1 /* $NetBSD: tcp_input.c,v 1.285 2008/04/23 06:09:05 thorpej 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, 2006 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 * This code is derived from software contributed to The NetBSD Foundation 82 * by Rui Paulo. 83 * 84 * Redistribution and use in source and binary forms, with or without 85 * modification, are permitted provided that the following conditions 86 * are met: 87 * 1. Redistributions of source code must retain the above copyright 88 * notice, this list of conditions and the following disclaimer. 89 * 2. Redistributions in binary form must reproduce the above copyright 90 * notice, this list of conditions and the following disclaimer in the 91 * documentation and/or other materials provided with the distribution. 92 * 3. All advertising materials mentioning features or use of this software 93 * must display the following acknowledgement: 94 * This product includes software developed by the NetBSD 95 * Foundation, Inc. and its contributors. 96 * 4. Neither the name of The NetBSD Foundation nor the names of its 97 * contributors may be used to endorse or promote products derived 98 * from this software without specific prior written permission. 99 * 100 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 101 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 102 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 103 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 104 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 105 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 106 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 107 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 108 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 109 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 110 * POSSIBILITY OF SUCH DAMAGE. 111 */ 112 113 /* 114 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 115 * The Regents of the University of California. All rights reserved. 116 * 117 * Redistribution and use in source and binary forms, with or without 118 * modification, are permitted provided that the following conditions 119 * are met: 120 * 1. Redistributions of source code must retain the above copyright 121 * notice, this list of conditions and the following disclaimer. 122 * 2. Redistributions in binary form must reproduce the above copyright 123 * notice, this list of conditions and the following disclaimer in the 124 * documentation and/or other materials provided with the distribution. 125 * 3. Neither the name of the University nor the names of its contributors 126 * may be used to endorse or promote products derived from this software 127 * without specific prior written permission. 128 * 129 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 130 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 131 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 132 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 133 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 134 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 135 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 136 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 137 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 138 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 139 * SUCH DAMAGE. 140 * 141 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 142 */ 143 144 /* 145 * TODO list for SYN cache stuff: 146 * 147 * Find room for a "state" field, which is needed to keep a 148 * compressed state for TIME_WAIT TCBs. It's been noted already 149 * that this is fairly important for very high-volume web and 150 * mail servers, which use a large number of short-lived 151 * connections. 152 */ 153 154 #include <sys/cdefs.h> 155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.285 2008/04/23 06:09:05 thorpej Exp $"); 156 157 #include "opt_inet.h" 158 #include "opt_ipsec.h" 159 #include "opt_inet_csum.h" 160 #include "opt_tcp_debug.h" 161 162 #include <sys/param.h> 163 #include <sys/systm.h> 164 #include <sys/malloc.h> 165 #include <sys/mbuf.h> 166 #include <sys/protosw.h> 167 #include <sys/socket.h> 168 #include <sys/socketvar.h> 169 #include <sys/errno.h> 170 #include <sys/syslog.h> 171 #include <sys/pool.h> 172 #include <sys/domain.h> 173 #include <sys/kernel.h> 174 #ifdef TCP_SIGNATURE 175 #include <sys/md5.h> 176 #endif 177 #include <sys/lwp.h> /* for lwp0 */ 178 179 #include <net/if.h> 180 #include <net/route.h> 181 #include <net/if_types.h> 182 183 #include <netinet/in.h> 184 #include <netinet/in_systm.h> 185 #include <netinet/ip.h> 186 #include <netinet/in_pcb.h> 187 #include <netinet/in_var.h> 188 #include <netinet/ip_var.h> 189 #include <netinet/in_offload.h> 190 191 #ifdef INET6 192 #ifndef INET 193 #include <netinet/in.h> 194 #endif 195 #include <netinet/ip6.h> 196 #include <netinet6/ip6_var.h> 197 #include <netinet6/in6_pcb.h> 198 #include <netinet6/ip6_var.h> 199 #include <netinet6/in6_var.h> 200 #include <netinet/icmp6.h> 201 #include <netinet6/nd6.h> 202 #ifdef TCP_SIGNATURE 203 #include <netinet6/scope6_var.h> 204 #endif 205 #endif 206 207 #ifndef INET6 208 /* always need ip6.h for IP6_EXTHDR_GET */ 209 #include <netinet/ip6.h> 210 #endif 211 212 #include <netinet/tcp.h> 213 #include <netinet/tcp_fsm.h> 214 #include <netinet/tcp_seq.h> 215 #include <netinet/tcp_timer.h> 216 #include <netinet/tcp_var.h> 217 #include <netinet/tcp_private.h> 218 #include <netinet/tcpip.h> 219 #include <netinet/tcp_congctl.h> 220 #include <netinet/tcp_debug.h> 221 222 #include <machine/stdarg.h> 223 224 #ifdef IPSEC 225 #include <netinet6/ipsec.h> 226 #include <netinet6/ipsec_private.h> 227 #include <netkey/key.h> 228 #endif /*IPSEC*/ 229 #ifdef INET6 230 #include "faith.h" 231 #if defined(NFAITH) && NFAITH > 0 232 #include <net/if_faith.h> 233 #endif 234 #endif /* IPSEC */ 235 236 #ifdef FAST_IPSEC 237 #include <netipsec/ipsec.h> 238 #include <netipsec/ipsec_var.h> 239 #include <netipsec/ipsec_private.h> 240 #include <netipsec/key.h> 241 #ifdef INET6 242 #include <netipsec/ipsec6.h> 243 #endif 244 #endif /* FAST_IPSEC*/ 245 246 int tcprexmtthresh = 3; 247 int tcp_log_refused; 248 249 int tcp_do_autorcvbuf = 0; 250 int tcp_autorcvbuf_inc = 16 * 1024; 251 int tcp_autorcvbuf_max = 256 * 1024; 252 253 static int tcp_rst_ppslim_count = 0; 254 static struct timeval tcp_rst_ppslim_last; 255 static int tcp_ackdrop_ppslim_count = 0; 256 static struct timeval tcp_ackdrop_ppslim_last; 257 258 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 259 260 /* for modulo comparisons of timestamps */ 261 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 262 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 263 264 /* 265 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 266 */ 267 #ifdef INET6 268 static inline void 269 nd6_hint(struct tcpcb *tp) 270 { 271 struct rtentry *rt; 272 273 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 274 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 275 nd6_nud_hint(rt, NULL, 0); 276 } 277 #else 278 static inline void 279 nd6_hint(struct tcpcb *tp) 280 { 281 } 282 #endif 283 284 /* 285 * Compute ACK transmission behavior. Delay the ACK unless 286 * we have already delayed an ACK (must send an ACK every two segments). 287 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 288 * option is enabled. 289 */ 290 static void 291 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 292 { 293 294 if (tp->t_flags & TF_DELACK || 295 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 296 tp->t_flags |= TF_ACKNOW; 297 else 298 TCP_SET_DELACK(tp); 299 } 300 301 static void 302 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 303 { 304 305 /* 306 * If we had a pending ICMP message that refers to data that have 307 * just been acknowledged, disregard the recorded ICMP message. 308 */ 309 if ((tp->t_flags & TF_PMTUD_PEND) && 310 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 311 tp->t_flags &= ~TF_PMTUD_PEND; 312 313 /* 314 * Keep track of the largest chunk of data 315 * acknowledged since last PMTU update 316 */ 317 if (tp->t_pmtud_mss_acked < acked) 318 tp->t_pmtud_mss_acked = acked; 319 } 320 321 /* 322 * Convert TCP protocol fields to host order for easier processing. 323 */ 324 static void 325 tcp_fields_to_host(struct tcphdr *th) 326 { 327 328 NTOHL(th->th_seq); 329 NTOHL(th->th_ack); 330 NTOHS(th->th_win); 331 NTOHS(th->th_urp); 332 } 333 334 /* 335 * ... and reverse the above. 336 */ 337 static void 338 tcp_fields_to_net(struct tcphdr *th) 339 { 340 341 HTONL(th->th_seq); 342 HTONL(th->th_ack); 343 HTONS(th->th_win); 344 HTONS(th->th_urp); 345 } 346 347 #ifdef TCP_CSUM_COUNTERS 348 #include <sys/device.h> 349 350 #if defined(INET) 351 extern struct evcnt tcp_hwcsum_ok; 352 extern struct evcnt tcp_hwcsum_bad; 353 extern struct evcnt tcp_hwcsum_data; 354 extern struct evcnt tcp_swcsum; 355 #endif /* defined(INET) */ 356 #if defined(INET6) 357 extern struct evcnt tcp6_hwcsum_ok; 358 extern struct evcnt tcp6_hwcsum_bad; 359 extern struct evcnt tcp6_hwcsum_data; 360 extern struct evcnt tcp6_swcsum; 361 #endif /* defined(INET6) */ 362 363 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 364 365 #else 366 367 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 368 369 #endif /* TCP_CSUM_COUNTERS */ 370 371 #ifdef TCP_REASS_COUNTERS 372 #include <sys/device.h> 373 374 extern struct evcnt tcp_reass_; 375 extern struct evcnt tcp_reass_empty; 376 extern struct evcnt tcp_reass_iteration[8]; 377 extern struct evcnt tcp_reass_prependfirst; 378 extern struct evcnt tcp_reass_prepend; 379 extern struct evcnt tcp_reass_insert; 380 extern struct evcnt tcp_reass_inserttail; 381 extern struct evcnt tcp_reass_append; 382 extern struct evcnt tcp_reass_appendtail; 383 extern struct evcnt tcp_reass_overlaptail; 384 extern struct evcnt tcp_reass_overlapfront; 385 extern struct evcnt tcp_reass_segdup; 386 extern struct evcnt tcp_reass_fragdup; 387 388 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 389 390 #else 391 392 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 393 394 #endif /* TCP_REASS_COUNTERS */ 395 396 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 397 int *); 398 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 399 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 400 401 #ifdef INET 402 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 403 #endif 404 #ifdef INET6 405 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 406 #endif 407 408 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 409 410 #if defined(MBUFTRACE) 411 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 412 #endif /* defined(MBUFTRACE) */ 413 414 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 415 NULL, IPL_VM); 416 417 struct ipqent * 418 tcpipqent_alloc(void) 419 { 420 struct ipqent *ipqe; 421 int s; 422 423 s = splvm(); 424 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 425 splx(s); 426 427 return ipqe; 428 } 429 430 void 431 tcpipqent_free(struct ipqent *ipqe) 432 { 433 int s; 434 435 s = splvm(); 436 pool_put(&tcpipqent_pool, ipqe); 437 splx(s); 438 } 439 440 static int 441 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 442 { 443 struct ipqent *p, *q, *nq, *tiqe = NULL; 444 struct socket *so = NULL; 445 int pkt_flags; 446 tcp_seq pkt_seq; 447 unsigned pkt_len; 448 u_long rcvpartdupbyte = 0; 449 u_long rcvoobyte; 450 #ifdef TCP_REASS_COUNTERS 451 u_int count = 0; 452 #endif 453 uint64_t *tcps; 454 455 if (tp->t_inpcb) 456 so = tp->t_inpcb->inp_socket; 457 #ifdef INET6 458 else if (tp->t_in6pcb) 459 so = tp->t_in6pcb->in6p_socket; 460 #endif 461 462 TCP_REASS_LOCK_CHECK(tp); 463 464 /* 465 * Call with th==0 after become established to 466 * force pre-ESTABLISHED data up to user socket. 467 */ 468 if (th == 0) 469 goto present; 470 471 m_claimm(m, &tcp_reass_mowner); 472 473 rcvoobyte = *tlen; 474 /* 475 * Copy these to local variables because the tcpiphdr 476 * gets munged while we are collapsing mbufs. 477 */ 478 pkt_seq = th->th_seq; 479 pkt_len = *tlen; 480 pkt_flags = th->th_flags; 481 482 TCP_REASS_COUNTER_INCR(&tcp_reass_); 483 484 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 485 /* 486 * When we miss a packet, the vast majority of time we get 487 * packets that follow it in order. So optimize for that. 488 */ 489 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 490 p->ipqe_len += pkt_len; 491 p->ipqe_flags |= pkt_flags; 492 m_cat(p->ipre_mlast, m); 493 TRAVERSE(p->ipre_mlast); 494 m = NULL; 495 tiqe = p; 496 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 497 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 498 goto skip_replacement; 499 } 500 /* 501 * While we're here, if the pkt is completely beyond 502 * anything we have, just insert it at the tail. 503 */ 504 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 505 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 506 goto insert_it; 507 } 508 } 509 510 q = TAILQ_FIRST(&tp->segq); 511 512 if (q != NULL) { 513 /* 514 * If this segment immediately precedes the first out-of-order 515 * block, simply slap the segment in front of it and (mostly) 516 * skip the complicated logic. 517 */ 518 if (pkt_seq + pkt_len == q->ipqe_seq) { 519 q->ipqe_seq = pkt_seq; 520 q->ipqe_len += pkt_len; 521 q->ipqe_flags |= pkt_flags; 522 m_cat(m, q->ipqe_m); 523 q->ipqe_m = m; 524 q->ipre_mlast = m; /* last mbuf may have changed */ 525 TRAVERSE(q->ipre_mlast); 526 tiqe = q; 527 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 528 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 529 goto skip_replacement; 530 } 531 } else { 532 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 533 } 534 535 /* 536 * Find a segment which begins after this one does. 537 */ 538 for (p = NULL; q != NULL; q = nq) { 539 nq = TAILQ_NEXT(q, ipqe_q); 540 #ifdef TCP_REASS_COUNTERS 541 count++; 542 #endif 543 /* 544 * If the received segment is just right after this 545 * fragment, merge the two together and then check 546 * for further overlaps. 547 */ 548 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 549 #ifdef TCPREASS_DEBUG 550 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 551 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 552 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 553 #endif 554 pkt_len += q->ipqe_len; 555 pkt_flags |= q->ipqe_flags; 556 pkt_seq = q->ipqe_seq; 557 m_cat(q->ipre_mlast, m); 558 TRAVERSE(q->ipre_mlast); 559 m = q->ipqe_m; 560 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 561 goto free_ipqe; 562 } 563 /* 564 * If the received segment is completely past this 565 * fragment, we need to go the next fragment. 566 */ 567 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 568 p = q; 569 continue; 570 } 571 /* 572 * If the fragment is past the received segment, 573 * it (or any following) can't be concatenated. 574 */ 575 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 576 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 577 break; 578 } 579 580 /* 581 * We've received all the data in this segment before. 582 * mark it as a duplicate and return. 583 */ 584 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 585 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 586 tcps = TCP_STAT_GETREF(); 587 tcps[TCP_STAT_RCVDUPPACK]++; 588 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 589 TCP_STAT_PUTREF(); 590 tcp_new_dsack(tp, pkt_seq, pkt_len); 591 m_freem(m); 592 if (tiqe != NULL) { 593 tcpipqent_free(tiqe); 594 } 595 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 596 return (0); 597 } 598 /* 599 * Received segment completely overlaps this fragment 600 * so we drop the fragment (this keeps the temporal 601 * ordering of segments correct). 602 */ 603 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 604 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 605 rcvpartdupbyte += q->ipqe_len; 606 m_freem(q->ipqe_m); 607 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 608 goto free_ipqe; 609 } 610 /* 611 * RX'ed segment extends past the end of the 612 * fragment. Drop the overlapping bytes. Then 613 * merge the fragment and segment then treat as 614 * a longer received packet. 615 */ 616 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 617 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 618 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 619 #ifdef TCPREASS_DEBUG 620 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 621 tp, overlap, 622 pkt_seq, pkt_seq + pkt_len, pkt_len); 623 #endif 624 m_adj(m, overlap); 625 rcvpartdupbyte += overlap; 626 m_cat(q->ipre_mlast, m); 627 TRAVERSE(q->ipre_mlast); 628 m = q->ipqe_m; 629 pkt_seq = q->ipqe_seq; 630 pkt_len += q->ipqe_len - overlap; 631 rcvoobyte -= overlap; 632 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 633 goto free_ipqe; 634 } 635 /* 636 * RX'ed segment extends past the front of the 637 * fragment. Drop the overlapping bytes on the 638 * received packet. The packet will then be 639 * contatentated with this fragment a bit later. 640 */ 641 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 642 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 643 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 644 #ifdef TCPREASS_DEBUG 645 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 646 tp, overlap, 647 pkt_seq, pkt_seq + pkt_len, pkt_len); 648 #endif 649 m_adj(m, -overlap); 650 pkt_len -= overlap; 651 rcvpartdupbyte += overlap; 652 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 653 rcvoobyte -= overlap; 654 } 655 /* 656 * If the received segment immediates precedes this 657 * fragment then tack the fragment onto this segment 658 * and reinsert the data. 659 */ 660 if (q->ipqe_seq == pkt_seq + pkt_len) { 661 #ifdef TCPREASS_DEBUG 662 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 663 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 664 pkt_seq, pkt_seq + pkt_len, pkt_len); 665 #endif 666 pkt_len += q->ipqe_len; 667 pkt_flags |= q->ipqe_flags; 668 m_cat(m, q->ipqe_m); 669 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 670 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 671 tp->t_segqlen--; 672 KASSERT(tp->t_segqlen >= 0); 673 KASSERT(tp->t_segqlen != 0 || 674 (TAILQ_EMPTY(&tp->segq) && 675 TAILQ_EMPTY(&tp->timeq))); 676 if (tiqe == NULL) { 677 tiqe = q; 678 } else { 679 tcpipqent_free(q); 680 } 681 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 682 break; 683 } 684 /* 685 * If the fragment is before the segment, remember it. 686 * When this loop is terminated, p will contain the 687 * pointer to fragment that is right before the received 688 * segment. 689 */ 690 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 691 p = q; 692 693 continue; 694 695 /* 696 * This is a common operation. It also will allow 697 * to save doing a malloc/free in most instances. 698 */ 699 free_ipqe: 700 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 701 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 702 tp->t_segqlen--; 703 KASSERT(tp->t_segqlen >= 0); 704 KASSERT(tp->t_segqlen != 0 || 705 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 706 if (tiqe == NULL) { 707 tiqe = q; 708 } else { 709 tcpipqent_free(q); 710 } 711 } 712 713 #ifdef TCP_REASS_COUNTERS 714 if (count > 7) 715 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 716 else if (count > 0) 717 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 718 #endif 719 720 insert_it: 721 722 /* 723 * Allocate a new queue entry since the received segment did not 724 * collapse onto any other out-of-order block; thus we are allocating 725 * a new block. If it had collapsed, tiqe would not be NULL and 726 * we would be reusing it. 727 * XXX If we can't, just drop the packet. XXX 728 */ 729 if (tiqe == NULL) { 730 tiqe = tcpipqent_alloc(); 731 if (tiqe == NULL) { 732 TCP_STATINC(TCP_STAT_RCVMEMDROP); 733 m_freem(m); 734 return (0); 735 } 736 } 737 738 /* 739 * Update the counters. 740 */ 741 tcps = TCP_STAT_GETREF(); 742 tcps[TCP_STAT_RCVOOPACK]++; 743 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 744 if (rcvpartdupbyte) { 745 tcps[TCP_STAT_RCVPARTDUPPACK]++; 746 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 747 } 748 TCP_STAT_PUTREF(); 749 750 /* 751 * Insert the new fragment queue entry into both queues. 752 */ 753 tiqe->ipqe_m = m; 754 tiqe->ipre_mlast = m; 755 tiqe->ipqe_seq = pkt_seq; 756 tiqe->ipqe_len = pkt_len; 757 tiqe->ipqe_flags = pkt_flags; 758 if (p == NULL) { 759 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 760 #ifdef TCPREASS_DEBUG 761 if (tiqe->ipqe_seq != tp->rcv_nxt) 762 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 763 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 764 #endif 765 } else { 766 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 767 #ifdef TCPREASS_DEBUG 768 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 769 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 770 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 771 #endif 772 } 773 tp->t_segqlen++; 774 775 skip_replacement: 776 777 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 778 779 present: 780 /* 781 * Present data to user, advancing rcv_nxt through 782 * completed sequence space. 783 */ 784 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 785 return (0); 786 q = TAILQ_FIRST(&tp->segq); 787 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 788 return (0); 789 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 790 return (0); 791 792 tp->rcv_nxt += q->ipqe_len; 793 pkt_flags = q->ipqe_flags & TH_FIN; 794 nd6_hint(tp); 795 796 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 797 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 798 tp->t_segqlen--; 799 KASSERT(tp->t_segqlen >= 0); 800 KASSERT(tp->t_segqlen != 0 || 801 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 802 if (so->so_state & SS_CANTRCVMORE) 803 m_freem(q->ipqe_m); 804 else 805 sbappendstream(&so->so_rcv, q->ipqe_m); 806 tcpipqent_free(q); 807 sorwakeup(so); 808 return (pkt_flags); 809 } 810 811 #ifdef INET6 812 int 813 tcp6_input(struct mbuf **mp, int *offp, int proto) 814 { 815 struct mbuf *m = *mp; 816 817 /* 818 * draft-itojun-ipv6-tcp-to-anycast 819 * better place to put this in? 820 */ 821 if (m->m_flags & M_ANYCAST6) { 822 struct ip6_hdr *ip6; 823 if (m->m_len < sizeof(struct ip6_hdr)) { 824 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 825 TCP_STATINC(TCP_STAT_RCVSHORT); 826 return IPPROTO_DONE; 827 } 828 } 829 ip6 = mtod(m, struct ip6_hdr *); 830 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 831 (char *)&ip6->ip6_dst - (char *)ip6); 832 return IPPROTO_DONE; 833 } 834 835 tcp_input(m, *offp, proto); 836 return IPPROTO_DONE; 837 } 838 #endif 839 840 #ifdef INET 841 static void 842 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 843 { 844 char src[4*sizeof "123"]; 845 char dst[4*sizeof "123"]; 846 847 if (ip) { 848 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 849 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 850 } 851 else { 852 strlcpy(src, "(unknown)", sizeof(src)); 853 strlcpy(dst, "(unknown)", sizeof(dst)); 854 } 855 log(LOG_INFO, 856 "Connection attempt to TCP %s:%d from %s:%d\n", 857 dst, ntohs(th->th_dport), 858 src, ntohs(th->th_sport)); 859 } 860 #endif 861 862 #ifdef INET6 863 static void 864 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 865 { 866 char src[INET6_ADDRSTRLEN]; 867 char dst[INET6_ADDRSTRLEN]; 868 869 if (ip6) { 870 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 871 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 872 } 873 else { 874 strlcpy(src, "(unknown v6)", sizeof(src)); 875 strlcpy(dst, "(unknown v6)", sizeof(dst)); 876 } 877 log(LOG_INFO, 878 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 879 dst, ntohs(th->th_dport), 880 src, ntohs(th->th_sport)); 881 } 882 #endif 883 884 /* 885 * Checksum extended TCP header and data. 886 */ 887 int 888 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 889 int toff, int off, int tlen) 890 { 891 892 /* 893 * XXX it's better to record and check if this mbuf is 894 * already checked. 895 */ 896 897 switch (af) { 898 #ifdef INET 899 case AF_INET: 900 switch (m->m_pkthdr.csum_flags & 901 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 902 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 903 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 904 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 905 goto badcsum; 906 907 case M_CSUM_TCPv4|M_CSUM_DATA: { 908 u_int32_t hw_csum = m->m_pkthdr.csum_data; 909 910 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 911 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 912 const struct ip *ip = 913 mtod(m, const struct ip *); 914 915 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 916 ip->ip_dst.s_addr, 917 htons(hw_csum + tlen + off + IPPROTO_TCP)); 918 } 919 if ((hw_csum ^ 0xffff) != 0) 920 goto badcsum; 921 break; 922 } 923 924 case M_CSUM_TCPv4: 925 /* Checksum was okay. */ 926 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 927 break; 928 929 default: 930 /* 931 * Must compute it ourselves. Maybe skip checksum 932 * on loopback interfaces. 933 */ 934 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 935 IFF_LOOPBACK) || 936 tcp_do_loopback_cksum)) { 937 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 938 if (in4_cksum(m, IPPROTO_TCP, toff, 939 tlen + off) != 0) 940 goto badcsum; 941 } 942 break; 943 } 944 break; 945 #endif /* INET4 */ 946 947 #ifdef INET6 948 case AF_INET6: 949 switch (m->m_pkthdr.csum_flags & 950 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 951 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 952 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 953 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 954 goto badcsum; 955 956 #if 0 /* notyet */ 957 case M_CSUM_TCPv6|M_CSUM_DATA: 958 #endif 959 960 case M_CSUM_TCPv6: 961 /* Checksum was okay. */ 962 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 963 break; 964 965 default: 966 /* 967 * Must compute it ourselves. Maybe skip checksum 968 * on loopback interfaces. 969 */ 970 if (__predict_true((m->m_flags & M_LOOP) == 0 || 971 tcp_do_loopback_cksum)) { 972 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 973 if (in6_cksum(m, IPPROTO_TCP, toff, 974 tlen + off) != 0) 975 goto badcsum; 976 } 977 } 978 break; 979 #endif /* INET6 */ 980 } 981 982 return 0; 983 984 badcsum: 985 TCP_STATINC(TCP_STAT_RCVBADSUM); 986 return -1; 987 } 988 989 /* 990 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 991 */ 992 void 993 tcp_input(struct mbuf *m, ...) 994 { 995 struct tcphdr *th; 996 struct ip *ip; 997 struct inpcb *inp; 998 #ifdef INET6 999 struct ip6_hdr *ip6; 1000 struct in6pcb *in6p; 1001 #endif 1002 u_int8_t *optp = NULL; 1003 int optlen = 0; 1004 int len, tlen, toff, hdroptlen = 0; 1005 struct tcpcb *tp = 0; 1006 int tiflags; 1007 struct socket *so = NULL; 1008 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 1009 #ifdef TCP_DEBUG 1010 short ostate = 0; 1011 #endif 1012 u_long tiwin; 1013 struct tcp_opt_info opti; 1014 int off, iphlen; 1015 va_list ap; 1016 int af; /* af on the wire */ 1017 struct mbuf *tcp_saveti = NULL; 1018 uint32_t ts_rtt; 1019 uint8_t iptos; 1020 uint64_t *tcps; 1021 1022 MCLAIM(m, &tcp_rx_mowner); 1023 va_start(ap, m); 1024 toff = va_arg(ap, int); 1025 (void)va_arg(ap, int); /* ignore value, advance ap */ 1026 va_end(ap); 1027 1028 TCP_STATINC(TCP_STAT_RCVTOTAL); 1029 1030 bzero(&opti, sizeof(opti)); 1031 opti.ts_present = 0; 1032 opti.maxseg = 0; 1033 1034 /* 1035 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1036 * 1037 * TCP is, by definition, unicast, so we reject all 1038 * multicast outright. 1039 * 1040 * Note, there are additional src/dst address checks in 1041 * the AF-specific code below. 1042 */ 1043 if (m->m_flags & (M_BCAST|M_MCAST)) { 1044 /* XXX stat */ 1045 goto drop; 1046 } 1047 #ifdef INET6 1048 if (m->m_flags & M_ANYCAST6) { 1049 /* XXX stat */ 1050 goto drop; 1051 } 1052 #endif 1053 1054 /* 1055 * Get IP and TCP header. 1056 * Note: IP leaves IP header in first mbuf. 1057 */ 1058 ip = mtod(m, struct ip *); 1059 #ifdef INET6 1060 ip6 = NULL; 1061 #endif 1062 switch (ip->ip_v) { 1063 #ifdef INET 1064 case 4: 1065 af = AF_INET; 1066 iphlen = sizeof(struct ip); 1067 ip = mtod(m, struct ip *); 1068 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1069 sizeof(struct tcphdr)); 1070 if (th == NULL) { 1071 TCP_STATINC(TCP_STAT_RCVSHORT); 1072 return; 1073 } 1074 /* We do the checksum after PCB lookup... */ 1075 len = ntohs(ip->ip_len); 1076 tlen = len - toff; 1077 iptos = ip->ip_tos; 1078 break; 1079 #endif 1080 #ifdef INET6 1081 case 6: 1082 ip = NULL; 1083 iphlen = sizeof(struct ip6_hdr); 1084 af = AF_INET6; 1085 ip6 = mtod(m, struct ip6_hdr *); 1086 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1087 sizeof(struct tcphdr)); 1088 if (th == NULL) { 1089 TCP_STATINC(TCP_STAT_RCVSHORT); 1090 return; 1091 } 1092 1093 /* Be proactive about malicious use of IPv4 mapped address */ 1094 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1095 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1096 /* XXX stat */ 1097 goto drop; 1098 } 1099 1100 /* 1101 * Be proactive about unspecified IPv6 address in source. 1102 * As we use all-zero to indicate unbounded/unconnected pcb, 1103 * unspecified IPv6 address can be used to confuse us. 1104 * 1105 * Note that packets with unspecified IPv6 destination is 1106 * already dropped in ip6_input. 1107 */ 1108 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1109 /* XXX stat */ 1110 goto drop; 1111 } 1112 1113 /* 1114 * Make sure destination address is not multicast. 1115 * Source address checked in ip6_input(). 1116 */ 1117 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1118 /* XXX stat */ 1119 goto drop; 1120 } 1121 1122 /* We do the checksum after PCB lookup... */ 1123 len = m->m_pkthdr.len; 1124 tlen = len - toff; 1125 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1126 break; 1127 #endif 1128 default: 1129 m_freem(m); 1130 return; 1131 } 1132 1133 KASSERT(TCP_HDR_ALIGNED_P(th)); 1134 1135 /* 1136 * Check that TCP offset makes sense, 1137 * pull out TCP options and adjust length. XXX 1138 */ 1139 off = th->th_off << 2; 1140 if (off < sizeof (struct tcphdr) || off > tlen) { 1141 TCP_STATINC(TCP_STAT_RCVBADOFF); 1142 goto drop; 1143 } 1144 tlen -= off; 1145 1146 /* 1147 * tcp_input() has been modified to use tlen to mean the TCP data 1148 * length throughout the function. Other functions can use 1149 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1150 * rja 1151 */ 1152 1153 if (off > sizeof (struct tcphdr)) { 1154 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1155 if (th == NULL) { 1156 TCP_STATINC(TCP_STAT_RCVSHORT); 1157 return; 1158 } 1159 /* 1160 * NOTE: ip/ip6 will not be affected by m_pulldown() 1161 * (as they're before toff) and we don't need to update those. 1162 */ 1163 KASSERT(TCP_HDR_ALIGNED_P(th)); 1164 optlen = off - sizeof (struct tcphdr); 1165 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1166 /* 1167 * Do quick retrieval of timestamp options ("options 1168 * prediction?"). If timestamp is the only option and it's 1169 * formatted as recommended in RFC 1323 appendix A, we 1170 * quickly get the values now and not bother calling 1171 * tcp_dooptions(), etc. 1172 */ 1173 if ((optlen == TCPOLEN_TSTAMP_APPA || 1174 (optlen > TCPOLEN_TSTAMP_APPA && 1175 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1176 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1177 (th->th_flags & TH_SYN) == 0) { 1178 opti.ts_present = 1; 1179 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1180 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1181 optp = NULL; /* we've parsed the options */ 1182 } 1183 } 1184 tiflags = th->th_flags; 1185 1186 /* 1187 * Locate pcb for segment. 1188 */ 1189 findpcb: 1190 inp = NULL; 1191 #ifdef INET6 1192 in6p = NULL; 1193 #endif 1194 switch (af) { 1195 #ifdef INET 1196 case AF_INET: 1197 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1198 ip->ip_dst, th->th_dport); 1199 if (inp == 0) { 1200 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1201 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1202 } 1203 #ifdef INET6 1204 if (inp == 0) { 1205 struct in6_addr s, d; 1206 1207 /* mapped addr case */ 1208 bzero(&s, sizeof(s)); 1209 s.s6_addr16[5] = htons(0xffff); 1210 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1211 bzero(&d, sizeof(d)); 1212 d.s6_addr16[5] = htons(0xffff); 1213 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1214 in6p = in6_pcblookup_connect(&tcbtable, &s, 1215 th->th_sport, &d, th->th_dport, 0); 1216 if (in6p == 0) { 1217 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1218 in6p = in6_pcblookup_bind(&tcbtable, &d, 1219 th->th_dport, 0); 1220 } 1221 } 1222 #endif 1223 #ifndef INET6 1224 if (inp == 0) 1225 #else 1226 if (inp == 0 && in6p == 0) 1227 #endif 1228 { 1229 TCP_STATINC(TCP_STAT_NOPORT); 1230 if (tcp_log_refused && 1231 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1232 tcp4_log_refused(ip, th); 1233 } 1234 tcp_fields_to_host(th); 1235 goto dropwithreset_ratelim; 1236 } 1237 #if defined(IPSEC) || defined(FAST_IPSEC) 1238 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1239 ipsec4_in_reject(m, inp)) { 1240 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1241 goto drop; 1242 } 1243 #ifdef INET6 1244 else if (in6p && 1245 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1246 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1247 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1248 goto drop; 1249 } 1250 #endif 1251 #endif /*IPSEC*/ 1252 break; 1253 #endif /*INET*/ 1254 #ifdef INET6 1255 case AF_INET6: 1256 { 1257 int faith; 1258 1259 #if defined(NFAITH) && NFAITH > 0 1260 faith = faithprefix(&ip6->ip6_dst); 1261 #else 1262 faith = 0; 1263 #endif 1264 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1265 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1266 if (in6p == NULL) { 1267 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1268 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1269 th->th_dport, faith); 1270 } 1271 if (in6p == NULL) { 1272 TCP_STATINC(TCP_STAT_NOPORT); 1273 if (tcp_log_refused && 1274 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1275 tcp6_log_refused(ip6, th); 1276 } 1277 tcp_fields_to_host(th); 1278 goto dropwithreset_ratelim; 1279 } 1280 #if defined(IPSEC) || defined(FAST_IPSEC) 1281 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1282 ipsec6_in_reject(m, in6p)) { 1283 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1284 goto drop; 1285 } 1286 #endif /*IPSEC*/ 1287 break; 1288 } 1289 #endif 1290 } 1291 1292 /* 1293 * If the state is CLOSED (i.e., TCB does not exist) then 1294 * all data in the incoming segment is discarded. 1295 * If the TCB exists but is in CLOSED state, it is embryonic, 1296 * but should either do a listen or a connect soon. 1297 */ 1298 tp = NULL; 1299 so = NULL; 1300 if (inp) { 1301 tp = intotcpcb(inp); 1302 so = inp->inp_socket; 1303 } 1304 #ifdef INET6 1305 else if (in6p) { 1306 tp = in6totcpcb(in6p); 1307 so = in6p->in6p_socket; 1308 } 1309 #endif 1310 if (tp == 0) { 1311 tcp_fields_to_host(th); 1312 goto dropwithreset_ratelim; 1313 } 1314 if (tp->t_state == TCPS_CLOSED) 1315 goto drop; 1316 1317 /* 1318 * Checksum extended TCP header and data. 1319 */ 1320 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1321 goto badcsum; 1322 1323 tcp_fields_to_host(th); 1324 1325 /* Unscale the window into a 32-bit value. */ 1326 if ((tiflags & TH_SYN) == 0) 1327 tiwin = th->th_win << tp->snd_scale; 1328 else 1329 tiwin = th->th_win; 1330 1331 #ifdef INET6 1332 /* save packet options if user wanted */ 1333 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1334 if (in6p->in6p_options) { 1335 m_freem(in6p->in6p_options); 1336 in6p->in6p_options = 0; 1337 } 1338 KASSERT(ip6 != NULL); 1339 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1340 } 1341 #endif 1342 1343 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1344 union syn_cache_sa src; 1345 union syn_cache_sa dst; 1346 1347 bzero(&src, sizeof(src)); 1348 bzero(&dst, sizeof(dst)); 1349 switch (af) { 1350 #ifdef INET 1351 case AF_INET: 1352 src.sin.sin_len = sizeof(struct sockaddr_in); 1353 src.sin.sin_family = AF_INET; 1354 src.sin.sin_addr = ip->ip_src; 1355 src.sin.sin_port = th->th_sport; 1356 1357 dst.sin.sin_len = sizeof(struct sockaddr_in); 1358 dst.sin.sin_family = AF_INET; 1359 dst.sin.sin_addr = ip->ip_dst; 1360 dst.sin.sin_port = th->th_dport; 1361 break; 1362 #endif 1363 #ifdef INET6 1364 case AF_INET6: 1365 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1366 src.sin6.sin6_family = AF_INET6; 1367 src.sin6.sin6_addr = ip6->ip6_src; 1368 src.sin6.sin6_port = th->th_sport; 1369 1370 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1371 dst.sin6.sin6_family = AF_INET6; 1372 dst.sin6.sin6_addr = ip6->ip6_dst; 1373 dst.sin6.sin6_port = th->th_dport; 1374 break; 1375 #endif /* INET6 */ 1376 default: 1377 goto badsyn; /*sanity*/ 1378 } 1379 1380 if (so->so_options & SO_DEBUG) { 1381 #ifdef TCP_DEBUG 1382 ostate = tp->t_state; 1383 #endif 1384 1385 tcp_saveti = NULL; 1386 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1387 goto nosave; 1388 1389 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1390 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1391 if (!tcp_saveti) 1392 goto nosave; 1393 } else { 1394 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1395 if (!tcp_saveti) 1396 goto nosave; 1397 MCLAIM(m, &tcp_mowner); 1398 tcp_saveti->m_len = iphlen; 1399 m_copydata(m, 0, iphlen, 1400 mtod(tcp_saveti, void *)); 1401 } 1402 1403 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1404 m_freem(tcp_saveti); 1405 tcp_saveti = NULL; 1406 } else { 1407 tcp_saveti->m_len += sizeof(struct tcphdr); 1408 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1409 sizeof(struct tcphdr)); 1410 } 1411 nosave:; 1412 } 1413 if (so->so_options & SO_ACCEPTCONN) { 1414 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1415 if (tiflags & TH_RST) { 1416 syn_cache_reset(&src.sa, &dst.sa, th); 1417 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1418 (TH_ACK|TH_SYN)) { 1419 /* 1420 * Received a SYN,ACK. This should 1421 * never happen while we are in 1422 * LISTEN. Send an RST. 1423 */ 1424 goto badsyn; 1425 } else if (tiflags & TH_ACK) { 1426 so = syn_cache_get(&src.sa, &dst.sa, 1427 th, toff, tlen, so, m); 1428 if (so == NULL) { 1429 /* 1430 * We don't have a SYN for 1431 * this ACK; send an RST. 1432 */ 1433 goto badsyn; 1434 } else if (so == 1435 (struct socket *)(-1)) { 1436 /* 1437 * We were unable to create 1438 * the connection. If the 1439 * 3-way handshake was 1440 * completed, and RST has 1441 * been sent to the peer. 1442 * Since the mbuf might be 1443 * in use for the reply, 1444 * do not free it. 1445 */ 1446 m = NULL; 1447 } else { 1448 /* 1449 * We have created a 1450 * full-blown connection. 1451 */ 1452 tp = NULL; 1453 inp = NULL; 1454 #ifdef INET6 1455 in6p = NULL; 1456 #endif 1457 switch (so->so_proto->pr_domain->dom_family) { 1458 #ifdef INET 1459 case AF_INET: 1460 inp = sotoinpcb(so); 1461 tp = intotcpcb(inp); 1462 break; 1463 #endif 1464 #ifdef INET6 1465 case AF_INET6: 1466 in6p = sotoin6pcb(so); 1467 tp = in6totcpcb(in6p); 1468 break; 1469 #endif 1470 } 1471 if (tp == NULL) 1472 goto badsyn; /*XXX*/ 1473 tiwin <<= tp->snd_scale; 1474 goto after_listen; 1475 } 1476 } else { 1477 /* 1478 * None of RST, SYN or ACK was set. 1479 * This is an invalid packet for a 1480 * TCB in LISTEN state. Send a RST. 1481 */ 1482 goto badsyn; 1483 } 1484 } else { 1485 /* 1486 * Received a SYN. 1487 * 1488 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1489 */ 1490 if (m->m_flags & (M_BCAST|M_MCAST)) 1491 goto drop; 1492 1493 switch (af) { 1494 #ifdef INET6 1495 case AF_INET6: 1496 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1497 goto drop; 1498 break; 1499 #endif /* INET6 */ 1500 case AF_INET: 1501 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1502 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1503 goto drop; 1504 break; 1505 } 1506 1507 #ifdef INET6 1508 /* 1509 * If deprecated address is forbidden, we do 1510 * not accept SYN to deprecated interface 1511 * address to prevent any new inbound 1512 * connection from getting established. 1513 * When we do not accept SYN, we send a TCP 1514 * RST, with deprecated source address (instead 1515 * of dropping it). We compromise it as it is 1516 * much better for peer to send a RST, and 1517 * RST will be the final packet for the 1518 * exchange. 1519 * 1520 * If we do not forbid deprecated addresses, we 1521 * accept the SYN packet. RFC2462 does not 1522 * suggest dropping SYN in this case. 1523 * If we decipher RFC2462 5.5.4, it says like 1524 * this: 1525 * 1. use of deprecated addr with existing 1526 * communication is okay - "SHOULD continue 1527 * to be used" 1528 * 2. use of it with new communication: 1529 * (2a) "SHOULD NOT be used if alternate 1530 * address with sufficient scope is 1531 * available" 1532 * (2b) nothing mentioned otherwise. 1533 * Here we fall into (2b) case as we have no 1534 * choice in our source address selection - we 1535 * must obey the peer. 1536 * 1537 * The wording in RFC2462 is confusing, and 1538 * there are multiple description text for 1539 * deprecated address handling - worse, they 1540 * are not exactly the same. I believe 5.5.4 1541 * is the best one, so we follow 5.5.4. 1542 */ 1543 if (af == AF_INET6 && !ip6_use_deprecated) { 1544 struct in6_ifaddr *ia6; 1545 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1546 &ip6->ip6_dst)) && 1547 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1548 tp = NULL; 1549 goto dropwithreset; 1550 } 1551 } 1552 #endif 1553 1554 #if defined(IPSEC) || defined(FAST_IPSEC) 1555 switch (af) { 1556 #ifdef INET 1557 case AF_INET: 1558 if (ipsec4_in_reject_so(m, so)) { 1559 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1560 tp = NULL; 1561 goto dropwithreset; 1562 } 1563 break; 1564 #endif 1565 #ifdef INET6 1566 case AF_INET6: 1567 if (ipsec6_in_reject_so(m, so)) { 1568 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1569 tp = NULL; 1570 goto dropwithreset; 1571 } 1572 break; 1573 #endif /*INET6*/ 1574 } 1575 #endif /*IPSEC*/ 1576 1577 /* 1578 * LISTEN socket received a SYN 1579 * from itself? This can't possibly 1580 * be valid; drop the packet. 1581 */ 1582 if (th->th_sport == th->th_dport) { 1583 int i; 1584 1585 switch (af) { 1586 #ifdef INET 1587 case AF_INET: 1588 i = in_hosteq(ip->ip_src, ip->ip_dst); 1589 break; 1590 #endif 1591 #ifdef INET6 1592 case AF_INET6: 1593 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1594 break; 1595 #endif 1596 default: 1597 i = 1; 1598 } 1599 if (i) { 1600 TCP_STATINC(TCP_STAT_BADSYN); 1601 goto drop; 1602 } 1603 } 1604 1605 /* 1606 * SYN looks ok; create compressed TCP 1607 * state for it. 1608 */ 1609 if (so->so_qlen <= so->so_qlimit && 1610 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1611 so, m, optp, optlen, &opti)) 1612 m = NULL; 1613 } 1614 goto drop; 1615 } 1616 } 1617 1618 after_listen: 1619 #ifdef DIAGNOSTIC 1620 /* 1621 * Should not happen now that all embryonic connections 1622 * are handled with compressed state. 1623 */ 1624 if (tp->t_state == TCPS_LISTEN) 1625 panic("tcp_input: TCPS_LISTEN"); 1626 #endif 1627 1628 /* 1629 * Segment received on connection. 1630 * Reset idle time and keep-alive timer. 1631 */ 1632 tp->t_rcvtime = tcp_now; 1633 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1634 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1635 1636 /* 1637 * Process options. 1638 */ 1639 #ifdef TCP_SIGNATURE 1640 if (optp || (tp->t_flags & TF_SIGNATURE)) 1641 #else 1642 if (optp) 1643 #endif 1644 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1645 goto drop; 1646 1647 if (TCP_SACK_ENABLED(tp)) { 1648 tcp_del_sackholes(tp, th); 1649 } 1650 1651 if (TCP_ECN_ALLOWED(tp)) { 1652 switch (iptos & IPTOS_ECN_MASK) { 1653 case IPTOS_ECN_CE: 1654 tp->t_flags |= TF_ECN_SND_ECE; 1655 TCP_STATINC(TCP_STAT_ECN_CE); 1656 break; 1657 case IPTOS_ECN_ECT0: 1658 TCP_STATINC(TCP_STAT_ECN_ECT); 1659 break; 1660 case IPTOS_ECN_ECT1: 1661 /* XXX: ignore for now -- rpaulo */ 1662 break; 1663 } 1664 1665 if (tiflags & TH_CWR) 1666 tp->t_flags &= ~TF_ECN_SND_ECE; 1667 1668 /* 1669 * Congestion experienced. 1670 * Ignore if we are already trying to recover. 1671 */ 1672 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1673 tp->t_congctl->cong_exp(tp); 1674 } 1675 1676 if (opti.ts_present && opti.ts_ecr) { 1677 /* 1678 * Calculate the RTT from the returned time stamp and the 1679 * connection's time base. If the time stamp is later than 1680 * the current time, or is extremely old, fall back to non-1323 1681 * RTT calculation. Since ts_ecr is unsigned, we can test both 1682 * at the same time. 1683 */ 1684 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1685 if (ts_rtt > TCP_PAWS_IDLE) 1686 ts_rtt = 0; 1687 } else { 1688 ts_rtt = 0; 1689 } 1690 1691 /* 1692 * Header prediction: check for the two common cases 1693 * of a uni-directional data xfer. If the packet has 1694 * no control flags, is in-sequence, the window didn't 1695 * change and we're not retransmitting, it's a 1696 * candidate. If the length is zero and the ack moved 1697 * forward, we're the sender side of the xfer. Just 1698 * free the data acked & wake any higher level process 1699 * that was blocked waiting for space. If the length 1700 * is non-zero and the ack didn't move, we're the 1701 * receiver side. If we're getting packets in-order 1702 * (the reassembly queue is empty), add the data to 1703 * the socket buffer and note that we need a delayed ack. 1704 */ 1705 if (tp->t_state == TCPS_ESTABLISHED && 1706 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1707 == TH_ACK && 1708 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1709 th->th_seq == tp->rcv_nxt && 1710 tiwin && tiwin == tp->snd_wnd && 1711 tp->snd_nxt == tp->snd_max) { 1712 1713 /* 1714 * If last ACK falls within this segment's sequence numbers, 1715 * record the timestamp. 1716 * NOTE that the test is modified according to the latest 1717 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1718 * 1719 * note that we already know 1720 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1721 */ 1722 if (opti.ts_present && 1723 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1724 tp->ts_recent_age = tcp_now; 1725 tp->ts_recent = opti.ts_val; 1726 } 1727 1728 if (tlen == 0) { 1729 /* Ack prediction. */ 1730 if (SEQ_GT(th->th_ack, tp->snd_una) && 1731 SEQ_LEQ(th->th_ack, tp->snd_max) && 1732 tp->snd_cwnd >= tp->snd_wnd && 1733 tp->t_partialacks < 0) { 1734 /* 1735 * this is a pure ack for outstanding data. 1736 */ 1737 if (ts_rtt) 1738 tcp_xmit_timer(tp, ts_rtt); 1739 else if (tp->t_rtttime && 1740 SEQ_GT(th->th_ack, tp->t_rtseq)) 1741 tcp_xmit_timer(tp, 1742 tcp_now - tp->t_rtttime); 1743 acked = th->th_ack - tp->snd_una; 1744 tcps = TCP_STAT_GETREF(); 1745 tcps[TCP_STAT_PREDACK]++; 1746 tcps[TCP_STAT_RCVACKPACK]++; 1747 tcps[TCP_STAT_RCVACKBYTE] += acked; 1748 TCP_STAT_PUTREF(); 1749 nd6_hint(tp); 1750 1751 if (acked > (tp->t_lastoff - tp->t_inoff)) 1752 tp->t_lastm = NULL; 1753 sbdrop(&so->so_snd, acked); 1754 tp->t_lastoff -= acked; 1755 1756 icmp_check(tp, th, acked); 1757 1758 tp->snd_una = th->th_ack; 1759 tp->snd_fack = tp->snd_una; 1760 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1761 tp->snd_high = tp->snd_una; 1762 m_freem(m); 1763 1764 /* 1765 * If all outstanding data are acked, stop 1766 * retransmit timer, otherwise restart timer 1767 * using current (possibly backed-off) value. 1768 * If process is waiting for space, 1769 * wakeup/selnotify/signal. If data 1770 * are ready to send, let tcp_output 1771 * decide between more output or persist. 1772 */ 1773 if (tp->snd_una == tp->snd_max) 1774 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1775 else if (TCP_TIMER_ISARMED(tp, 1776 TCPT_PERSIST) == 0) 1777 TCP_TIMER_ARM(tp, TCPT_REXMT, 1778 tp->t_rxtcur); 1779 1780 sowwakeup(so); 1781 if (so->so_snd.sb_cc) 1782 (void) tcp_output(tp); 1783 if (tcp_saveti) 1784 m_freem(tcp_saveti); 1785 return; 1786 } 1787 } else if (th->th_ack == tp->snd_una && 1788 TAILQ_FIRST(&tp->segq) == NULL && 1789 tlen <= sbspace(&so->so_rcv)) { 1790 int newsize = 0; /* automatic sockbuf scaling */ 1791 1792 /* 1793 * this is a pure, in-sequence data packet 1794 * with nothing on the reassembly queue and 1795 * we have enough buffer space to take it. 1796 */ 1797 tp->rcv_nxt += tlen; 1798 tcps = TCP_STAT_GETREF(); 1799 tcps[TCP_STAT_PREDDAT]++; 1800 tcps[TCP_STAT_RCVPACK]++; 1801 tcps[TCP_STAT_RCVBYTE] += tlen; 1802 TCP_STAT_PUTREF(); 1803 nd6_hint(tp); 1804 1805 /* 1806 * Automatic sizing enables the performance of large buffers 1807 * and most of the efficiency of small ones by only allocating 1808 * space when it is needed. 1809 * 1810 * On the receive side the socket buffer memory is only rarely 1811 * used to any significant extent. This allows us to be much 1812 * more aggressive in scaling the receive socket buffer. For 1813 * the case that the buffer space is actually used to a large 1814 * extent and we run out of kernel memory we can simply drop 1815 * the new segments; TCP on the sender will just retransmit it 1816 * later. Setting the buffer size too big may only consume too 1817 * much kernel memory if the application doesn't read() from 1818 * the socket or packet loss or reordering makes use of the 1819 * reassembly queue. 1820 * 1821 * The criteria to step up the receive buffer one notch are: 1822 * 1. the number of bytes received during the time it takes 1823 * one timestamp to be reflected back to us (the RTT); 1824 * 2. received bytes per RTT is within seven eighth of the 1825 * current socket buffer size; 1826 * 3. receive buffer size has not hit maximal automatic size; 1827 * 1828 * This algorithm does one step per RTT at most and only if 1829 * we receive a bulk stream w/o packet losses or reorderings. 1830 * Shrinking the buffer during idle times is not necessary as 1831 * it doesn't consume any memory when idle. 1832 * 1833 * TODO: Only step up if the application is actually serving 1834 * the buffer to better manage the socket buffer resources. 1835 */ 1836 if (tcp_do_autorcvbuf && 1837 opti.ts_ecr && 1838 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1839 if (opti.ts_ecr > tp->rfbuf_ts && 1840 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 1841 if (tp->rfbuf_cnt > 1842 (so->so_rcv.sb_hiwat / 8 * 7) && 1843 so->so_rcv.sb_hiwat < 1844 tcp_autorcvbuf_max) { 1845 newsize = 1846 min(so->so_rcv.sb_hiwat + 1847 tcp_autorcvbuf_inc, 1848 tcp_autorcvbuf_max); 1849 } 1850 /* Start over with next RTT. */ 1851 tp->rfbuf_ts = 0; 1852 tp->rfbuf_cnt = 0; 1853 } else 1854 tp->rfbuf_cnt += tlen; /* add up */ 1855 } 1856 1857 /* 1858 * Drop TCP, IP headers and TCP options then add data 1859 * to socket buffer. 1860 */ 1861 if (so->so_state & SS_CANTRCVMORE) 1862 m_freem(m); 1863 else { 1864 /* 1865 * Set new socket buffer size. 1866 * Give up when limit is reached. 1867 */ 1868 if (newsize) 1869 if (!sbreserve(&so->so_rcv, 1870 newsize, so)) 1871 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1872 m_adj(m, toff + off); 1873 sbappendstream(&so->so_rcv, m); 1874 } 1875 sorwakeup(so); 1876 tcp_setup_ack(tp, th); 1877 if (tp->t_flags & TF_ACKNOW) 1878 (void) tcp_output(tp); 1879 if (tcp_saveti) 1880 m_freem(tcp_saveti); 1881 return; 1882 } 1883 } 1884 1885 /* 1886 * Compute mbuf offset to TCP data segment. 1887 */ 1888 hdroptlen = toff + off; 1889 1890 /* 1891 * Calculate amount of space in receive window, 1892 * and then do TCP input processing. 1893 * Receive window is amount of space in rcv queue, 1894 * but not less than advertised window. 1895 */ 1896 { int win; 1897 1898 win = sbspace(&so->so_rcv); 1899 if (win < 0) 1900 win = 0; 1901 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1902 } 1903 1904 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1905 tp->rfbuf_ts = 0; 1906 tp->rfbuf_cnt = 0; 1907 1908 switch (tp->t_state) { 1909 /* 1910 * If the state is SYN_SENT: 1911 * if seg contains an ACK, but not for our SYN, drop the input. 1912 * if seg contains a RST, then drop the connection. 1913 * if seg does not contain SYN, then drop it. 1914 * Otherwise this is an acceptable SYN segment 1915 * initialize tp->rcv_nxt and tp->irs 1916 * if seg contains ack then advance tp->snd_una 1917 * if seg contains a ECE and ECN support is enabled, the stream 1918 * is ECN capable. 1919 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1920 * arrange for segment to be acked (eventually) 1921 * continue processing rest of data/controls, beginning with URG 1922 */ 1923 case TCPS_SYN_SENT: 1924 if ((tiflags & TH_ACK) && 1925 (SEQ_LEQ(th->th_ack, tp->iss) || 1926 SEQ_GT(th->th_ack, tp->snd_max))) 1927 goto dropwithreset; 1928 if (tiflags & TH_RST) { 1929 if (tiflags & TH_ACK) 1930 tp = tcp_drop(tp, ECONNREFUSED); 1931 goto drop; 1932 } 1933 if ((tiflags & TH_SYN) == 0) 1934 goto drop; 1935 if (tiflags & TH_ACK) { 1936 tp->snd_una = th->th_ack; 1937 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1938 tp->snd_nxt = tp->snd_una; 1939 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1940 tp->snd_high = tp->snd_una; 1941 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1942 1943 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1944 tp->t_flags |= TF_ECN_PERMIT; 1945 TCP_STATINC(TCP_STAT_ECN_SHS); 1946 } 1947 1948 } 1949 tp->irs = th->th_seq; 1950 tcp_rcvseqinit(tp); 1951 tp->t_flags |= TF_ACKNOW; 1952 tcp_mss_from_peer(tp, opti.maxseg); 1953 1954 /* 1955 * Initialize the initial congestion window. If we 1956 * had to retransmit the SYN, we must initialize cwnd 1957 * to 1 segment (i.e. the Loss Window). 1958 */ 1959 if (tp->t_flags & TF_SYN_REXMT) 1960 tp->snd_cwnd = tp->t_peermss; 1961 else { 1962 int ss = tcp_init_win; 1963 #ifdef INET 1964 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1965 ss = tcp_init_win_local; 1966 #endif 1967 #ifdef INET6 1968 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1969 ss = tcp_init_win_local; 1970 #endif 1971 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1972 } 1973 1974 tcp_rmx_rtt(tp); 1975 if (tiflags & TH_ACK) { 1976 TCP_STATINC(TCP_STAT_CONNECTS); 1977 soisconnected(so); 1978 tcp_established(tp); 1979 /* Do window scaling on this connection? */ 1980 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1981 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1982 tp->snd_scale = tp->requested_s_scale; 1983 tp->rcv_scale = tp->request_r_scale; 1984 } 1985 TCP_REASS_LOCK(tp); 1986 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1987 TCP_REASS_UNLOCK(tp); 1988 /* 1989 * if we didn't have to retransmit the SYN, 1990 * use its rtt as our initial srtt & rtt var. 1991 */ 1992 if (tp->t_rtttime) 1993 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1994 } else 1995 tp->t_state = TCPS_SYN_RECEIVED; 1996 1997 /* 1998 * Advance th->th_seq to correspond to first data byte. 1999 * If data, trim to stay within window, 2000 * dropping FIN if necessary. 2001 */ 2002 th->th_seq++; 2003 if (tlen > tp->rcv_wnd) { 2004 todrop = tlen - tp->rcv_wnd; 2005 m_adj(m, -todrop); 2006 tlen = tp->rcv_wnd; 2007 tiflags &= ~TH_FIN; 2008 tcps = TCP_STAT_GETREF(); 2009 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2010 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2011 TCP_STAT_PUTREF(); 2012 } 2013 tp->snd_wl1 = th->th_seq - 1; 2014 tp->rcv_up = th->th_seq; 2015 goto step6; 2016 2017 /* 2018 * If the state is SYN_RECEIVED: 2019 * If seg contains an ACK, but not for our SYN, drop the input 2020 * and generate an RST. See page 36, rfc793 2021 */ 2022 case TCPS_SYN_RECEIVED: 2023 if ((tiflags & TH_ACK) && 2024 (SEQ_LEQ(th->th_ack, tp->iss) || 2025 SEQ_GT(th->th_ack, tp->snd_max))) 2026 goto dropwithreset; 2027 break; 2028 } 2029 2030 /* 2031 * States other than LISTEN or SYN_SENT. 2032 * First check timestamp, if present. 2033 * Then check that at least some bytes of segment are within 2034 * receive window. If segment begins before rcv_nxt, 2035 * drop leading data (and SYN); if nothing left, just ack. 2036 * 2037 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2038 * and it's less than ts_recent, drop it. 2039 */ 2040 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2041 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2042 2043 /* Check to see if ts_recent is over 24 days old. */ 2044 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2045 /* 2046 * Invalidate ts_recent. If this segment updates 2047 * ts_recent, the age will be reset later and ts_recent 2048 * will get a valid value. If it does not, setting 2049 * ts_recent to zero will at least satisfy the 2050 * requirement that zero be placed in the timestamp 2051 * echo reply when ts_recent isn't valid. The 2052 * age isn't reset until we get a valid ts_recent 2053 * because we don't want out-of-order segments to be 2054 * dropped when ts_recent is old. 2055 */ 2056 tp->ts_recent = 0; 2057 } else { 2058 tcps = TCP_STAT_GETREF(); 2059 tcps[TCP_STAT_RCVDUPPACK]++; 2060 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2061 tcps[TCP_STAT_PAWSDROP]++; 2062 TCP_STAT_PUTREF(); 2063 tcp_new_dsack(tp, th->th_seq, tlen); 2064 goto dropafterack; 2065 } 2066 } 2067 2068 todrop = tp->rcv_nxt - th->th_seq; 2069 dupseg = false; 2070 if (todrop > 0) { 2071 if (tiflags & TH_SYN) { 2072 tiflags &= ~TH_SYN; 2073 th->th_seq++; 2074 if (th->th_urp > 1) 2075 th->th_urp--; 2076 else { 2077 tiflags &= ~TH_URG; 2078 th->th_urp = 0; 2079 } 2080 todrop--; 2081 } 2082 if (todrop > tlen || 2083 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2084 /* 2085 * Any valid FIN or RST must be to the left of the 2086 * window. At this point the FIN or RST must be a 2087 * duplicate or out of sequence; drop it. 2088 */ 2089 if (tiflags & TH_RST) 2090 goto drop; 2091 tiflags &= ~(TH_FIN|TH_RST); 2092 /* 2093 * Send an ACK to resynchronize and drop any data. 2094 * But keep on processing for RST or ACK. 2095 */ 2096 tp->t_flags |= TF_ACKNOW; 2097 todrop = tlen; 2098 dupseg = true; 2099 tcps = TCP_STAT_GETREF(); 2100 tcps[TCP_STAT_RCVDUPPACK]++; 2101 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2102 TCP_STAT_PUTREF(); 2103 } else if ((tiflags & TH_RST) && 2104 th->th_seq != tp->last_ack_sent) { 2105 /* 2106 * Test for reset before adjusting the sequence 2107 * number for overlapping data. 2108 */ 2109 goto dropafterack_ratelim; 2110 } else { 2111 tcps = TCP_STAT_GETREF(); 2112 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2113 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2114 TCP_STAT_PUTREF(); 2115 } 2116 tcp_new_dsack(tp, th->th_seq, todrop); 2117 hdroptlen += todrop; /*drop from head afterwards*/ 2118 th->th_seq += todrop; 2119 tlen -= todrop; 2120 if (th->th_urp > todrop) 2121 th->th_urp -= todrop; 2122 else { 2123 tiflags &= ~TH_URG; 2124 th->th_urp = 0; 2125 } 2126 } 2127 2128 /* 2129 * If new data are received on a connection after the 2130 * user processes are gone, then RST the other end. 2131 */ 2132 if ((so->so_state & SS_NOFDREF) && 2133 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2134 tp = tcp_close(tp); 2135 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2136 goto dropwithreset; 2137 } 2138 2139 /* 2140 * If segment ends after window, drop trailing data 2141 * (and PUSH and FIN); if nothing left, just ACK. 2142 */ 2143 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2144 if (todrop > 0) { 2145 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2146 if (todrop >= tlen) { 2147 /* 2148 * The segment actually starts after the window. 2149 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2150 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2151 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2152 */ 2153 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2154 /* 2155 * If a new connection request is received 2156 * while in TIME_WAIT, drop the old connection 2157 * and start over if the sequence numbers 2158 * are above the previous ones. 2159 * 2160 * NOTE: We will checksum the packet again, and 2161 * so we need to put the header fields back into 2162 * network order! 2163 * XXX This kind of sucks, but we don't expect 2164 * XXX this to happen very often, so maybe it 2165 * XXX doesn't matter so much. 2166 */ 2167 if (tiflags & TH_SYN && 2168 tp->t_state == TCPS_TIME_WAIT && 2169 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2170 tp = tcp_close(tp); 2171 tcp_fields_to_net(th); 2172 goto findpcb; 2173 } 2174 /* 2175 * If window is closed can only take segments at 2176 * window edge, and have to drop data and PUSH from 2177 * incoming segments. Continue processing, but 2178 * remember to ack. Otherwise, drop segment 2179 * and (if not RST) ack. 2180 */ 2181 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2182 tp->t_flags |= TF_ACKNOW; 2183 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2184 } else 2185 goto dropafterack; 2186 } else 2187 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2188 m_adj(m, -todrop); 2189 tlen -= todrop; 2190 tiflags &= ~(TH_PUSH|TH_FIN); 2191 } 2192 2193 /* 2194 * If last ACK falls within this segment's sequence numbers, 2195 * record the timestamp. 2196 * NOTE: 2197 * 1) That the test incorporates suggestions from the latest 2198 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2199 * 2) That updating only on newer timestamps interferes with 2200 * our earlier PAWS tests, so this check should be solely 2201 * predicated on the sequence space of this segment. 2202 * 3) That we modify the segment boundary check to be 2203 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2204 * instead of RFC1323's 2205 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2206 * This modified check allows us to overcome RFC1323's 2207 * limitations as described in Stevens TCP/IP Illustrated 2208 * Vol. 2 p.869. In such cases, we can still calculate the 2209 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2210 */ 2211 if (opti.ts_present && 2212 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2213 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2214 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2215 tp->ts_recent_age = tcp_now; 2216 tp->ts_recent = opti.ts_val; 2217 } 2218 2219 /* 2220 * If the RST bit is set examine the state: 2221 * SYN_RECEIVED STATE: 2222 * If passive open, return to LISTEN state. 2223 * If active open, inform user that connection was refused. 2224 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2225 * Inform user that connection was reset, and close tcb. 2226 * CLOSING, LAST_ACK, TIME_WAIT STATES 2227 * Close the tcb. 2228 */ 2229 if (tiflags & TH_RST) { 2230 if (th->th_seq != tp->last_ack_sent) 2231 goto dropafterack_ratelim; 2232 2233 switch (tp->t_state) { 2234 case TCPS_SYN_RECEIVED: 2235 so->so_error = ECONNREFUSED; 2236 goto close; 2237 2238 case TCPS_ESTABLISHED: 2239 case TCPS_FIN_WAIT_1: 2240 case TCPS_FIN_WAIT_2: 2241 case TCPS_CLOSE_WAIT: 2242 so->so_error = ECONNRESET; 2243 close: 2244 tp->t_state = TCPS_CLOSED; 2245 TCP_STATINC(TCP_STAT_DROPS); 2246 tp = tcp_close(tp); 2247 goto drop; 2248 2249 case TCPS_CLOSING: 2250 case TCPS_LAST_ACK: 2251 case TCPS_TIME_WAIT: 2252 tp = tcp_close(tp); 2253 goto drop; 2254 } 2255 } 2256 2257 /* 2258 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2259 * we must be in a synchronized state. RFC791 states (under RST 2260 * generation) that any unacceptable segment (an out-of-order SYN 2261 * qualifies) received in a synchronized state must elicit only an 2262 * empty acknowledgment segment ... and the connection remains in 2263 * the same state. 2264 */ 2265 if (tiflags & TH_SYN) { 2266 if (tp->rcv_nxt == th->th_seq) { 2267 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2268 TH_ACK); 2269 if (tcp_saveti) 2270 m_freem(tcp_saveti); 2271 return; 2272 } 2273 2274 goto dropafterack_ratelim; 2275 } 2276 2277 /* 2278 * If the ACK bit is off we drop the segment and return. 2279 */ 2280 if ((tiflags & TH_ACK) == 0) { 2281 if (tp->t_flags & TF_ACKNOW) 2282 goto dropafterack; 2283 else 2284 goto drop; 2285 } 2286 2287 /* 2288 * Ack processing. 2289 */ 2290 switch (tp->t_state) { 2291 2292 /* 2293 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2294 * ESTABLISHED state and continue processing, otherwise 2295 * send an RST. 2296 */ 2297 case TCPS_SYN_RECEIVED: 2298 if (SEQ_GT(tp->snd_una, th->th_ack) || 2299 SEQ_GT(th->th_ack, tp->snd_max)) 2300 goto dropwithreset; 2301 TCP_STATINC(TCP_STAT_CONNECTS); 2302 soisconnected(so); 2303 tcp_established(tp); 2304 /* Do window scaling? */ 2305 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2306 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2307 tp->snd_scale = tp->requested_s_scale; 2308 tp->rcv_scale = tp->request_r_scale; 2309 } 2310 TCP_REASS_LOCK(tp); 2311 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2312 TCP_REASS_UNLOCK(tp); 2313 tp->snd_wl1 = th->th_seq - 1; 2314 /* fall into ... */ 2315 2316 /* 2317 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2318 * ACKs. If the ack is in the range 2319 * tp->snd_una < th->th_ack <= tp->snd_max 2320 * then advance tp->snd_una to th->th_ack and drop 2321 * data from the retransmission queue. If this ACK reflects 2322 * more up to date window information we update our window information. 2323 */ 2324 case TCPS_ESTABLISHED: 2325 case TCPS_FIN_WAIT_1: 2326 case TCPS_FIN_WAIT_2: 2327 case TCPS_CLOSE_WAIT: 2328 case TCPS_CLOSING: 2329 case TCPS_LAST_ACK: 2330 case TCPS_TIME_WAIT: 2331 2332 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2333 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2334 TCP_STATINC(TCP_STAT_RCVDUPPACK); 2335 /* 2336 * If we have outstanding data (other than 2337 * a window probe), this is a completely 2338 * duplicate ack (ie, window info didn't 2339 * change), the ack is the biggest we've 2340 * seen and we've seen exactly our rexmt 2341 * threshhold of them, assume a packet 2342 * has been dropped and retransmit it. 2343 * Kludge snd_nxt & the congestion 2344 * window so we send only this one 2345 * packet. 2346 */ 2347 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2348 th->th_ack != tp->snd_una) 2349 tp->t_dupacks = 0; 2350 else if (tp->t_partialacks < 0 && 2351 (++tp->t_dupacks == tcprexmtthresh || 2352 TCP_FACK_FASTRECOV(tp))) { 2353 /* 2354 * Do the fast retransmit, and adjust 2355 * congestion control paramenters. 2356 */ 2357 if (tp->t_congctl->fast_retransmit(tp, th)) { 2358 /* False fast retransmit */ 2359 break; 2360 } else 2361 goto drop; 2362 } else if (tp->t_dupacks > tcprexmtthresh) { 2363 tp->snd_cwnd += tp->t_segsz; 2364 (void) tcp_output(tp); 2365 goto drop; 2366 } 2367 } else { 2368 /* 2369 * If the ack appears to be very old, only 2370 * allow data that is in-sequence. This 2371 * makes it somewhat more difficult to insert 2372 * forged data by guessing sequence numbers. 2373 * Sent an ack to try to update the send 2374 * sequence number on the other side. 2375 */ 2376 if (tlen && th->th_seq != tp->rcv_nxt && 2377 SEQ_LT(th->th_ack, 2378 tp->snd_una - tp->max_sndwnd)) 2379 goto dropafterack; 2380 } 2381 break; 2382 } 2383 /* 2384 * If the congestion window was inflated to account 2385 * for the other side's cached packets, retract it. 2386 */ 2387 /* XXX: make SACK have his own congestion control 2388 * struct -- rpaulo */ 2389 if (TCP_SACK_ENABLED(tp)) 2390 tcp_sack_newack(tp, th); 2391 else 2392 tp->t_congctl->fast_retransmit_newack(tp, th); 2393 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2394 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2395 goto dropafterack; 2396 } 2397 acked = th->th_ack - tp->snd_una; 2398 tcps = TCP_STAT_GETREF(); 2399 tcps[TCP_STAT_RCVACKPACK]++; 2400 tcps[TCP_STAT_RCVACKBYTE] += acked; 2401 TCP_STAT_PUTREF(); 2402 2403 /* 2404 * If we have a timestamp reply, update smoothed 2405 * round trip time. If no timestamp is present but 2406 * transmit timer is running and timed sequence 2407 * number was acked, update smoothed round trip time. 2408 * Since we now have an rtt measurement, cancel the 2409 * timer backoff (cf., Phil Karn's retransmit alg.). 2410 * Recompute the initial retransmit timer. 2411 */ 2412 if (ts_rtt) 2413 tcp_xmit_timer(tp, ts_rtt); 2414 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2415 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2416 2417 /* 2418 * If all outstanding data is acked, stop retransmit 2419 * timer and remember to restart (more output or persist). 2420 * If there is more data to be acked, restart retransmit 2421 * timer, using current (possibly backed-off) value. 2422 */ 2423 if (th->th_ack == tp->snd_max) { 2424 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2425 needoutput = 1; 2426 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2427 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2428 2429 /* 2430 * New data has been acked, adjust the congestion window. 2431 */ 2432 tp->t_congctl->newack(tp, th); 2433 2434 nd6_hint(tp); 2435 if (acked > so->so_snd.sb_cc) { 2436 tp->snd_wnd -= so->so_snd.sb_cc; 2437 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2438 ourfinisacked = 1; 2439 } else { 2440 if (acked > (tp->t_lastoff - tp->t_inoff)) 2441 tp->t_lastm = NULL; 2442 sbdrop(&so->so_snd, acked); 2443 tp->t_lastoff -= acked; 2444 tp->snd_wnd -= acked; 2445 ourfinisacked = 0; 2446 } 2447 sowwakeup(so); 2448 2449 icmp_check(tp, th, acked); 2450 2451 tp->snd_una = th->th_ack; 2452 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2453 tp->snd_fack = tp->snd_una; 2454 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2455 tp->snd_nxt = tp->snd_una; 2456 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2457 tp->snd_high = tp->snd_una; 2458 2459 switch (tp->t_state) { 2460 2461 /* 2462 * In FIN_WAIT_1 STATE in addition to the processing 2463 * for the ESTABLISHED state if our FIN is now acknowledged 2464 * then enter FIN_WAIT_2. 2465 */ 2466 case TCPS_FIN_WAIT_1: 2467 if (ourfinisacked) { 2468 /* 2469 * If we can't receive any more 2470 * data, then closing user can proceed. 2471 * Starting the timer is contrary to the 2472 * specification, but if we don't get a FIN 2473 * we'll hang forever. 2474 */ 2475 if (so->so_state & SS_CANTRCVMORE) { 2476 soisdisconnected(so); 2477 if (tp->t_maxidle > 0) 2478 TCP_TIMER_ARM(tp, TCPT_2MSL, 2479 tp->t_maxidle); 2480 } 2481 tp->t_state = TCPS_FIN_WAIT_2; 2482 } 2483 break; 2484 2485 /* 2486 * In CLOSING STATE in addition to the processing for 2487 * the ESTABLISHED state if the ACK acknowledges our FIN 2488 * then enter the TIME-WAIT state, otherwise ignore 2489 * the segment. 2490 */ 2491 case TCPS_CLOSING: 2492 if (ourfinisacked) { 2493 tp->t_state = TCPS_TIME_WAIT; 2494 tcp_canceltimers(tp); 2495 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2496 soisdisconnected(so); 2497 } 2498 break; 2499 2500 /* 2501 * In LAST_ACK, we may still be waiting for data to drain 2502 * and/or to be acked, as well as for the ack of our FIN. 2503 * If our FIN is now acknowledged, delete the TCB, 2504 * enter the closed state and return. 2505 */ 2506 case TCPS_LAST_ACK: 2507 if (ourfinisacked) { 2508 tp = tcp_close(tp); 2509 goto drop; 2510 } 2511 break; 2512 2513 /* 2514 * In TIME_WAIT state the only thing that should arrive 2515 * is a retransmission of the remote FIN. Acknowledge 2516 * it and restart the finack timer. 2517 */ 2518 case TCPS_TIME_WAIT: 2519 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2520 goto dropafterack; 2521 } 2522 } 2523 2524 step6: 2525 /* 2526 * Update window information. 2527 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2528 */ 2529 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2530 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2531 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2532 /* keep track of pure window updates */ 2533 if (tlen == 0 && 2534 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2535 TCP_STATINC(TCP_STAT_RCVWINUPD); 2536 tp->snd_wnd = tiwin; 2537 tp->snd_wl1 = th->th_seq; 2538 tp->snd_wl2 = th->th_ack; 2539 if (tp->snd_wnd > tp->max_sndwnd) 2540 tp->max_sndwnd = tp->snd_wnd; 2541 needoutput = 1; 2542 } 2543 2544 /* 2545 * Process segments with URG. 2546 */ 2547 if ((tiflags & TH_URG) && th->th_urp && 2548 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2549 /* 2550 * This is a kludge, but if we receive and accept 2551 * random urgent pointers, we'll crash in 2552 * soreceive. It's hard to imagine someone 2553 * actually wanting to send this much urgent data. 2554 */ 2555 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2556 th->th_urp = 0; /* XXX */ 2557 tiflags &= ~TH_URG; /* XXX */ 2558 goto dodata; /* XXX */ 2559 } 2560 /* 2561 * If this segment advances the known urgent pointer, 2562 * then mark the data stream. This should not happen 2563 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2564 * a FIN has been received from the remote side. 2565 * In these states we ignore the URG. 2566 * 2567 * According to RFC961 (Assigned Protocols), 2568 * the urgent pointer points to the last octet 2569 * of urgent data. We continue, however, 2570 * to consider it to indicate the first octet 2571 * of data past the urgent section as the original 2572 * spec states (in one of two places). 2573 */ 2574 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2575 tp->rcv_up = th->th_seq + th->th_urp; 2576 so->so_oobmark = so->so_rcv.sb_cc + 2577 (tp->rcv_up - tp->rcv_nxt) - 1; 2578 if (so->so_oobmark == 0) 2579 so->so_state |= SS_RCVATMARK; 2580 sohasoutofband(so); 2581 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2582 } 2583 /* 2584 * Remove out of band data so doesn't get presented to user. 2585 * This can happen independent of advancing the URG pointer, 2586 * but if two URG's are pending at once, some out-of-band 2587 * data may creep in... ick. 2588 */ 2589 if (th->th_urp <= (u_int16_t) tlen 2590 #ifdef SO_OOBINLINE 2591 && (so->so_options & SO_OOBINLINE) == 0 2592 #endif 2593 ) 2594 tcp_pulloutofband(so, th, m, hdroptlen); 2595 } else 2596 /* 2597 * If no out of band data is expected, 2598 * pull receive urgent pointer along 2599 * with the receive window. 2600 */ 2601 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2602 tp->rcv_up = tp->rcv_nxt; 2603 dodata: /* XXX */ 2604 2605 /* 2606 * Process the segment text, merging it into the TCP sequencing queue, 2607 * and arranging for acknowledgement of receipt if necessary. 2608 * This process logically involves adjusting tp->rcv_wnd as data 2609 * is presented to the user (this happens in tcp_usrreq.c, 2610 * case PRU_RCVD). If a FIN has already been received on this 2611 * connection then we just ignore the text. 2612 */ 2613 if ((tlen || (tiflags & TH_FIN)) && 2614 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2615 /* 2616 * Insert segment ti into reassembly queue of tcp with 2617 * control block tp. Return TH_FIN if reassembly now includes 2618 * a segment with FIN. The macro form does the common case 2619 * inline (segment is the next to be received on an 2620 * established connection, and the queue is empty), 2621 * avoiding linkage into and removal from the queue and 2622 * repetition of various conversions. 2623 * Set DELACK for segments received in order, but ack 2624 * immediately when segments are out of order 2625 * (so fast retransmit can work). 2626 */ 2627 /* NOTE: this was TCP_REASS() macro, but used only once */ 2628 TCP_REASS_LOCK(tp); 2629 if (th->th_seq == tp->rcv_nxt && 2630 TAILQ_FIRST(&tp->segq) == NULL && 2631 tp->t_state == TCPS_ESTABLISHED) { 2632 tcp_setup_ack(tp, th); 2633 tp->rcv_nxt += tlen; 2634 tiflags = th->th_flags & TH_FIN; 2635 tcps = TCP_STAT_GETREF(); 2636 tcps[TCP_STAT_RCVPACK]++; 2637 tcps[TCP_STAT_RCVBYTE] += tlen; 2638 TCP_STAT_PUTREF(); 2639 nd6_hint(tp); 2640 if (so->so_state & SS_CANTRCVMORE) 2641 m_freem(m); 2642 else { 2643 m_adj(m, hdroptlen); 2644 sbappendstream(&(so)->so_rcv, m); 2645 } 2646 sorwakeup(so); 2647 } else { 2648 m_adj(m, hdroptlen); 2649 tiflags = tcp_reass(tp, th, m, &tlen); 2650 tp->t_flags |= TF_ACKNOW; 2651 } 2652 TCP_REASS_UNLOCK(tp); 2653 2654 /* 2655 * Note the amount of data that peer has sent into 2656 * our window, in order to estimate the sender's 2657 * buffer size. 2658 */ 2659 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2660 } else { 2661 m_freem(m); 2662 m = NULL; 2663 tiflags &= ~TH_FIN; 2664 } 2665 2666 /* 2667 * If FIN is received ACK the FIN and let the user know 2668 * that the connection is closing. Ignore a FIN received before 2669 * the connection is fully established. 2670 */ 2671 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2672 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2673 socantrcvmore(so); 2674 tp->t_flags |= TF_ACKNOW; 2675 tp->rcv_nxt++; 2676 } 2677 switch (tp->t_state) { 2678 2679 /* 2680 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2681 */ 2682 case TCPS_ESTABLISHED: 2683 tp->t_state = TCPS_CLOSE_WAIT; 2684 break; 2685 2686 /* 2687 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2688 * enter the CLOSING state. 2689 */ 2690 case TCPS_FIN_WAIT_1: 2691 tp->t_state = TCPS_CLOSING; 2692 break; 2693 2694 /* 2695 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2696 * starting the time-wait timer, turning off the other 2697 * standard timers. 2698 */ 2699 case TCPS_FIN_WAIT_2: 2700 tp->t_state = TCPS_TIME_WAIT; 2701 tcp_canceltimers(tp); 2702 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2703 soisdisconnected(so); 2704 break; 2705 2706 /* 2707 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2708 */ 2709 case TCPS_TIME_WAIT: 2710 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2711 break; 2712 } 2713 } 2714 #ifdef TCP_DEBUG 2715 if (so->so_options & SO_DEBUG) 2716 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2717 #endif 2718 2719 /* 2720 * Return any desired output. 2721 */ 2722 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2723 (void) tcp_output(tp); 2724 } 2725 if (tcp_saveti) 2726 m_freem(tcp_saveti); 2727 return; 2728 2729 badsyn: 2730 /* 2731 * Received a bad SYN. Increment counters and dropwithreset. 2732 */ 2733 TCP_STATINC(TCP_STAT_BADSYN); 2734 tp = NULL; 2735 goto dropwithreset; 2736 2737 dropafterack: 2738 /* 2739 * Generate an ACK dropping incoming segment if it occupies 2740 * sequence space, where the ACK reflects our state. 2741 */ 2742 if (tiflags & TH_RST) 2743 goto drop; 2744 goto dropafterack2; 2745 2746 dropafterack_ratelim: 2747 /* 2748 * We may want to rate-limit ACKs against SYN/RST attack. 2749 */ 2750 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2751 tcp_ackdrop_ppslim) == 0) { 2752 /* XXX stat */ 2753 goto drop; 2754 } 2755 /* ...fall into dropafterack2... */ 2756 2757 dropafterack2: 2758 m_freem(m); 2759 tp->t_flags |= TF_ACKNOW; 2760 (void) tcp_output(tp); 2761 if (tcp_saveti) 2762 m_freem(tcp_saveti); 2763 return; 2764 2765 dropwithreset_ratelim: 2766 /* 2767 * We may want to rate-limit RSTs in certain situations, 2768 * particularly if we are sending an RST in response to 2769 * an attempt to connect to or otherwise communicate with 2770 * a port for which we have no socket. 2771 */ 2772 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2773 tcp_rst_ppslim) == 0) { 2774 /* XXX stat */ 2775 goto drop; 2776 } 2777 /* ...fall into dropwithreset... */ 2778 2779 dropwithreset: 2780 /* 2781 * Generate a RST, dropping incoming segment. 2782 * Make ACK acceptable to originator of segment. 2783 */ 2784 if (tiflags & TH_RST) 2785 goto drop; 2786 2787 switch (af) { 2788 #ifdef INET6 2789 case AF_INET6: 2790 /* For following calls to tcp_respond */ 2791 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2792 goto drop; 2793 break; 2794 #endif /* INET6 */ 2795 case AF_INET: 2796 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2797 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2798 goto drop; 2799 } 2800 2801 if (tiflags & TH_ACK) 2802 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2803 else { 2804 if (tiflags & TH_SYN) 2805 tlen++; 2806 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2807 TH_RST|TH_ACK); 2808 } 2809 if (tcp_saveti) 2810 m_freem(tcp_saveti); 2811 return; 2812 2813 badcsum: 2814 drop: 2815 /* 2816 * Drop space held by incoming segment and return. 2817 */ 2818 if (tp) { 2819 if (tp->t_inpcb) 2820 so = tp->t_inpcb->inp_socket; 2821 #ifdef INET6 2822 else if (tp->t_in6pcb) 2823 so = tp->t_in6pcb->in6p_socket; 2824 #endif 2825 else 2826 so = NULL; 2827 #ifdef TCP_DEBUG 2828 if (so && (so->so_options & SO_DEBUG) != 0) 2829 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2830 #endif 2831 } 2832 if (tcp_saveti) 2833 m_freem(tcp_saveti); 2834 m_freem(m); 2835 return; 2836 } 2837 2838 #ifdef TCP_SIGNATURE 2839 int 2840 tcp_signature_apply(void *fstate, void *data, u_int len) 2841 { 2842 2843 MD5Update(fstate, (u_char *)data, len); 2844 return (0); 2845 } 2846 2847 struct secasvar * 2848 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2849 { 2850 struct secasvar *sav; 2851 #ifdef FAST_IPSEC 2852 union sockaddr_union dst; 2853 #endif 2854 struct ip *ip; 2855 struct ip6_hdr *ip6; 2856 2857 ip = mtod(m, struct ip *); 2858 switch (ip->ip_v) { 2859 case 4: 2860 ip = mtod(m, struct ip *); 2861 ip6 = NULL; 2862 break; 2863 case 6: 2864 ip = NULL; 2865 ip6 = mtod(m, struct ip6_hdr *); 2866 break; 2867 default: 2868 return (NULL); 2869 } 2870 2871 #ifdef FAST_IPSEC 2872 /* Extract the destination from the IP header in the mbuf. */ 2873 bzero(&dst, sizeof(union sockaddr_union)); 2874 if (ip !=NULL) { 2875 dst.sa.sa_len = sizeof(struct sockaddr_in); 2876 dst.sa.sa_family = AF_INET; 2877 dst.sin.sin_addr = ip->ip_dst; 2878 } else { 2879 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2880 dst.sa.sa_family = AF_INET6; 2881 dst.sin6.sin6_addr = ip6->ip6_dst; 2882 } 2883 2884 /* 2885 * Look up an SADB entry which matches the address of the peer. 2886 */ 2887 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2888 #else 2889 if (ip) 2890 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 2891 (void *)&ip->ip_dst, IPPROTO_TCP, 2892 htonl(TCP_SIG_SPI), 0, 0); 2893 else 2894 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 2895 (void *)&ip6->ip6_dst, IPPROTO_TCP, 2896 htonl(TCP_SIG_SPI), 0, 0); 2897 #endif 2898 2899 return (sav); /* freesav must be performed by caller */ 2900 } 2901 2902 int 2903 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2904 struct secasvar *sav, char *sig) 2905 { 2906 MD5_CTX ctx; 2907 struct ip *ip; 2908 struct ipovly *ipovly; 2909 struct ip6_hdr *ip6; 2910 struct ippseudo ippseudo; 2911 struct ip6_hdr_pseudo ip6pseudo; 2912 struct tcphdr th0; 2913 int l, tcphdrlen; 2914 2915 if (sav == NULL) 2916 return (-1); 2917 2918 tcphdrlen = th->th_off * 4; 2919 2920 switch (mtod(m, struct ip *)->ip_v) { 2921 case 4: 2922 ip = mtod(m, struct ip *); 2923 ip6 = NULL; 2924 break; 2925 case 6: 2926 ip = NULL; 2927 ip6 = mtod(m, struct ip6_hdr *); 2928 break; 2929 default: 2930 return (-1); 2931 } 2932 2933 MD5Init(&ctx); 2934 2935 if (ip) { 2936 memset(&ippseudo, 0, sizeof(ippseudo)); 2937 ipovly = (struct ipovly *)ip; 2938 ippseudo.ippseudo_src = ipovly->ih_src; 2939 ippseudo.ippseudo_dst = ipovly->ih_dst; 2940 ippseudo.ippseudo_pad = 0; 2941 ippseudo.ippseudo_p = IPPROTO_TCP; 2942 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2943 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2944 } else { 2945 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2946 ip6pseudo.ip6ph_src = ip6->ip6_src; 2947 in6_clearscope(&ip6pseudo.ip6ph_src); 2948 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2949 in6_clearscope(&ip6pseudo.ip6ph_dst); 2950 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2951 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2952 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2953 } 2954 2955 th0 = *th; 2956 th0.th_sum = 0; 2957 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2958 2959 l = m->m_pkthdr.len - thoff - tcphdrlen; 2960 if (l > 0) 2961 m_apply(m, thoff + tcphdrlen, 2962 m->m_pkthdr.len - thoff - tcphdrlen, 2963 tcp_signature_apply, &ctx); 2964 2965 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2966 MD5Final(sig, &ctx); 2967 2968 return (0); 2969 } 2970 #endif 2971 2972 static int 2973 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 2974 struct tcphdr *th, 2975 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2976 { 2977 u_int16_t mss; 2978 int opt, optlen = 0; 2979 #ifdef TCP_SIGNATURE 2980 void *sigp = NULL; 2981 char sigbuf[TCP_SIGLEN]; 2982 struct secasvar *sav = NULL; 2983 #endif 2984 2985 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2986 opt = cp[0]; 2987 if (opt == TCPOPT_EOL) 2988 break; 2989 if (opt == TCPOPT_NOP) 2990 optlen = 1; 2991 else { 2992 if (cnt < 2) 2993 break; 2994 optlen = cp[1]; 2995 if (optlen < 2 || optlen > cnt) 2996 break; 2997 } 2998 switch (opt) { 2999 3000 default: 3001 continue; 3002 3003 case TCPOPT_MAXSEG: 3004 if (optlen != TCPOLEN_MAXSEG) 3005 continue; 3006 if (!(th->th_flags & TH_SYN)) 3007 continue; 3008 if (TCPS_HAVERCVDSYN(tp->t_state)) 3009 continue; 3010 bcopy(cp + 2, &mss, sizeof(mss)); 3011 oi->maxseg = ntohs(mss); 3012 break; 3013 3014 case TCPOPT_WINDOW: 3015 if (optlen != TCPOLEN_WINDOW) 3016 continue; 3017 if (!(th->th_flags & TH_SYN)) 3018 continue; 3019 if (TCPS_HAVERCVDSYN(tp->t_state)) 3020 continue; 3021 tp->t_flags |= TF_RCVD_SCALE; 3022 tp->requested_s_scale = cp[2]; 3023 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3024 #if 0 /*XXX*/ 3025 char *p; 3026 3027 if (ip) 3028 p = ntohl(ip->ip_src); 3029 #ifdef INET6 3030 else if (ip6) 3031 p = ip6_sprintf(&ip6->ip6_src); 3032 #endif 3033 else 3034 p = "(unknown)"; 3035 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3036 "assuming %d\n", 3037 tp->requested_s_scale, p, 3038 TCP_MAX_WINSHIFT); 3039 #else 3040 log(LOG_ERR, "TCP: invalid wscale %d, " 3041 "assuming %d\n", 3042 tp->requested_s_scale, 3043 TCP_MAX_WINSHIFT); 3044 #endif 3045 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3046 } 3047 break; 3048 3049 case TCPOPT_TIMESTAMP: 3050 if (optlen != TCPOLEN_TIMESTAMP) 3051 continue; 3052 oi->ts_present = 1; 3053 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3054 NTOHL(oi->ts_val); 3055 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3056 NTOHL(oi->ts_ecr); 3057 3058 if (!(th->th_flags & TH_SYN)) 3059 continue; 3060 if (TCPS_HAVERCVDSYN(tp->t_state)) 3061 continue; 3062 /* 3063 * A timestamp received in a SYN makes 3064 * it ok to send timestamp requests and replies. 3065 */ 3066 tp->t_flags |= TF_RCVD_TSTMP; 3067 tp->ts_recent = oi->ts_val; 3068 tp->ts_recent_age = tcp_now; 3069 break; 3070 3071 case TCPOPT_SACK_PERMITTED: 3072 if (optlen != TCPOLEN_SACK_PERMITTED) 3073 continue; 3074 if (!(th->th_flags & TH_SYN)) 3075 continue; 3076 if (TCPS_HAVERCVDSYN(tp->t_state)) 3077 continue; 3078 if (tcp_do_sack) { 3079 tp->t_flags |= TF_SACK_PERMIT; 3080 tp->t_flags |= TF_WILL_SACK; 3081 } 3082 break; 3083 3084 case TCPOPT_SACK: 3085 tcp_sack_option(tp, th, cp, optlen); 3086 break; 3087 #ifdef TCP_SIGNATURE 3088 case TCPOPT_SIGNATURE: 3089 if (optlen != TCPOLEN_SIGNATURE) 3090 continue; 3091 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 3092 return (-1); 3093 3094 sigp = sigbuf; 3095 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3096 tp->t_flags |= TF_SIGNATURE; 3097 break; 3098 #endif 3099 } 3100 } 3101 3102 #ifdef TCP_SIGNATURE 3103 if (tp->t_flags & TF_SIGNATURE) { 3104 3105 sav = tcp_signature_getsav(m, th); 3106 3107 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3108 return (-1); 3109 } 3110 3111 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3112 if (sav == NULL) 3113 return (-1); 3114 #ifdef FAST_IPSEC 3115 KEY_FREESAV(&sav); 3116 #else 3117 key_freesav(sav); 3118 #endif 3119 return (-1); 3120 } 3121 3122 if (sigp) { 3123 char sig[TCP_SIGLEN]; 3124 3125 tcp_fields_to_net(th); 3126 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3127 tcp_fields_to_host(th); 3128 if (sav == NULL) 3129 return (-1); 3130 #ifdef FAST_IPSEC 3131 KEY_FREESAV(&sav); 3132 #else 3133 key_freesav(sav); 3134 #endif 3135 return (-1); 3136 } 3137 tcp_fields_to_host(th); 3138 3139 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3140 TCP_STATINC(TCP_STAT_BADSIG); 3141 if (sav == NULL) 3142 return (-1); 3143 #ifdef FAST_IPSEC 3144 KEY_FREESAV(&sav); 3145 #else 3146 key_freesav(sav); 3147 #endif 3148 return (-1); 3149 } else 3150 TCP_STATINC(TCP_STAT_GOODSIG); 3151 3152 key_sa_recordxfer(sav, m); 3153 #ifdef FAST_IPSEC 3154 KEY_FREESAV(&sav); 3155 #else 3156 key_freesav(sav); 3157 #endif 3158 } 3159 #endif 3160 3161 return (0); 3162 } 3163 3164 /* 3165 * Pull out of band byte out of a segment so 3166 * it doesn't appear in the user's data queue. 3167 * It is still reflected in the segment length for 3168 * sequencing purposes. 3169 */ 3170 void 3171 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3172 struct mbuf *m, int off) 3173 { 3174 int cnt = off + th->th_urp - 1; 3175 3176 while (cnt >= 0) { 3177 if (m->m_len > cnt) { 3178 char *cp = mtod(m, char *) + cnt; 3179 struct tcpcb *tp = sototcpcb(so); 3180 3181 tp->t_iobc = *cp; 3182 tp->t_oobflags |= TCPOOB_HAVEDATA; 3183 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3184 m->m_len--; 3185 return; 3186 } 3187 cnt -= m->m_len; 3188 m = m->m_next; 3189 if (m == 0) 3190 break; 3191 } 3192 panic("tcp_pulloutofband"); 3193 } 3194 3195 /* 3196 * Collect new round-trip time estimate 3197 * and update averages and current timeout. 3198 */ 3199 void 3200 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3201 { 3202 int32_t delta; 3203 3204 TCP_STATINC(TCP_STAT_RTTUPDATED); 3205 if (tp->t_srtt != 0) { 3206 /* 3207 * srtt is stored as fixed point with 3 bits after the 3208 * binary point (i.e., scaled by 8). The following magic 3209 * is equivalent to the smoothing algorithm in rfc793 with 3210 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3211 * point). Adjust rtt to origin 0. 3212 */ 3213 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3214 if ((tp->t_srtt += delta) <= 0) 3215 tp->t_srtt = 1 << 2; 3216 /* 3217 * We accumulate a smoothed rtt variance (actually, a 3218 * smoothed mean difference), then set the retransmit 3219 * timer to smoothed rtt + 4 times the smoothed variance. 3220 * rttvar is stored as fixed point with 2 bits after the 3221 * binary point (scaled by 4). The following is 3222 * equivalent to rfc793 smoothing with an alpha of .75 3223 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3224 * rfc793's wired-in beta. 3225 */ 3226 if (delta < 0) 3227 delta = -delta; 3228 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3229 if ((tp->t_rttvar += delta) <= 0) 3230 tp->t_rttvar = 1 << 2; 3231 } else { 3232 /* 3233 * No rtt measurement yet - use the unsmoothed rtt. 3234 * Set the variance to half the rtt (so our first 3235 * retransmit happens at 3*rtt). 3236 */ 3237 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3238 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3239 } 3240 tp->t_rtttime = 0; 3241 tp->t_rxtshift = 0; 3242 3243 /* 3244 * the retransmit should happen at rtt + 4 * rttvar. 3245 * Because of the way we do the smoothing, srtt and rttvar 3246 * will each average +1/2 tick of bias. When we compute 3247 * the retransmit timer, we want 1/2 tick of rounding and 3248 * 1 extra tick because of +-1/2 tick uncertainty in the 3249 * firing of the timer. The bias will give us exactly the 3250 * 1.5 tick we need. But, because the bias is 3251 * statistical, we have to test that we don't drop below 3252 * the minimum feasible timer (which is 2 ticks). 3253 */ 3254 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3255 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3256 3257 /* 3258 * We received an ack for a packet that wasn't retransmitted; 3259 * it is probably safe to discard any error indications we've 3260 * received recently. This isn't quite right, but close enough 3261 * for now (a route might have failed after we sent a segment, 3262 * and the return path might not be symmetrical). 3263 */ 3264 tp->t_softerror = 0; 3265 } 3266 3267 3268 /* 3269 * TCP compressed state engine. Currently used to hold compressed 3270 * state for SYN_RECEIVED. 3271 */ 3272 3273 u_long syn_cache_count; 3274 u_int32_t syn_hash1, syn_hash2; 3275 3276 #define SYN_HASH(sa, sp, dp) \ 3277 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3278 ((u_int32_t)(sp)))^syn_hash2))) 3279 #ifndef INET6 3280 #define SYN_HASHALL(hash, src, dst) \ 3281 do { \ 3282 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3283 ((const struct sockaddr_in *)(src))->sin_port, \ 3284 ((const struct sockaddr_in *)(dst))->sin_port); \ 3285 } while (/*CONSTCOND*/ 0) 3286 #else 3287 #define SYN_HASH6(sa, sp, dp) \ 3288 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3289 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3290 & 0x7fffffff) 3291 3292 #define SYN_HASHALL(hash, src, dst) \ 3293 do { \ 3294 switch ((src)->sa_family) { \ 3295 case AF_INET: \ 3296 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3297 ((const struct sockaddr_in *)(src))->sin_port, \ 3298 ((const struct sockaddr_in *)(dst))->sin_port); \ 3299 break; \ 3300 case AF_INET6: \ 3301 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3302 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3303 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3304 break; \ 3305 default: \ 3306 hash = 0; \ 3307 } \ 3308 } while (/*CONSTCOND*/0) 3309 #endif /* INET6 */ 3310 3311 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL, 3312 IPL_SOFTNET); 3313 3314 /* 3315 * We don't estimate RTT with SYNs, so each packet starts with the default 3316 * RTT and each timer step has a fixed timeout value. 3317 */ 3318 #define SYN_CACHE_TIMER_ARM(sc) \ 3319 do { \ 3320 TCPT_RANGESET((sc)->sc_rxtcur, \ 3321 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3322 TCPTV_REXMTMAX); \ 3323 callout_reset(&(sc)->sc_timer, \ 3324 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3325 } while (/*CONSTCOND*/0) 3326 3327 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3328 3329 static inline void 3330 syn_cache_rm(struct syn_cache *sc) 3331 { 3332 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3333 sc, sc_bucketq); 3334 sc->sc_tp = NULL; 3335 LIST_REMOVE(sc, sc_tpq); 3336 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3337 callout_stop(&sc->sc_timer); 3338 syn_cache_count--; 3339 } 3340 3341 static inline void 3342 syn_cache_put(struct syn_cache *sc) 3343 { 3344 if (sc->sc_ipopts) 3345 (void) m_free(sc->sc_ipopts); 3346 rtcache_free(&sc->sc_route); 3347 if (callout_invoking(&sc->sc_timer)) 3348 sc->sc_flags |= SCF_DEAD; 3349 else { 3350 callout_destroy(&sc->sc_timer); 3351 pool_put(&syn_cache_pool, sc); 3352 } 3353 } 3354 3355 void 3356 syn_cache_init(void) 3357 { 3358 int i; 3359 3360 /* Initialize the hash buckets. */ 3361 for (i = 0; i < tcp_syn_cache_size; i++) 3362 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3363 } 3364 3365 void 3366 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3367 { 3368 struct syn_cache_head *scp; 3369 struct syn_cache *sc2; 3370 int s; 3371 3372 /* 3373 * If there are no entries in the hash table, reinitialize 3374 * the hash secrets. 3375 */ 3376 if (syn_cache_count == 0) { 3377 syn_hash1 = arc4random(); 3378 syn_hash2 = arc4random(); 3379 } 3380 3381 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3382 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3383 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3384 3385 /* 3386 * Make sure that we don't overflow the per-bucket 3387 * limit or the total cache size limit. 3388 */ 3389 s = splsoftnet(); 3390 if (scp->sch_length >= tcp_syn_bucket_limit) { 3391 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3392 /* 3393 * The bucket is full. Toss the oldest element in the 3394 * bucket. This will be the first entry in the bucket. 3395 */ 3396 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3397 #ifdef DIAGNOSTIC 3398 /* 3399 * This should never happen; we should always find an 3400 * entry in our bucket. 3401 */ 3402 if (sc2 == NULL) 3403 panic("syn_cache_insert: bucketoverflow: impossible"); 3404 #endif 3405 syn_cache_rm(sc2); 3406 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3407 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3408 struct syn_cache_head *scp2, *sce; 3409 3410 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3411 /* 3412 * The cache is full. Toss the oldest entry in the 3413 * first non-empty bucket we can find. 3414 * 3415 * XXX We would really like to toss the oldest 3416 * entry in the cache, but we hope that this 3417 * condition doesn't happen very often. 3418 */ 3419 scp2 = scp; 3420 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3421 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3422 for (++scp2; scp2 != scp; scp2++) { 3423 if (scp2 >= sce) 3424 scp2 = &tcp_syn_cache[0]; 3425 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3426 break; 3427 } 3428 #ifdef DIAGNOSTIC 3429 /* 3430 * This should never happen; we should always find a 3431 * non-empty bucket. 3432 */ 3433 if (scp2 == scp) 3434 panic("syn_cache_insert: cacheoverflow: " 3435 "impossible"); 3436 #endif 3437 } 3438 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3439 syn_cache_rm(sc2); 3440 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3441 } 3442 3443 /* 3444 * Initialize the entry's timer. 3445 */ 3446 sc->sc_rxttot = 0; 3447 sc->sc_rxtshift = 0; 3448 SYN_CACHE_TIMER_ARM(sc); 3449 3450 /* Link it from tcpcb entry */ 3451 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3452 3453 /* Put it into the bucket. */ 3454 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3455 scp->sch_length++; 3456 syn_cache_count++; 3457 3458 TCP_STATINC(TCP_STAT_SC_ADDED); 3459 splx(s); 3460 } 3461 3462 /* 3463 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3464 * If we have retransmitted an entry the maximum number of times, expire 3465 * that entry. 3466 */ 3467 void 3468 syn_cache_timer(void *arg) 3469 { 3470 struct syn_cache *sc = arg; 3471 int s; 3472 3473 s = splsoftnet(); 3474 callout_ack(&sc->sc_timer); 3475 3476 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3477 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3478 callout_destroy(&sc->sc_timer); 3479 pool_put(&syn_cache_pool, sc); 3480 splx(s); 3481 return; 3482 } 3483 3484 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3485 /* Drop it -- too many retransmissions. */ 3486 goto dropit; 3487 } 3488 3489 /* 3490 * Compute the total amount of time this entry has 3491 * been on a queue. If this entry has been on longer 3492 * than the keep alive timer would allow, expire it. 3493 */ 3494 sc->sc_rxttot += sc->sc_rxtcur; 3495 if (sc->sc_rxttot >= tcp_keepinit) 3496 goto dropit; 3497 3498 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3499 (void) syn_cache_respond(sc, NULL); 3500 3501 /* Advance the timer back-off. */ 3502 sc->sc_rxtshift++; 3503 SYN_CACHE_TIMER_ARM(sc); 3504 3505 splx(s); 3506 return; 3507 3508 dropit: 3509 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3510 syn_cache_rm(sc); 3511 syn_cache_put(sc); /* calls pool_put but see spl above */ 3512 splx(s); 3513 } 3514 3515 /* 3516 * Remove syn cache created by the specified tcb entry, 3517 * because this does not make sense to keep them 3518 * (if there's no tcb entry, syn cache entry will never be used) 3519 */ 3520 void 3521 syn_cache_cleanup(struct tcpcb *tp) 3522 { 3523 struct syn_cache *sc, *nsc; 3524 int s; 3525 3526 s = splsoftnet(); 3527 3528 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3529 nsc = LIST_NEXT(sc, sc_tpq); 3530 3531 #ifdef DIAGNOSTIC 3532 if (sc->sc_tp != tp) 3533 panic("invalid sc_tp in syn_cache_cleanup"); 3534 #endif 3535 syn_cache_rm(sc); 3536 syn_cache_put(sc); /* calls pool_put but see spl above */ 3537 } 3538 /* just for safety */ 3539 LIST_INIT(&tp->t_sc); 3540 3541 splx(s); 3542 } 3543 3544 /* 3545 * Find an entry in the syn cache. 3546 */ 3547 struct syn_cache * 3548 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3549 struct syn_cache_head **headp) 3550 { 3551 struct syn_cache *sc; 3552 struct syn_cache_head *scp; 3553 u_int32_t hash; 3554 int s; 3555 3556 SYN_HASHALL(hash, src, dst); 3557 3558 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3559 *headp = scp; 3560 s = splsoftnet(); 3561 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3562 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3563 if (sc->sc_hash != hash) 3564 continue; 3565 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3566 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3567 splx(s); 3568 return (sc); 3569 } 3570 } 3571 splx(s); 3572 return (NULL); 3573 } 3574 3575 /* 3576 * This function gets called when we receive an ACK for a 3577 * socket in the LISTEN state. We look up the connection 3578 * in the syn cache, and if its there, we pull it out of 3579 * the cache and turn it into a full-blown connection in 3580 * the SYN-RECEIVED state. 3581 * 3582 * The return values may not be immediately obvious, and their effects 3583 * can be subtle, so here they are: 3584 * 3585 * NULL SYN was not found in cache; caller should drop the 3586 * packet and send an RST. 3587 * 3588 * -1 We were unable to create the new connection, and are 3589 * aborting it. An ACK,RST is being sent to the peer 3590 * (unless we got screwey sequence numbners; see below), 3591 * because the 3-way handshake has been completed. Caller 3592 * should not free the mbuf, since we may be using it. If 3593 * we are not, we will free it. 3594 * 3595 * Otherwise, the return value is a pointer to the new socket 3596 * associated with the connection. 3597 */ 3598 struct socket * 3599 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3600 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3601 struct socket *so, struct mbuf *m) 3602 { 3603 struct syn_cache *sc; 3604 struct syn_cache_head *scp; 3605 struct inpcb *inp = NULL; 3606 #ifdef INET6 3607 struct in6pcb *in6p = NULL; 3608 #endif 3609 struct tcpcb *tp = 0; 3610 struct mbuf *am; 3611 int s; 3612 struct socket *oso; 3613 3614 s = splsoftnet(); 3615 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3616 splx(s); 3617 return (NULL); 3618 } 3619 3620 /* 3621 * Verify the sequence and ack numbers. Try getting the correct 3622 * response again. 3623 */ 3624 if ((th->th_ack != sc->sc_iss + 1) || 3625 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3626 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3627 (void) syn_cache_respond(sc, m); 3628 splx(s); 3629 return ((struct socket *)(-1)); 3630 } 3631 3632 /* Remove this cache entry */ 3633 syn_cache_rm(sc); 3634 splx(s); 3635 3636 /* 3637 * Ok, create the full blown connection, and set things up 3638 * as they would have been set up if we had created the 3639 * connection when the SYN arrived. If we can't create 3640 * the connection, abort it. 3641 */ 3642 /* 3643 * inp still has the OLD in_pcb stuff, set the 3644 * v6-related flags on the new guy, too. This is 3645 * done particularly for the case where an AF_INET6 3646 * socket is bound only to a port, and a v4 connection 3647 * comes in on that port. 3648 * we also copy the flowinfo from the original pcb 3649 * to the new one. 3650 */ 3651 oso = so; 3652 so = sonewconn(so, SS_ISCONNECTED); 3653 if (so == NULL) 3654 goto resetandabort; 3655 3656 switch (so->so_proto->pr_domain->dom_family) { 3657 #ifdef INET 3658 case AF_INET: 3659 inp = sotoinpcb(so); 3660 break; 3661 #endif 3662 #ifdef INET6 3663 case AF_INET6: 3664 in6p = sotoin6pcb(so); 3665 break; 3666 #endif 3667 } 3668 switch (src->sa_family) { 3669 #ifdef INET 3670 case AF_INET: 3671 if (inp) { 3672 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3673 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3674 inp->inp_options = ip_srcroute(); 3675 in_pcbstate(inp, INP_BOUND); 3676 if (inp->inp_options == NULL) { 3677 inp->inp_options = sc->sc_ipopts; 3678 sc->sc_ipopts = NULL; 3679 } 3680 } 3681 #ifdef INET6 3682 else if (in6p) { 3683 /* IPv4 packet to AF_INET6 socket */ 3684 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3685 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3686 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3687 &in6p->in6p_laddr.s6_addr32[3], 3688 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3689 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3690 in6totcpcb(in6p)->t_family = AF_INET; 3691 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3692 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3693 else 3694 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3695 in6_pcbstate(in6p, IN6P_BOUND); 3696 } 3697 #endif 3698 break; 3699 #endif 3700 #ifdef INET6 3701 case AF_INET6: 3702 if (in6p) { 3703 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3704 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3705 in6_pcbstate(in6p, IN6P_BOUND); 3706 } 3707 break; 3708 #endif 3709 } 3710 #ifdef INET6 3711 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3712 struct in6pcb *oin6p = sotoin6pcb(oso); 3713 /* inherit socket options from the listening socket */ 3714 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3715 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3716 m_freem(in6p->in6p_options); 3717 in6p->in6p_options = 0; 3718 } 3719 ip6_savecontrol(in6p, &in6p->in6p_options, 3720 mtod(m, struct ip6_hdr *), m); 3721 } 3722 #endif 3723 3724 #if defined(IPSEC) || defined(FAST_IPSEC) 3725 /* 3726 * we make a copy of policy, instead of sharing the policy, 3727 * for better behavior in terms of SA lookup and dead SA removal. 3728 */ 3729 if (inp) { 3730 /* copy old policy into new socket's */ 3731 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3732 printf("tcp_input: could not copy policy\n"); 3733 } 3734 #ifdef INET6 3735 else if (in6p) { 3736 /* copy old policy into new socket's */ 3737 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3738 in6p->in6p_sp)) 3739 printf("tcp_input: could not copy policy\n"); 3740 } 3741 #endif 3742 #endif 3743 3744 /* 3745 * Give the new socket our cached route reference. 3746 */ 3747 if (inp) { 3748 rtcache_copy(&inp->inp_route, &sc->sc_route); 3749 rtcache_free(&sc->sc_route); 3750 } 3751 #ifdef INET6 3752 else { 3753 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3754 rtcache_free(&sc->sc_route); 3755 } 3756 #endif 3757 3758 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3759 if (am == NULL) 3760 goto resetandabort; 3761 MCLAIM(am, &tcp_mowner); 3762 am->m_len = src->sa_len; 3763 bcopy(src, mtod(am, void *), src->sa_len); 3764 if (inp) { 3765 if (in_pcbconnect(inp, am, &lwp0)) { 3766 (void) m_free(am); 3767 goto resetandabort; 3768 } 3769 } 3770 #ifdef INET6 3771 else if (in6p) { 3772 if (src->sa_family == AF_INET) { 3773 /* IPv4 packet to AF_INET6 socket */ 3774 struct sockaddr_in6 *sin6; 3775 sin6 = mtod(am, struct sockaddr_in6 *); 3776 am->m_len = sizeof(*sin6); 3777 bzero(sin6, sizeof(*sin6)); 3778 sin6->sin6_family = AF_INET6; 3779 sin6->sin6_len = sizeof(*sin6); 3780 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3781 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3782 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3783 &sin6->sin6_addr.s6_addr32[3], 3784 sizeof(sin6->sin6_addr.s6_addr32[3])); 3785 } 3786 if (in6_pcbconnect(in6p, am, NULL)) { 3787 (void) m_free(am); 3788 goto resetandabort; 3789 } 3790 } 3791 #endif 3792 else { 3793 (void) m_free(am); 3794 goto resetandabort; 3795 } 3796 (void) m_free(am); 3797 3798 if (inp) 3799 tp = intotcpcb(inp); 3800 #ifdef INET6 3801 else if (in6p) 3802 tp = in6totcpcb(in6p); 3803 #endif 3804 else 3805 tp = NULL; 3806 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3807 if (sc->sc_request_r_scale != 15) { 3808 tp->requested_s_scale = sc->sc_requested_s_scale; 3809 tp->request_r_scale = sc->sc_request_r_scale; 3810 tp->snd_scale = sc->sc_requested_s_scale; 3811 tp->rcv_scale = sc->sc_request_r_scale; 3812 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3813 } 3814 if (sc->sc_flags & SCF_TIMESTAMP) 3815 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3816 tp->ts_timebase = sc->sc_timebase; 3817 3818 tp->t_template = tcp_template(tp); 3819 if (tp->t_template == 0) { 3820 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3821 so = NULL; 3822 m_freem(m); 3823 goto abort; 3824 } 3825 3826 tp->iss = sc->sc_iss; 3827 tp->irs = sc->sc_irs; 3828 tcp_sendseqinit(tp); 3829 tcp_rcvseqinit(tp); 3830 tp->t_state = TCPS_SYN_RECEIVED; 3831 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 3832 TCP_STATINC(TCP_STAT_ACCEPTS); 3833 3834 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3835 tp->t_flags |= TF_WILL_SACK; 3836 3837 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3838 tp->t_flags |= TF_ECN_PERMIT; 3839 3840 #ifdef TCP_SIGNATURE 3841 if (sc->sc_flags & SCF_SIGNATURE) 3842 tp->t_flags |= TF_SIGNATURE; 3843 #endif 3844 3845 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3846 tp->t_ourmss = sc->sc_ourmaxseg; 3847 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3848 3849 /* 3850 * Initialize the initial congestion window. If we 3851 * had to retransmit the SYN,ACK, we must initialize cwnd 3852 * to 1 segment (i.e. the Loss Window). 3853 */ 3854 if (sc->sc_rxtshift) 3855 tp->snd_cwnd = tp->t_peermss; 3856 else { 3857 int ss = tcp_init_win; 3858 #ifdef INET 3859 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3860 ss = tcp_init_win_local; 3861 #endif 3862 #ifdef INET6 3863 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3864 ss = tcp_init_win_local; 3865 #endif 3866 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3867 } 3868 3869 tcp_rmx_rtt(tp); 3870 tp->snd_wl1 = sc->sc_irs; 3871 tp->rcv_up = sc->sc_irs + 1; 3872 3873 /* 3874 * This is what whould have happened in tcp_output() when 3875 * the SYN,ACK was sent. 3876 */ 3877 tp->snd_up = tp->snd_una; 3878 tp->snd_max = tp->snd_nxt = tp->iss+1; 3879 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3880 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3881 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3882 tp->last_ack_sent = tp->rcv_nxt; 3883 tp->t_partialacks = -1; 3884 tp->t_dupacks = 0; 3885 3886 TCP_STATINC(TCP_STAT_SC_COMPLETED); 3887 s = splsoftnet(); 3888 syn_cache_put(sc); 3889 splx(s); 3890 return (so); 3891 3892 resetandabort: 3893 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3894 abort: 3895 if (so != NULL) 3896 (void) soabort(so); 3897 s = splsoftnet(); 3898 syn_cache_put(sc); 3899 splx(s); 3900 TCP_STATINC(TCP_STAT_SC_ABORTED); 3901 return ((struct socket *)(-1)); 3902 } 3903 3904 /* 3905 * This function is called when we get a RST for a 3906 * non-existent connection, so that we can see if the 3907 * connection is in the syn cache. If it is, zap it. 3908 */ 3909 3910 void 3911 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3912 { 3913 struct syn_cache *sc; 3914 struct syn_cache_head *scp; 3915 int s = splsoftnet(); 3916 3917 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3918 splx(s); 3919 return; 3920 } 3921 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3922 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3923 splx(s); 3924 return; 3925 } 3926 syn_cache_rm(sc); 3927 TCP_STATINC(TCP_STAT_SC_RESET); 3928 syn_cache_put(sc); /* calls pool_put but see spl above */ 3929 splx(s); 3930 } 3931 3932 void 3933 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3934 struct tcphdr *th) 3935 { 3936 struct syn_cache *sc; 3937 struct syn_cache_head *scp; 3938 int s; 3939 3940 s = splsoftnet(); 3941 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3942 splx(s); 3943 return; 3944 } 3945 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3946 if (ntohl (th->th_seq) != sc->sc_iss) { 3947 splx(s); 3948 return; 3949 } 3950 3951 /* 3952 * If we've retransmitted 3 times and this is our second error, 3953 * we remove the entry. Otherwise, we allow it to continue on. 3954 * This prevents us from incorrectly nuking an entry during a 3955 * spurious network outage. 3956 * 3957 * See tcp_notify(). 3958 */ 3959 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3960 sc->sc_flags |= SCF_UNREACH; 3961 splx(s); 3962 return; 3963 } 3964 3965 syn_cache_rm(sc); 3966 TCP_STATINC(TCP_STAT_SC_UNREACH); 3967 syn_cache_put(sc); /* calls pool_put but see spl above */ 3968 splx(s); 3969 } 3970 3971 /* 3972 * Given a LISTEN socket and an inbound SYN request, add 3973 * this to the syn cache, and send back a segment: 3974 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3975 * to the source. 3976 * 3977 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3978 * Doing so would require that we hold onto the data and deliver it 3979 * to the application. However, if we are the target of a SYN-flood 3980 * DoS attack, an attacker could send data which would eventually 3981 * consume all available buffer space if it were ACKed. By not ACKing 3982 * the data, we avoid this DoS scenario. 3983 */ 3984 3985 int 3986 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3987 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3988 int optlen, struct tcp_opt_info *oi) 3989 { 3990 struct tcpcb tb, *tp; 3991 long win; 3992 struct syn_cache *sc; 3993 struct syn_cache_head *scp; 3994 struct mbuf *ipopts; 3995 struct tcp_opt_info opti; 3996 int s; 3997 3998 tp = sototcpcb(so); 3999 4000 bzero(&opti, sizeof(opti)); 4001 4002 /* 4003 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4004 * 4005 * Note this check is performed in tcp_input() very early on. 4006 */ 4007 4008 /* 4009 * Initialize some local state. 4010 */ 4011 win = sbspace(&so->so_rcv); 4012 if (win > TCP_MAXWIN) 4013 win = TCP_MAXWIN; 4014 4015 switch (src->sa_family) { 4016 #ifdef INET 4017 case AF_INET: 4018 /* 4019 * Remember the IP options, if any. 4020 */ 4021 ipopts = ip_srcroute(); 4022 break; 4023 #endif 4024 default: 4025 ipopts = NULL; 4026 } 4027 4028 #ifdef TCP_SIGNATURE 4029 if (optp || (tp->t_flags & TF_SIGNATURE)) 4030 #else 4031 if (optp) 4032 #endif 4033 { 4034 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4035 #ifdef TCP_SIGNATURE 4036 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4037 #endif 4038 tb.t_state = TCPS_LISTEN; 4039 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4040 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4041 return (0); 4042 } else 4043 tb.t_flags = 0; 4044 4045 /* 4046 * See if we already have an entry for this connection. 4047 * If we do, resend the SYN,ACK. We do not count this 4048 * as a retransmission (XXX though maybe we should). 4049 */ 4050 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4051 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4052 if (ipopts) { 4053 /* 4054 * If we were remembering a previous source route, 4055 * forget it and use the new one we've been given. 4056 */ 4057 if (sc->sc_ipopts) 4058 (void) m_free(sc->sc_ipopts); 4059 sc->sc_ipopts = ipopts; 4060 } 4061 sc->sc_timestamp = tb.ts_recent; 4062 if (syn_cache_respond(sc, m) == 0) { 4063 uint64_t *tcps = TCP_STAT_GETREF(); 4064 tcps[TCP_STAT_SNDACKS]++; 4065 tcps[TCP_STAT_SNDTOTAL]++; 4066 TCP_STAT_PUTREF(); 4067 } 4068 return (1); 4069 } 4070 4071 s = splsoftnet(); 4072 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4073 splx(s); 4074 if (sc == NULL) { 4075 if (ipopts) 4076 (void) m_free(ipopts); 4077 return (0); 4078 } 4079 4080 /* 4081 * Fill in the cache, and put the necessary IP and TCP 4082 * options into the reply. 4083 */ 4084 bzero(sc, sizeof(struct syn_cache)); 4085 callout_init(&sc->sc_timer, 0); 4086 bcopy(src, &sc->sc_src, src->sa_len); 4087 bcopy(dst, &sc->sc_dst, dst->sa_len); 4088 sc->sc_flags = 0; 4089 sc->sc_ipopts = ipopts; 4090 sc->sc_irs = th->th_seq; 4091 switch (src->sa_family) { 4092 #ifdef INET 4093 case AF_INET: 4094 { 4095 struct sockaddr_in *srcin = (void *) src; 4096 struct sockaddr_in *dstin = (void *) dst; 4097 4098 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4099 &srcin->sin_addr, dstin->sin_port, 4100 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4101 break; 4102 } 4103 #endif /* INET */ 4104 #ifdef INET6 4105 case AF_INET6: 4106 { 4107 struct sockaddr_in6 *srcin6 = (void *) src; 4108 struct sockaddr_in6 *dstin6 = (void *) dst; 4109 4110 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4111 &srcin6->sin6_addr, dstin6->sin6_port, 4112 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4113 break; 4114 } 4115 #endif /* INET6 */ 4116 } 4117 sc->sc_peermaxseg = oi->maxseg; 4118 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4119 m->m_pkthdr.rcvif : NULL, 4120 sc->sc_src.sa.sa_family); 4121 sc->sc_win = win; 4122 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4123 sc->sc_timestamp = tb.ts_recent; 4124 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4125 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4126 sc->sc_flags |= SCF_TIMESTAMP; 4127 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4128 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4129 sc->sc_requested_s_scale = tb.requested_s_scale; 4130 sc->sc_request_r_scale = 0; 4131 /* 4132 * Pick the smallest possible scaling factor that 4133 * will still allow us to scale up to sb_max. 4134 * 4135 * We do this because there are broken firewalls that 4136 * will corrupt the window scale option, leading to 4137 * the other endpoint believing that our advertised 4138 * window is unscaled. At scale factors larger than 4139 * 5 the unscaled window will drop below 1500 bytes, 4140 * leading to serious problems when traversing these 4141 * broken firewalls. 4142 * 4143 * With the default sbmax of 256K, a scale factor 4144 * of 3 will be chosen by this algorithm. Those who 4145 * choose a larger sbmax should watch out 4146 * for the compatiblity problems mentioned above. 4147 * 4148 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4149 * or <SYN,ACK>) segment itself is never scaled. 4150 */ 4151 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4152 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4153 sc->sc_request_r_scale++; 4154 } else { 4155 sc->sc_requested_s_scale = 15; 4156 sc->sc_request_r_scale = 15; 4157 } 4158 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4159 sc->sc_flags |= SCF_SACK_PERMIT; 4160 4161 /* 4162 * ECN setup packet recieved. 4163 */ 4164 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4165 sc->sc_flags |= SCF_ECN_PERMIT; 4166 4167 #ifdef TCP_SIGNATURE 4168 if (tb.t_flags & TF_SIGNATURE) 4169 sc->sc_flags |= SCF_SIGNATURE; 4170 #endif 4171 sc->sc_tp = tp; 4172 if (syn_cache_respond(sc, m) == 0) { 4173 uint64_t *tcps = TCP_STAT_GETREF(); 4174 tcps[TCP_STAT_SNDACKS]++; 4175 tcps[TCP_STAT_SNDTOTAL]++; 4176 TCP_STAT_PUTREF(); 4177 syn_cache_insert(sc, tp); 4178 } else { 4179 s = splsoftnet(); 4180 syn_cache_put(sc); 4181 splx(s); 4182 TCP_STATINC(TCP_STAT_SC_DROPPED); 4183 } 4184 return (1); 4185 } 4186 4187 int 4188 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4189 { 4190 #ifdef INET6 4191 struct rtentry *rt; 4192 #endif 4193 struct route *ro; 4194 u_int8_t *optp; 4195 int optlen, error; 4196 u_int16_t tlen; 4197 struct ip *ip = NULL; 4198 #ifdef INET6 4199 struct ip6_hdr *ip6 = NULL; 4200 #endif 4201 struct tcpcb *tp = NULL; 4202 struct tcphdr *th; 4203 u_int hlen; 4204 struct socket *so; 4205 4206 ro = &sc->sc_route; 4207 switch (sc->sc_src.sa.sa_family) { 4208 case AF_INET: 4209 hlen = sizeof(struct ip); 4210 break; 4211 #ifdef INET6 4212 case AF_INET6: 4213 hlen = sizeof(struct ip6_hdr); 4214 break; 4215 #endif 4216 default: 4217 if (m) 4218 m_freem(m); 4219 return (EAFNOSUPPORT); 4220 } 4221 4222 /* Compute the size of the TCP options. */ 4223 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4224 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4225 #ifdef TCP_SIGNATURE 4226 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4227 #endif 4228 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4229 4230 tlen = hlen + sizeof(struct tcphdr) + optlen; 4231 4232 /* 4233 * Create the IP+TCP header from scratch. 4234 */ 4235 if (m) 4236 m_freem(m); 4237 #ifdef DIAGNOSTIC 4238 if (max_linkhdr + tlen > MCLBYTES) 4239 return (ENOBUFS); 4240 #endif 4241 MGETHDR(m, M_DONTWAIT, MT_DATA); 4242 if (m && tlen > MHLEN) { 4243 MCLGET(m, M_DONTWAIT); 4244 if ((m->m_flags & M_EXT) == 0) { 4245 m_freem(m); 4246 m = NULL; 4247 } 4248 } 4249 if (m == NULL) 4250 return (ENOBUFS); 4251 MCLAIM(m, &tcp_tx_mowner); 4252 4253 /* Fixup the mbuf. */ 4254 m->m_data += max_linkhdr; 4255 m->m_len = m->m_pkthdr.len = tlen; 4256 if (sc->sc_tp) { 4257 tp = sc->sc_tp; 4258 if (tp->t_inpcb) 4259 so = tp->t_inpcb->inp_socket; 4260 #ifdef INET6 4261 else if (tp->t_in6pcb) 4262 so = tp->t_in6pcb->in6p_socket; 4263 #endif 4264 else 4265 so = NULL; 4266 } else 4267 so = NULL; 4268 m->m_pkthdr.rcvif = NULL; 4269 memset(mtod(m, u_char *), 0, tlen); 4270 4271 switch (sc->sc_src.sa.sa_family) { 4272 case AF_INET: 4273 ip = mtod(m, struct ip *); 4274 ip->ip_v = 4; 4275 ip->ip_dst = sc->sc_src.sin.sin_addr; 4276 ip->ip_src = sc->sc_dst.sin.sin_addr; 4277 ip->ip_p = IPPROTO_TCP; 4278 th = (struct tcphdr *)(ip + 1); 4279 th->th_dport = sc->sc_src.sin.sin_port; 4280 th->th_sport = sc->sc_dst.sin.sin_port; 4281 break; 4282 #ifdef INET6 4283 case AF_INET6: 4284 ip6 = mtod(m, struct ip6_hdr *); 4285 ip6->ip6_vfc = IPV6_VERSION; 4286 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4287 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4288 ip6->ip6_nxt = IPPROTO_TCP; 4289 /* ip6_plen will be updated in ip6_output() */ 4290 th = (struct tcphdr *)(ip6 + 1); 4291 th->th_dport = sc->sc_src.sin6.sin6_port; 4292 th->th_sport = sc->sc_dst.sin6.sin6_port; 4293 break; 4294 #endif 4295 default: 4296 th = NULL; 4297 } 4298 4299 th->th_seq = htonl(sc->sc_iss); 4300 th->th_ack = htonl(sc->sc_irs + 1); 4301 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4302 th->th_flags = TH_SYN|TH_ACK; 4303 th->th_win = htons(sc->sc_win); 4304 /* th_sum already 0 */ 4305 /* th_urp already 0 */ 4306 4307 /* Tack on the TCP options. */ 4308 optp = (u_int8_t *)(th + 1); 4309 *optp++ = TCPOPT_MAXSEG; 4310 *optp++ = 4; 4311 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4312 *optp++ = sc->sc_ourmaxseg & 0xff; 4313 4314 if (sc->sc_request_r_scale != 15) { 4315 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4316 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4317 sc->sc_request_r_scale); 4318 optp += 4; 4319 } 4320 4321 if (sc->sc_flags & SCF_TIMESTAMP) { 4322 u_int32_t *lp = (u_int32_t *)(optp); 4323 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4324 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4325 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4326 *lp = htonl(sc->sc_timestamp); 4327 optp += TCPOLEN_TSTAMP_APPA; 4328 } 4329 4330 if (sc->sc_flags & SCF_SACK_PERMIT) { 4331 u_int8_t *p = optp; 4332 4333 /* Let the peer know that we will SACK. */ 4334 p[0] = TCPOPT_SACK_PERMITTED; 4335 p[1] = 2; 4336 p[2] = TCPOPT_NOP; 4337 p[3] = TCPOPT_NOP; 4338 optp += 4; 4339 } 4340 4341 /* 4342 * Send ECN SYN-ACK setup packet. 4343 * Routes can be asymetric, so, even if we receive a packet 4344 * with ECE and CWR set, we must not assume no one will block 4345 * the ECE packet we are about to send. 4346 */ 4347 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4348 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4349 th->th_flags |= TH_ECE; 4350 TCP_STATINC(TCP_STAT_ECN_SHS); 4351 4352 /* 4353 * draft-ietf-tcpm-ecnsyn-00.txt 4354 * 4355 * "[...] a TCP node MAY respond to an ECN-setup 4356 * SYN packet by setting ECT in the responding 4357 * ECN-setup SYN/ACK packet, indicating to routers 4358 * that the SYN/ACK packet is ECN-Capable. 4359 * This allows a congested router along the path 4360 * to mark the packet instead of dropping the 4361 * packet as an indication of congestion." 4362 * 4363 * "[...] There can be a great benefit in setting 4364 * an ECN-capable codepoint in SYN/ACK packets [...] 4365 * Congestion is most likely to occur in 4366 * the server-to-client direction. As a result, 4367 * setting an ECN-capable codepoint in SYN/ACK 4368 * packets can reduce the occurence of three-second 4369 * retransmit timeouts resulting from the drop 4370 * of SYN/ACK packets." 4371 * 4372 * Page 4 and 6, January 2006. 4373 */ 4374 4375 switch (sc->sc_src.sa.sa_family) { 4376 #ifdef INET 4377 case AF_INET: 4378 ip->ip_tos |= IPTOS_ECN_ECT0; 4379 break; 4380 #endif 4381 #ifdef INET6 4382 case AF_INET6: 4383 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4384 break; 4385 #endif 4386 } 4387 TCP_STATINC(TCP_STAT_ECN_ECT); 4388 } 4389 4390 #ifdef TCP_SIGNATURE 4391 if (sc->sc_flags & SCF_SIGNATURE) { 4392 struct secasvar *sav; 4393 u_int8_t *sigp; 4394 4395 sav = tcp_signature_getsav(m, th); 4396 4397 if (sav == NULL) { 4398 if (m) 4399 m_freem(m); 4400 return (EPERM); 4401 } 4402 4403 *optp++ = TCPOPT_SIGNATURE; 4404 *optp++ = TCPOLEN_SIGNATURE; 4405 sigp = optp; 4406 bzero(optp, TCP_SIGLEN); 4407 optp += TCP_SIGLEN; 4408 *optp++ = TCPOPT_NOP; 4409 *optp++ = TCPOPT_EOL; 4410 4411 (void)tcp_signature(m, th, hlen, sav, sigp); 4412 4413 key_sa_recordxfer(sav, m); 4414 #ifdef FAST_IPSEC 4415 KEY_FREESAV(&sav); 4416 #else 4417 key_freesav(sav); 4418 #endif 4419 } 4420 #endif 4421 4422 /* Compute the packet's checksum. */ 4423 switch (sc->sc_src.sa.sa_family) { 4424 case AF_INET: 4425 ip->ip_len = htons(tlen - hlen); 4426 th->th_sum = 0; 4427 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4428 break; 4429 #ifdef INET6 4430 case AF_INET6: 4431 ip6->ip6_plen = htons(tlen - hlen); 4432 th->th_sum = 0; 4433 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4434 break; 4435 #endif 4436 } 4437 4438 /* 4439 * Fill in some straggling IP bits. Note the stack expects 4440 * ip_len to be in host order, for convenience. 4441 */ 4442 switch (sc->sc_src.sa.sa_family) { 4443 #ifdef INET 4444 case AF_INET: 4445 ip->ip_len = htons(tlen); 4446 ip->ip_ttl = ip_defttl; 4447 /* XXX tos? */ 4448 break; 4449 #endif 4450 #ifdef INET6 4451 case AF_INET6: 4452 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4453 ip6->ip6_vfc |= IPV6_VERSION; 4454 ip6->ip6_plen = htons(tlen - hlen); 4455 /* ip6_hlim will be initialized afterwards */ 4456 /* XXX flowlabel? */ 4457 break; 4458 #endif 4459 } 4460 4461 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4462 tp = sc->sc_tp; 4463 4464 switch (sc->sc_src.sa.sa_family) { 4465 #ifdef INET 4466 case AF_INET: 4467 error = ip_output(m, sc->sc_ipopts, ro, 4468 (ip_mtudisc ? IP_MTUDISC : 0), 4469 (struct ip_moptions *)NULL, so); 4470 break; 4471 #endif 4472 #ifdef INET6 4473 case AF_INET6: 4474 ip6->ip6_hlim = in6_selecthlim(NULL, 4475 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4476 : NULL); 4477 4478 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4479 break; 4480 #endif 4481 default: 4482 error = EAFNOSUPPORT; 4483 break; 4484 } 4485 return (error); 4486 } 4487