 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3838 ss = tcp_init_win_local; 3839 #endif 3840 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3841 } 3842 3843 tcp_rmx_rtt(tp); 3844 tp->snd_wl1 = sc->sc_irs; 3845 tp->rcv_up = sc->sc_irs + 1; 3846 3847 /* 3848 * This is what whould have happened in tcp_output() when 3849 * the SYN,ACK was sent. 3850 */ 3851 tp->snd_up = tp->snd_una; 3852 tp->snd_max = tp->snd_nxt = tp->iss+1; 3853 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3854 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3855 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3856 tp->last_ack_sent = tp->rcv_nxt; 3857 tp->t_partialacks = -1; 3858 tp->t_dupacks = 0; 3859 3860 tcpstat.tcps_sc_completed++; 3861 SYN_CACHE_PUT(sc); 3862 return (so); 3863 3864 resetandabort: 3865 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3866 abort: 3867 if (so != NULL) 3868 (void) soabort(so); 3869 SYN_CACHE_PUT(sc); 3870 tcpstat.tcps_sc_aborted++; 3871 return ((struct socket *)(-1)); 3872 } 3873 3874 /* 3875 * This function is called when we get a RST for a 3876 * non-existent connection, so that we can see if the 3877 * connection is in the syn cache. If it is, zap it. 3878 */ 3879 3880 void 3881 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3882 { 3883 struct syn_cache *sc; 3884 struct syn_cache_head *scp; 3885 int s = splsoftnet(); 3886 3887 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3888 splx(s); 3889 return; 3890 } 3891 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3892 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3893 splx(s); 3894 return; 3895 } 3896 SYN_CACHE_RM(sc); 3897 splx(s); 3898 tcpstat.tcps_sc_reset++; 3899 SYN_CACHE_PUT(sc); 3900 } 3901 3902 void 3903 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3904 struct tcphdr *th) 3905 { 3906 struct syn_cache *sc; 3907 struct syn_cache_head *scp; 3908 int s; 3909 3910 s = splsoftnet(); 3911 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3912 splx(s); 3913 return; 3914 } 3915 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3916 if (ntohl (th->th_seq) != sc->sc_iss) { 3917 splx(s); 3918 return; 3919 } 3920 3921 /* 3922 * If we've retransmitted 3 times and this is our second error, 3923 * we remove the entry. Otherwise, we allow it to continue on. 3924 * This prevents us from incorrectly nuking an entry during a 3925 * spurious network outage. 3926 * 3927 * See tcp_notify(). 3928 */ 3929 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3930 sc->sc_flags |= SCF_UNREACH; 3931 splx(s); 3932 return; 3933 } 3934 3935 SYN_CACHE_RM(sc); 3936 splx(s); 3937 tcpstat.tcps_sc_unreach++; 3938 SYN_CACHE_PUT(sc); 3939 } 3940 3941 /* 3942 * Given a LISTEN socket and an inbound SYN request, add 3943 * this to the syn cache, and send back a segment: 3944 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3945 * to the source. 3946 * 3947 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3948 * Doing so would require that we hold onto the data and deliver it 3949 * to the application. However, if we are the target of a SYN-flood 3950 * DoS attack, an attacker could send data which would eventually 3951 * consume all available buffer space if it were ACKed. By not ACKing 3952 * the data, we avoid this DoS scenario. 3953 */ 3954 3955 int 3956 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3957 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3958 int optlen, struct tcp_opt_info *oi) 3959 { 3960 struct tcpcb tb, *tp; 3961 long win; 3962 struct syn_cache *sc; 3963 struct syn_cache_head *scp; 3964 struct mbuf *ipopts; 3965 struct tcp_opt_info opti; 3966 3967 tp = sototcpcb(so); 3968 3969 bzero(&opti, sizeof(opti)); 3970 3971 /* 3972 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3973 * 3974 * Note this check is performed in tcp_input() very early on. 3975 */ 3976 3977 /* 3978 * Initialize some local state. 3979 */ 3980 win = sbspace(&so->so_rcv); 3981 if (win > TCP_MAXWIN) 3982 win = TCP_MAXWIN; 3983 3984 switch (src->sa_family) { 3985 #ifdef INET 3986 case AF_INET: 3987 /* 3988 * Remember the IP options, if any. 3989 */ 3990 ipopts = ip_srcroute(); 3991 break; 3992 #endif 3993 default: 3994 ipopts = NULL; 3995 } 3996 3997 #ifdef TCP_SIGNATURE 3998 if (optp || (tp->t_flags & TF_SIGNATURE)) 3999 #else 4000 if (optp) 4001 #endif 4002 { 4003 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4004 #ifdef TCP_SIGNATURE 4005 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4006 #endif 4007 tb.t_state = TCPS_LISTEN; 4008 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4009 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4010 return (0); 4011 } else 4012 tb.t_flags = 0; 4013 4014 /* 4015 * See if we already have an entry for this connection. 4016 * If we do, resend the SYN,ACK. We do not count this 4017 * as a retransmission (XXX though maybe we should). 4018 */ 4019 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4020 tcpstat.tcps_sc_dupesyn++; 4021 if (ipopts) { 4022 /* 4023 * If we were remembering a previous source route, 4024 * forget it and use the new one we've been given. 4025 */ 4026 if (sc->sc_ipopts) 4027 (void) m_free(sc->sc_ipopts); 4028 sc->sc_ipopts = ipopts; 4029 } 4030 sc->sc_timestamp = tb.ts_recent; 4031 if (syn_cache_respond(sc, m) == 0) { 4032 tcpstat.tcps_sndacks++; 4033 tcpstat.tcps_sndtotal++; 4034 } 4035 return (1); 4036 } 4037 4038 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4039 if (sc == NULL) { 4040 if (ipopts) 4041 (void) m_free(ipopts); 4042 return (0); 4043 } 4044 4045 /* 4046 * Fill in the cache, and put the necessary IP and TCP 4047 * options into the reply. 4048 */ 4049 bzero(sc, sizeof(struct syn_cache)); 4050 callout_init(&sc->sc_timer); 4051 bcopy(src, &sc->sc_src, src->sa_len); 4052 bcopy(dst, &sc->sc_dst, dst->sa_len); 4053 sc->sc_flags = 0; 4054 sc->sc_ipopts = ipopts; 4055 sc->sc_irs = th->th_seq; 4056 switch (src->sa_family) { 4057 #ifdef INET 4058 case AF_INET: 4059 { 4060 struct sockaddr_in *srcin = (void *) src; 4061 struct sockaddr_in *dstin = (void *) dst; 4062 4063 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4064 &srcin->sin_addr, dstin->sin_port, 4065 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4066 break; 4067 } 4068 #endif /* INET */ 4069 #ifdef INET6 4070 case AF_INET6: 4071 { 4072 struct sockaddr_in6 *srcin6 = (void *) src; 4073 struct sockaddr_in6 *dstin6 = (void *) dst; 4074 4075 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4076 &srcin6->sin6_addr, dstin6->sin6_port, 4077 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4078 break; 4079 } 4080 #endif /* INET6 */ 4081 } 4082 sc->sc_peermaxseg = oi->maxseg; 4083 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4084 m->m_pkthdr.rcvif : NULL, 4085 sc->sc_src.sa.sa_family); 4086 sc->sc_win = win; 4087 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4088 sc->sc_timestamp = tb.ts_recent; 4089 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4090 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4091 sc->sc_flags |= SCF_TIMESTAMP; 4092 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4093 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4094 sc->sc_requested_s_scale = tb.requested_s_scale; 4095 sc->sc_request_r_scale = 0; 4096 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4097 TCP_MAXWIN << sc->sc_request_r_scale < 4098 so->so_rcv.sb_hiwat) 4099 sc->sc_request_r_scale++; 4100 } else { 4101 sc->sc_requested_s_scale = 15; 4102 sc->sc_request_r_scale = 15; 4103 } 4104 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4105 sc->sc_flags |= SCF_SACK_PERMIT; 4106 #ifdef TCP_SIGNATURE 4107 if (tb.t_flags & TF_SIGNATURE) 4108 sc->sc_flags |= SCF_SIGNATURE; 4109 #endif 4110 sc->sc_tp = tp; 4111 if (syn_cache_respond(sc, m) == 0) { 4112 syn_cache_insert(sc, tp); 4113 tcpstat.tcps_sndacks++; 4114 tcpstat.tcps_sndtotal++; 4115 } else { 4116 SYN_CACHE_PUT(sc); 4117 tcpstat.tcps_sc_dropped++; 4118 } 4119 return (1); 4120 } 4121 4122 int 4123 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4124 { 4125 struct route *ro; 4126 u_int8_t *optp; 4127 int optlen, error; 4128 u_int16_t tlen; 4129 struct ip *ip = NULL; 4130 #ifdef INET6 4131 struct ip6_hdr *ip6 = NULL; 4132 #endif 4133 struct tcpcb *tp; 4134 struct tcphdr *th; 4135 u_int hlen; 4136 struct socket *so; 4137 4138 switch (sc->sc_src.sa.sa_family) { 4139 case AF_INET: 4140 hlen = sizeof(struct ip); 4141 ro = &sc->sc_route4; 4142 break; 4143 #ifdef INET6 4144 case AF_INET6: 4145 hlen = sizeof(struct ip6_hdr); 4146 ro = (struct route *)&sc->sc_route6; 4147 break; 4148 #endif 4149 default: 4150 if (m) 4151 m_freem(m); 4152 return (EAFNOSUPPORT); 4153 } 4154 4155 /* Compute the size of the TCP options. */ 4156 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4157 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4158 #ifdef TCP_SIGNATURE 4159 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4160 #endif 4161 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4162 4163 tlen = hlen + sizeof(struct tcphdr) + optlen; 4164 4165 /* 4166 * Create the IP+TCP header from scratch. 4167 */ 4168 if (m) 4169 m_freem(m); 4170 #ifdef DIAGNOSTIC 4171 if (max_linkhdr + tlen > MCLBYTES) 4172 return (ENOBUFS); 4173 #endif 4174 MGETHDR(m, M_DONTWAIT, MT_DATA); 4175 if (m && tlen > MHLEN) { 4176 MCLGET(m, M_DONTWAIT); 4177 if ((m->m_flags & M_EXT) == 0) { 4178 m_freem(m); 4179 m = NULL; 4180 } 4181 } 4182 if (m == NULL) 4183 return (ENOBUFS); 4184 MCLAIM(m, &tcp_tx_mowner); 4185 4186 /* Fixup the mbuf. */ 4187 m->m_data += max_linkhdr; 4188 m->m_len = m->m_pkthdr.len = tlen; 4189 if (sc->sc_tp) { 4190 tp = sc->sc_tp; 4191 if (tp->t_inpcb) 4192 so = tp->t_inpcb->inp_socket; 4193 #ifdef INET6 4194 else if (tp->t_in6pcb) 4195 so = tp->t_in6pcb->in6p_socket; 4196 #endif 4197 else 4198 so = NULL; 4199 } else 4200 so = NULL; 4201 m->m_pkthdr.rcvif = NULL; 4202 memset(mtod(m, u_char *), 0, tlen); 4203 4204 switch (sc->sc_src.sa.sa_family) { 4205 case AF_INET: 4206 ip = mtod(m, struct ip *); 4207 ip->ip_v = 4; 4208 ip->ip_dst = sc->sc_src.sin.sin_addr; 4209 ip->ip_src = sc->sc_dst.sin.sin_addr; 4210 ip->ip_p = IPPROTO_TCP; 4211 th = (struct tcphdr *)(ip + 1); 4212 th->th_dport = sc->sc_src.sin.sin_port; 4213 th->th_sport = sc->sc_dst.sin.sin_port; 4214 break; 4215 #ifdef INET6 4216 case AF_INET6: 4217 ip6 = mtod(m, struct ip6_hdr *); 4218 ip6->ip6_vfc = IPV6_VERSION; 4219 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4220 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4221 ip6->ip6_nxt = IPPROTO_TCP; 4222 /* ip6_plen will be updated in ip6_output() */ 4223 th = (struct tcphdr *)(ip6 + 1); 4224 th->th_dport = sc->sc_src.sin6.sin6_port; 4225 th->th_sport = sc->sc_dst.sin6.sin6_port; 4226 break; 4227 #endif 4228 default: 4229 th = NULL; 4230 } 4231 4232 th->th_seq = htonl(sc->sc_iss); 4233 th->th_ack = htonl(sc->sc_irs + 1); 4234 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4235 th->th_flags = TH_SYN|TH_ACK; 4236 th->th_win = htons(sc->sc_win); 4237 /* th_sum already 0 */ 4238 /* th_urp already 0 */ 4239 4240 /* Tack on the TCP options. */ 4241 optp = (u_int8_t *)(th + 1); 4242 *optp++ = TCPOPT_MAXSEG; 4243 *optp++ = 4; 4244 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4245 *optp++ = sc->sc_ourmaxseg & 0xff; 4246 4247 if (sc->sc_request_r_scale != 15) { 4248 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4249 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4250 sc->sc_request_r_scale); 4251 optp += 4; 4252 } 4253 4254 if (sc->sc_flags & SCF_TIMESTAMP) { 4255 u_int32_t *lp = (u_int32_t *)(optp); 4256 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4257 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4258 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4259 *lp = htonl(sc->sc_timestamp); 4260 optp += TCPOLEN_TSTAMP_APPA; 4261 } 4262 4263 if (sc->sc_flags & SCF_SACK_PERMIT) { 4264 u_int8_t *p = optp; 4265 4266 /* Let the peer know that we will SACK. */ 4267 p[0] = TCPOPT_SACK_PERMITTED; 4268 p[1] = 2; 4269 p[2] = TCPOPT_NOP; 4270 p[3] = TCPOPT_NOP; 4271 optp += 4; 4272 } 4273 4274 #ifdef TCP_SIGNATURE 4275 if (sc->sc_flags & SCF_SIGNATURE) { 4276 struct secasvar *sav; 4277 u_int8_t *sigp; 4278 4279 sav = tcp_signature_getsav(m, th); 4280 4281 if (sav == NULL) { 4282 if (m) 4283 m_freem(m); 4284 return (EPERM); 4285 } 4286 4287 *optp++ = TCPOPT_SIGNATURE; 4288 *optp++ = TCPOLEN_SIGNATURE; 4289 sigp = optp; 4290 bzero(optp, TCP_SIGLEN); 4291 optp += TCP_SIGLEN; 4292 *optp++ = TCPOPT_NOP; 4293 *optp++ = TCPOPT_EOL; 4294 4295 (void)tcp_signature(m, th, hlen, sav, sigp); 4296 4297 key_sa_recordxfer(sav, m); 4298 #ifdef FAST_IPSEC 4299 KEY_FREESAV(&sav); 4300 #else 4301 key_freesav(sav); 4302 #endif 4303 } 4304 #endif 4305 4306 /* Compute the packet's checksum. */ 4307 switch (sc->sc_src.sa.sa_family) { 4308 case AF_INET: 4309 ip->ip_len = htons(tlen - hlen); 4310 th->th_sum = 0; 4311 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4312 break; 4313 #ifdef INET6 4314 case AF_INET6: 4315 ip6->ip6_plen = htons(tlen - hlen); 4316 th->th_sum = 0; 4317 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4318 break; 4319 #endif 4320 } 4321 4322 /* 4323 * Fill in some straggling IP bits. Note the stack expects 4324 * ip_len to be in host order, for convenience. 4325 */ 4326 switch (sc->sc_src.sa.sa_family) { 4327 #ifdef INET 4328 case AF_INET: 4329 ip->ip_len = htons(tlen); 4330 ip->ip_ttl = ip_defttl; 4331 /* XXX tos? */ 4332 break; 4333 #endif 4334 #ifdef INET6 4335 case AF_INET6: 4336 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4337 ip6->ip6_vfc |= IPV6_VERSION; 4338 ip6->ip6_plen = htons(tlen - hlen); 4339 /* ip6_hlim will be initialized afterwards */ 4340 /* XXX flowlabel? */ 4341 break; 4342 #endif 4343 } 4344 4345 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4346 tp = sc->sc_tp; 4347 4348 switch (sc->sc_src.sa.sa_family) { 4349 #ifdef INET 4350 case AF_INET: 4351 error = ip_output(m, sc->sc_ipopts, ro, 4352 (ip_mtudisc ? IP_MTUDISC : 0), 4353 (struct ip_moptions *)NULL, so); 4354 break; 4355 #endif 4356 #ifdef INET6 4357 case AF_INET6: 4358 ip6->ip6_hlim = in6_selecthlim(NULL, 4359 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4360 4361 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4362 (struct ip6_moptions *)0, so, NULL); 4363 break; 4364 #endif 4365 default: 4366 error = EAFNOSUPPORT; 4367 break; 4368 } 4369 return (error); 4370 } 4371