1 /* $NetBSD: tcp_input.c,v 1.287 2008/04/28 20:24:09 martin 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 * 93 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 94 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 95 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 96 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 97 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 98 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 99 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 100 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 101 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 102 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 103 * POSSIBILITY OF SUCH DAMAGE. 104 */ 105 106 /* 107 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 108 * The Regents of the University of California. All rights reserved. 109 * 110 * Redistribution and use in source and binary forms, with or without 111 * modification, are permitted provided that the following conditions 112 * are met: 113 * 1. Redistributions of source code must retain the above copyright 114 * notice, this list of conditions and the following disclaimer. 115 * 2. Redistributions in binary form must reproduce the above copyright 116 * notice, this list of conditions and the following disclaimer in the 117 * documentation and/or other materials provided with the distribution. 118 * 3. Neither the name of the University nor the names of its contributors 119 * may be used to endorse or promote products derived from this software 120 * without specific prior written permission. 121 * 122 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 123 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 124 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 125 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 126 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 127 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 128 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 129 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 130 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 131 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 132 * SUCH DAMAGE. 133 * 134 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 135 */ 136 137 /* 138 * TODO list for SYN cache stuff: 139 * 140 * Find room for a "state" field, which is needed to keep a 141 * compressed state for TIME_WAIT TCBs. It's been noted already 142 * that this is fairly important for very high-volume web and 143 * mail servers, which use a large number of short-lived 144 * connections. 145 */ 146 147 #include <sys/cdefs.h> 148 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.287 2008/04/28 20:24:09 martin Exp $"); 149 150 #include "opt_inet.h" 151 #include "opt_ipsec.h" 152 #include "opt_inet_csum.h" 153 #include "opt_tcp_debug.h" 154 155 #include <sys/param.h> 156 #include <sys/systm.h> 157 #include <sys/malloc.h> 158 #include <sys/mbuf.h> 159 #include <sys/protosw.h> 160 #include <sys/socket.h> 161 #include <sys/socketvar.h> 162 #include <sys/errno.h> 163 #include <sys/syslog.h> 164 #include <sys/pool.h> 165 #include <sys/domain.h> 166 #include <sys/kernel.h> 167 #ifdef TCP_SIGNATURE 168 #include <sys/md5.h> 169 #endif 170 #include <sys/lwp.h> /* for lwp0 */ 171 172 #include <net/if.h> 173 #include <net/route.h> 174 #include <net/if_types.h> 175 176 #include <netinet/in.h> 177 #include <netinet/in_systm.h> 178 #include <netinet/ip.h> 179 #include <netinet/in_pcb.h> 180 #include <netinet/in_var.h> 181 #include <netinet/ip_var.h> 182 #include <netinet/in_offload.h> 183 184 #ifdef INET6 185 #ifndef INET 186 #include <netinet/in.h> 187 #endif 188 #include <netinet/ip6.h> 189 #include <netinet6/ip6_var.h> 190 #include <netinet6/in6_pcb.h> 191 #include <netinet6/ip6_var.h> 192 #include <netinet6/in6_var.h> 193 #include <netinet/icmp6.h> 194 #include <netinet6/nd6.h> 195 #ifdef TCP_SIGNATURE 196 #include <netinet6/scope6_var.h> 197 #endif 198 #endif 199 200 #ifndef INET6 201 /* always need ip6.h for IP6_EXTHDR_GET */ 202 #include <netinet/ip6.h> 203 #endif 204 205 #include <netinet/tcp.h> 206 #include <netinet/tcp_fsm.h> 207 #include <netinet/tcp_seq.h> 208 #include <netinet/tcp_timer.h> 209 #include <netinet/tcp_var.h> 210 #include <netinet/tcp_private.h> 211 #include <netinet/tcpip.h> 212 #include <netinet/tcp_congctl.h> 213 #include <netinet/tcp_debug.h> 214 215 #include <machine/stdarg.h> 216 217 #ifdef IPSEC 218 #include <netinet6/ipsec.h> 219 #include <netinet6/ipsec_private.h> 220 #include <netkey/key.h> 221 #endif /*IPSEC*/ 222 #ifdef INET6 223 #include "faith.h" 224 #if defined(NFAITH) && NFAITH > 0 225 #include <net/if_faith.h> 226 #endif 227 #endif /* IPSEC */ 228 229 #ifdef FAST_IPSEC 230 #include <netipsec/ipsec.h> 231 #include <netipsec/ipsec_var.h> 232 #include <netipsec/ipsec_private.h> 233 #include <netipsec/key.h> 234 #ifdef INET6 235 #include <netipsec/ipsec6.h> 236 #endif 237 #endif /* FAST_IPSEC*/ 238 239 int tcprexmtthresh = 3; 240 int tcp_log_refused; 241 242 int tcp_do_autorcvbuf = 0; 243 int tcp_autorcvbuf_inc = 16 * 1024; 244 int tcp_autorcvbuf_max = 256 * 1024; 245 246 static int tcp_rst_ppslim_count = 0; 247 static struct timeval tcp_rst_ppslim_last; 248 static int tcp_ackdrop_ppslim_count = 0; 249 static struct timeval tcp_ackdrop_ppslim_last; 250 251 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 252 253 /* for modulo comparisons of timestamps */ 254 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 255 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 256 257 /* 258 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 259 */ 260 #ifdef INET6 261 static inline void 262 nd6_hint(struct tcpcb *tp) 263 { 264 struct rtentry *rt; 265 266 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 267 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 268 nd6_nud_hint(rt, NULL, 0); 269 } 270 #else 271 static inline void 272 nd6_hint(struct tcpcb *tp) 273 { 274 } 275 #endif 276 277 /* 278 * Compute ACK transmission behavior. Delay the ACK unless 279 * we have already delayed an ACK (must send an ACK every two segments). 280 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 281 * option is enabled. 282 */ 283 static void 284 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 285 { 286 287 if (tp->t_flags & TF_DELACK || 288 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 289 tp->t_flags |= TF_ACKNOW; 290 else 291 TCP_SET_DELACK(tp); 292 } 293 294 static void 295 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 296 { 297 298 /* 299 * If we had a pending ICMP message that refers to data that have 300 * just been acknowledged, disregard the recorded ICMP message. 301 */ 302 if ((tp->t_flags & TF_PMTUD_PEND) && 303 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 304 tp->t_flags &= ~TF_PMTUD_PEND; 305 306 /* 307 * Keep track of the largest chunk of data 308 * acknowledged since last PMTU update 309 */ 310 if (tp->t_pmtud_mss_acked < acked) 311 tp->t_pmtud_mss_acked = acked; 312 } 313 314 /* 315 * Convert TCP protocol fields to host order for easier processing. 316 */ 317 static void 318 tcp_fields_to_host(struct tcphdr *th) 319 { 320 321 NTOHL(th->th_seq); 322 NTOHL(th->th_ack); 323 NTOHS(th->th_win); 324 NTOHS(th->th_urp); 325 } 326 327 /* 328 * ... and reverse the above. 329 */ 330 static void 331 tcp_fields_to_net(struct tcphdr *th) 332 { 333 334 HTONL(th->th_seq); 335 HTONL(th->th_ack); 336 HTONS(th->th_win); 337 HTONS(th->th_urp); 338 } 339 340 #ifdef TCP_CSUM_COUNTERS 341 #include <sys/device.h> 342 343 #if defined(INET) 344 extern struct evcnt tcp_hwcsum_ok; 345 extern struct evcnt tcp_hwcsum_bad; 346 extern struct evcnt tcp_hwcsum_data; 347 extern struct evcnt tcp_swcsum; 348 #endif /* defined(INET) */ 349 #if defined(INET6) 350 extern struct evcnt tcp6_hwcsum_ok; 351 extern struct evcnt tcp6_hwcsum_bad; 352 extern struct evcnt tcp6_hwcsum_data; 353 extern struct evcnt tcp6_swcsum; 354 #endif /* defined(INET6) */ 355 356 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 357 358 #else 359 360 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 361 362 #endif /* TCP_CSUM_COUNTERS */ 363 364 #ifdef TCP_REASS_COUNTERS 365 #include <sys/device.h> 366 367 extern struct evcnt tcp_reass_; 368 extern struct evcnt tcp_reass_empty; 369 extern struct evcnt tcp_reass_iteration[8]; 370 extern struct evcnt tcp_reass_prependfirst; 371 extern struct evcnt tcp_reass_prepend; 372 extern struct evcnt tcp_reass_insert; 373 extern struct evcnt tcp_reass_inserttail; 374 extern struct evcnt tcp_reass_append; 375 extern struct evcnt tcp_reass_appendtail; 376 extern struct evcnt tcp_reass_overlaptail; 377 extern struct evcnt tcp_reass_overlapfront; 378 extern struct evcnt tcp_reass_segdup; 379 extern struct evcnt tcp_reass_fragdup; 380 381 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 382 383 #else 384 385 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 386 387 #endif /* TCP_REASS_COUNTERS */ 388 389 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 390 int *); 391 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 392 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 393 394 #ifdef INET 395 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 396 #endif 397 #ifdef INET6 398 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 399 #endif 400 401 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 402 403 #if defined(MBUFTRACE) 404 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 405 #endif /* defined(MBUFTRACE) */ 406 407 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 408 NULL, IPL_VM); 409 410 struct ipqent * 411 tcpipqent_alloc(void) 412 { 413 struct ipqent *ipqe; 414 int s; 415 416 s = splvm(); 417 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 418 splx(s); 419 420 return ipqe; 421 } 422 423 void 424 tcpipqent_free(struct ipqent *ipqe) 425 { 426 int s; 427 428 s = splvm(); 429 pool_put(&tcpipqent_pool, ipqe); 430 splx(s); 431 } 432 433 static int 434 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 435 { 436 struct ipqent *p, *q, *nq, *tiqe = NULL; 437 struct socket *so = NULL; 438 int pkt_flags; 439 tcp_seq pkt_seq; 440 unsigned pkt_len; 441 u_long rcvpartdupbyte = 0; 442 u_long rcvoobyte; 443 #ifdef TCP_REASS_COUNTERS 444 u_int count = 0; 445 #endif 446 uint64_t *tcps; 447 448 if (tp->t_inpcb) 449 so = tp->t_inpcb->inp_socket; 450 #ifdef INET6 451 else if (tp->t_in6pcb) 452 so = tp->t_in6pcb->in6p_socket; 453 #endif 454 455 TCP_REASS_LOCK_CHECK(tp); 456 457 /* 458 * Call with th==0 after become established to 459 * force pre-ESTABLISHED data up to user socket. 460 */ 461 if (th == 0) 462 goto present; 463 464 m_claimm(m, &tcp_reass_mowner); 465 466 rcvoobyte = *tlen; 467 /* 468 * Copy these to local variables because the tcpiphdr 469 * gets munged while we are collapsing mbufs. 470 */ 471 pkt_seq = th->th_seq; 472 pkt_len = *tlen; 473 pkt_flags = th->th_flags; 474 475 TCP_REASS_COUNTER_INCR(&tcp_reass_); 476 477 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 478 /* 479 * When we miss a packet, the vast majority of time we get 480 * packets that follow it in order. So optimize for that. 481 */ 482 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 483 p->ipqe_len += pkt_len; 484 p->ipqe_flags |= pkt_flags; 485 m_cat(p->ipre_mlast, m); 486 TRAVERSE(p->ipre_mlast); 487 m = NULL; 488 tiqe = p; 489 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 490 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 491 goto skip_replacement; 492 } 493 /* 494 * While we're here, if the pkt is completely beyond 495 * anything we have, just insert it at the tail. 496 */ 497 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 498 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 499 goto insert_it; 500 } 501 } 502 503 q = TAILQ_FIRST(&tp->segq); 504 505 if (q != NULL) { 506 /* 507 * If this segment immediately precedes the first out-of-order 508 * block, simply slap the segment in front of it and (mostly) 509 * skip the complicated logic. 510 */ 511 if (pkt_seq + pkt_len == q->ipqe_seq) { 512 q->ipqe_seq = pkt_seq; 513 q->ipqe_len += pkt_len; 514 q->ipqe_flags |= pkt_flags; 515 m_cat(m, q->ipqe_m); 516 q->ipqe_m = m; 517 q->ipre_mlast = m; /* last mbuf may have changed */ 518 TRAVERSE(q->ipre_mlast); 519 tiqe = q; 520 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 521 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 522 goto skip_replacement; 523 } 524 } else { 525 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 526 } 527 528 /* 529 * Find a segment which begins after this one does. 530 */ 531 for (p = NULL; q != NULL; q = nq) { 532 nq = TAILQ_NEXT(q, ipqe_q); 533 #ifdef TCP_REASS_COUNTERS 534 count++; 535 #endif 536 /* 537 * If the received segment is just right after this 538 * fragment, merge the two together and then check 539 * for further overlaps. 540 */ 541 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 542 #ifdef TCPREASS_DEBUG 543 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 544 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 545 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 546 #endif 547 pkt_len += q->ipqe_len; 548 pkt_flags |= q->ipqe_flags; 549 pkt_seq = q->ipqe_seq; 550 m_cat(q->ipre_mlast, m); 551 TRAVERSE(q->ipre_mlast); 552 m = q->ipqe_m; 553 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 554 goto free_ipqe; 555 } 556 /* 557 * If the received segment is completely past this 558 * fragment, we need to go the next fragment. 559 */ 560 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 561 p = q; 562 continue; 563 } 564 /* 565 * If the fragment is past the received segment, 566 * it (or any following) can't be concatenated. 567 */ 568 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 569 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 570 break; 571 } 572 573 /* 574 * We've received all the data in this segment before. 575 * mark it as a duplicate and return. 576 */ 577 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 578 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 579 tcps = TCP_STAT_GETREF(); 580 tcps[TCP_STAT_RCVDUPPACK]++; 581 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 582 TCP_STAT_PUTREF(); 583 tcp_new_dsack(tp, pkt_seq, pkt_len); 584 m_freem(m); 585 if (tiqe != NULL) { 586 tcpipqent_free(tiqe); 587 } 588 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 589 return (0); 590 } 591 /* 592 * Received segment completely overlaps this fragment 593 * so we drop the fragment (this keeps the temporal 594 * ordering of segments correct). 595 */ 596 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 597 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 598 rcvpartdupbyte += q->ipqe_len; 599 m_freem(q->ipqe_m); 600 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 601 goto free_ipqe; 602 } 603 /* 604 * RX'ed segment extends past the end of the 605 * fragment. Drop the overlapping bytes. Then 606 * merge the fragment and segment then treat as 607 * a longer received packet. 608 */ 609 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 610 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 611 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 612 #ifdef TCPREASS_DEBUG 613 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 614 tp, overlap, 615 pkt_seq, pkt_seq + pkt_len, pkt_len); 616 #endif 617 m_adj(m, overlap); 618 rcvpartdupbyte += overlap; 619 m_cat(q->ipre_mlast, m); 620 TRAVERSE(q->ipre_mlast); 621 m = q->ipqe_m; 622 pkt_seq = q->ipqe_seq; 623 pkt_len += q->ipqe_len - overlap; 624 rcvoobyte -= overlap; 625 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 626 goto free_ipqe; 627 } 628 /* 629 * RX'ed segment extends past the front of the 630 * fragment. Drop the overlapping bytes on the 631 * received packet. The packet will then be 632 * contatentated with this fragment a bit later. 633 */ 634 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 635 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 636 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 637 #ifdef TCPREASS_DEBUG 638 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 639 tp, overlap, 640 pkt_seq, pkt_seq + pkt_len, pkt_len); 641 #endif 642 m_adj(m, -overlap); 643 pkt_len -= overlap; 644 rcvpartdupbyte += overlap; 645 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 646 rcvoobyte -= overlap; 647 } 648 /* 649 * If the received segment immediates precedes this 650 * fragment then tack the fragment onto this segment 651 * and reinsert the data. 652 */ 653 if (q->ipqe_seq == pkt_seq + pkt_len) { 654 #ifdef TCPREASS_DEBUG 655 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 656 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 657 pkt_seq, pkt_seq + pkt_len, pkt_len); 658 #endif 659 pkt_len += q->ipqe_len; 660 pkt_flags |= q->ipqe_flags; 661 m_cat(m, q->ipqe_m); 662 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 663 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 664 tp->t_segqlen--; 665 KASSERT(tp->t_segqlen >= 0); 666 KASSERT(tp->t_segqlen != 0 || 667 (TAILQ_EMPTY(&tp->segq) && 668 TAILQ_EMPTY(&tp->timeq))); 669 if (tiqe == NULL) { 670 tiqe = q; 671 } else { 672 tcpipqent_free(q); 673 } 674 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 675 break; 676 } 677 /* 678 * If the fragment is before the segment, remember it. 679 * When this loop is terminated, p will contain the 680 * pointer to fragment that is right before the received 681 * segment. 682 */ 683 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 684 p = q; 685 686 continue; 687 688 /* 689 * This is a common operation. It also will allow 690 * to save doing a malloc/free in most instances. 691 */ 692 free_ipqe: 693 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 694 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 695 tp->t_segqlen--; 696 KASSERT(tp->t_segqlen >= 0); 697 KASSERT(tp->t_segqlen != 0 || 698 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 699 if (tiqe == NULL) { 700 tiqe = q; 701 } else { 702 tcpipqent_free(q); 703 } 704 } 705 706 #ifdef TCP_REASS_COUNTERS 707 if (count > 7) 708 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 709 else if (count > 0) 710 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 711 #endif 712 713 insert_it: 714 715 /* 716 * Allocate a new queue entry since the received segment did not 717 * collapse onto any other out-of-order block; thus we are allocating 718 * a new block. If it had collapsed, tiqe would not be NULL and 719 * we would be reusing it. 720 * XXX If we can't, just drop the packet. XXX 721 */ 722 if (tiqe == NULL) { 723 tiqe = tcpipqent_alloc(); 724 if (tiqe == NULL) { 725 TCP_STATINC(TCP_STAT_RCVMEMDROP); 726 m_freem(m); 727 return (0); 728 } 729 } 730 731 /* 732 * Update the counters. 733 */ 734 tcps = TCP_STAT_GETREF(); 735 tcps[TCP_STAT_RCVOOPACK]++; 736 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 737 if (rcvpartdupbyte) { 738 tcps[TCP_STAT_RCVPARTDUPPACK]++; 739 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 740 } 741 TCP_STAT_PUTREF(); 742 743 /* 744 * Insert the new fragment queue entry into both queues. 745 */ 746 tiqe->ipqe_m = m; 747 tiqe->ipre_mlast = m; 748 tiqe->ipqe_seq = pkt_seq; 749 tiqe->ipqe_len = pkt_len; 750 tiqe->ipqe_flags = pkt_flags; 751 if (p == NULL) { 752 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 753 #ifdef TCPREASS_DEBUG 754 if (tiqe->ipqe_seq != tp->rcv_nxt) 755 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 756 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 757 #endif 758 } else { 759 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 760 #ifdef TCPREASS_DEBUG 761 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 762 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 763 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 764 #endif 765 } 766 tp->t_segqlen++; 767 768 skip_replacement: 769 770 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 771 772 present: 773 /* 774 * Present data to user, advancing rcv_nxt through 775 * completed sequence space. 776 */ 777 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 778 return (0); 779 q = TAILQ_FIRST(&tp->segq); 780 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 781 return (0); 782 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 783 return (0); 784 785 tp->rcv_nxt += q->ipqe_len; 786 pkt_flags = q->ipqe_flags & TH_FIN; 787 nd6_hint(tp); 788 789 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 790 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 791 tp->t_segqlen--; 792 KASSERT(tp->t_segqlen >= 0); 793 KASSERT(tp->t_segqlen != 0 || 794 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 795 if (so->so_state & SS_CANTRCVMORE) 796 m_freem(q->ipqe_m); 797 else 798 sbappendstream(&so->so_rcv, q->ipqe_m); 799 tcpipqent_free(q); 800 sorwakeup(so); 801 return (pkt_flags); 802 } 803 804 #ifdef INET6 805 int 806 tcp6_input(struct mbuf **mp, int *offp, int proto) 807 { 808 struct mbuf *m = *mp; 809 810 /* 811 * draft-itojun-ipv6-tcp-to-anycast 812 * better place to put this in? 813 */ 814 if (m->m_flags & M_ANYCAST6) { 815 struct ip6_hdr *ip6; 816 if (m->m_len < sizeof(struct ip6_hdr)) { 817 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 818 TCP_STATINC(TCP_STAT_RCVSHORT); 819 return IPPROTO_DONE; 820 } 821 } 822 ip6 = mtod(m, struct ip6_hdr *); 823 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 824 (char *)&ip6->ip6_dst - (char *)ip6); 825 return IPPROTO_DONE; 826 } 827 828 tcp_input(m, *offp, proto); 829 return IPPROTO_DONE; 830 } 831 #endif 832 833 #ifdef INET 834 static void 835 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 836 { 837 char src[4*sizeof "123"]; 838 char dst[4*sizeof "123"]; 839 840 if (ip) { 841 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 842 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 843 } 844 else { 845 strlcpy(src, "(unknown)", sizeof(src)); 846 strlcpy(dst, "(unknown)", sizeof(dst)); 847 } 848 log(LOG_INFO, 849 "Connection attempt to TCP %s:%d from %s:%d\n", 850 dst, ntohs(th->th_dport), 851 src, ntohs(th->th_sport)); 852 } 853 #endif 854 855 #ifdef INET6 856 static void 857 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 858 { 859 char src[INET6_ADDRSTRLEN]; 860 char dst[INET6_ADDRSTRLEN]; 861 862 if (ip6) { 863 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 864 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 865 } 866 else { 867 strlcpy(src, "(unknown v6)", sizeof(src)); 868 strlcpy(dst, "(unknown v6)", sizeof(dst)); 869 } 870 log(LOG_INFO, 871 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 872 dst, ntohs(th->th_dport), 873 src, ntohs(th->th_sport)); 874 } 875 #endif 876 877 /* 878 * Checksum extended TCP header and data. 879 */ 880 int 881 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 882 int toff, int off, int tlen) 883 { 884 885 /* 886 * XXX it's better to record and check if this mbuf is 887 * already checked. 888 */ 889 890 switch (af) { 891 #ifdef INET 892 case AF_INET: 893 switch (m->m_pkthdr.csum_flags & 894 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 895 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 896 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 897 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 898 goto badcsum; 899 900 case M_CSUM_TCPv4|M_CSUM_DATA: { 901 u_int32_t hw_csum = m->m_pkthdr.csum_data; 902 903 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 904 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 905 const struct ip *ip = 906 mtod(m, const struct ip *); 907 908 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 909 ip->ip_dst.s_addr, 910 htons(hw_csum + tlen + off + IPPROTO_TCP)); 911 } 912 if ((hw_csum ^ 0xffff) != 0) 913 goto badcsum; 914 break; 915 } 916 917 case M_CSUM_TCPv4: 918 /* Checksum was okay. */ 919 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 920 break; 921 922 default: 923 /* 924 * Must compute it ourselves. Maybe skip checksum 925 * on loopback interfaces. 926 */ 927 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 928 IFF_LOOPBACK) || 929 tcp_do_loopback_cksum)) { 930 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 931 if (in4_cksum(m, IPPROTO_TCP, toff, 932 tlen + off) != 0) 933 goto badcsum; 934 } 935 break; 936 } 937 break; 938 #endif /* INET4 */ 939 940 #ifdef INET6 941 case AF_INET6: 942 switch (m->m_pkthdr.csum_flags & 943 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 944 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 945 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 946 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 947 goto badcsum; 948 949 #if 0 /* notyet */ 950 case M_CSUM_TCPv6|M_CSUM_DATA: 951 #endif 952 953 case M_CSUM_TCPv6: 954 /* Checksum was okay. */ 955 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 956 break; 957 958 default: 959 /* 960 * Must compute it ourselves. Maybe skip checksum 961 * on loopback interfaces. 962 */ 963 if (__predict_true((m->m_flags & M_LOOP) == 0 || 964 tcp_do_loopback_cksum)) { 965 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 966 if (in6_cksum(m, IPPROTO_TCP, toff, 967 tlen + off) != 0) 968 goto badcsum; 969 } 970 } 971 break; 972 #endif /* INET6 */ 973 } 974 975 return 0; 976 977 badcsum: 978 TCP_STATINC(TCP_STAT_RCVBADSUM); 979 return -1; 980 } 981 982 /* 983 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 984 */ 985 void 986 tcp_input(struct mbuf *m, ...) 987 { 988 struct tcphdr *th; 989 struct ip *ip; 990 struct inpcb *inp; 991 #ifdef INET6 992 struct ip6_hdr *ip6; 993 struct in6pcb *in6p; 994 #endif 995 u_int8_t *optp = NULL; 996 int optlen = 0; 997 int len, tlen, toff, hdroptlen = 0; 998 struct tcpcb *tp = 0; 999 int tiflags; 1000 struct socket *so = NULL; 1001 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 1002 #ifdef TCP_DEBUG 1003 short ostate = 0; 1004 #endif 1005 u_long tiwin; 1006 struct tcp_opt_info opti; 1007 int off, iphlen; 1008 va_list ap; 1009 int af; /* af on the wire */ 1010 struct mbuf *tcp_saveti = NULL; 1011 uint32_t ts_rtt; 1012 uint8_t iptos; 1013 uint64_t *tcps; 1014 1015 MCLAIM(m, &tcp_rx_mowner); 1016 va_start(ap, m); 1017 toff = va_arg(ap, int); 1018 (void)va_arg(ap, int); /* ignore value, advance ap */ 1019 va_end(ap); 1020 1021 TCP_STATINC(TCP_STAT_RCVTOTAL); 1022 1023 bzero(&opti, sizeof(opti)); 1024 opti.ts_present = 0; 1025 opti.maxseg = 0; 1026 1027 /* 1028 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1029 * 1030 * TCP is, by definition, unicast, so we reject all 1031 * multicast outright. 1032 * 1033 * Note, there are additional src/dst address checks in 1034 * the AF-specific code below. 1035 */ 1036 if (m->m_flags & (M_BCAST|M_MCAST)) { 1037 /* XXX stat */ 1038 goto drop; 1039 } 1040 #ifdef INET6 1041 if (m->m_flags & M_ANYCAST6) { 1042 /* XXX stat */ 1043 goto drop; 1044 } 1045 #endif 1046 1047 /* 1048 * Get IP and TCP header. 1049 * Note: IP leaves IP header in first mbuf. 1050 */ 1051 ip = mtod(m, struct ip *); 1052 #ifdef INET6 1053 ip6 = NULL; 1054 #endif 1055 switch (ip->ip_v) { 1056 #ifdef INET 1057 case 4: 1058 af = AF_INET; 1059 iphlen = sizeof(struct ip); 1060 ip = mtod(m, struct ip *); 1061 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1062 sizeof(struct tcphdr)); 1063 if (th == NULL) { 1064 TCP_STATINC(TCP_STAT_RCVSHORT); 1065 return; 1066 } 1067 /* We do the checksum after PCB lookup... */ 1068 len = ntohs(ip->ip_len); 1069 tlen = len - toff; 1070 iptos = ip->ip_tos; 1071 break; 1072 #endif 1073 #ifdef INET6 1074 case 6: 1075 ip = NULL; 1076 iphlen = sizeof(struct ip6_hdr); 1077 af = AF_INET6; 1078 ip6 = mtod(m, struct ip6_hdr *); 1079 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1080 sizeof(struct tcphdr)); 1081 if (th == NULL) { 1082 TCP_STATINC(TCP_STAT_RCVSHORT); 1083 return; 1084 } 1085 1086 /* Be proactive about malicious use of IPv4 mapped address */ 1087 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1088 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1089 /* XXX stat */ 1090 goto drop; 1091 } 1092 1093 /* 1094 * Be proactive about unspecified IPv6 address in source. 1095 * As we use all-zero to indicate unbounded/unconnected pcb, 1096 * unspecified IPv6 address can be used to confuse us. 1097 * 1098 * Note that packets with unspecified IPv6 destination is 1099 * already dropped in ip6_input. 1100 */ 1101 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1102 /* XXX stat */ 1103 goto drop; 1104 } 1105 1106 /* 1107 * Make sure destination address is not multicast. 1108 * Source address checked in ip6_input(). 1109 */ 1110 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1111 /* XXX stat */ 1112 goto drop; 1113 } 1114 1115 /* We do the checksum after PCB lookup... */ 1116 len = m->m_pkthdr.len; 1117 tlen = len - toff; 1118 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1119 break; 1120 #endif 1121 default: 1122 m_freem(m); 1123 return; 1124 } 1125 1126 KASSERT(TCP_HDR_ALIGNED_P(th)); 1127 1128 /* 1129 * Check that TCP offset makes sense, 1130 * pull out TCP options and adjust length. XXX 1131 */ 1132 off = th->th_off << 2; 1133 if (off < sizeof (struct tcphdr) || off > tlen) { 1134 TCP_STATINC(TCP_STAT_RCVBADOFF); 1135 goto drop; 1136 } 1137 tlen -= off; 1138 1139 /* 1140 * tcp_input() has been modified to use tlen to mean the TCP data 1141 * length throughout the function. Other functions can use 1142 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1143 * rja 1144 */ 1145 1146 if (off > sizeof (struct tcphdr)) { 1147 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1148 if (th == NULL) { 1149 TCP_STATINC(TCP_STAT_RCVSHORT); 1150 return; 1151 } 1152 /* 1153 * NOTE: ip/ip6 will not be affected by m_pulldown() 1154 * (as they're before toff) and we don't need to update those. 1155 */ 1156 KASSERT(TCP_HDR_ALIGNED_P(th)); 1157 optlen = off - sizeof (struct tcphdr); 1158 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1159 /* 1160 * Do quick retrieval of timestamp options ("options 1161 * prediction?"). If timestamp is the only option and it's 1162 * formatted as recommended in RFC 1323 appendix A, we 1163 * quickly get the values now and not bother calling 1164 * tcp_dooptions(), etc. 1165 */ 1166 if ((optlen == TCPOLEN_TSTAMP_APPA || 1167 (optlen > TCPOLEN_TSTAMP_APPA && 1168 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1169 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1170 (th->th_flags & TH_SYN) == 0) { 1171 opti.ts_present = 1; 1172 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1173 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1174 optp = NULL; /* we've parsed the options */ 1175 } 1176 } 1177 tiflags = th->th_flags; 1178 1179 /* 1180 * Locate pcb for segment. 1181 */ 1182 findpcb: 1183 inp = NULL; 1184 #ifdef INET6 1185 in6p = NULL; 1186 #endif 1187 switch (af) { 1188 #ifdef INET 1189 case AF_INET: 1190 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1191 ip->ip_dst, th->th_dport); 1192 if (inp == 0) { 1193 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1194 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1195 } 1196 #ifdef INET6 1197 if (inp == 0) { 1198 struct in6_addr s, d; 1199 1200 /* mapped addr case */ 1201 bzero(&s, sizeof(s)); 1202 s.s6_addr16[5] = htons(0xffff); 1203 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1204 bzero(&d, sizeof(d)); 1205 d.s6_addr16[5] = htons(0xffff); 1206 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1207 in6p = in6_pcblookup_connect(&tcbtable, &s, 1208 th->th_sport, &d, th->th_dport, 0); 1209 if (in6p == 0) { 1210 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1211 in6p = in6_pcblookup_bind(&tcbtable, &d, 1212 th->th_dport, 0); 1213 } 1214 } 1215 #endif 1216 #ifndef INET6 1217 if (inp == 0) 1218 #else 1219 if (inp == 0 && in6p == 0) 1220 #endif 1221 { 1222 TCP_STATINC(TCP_STAT_NOPORT); 1223 if (tcp_log_refused && 1224 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1225 tcp4_log_refused(ip, th); 1226 } 1227 tcp_fields_to_host(th); 1228 goto dropwithreset_ratelim; 1229 } 1230 #if defined(IPSEC) || defined(FAST_IPSEC) 1231 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1232 ipsec4_in_reject(m, inp)) { 1233 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1234 goto drop; 1235 } 1236 #ifdef INET6 1237 else if (in6p && 1238 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1239 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1240 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1241 goto drop; 1242 } 1243 #endif 1244 #endif /*IPSEC*/ 1245 break; 1246 #endif /*INET*/ 1247 #ifdef INET6 1248 case AF_INET6: 1249 { 1250 int faith; 1251 1252 #if defined(NFAITH) && NFAITH > 0 1253 faith = faithprefix(&ip6->ip6_dst); 1254 #else 1255 faith = 0; 1256 #endif 1257 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1258 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1259 if (in6p == NULL) { 1260 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1261 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1262 th->th_dport, faith); 1263 } 1264 if (in6p == NULL) { 1265 TCP_STATINC(TCP_STAT_NOPORT); 1266 if (tcp_log_refused && 1267 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1268 tcp6_log_refused(ip6, th); 1269 } 1270 tcp_fields_to_host(th); 1271 goto dropwithreset_ratelim; 1272 } 1273 #if defined(IPSEC) || defined(FAST_IPSEC) 1274 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1275 ipsec6_in_reject(m, in6p)) { 1276 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1277 goto drop; 1278 } 1279 #endif /*IPSEC*/ 1280 break; 1281 } 1282 #endif 1283 } 1284 1285 /* 1286 * If the state is CLOSED (i.e., TCB does not exist) then 1287 * all data in the incoming segment is discarded. 1288 * If the TCB exists but is in CLOSED state, it is embryonic, 1289 * but should either do a listen or a connect soon. 1290 */ 1291 tp = NULL; 1292 so = NULL; 1293 if (inp) { 1294 tp = intotcpcb(inp); 1295 so = inp->inp_socket; 1296 } 1297 #ifdef INET6 1298 else if (in6p) { 1299 tp = in6totcpcb(in6p); 1300 so = in6p->in6p_socket; 1301 } 1302 #endif 1303 if (tp == 0) { 1304 tcp_fields_to_host(th); 1305 goto dropwithreset_ratelim; 1306 } 1307 if (tp->t_state == TCPS_CLOSED) 1308 goto drop; 1309 1310 /* 1311 * Checksum extended TCP header and data. 1312 */ 1313 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1314 goto badcsum; 1315 1316 tcp_fields_to_host(th); 1317 1318 /* Unscale the window into a 32-bit value. */ 1319 if ((tiflags & TH_SYN) == 0) 1320 tiwin = th->th_win << tp->snd_scale; 1321 else 1322 tiwin = th->th_win; 1323 1324 #ifdef INET6 1325 /* save packet options if user wanted */ 1326 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1327 if (in6p->in6p_options) { 1328 m_freem(in6p->in6p_options); 1329 in6p->in6p_options = 0; 1330 } 1331 KASSERT(ip6 != NULL); 1332 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1333 } 1334 #endif 1335 1336 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1337 union syn_cache_sa src; 1338 union syn_cache_sa dst; 1339 1340 bzero(&src, sizeof(src)); 1341 bzero(&dst, sizeof(dst)); 1342 switch (af) { 1343 #ifdef INET 1344 case AF_INET: 1345 src.sin.sin_len = sizeof(struct sockaddr_in); 1346 src.sin.sin_family = AF_INET; 1347 src.sin.sin_addr = ip->ip_src; 1348 src.sin.sin_port = th->th_sport; 1349 1350 dst.sin.sin_len = sizeof(struct sockaddr_in); 1351 dst.sin.sin_family = AF_INET; 1352 dst.sin.sin_addr = ip->ip_dst; 1353 dst.sin.sin_port = th->th_dport; 1354 break; 1355 #endif 1356 #ifdef INET6 1357 case AF_INET6: 1358 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1359 src.sin6.sin6_family = AF_INET6; 1360 src.sin6.sin6_addr = ip6->ip6_src; 1361 src.sin6.sin6_port = th->th_sport; 1362 1363 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1364 dst.sin6.sin6_family = AF_INET6; 1365 dst.sin6.sin6_addr = ip6->ip6_dst; 1366 dst.sin6.sin6_port = th->th_dport; 1367 break; 1368 #endif /* INET6 */ 1369 default: 1370 goto badsyn; /*sanity*/ 1371 } 1372 1373 if (so->so_options & SO_DEBUG) { 1374 #ifdef TCP_DEBUG 1375 ostate = tp->t_state; 1376 #endif 1377 1378 tcp_saveti = NULL; 1379 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1380 goto nosave; 1381 1382 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1383 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1384 if (!tcp_saveti) 1385 goto nosave; 1386 } else { 1387 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1388 if (!tcp_saveti) 1389 goto nosave; 1390 MCLAIM(m, &tcp_mowner); 1391 tcp_saveti->m_len = iphlen; 1392 m_copydata(m, 0, iphlen, 1393 mtod(tcp_saveti, void *)); 1394 } 1395 1396 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1397 m_freem(tcp_saveti); 1398 tcp_saveti = NULL; 1399 } else { 1400 tcp_saveti->m_len += sizeof(struct tcphdr); 1401 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1402 sizeof(struct tcphdr)); 1403 } 1404 nosave:; 1405 } 1406 if (so->so_options & SO_ACCEPTCONN) { 1407 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1408 if (tiflags & TH_RST) { 1409 syn_cache_reset(&src.sa, &dst.sa, th); 1410 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1411 (TH_ACK|TH_SYN)) { 1412 /* 1413 * Received a SYN,ACK. This should 1414 * never happen while we are in 1415 * LISTEN. Send an RST. 1416 */ 1417 goto badsyn; 1418 } else if (tiflags & TH_ACK) { 1419 so = syn_cache_get(&src.sa, &dst.sa, 1420 th, toff, tlen, so, m); 1421 if (so == NULL) { 1422 /* 1423 * We don't have a SYN for 1424 * this ACK; send an RST. 1425 */ 1426 goto badsyn; 1427 } else if (so == 1428 (struct socket *)(-1)) { 1429 /* 1430 * We were unable to create 1431 * the connection. If the 1432 * 3-way handshake was 1433 * completed, and RST has 1434 * been sent to the peer. 1435 * Since the mbuf might be 1436 * in use for the reply, 1437 * do not free it. 1438 */ 1439 m = NULL; 1440 } else { 1441 /* 1442 * We have created a 1443 * full-blown connection. 1444 */ 1445 tp = NULL; 1446 inp = NULL; 1447 #ifdef INET6 1448 in6p = NULL; 1449 #endif 1450 switch (so->so_proto->pr_domain->dom_family) { 1451 #ifdef INET 1452 case AF_INET: 1453 inp = sotoinpcb(so); 1454 tp = intotcpcb(inp); 1455 break; 1456 #endif 1457 #ifdef INET6 1458 case AF_INET6: 1459 in6p = sotoin6pcb(so); 1460 tp = in6totcpcb(in6p); 1461 break; 1462 #endif 1463 } 1464 if (tp == NULL) 1465 goto badsyn; /*XXX*/ 1466 tiwin <<= tp->snd_scale; 1467 goto after_listen; 1468 } 1469 } else { 1470 /* 1471 * None of RST, SYN or ACK was set. 1472 * This is an invalid packet for a 1473 * TCB in LISTEN state. Send a RST. 1474 */ 1475 goto badsyn; 1476 } 1477 } else { 1478 /* 1479 * Received a SYN. 1480 * 1481 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1482 */ 1483 if (m->m_flags & (M_BCAST|M_MCAST)) 1484 goto drop; 1485 1486 switch (af) { 1487 #ifdef INET6 1488 case AF_INET6: 1489 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1490 goto drop; 1491 break; 1492 #endif /* INET6 */ 1493 case AF_INET: 1494 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1495 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1496 goto drop; 1497 break; 1498 } 1499 1500 #ifdef INET6 1501 /* 1502 * If deprecated address is forbidden, we do 1503 * not accept SYN to deprecated interface 1504 * address to prevent any new inbound 1505 * connection from getting established. 1506 * When we do not accept SYN, we send a TCP 1507 * RST, with deprecated source address (instead 1508 * of dropping it). We compromise it as it is 1509 * much better for peer to send a RST, and 1510 * RST will be the final packet for the 1511 * exchange. 1512 * 1513 * If we do not forbid deprecated addresses, we 1514 * accept the SYN packet. RFC2462 does not 1515 * suggest dropping SYN in this case. 1516 * If we decipher RFC2462 5.5.4, it says like 1517 * this: 1518 * 1. use of deprecated addr with existing 1519 * communication is okay - "SHOULD continue 1520 * to be used" 1521 * 2. use of it with new communication: 1522 * (2a) "SHOULD NOT be used if alternate 1523 * address with sufficient scope is 1524 * available" 1525 * (2b) nothing mentioned otherwise. 1526 * Here we fall into (2b) case as we have no 1527 * choice in our source address selection - we 1528 * must obey the peer. 1529 * 1530 * The wording in RFC2462 is confusing, and 1531 * there are multiple description text for 1532 * deprecated address handling - worse, they 1533 * are not exactly the same. I believe 5.5.4 1534 * is the best one, so we follow 5.5.4. 1535 */ 1536 if (af == AF_INET6 && !ip6_use_deprecated) { 1537 struct in6_ifaddr *ia6; 1538 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1539 &ip6->ip6_dst)) && 1540 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1541 tp = NULL; 1542 goto dropwithreset; 1543 } 1544 } 1545 #endif 1546 1547 #if defined(IPSEC) || defined(FAST_IPSEC) 1548 switch (af) { 1549 #ifdef INET 1550 case AF_INET: 1551 if (ipsec4_in_reject_so(m, so)) { 1552 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1553 tp = NULL; 1554 goto dropwithreset; 1555 } 1556 break; 1557 #endif 1558 #ifdef INET6 1559 case AF_INET6: 1560 if (ipsec6_in_reject_so(m, so)) { 1561 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1562 tp = NULL; 1563 goto dropwithreset; 1564 } 1565 break; 1566 #endif /*INET6*/ 1567 } 1568 #endif /*IPSEC*/ 1569 1570 /* 1571 * LISTEN socket received a SYN 1572 * from itself? This can't possibly 1573 * be valid; drop the packet. 1574 */ 1575 if (th->th_sport == th->th_dport) { 1576 int i; 1577 1578 switch (af) { 1579 #ifdef INET 1580 case AF_INET: 1581 i = in_hosteq(ip->ip_src, ip->ip_dst); 1582 break; 1583 #endif 1584 #ifdef INET6 1585 case AF_INET6: 1586 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1587 break; 1588 #endif 1589 default: 1590 i = 1; 1591 } 1592 if (i) { 1593 TCP_STATINC(TCP_STAT_BADSYN); 1594 goto drop; 1595 } 1596 } 1597 1598 /* 1599 * SYN looks ok; create compressed TCP 1600 * state for it. 1601 */ 1602 if (so->so_qlen <= so->so_qlimit && 1603 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1604 so, m, optp, optlen, &opti)) 1605 m = NULL; 1606 } 1607 goto drop; 1608 } 1609 } 1610 1611 after_listen: 1612 #ifdef DIAGNOSTIC 1613 /* 1614 * Should not happen now that all embryonic connections 1615 * are handled with compressed state. 1616 */ 1617 if (tp->t_state == TCPS_LISTEN) 1618 panic("tcp_input: TCPS_LISTEN"); 1619 #endif 1620 1621 /* 1622 * Segment received on connection. 1623 * Reset idle time and keep-alive timer. 1624 */ 1625 tp->t_rcvtime = tcp_now; 1626 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1627 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1628 1629 /* 1630 * Process options. 1631 */ 1632 #ifdef TCP_SIGNATURE 1633 if (optp || (tp->t_flags & TF_SIGNATURE)) 1634 #else 1635 if (optp) 1636 #endif 1637 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1638 goto drop; 1639 1640 if (TCP_SACK_ENABLED(tp)) { 1641 tcp_del_sackholes(tp, th); 1642 } 1643 1644 if (TCP_ECN_ALLOWED(tp)) { 1645 switch (iptos & IPTOS_ECN_MASK) { 1646 case IPTOS_ECN_CE: 1647 tp->t_flags |= TF_ECN_SND_ECE; 1648 TCP_STATINC(TCP_STAT_ECN_CE); 1649 break; 1650 case IPTOS_ECN_ECT0: 1651 TCP_STATINC(TCP_STAT_ECN_ECT); 1652 break; 1653 case IPTOS_ECN_ECT1: 1654 /* XXX: ignore for now -- rpaulo */ 1655 break; 1656 } 1657 1658 if (tiflags & TH_CWR) 1659 tp->t_flags &= ~TF_ECN_SND_ECE; 1660 1661 /* 1662 * Congestion experienced. 1663 * Ignore if we are already trying to recover. 1664 */ 1665 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1666 tp->t_congctl->cong_exp(tp); 1667 } 1668 1669 if (opti.ts_present && opti.ts_ecr) { 1670 /* 1671 * Calculate the RTT from the returned time stamp and the 1672 * connection's time base. If the time stamp is later than 1673 * the current time, or is extremely old, fall back to non-1323 1674 * RTT calculation. Since ts_ecr is unsigned, we can test both 1675 * at the same time. 1676 */ 1677 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1678 if (ts_rtt > TCP_PAWS_IDLE) 1679 ts_rtt = 0; 1680 } else { 1681 ts_rtt = 0; 1682 } 1683 1684 /* 1685 * Header prediction: check for the two common cases 1686 * of a uni-directional data xfer. If the packet has 1687 * no control flags, is in-sequence, the window didn't 1688 * change and we're not retransmitting, it's a 1689 * candidate. If the length is zero and the ack moved 1690 * forward, we're the sender side of the xfer. Just 1691 * free the data acked & wake any higher level process 1692 * that was blocked waiting for space. If the length 1693 * is non-zero and the ack didn't move, we're the 1694 * receiver side. If we're getting packets in-order 1695 * (the reassembly queue is empty), add the data to 1696 * the socket buffer and note that we need a delayed ack. 1697 */ 1698 if (tp->t_state == TCPS_ESTABLISHED && 1699 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1700 == TH_ACK && 1701 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1702 th->th_seq == tp->rcv_nxt && 1703 tiwin && tiwin == tp->snd_wnd && 1704 tp->snd_nxt == tp->snd_max) { 1705 1706 /* 1707 * If last ACK falls within this segment's sequence numbers, 1708 * record the timestamp. 1709 * NOTE that the test is modified according to the latest 1710 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1711 * 1712 * note that we already know 1713 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1714 */ 1715 if (opti.ts_present && 1716 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1717 tp->ts_recent_age = tcp_now; 1718 tp->ts_recent = opti.ts_val; 1719 } 1720 1721 if (tlen == 0) { 1722 /* Ack prediction. */ 1723 if (SEQ_GT(th->th_ack, tp->snd_una) && 1724 SEQ_LEQ(th->th_ack, tp->snd_max) && 1725 tp->snd_cwnd >= tp->snd_wnd && 1726 tp->t_partialacks < 0) { 1727 /* 1728 * this is a pure ack for outstanding data. 1729 */ 1730 if (ts_rtt) 1731 tcp_xmit_timer(tp, ts_rtt); 1732 else if (tp->t_rtttime && 1733 SEQ_GT(th->th_ack, tp->t_rtseq)) 1734 tcp_xmit_timer(tp, 1735 tcp_now - tp->t_rtttime); 1736 acked = th->th_ack - tp->snd_una; 1737 tcps = TCP_STAT_GETREF(); 1738 tcps[TCP_STAT_PREDACK]++; 1739 tcps[TCP_STAT_RCVACKPACK]++; 1740 tcps[TCP_STAT_RCVACKBYTE] += acked; 1741 TCP_STAT_PUTREF(); 1742 nd6_hint(tp); 1743 1744 if (acked > (tp->t_lastoff - tp->t_inoff)) 1745 tp->t_lastm = NULL; 1746 sbdrop(&so->so_snd, acked); 1747 tp->t_lastoff -= acked; 1748 1749 icmp_check(tp, th, acked); 1750 1751 tp->snd_una = th->th_ack; 1752 tp->snd_fack = tp->snd_una; 1753 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1754 tp->snd_high = tp->snd_una; 1755 m_freem(m); 1756 1757 /* 1758 * If all outstanding data are acked, stop 1759 * retransmit timer, otherwise restart timer 1760 * using current (possibly backed-off) value. 1761 * If process is waiting for space, 1762 * wakeup/selnotify/signal. If data 1763 * are ready to send, let tcp_output 1764 * decide between more output or persist. 1765 */ 1766 if (tp->snd_una == tp->snd_max) 1767 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1768 else if (TCP_TIMER_ISARMED(tp, 1769 TCPT_PERSIST) == 0) 1770 TCP_TIMER_ARM(tp, TCPT_REXMT, 1771 tp->t_rxtcur); 1772 1773 sowwakeup(so); 1774 if (so->so_snd.sb_cc) 1775 (void) tcp_output(tp); 1776 if (tcp_saveti) 1777 m_freem(tcp_saveti); 1778 return; 1779 } 1780 } else if (th->th_ack == tp->snd_una && 1781 TAILQ_FIRST(&tp->segq) == NULL && 1782 tlen <= sbspace(&so->so_rcv)) { 1783 int newsize = 0; /* automatic sockbuf scaling */ 1784 1785 /* 1786 * this is a pure, in-sequence data packet 1787 * with nothing on the reassembly queue and 1788 * we have enough buffer space to take it. 1789 */ 1790 tp->rcv_nxt += tlen; 1791 tcps = TCP_STAT_GETREF(); 1792 tcps[TCP_STAT_PREDDAT]++; 1793 tcps[TCP_STAT_RCVPACK]++; 1794 tcps[TCP_STAT_RCVBYTE] += tlen; 1795 TCP_STAT_PUTREF(); 1796 nd6_hint(tp); 1797 1798 /* 1799 * Automatic sizing enables the performance of large buffers 1800 * and most of the efficiency of small ones by only allocating 1801 * space when it is needed. 1802 * 1803 * On the receive side the socket buffer memory is only rarely 1804 * used to any significant extent. This allows us to be much 1805 * more aggressive in scaling the receive socket buffer. For 1806 * the case that the buffer space is actually used to a large 1807 * extent and we run out of kernel memory we can simply drop 1808 * the new segments; TCP on the sender will just retransmit it 1809 * later. Setting the buffer size too big may only consume too 1810 * much kernel memory if the application doesn't read() from 1811 * the socket or packet loss or reordering makes use of the 1812 * reassembly queue. 1813 * 1814 * The criteria to step up the receive buffer one notch are: 1815 * 1. the number of bytes received during the time it takes 1816 * one timestamp to be reflected back to us (the RTT); 1817 * 2. received bytes per RTT is within seven eighth of the 1818 * current socket buffer size; 1819 * 3. receive buffer size has not hit maximal automatic size; 1820 * 1821 * This algorithm does one step per RTT at most and only if 1822 * we receive a bulk stream w/o packet losses or reorderings. 1823 * Shrinking the buffer during idle times is not necessary as 1824 * it doesn't consume any memory when idle. 1825 * 1826 * TODO: Only step up if the application is actually serving 1827 * the buffer to better manage the socket buffer resources. 1828 */ 1829 if (tcp_do_autorcvbuf && 1830 opti.ts_ecr && 1831 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1832 if (opti.ts_ecr > tp->rfbuf_ts && 1833 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 1834 if (tp->rfbuf_cnt > 1835 (so->so_rcv.sb_hiwat / 8 * 7) && 1836 so->so_rcv.sb_hiwat < 1837 tcp_autorcvbuf_max) { 1838 newsize = 1839 min(so->so_rcv.sb_hiwat + 1840 tcp_autorcvbuf_inc, 1841 tcp_autorcvbuf_max); 1842 } 1843 /* Start over with next RTT. */ 1844 tp->rfbuf_ts = 0; 1845 tp->rfbuf_cnt = 0; 1846 } else 1847 tp->rfbuf_cnt += tlen; /* add up */ 1848 } 1849 1850 /* 1851 * Drop TCP, IP headers and TCP options then add data 1852 * to socket buffer. 1853 */ 1854 if (so->so_state & SS_CANTRCVMORE) 1855 m_freem(m); 1856 else { 1857 /* 1858 * Set new socket buffer size. 1859 * Give up when limit is reached. 1860 */ 1861 if (newsize) 1862 if (!sbreserve(&so->so_rcv, 1863 newsize, so)) 1864 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1865 m_adj(m, toff + off); 1866 sbappendstream(&so->so_rcv, m); 1867 } 1868 sorwakeup(so); 1869 tcp_setup_ack(tp, th); 1870 if (tp->t_flags & TF_ACKNOW) 1871 (void) tcp_output(tp); 1872 if (tcp_saveti) 1873 m_freem(tcp_saveti); 1874 return; 1875 } 1876 } 1877 1878 /* 1879 * Compute mbuf offset to TCP data segment. 1880 */ 1881 hdroptlen = toff + off; 1882 1883 /* 1884 * Calculate amount of space in receive window, 1885 * and then do TCP input processing. 1886 * Receive window is amount of space in rcv queue, 1887 * but not less than advertised window. 1888 */ 1889 { int win; 1890 1891 win = sbspace(&so->so_rcv); 1892 if (win < 0) 1893 win = 0; 1894 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1895 } 1896 1897 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1898 tp->rfbuf_ts = 0; 1899 tp->rfbuf_cnt = 0; 1900 1901 switch (tp->t_state) { 1902 /* 1903 * If the state is SYN_SENT: 1904 * if seg contains an ACK, but not for our SYN, drop the input. 1905 * if seg contains a RST, then drop the connection. 1906 * if seg does not contain SYN, then drop it. 1907 * Otherwise this is an acceptable SYN segment 1908 * initialize tp->rcv_nxt and tp->irs 1909 * if seg contains ack then advance tp->snd_una 1910 * if seg contains a ECE and ECN support is enabled, the stream 1911 * is ECN capable. 1912 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1913 * arrange for segment to be acked (eventually) 1914 * continue processing rest of data/controls, beginning with URG 1915 */ 1916 case TCPS_SYN_SENT: 1917 if ((tiflags & TH_ACK) && 1918 (SEQ_LEQ(th->th_ack, tp->iss) || 1919 SEQ_GT(th->th_ack, tp->snd_max))) 1920 goto dropwithreset; 1921 if (tiflags & TH_RST) { 1922 if (tiflags & TH_ACK) 1923 tp = tcp_drop(tp, ECONNREFUSED); 1924 goto drop; 1925 } 1926 if ((tiflags & TH_SYN) == 0) 1927 goto drop; 1928 if (tiflags & TH_ACK) { 1929 tp->snd_una = th->th_ack; 1930 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1931 tp->snd_nxt = tp->snd_una; 1932 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1933 tp->snd_high = tp->snd_una; 1934 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1935 1936 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1937 tp->t_flags |= TF_ECN_PERMIT; 1938 TCP_STATINC(TCP_STAT_ECN_SHS); 1939 } 1940 1941 } 1942 tp->irs = th->th_seq; 1943 tcp_rcvseqinit(tp); 1944 tp->t_flags |= TF_ACKNOW; 1945 tcp_mss_from_peer(tp, opti.maxseg); 1946 1947 /* 1948 * Initialize the initial congestion window. If we 1949 * had to retransmit the SYN, we must initialize cwnd 1950 * to 1 segment (i.e. the Loss Window). 1951 */ 1952 if (tp->t_flags & TF_SYN_REXMT) 1953 tp->snd_cwnd = tp->t_peermss; 1954 else { 1955 int ss = tcp_init_win; 1956 #ifdef INET 1957 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1958 ss = tcp_init_win_local; 1959 #endif 1960 #ifdef INET6 1961 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1962 ss = tcp_init_win_local; 1963 #endif 1964 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1965 } 1966 1967 tcp_rmx_rtt(tp); 1968 if (tiflags & TH_ACK) { 1969 TCP_STATINC(TCP_STAT_CONNECTS); 1970 soisconnected(so); 1971 tcp_established(tp); 1972 /* Do window scaling on this connection? */ 1973 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1974 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1975 tp->snd_scale = tp->requested_s_scale; 1976 tp->rcv_scale = tp->request_r_scale; 1977 } 1978 TCP_REASS_LOCK(tp); 1979 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1980 TCP_REASS_UNLOCK(tp); 1981 /* 1982 * if we didn't have to retransmit the SYN, 1983 * use its rtt as our initial srtt & rtt var. 1984 */ 1985 if (tp->t_rtttime) 1986 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1987 } else 1988 tp->t_state = TCPS_SYN_RECEIVED; 1989 1990 /* 1991 * Advance th->th_seq to correspond to first data byte. 1992 * If data, trim to stay within window, 1993 * dropping FIN if necessary. 1994 */ 1995 th->th_seq++; 1996 if (tlen > tp->rcv_wnd) { 1997 todrop = tlen - tp->rcv_wnd; 1998 m_adj(m, -todrop); 1999 tlen = tp->rcv_wnd; 2000 tiflags &= ~TH_FIN; 2001 tcps = TCP_STAT_GETREF(); 2002 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2003 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2004 TCP_STAT_PUTREF(); 2005 } 2006 tp->snd_wl1 = th->th_seq - 1; 2007 tp->rcv_up = th->th_seq; 2008 goto step6; 2009 2010 /* 2011 * If the state is SYN_RECEIVED: 2012 * If seg contains an ACK, but not for our SYN, drop the input 2013 * and generate an RST. See page 36, rfc793 2014 */ 2015 case TCPS_SYN_RECEIVED: 2016 if ((tiflags & TH_ACK) && 2017 (SEQ_LEQ(th->th_ack, tp->iss) || 2018 SEQ_GT(th->th_ack, tp->snd_max))) 2019 goto dropwithreset; 2020 break; 2021 } 2022 2023 /* 2024 * States other than LISTEN or SYN_SENT. 2025 * First check timestamp, if present. 2026 * Then check that at least some bytes of segment are within 2027 * receive window. If segment begins before rcv_nxt, 2028 * drop leading data (and SYN); if nothing left, just ack. 2029 * 2030 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2031 * and it's less than ts_recent, drop it. 2032 */ 2033 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2034 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2035 2036 /* Check to see if ts_recent is over 24 days old. */ 2037 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2038 /* 2039 * Invalidate ts_recent. If this segment updates 2040 * ts_recent, the age will be reset later and ts_recent 2041 * will get a valid value. If it does not, setting 2042 * ts_recent to zero will at least satisfy the 2043 * requirement that zero be placed in the timestamp 2044 * echo reply when ts_recent isn't valid. The 2045 * age isn't reset until we get a valid ts_recent 2046 * because we don't want out-of-order segments to be 2047 * dropped when ts_recent is old. 2048 */ 2049 tp->ts_recent = 0; 2050 } else { 2051 tcps = TCP_STAT_GETREF(); 2052 tcps[TCP_STAT_RCVDUPPACK]++; 2053 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2054 tcps[TCP_STAT_PAWSDROP]++; 2055 TCP_STAT_PUTREF(); 2056 tcp_new_dsack(tp, th->th_seq, tlen); 2057 goto dropafterack; 2058 } 2059 } 2060 2061 todrop = tp->rcv_nxt - th->th_seq; 2062 dupseg = false; 2063 if (todrop > 0) { 2064 if (tiflags & TH_SYN) { 2065 tiflags &= ~TH_SYN; 2066 th->th_seq++; 2067 if (th->th_urp > 1) 2068 th->th_urp--; 2069 else { 2070 tiflags &= ~TH_URG; 2071 th->th_urp = 0; 2072 } 2073 todrop--; 2074 } 2075 if (todrop > tlen || 2076 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2077 /* 2078 * Any valid FIN or RST must be to the left of the 2079 * window. At this point the FIN or RST must be a 2080 * duplicate or out of sequence; drop it. 2081 */ 2082 if (tiflags & TH_RST) 2083 goto drop; 2084 tiflags &= ~(TH_FIN|TH_RST); 2085 /* 2086 * Send an ACK to resynchronize and drop any data. 2087 * But keep on processing for RST or ACK. 2088 */ 2089 tp->t_flags |= TF_ACKNOW; 2090 todrop = tlen; 2091 dupseg = true; 2092 tcps = TCP_STAT_GETREF(); 2093 tcps[TCP_STAT_RCVDUPPACK]++; 2094 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2095 TCP_STAT_PUTREF(); 2096 } else if ((tiflags & TH_RST) && 2097 th->th_seq != tp->last_ack_sent) { 2098 /* 2099 * Test for reset before adjusting the sequence 2100 * number for overlapping data. 2101 */ 2102 goto dropafterack_ratelim; 2103 } else { 2104 tcps = TCP_STAT_GETREF(); 2105 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2106 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2107 TCP_STAT_PUTREF(); 2108 } 2109 tcp_new_dsack(tp, th->th_seq, todrop); 2110 hdroptlen += todrop; /*drop from head afterwards*/ 2111 th->th_seq += todrop; 2112 tlen -= todrop; 2113 if (th->th_urp > todrop) 2114 th->th_urp -= todrop; 2115 else { 2116 tiflags &= ~TH_URG; 2117 th->th_urp = 0; 2118 } 2119 } 2120 2121 /* 2122 * If new data are received on a connection after the 2123 * user processes are gone, then RST the other end. 2124 */ 2125 if ((so->so_state & SS_NOFDREF) && 2126 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2127 tp = tcp_close(tp); 2128 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2129 goto dropwithreset; 2130 } 2131 2132 /* 2133 * If segment ends after window, drop trailing data 2134 * (and PUSH and FIN); if nothing left, just ACK. 2135 */ 2136 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2137 if (todrop > 0) { 2138 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2139 if (todrop >= tlen) { 2140 /* 2141 * The segment actually starts after the window. 2142 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2143 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2144 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2145 */ 2146 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2147 /* 2148 * If a new connection request is received 2149 * while in TIME_WAIT, drop the old connection 2150 * and start over if the sequence numbers 2151 * are above the previous ones. 2152 * 2153 * NOTE: We will checksum the packet again, and 2154 * so we need to put the header fields back into 2155 * network order! 2156 * XXX This kind of sucks, but we don't expect 2157 * XXX this to happen very often, so maybe it 2158 * XXX doesn't matter so much. 2159 */ 2160 if (tiflags & TH_SYN && 2161 tp->t_state == TCPS_TIME_WAIT && 2162 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2163 tp = tcp_close(tp); 2164 tcp_fields_to_net(th); 2165 goto findpcb; 2166 } 2167 /* 2168 * If window is closed can only take segments at 2169 * window edge, and have to drop data and PUSH from 2170 * incoming segments. Continue processing, but 2171 * remember to ack. Otherwise, drop segment 2172 * and (if not RST) ack. 2173 */ 2174 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2175 tp->t_flags |= TF_ACKNOW; 2176 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2177 } else 2178 goto dropafterack; 2179 } else 2180 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2181 m_adj(m, -todrop); 2182 tlen -= todrop; 2183 tiflags &= ~(TH_PUSH|TH_FIN); 2184 } 2185 2186 /* 2187 * If last ACK falls within this segment's sequence numbers, 2188 * record the timestamp. 2189 * NOTE: 2190 * 1) That the test incorporates suggestions from the latest 2191 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2192 * 2) That updating only on newer timestamps interferes with 2193 * our earlier PAWS tests, so this check should be solely 2194 * predicated on the sequence space of this segment. 2195 * 3) That we modify the segment boundary check to be 2196 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2197 * instead of RFC1323's 2198 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2199 * This modified check allows us to overcome RFC1323's 2200 * limitations as described in Stevens TCP/IP Illustrated 2201 * Vol. 2 p.869. In such cases, we can still calculate the 2202 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2203 */ 2204 if (opti.ts_present && 2205 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2206 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2207 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2208 tp->ts_recent_age = tcp_now; 2209 tp->ts_recent = opti.ts_val; 2210 } 2211 2212 /* 2213 * If the RST bit is set examine the state: 2214 * SYN_RECEIVED STATE: 2215 * If passive open, return to LISTEN state. 2216 * If active open, inform user that connection was refused. 2217 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2218 * Inform user that connection was reset, and close tcb. 2219 * CLOSING, LAST_ACK, TIME_WAIT STATES 2220 * Close the tcb. 2221 */ 2222 if (tiflags & TH_RST) { 2223 if (th->th_seq != tp->last_ack_sent) 2224 goto dropafterack_ratelim; 2225 2226 switch (tp->t_state) { 2227 case TCPS_SYN_RECEIVED: 2228 so->so_error = ECONNREFUSED; 2229 goto close; 2230 2231 case TCPS_ESTABLISHED: 2232 case TCPS_FIN_WAIT_1: 2233 case TCPS_FIN_WAIT_2: 2234 case TCPS_CLOSE_WAIT: 2235 so->so_error = ECONNRESET; 2236 close: 2237 tp->t_state = TCPS_CLOSED; 2238 TCP_STATINC(TCP_STAT_DROPS); 2239 tp = tcp_close(tp); 2240 goto drop; 2241 2242 case TCPS_CLOSING: 2243 case TCPS_LAST_ACK: 2244 case TCPS_TIME_WAIT: 2245 tp = tcp_close(tp); 2246 goto drop; 2247 } 2248 } 2249 2250 /* 2251 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2252 * we must be in a synchronized state. RFC791 states (under RST 2253 * generation) that any unacceptable segment (an out-of-order SYN 2254 * qualifies) received in a synchronized state must elicit only an 2255 * empty acknowledgment segment ... and the connection remains in 2256 * the same state. 2257 */ 2258 if (tiflags & TH_SYN) { 2259 if (tp->rcv_nxt == th->th_seq) { 2260 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2261 TH_ACK); 2262 if (tcp_saveti) 2263 m_freem(tcp_saveti); 2264 return; 2265 } 2266 2267 goto dropafterack_ratelim; 2268 } 2269 2270 /* 2271 * If the ACK bit is off we drop the segment and return. 2272 */ 2273 if ((tiflags & TH_ACK) == 0) { 2274 if (tp->t_flags & TF_ACKNOW) 2275 goto dropafterack; 2276 else 2277 goto drop; 2278 } 2279 2280 /* 2281 * Ack processing. 2282 */ 2283 switch (tp->t_state) { 2284 2285 /* 2286 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2287 * ESTABLISHED state and continue processing, otherwise 2288 * send an RST. 2289 */ 2290 case TCPS_SYN_RECEIVED: 2291 if (SEQ_GT(tp->snd_una, th->th_ack) || 2292 SEQ_GT(th->th_ack, tp->snd_max)) 2293 goto dropwithreset; 2294 TCP_STATINC(TCP_STAT_CONNECTS); 2295 soisconnected(so); 2296 tcp_established(tp); 2297 /* Do window scaling? */ 2298 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2299 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2300 tp->snd_scale = tp->requested_s_scale; 2301 tp->rcv_scale = tp->request_r_scale; 2302 } 2303 TCP_REASS_LOCK(tp); 2304 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2305 TCP_REASS_UNLOCK(tp); 2306 tp->snd_wl1 = th->th_seq - 1; 2307 /* fall into ... */ 2308 2309 /* 2310 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2311 * ACKs. If the ack is in the range 2312 * tp->snd_una < th->th_ack <= tp->snd_max 2313 * then advance tp->snd_una to th->th_ack and drop 2314 * data from the retransmission queue. If this ACK reflects 2315 * more up to date window information we update our window information. 2316 */ 2317 case TCPS_ESTABLISHED: 2318 case TCPS_FIN_WAIT_1: 2319 case TCPS_FIN_WAIT_2: 2320 case TCPS_CLOSE_WAIT: 2321 case TCPS_CLOSING: 2322 case TCPS_LAST_ACK: 2323 case TCPS_TIME_WAIT: 2324 2325 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2326 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2327 TCP_STATINC(TCP_STAT_RCVDUPPACK); 2328 /* 2329 * If we have outstanding data (other than 2330 * a window probe), this is a completely 2331 * duplicate ack (ie, window info didn't 2332 * change), the ack is the biggest we've 2333 * seen and we've seen exactly our rexmt 2334 * threshhold of them, assume a packet 2335 * has been dropped and retransmit it. 2336 * Kludge snd_nxt & the congestion 2337 * window so we send only this one 2338 * packet. 2339 */ 2340 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2341 th->th_ack != tp->snd_una) 2342 tp->t_dupacks = 0; 2343 else if (tp->t_partialacks < 0 && 2344 (++tp->t_dupacks == tcprexmtthresh || 2345 TCP_FACK_FASTRECOV(tp))) { 2346 /* 2347 * Do the fast retransmit, and adjust 2348 * congestion control paramenters. 2349 */ 2350 if (tp->t_congctl->fast_retransmit(tp, th)) { 2351 /* False fast retransmit */ 2352 break; 2353 } else 2354 goto drop; 2355 } else if (tp->t_dupacks > tcprexmtthresh) { 2356 tp->snd_cwnd += tp->t_segsz; 2357 (void) tcp_output(tp); 2358 goto drop; 2359 } 2360 } else { 2361 /* 2362 * If the ack appears to be very old, only 2363 * allow data that is in-sequence. This 2364 * makes it somewhat more difficult to insert 2365 * forged data by guessing sequence numbers. 2366 * Sent an ack to try to update the send 2367 * sequence number on the other side. 2368 */ 2369 if (tlen && th->th_seq != tp->rcv_nxt && 2370 SEQ_LT(th->th_ack, 2371 tp->snd_una - tp->max_sndwnd)) 2372 goto dropafterack; 2373 } 2374 break; 2375 } 2376 /* 2377 * If the congestion window was inflated to account 2378 * for the other side's cached packets, retract it. 2379 */ 2380 /* XXX: make SACK have his own congestion control 2381 * struct -- rpaulo */ 2382 if (TCP_SACK_ENABLED(tp)) 2383 tcp_sack_newack(tp, th); 2384 else 2385 tp->t_congctl->fast_retransmit_newack(tp, th); 2386 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2387 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2388 goto dropafterack; 2389 } 2390 acked = th->th_ack - tp->snd_una; 2391 tcps = TCP_STAT_GETREF(); 2392 tcps[TCP_STAT_RCVACKPACK]++; 2393 tcps[TCP_STAT_RCVACKBYTE] += acked; 2394 TCP_STAT_PUTREF(); 2395 2396 /* 2397 * If we have a timestamp reply, update smoothed 2398 * round trip time. If no timestamp is present but 2399 * transmit timer is running and timed sequence 2400 * number was acked, update smoothed round trip time. 2401 * Since we now have an rtt measurement, cancel the 2402 * timer backoff (cf., Phil Karn's retransmit alg.). 2403 * Recompute the initial retransmit timer. 2404 */ 2405 if (ts_rtt) 2406 tcp_xmit_timer(tp, ts_rtt); 2407 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2408 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2409 2410 /* 2411 * If all outstanding data is acked, stop retransmit 2412 * timer and remember to restart (more output or persist). 2413 * If there is more data to be acked, restart retransmit 2414 * timer, using current (possibly backed-off) value. 2415 */ 2416 if (th->th_ack == tp->snd_max) { 2417 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2418 needoutput = 1; 2419 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2420 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2421 2422 /* 2423 * New data has been acked, adjust the congestion window. 2424 */ 2425 tp->t_congctl->newack(tp, th); 2426 2427 nd6_hint(tp); 2428 if (acked > so->so_snd.sb_cc) { 2429 tp->snd_wnd -= so->so_snd.sb_cc; 2430 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2431 ourfinisacked = 1; 2432 } else { 2433 if (acked > (tp->t_lastoff - tp->t_inoff)) 2434 tp->t_lastm = NULL; 2435 sbdrop(&so->so_snd, acked); 2436 tp->t_lastoff -= acked; 2437 tp->snd_wnd -= acked; 2438 ourfinisacked = 0; 2439 } 2440 sowwakeup(so); 2441 2442 icmp_check(tp, th, acked); 2443 2444 tp->snd_una = th->th_ack; 2445 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2446 tp->snd_fack = tp->snd_una; 2447 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2448 tp->snd_nxt = tp->snd_una; 2449 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2450 tp->snd_high = tp->snd_una; 2451 2452 switch (tp->t_state) { 2453 2454 /* 2455 * In FIN_WAIT_1 STATE in addition to the processing 2456 * for the ESTABLISHED state if our FIN is now acknowledged 2457 * then enter FIN_WAIT_2. 2458 */ 2459 case TCPS_FIN_WAIT_1: 2460 if (ourfinisacked) { 2461 /* 2462 * If we can't receive any more 2463 * data, then closing user can proceed. 2464 * Starting the timer is contrary to the 2465 * specification, but if we don't get a FIN 2466 * we'll hang forever. 2467 */ 2468 if (so->so_state & SS_CANTRCVMORE) { 2469 soisdisconnected(so); 2470 if (tp->t_maxidle > 0) 2471 TCP_TIMER_ARM(tp, TCPT_2MSL, 2472 tp->t_maxidle); 2473 } 2474 tp->t_state = TCPS_FIN_WAIT_2; 2475 } 2476 break; 2477 2478 /* 2479 * In CLOSING STATE in addition to the processing for 2480 * the ESTABLISHED state if the ACK acknowledges our FIN 2481 * then enter the TIME-WAIT state, otherwise ignore 2482 * the segment. 2483 */ 2484 case TCPS_CLOSING: 2485 if (ourfinisacked) { 2486 tp->t_state = TCPS_TIME_WAIT; 2487 tcp_canceltimers(tp); 2488 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2489 soisdisconnected(so); 2490 } 2491 break; 2492 2493 /* 2494 * In LAST_ACK, we may still be waiting for data to drain 2495 * and/or to be acked, as well as for the ack of our FIN. 2496 * If our FIN is now acknowledged, delete the TCB, 2497 * enter the closed state and return. 2498 */ 2499 case TCPS_LAST_ACK: 2500 if (ourfinisacked) { 2501 tp = tcp_close(tp); 2502 goto drop; 2503 } 2504 break; 2505 2506 /* 2507 * In TIME_WAIT state the only thing that should arrive 2508 * is a retransmission of the remote FIN. Acknowledge 2509 * it and restart the finack timer. 2510 */ 2511 case TCPS_TIME_WAIT: 2512 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2513 goto dropafterack; 2514 } 2515 } 2516 2517 step6: 2518 /* 2519 * Update window information. 2520 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2521 */ 2522 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2523 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2524 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2525 /* keep track of pure window updates */ 2526 if (tlen == 0 && 2527 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2528 TCP_STATINC(TCP_STAT_RCVWINUPD); 2529 tp->snd_wnd = tiwin; 2530 tp->snd_wl1 = th->th_seq; 2531 tp->snd_wl2 = th->th_ack; 2532 if (tp->snd_wnd > tp->max_sndwnd) 2533 tp->max_sndwnd = tp->snd_wnd; 2534 needoutput = 1; 2535 } 2536 2537 /* 2538 * Process segments with URG. 2539 */ 2540 if ((tiflags & TH_URG) && th->th_urp && 2541 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2542 /* 2543 * This is a kludge, but if we receive and accept 2544 * random urgent pointers, we'll crash in 2545 * soreceive. It's hard to imagine someone 2546 * actually wanting to send this much urgent data. 2547 */ 2548 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2549 th->th_urp = 0; /* XXX */ 2550 tiflags &= ~TH_URG; /* XXX */ 2551 goto dodata; /* XXX */ 2552 } 2553 /* 2554 * If this segment advances the known urgent pointer, 2555 * then mark the data stream. This should not happen 2556 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2557 * a FIN has been received from the remote side. 2558 * In these states we ignore the URG. 2559 * 2560 * According to RFC961 (Assigned Protocols), 2561 * the urgent pointer points to the last octet 2562 * of urgent data. We continue, however, 2563 * to consider it to indicate the first octet 2564 * of data past the urgent section as the original 2565 * spec states (in one of two places). 2566 */ 2567 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2568 tp->rcv_up = th->th_seq + th->th_urp; 2569 so->so_oobmark = so->so_rcv.sb_cc + 2570 (tp->rcv_up - tp->rcv_nxt) - 1; 2571 if (so->so_oobmark == 0) 2572 so->so_state |= SS_RCVATMARK; 2573 sohasoutofband(so); 2574 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2575 } 2576 /* 2577 * Remove out of band data so doesn't get presented to user. 2578 * This can happen independent of advancing the URG pointer, 2579 * but if two URG's are pending at once, some out-of-band 2580 * data may creep in... ick. 2581 */ 2582 if (th->th_urp <= (u_int16_t) tlen 2583 #ifdef SO_OOBINLINE 2584 && (so->so_options & SO_OOBINLINE) == 0 2585 #endif 2586 ) 2587 tcp_pulloutofband(so, th, m, hdroptlen); 2588 } else 2589 /* 2590 * If no out of band data is expected, 2591 * pull receive urgent pointer along 2592 * with the receive window. 2593 */ 2594 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2595 tp->rcv_up = tp->rcv_nxt; 2596 dodata: /* XXX */ 2597 2598 /* 2599 * Process the segment text, merging it into the TCP sequencing queue, 2600 * and arranging for acknowledgement of receipt if necessary. 2601 * This process logically involves adjusting tp->rcv_wnd as data 2602 * is presented to the user (this happens in tcp_usrreq.c, 2603 * case PRU_RCVD). If a FIN has already been received on this 2604 * connection then we just ignore the text. 2605 */ 2606 if ((tlen || (tiflags & TH_FIN)) && 2607 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2608 /* 2609 * Insert segment ti into reassembly queue of tcp with 2610 * control block tp. Return TH_FIN if reassembly now includes 2611 * a segment with FIN. The macro form does the common case 2612 * inline (segment is the next to be received on an 2613 * established connection, and the queue is empty), 2614 * avoiding linkage into and removal from the queue and 2615 * repetition of various conversions. 2616 * Set DELACK for segments received in order, but ack 2617 * immediately when segments are out of order 2618 * (so fast retransmit can work). 2619 */ 2620 /* NOTE: this was TCP_REASS() macro, but used only once */ 2621 TCP_REASS_LOCK(tp); 2622 if (th->th_seq == tp->rcv_nxt && 2623 TAILQ_FIRST(&tp->segq) == NULL && 2624 tp->t_state == TCPS_ESTABLISHED) { 2625 tcp_setup_ack(tp, th); 2626 tp->rcv_nxt += tlen; 2627 tiflags = th->th_flags & TH_FIN; 2628 tcps = TCP_STAT_GETREF(); 2629 tcps[TCP_STAT_RCVPACK]++; 2630 tcps[TCP_STAT_RCVBYTE] += tlen; 2631 TCP_STAT_PUTREF(); 2632 nd6_hint(tp); 2633 if (so->so_state & SS_CANTRCVMORE) 2634 m_freem(m); 2635 else { 2636 m_adj(m, hdroptlen); 2637 sbappendstream(&(so)->so_rcv, m); 2638 } 2639 sorwakeup(so); 2640 } else { 2641 m_adj(m, hdroptlen); 2642 tiflags = tcp_reass(tp, th, m, &tlen); 2643 tp->t_flags |= TF_ACKNOW; 2644 } 2645 TCP_REASS_UNLOCK(tp); 2646 2647 /* 2648 * Note the amount of data that peer has sent into 2649 * our window, in order to estimate the sender's 2650 * buffer size. 2651 */ 2652 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2653 } else { 2654 m_freem(m); 2655 m = NULL; 2656 tiflags &= ~TH_FIN; 2657 } 2658 2659 /* 2660 * If FIN is received ACK the FIN and let the user know 2661 * that the connection is closing. Ignore a FIN received before 2662 * the connection is fully established. 2663 */ 2664 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2665 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2666 socantrcvmore(so); 2667 tp->t_flags |= TF_ACKNOW; 2668 tp->rcv_nxt++; 2669 } 2670 switch (tp->t_state) { 2671 2672 /* 2673 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2674 */ 2675 case TCPS_ESTABLISHED: 2676 tp->t_state = TCPS_CLOSE_WAIT; 2677 break; 2678 2679 /* 2680 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2681 * enter the CLOSING state. 2682 */ 2683 case TCPS_FIN_WAIT_1: 2684 tp->t_state = TCPS_CLOSING; 2685 break; 2686 2687 /* 2688 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2689 * starting the time-wait timer, turning off the other 2690 * standard timers. 2691 */ 2692 case TCPS_FIN_WAIT_2: 2693 tp->t_state = TCPS_TIME_WAIT; 2694 tcp_canceltimers(tp); 2695 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2696 soisdisconnected(so); 2697 break; 2698 2699 /* 2700 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2701 */ 2702 case TCPS_TIME_WAIT: 2703 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2704 break; 2705 } 2706 } 2707 #ifdef TCP_DEBUG 2708 if (so->so_options & SO_DEBUG) 2709 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2710 #endif 2711 2712 /* 2713 * Return any desired output. 2714 */ 2715 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2716 (void) tcp_output(tp); 2717 } 2718 if (tcp_saveti) 2719 m_freem(tcp_saveti); 2720 return; 2721 2722 badsyn: 2723 /* 2724 * Received a bad SYN. Increment counters and dropwithreset. 2725 */ 2726 TCP_STATINC(TCP_STAT_BADSYN); 2727 tp = NULL; 2728 goto dropwithreset; 2729 2730 dropafterack: 2731 /* 2732 * Generate an ACK dropping incoming segment if it occupies 2733 * sequence space, where the ACK reflects our state. 2734 */ 2735 if (tiflags & TH_RST) 2736 goto drop; 2737 goto dropafterack2; 2738 2739 dropafterack_ratelim: 2740 /* 2741 * We may want to rate-limit ACKs against SYN/RST attack. 2742 */ 2743 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2744 tcp_ackdrop_ppslim) == 0) { 2745 /* XXX stat */ 2746 goto drop; 2747 } 2748 /* ...fall into dropafterack2... */ 2749 2750 dropafterack2: 2751 m_freem(m); 2752 tp->t_flags |= TF_ACKNOW; 2753 (void) tcp_output(tp); 2754 if (tcp_saveti) 2755 m_freem(tcp_saveti); 2756 return; 2757 2758 dropwithreset_ratelim: 2759 /* 2760 * We may want to rate-limit RSTs in certain situations, 2761 * particularly if we are sending an RST in response to 2762 * an attempt to connect to or otherwise communicate with 2763 * a port for which we have no socket. 2764 */ 2765 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2766 tcp_rst_ppslim) == 0) { 2767 /* XXX stat */ 2768 goto drop; 2769 } 2770 /* ...fall into dropwithreset... */ 2771 2772 dropwithreset: 2773 /* 2774 * Generate a RST, dropping incoming segment. 2775 * Make ACK acceptable to originator of segment. 2776 */ 2777 if (tiflags & TH_RST) 2778 goto drop; 2779 2780 switch (af) { 2781 #ifdef INET6 2782 case AF_INET6: 2783 /* For following calls to tcp_respond */ 2784 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2785 goto drop; 2786 break; 2787 #endif /* INET6 */ 2788 case AF_INET: 2789 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2790 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2791 goto drop; 2792 } 2793 2794 if (tiflags & TH_ACK) 2795 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2796 else { 2797 if (tiflags & TH_SYN) 2798 tlen++; 2799 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2800 TH_RST|TH_ACK); 2801 } 2802 if (tcp_saveti) 2803 m_freem(tcp_saveti); 2804 return; 2805 2806 badcsum: 2807 drop: 2808 /* 2809 * Drop space held by incoming segment and return. 2810 */ 2811 if (tp) { 2812 if (tp->t_inpcb) 2813 so = tp->t_inpcb->inp_socket; 2814 #ifdef INET6 2815 else if (tp->t_in6pcb) 2816 so = tp->t_in6pcb->in6p_socket; 2817 #endif 2818 else 2819 so = NULL; 2820 #ifdef TCP_DEBUG 2821 if (so && (so->so_options & SO_DEBUG) != 0) 2822 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2823 #endif 2824 } 2825 if (tcp_saveti) 2826 m_freem(tcp_saveti); 2827 m_freem(m); 2828 return; 2829 } 2830 2831 #ifdef TCP_SIGNATURE 2832 int 2833 tcp_signature_apply(void *fstate, void *data, u_int len) 2834 { 2835 2836 MD5Update(fstate, (u_char *)data, len); 2837 return (0); 2838 } 2839 2840 struct secasvar * 2841 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2842 { 2843 struct secasvar *sav; 2844 #ifdef FAST_IPSEC 2845 union sockaddr_union dst; 2846 #endif 2847 struct ip *ip; 2848 struct ip6_hdr *ip6; 2849 2850 ip = mtod(m, struct ip *); 2851 switch (ip->ip_v) { 2852 case 4: 2853 ip = mtod(m, struct ip *); 2854 ip6 = NULL; 2855 break; 2856 case 6: 2857 ip = NULL; 2858 ip6 = mtod(m, struct ip6_hdr *); 2859 break; 2860 default: 2861 return (NULL); 2862 } 2863 2864 #ifdef FAST_IPSEC 2865 /* Extract the destination from the IP header in the mbuf. */ 2866 bzero(&dst, sizeof(union sockaddr_union)); 2867 if (ip !=NULL) { 2868 dst.sa.sa_len = sizeof(struct sockaddr_in); 2869 dst.sa.sa_family = AF_INET; 2870 dst.sin.sin_addr = ip->ip_dst; 2871 } else { 2872 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2873 dst.sa.sa_family = AF_INET6; 2874 dst.sin6.sin6_addr = ip6->ip6_dst; 2875 } 2876 2877 /* 2878 * Look up an SADB entry which matches the address of the peer. 2879 */ 2880 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2881 #else 2882 if (ip) 2883 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 2884 (void *)&ip->ip_dst, IPPROTO_TCP, 2885 htonl(TCP_SIG_SPI), 0, 0); 2886 else 2887 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 2888 (void *)&ip6->ip6_dst, IPPROTO_TCP, 2889 htonl(TCP_SIG_SPI), 0, 0); 2890 #endif 2891 2892 return (sav); /* freesav must be performed by caller */ 2893 } 2894 2895 int 2896 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2897 struct secasvar *sav, char *sig) 2898 { 2899 MD5_CTX ctx; 2900 struct ip *ip; 2901 struct ipovly *ipovly; 2902 struct ip6_hdr *ip6; 2903 struct ippseudo ippseudo; 2904 struct ip6_hdr_pseudo ip6pseudo; 2905 struct tcphdr th0; 2906 int l, tcphdrlen; 2907 2908 if (sav == NULL) 2909 return (-1); 2910 2911 tcphdrlen = th->th_off * 4; 2912 2913 switch (mtod(m, struct ip *)->ip_v) { 2914 case 4: 2915 ip = mtod(m, struct ip *); 2916 ip6 = NULL; 2917 break; 2918 case 6: 2919 ip = NULL; 2920 ip6 = mtod(m, struct ip6_hdr *); 2921 break; 2922 default: 2923 return (-1); 2924 } 2925 2926 MD5Init(&ctx); 2927 2928 if (ip) { 2929 memset(&ippseudo, 0, sizeof(ippseudo)); 2930 ipovly = (struct ipovly *)ip; 2931 ippseudo.ippseudo_src = ipovly->ih_src; 2932 ippseudo.ippseudo_dst = ipovly->ih_dst; 2933 ippseudo.ippseudo_pad = 0; 2934 ippseudo.ippseudo_p = IPPROTO_TCP; 2935 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2936 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2937 } else { 2938 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2939 ip6pseudo.ip6ph_src = ip6->ip6_src; 2940 in6_clearscope(&ip6pseudo.ip6ph_src); 2941 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2942 in6_clearscope(&ip6pseudo.ip6ph_dst); 2943 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2944 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2945 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2946 } 2947 2948 th0 = *th; 2949 th0.th_sum = 0; 2950 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2951 2952 l = m->m_pkthdr.len - thoff - tcphdrlen; 2953 if (l > 0) 2954 m_apply(m, thoff + tcphdrlen, 2955 m->m_pkthdr.len - thoff - tcphdrlen, 2956 tcp_signature_apply, &ctx); 2957 2958 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2959 MD5Final(sig, &ctx); 2960 2961 return (0); 2962 } 2963 #endif 2964 2965 static int 2966 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 2967 struct tcphdr *th, 2968 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2969 { 2970 u_int16_t mss; 2971 int opt, optlen = 0; 2972 #ifdef TCP_SIGNATURE 2973 void *sigp = NULL; 2974 char sigbuf[TCP_SIGLEN]; 2975 struct secasvar *sav = NULL; 2976 #endif 2977 2978 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2979 opt = cp[0]; 2980 if (opt == TCPOPT_EOL) 2981 break; 2982 if (opt == TCPOPT_NOP) 2983 optlen = 1; 2984 else { 2985 if (cnt < 2) 2986 break; 2987 optlen = cp[1]; 2988 if (optlen < 2 || optlen > cnt) 2989 break; 2990 } 2991 switch (opt) { 2992 2993 default: 2994 continue; 2995 2996 case TCPOPT_MAXSEG: 2997 if (optlen != TCPOLEN_MAXSEG) 2998 continue; 2999 if (!(th->th_flags & TH_SYN)) 3000 continue; 3001 if (TCPS_HAVERCVDSYN(tp->t_state)) 3002 continue; 3003 bcopy(cp + 2, &mss, sizeof(mss)); 3004 oi->maxseg = ntohs(mss); 3005 break; 3006 3007 case TCPOPT_WINDOW: 3008 if (optlen != TCPOLEN_WINDOW) 3009 continue; 3010 if (!(th->th_flags & TH_SYN)) 3011 continue; 3012 if (TCPS_HAVERCVDSYN(tp->t_state)) 3013 continue; 3014 tp->t_flags |= TF_RCVD_SCALE; 3015 tp->requested_s_scale = cp[2]; 3016 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3017 #if 0 /*XXX*/ 3018 char *p; 3019 3020 if (ip) 3021 p = ntohl(ip->ip_src); 3022 #ifdef INET6 3023 else if (ip6) 3024 p = ip6_sprintf(&ip6->ip6_src); 3025 #endif 3026 else 3027 p = "(unknown)"; 3028 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3029 "assuming %d\n", 3030 tp->requested_s_scale, p, 3031 TCP_MAX_WINSHIFT); 3032 #else 3033 log(LOG_ERR, "TCP: invalid wscale %d, " 3034 "assuming %d\n", 3035 tp->requested_s_scale, 3036 TCP_MAX_WINSHIFT); 3037 #endif 3038 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3039 } 3040 break; 3041 3042 case TCPOPT_TIMESTAMP: 3043 if (optlen != TCPOLEN_TIMESTAMP) 3044 continue; 3045 oi->ts_present = 1; 3046 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3047 NTOHL(oi->ts_val); 3048 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3049 NTOHL(oi->ts_ecr); 3050 3051 if (!(th->th_flags & TH_SYN)) 3052 continue; 3053 if (TCPS_HAVERCVDSYN(tp->t_state)) 3054 continue; 3055 /* 3056 * A timestamp received in a SYN makes 3057 * it ok to send timestamp requests and replies. 3058 */ 3059 tp->t_flags |= TF_RCVD_TSTMP; 3060 tp->ts_recent = oi->ts_val; 3061 tp->ts_recent_age = tcp_now; 3062 break; 3063 3064 case TCPOPT_SACK_PERMITTED: 3065 if (optlen != TCPOLEN_SACK_PERMITTED) 3066 continue; 3067 if (!(th->th_flags & TH_SYN)) 3068 continue; 3069 if (TCPS_HAVERCVDSYN(tp->t_state)) 3070 continue; 3071 if (tcp_do_sack) { 3072 tp->t_flags |= TF_SACK_PERMIT; 3073 tp->t_flags |= TF_WILL_SACK; 3074 } 3075 break; 3076 3077 case TCPOPT_SACK: 3078 tcp_sack_option(tp, th, cp, optlen); 3079 break; 3080 #ifdef TCP_SIGNATURE 3081 case TCPOPT_SIGNATURE: 3082 if (optlen != TCPOLEN_SIGNATURE) 3083 continue; 3084 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 3085 return (-1); 3086 3087 sigp = sigbuf; 3088 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3089 tp->t_flags |= TF_SIGNATURE; 3090 break; 3091 #endif 3092 } 3093 } 3094 3095 #ifdef TCP_SIGNATURE 3096 if (tp->t_flags & TF_SIGNATURE) { 3097 3098 sav = tcp_signature_getsav(m, th); 3099 3100 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3101 return (-1); 3102 } 3103 3104 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3105 if (sav == NULL) 3106 return (-1); 3107 #ifdef FAST_IPSEC 3108 KEY_FREESAV(&sav); 3109 #else 3110 key_freesav(sav); 3111 #endif 3112 return (-1); 3113 } 3114 3115 if (sigp) { 3116 char sig[TCP_SIGLEN]; 3117 3118 tcp_fields_to_net(th); 3119 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3120 tcp_fields_to_host(th); 3121 if (sav == NULL) 3122 return (-1); 3123 #ifdef FAST_IPSEC 3124 KEY_FREESAV(&sav); 3125 #else 3126 key_freesav(sav); 3127 #endif 3128 return (-1); 3129 } 3130 tcp_fields_to_host(th); 3131 3132 if (bcmp(sig, sigp, TCP_SIGLEN)) { 3133 TCP_STATINC(TCP_STAT_BADSIG); 3134 if (sav == NULL) 3135 return (-1); 3136 #ifdef FAST_IPSEC 3137 KEY_FREESAV(&sav); 3138 #else 3139 key_freesav(sav); 3140 #endif 3141 return (-1); 3142 } else 3143 TCP_STATINC(TCP_STAT_GOODSIG); 3144 3145 key_sa_recordxfer(sav, m); 3146 #ifdef FAST_IPSEC 3147 KEY_FREESAV(&sav); 3148 #else 3149 key_freesav(sav); 3150 #endif 3151 } 3152 #endif 3153 3154 return (0); 3155 } 3156 3157 /* 3158 * Pull out of band byte out of a segment so 3159 * it doesn't appear in the user's data queue. 3160 * It is still reflected in the segment length for 3161 * sequencing purposes. 3162 */ 3163 void 3164 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3165 struct mbuf *m, int off) 3166 { 3167 int cnt = off + th->th_urp - 1; 3168 3169 while (cnt >= 0) { 3170 if (m->m_len > cnt) { 3171 char *cp = mtod(m, char *) + cnt; 3172 struct tcpcb *tp = sototcpcb(so); 3173 3174 tp->t_iobc = *cp; 3175 tp->t_oobflags |= TCPOOB_HAVEDATA; 3176 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3177 m->m_len--; 3178 return; 3179 } 3180 cnt -= m->m_len; 3181 m = m->m_next; 3182 if (m == 0) 3183 break; 3184 } 3185 panic("tcp_pulloutofband"); 3186 } 3187 3188 /* 3189 * Collect new round-trip time estimate 3190 * and update averages and current timeout. 3191 */ 3192 void 3193 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3194 { 3195 int32_t delta; 3196 3197 TCP_STATINC(TCP_STAT_RTTUPDATED); 3198 if (tp->t_srtt != 0) { 3199 /* 3200 * srtt is stored as fixed point with 3 bits after the 3201 * binary point (i.e., scaled by 8). The following magic 3202 * is equivalent to the smoothing algorithm in rfc793 with 3203 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3204 * point). Adjust rtt to origin 0. 3205 */ 3206 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3207 if ((tp->t_srtt += delta) <= 0) 3208 tp->t_srtt = 1 << 2; 3209 /* 3210 * We accumulate a smoothed rtt variance (actually, a 3211 * smoothed mean difference), then set the retransmit 3212 * timer to smoothed rtt + 4 times the smoothed variance. 3213 * rttvar is stored as fixed point with 2 bits after the 3214 * binary point (scaled by 4). The following is 3215 * equivalent to rfc793 smoothing with an alpha of .75 3216 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3217 * rfc793's wired-in beta. 3218 */ 3219 if (delta < 0) 3220 delta = -delta; 3221 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3222 if ((tp->t_rttvar += delta) <= 0) 3223 tp->t_rttvar = 1 << 2; 3224 } else { 3225 /* 3226 * No rtt measurement yet - use the unsmoothed rtt. 3227 * Set the variance to half the rtt (so our first 3228 * retransmit happens at 3*rtt). 3229 */ 3230 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3231 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3232 } 3233 tp->t_rtttime = 0; 3234 tp->t_rxtshift = 0; 3235 3236 /* 3237 * the retransmit should happen at rtt + 4 * rttvar. 3238 * Because of the way we do the smoothing, srtt and rttvar 3239 * will each average +1/2 tick of bias. When we compute 3240 * the retransmit timer, we want 1/2 tick of rounding and 3241 * 1 extra tick because of +-1/2 tick uncertainty in the 3242 * firing of the timer. The bias will give us exactly the 3243 * 1.5 tick we need. But, because the bias is 3244 * statistical, we have to test that we don't drop below 3245 * the minimum feasible timer (which is 2 ticks). 3246 */ 3247 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3248 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3249 3250 /* 3251 * We received an ack for a packet that wasn't retransmitted; 3252 * it is probably safe to discard any error indications we've 3253 * received recently. This isn't quite right, but close enough 3254 * for now (a route might have failed after we sent a segment, 3255 * and the return path might not be symmetrical). 3256 */ 3257 tp->t_softerror = 0; 3258 } 3259 3260 3261 /* 3262 * TCP compressed state engine. Currently used to hold compressed 3263 * state for SYN_RECEIVED. 3264 */ 3265 3266 u_long syn_cache_count; 3267 u_int32_t syn_hash1, syn_hash2; 3268 3269 #define SYN_HASH(sa, sp, dp) \ 3270 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3271 ((u_int32_t)(sp)))^syn_hash2))) 3272 #ifndef INET6 3273 #define SYN_HASHALL(hash, src, dst) \ 3274 do { \ 3275 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3276 ((const struct sockaddr_in *)(src))->sin_port, \ 3277 ((const struct sockaddr_in *)(dst))->sin_port); \ 3278 } while (/*CONSTCOND*/ 0) 3279 #else 3280 #define SYN_HASH6(sa, sp, dp) \ 3281 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3282 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3283 & 0x7fffffff) 3284 3285 #define SYN_HASHALL(hash, src, dst) \ 3286 do { \ 3287 switch ((src)->sa_family) { \ 3288 case AF_INET: \ 3289 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3290 ((const struct sockaddr_in *)(src))->sin_port, \ 3291 ((const struct sockaddr_in *)(dst))->sin_port); \ 3292 break; \ 3293 case AF_INET6: \ 3294 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3295 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3296 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3297 break; \ 3298 default: \ 3299 hash = 0; \ 3300 } \ 3301 } while (/*CONSTCOND*/0) 3302 #endif /* INET6 */ 3303 3304 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL, 3305 IPL_SOFTNET); 3306 3307 /* 3308 * We don't estimate RTT with SYNs, so each packet starts with the default 3309 * RTT and each timer step has a fixed timeout value. 3310 */ 3311 #define SYN_CACHE_TIMER_ARM(sc) \ 3312 do { \ 3313 TCPT_RANGESET((sc)->sc_rxtcur, \ 3314 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3315 TCPTV_REXMTMAX); \ 3316 callout_reset(&(sc)->sc_timer, \ 3317 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3318 } while (/*CONSTCOND*/0) 3319 3320 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3321 3322 static inline void 3323 syn_cache_rm(struct syn_cache *sc) 3324 { 3325 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3326 sc, sc_bucketq); 3327 sc->sc_tp = NULL; 3328 LIST_REMOVE(sc, sc_tpq); 3329 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3330 callout_stop(&sc->sc_timer); 3331 syn_cache_count--; 3332 } 3333 3334 static inline void 3335 syn_cache_put(struct syn_cache *sc) 3336 { 3337 if (sc->sc_ipopts) 3338 (void) m_free(sc->sc_ipopts); 3339 rtcache_free(&sc->sc_route); 3340 if (callout_invoking(&sc->sc_timer)) 3341 sc->sc_flags |= SCF_DEAD; 3342 else { 3343 callout_destroy(&sc->sc_timer); 3344 pool_put(&syn_cache_pool, sc); 3345 } 3346 } 3347 3348 void 3349 syn_cache_init(void) 3350 { 3351 int i; 3352 3353 /* Initialize the hash buckets. */ 3354 for (i = 0; i < tcp_syn_cache_size; i++) 3355 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3356 } 3357 3358 void 3359 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3360 { 3361 struct syn_cache_head *scp; 3362 struct syn_cache *sc2; 3363 int s; 3364 3365 /* 3366 * If there are no entries in the hash table, reinitialize 3367 * the hash secrets. 3368 */ 3369 if (syn_cache_count == 0) { 3370 syn_hash1 = arc4random(); 3371 syn_hash2 = arc4random(); 3372 } 3373 3374 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3375 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3376 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3377 3378 /* 3379 * Make sure that we don't overflow the per-bucket 3380 * limit or the total cache size limit. 3381 */ 3382 s = splsoftnet(); 3383 if (scp->sch_length >= tcp_syn_bucket_limit) { 3384 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3385 /* 3386 * The bucket is full. Toss the oldest element in the 3387 * bucket. This will be the first entry in the bucket. 3388 */ 3389 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3390 #ifdef DIAGNOSTIC 3391 /* 3392 * This should never happen; we should always find an 3393 * entry in our bucket. 3394 */ 3395 if (sc2 == NULL) 3396 panic("syn_cache_insert: bucketoverflow: impossible"); 3397 #endif 3398 syn_cache_rm(sc2); 3399 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3400 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3401 struct syn_cache_head *scp2, *sce; 3402 3403 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3404 /* 3405 * The cache is full. Toss the oldest entry in the 3406 * first non-empty bucket we can find. 3407 * 3408 * XXX We would really like to toss the oldest 3409 * entry in the cache, but we hope that this 3410 * condition doesn't happen very often. 3411 */ 3412 scp2 = scp; 3413 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3414 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3415 for (++scp2; scp2 != scp; scp2++) { 3416 if (scp2 >= sce) 3417 scp2 = &tcp_syn_cache[0]; 3418 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3419 break; 3420 } 3421 #ifdef DIAGNOSTIC 3422 /* 3423 * This should never happen; we should always find a 3424 * non-empty bucket. 3425 */ 3426 if (scp2 == scp) 3427 panic("syn_cache_insert: cacheoverflow: " 3428 "impossible"); 3429 #endif 3430 } 3431 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3432 syn_cache_rm(sc2); 3433 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3434 } 3435 3436 /* 3437 * Initialize the entry's timer. 3438 */ 3439 sc->sc_rxttot = 0; 3440 sc->sc_rxtshift = 0; 3441 SYN_CACHE_TIMER_ARM(sc); 3442 3443 /* Link it from tcpcb entry */ 3444 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3445 3446 /* Put it into the bucket. */ 3447 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3448 scp->sch_length++; 3449 syn_cache_count++; 3450 3451 TCP_STATINC(TCP_STAT_SC_ADDED); 3452 splx(s); 3453 } 3454 3455 /* 3456 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3457 * If we have retransmitted an entry the maximum number of times, expire 3458 * that entry. 3459 */ 3460 void 3461 syn_cache_timer(void *arg) 3462 { 3463 struct syn_cache *sc = arg; 3464 3465 mutex_enter(softnet_lock); 3466 KERNEL_LOCK(1, NULL); 3467 callout_ack(&sc->sc_timer); 3468 3469 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3470 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3471 callout_destroy(&sc->sc_timer); 3472 pool_put(&syn_cache_pool, sc); 3473 KERNEL_UNLOCK_ONE(NULL); 3474 mutex_exit(softnet_lock); 3475 return; 3476 } 3477 3478 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3479 /* Drop it -- too many retransmissions. */ 3480 goto dropit; 3481 } 3482 3483 /* 3484 * Compute the total amount of time this entry has 3485 * been on a queue. If this entry has been on longer 3486 * than the keep alive timer would allow, expire it. 3487 */ 3488 sc->sc_rxttot += sc->sc_rxtcur; 3489 if (sc->sc_rxttot >= tcp_keepinit) 3490 goto dropit; 3491 3492 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3493 (void) syn_cache_respond(sc, NULL); 3494 3495 /* Advance the timer back-off. */ 3496 sc->sc_rxtshift++; 3497 SYN_CACHE_TIMER_ARM(sc); 3498 3499 KERNEL_UNLOCK_ONE(NULL); 3500 mutex_exit(softnet_lock); 3501 return; 3502 3503 dropit: 3504 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3505 syn_cache_rm(sc); 3506 syn_cache_put(sc); /* calls pool_put but see spl above */ 3507 KERNEL_UNLOCK_ONE(NULL); 3508 mutex_exit(softnet_lock); 3509 } 3510 3511 /* 3512 * Remove syn cache created by the specified tcb entry, 3513 * because this does not make sense to keep them 3514 * (if there's no tcb entry, syn cache entry will never be used) 3515 */ 3516 void 3517 syn_cache_cleanup(struct tcpcb *tp) 3518 { 3519 struct syn_cache *sc, *nsc; 3520 int s; 3521 3522 s = splsoftnet(); 3523 3524 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3525 nsc = LIST_NEXT(sc, sc_tpq); 3526 3527 #ifdef DIAGNOSTIC 3528 if (sc->sc_tp != tp) 3529 panic("invalid sc_tp in syn_cache_cleanup"); 3530 #endif 3531 syn_cache_rm(sc); 3532 syn_cache_put(sc); /* calls pool_put but see spl above */ 3533 } 3534 /* just for safety */ 3535 LIST_INIT(&tp->t_sc); 3536 3537 splx(s); 3538 } 3539 3540 /* 3541 * Find an entry in the syn cache. 3542 */ 3543 struct syn_cache * 3544 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3545 struct syn_cache_head **headp) 3546 { 3547 struct syn_cache *sc; 3548 struct syn_cache_head *scp; 3549 u_int32_t hash; 3550 int s; 3551 3552 SYN_HASHALL(hash, src, dst); 3553 3554 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3555 *headp = scp; 3556 s = splsoftnet(); 3557 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3558 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3559 if (sc->sc_hash != hash) 3560 continue; 3561 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3562 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3563 splx(s); 3564 return (sc); 3565 } 3566 } 3567 splx(s); 3568 return (NULL); 3569 } 3570 3571 /* 3572 * This function gets called when we receive an ACK for a 3573 * socket in the LISTEN state. We look up the connection 3574 * in the syn cache, and if its there, we pull it out of 3575 * the cache and turn it into a full-blown connection in 3576 * the SYN-RECEIVED state. 3577 * 3578 * The return values may not be immediately obvious, and their effects 3579 * can be subtle, so here they are: 3580 * 3581 * NULL SYN was not found in cache; caller should drop the 3582 * packet and send an RST. 3583 * 3584 * -1 We were unable to create the new connection, and are 3585 * aborting it. An ACK,RST is being sent to the peer 3586 * (unless we got screwey sequence numbners; see below), 3587 * because the 3-way handshake has been completed. Caller 3588 * should not free the mbuf, since we may be using it. If 3589 * we are not, we will free it. 3590 * 3591 * Otherwise, the return value is a pointer to the new socket 3592 * associated with the connection. 3593 */ 3594 struct socket * 3595 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3596 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3597 struct socket *so, struct mbuf *m) 3598 { 3599 struct syn_cache *sc; 3600 struct syn_cache_head *scp; 3601 struct inpcb *inp = NULL; 3602 #ifdef INET6 3603 struct in6pcb *in6p = NULL; 3604 #endif 3605 struct tcpcb *tp = 0; 3606 struct mbuf *am; 3607 int s; 3608 struct socket *oso; 3609 3610 s = splsoftnet(); 3611 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3612 splx(s); 3613 return (NULL); 3614 } 3615 3616 /* 3617 * Verify the sequence and ack numbers. Try getting the correct 3618 * response again. 3619 */ 3620 if ((th->th_ack != sc->sc_iss + 1) || 3621 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3622 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3623 (void) syn_cache_respond(sc, m); 3624 splx(s); 3625 return ((struct socket *)(-1)); 3626 } 3627 3628 /* Remove this cache entry */ 3629 syn_cache_rm(sc); 3630 splx(s); 3631 3632 /* 3633 * Ok, create the full blown connection, and set things up 3634 * as they would have been set up if we had created the 3635 * connection when the SYN arrived. If we can't create 3636 * the connection, abort it. 3637 */ 3638 /* 3639 * inp still has the OLD in_pcb stuff, set the 3640 * v6-related flags on the new guy, too. This is 3641 * done particularly for the case where an AF_INET6 3642 * socket is bound only to a port, and a v4 connection 3643 * comes in on that port. 3644 * we also copy the flowinfo from the original pcb 3645 * to the new one. 3646 */ 3647 oso = so; 3648 so = sonewconn(so, SS_ISCONNECTED); 3649 if (so == NULL) 3650 goto resetandabort; 3651 3652 switch (so->so_proto->pr_domain->dom_family) { 3653 #ifdef INET 3654 case AF_INET: 3655 inp = sotoinpcb(so); 3656 break; 3657 #endif 3658 #ifdef INET6 3659 case AF_INET6: 3660 in6p = sotoin6pcb(so); 3661 break; 3662 #endif 3663 } 3664 switch (src->sa_family) { 3665 #ifdef INET 3666 case AF_INET: 3667 if (inp) { 3668 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3669 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3670 inp->inp_options = ip_srcroute(); 3671 in_pcbstate(inp, INP_BOUND); 3672 if (inp->inp_options == NULL) { 3673 inp->inp_options = sc->sc_ipopts; 3674 sc->sc_ipopts = NULL; 3675 } 3676 } 3677 #ifdef INET6 3678 else if (in6p) { 3679 /* IPv4 packet to AF_INET6 socket */ 3680 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3681 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3682 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3683 &in6p->in6p_laddr.s6_addr32[3], 3684 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3685 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3686 in6totcpcb(in6p)->t_family = AF_INET; 3687 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3688 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3689 else 3690 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3691 in6_pcbstate(in6p, IN6P_BOUND); 3692 } 3693 #endif 3694 break; 3695 #endif 3696 #ifdef INET6 3697 case AF_INET6: 3698 if (in6p) { 3699 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3700 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3701 in6_pcbstate(in6p, IN6P_BOUND); 3702 } 3703 break; 3704 #endif 3705 } 3706 #ifdef INET6 3707 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3708 struct in6pcb *oin6p = sotoin6pcb(oso); 3709 /* inherit socket options from the listening socket */ 3710 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3711 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3712 m_freem(in6p->in6p_options); 3713 in6p->in6p_options = 0; 3714 } 3715 ip6_savecontrol(in6p, &in6p->in6p_options, 3716 mtod(m, struct ip6_hdr *), m); 3717 } 3718 #endif 3719 3720 #if defined(IPSEC) || defined(FAST_IPSEC) 3721 /* 3722 * we make a copy of policy, instead of sharing the policy, 3723 * for better behavior in terms of SA lookup and dead SA removal. 3724 */ 3725 if (inp) { 3726 /* copy old policy into new socket's */ 3727 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3728 printf("tcp_input: could not copy policy\n"); 3729 } 3730 #ifdef INET6 3731 else if (in6p) { 3732 /* copy old policy into new socket's */ 3733 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3734 in6p->in6p_sp)) 3735 printf("tcp_input: could not copy policy\n"); 3736 } 3737 #endif 3738 #endif 3739 3740 /* 3741 * Give the new socket our cached route reference. 3742 */ 3743 if (inp) { 3744 rtcache_copy(&inp->inp_route, &sc->sc_route); 3745 rtcache_free(&sc->sc_route); 3746 } 3747 #ifdef INET6 3748 else { 3749 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3750 rtcache_free(&sc->sc_route); 3751 } 3752 #endif 3753 3754 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3755 if (am == NULL) 3756 goto resetandabort; 3757 MCLAIM(am, &tcp_mowner); 3758 am->m_len = src->sa_len; 3759 bcopy(src, mtod(am, void *), src->sa_len); 3760 if (inp) { 3761 if (in_pcbconnect(inp, am, &lwp0)) { 3762 (void) m_free(am); 3763 goto resetandabort; 3764 } 3765 } 3766 #ifdef INET6 3767 else if (in6p) { 3768 if (src->sa_family == AF_INET) { 3769 /* IPv4 packet to AF_INET6 socket */ 3770 struct sockaddr_in6 *sin6; 3771 sin6 = mtod(am, struct sockaddr_in6 *); 3772 am->m_len = sizeof(*sin6); 3773 bzero(sin6, sizeof(*sin6)); 3774 sin6->sin6_family = AF_INET6; 3775 sin6->sin6_len = sizeof(*sin6); 3776 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3777 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3778 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3779 &sin6->sin6_addr.s6_addr32[3], 3780 sizeof(sin6->sin6_addr.s6_addr32[3])); 3781 } 3782 if (in6_pcbconnect(in6p, am, NULL)) { 3783 (void) m_free(am); 3784 goto resetandabort; 3785 } 3786 } 3787 #endif 3788 else { 3789 (void) m_free(am); 3790 goto resetandabort; 3791 } 3792 (void) m_free(am); 3793 3794 if (inp) 3795 tp = intotcpcb(inp); 3796 #ifdef INET6 3797 else if (in6p) 3798 tp = in6totcpcb(in6p); 3799 #endif 3800 else 3801 tp = NULL; 3802 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3803 if (sc->sc_request_r_scale != 15) { 3804 tp->requested_s_scale = sc->sc_requested_s_scale; 3805 tp->request_r_scale = sc->sc_request_r_scale; 3806 tp->snd_scale = sc->sc_requested_s_scale; 3807 tp->rcv_scale = sc->sc_request_r_scale; 3808 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3809 } 3810 if (sc->sc_flags & SCF_TIMESTAMP) 3811 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3812 tp->ts_timebase = sc->sc_timebase; 3813 3814 tp->t_template = tcp_template(tp); 3815 if (tp->t_template == 0) { 3816 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3817 so = NULL; 3818 m_freem(m); 3819 goto abort; 3820 } 3821 3822 tp->iss = sc->sc_iss; 3823 tp->irs = sc->sc_irs; 3824 tcp_sendseqinit(tp); 3825 tcp_rcvseqinit(tp); 3826 tp->t_state = TCPS_SYN_RECEIVED; 3827 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 3828 TCP_STATINC(TCP_STAT_ACCEPTS); 3829 3830 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3831 tp->t_flags |= TF_WILL_SACK; 3832 3833 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3834 tp->t_flags |= TF_ECN_PERMIT; 3835 3836 #ifdef TCP_SIGNATURE 3837 if (sc->sc_flags & SCF_SIGNATURE) 3838 tp->t_flags |= TF_SIGNATURE; 3839 #endif 3840 3841 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3842 tp->t_ourmss = sc->sc_ourmaxseg; 3843 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3844 3845 /* 3846 * Initialize the initial congestion window. If we 3847 * had to retransmit the SYN,ACK, we must initialize cwnd 3848 * to 1 segment (i.e. the Loss Window). 3849 */ 3850 if (sc->sc_rxtshift) 3851 tp->snd_cwnd = tp->t_peermss; 3852 else { 3853 int ss = tcp_init_win; 3854 #ifdef INET 3855 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3856 ss = tcp_init_win_local; 3857 #endif 3858 #ifdef INET6 3859 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3860 ss = tcp_init_win_local; 3861 #endif 3862 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3863 } 3864 3865 tcp_rmx_rtt(tp); 3866 tp->snd_wl1 = sc->sc_irs; 3867 tp->rcv_up = sc->sc_irs + 1; 3868 3869 /* 3870 * This is what whould have happened in tcp_output() when 3871 * the SYN,ACK was sent. 3872 */ 3873 tp->snd_up = tp->snd_una; 3874 tp->snd_max = tp->snd_nxt = tp->iss+1; 3875 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3876 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3877 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3878 tp->last_ack_sent = tp->rcv_nxt; 3879 tp->t_partialacks = -1; 3880 tp->t_dupacks = 0; 3881 3882 TCP_STATINC(TCP_STAT_SC_COMPLETED); 3883 s = splsoftnet(); 3884 syn_cache_put(sc); 3885 splx(s); 3886 return (so); 3887 3888 resetandabort: 3889 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3890 abort: 3891 if (so != NULL) 3892 (void) soabort(so); 3893 s = splsoftnet(); 3894 syn_cache_put(sc); 3895 splx(s); 3896 TCP_STATINC(TCP_STAT_SC_ABORTED); 3897 return ((struct socket *)(-1)); 3898 } 3899 3900 /* 3901 * This function is called when we get a RST for a 3902 * non-existent connection, so that we can see if the 3903 * connection is in the syn cache. If it is, zap it. 3904 */ 3905 3906 void 3907 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3908 { 3909 struct syn_cache *sc; 3910 struct syn_cache_head *scp; 3911 int s = splsoftnet(); 3912 3913 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3914 splx(s); 3915 return; 3916 } 3917 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3918 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3919 splx(s); 3920 return; 3921 } 3922 syn_cache_rm(sc); 3923 TCP_STATINC(TCP_STAT_SC_RESET); 3924 syn_cache_put(sc); /* calls pool_put but see spl above */ 3925 splx(s); 3926 } 3927 3928 void 3929 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3930 struct tcphdr *th) 3931 { 3932 struct syn_cache *sc; 3933 struct syn_cache_head *scp; 3934 int s; 3935 3936 s = splsoftnet(); 3937 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3938 splx(s); 3939 return; 3940 } 3941 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3942 if (ntohl (th->th_seq) != sc->sc_iss) { 3943 splx(s); 3944 return; 3945 } 3946 3947 /* 3948 * If we've retransmitted 3 times and this is our second error, 3949 * we remove the entry. Otherwise, we allow it to continue on. 3950 * This prevents us from incorrectly nuking an entry during a 3951 * spurious network outage. 3952 * 3953 * See tcp_notify(). 3954 */ 3955 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3956 sc->sc_flags |= SCF_UNREACH; 3957 splx(s); 3958 return; 3959 } 3960 3961 syn_cache_rm(sc); 3962 TCP_STATINC(TCP_STAT_SC_UNREACH); 3963 syn_cache_put(sc); /* calls pool_put but see spl above */ 3964 splx(s); 3965 } 3966 3967 /* 3968 * Given a LISTEN socket and an inbound SYN request, add 3969 * this to the syn cache, and send back a segment: 3970 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3971 * to the source. 3972 * 3973 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3974 * Doing so would require that we hold onto the data and deliver it 3975 * to the application. However, if we are the target of a SYN-flood 3976 * DoS attack, an attacker could send data which would eventually 3977 * consume all available buffer space if it were ACKed. By not ACKing 3978 * the data, we avoid this DoS scenario. 3979 */ 3980 3981 int 3982 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3983 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3984 int optlen, struct tcp_opt_info *oi) 3985 { 3986 struct tcpcb tb, *tp; 3987 long win; 3988 struct syn_cache *sc; 3989 struct syn_cache_head *scp; 3990 struct mbuf *ipopts; 3991 struct tcp_opt_info opti; 3992 int s; 3993 3994 tp = sototcpcb(so); 3995 3996 bzero(&opti, sizeof(opti)); 3997 3998 /* 3999 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4000 * 4001 * Note this check is performed in tcp_input() very early on. 4002 */ 4003 4004 /* 4005 * Initialize some local state. 4006 */ 4007 win = sbspace(&so->so_rcv); 4008 if (win > TCP_MAXWIN) 4009 win = TCP_MAXWIN; 4010 4011 switch (src->sa_family) { 4012 #ifdef INET 4013 case AF_INET: 4014 /* 4015 * Remember the IP options, if any. 4016 */ 4017 ipopts = ip_srcroute(); 4018 break; 4019 #endif 4020 default: 4021 ipopts = NULL; 4022 } 4023 4024 #ifdef TCP_SIGNATURE 4025 if (optp || (tp->t_flags & TF_SIGNATURE)) 4026 #else 4027 if (optp) 4028 #endif 4029 { 4030 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4031 #ifdef TCP_SIGNATURE 4032 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4033 #endif 4034 tb.t_state = TCPS_LISTEN; 4035 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4036 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4037 return (0); 4038 } else 4039 tb.t_flags = 0; 4040 4041 /* 4042 * See if we already have an entry for this connection. 4043 * If we do, resend the SYN,ACK. We do not count this 4044 * as a retransmission (XXX though maybe we should). 4045 */ 4046 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4047 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4048 if (ipopts) { 4049 /* 4050 * If we were remembering a previous source route, 4051 * forget it and use the new one we've been given. 4052 */ 4053 if (sc->sc_ipopts) 4054 (void) m_free(sc->sc_ipopts); 4055 sc->sc_ipopts = ipopts; 4056 } 4057 sc->sc_timestamp = tb.ts_recent; 4058 if (syn_cache_respond(sc, m) == 0) { 4059 uint64_t *tcps = TCP_STAT_GETREF(); 4060 tcps[TCP_STAT_SNDACKS]++; 4061 tcps[TCP_STAT_SNDTOTAL]++; 4062 TCP_STAT_PUTREF(); 4063 } 4064 return (1); 4065 } 4066 4067 s = splsoftnet(); 4068 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4069 splx(s); 4070 if (sc == NULL) { 4071 if (ipopts) 4072 (void) m_free(ipopts); 4073 return (0); 4074 } 4075 4076 /* 4077 * Fill in the cache, and put the necessary IP and TCP 4078 * options into the reply. 4079 */ 4080 bzero(sc, sizeof(struct syn_cache)); 4081 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4082 bcopy(src, &sc->sc_src, src->sa_len); 4083 bcopy(dst, &sc->sc_dst, dst->sa_len); 4084 sc->sc_flags = 0; 4085 sc->sc_ipopts = ipopts; 4086 sc->sc_irs = th->th_seq; 4087 switch (src->sa_family) { 4088 #ifdef INET 4089 case AF_INET: 4090 { 4091 struct sockaddr_in *srcin = (void *) src; 4092 struct sockaddr_in *dstin = (void *) dst; 4093 4094 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4095 &srcin->sin_addr, dstin->sin_port, 4096 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4097 break; 4098 } 4099 #endif /* INET */ 4100 #ifdef INET6 4101 case AF_INET6: 4102 { 4103 struct sockaddr_in6 *srcin6 = (void *) src; 4104 struct sockaddr_in6 *dstin6 = (void *) dst; 4105 4106 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4107 &srcin6->sin6_addr, dstin6->sin6_port, 4108 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4109 break; 4110 } 4111 #endif /* INET6 */ 4112 } 4113 sc->sc_peermaxseg = oi->maxseg; 4114 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4115 m->m_pkthdr.rcvif : NULL, 4116 sc->sc_src.sa.sa_family); 4117 sc->sc_win = win; 4118 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4119 sc->sc_timestamp = tb.ts_recent; 4120 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4121 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4122 sc->sc_flags |= SCF_TIMESTAMP; 4123 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4124 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4125 sc->sc_requested_s_scale = tb.requested_s_scale; 4126 sc->sc_request_r_scale = 0; 4127 /* 4128 * Pick the smallest possible scaling factor that 4129 * will still allow us to scale up to sb_max. 4130 * 4131 * We do this because there are broken firewalls that 4132 * will corrupt the window scale option, leading to 4133 * the other endpoint believing that our advertised 4134 * window is unscaled. At scale factors larger than 4135 * 5 the unscaled window will drop below 1500 bytes, 4136 * leading to serious problems when traversing these 4137 * broken firewalls. 4138 * 4139 * With the default sbmax of 256K, a scale factor 4140 * of 3 will be chosen by this algorithm. Those who 4141 * choose a larger sbmax should watch out 4142 * for the compatiblity problems mentioned above. 4143 * 4144 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4145 * or <SYN,ACK>) segment itself is never scaled. 4146 */ 4147 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4148 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4149 sc->sc_request_r_scale++; 4150 } else { 4151 sc->sc_requested_s_scale = 15; 4152 sc->sc_request_r_scale = 15; 4153 } 4154 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4155 sc->sc_flags |= SCF_SACK_PERMIT; 4156 4157 /* 4158 * ECN setup packet recieved. 4159 */ 4160 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4161 sc->sc_flags |= SCF_ECN_PERMIT; 4162 4163 #ifdef TCP_SIGNATURE 4164 if (tb.t_flags & TF_SIGNATURE) 4165 sc->sc_flags |= SCF_SIGNATURE; 4166 #endif 4167 sc->sc_tp = tp; 4168 if (syn_cache_respond(sc, m) == 0) { 4169 uint64_t *tcps = TCP_STAT_GETREF(); 4170 tcps[TCP_STAT_SNDACKS]++; 4171 tcps[TCP_STAT_SNDTOTAL]++; 4172 TCP_STAT_PUTREF(); 4173 syn_cache_insert(sc, tp); 4174 } else { 4175 s = splsoftnet(); 4176 syn_cache_put(sc); 4177 splx(s); 4178 TCP_STATINC(TCP_STAT_SC_DROPPED); 4179 } 4180 return (1); 4181 } 4182 4183 int 4184 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4185 { 4186 #ifdef INET6 4187 struct rtentry *rt; 4188 #endif 4189 struct route *ro; 4190 u_int8_t *optp; 4191 int optlen, error; 4192 u_int16_t tlen; 4193 struct ip *ip = NULL; 4194 #ifdef INET6 4195 struct ip6_hdr *ip6 = NULL; 4196 #endif 4197 struct tcpcb *tp = NULL; 4198 struct tcphdr *th; 4199 u_int hlen; 4200 struct socket *so; 4201 4202 ro = &sc->sc_route; 4203 switch (sc->sc_src.sa.sa_family) { 4204 case AF_INET: 4205 hlen = sizeof(struct ip); 4206 break; 4207 #ifdef INET6 4208 case AF_INET6: 4209 hlen = sizeof(struct ip6_hdr); 4210 break; 4211 #endif 4212 default: 4213 if (m) 4214 m_freem(m); 4215 return (EAFNOSUPPORT); 4216 } 4217 4218 /* Compute the size of the TCP options. */ 4219 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4220 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4221 #ifdef TCP_SIGNATURE 4222 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4223 #endif 4224 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4225 4226 tlen = hlen + sizeof(struct tcphdr) + optlen; 4227 4228 /* 4229 * Create the IP+TCP header from scratch. 4230 */ 4231 if (m) 4232 m_freem(m); 4233 #ifdef DIAGNOSTIC 4234 if (max_linkhdr + tlen > MCLBYTES) 4235 return (ENOBUFS); 4236 #endif 4237 MGETHDR(m, M_DONTWAIT, MT_DATA); 4238 if (m && tlen > MHLEN) { 4239 MCLGET(m, M_DONTWAIT); 4240 if ((m->m_flags & M_EXT) == 0) { 4241 m_freem(m); 4242 m = NULL; 4243 } 4244 } 4245 if (m == NULL) 4246 return (ENOBUFS); 4247 MCLAIM(m, &tcp_tx_mowner); 4248 4249 /* Fixup the mbuf. */ 4250 m->m_data += max_linkhdr; 4251 m->m_len = m->m_pkthdr.len = tlen; 4252 if (sc->sc_tp) { 4253 tp = sc->sc_tp; 4254 if (tp->t_inpcb) 4255 so = tp->t_inpcb->inp_socket; 4256 #ifdef INET6 4257 else if (tp->t_in6pcb) 4258 so = tp->t_in6pcb->in6p_socket; 4259 #endif 4260 else 4261 so = NULL; 4262 } else 4263 so = NULL; 4264 m->m_pkthdr.rcvif = NULL; 4265 memset(mtod(m, u_char *), 0, tlen); 4266 4267 switch (sc->sc_src.sa.sa_family) { 4268 case AF_INET: 4269 ip = mtod(m, struct ip *); 4270 ip->ip_v = 4; 4271 ip->ip_dst = sc->sc_src.sin.sin_addr; 4272 ip->ip_src = sc->sc_dst.sin.sin_addr; 4273 ip->ip_p = IPPROTO_TCP; 4274 th = (struct tcphdr *)(ip + 1); 4275 th->th_dport = sc->sc_src.sin.sin_port; 4276 th->th_sport = sc->sc_dst.sin.sin_port; 4277 break; 4278 #ifdef INET6 4279 case AF_INET6: 4280 ip6 = mtod(m, struct ip6_hdr *); 4281 ip6->ip6_vfc = IPV6_VERSION; 4282 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4283 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4284 ip6->ip6_nxt = IPPROTO_TCP; 4285 /* ip6_plen will be updated in ip6_output() */ 4286 th = (struct tcphdr *)(ip6 + 1); 4287 th->th_dport = sc->sc_src.sin6.sin6_port; 4288 th->th_sport = sc->sc_dst.sin6.sin6_port; 4289 break; 4290 #endif 4291 default: 4292 th = NULL; 4293 } 4294 4295 th->th_seq = htonl(sc->sc_iss); 4296 th->th_ack = htonl(sc->sc_irs + 1); 4297 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4298 th->th_flags = TH_SYN|TH_ACK; 4299 th->th_win = htons(sc->sc_win); 4300 /* th_sum already 0 */ 4301 /* th_urp already 0 */ 4302 4303 /* Tack on the TCP options. */ 4304 optp = (u_int8_t *)(th + 1); 4305 *optp++ = TCPOPT_MAXSEG; 4306 *optp++ = 4; 4307 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4308 *optp++ = sc->sc_ourmaxseg & 0xff; 4309 4310 if (sc->sc_request_r_scale != 15) { 4311 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4312 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4313 sc->sc_request_r_scale); 4314 optp += 4; 4315 } 4316 4317 if (sc->sc_flags & SCF_TIMESTAMP) { 4318 u_int32_t *lp = (u_int32_t *)(optp); 4319 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4320 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4321 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4322 *lp = htonl(sc->sc_timestamp); 4323 optp += TCPOLEN_TSTAMP_APPA; 4324 } 4325 4326 if (sc->sc_flags & SCF_SACK_PERMIT) { 4327 u_int8_t *p = optp; 4328 4329 /* Let the peer know that we will SACK. */ 4330 p[0] = TCPOPT_SACK_PERMITTED; 4331 p[1] = 2; 4332 p[2] = TCPOPT_NOP; 4333 p[3] = TCPOPT_NOP; 4334 optp += 4; 4335 } 4336 4337 /* 4338 * Send ECN SYN-ACK setup packet. 4339 * Routes can be asymetric, so, even if we receive a packet 4340 * with ECE and CWR set, we must not assume no one will block 4341 * the ECE packet we are about to send. 4342 */ 4343 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4344 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4345 th->th_flags |= TH_ECE; 4346 TCP_STATINC(TCP_STAT_ECN_SHS); 4347 4348 /* 4349 * draft-ietf-tcpm-ecnsyn-00.txt 4350 * 4351 * "[...] a TCP node MAY respond to an ECN-setup 4352 * SYN packet by setting ECT in the responding 4353 * ECN-setup SYN/ACK packet, indicating to routers 4354 * that the SYN/ACK packet is ECN-Capable. 4355 * This allows a congested router along the path 4356 * to mark the packet instead of dropping the 4357 * packet as an indication of congestion." 4358 * 4359 * "[...] There can be a great benefit in setting 4360 * an ECN-capable codepoint in SYN/ACK packets [...] 4361 * Congestion is most likely to occur in 4362 * the server-to-client direction. As a result, 4363 * setting an ECN-capable codepoint in SYN/ACK 4364 * packets can reduce the occurence of three-second 4365 * retransmit timeouts resulting from the drop 4366 * of SYN/ACK packets." 4367 * 4368 * Page 4 and 6, January 2006. 4369 */ 4370 4371 switch (sc->sc_src.sa.sa_family) { 4372 #ifdef INET 4373 case AF_INET: 4374 ip->ip_tos |= IPTOS_ECN_ECT0; 4375 break; 4376 #endif 4377 #ifdef INET6 4378 case AF_INET6: 4379 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4380 break; 4381 #endif 4382 } 4383 TCP_STATINC(TCP_STAT_ECN_ECT); 4384 } 4385 4386 #ifdef TCP_SIGNATURE 4387 if (sc->sc_flags & SCF_SIGNATURE) { 4388 struct secasvar *sav; 4389 u_int8_t *sigp; 4390 4391 sav = tcp_signature_getsav(m, th); 4392 4393 if (sav == NULL) { 4394 if (m) 4395 m_freem(m); 4396 return (EPERM); 4397 } 4398 4399 *optp++ = TCPOPT_SIGNATURE; 4400 *optp++ = TCPOLEN_SIGNATURE; 4401 sigp = optp; 4402 bzero(optp, TCP_SIGLEN); 4403 optp += TCP_SIGLEN; 4404 *optp++ = TCPOPT_NOP; 4405 *optp++ = TCPOPT_EOL; 4406 4407 (void)tcp_signature(m, th, hlen, sav, sigp); 4408 4409 key_sa_recordxfer(sav, m); 4410 #ifdef FAST_IPSEC 4411 KEY_FREESAV(&sav); 4412 #else 4413 key_freesav(sav); 4414 #endif 4415 } 4416 #endif 4417 4418 /* Compute the packet's checksum. */ 4419 switch (sc->sc_src.sa.sa_family) { 4420 case AF_INET: 4421 ip->ip_len = htons(tlen - hlen); 4422 th->th_sum = 0; 4423 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4424 break; 4425 #ifdef INET6 4426 case AF_INET6: 4427 ip6->ip6_plen = htons(tlen - hlen); 4428 th->th_sum = 0; 4429 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4430 break; 4431 #endif 4432 } 4433 4434 /* 4435 * Fill in some straggling IP bits. Note the stack expects 4436 * ip_len to be in host order, for convenience. 4437 */ 4438 switch (sc->sc_src.sa.sa_family) { 4439 #ifdef INET 4440 case AF_INET: 4441 ip->ip_len = htons(tlen); 4442 ip->ip_ttl = ip_defttl; 4443 /* XXX tos? */ 4444 break; 4445 #endif 4446 #ifdef INET6 4447 case AF_INET6: 4448 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4449 ip6->ip6_vfc |= IPV6_VERSION; 4450 ip6->ip6_plen = htons(tlen - hlen); 4451 /* ip6_hlim will be initialized afterwards */ 4452 /* XXX flowlabel? */ 4453 break; 4454 #endif 4455 } 4456 4457 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4458 tp = sc->sc_tp; 4459 4460 switch (sc->sc_src.sa.sa_family) { 4461 #ifdef INET 4462 case AF_INET: 4463 error = ip_output(m, sc->sc_ipopts, ro, 4464 (ip_mtudisc ? IP_MTUDISC : 0), 4465 (struct ip_moptions *)NULL, so); 4466 break; 4467 #endif 4468 #ifdef INET6 4469 case AF_INET6: 4470 ip6->ip6_hlim = in6_selecthlim(NULL, 4471 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4472 : NULL); 4473 4474 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4475 break; 4476 #endif 4477 default: 4478 error = EAFNOSUPPORT; 4479 break; 4480 } 4481 return (error); 4482 } 4483