1 /* $NetBSD: tcp_input.c,v 1.306 2010/12/02 19:07:27 plunky 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 1.306 2010/12/02 19:07:27 plunky 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 = 1; 243 int tcp_autorcvbuf_inc = 16 * 1024; 244 int tcp_autorcvbuf_max = 256 * 1024; 245 int tcp_msl = (TCPTV_MSL / PR_SLOWHZ); 246 247 static int tcp_rst_ppslim_count = 0; 248 static struct timeval tcp_rst_ppslim_last; 249 static int tcp_ackdrop_ppslim_count = 0; 250 static struct timeval tcp_ackdrop_ppslim_last; 251 252 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 253 254 /* for modulo comparisons of timestamps */ 255 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 256 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 257 258 /* 259 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 260 */ 261 #ifdef INET6 262 static inline void 263 nd6_hint(struct tcpcb *tp) 264 { 265 struct rtentry *rt; 266 267 if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 && 268 (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL) 269 nd6_nud_hint(rt, NULL, 0); 270 } 271 #else 272 static inline void 273 nd6_hint(struct tcpcb *tp) 274 { 275 } 276 #endif 277 278 /* 279 * Compute ACK transmission behavior. Delay the ACK unless 280 * we have already delayed an ACK (must send an ACK every two segments). 281 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 282 * option is enabled. 283 */ 284 static void 285 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th) 286 { 287 288 if (tp->t_flags & TF_DELACK || 289 (tcp_ack_on_push && th->th_flags & TH_PUSH)) 290 tp->t_flags |= TF_ACKNOW; 291 else 292 TCP_SET_DELACK(tp); 293 } 294 295 static void 296 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked) 297 { 298 299 /* 300 * If we had a pending ICMP message that refers to data that have 301 * just been acknowledged, disregard the recorded ICMP message. 302 */ 303 if ((tp->t_flags & TF_PMTUD_PEND) && 304 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 305 tp->t_flags &= ~TF_PMTUD_PEND; 306 307 /* 308 * Keep track of the largest chunk of data 309 * acknowledged since last PMTU update 310 */ 311 if (tp->t_pmtud_mss_acked < acked) 312 tp->t_pmtud_mss_acked = acked; 313 } 314 315 /* 316 * Convert TCP protocol fields to host order for easier processing. 317 */ 318 static void 319 tcp_fields_to_host(struct tcphdr *th) 320 { 321 322 NTOHL(th->th_seq); 323 NTOHL(th->th_ack); 324 NTOHS(th->th_win); 325 NTOHS(th->th_urp); 326 } 327 328 /* 329 * ... and reverse the above. 330 */ 331 static void 332 tcp_fields_to_net(struct tcphdr *th) 333 { 334 335 HTONL(th->th_seq); 336 HTONL(th->th_ack); 337 HTONS(th->th_win); 338 HTONS(th->th_urp); 339 } 340 341 #ifdef TCP_CSUM_COUNTERS 342 #include <sys/device.h> 343 344 #if defined(INET) 345 extern struct evcnt tcp_hwcsum_ok; 346 extern struct evcnt tcp_hwcsum_bad; 347 extern struct evcnt tcp_hwcsum_data; 348 extern struct evcnt tcp_swcsum; 349 #endif /* defined(INET) */ 350 #if defined(INET6) 351 extern struct evcnt tcp6_hwcsum_ok; 352 extern struct evcnt tcp6_hwcsum_bad; 353 extern struct evcnt tcp6_hwcsum_data; 354 extern struct evcnt tcp6_swcsum; 355 #endif /* defined(INET6) */ 356 357 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 358 359 #else 360 361 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 362 363 #endif /* TCP_CSUM_COUNTERS */ 364 365 #ifdef TCP_REASS_COUNTERS 366 #include <sys/device.h> 367 368 extern struct evcnt tcp_reass_; 369 extern struct evcnt tcp_reass_empty; 370 extern struct evcnt tcp_reass_iteration[8]; 371 extern struct evcnt tcp_reass_prependfirst; 372 extern struct evcnt tcp_reass_prepend; 373 extern struct evcnt tcp_reass_insert; 374 extern struct evcnt tcp_reass_inserttail; 375 extern struct evcnt tcp_reass_append; 376 extern struct evcnt tcp_reass_appendtail; 377 extern struct evcnt tcp_reass_overlaptail; 378 extern struct evcnt tcp_reass_overlapfront; 379 extern struct evcnt tcp_reass_segdup; 380 extern struct evcnt tcp_reass_fragdup; 381 382 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 383 384 #else 385 386 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 387 388 #endif /* TCP_REASS_COUNTERS */ 389 390 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *, 391 int *); 392 static int tcp_dooptions(struct tcpcb *, const u_char *, int, 393 struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *); 394 395 #ifdef INET 396 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 397 #endif 398 #ifdef INET6 399 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 400 #endif 401 402 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 403 404 #if defined(MBUFTRACE) 405 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass"); 406 #endif /* defined(MBUFTRACE) */ 407 408 static struct pool tcpipqent_pool; 409 410 void 411 tcpipqent_init(void) 412 { 413 414 pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", 415 NULL, IPL_VM); 416 } 417 418 struct ipqent * 419 tcpipqent_alloc(void) 420 { 421 struct ipqent *ipqe; 422 int s; 423 424 s = splvm(); 425 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 426 splx(s); 427 428 return ipqe; 429 } 430 431 void 432 tcpipqent_free(struct ipqent *ipqe) 433 { 434 int s; 435 436 s = splvm(); 437 pool_put(&tcpipqent_pool, ipqe); 438 splx(s); 439 } 440 441 static int 442 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen) 443 { 444 struct ipqent *p, *q, *nq, *tiqe = NULL; 445 struct socket *so = NULL; 446 int pkt_flags; 447 tcp_seq pkt_seq; 448 unsigned pkt_len; 449 u_long rcvpartdupbyte = 0; 450 u_long rcvoobyte; 451 #ifdef TCP_REASS_COUNTERS 452 u_int count = 0; 453 #endif 454 uint64_t *tcps; 455 456 if (tp->t_inpcb) 457 so = tp->t_inpcb->inp_socket; 458 #ifdef INET6 459 else if (tp->t_in6pcb) 460 so = tp->t_in6pcb->in6p_socket; 461 #endif 462 463 TCP_REASS_LOCK_CHECK(tp); 464 465 /* 466 * Call with th==0 after become established to 467 * force pre-ESTABLISHED data up to user socket. 468 */ 469 if (th == 0) 470 goto present; 471 472 m_claimm(m, &tcp_reass_mowner); 473 474 rcvoobyte = *tlen; 475 /* 476 * Copy these to local variables because the tcpiphdr 477 * gets munged while we are collapsing mbufs. 478 */ 479 pkt_seq = th->th_seq; 480 pkt_len = *tlen; 481 pkt_flags = th->th_flags; 482 483 TCP_REASS_COUNTER_INCR(&tcp_reass_); 484 485 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 486 /* 487 * When we miss a packet, the vast majority of time we get 488 * packets that follow it in order. So optimize for that. 489 */ 490 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 491 p->ipqe_len += pkt_len; 492 p->ipqe_flags |= pkt_flags; 493 m_cat(p->ipre_mlast, m); 494 TRAVERSE(p->ipre_mlast); 495 m = NULL; 496 tiqe = p; 497 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 498 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 499 goto skip_replacement; 500 } 501 /* 502 * While we're here, if the pkt is completely beyond 503 * anything we have, just insert it at the tail. 504 */ 505 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 506 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 507 goto insert_it; 508 } 509 } 510 511 q = TAILQ_FIRST(&tp->segq); 512 513 if (q != NULL) { 514 /* 515 * If this segment immediately precedes the first out-of-order 516 * block, simply slap the segment in front of it and (mostly) 517 * skip the complicated logic. 518 */ 519 if (pkt_seq + pkt_len == q->ipqe_seq) { 520 q->ipqe_seq = pkt_seq; 521 q->ipqe_len += pkt_len; 522 q->ipqe_flags |= pkt_flags; 523 m_cat(m, q->ipqe_m); 524 q->ipqe_m = m; 525 q->ipre_mlast = m; /* last mbuf may have changed */ 526 TRAVERSE(q->ipre_mlast); 527 tiqe = q; 528 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 529 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 530 goto skip_replacement; 531 } 532 } else { 533 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 534 } 535 536 /* 537 * Find a segment which begins after this one does. 538 */ 539 for (p = NULL; q != NULL; q = nq) { 540 nq = TAILQ_NEXT(q, ipqe_q); 541 #ifdef TCP_REASS_COUNTERS 542 count++; 543 #endif 544 /* 545 * If the received segment is just right after this 546 * fragment, merge the two together and then check 547 * for further overlaps. 548 */ 549 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 552 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 553 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 554 #endif 555 pkt_len += q->ipqe_len; 556 pkt_flags |= q->ipqe_flags; 557 pkt_seq = q->ipqe_seq; 558 m_cat(q->ipre_mlast, m); 559 TRAVERSE(q->ipre_mlast); 560 m = q->ipqe_m; 561 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 562 goto free_ipqe; 563 } 564 /* 565 * If the received segment is completely past this 566 * fragment, we need to go the next fragment. 567 */ 568 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 569 p = q; 570 continue; 571 } 572 /* 573 * If the fragment is past the received segment, 574 * it (or any following) can't be concatenated. 575 */ 576 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 577 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 578 break; 579 } 580 581 /* 582 * We've received all the data in this segment before. 583 * mark it as a duplicate and return. 584 */ 585 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 586 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 587 tcps = TCP_STAT_GETREF(); 588 tcps[TCP_STAT_RCVDUPPACK]++; 589 tcps[TCP_STAT_RCVDUPBYTE] += pkt_len; 590 TCP_STAT_PUTREF(); 591 tcp_new_dsack(tp, pkt_seq, pkt_len); 592 m_freem(m); 593 if (tiqe != NULL) { 594 tcpipqent_free(tiqe); 595 } 596 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 597 return (0); 598 } 599 /* 600 * Received segment completely overlaps this fragment 601 * so we drop the fragment (this keeps the temporal 602 * ordering of segments correct). 603 */ 604 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 605 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 606 rcvpartdupbyte += q->ipqe_len; 607 m_freem(q->ipqe_m); 608 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 609 goto free_ipqe; 610 } 611 /* 612 * RX'ed segment extends past the end of the 613 * fragment. Drop the overlapping bytes. Then 614 * merge the fragment and segment then treat as 615 * a longer received packet. 616 */ 617 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 618 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 619 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 620 #ifdef TCPREASS_DEBUG 621 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 622 tp, overlap, 623 pkt_seq, pkt_seq + pkt_len, pkt_len); 624 #endif 625 m_adj(m, overlap); 626 rcvpartdupbyte += overlap; 627 m_cat(q->ipre_mlast, m); 628 TRAVERSE(q->ipre_mlast); 629 m = q->ipqe_m; 630 pkt_seq = q->ipqe_seq; 631 pkt_len += q->ipqe_len - overlap; 632 rcvoobyte -= overlap; 633 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 634 goto free_ipqe; 635 } 636 /* 637 * RX'ed segment extends past the front of the 638 * fragment. Drop the overlapping bytes on the 639 * received packet. The packet will then be 640 * contatentated with this fragment a bit later. 641 */ 642 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 643 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 644 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 645 #ifdef TCPREASS_DEBUG 646 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 647 tp, overlap, 648 pkt_seq, pkt_seq + pkt_len, pkt_len); 649 #endif 650 m_adj(m, -overlap); 651 pkt_len -= overlap; 652 rcvpartdupbyte += overlap; 653 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 654 rcvoobyte -= overlap; 655 } 656 /* 657 * If the received segment immediates precedes this 658 * fragment then tack the fragment onto this segment 659 * and reinsert the data. 660 */ 661 if (q->ipqe_seq == pkt_seq + pkt_len) { 662 #ifdef TCPREASS_DEBUG 663 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 664 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 665 pkt_seq, pkt_seq + pkt_len, pkt_len); 666 #endif 667 pkt_len += q->ipqe_len; 668 pkt_flags |= q->ipqe_flags; 669 m_cat(m, q->ipqe_m); 670 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 671 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 672 tp->t_segqlen--; 673 KASSERT(tp->t_segqlen >= 0); 674 KASSERT(tp->t_segqlen != 0 || 675 (TAILQ_EMPTY(&tp->segq) && 676 TAILQ_EMPTY(&tp->timeq))); 677 if (tiqe == NULL) { 678 tiqe = q; 679 } else { 680 tcpipqent_free(q); 681 } 682 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 683 break; 684 } 685 /* 686 * If the fragment is before the segment, remember it. 687 * When this loop is terminated, p will contain the 688 * pointer to fragment that is right before the received 689 * segment. 690 */ 691 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 692 p = q; 693 694 continue; 695 696 /* 697 * This is a common operation. It also will allow 698 * to save doing a malloc/free in most instances. 699 */ 700 free_ipqe: 701 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 702 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 703 tp->t_segqlen--; 704 KASSERT(tp->t_segqlen >= 0); 705 KASSERT(tp->t_segqlen != 0 || 706 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 707 if (tiqe == NULL) { 708 tiqe = q; 709 } else { 710 tcpipqent_free(q); 711 } 712 } 713 714 #ifdef TCP_REASS_COUNTERS 715 if (count > 7) 716 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 717 else if (count > 0) 718 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 719 #endif 720 721 insert_it: 722 723 /* 724 * Allocate a new queue entry since the received segment did not 725 * collapse onto any other out-of-order block; thus we are allocating 726 * a new block. If it had collapsed, tiqe would not be NULL and 727 * we would be reusing it. 728 * XXX If we can't, just drop the packet. XXX 729 */ 730 if (tiqe == NULL) { 731 tiqe = tcpipqent_alloc(); 732 if (tiqe == NULL) { 733 TCP_STATINC(TCP_STAT_RCVMEMDROP); 734 m_freem(m); 735 return (0); 736 } 737 } 738 739 /* 740 * Update the counters. 741 */ 742 tcps = TCP_STAT_GETREF(); 743 tcps[TCP_STAT_RCVOOPACK]++; 744 tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte; 745 if (rcvpartdupbyte) { 746 tcps[TCP_STAT_RCVPARTDUPPACK]++; 747 tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte; 748 } 749 TCP_STAT_PUTREF(); 750 751 /* 752 * Insert the new fragment queue entry into both queues. 753 */ 754 tiqe->ipqe_m = m; 755 tiqe->ipre_mlast = m; 756 tiqe->ipqe_seq = pkt_seq; 757 tiqe->ipqe_len = pkt_len; 758 tiqe->ipqe_flags = pkt_flags; 759 if (p == NULL) { 760 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 761 #ifdef TCPREASS_DEBUG 762 if (tiqe->ipqe_seq != tp->rcv_nxt) 763 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 764 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 765 #endif 766 } else { 767 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 768 #ifdef TCPREASS_DEBUG 769 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 770 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 771 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 772 #endif 773 } 774 tp->t_segqlen++; 775 776 skip_replacement: 777 778 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 779 780 present: 781 /* 782 * Present data to user, advancing rcv_nxt through 783 * completed sequence space. 784 */ 785 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 786 return (0); 787 q = TAILQ_FIRST(&tp->segq); 788 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 789 return (0); 790 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 791 return (0); 792 793 tp->rcv_nxt += q->ipqe_len; 794 pkt_flags = q->ipqe_flags & TH_FIN; 795 nd6_hint(tp); 796 797 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 798 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 799 tp->t_segqlen--; 800 KASSERT(tp->t_segqlen >= 0); 801 KASSERT(tp->t_segqlen != 0 || 802 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 803 if (so->so_state & SS_CANTRCVMORE) 804 m_freem(q->ipqe_m); 805 else 806 sbappendstream(&so->so_rcv, q->ipqe_m); 807 tcpipqent_free(q); 808 sorwakeup(so); 809 return (pkt_flags); 810 } 811 812 #ifdef INET6 813 int 814 tcp6_input(struct mbuf **mp, int *offp, int proto) 815 { 816 struct mbuf *m = *mp; 817 818 /* 819 * draft-itojun-ipv6-tcp-to-anycast 820 * better place to put this in? 821 */ 822 if (m->m_flags & M_ANYCAST6) { 823 struct ip6_hdr *ip6; 824 if (m->m_len < sizeof(struct ip6_hdr)) { 825 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 826 TCP_STATINC(TCP_STAT_RCVSHORT); 827 return IPPROTO_DONE; 828 } 829 } 830 ip6 = mtod(m, struct ip6_hdr *); 831 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 832 (char *)&ip6->ip6_dst - (char *)ip6); 833 return IPPROTO_DONE; 834 } 835 836 tcp_input(m, *offp, proto); 837 return IPPROTO_DONE; 838 } 839 #endif 840 841 #ifdef INET 842 static void 843 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 844 { 845 char src[4*sizeof "123"]; 846 char dst[4*sizeof "123"]; 847 848 if (ip) { 849 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 850 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 851 } 852 else { 853 strlcpy(src, "(unknown)", sizeof(src)); 854 strlcpy(dst, "(unknown)", sizeof(dst)); 855 } 856 log(LOG_INFO, 857 "Connection attempt to TCP %s:%d from %s:%d\n", 858 dst, ntohs(th->th_dport), 859 src, ntohs(th->th_sport)); 860 } 861 #endif 862 863 #ifdef INET6 864 static void 865 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 866 { 867 char src[INET6_ADDRSTRLEN]; 868 char dst[INET6_ADDRSTRLEN]; 869 870 if (ip6) { 871 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 872 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 873 } 874 else { 875 strlcpy(src, "(unknown v6)", sizeof(src)); 876 strlcpy(dst, "(unknown v6)", sizeof(dst)); 877 } 878 log(LOG_INFO, 879 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 880 dst, ntohs(th->th_dport), 881 src, ntohs(th->th_sport)); 882 } 883 #endif 884 885 /* 886 * Checksum extended TCP header and data. 887 */ 888 int 889 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, 890 int toff, int off, int tlen) 891 { 892 893 /* 894 * XXX it's better to record and check if this mbuf is 895 * already checked. 896 */ 897 898 switch (af) { 899 #ifdef INET 900 case AF_INET: 901 switch (m->m_pkthdr.csum_flags & 902 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 903 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 904 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 905 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 906 goto badcsum; 907 908 case M_CSUM_TCPv4|M_CSUM_DATA: { 909 u_int32_t hw_csum = m->m_pkthdr.csum_data; 910 911 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 912 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 913 const struct ip *ip = 914 mtod(m, const struct ip *); 915 916 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 917 ip->ip_dst.s_addr, 918 htons(hw_csum + tlen + off + IPPROTO_TCP)); 919 } 920 if ((hw_csum ^ 0xffff) != 0) 921 goto badcsum; 922 break; 923 } 924 925 case M_CSUM_TCPv4: 926 /* Checksum was okay. */ 927 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 928 break; 929 930 default: 931 /* 932 * Must compute it ourselves. Maybe skip checksum 933 * on loopback interfaces. 934 */ 935 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 936 IFF_LOOPBACK) || 937 tcp_do_loopback_cksum)) { 938 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 939 if (in4_cksum(m, IPPROTO_TCP, toff, 940 tlen + off) != 0) 941 goto badcsum; 942 } 943 break; 944 } 945 break; 946 #endif /* INET4 */ 947 948 #ifdef INET6 949 case AF_INET6: 950 switch (m->m_pkthdr.csum_flags & 951 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) | 952 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 953 case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD: 954 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad); 955 goto badcsum; 956 957 #if 0 /* notyet */ 958 case M_CSUM_TCPv6|M_CSUM_DATA: 959 #endif 960 961 case M_CSUM_TCPv6: 962 /* Checksum was okay. */ 963 TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok); 964 break; 965 966 default: 967 /* 968 * Must compute it ourselves. Maybe skip checksum 969 * on loopback interfaces. 970 */ 971 if (__predict_true((m->m_flags & M_LOOP) == 0 || 972 tcp_do_loopback_cksum)) { 973 TCP_CSUM_COUNTER_INCR(&tcp6_swcsum); 974 if (in6_cksum(m, IPPROTO_TCP, toff, 975 tlen + off) != 0) 976 goto badcsum; 977 } 978 } 979 break; 980 #endif /* INET6 */ 981 } 982 983 return 0; 984 985 badcsum: 986 TCP_STATINC(TCP_STAT_RCVBADSUM); 987 return -1; 988 } 989 990 /* 991 * TCP input routine, follows pages 65-76 of RFC 793 very closely. 992 */ 993 void 994 tcp_input(struct mbuf *m, ...) 995 { 996 struct tcphdr *th; 997 struct ip *ip; 998 struct inpcb *inp; 999 #ifdef INET6 1000 struct ip6_hdr *ip6; 1001 struct in6pcb *in6p; 1002 #endif 1003 u_int8_t *optp = NULL; 1004 int optlen = 0; 1005 int len, tlen, toff, hdroptlen = 0; 1006 struct tcpcb *tp = 0; 1007 int tiflags; 1008 struct socket *so = NULL; 1009 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 1010 #ifdef TCP_DEBUG 1011 short ostate = 0; 1012 #endif 1013 u_long tiwin; 1014 struct tcp_opt_info opti; 1015 int off, iphlen; 1016 va_list ap; 1017 int af; /* af on the wire */ 1018 struct mbuf *tcp_saveti = NULL; 1019 uint32_t ts_rtt; 1020 uint8_t iptos; 1021 uint64_t *tcps; 1022 1023 MCLAIM(m, &tcp_rx_mowner); 1024 va_start(ap, m); 1025 toff = va_arg(ap, int); 1026 (void)va_arg(ap, int); /* ignore value, advance ap */ 1027 va_end(ap); 1028 1029 TCP_STATINC(TCP_STAT_RCVTOTAL); 1030 1031 memset(&opti, 0, sizeof(opti)); 1032 opti.ts_present = 0; 1033 opti.maxseg = 0; 1034 1035 /* 1036 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 1037 * 1038 * TCP is, by definition, unicast, so we reject all 1039 * multicast outright. 1040 * 1041 * Note, there are additional src/dst address checks in 1042 * the AF-specific code below. 1043 */ 1044 if (m->m_flags & (M_BCAST|M_MCAST)) { 1045 /* XXX stat */ 1046 goto drop; 1047 } 1048 #ifdef INET6 1049 if (m->m_flags & M_ANYCAST6) { 1050 /* XXX stat */ 1051 goto drop; 1052 } 1053 #endif 1054 1055 /* 1056 * Get IP and TCP header. 1057 * Note: IP leaves IP header in first mbuf. 1058 */ 1059 ip = mtod(m, struct ip *); 1060 #ifdef INET6 1061 ip6 = NULL; 1062 #endif 1063 switch (ip->ip_v) { 1064 #ifdef INET 1065 case 4: 1066 af = AF_INET; 1067 iphlen = sizeof(struct ip); 1068 ip = mtod(m, struct ip *); 1069 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1070 sizeof(struct tcphdr)); 1071 if (th == NULL) { 1072 TCP_STATINC(TCP_STAT_RCVSHORT); 1073 return; 1074 } 1075 /* We do the checksum after PCB lookup... */ 1076 len = ntohs(ip->ip_len); 1077 tlen = len - toff; 1078 iptos = ip->ip_tos; 1079 break; 1080 #endif 1081 #ifdef INET6 1082 case 6: 1083 ip = NULL; 1084 iphlen = sizeof(struct ip6_hdr); 1085 af = AF_INET6; 1086 ip6 = mtod(m, struct ip6_hdr *); 1087 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 1088 sizeof(struct tcphdr)); 1089 if (th == NULL) { 1090 TCP_STATINC(TCP_STAT_RCVSHORT); 1091 return; 1092 } 1093 1094 /* Be proactive about malicious use of IPv4 mapped address */ 1095 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 1096 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1097 /* XXX stat */ 1098 goto drop; 1099 } 1100 1101 /* 1102 * Be proactive about unspecified IPv6 address in source. 1103 * As we use all-zero to indicate unbounded/unconnected pcb, 1104 * unspecified IPv6 address can be used to confuse us. 1105 * 1106 * Note that packets with unspecified IPv6 destination is 1107 * already dropped in ip6_input. 1108 */ 1109 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1110 /* XXX stat */ 1111 goto drop; 1112 } 1113 1114 /* 1115 * Make sure destination address is not multicast. 1116 * Source address checked in ip6_input(). 1117 */ 1118 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1119 /* XXX stat */ 1120 goto drop; 1121 } 1122 1123 /* We do the checksum after PCB lookup... */ 1124 len = m->m_pkthdr.len; 1125 tlen = len - toff; 1126 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1127 break; 1128 #endif 1129 default: 1130 m_freem(m); 1131 return; 1132 } 1133 1134 KASSERT(TCP_HDR_ALIGNED_P(th)); 1135 1136 /* 1137 * Check that TCP offset makes sense, 1138 * pull out TCP options and adjust length. XXX 1139 */ 1140 off = th->th_off << 2; 1141 if (off < sizeof (struct tcphdr) || off > tlen) { 1142 TCP_STATINC(TCP_STAT_RCVBADOFF); 1143 goto drop; 1144 } 1145 tlen -= off; 1146 1147 /* 1148 * tcp_input() has been modified to use tlen to mean the TCP data 1149 * length throughout the function. Other functions can use 1150 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1151 * rja 1152 */ 1153 1154 if (off > sizeof (struct tcphdr)) { 1155 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1156 if (th == NULL) { 1157 TCP_STATINC(TCP_STAT_RCVSHORT); 1158 return; 1159 } 1160 /* 1161 * NOTE: ip/ip6 will not be affected by m_pulldown() 1162 * (as they're before toff) and we don't need to update those. 1163 */ 1164 KASSERT(TCP_HDR_ALIGNED_P(th)); 1165 optlen = off - sizeof (struct tcphdr); 1166 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1167 /* 1168 * Do quick retrieval of timestamp options ("options 1169 * prediction?"). If timestamp is the only option and it's 1170 * formatted as recommended in RFC 1323 appendix A, we 1171 * quickly get the values now and not bother calling 1172 * tcp_dooptions(), etc. 1173 */ 1174 if ((optlen == TCPOLEN_TSTAMP_APPA || 1175 (optlen > TCPOLEN_TSTAMP_APPA && 1176 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1177 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1178 (th->th_flags & TH_SYN) == 0) { 1179 opti.ts_present = 1; 1180 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1181 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1182 optp = NULL; /* we've parsed the options */ 1183 } 1184 } 1185 tiflags = th->th_flags; 1186 1187 /* 1188 * Locate pcb for segment. 1189 */ 1190 findpcb: 1191 inp = NULL; 1192 #ifdef INET6 1193 in6p = NULL; 1194 #endif 1195 switch (af) { 1196 #ifdef INET 1197 case AF_INET: 1198 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1199 ip->ip_dst, th->th_dport); 1200 if (inp == 0) { 1201 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1202 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1203 } 1204 #ifdef INET6 1205 if (inp == 0) { 1206 struct in6_addr s, d; 1207 1208 /* mapped addr case */ 1209 memset(&s, 0, sizeof(s)); 1210 s.s6_addr16[5] = htons(0xffff); 1211 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1212 memset(&d, 0, sizeof(d)); 1213 d.s6_addr16[5] = htons(0xffff); 1214 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1215 in6p = in6_pcblookup_connect(&tcbtable, &s, 1216 th->th_sport, &d, th->th_dport, 0); 1217 if (in6p == 0) { 1218 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1219 in6p = in6_pcblookup_bind(&tcbtable, &d, 1220 th->th_dport, 0); 1221 } 1222 } 1223 #endif 1224 #ifndef INET6 1225 if (inp == 0) 1226 #else 1227 if (inp == 0 && in6p == 0) 1228 #endif 1229 { 1230 TCP_STATINC(TCP_STAT_NOPORT); 1231 if (tcp_log_refused && 1232 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1233 tcp4_log_refused(ip, th); 1234 } 1235 tcp_fields_to_host(th); 1236 goto dropwithreset_ratelim; 1237 } 1238 #if defined(IPSEC) || defined(FAST_IPSEC) 1239 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1240 ipsec4_in_reject(m, inp)) { 1241 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1242 goto drop; 1243 } 1244 #ifdef INET6 1245 else if (in6p && 1246 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1247 ipsec6_in_reject_so(m, in6p->in6p_socket)) { 1248 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1249 goto drop; 1250 } 1251 #endif 1252 #endif /*IPSEC*/ 1253 break; 1254 #endif /*INET*/ 1255 #ifdef INET6 1256 case AF_INET6: 1257 { 1258 int faith; 1259 1260 #if defined(NFAITH) && NFAITH > 0 1261 faith = faithprefix(&ip6->ip6_dst); 1262 #else 1263 faith = 0; 1264 #endif 1265 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1266 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1267 if (in6p == NULL) { 1268 TCP_STATINC(TCP_STAT_PCBHASHMISS); 1269 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1270 th->th_dport, faith); 1271 } 1272 if (in6p == NULL) { 1273 TCP_STATINC(TCP_STAT_NOPORT); 1274 if (tcp_log_refused && 1275 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1276 tcp6_log_refused(ip6, th); 1277 } 1278 tcp_fields_to_host(th); 1279 goto dropwithreset_ratelim; 1280 } 1281 #if defined(IPSEC) || defined(FAST_IPSEC) 1282 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1283 ipsec6_in_reject(m, in6p)) { 1284 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1285 goto drop; 1286 } 1287 #endif /*IPSEC*/ 1288 break; 1289 } 1290 #endif 1291 } 1292 1293 /* 1294 * If the state is CLOSED (i.e., TCB does not exist) then 1295 * all data in the incoming segment is discarded. 1296 * If the TCB exists but is in CLOSED state, it is embryonic, 1297 * but should either do a listen or a connect soon. 1298 */ 1299 tp = NULL; 1300 so = NULL; 1301 if (inp) { 1302 /* Check the minimum TTL for socket. */ 1303 if (ip->ip_ttl < inp->inp_ip_minttl) 1304 goto drop; 1305 1306 tp = intotcpcb(inp); 1307 so = inp->inp_socket; 1308 } 1309 #ifdef INET6 1310 else if (in6p) { 1311 tp = in6totcpcb(in6p); 1312 so = in6p->in6p_socket; 1313 } 1314 #endif 1315 if (tp == 0) { 1316 tcp_fields_to_host(th); 1317 goto dropwithreset_ratelim; 1318 } 1319 if (tp->t_state == TCPS_CLOSED) 1320 goto drop; 1321 1322 KASSERT(so->so_lock == softnet_lock); 1323 KASSERT(solocked(so)); 1324 1325 /* 1326 * Checksum extended TCP header and data. 1327 */ 1328 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1329 goto badcsum; 1330 1331 tcp_fields_to_host(th); 1332 1333 /* Unscale the window into a 32-bit value. */ 1334 if ((tiflags & TH_SYN) == 0) 1335 tiwin = th->th_win << tp->snd_scale; 1336 else 1337 tiwin = th->th_win; 1338 1339 #ifdef INET6 1340 /* save packet options if user wanted */ 1341 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1342 if (in6p->in6p_options) { 1343 m_freem(in6p->in6p_options); 1344 in6p->in6p_options = 0; 1345 } 1346 KASSERT(ip6 != NULL); 1347 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1348 } 1349 #endif 1350 1351 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1352 union syn_cache_sa src; 1353 union syn_cache_sa dst; 1354 1355 memset(&src, 0, sizeof(src)); 1356 memset(&dst, 0, sizeof(dst)); 1357 switch (af) { 1358 #ifdef INET 1359 case AF_INET: 1360 src.sin.sin_len = sizeof(struct sockaddr_in); 1361 src.sin.sin_family = AF_INET; 1362 src.sin.sin_addr = ip->ip_src; 1363 src.sin.sin_port = th->th_sport; 1364 1365 dst.sin.sin_len = sizeof(struct sockaddr_in); 1366 dst.sin.sin_family = AF_INET; 1367 dst.sin.sin_addr = ip->ip_dst; 1368 dst.sin.sin_port = th->th_dport; 1369 break; 1370 #endif 1371 #ifdef INET6 1372 case AF_INET6: 1373 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1374 src.sin6.sin6_family = AF_INET6; 1375 src.sin6.sin6_addr = ip6->ip6_src; 1376 src.sin6.sin6_port = th->th_sport; 1377 1378 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1379 dst.sin6.sin6_family = AF_INET6; 1380 dst.sin6.sin6_addr = ip6->ip6_dst; 1381 dst.sin6.sin6_port = th->th_dport; 1382 break; 1383 #endif /* INET6 */ 1384 default: 1385 goto badsyn; /*sanity*/ 1386 } 1387 1388 if (so->so_options & SO_DEBUG) { 1389 #ifdef TCP_DEBUG 1390 ostate = tp->t_state; 1391 #endif 1392 1393 tcp_saveti = NULL; 1394 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1395 goto nosave; 1396 1397 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1398 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1399 if (!tcp_saveti) 1400 goto nosave; 1401 } else { 1402 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1403 if (!tcp_saveti) 1404 goto nosave; 1405 MCLAIM(m, &tcp_mowner); 1406 tcp_saveti->m_len = iphlen; 1407 m_copydata(m, 0, iphlen, 1408 mtod(tcp_saveti, void *)); 1409 } 1410 1411 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1412 m_freem(tcp_saveti); 1413 tcp_saveti = NULL; 1414 } else { 1415 tcp_saveti->m_len += sizeof(struct tcphdr); 1416 memcpy(mtod(tcp_saveti, char *) + iphlen, th, 1417 sizeof(struct tcphdr)); 1418 } 1419 nosave:; 1420 } 1421 if (so->so_options & SO_ACCEPTCONN) { 1422 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1423 if (tiflags & TH_RST) { 1424 syn_cache_reset(&src.sa, &dst.sa, th); 1425 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1426 (TH_ACK|TH_SYN)) { 1427 /* 1428 * Received a SYN,ACK. This should 1429 * never happen while we are in 1430 * LISTEN. Send an RST. 1431 */ 1432 goto badsyn; 1433 } else if (tiflags & TH_ACK) { 1434 so = syn_cache_get(&src.sa, &dst.sa, 1435 th, toff, tlen, so, m); 1436 if (so == NULL) { 1437 /* 1438 * We don't have a SYN for 1439 * this ACK; send an RST. 1440 */ 1441 goto badsyn; 1442 } else if (so == 1443 (struct socket *)(-1)) { 1444 /* 1445 * We were unable to create 1446 * the connection. If the 1447 * 3-way handshake was 1448 * completed, and RST has 1449 * been sent to the peer. 1450 * Since the mbuf might be 1451 * in use for the reply, 1452 * do not free it. 1453 */ 1454 m = NULL; 1455 } else { 1456 /* 1457 * We have created a 1458 * full-blown connection. 1459 */ 1460 tp = NULL; 1461 inp = NULL; 1462 #ifdef INET6 1463 in6p = NULL; 1464 #endif 1465 switch (so->so_proto->pr_domain->dom_family) { 1466 #ifdef INET 1467 case AF_INET: 1468 inp = sotoinpcb(so); 1469 tp = intotcpcb(inp); 1470 break; 1471 #endif 1472 #ifdef INET6 1473 case AF_INET6: 1474 in6p = sotoin6pcb(so); 1475 tp = in6totcpcb(in6p); 1476 break; 1477 #endif 1478 } 1479 if (tp == NULL) 1480 goto badsyn; /*XXX*/ 1481 tiwin <<= tp->snd_scale; 1482 goto after_listen; 1483 } 1484 } else { 1485 /* 1486 * None of RST, SYN or ACK was set. 1487 * This is an invalid packet for a 1488 * TCB in LISTEN state. Send a RST. 1489 */ 1490 goto badsyn; 1491 } 1492 } else { 1493 /* 1494 * Received a SYN. 1495 * 1496 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1497 */ 1498 if (m->m_flags & (M_BCAST|M_MCAST)) 1499 goto drop; 1500 1501 switch (af) { 1502 #ifdef INET6 1503 case AF_INET6: 1504 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1505 goto drop; 1506 break; 1507 #endif /* INET6 */ 1508 case AF_INET: 1509 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1510 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1511 goto drop; 1512 break; 1513 } 1514 1515 #ifdef INET6 1516 /* 1517 * If deprecated address is forbidden, we do 1518 * not accept SYN to deprecated interface 1519 * address to prevent any new inbound 1520 * connection from getting established. 1521 * When we do not accept SYN, we send a TCP 1522 * RST, with deprecated source address (instead 1523 * of dropping it). We compromise it as it is 1524 * much better for peer to send a RST, and 1525 * RST will be the final packet for the 1526 * exchange. 1527 * 1528 * If we do not forbid deprecated addresses, we 1529 * accept the SYN packet. RFC2462 does not 1530 * suggest dropping SYN in this case. 1531 * If we decipher RFC2462 5.5.4, it says like 1532 * this: 1533 * 1. use of deprecated addr with existing 1534 * communication is okay - "SHOULD continue 1535 * to be used" 1536 * 2. use of it with new communication: 1537 * (2a) "SHOULD NOT be used if alternate 1538 * address with sufficient scope is 1539 * available" 1540 * (2b) nothing mentioned otherwise. 1541 * Here we fall into (2b) case as we have no 1542 * choice in our source address selection - we 1543 * must obey the peer. 1544 * 1545 * The wording in RFC2462 is confusing, and 1546 * there are multiple description text for 1547 * deprecated address handling - worse, they 1548 * are not exactly the same. I believe 5.5.4 1549 * is the best one, so we follow 5.5.4. 1550 */ 1551 if (af == AF_INET6 && !ip6_use_deprecated) { 1552 struct in6_ifaddr *ia6; 1553 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1554 &ip6->ip6_dst)) && 1555 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1556 tp = NULL; 1557 goto dropwithreset; 1558 } 1559 } 1560 #endif 1561 1562 #if defined(IPSEC) || defined(FAST_IPSEC) 1563 switch (af) { 1564 #ifdef INET 1565 case AF_INET: 1566 if (ipsec4_in_reject_so(m, so)) { 1567 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1568 tp = NULL; 1569 goto dropwithreset; 1570 } 1571 break; 1572 #endif 1573 #ifdef INET6 1574 case AF_INET6: 1575 if (ipsec6_in_reject_so(m, so)) { 1576 IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO); 1577 tp = NULL; 1578 goto dropwithreset; 1579 } 1580 break; 1581 #endif /*INET6*/ 1582 } 1583 #endif /*IPSEC*/ 1584 1585 /* 1586 * LISTEN socket received a SYN 1587 * from itself? This can't possibly 1588 * be valid; drop the packet. 1589 */ 1590 if (th->th_sport == th->th_dport) { 1591 int i; 1592 1593 switch (af) { 1594 #ifdef INET 1595 case AF_INET: 1596 i = in_hosteq(ip->ip_src, ip->ip_dst); 1597 break; 1598 #endif 1599 #ifdef INET6 1600 case AF_INET6: 1601 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1602 break; 1603 #endif 1604 default: 1605 i = 1; 1606 } 1607 if (i) { 1608 TCP_STATINC(TCP_STAT_BADSYN); 1609 goto drop; 1610 } 1611 } 1612 1613 /* 1614 * SYN looks ok; create compressed TCP 1615 * state for it. 1616 */ 1617 if (so->so_qlen <= so->so_qlimit && 1618 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1619 so, m, optp, optlen, &opti)) 1620 m = NULL; 1621 } 1622 goto drop; 1623 } 1624 } 1625 1626 after_listen: 1627 #ifdef DIAGNOSTIC 1628 /* 1629 * Should not happen now that all embryonic connections 1630 * are handled with compressed state. 1631 */ 1632 if (tp->t_state == TCPS_LISTEN) 1633 panic("tcp_input: TCPS_LISTEN"); 1634 #endif 1635 1636 /* 1637 * Segment received on connection. 1638 * Reset idle time and keep-alive timer. 1639 */ 1640 tp->t_rcvtime = tcp_now; 1641 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1642 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle); 1643 1644 /* 1645 * Process options. 1646 */ 1647 #ifdef TCP_SIGNATURE 1648 if (optp || (tp->t_flags & TF_SIGNATURE)) 1649 #else 1650 if (optp) 1651 #endif 1652 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1653 goto drop; 1654 1655 if (TCP_SACK_ENABLED(tp)) { 1656 tcp_del_sackholes(tp, th); 1657 } 1658 1659 if (TCP_ECN_ALLOWED(tp)) { 1660 if (tiflags & TH_CWR) { 1661 tp->t_flags &= ~TF_ECN_SND_ECE; 1662 } 1663 switch (iptos & IPTOS_ECN_MASK) { 1664 case IPTOS_ECN_CE: 1665 tp->t_flags |= TF_ECN_SND_ECE; 1666 TCP_STATINC(TCP_STAT_ECN_CE); 1667 break; 1668 case IPTOS_ECN_ECT0: 1669 TCP_STATINC(TCP_STAT_ECN_ECT); 1670 break; 1671 case IPTOS_ECN_ECT1: 1672 /* XXX: ignore for now -- rpaulo */ 1673 break; 1674 } 1675 /* 1676 * Congestion experienced. 1677 * Ignore if we are already trying to recover. 1678 */ 1679 if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover)) 1680 tp->t_congctl->cong_exp(tp); 1681 } 1682 1683 if (opti.ts_present && opti.ts_ecr) { 1684 /* 1685 * Calculate the RTT from the returned time stamp and the 1686 * connection's time base. If the time stamp is later than 1687 * the current time, or is extremely old, fall back to non-1323 1688 * RTT calculation. Since ts_ecr is unsigned, we can test both 1689 * at the same time. 1690 */ 1691 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1692 if (ts_rtt > TCP_PAWS_IDLE) 1693 ts_rtt = 0; 1694 } else { 1695 ts_rtt = 0; 1696 } 1697 1698 /* 1699 * Header prediction: check for the two common cases 1700 * of a uni-directional data xfer. If the packet has 1701 * no control flags, is in-sequence, the window didn't 1702 * change and we're not retransmitting, it's a 1703 * candidate. If the length is zero and the ack moved 1704 * forward, we're the sender side of the xfer. Just 1705 * free the data acked & wake any higher level process 1706 * that was blocked waiting for space. If the length 1707 * is non-zero and the ack didn't move, we're the 1708 * receiver side. If we're getting packets in-order 1709 * (the reassembly queue is empty), add the data to 1710 * the socket buffer and note that we need a delayed ack. 1711 */ 1712 if (tp->t_state == TCPS_ESTABLISHED && 1713 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) 1714 == TH_ACK && 1715 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1716 th->th_seq == tp->rcv_nxt && 1717 tiwin && tiwin == tp->snd_wnd && 1718 tp->snd_nxt == tp->snd_max) { 1719 1720 /* 1721 * If last ACK falls within this segment's sequence numbers, 1722 * record the timestamp. 1723 * NOTE that the test is modified according to the latest 1724 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1725 * 1726 * note that we already know 1727 * TSTMP_GEQ(opti.ts_val, tp->ts_recent) 1728 */ 1729 if (opti.ts_present && 1730 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1731 tp->ts_recent_age = tcp_now; 1732 tp->ts_recent = opti.ts_val; 1733 } 1734 1735 if (tlen == 0) { 1736 /* Ack prediction. */ 1737 if (SEQ_GT(th->th_ack, tp->snd_una) && 1738 SEQ_LEQ(th->th_ack, tp->snd_max) && 1739 tp->snd_cwnd >= tp->snd_wnd && 1740 tp->t_partialacks < 0) { 1741 /* 1742 * this is a pure ack for outstanding data. 1743 */ 1744 if (ts_rtt) 1745 tcp_xmit_timer(tp, ts_rtt); 1746 else if (tp->t_rtttime && 1747 SEQ_GT(th->th_ack, tp->t_rtseq)) 1748 tcp_xmit_timer(tp, 1749 tcp_now - tp->t_rtttime); 1750 acked = th->th_ack - tp->snd_una; 1751 tcps = TCP_STAT_GETREF(); 1752 tcps[TCP_STAT_PREDACK]++; 1753 tcps[TCP_STAT_RCVACKPACK]++; 1754 tcps[TCP_STAT_RCVACKBYTE] += acked; 1755 TCP_STAT_PUTREF(); 1756 nd6_hint(tp); 1757 1758 if (acked > (tp->t_lastoff - tp->t_inoff)) 1759 tp->t_lastm = NULL; 1760 sbdrop(&so->so_snd, acked); 1761 tp->t_lastoff -= acked; 1762 1763 icmp_check(tp, th, acked); 1764 1765 tp->snd_una = th->th_ack; 1766 tp->snd_fack = tp->snd_una; 1767 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1768 tp->snd_high = tp->snd_una; 1769 m_freem(m); 1770 1771 /* 1772 * If all outstanding data are acked, stop 1773 * retransmit timer, otherwise restart timer 1774 * using current (possibly backed-off) value. 1775 * If process is waiting for space, 1776 * wakeup/selnotify/signal. If data 1777 * are ready to send, let tcp_output 1778 * decide between more output or persist. 1779 */ 1780 if (tp->snd_una == tp->snd_max) 1781 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1782 else if (TCP_TIMER_ISARMED(tp, 1783 TCPT_PERSIST) == 0) 1784 TCP_TIMER_ARM(tp, TCPT_REXMT, 1785 tp->t_rxtcur); 1786 1787 sowwakeup(so); 1788 if (so->so_snd.sb_cc) { 1789 KERNEL_LOCK(1, NULL); 1790 (void) tcp_output(tp); 1791 KERNEL_UNLOCK_ONE(NULL); 1792 } 1793 if (tcp_saveti) 1794 m_freem(tcp_saveti); 1795 return; 1796 } 1797 } else if (th->th_ack == tp->snd_una && 1798 TAILQ_FIRST(&tp->segq) == NULL && 1799 tlen <= sbspace(&so->so_rcv)) { 1800 int newsize = 0; /* automatic sockbuf scaling */ 1801 1802 /* 1803 * this is a pure, in-sequence data packet 1804 * with nothing on the reassembly queue and 1805 * we have enough buffer space to take it. 1806 */ 1807 tp->rcv_nxt += tlen; 1808 tcps = TCP_STAT_GETREF(); 1809 tcps[TCP_STAT_PREDDAT]++; 1810 tcps[TCP_STAT_RCVPACK]++; 1811 tcps[TCP_STAT_RCVBYTE] += tlen; 1812 TCP_STAT_PUTREF(); 1813 nd6_hint(tp); 1814 1815 /* 1816 * Automatic sizing enables the performance of large buffers 1817 * and most of the efficiency of small ones by only allocating 1818 * space when it is needed. 1819 * 1820 * On the receive side the socket buffer memory is only rarely 1821 * used to any significant extent. This allows us to be much 1822 * more aggressive in scaling the receive socket buffer. For 1823 * the case that the buffer space is actually used to a large 1824 * extent and we run out of kernel memory we can simply drop 1825 * the new segments; TCP on the sender will just retransmit it 1826 * later. Setting the buffer size too big may only consume too 1827 * much kernel memory if the application doesn't read() from 1828 * the socket or packet loss or reordering makes use of the 1829 * reassembly queue. 1830 * 1831 * The criteria to step up the receive buffer one notch are: 1832 * 1. the number of bytes received during the time it takes 1833 * one timestamp to be reflected back to us (the RTT); 1834 * 2. received bytes per RTT is within seven eighth of the 1835 * current socket buffer size; 1836 * 3. receive buffer size has not hit maximal automatic size; 1837 * 1838 * This algorithm does one step per RTT at most and only if 1839 * we receive a bulk stream w/o packet losses or reorderings. 1840 * Shrinking the buffer during idle times is not necessary as 1841 * it doesn't consume any memory when idle. 1842 * 1843 * TODO: Only step up if the application is actually serving 1844 * the buffer to better manage the socket buffer resources. 1845 */ 1846 if (tcp_do_autorcvbuf && 1847 opti.ts_ecr && 1848 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1849 if (opti.ts_ecr > tp->rfbuf_ts && 1850 opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) { 1851 if (tp->rfbuf_cnt > 1852 (so->so_rcv.sb_hiwat / 8 * 7) && 1853 so->so_rcv.sb_hiwat < 1854 tcp_autorcvbuf_max) { 1855 newsize = 1856 min(so->so_rcv.sb_hiwat + 1857 tcp_autorcvbuf_inc, 1858 tcp_autorcvbuf_max); 1859 } 1860 /* Start over with next RTT. */ 1861 tp->rfbuf_ts = 0; 1862 tp->rfbuf_cnt = 0; 1863 } else 1864 tp->rfbuf_cnt += tlen; /* add up */ 1865 } 1866 1867 /* 1868 * Drop TCP, IP headers and TCP options then add data 1869 * to socket buffer. 1870 */ 1871 if (so->so_state & SS_CANTRCVMORE) 1872 m_freem(m); 1873 else { 1874 /* 1875 * Set new socket buffer size. 1876 * Give up when limit is reached. 1877 */ 1878 if (newsize) 1879 if (!sbreserve(&so->so_rcv, 1880 newsize, so)) 1881 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1882 m_adj(m, toff + off); 1883 sbappendstream(&so->so_rcv, m); 1884 } 1885 sorwakeup(so); 1886 tcp_setup_ack(tp, th); 1887 if (tp->t_flags & TF_ACKNOW) { 1888 KERNEL_LOCK(1, NULL); 1889 (void) tcp_output(tp); 1890 KERNEL_UNLOCK_ONE(NULL); 1891 } 1892 if (tcp_saveti) 1893 m_freem(tcp_saveti); 1894 return; 1895 } 1896 } 1897 1898 /* 1899 * Compute mbuf offset to TCP data segment. 1900 */ 1901 hdroptlen = toff + off; 1902 1903 /* 1904 * Calculate amount of space in receive window, 1905 * and then do TCP input processing. 1906 * Receive window is amount of space in rcv queue, 1907 * but not less than advertised window. 1908 */ 1909 { int win; 1910 1911 win = sbspace(&so->so_rcv); 1912 if (win < 0) 1913 win = 0; 1914 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1915 } 1916 1917 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1918 tp->rfbuf_ts = 0; 1919 tp->rfbuf_cnt = 0; 1920 1921 switch (tp->t_state) { 1922 /* 1923 * If the state is SYN_SENT: 1924 * if seg contains an ACK, but not for our SYN, drop the input. 1925 * if seg contains a RST, then drop the connection. 1926 * if seg does not contain SYN, then drop it. 1927 * Otherwise this is an acceptable SYN segment 1928 * initialize tp->rcv_nxt and tp->irs 1929 * if seg contains ack then advance tp->snd_una 1930 * if seg contains a ECE and ECN support is enabled, the stream 1931 * is ECN capable. 1932 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1933 * arrange for segment to be acked (eventually) 1934 * continue processing rest of data/controls, beginning with URG 1935 */ 1936 case TCPS_SYN_SENT: 1937 if ((tiflags & TH_ACK) && 1938 (SEQ_LEQ(th->th_ack, tp->iss) || 1939 SEQ_GT(th->th_ack, tp->snd_max))) 1940 goto dropwithreset; 1941 if (tiflags & TH_RST) { 1942 if (tiflags & TH_ACK) 1943 tp = tcp_drop(tp, ECONNREFUSED); 1944 goto drop; 1945 } 1946 if ((tiflags & TH_SYN) == 0) 1947 goto drop; 1948 if (tiflags & TH_ACK) { 1949 tp->snd_una = th->th_ack; 1950 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1951 tp->snd_nxt = tp->snd_una; 1952 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1953 tp->snd_high = tp->snd_una; 1954 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1955 1956 if ((tiflags & TH_ECE) && tcp_do_ecn) { 1957 tp->t_flags |= TF_ECN_PERMIT; 1958 TCP_STATINC(TCP_STAT_ECN_SHS); 1959 } 1960 1961 } 1962 tp->irs = th->th_seq; 1963 tcp_rcvseqinit(tp); 1964 tp->t_flags |= TF_ACKNOW; 1965 tcp_mss_from_peer(tp, opti.maxseg); 1966 1967 /* 1968 * Initialize the initial congestion window. If we 1969 * had to retransmit the SYN, we must initialize cwnd 1970 * to 1 segment (i.e. the Loss Window). 1971 */ 1972 if (tp->t_flags & TF_SYN_REXMT) 1973 tp->snd_cwnd = tp->t_peermss; 1974 else { 1975 int ss = tcp_init_win; 1976 #ifdef INET 1977 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1978 ss = tcp_init_win_local; 1979 #endif 1980 #ifdef INET6 1981 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1982 ss = tcp_init_win_local; 1983 #endif 1984 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1985 } 1986 1987 tcp_rmx_rtt(tp); 1988 if (tiflags & TH_ACK) { 1989 TCP_STATINC(TCP_STAT_CONNECTS); 1990 soisconnected(so); 1991 tcp_established(tp); 1992 /* Do window scaling on this connection? */ 1993 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1994 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1995 tp->snd_scale = tp->requested_s_scale; 1996 tp->rcv_scale = tp->request_r_scale; 1997 } 1998 TCP_REASS_LOCK(tp); 1999 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2000 TCP_REASS_UNLOCK(tp); 2001 /* 2002 * if we didn't have to retransmit the SYN, 2003 * use its rtt as our initial srtt & rtt var. 2004 */ 2005 if (tp->t_rtttime) 2006 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2007 } else 2008 tp->t_state = TCPS_SYN_RECEIVED; 2009 2010 /* 2011 * Advance th->th_seq to correspond to first data byte. 2012 * If data, trim to stay within window, 2013 * dropping FIN if necessary. 2014 */ 2015 th->th_seq++; 2016 if (tlen > tp->rcv_wnd) { 2017 todrop = tlen - tp->rcv_wnd; 2018 m_adj(m, -todrop); 2019 tlen = tp->rcv_wnd; 2020 tiflags &= ~TH_FIN; 2021 tcps = TCP_STAT_GETREF(); 2022 tcps[TCP_STAT_RCVPACKAFTERWIN]++; 2023 tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop; 2024 TCP_STAT_PUTREF(); 2025 } 2026 tp->snd_wl1 = th->th_seq - 1; 2027 tp->rcv_up = th->th_seq; 2028 goto step6; 2029 2030 /* 2031 * If the state is SYN_RECEIVED: 2032 * If seg contains an ACK, but not for our SYN, drop the input 2033 * and generate an RST. See page 36, rfc793 2034 */ 2035 case TCPS_SYN_RECEIVED: 2036 if ((tiflags & TH_ACK) && 2037 (SEQ_LEQ(th->th_ack, tp->iss) || 2038 SEQ_GT(th->th_ack, tp->snd_max))) 2039 goto dropwithreset; 2040 break; 2041 } 2042 2043 /* 2044 * States other than LISTEN or SYN_SENT. 2045 * First check timestamp, if present. 2046 * Then check that at least some bytes of segment are within 2047 * receive window. If segment begins before rcv_nxt, 2048 * drop leading data (and SYN); if nothing left, just ack. 2049 * 2050 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2051 * and it's less than ts_recent, drop it. 2052 */ 2053 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 2054 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 2055 2056 /* Check to see if ts_recent is over 24 days old. */ 2057 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 2058 /* 2059 * Invalidate ts_recent. If this segment updates 2060 * ts_recent, the age will be reset later and ts_recent 2061 * will get a valid value. If it does not, setting 2062 * ts_recent to zero will at least satisfy the 2063 * requirement that zero be placed in the timestamp 2064 * echo reply when ts_recent isn't valid. The 2065 * age isn't reset until we get a valid ts_recent 2066 * because we don't want out-of-order segments to be 2067 * dropped when ts_recent is old. 2068 */ 2069 tp->ts_recent = 0; 2070 } else { 2071 tcps = TCP_STAT_GETREF(); 2072 tcps[TCP_STAT_RCVDUPPACK]++; 2073 tcps[TCP_STAT_RCVDUPBYTE] += tlen; 2074 tcps[TCP_STAT_PAWSDROP]++; 2075 TCP_STAT_PUTREF(); 2076 tcp_new_dsack(tp, th->th_seq, tlen); 2077 goto dropafterack; 2078 } 2079 } 2080 2081 todrop = tp->rcv_nxt - th->th_seq; 2082 dupseg = false; 2083 if (todrop > 0) { 2084 if (tiflags & TH_SYN) { 2085 tiflags &= ~TH_SYN; 2086 th->th_seq++; 2087 if (th->th_urp > 1) 2088 th->th_urp--; 2089 else { 2090 tiflags &= ~TH_URG; 2091 th->th_urp = 0; 2092 } 2093 todrop--; 2094 } 2095 if (todrop > tlen || 2096 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 2097 /* 2098 * Any valid FIN or RST must be to the left of the 2099 * window. At this point the FIN or RST must be a 2100 * duplicate or out of sequence; drop it. 2101 */ 2102 if (tiflags & TH_RST) 2103 goto drop; 2104 tiflags &= ~(TH_FIN|TH_RST); 2105 /* 2106 * Send an ACK to resynchronize and drop any data. 2107 * But keep on processing for RST or ACK. 2108 */ 2109 tp->t_flags |= TF_ACKNOW; 2110 todrop = tlen; 2111 dupseg = true; 2112 tcps = TCP_STAT_GETREF(); 2113 tcps[TCP_STAT_RCVDUPPACK]++; 2114 tcps[TCP_STAT_RCVDUPBYTE] += todrop; 2115 TCP_STAT_PUTREF(); 2116 } else if ((tiflags & TH_RST) && 2117 th->th_seq != tp->rcv_nxt) { 2118 /* 2119 * Test for reset before adjusting the sequence 2120 * number for overlapping data. 2121 */ 2122 goto dropafterack_ratelim; 2123 } else { 2124 tcps = TCP_STAT_GETREF(); 2125 tcps[TCP_STAT_RCVPARTDUPPACK]++; 2126 tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop; 2127 TCP_STAT_PUTREF(); 2128 } 2129 tcp_new_dsack(tp, th->th_seq, todrop); 2130 hdroptlen += todrop; /*drop from head afterwards*/ 2131 th->th_seq += todrop; 2132 tlen -= todrop; 2133 if (th->th_urp > todrop) 2134 th->th_urp -= todrop; 2135 else { 2136 tiflags &= ~TH_URG; 2137 th->th_urp = 0; 2138 } 2139 } 2140 2141 /* 2142 * If new data are received on a connection after the 2143 * user processes are gone, then RST the other end. 2144 */ 2145 if ((so->so_state & SS_NOFDREF) && 2146 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2147 tp = tcp_close(tp); 2148 TCP_STATINC(TCP_STAT_RCVAFTERCLOSE); 2149 goto dropwithreset; 2150 } 2151 2152 /* 2153 * If segment ends after window, drop trailing data 2154 * (and PUSH and FIN); if nothing left, just ACK. 2155 */ 2156 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 2157 if (todrop > 0) { 2158 TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN); 2159 if (todrop >= tlen) { 2160 /* 2161 * The segment actually starts after the window. 2162 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 2163 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 2164 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 2165 */ 2166 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen); 2167 /* 2168 * If a new connection request is received 2169 * while in TIME_WAIT, drop the old connection 2170 * and start over if the sequence numbers 2171 * are above the previous ones. 2172 * 2173 * NOTE: We will checksum the packet again, and 2174 * so we need to put the header fields back into 2175 * network order! 2176 * XXX This kind of sucks, but we don't expect 2177 * XXX this to happen very often, so maybe it 2178 * XXX doesn't matter so much. 2179 */ 2180 if (tiflags & TH_SYN && 2181 tp->t_state == TCPS_TIME_WAIT && 2182 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 2183 tp = tcp_close(tp); 2184 tcp_fields_to_net(th); 2185 goto findpcb; 2186 } 2187 /* 2188 * If window is closed can only take segments at 2189 * window edge, and have to drop data and PUSH from 2190 * incoming segments. Continue processing, but 2191 * remember to ack. Otherwise, drop segment 2192 * and (if not RST) ack. 2193 */ 2194 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2195 tp->t_flags |= TF_ACKNOW; 2196 TCP_STATINC(TCP_STAT_RCVWINPROBE); 2197 } else 2198 goto dropafterack; 2199 } else 2200 TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop); 2201 m_adj(m, -todrop); 2202 tlen -= todrop; 2203 tiflags &= ~(TH_PUSH|TH_FIN); 2204 } 2205 2206 /* 2207 * If last ACK falls within this segment's sequence numbers, 2208 * record the timestamp. 2209 * NOTE: 2210 * 1) That the test incorporates suggestions from the latest 2211 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2212 * 2) That updating only on newer timestamps interferes with 2213 * our earlier PAWS tests, so this check should be solely 2214 * predicated on the sequence space of this segment. 2215 * 3) That we modify the segment boundary check to be 2216 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2217 * instead of RFC1323's 2218 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2219 * This modified check allows us to overcome RFC1323's 2220 * limitations as described in Stevens TCP/IP Illustrated 2221 * Vol. 2 p.869. In such cases, we can still calculate the 2222 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2223 */ 2224 if (opti.ts_present && 2225 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2226 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2227 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 2228 tp->ts_recent_age = tcp_now; 2229 tp->ts_recent = opti.ts_val; 2230 } 2231 2232 /* 2233 * If the RST bit is set examine the state: 2234 * SYN_RECEIVED STATE: 2235 * If passive open, return to LISTEN state. 2236 * If active open, inform user that connection was refused. 2237 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 2238 * Inform user that connection was reset, and close tcb. 2239 * CLOSING, LAST_ACK, TIME_WAIT STATES 2240 * Close the tcb. 2241 */ 2242 if (tiflags & TH_RST) { 2243 if (th->th_seq != tp->rcv_nxt) 2244 goto dropafterack_ratelim; 2245 2246 switch (tp->t_state) { 2247 case TCPS_SYN_RECEIVED: 2248 so->so_error = ECONNREFUSED; 2249 goto close; 2250 2251 case TCPS_ESTABLISHED: 2252 case TCPS_FIN_WAIT_1: 2253 case TCPS_FIN_WAIT_2: 2254 case TCPS_CLOSE_WAIT: 2255 so->so_error = ECONNRESET; 2256 close: 2257 tp->t_state = TCPS_CLOSED; 2258 TCP_STATINC(TCP_STAT_DROPS); 2259 tp = tcp_close(tp); 2260 goto drop; 2261 2262 case TCPS_CLOSING: 2263 case TCPS_LAST_ACK: 2264 case TCPS_TIME_WAIT: 2265 tp = tcp_close(tp); 2266 goto drop; 2267 } 2268 } 2269 2270 /* 2271 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2272 * we must be in a synchronized state. RFC791 states (under RST 2273 * generation) that any unacceptable segment (an out-of-order SYN 2274 * qualifies) received in a synchronized state must elicit only an 2275 * empty acknowledgment segment ... and the connection remains in 2276 * the same state. 2277 */ 2278 if (tiflags & TH_SYN) { 2279 if (tp->rcv_nxt == th->th_seq) { 2280 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2281 TH_ACK); 2282 if (tcp_saveti) 2283 m_freem(tcp_saveti); 2284 return; 2285 } 2286 2287 goto dropafterack_ratelim; 2288 } 2289 2290 /* 2291 * If the ACK bit is off we drop the segment and return. 2292 */ 2293 if ((tiflags & TH_ACK) == 0) { 2294 if (tp->t_flags & TF_ACKNOW) 2295 goto dropafterack; 2296 else 2297 goto drop; 2298 } 2299 2300 /* 2301 * Ack processing. 2302 */ 2303 switch (tp->t_state) { 2304 2305 /* 2306 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2307 * ESTABLISHED state and continue processing, otherwise 2308 * send an RST. 2309 */ 2310 case TCPS_SYN_RECEIVED: 2311 if (SEQ_GT(tp->snd_una, th->th_ack) || 2312 SEQ_GT(th->th_ack, tp->snd_max)) 2313 goto dropwithreset; 2314 TCP_STATINC(TCP_STAT_CONNECTS); 2315 soisconnected(so); 2316 tcp_established(tp); 2317 /* Do window scaling? */ 2318 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2319 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2320 tp->snd_scale = tp->requested_s_scale; 2321 tp->rcv_scale = tp->request_r_scale; 2322 } 2323 TCP_REASS_LOCK(tp); 2324 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2325 TCP_REASS_UNLOCK(tp); 2326 tp->snd_wl1 = th->th_seq - 1; 2327 /* fall into ... */ 2328 2329 /* 2330 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2331 * ACKs. If the ack is in the range 2332 * tp->snd_una < th->th_ack <= tp->snd_max 2333 * then advance tp->snd_una to th->th_ack and drop 2334 * data from the retransmission queue. If this ACK reflects 2335 * more up to date window information we update our window information. 2336 */ 2337 case TCPS_ESTABLISHED: 2338 case TCPS_FIN_WAIT_1: 2339 case TCPS_FIN_WAIT_2: 2340 case TCPS_CLOSE_WAIT: 2341 case TCPS_CLOSING: 2342 case TCPS_LAST_ACK: 2343 case TCPS_TIME_WAIT: 2344 2345 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2346 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2347 TCP_STATINC(TCP_STAT_RCVDUPPACK); 2348 /* 2349 * If we have outstanding data (other than 2350 * a window probe), this is a completely 2351 * duplicate ack (ie, window info didn't 2352 * change), the ack is the biggest we've 2353 * seen and we've seen exactly our rexmt 2354 * threshhold of them, assume a packet 2355 * has been dropped and retransmit it. 2356 * Kludge snd_nxt & the congestion 2357 * window so we send only this one 2358 * packet. 2359 */ 2360 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2361 th->th_ack != tp->snd_una) 2362 tp->t_dupacks = 0; 2363 else if (tp->t_partialacks < 0 && 2364 (++tp->t_dupacks == tcprexmtthresh || 2365 TCP_FACK_FASTRECOV(tp))) { 2366 /* 2367 * Do the fast retransmit, and adjust 2368 * congestion control paramenters. 2369 */ 2370 if (tp->t_congctl->fast_retransmit(tp, th)) { 2371 /* False fast retransmit */ 2372 break; 2373 } else 2374 goto drop; 2375 } else if (tp->t_dupacks > tcprexmtthresh) { 2376 tp->snd_cwnd += tp->t_segsz; 2377 KERNEL_LOCK(1, NULL); 2378 (void) tcp_output(tp); 2379 KERNEL_UNLOCK_ONE(NULL); 2380 goto drop; 2381 } 2382 } else { 2383 /* 2384 * If the ack appears to be very old, only 2385 * allow data that is in-sequence. This 2386 * makes it somewhat more difficult to insert 2387 * forged data by guessing sequence numbers. 2388 * Sent an ack to try to update the send 2389 * sequence number on the other side. 2390 */ 2391 if (tlen && th->th_seq != tp->rcv_nxt && 2392 SEQ_LT(th->th_ack, 2393 tp->snd_una - tp->max_sndwnd)) 2394 goto dropafterack; 2395 } 2396 break; 2397 } 2398 /* 2399 * If the congestion window was inflated to account 2400 * for the other side's cached packets, retract it. 2401 */ 2402 /* XXX: make SACK have his own congestion control 2403 * struct -- rpaulo */ 2404 if (TCP_SACK_ENABLED(tp)) 2405 tcp_sack_newack(tp, th); 2406 else 2407 tp->t_congctl->fast_retransmit_newack(tp, th); 2408 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2409 TCP_STATINC(TCP_STAT_RCVACKTOOMUCH); 2410 goto dropafterack; 2411 } 2412 acked = th->th_ack - tp->snd_una; 2413 tcps = TCP_STAT_GETREF(); 2414 tcps[TCP_STAT_RCVACKPACK]++; 2415 tcps[TCP_STAT_RCVACKBYTE] += acked; 2416 TCP_STAT_PUTREF(); 2417 2418 /* 2419 * If we have a timestamp reply, update smoothed 2420 * round trip time. If no timestamp is present but 2421 * transmit timer is running and timed sequence 2422 * number was acked, update smoothed round trip time. 2423 * Since we now have an rtt measurement, cancel the 2424 * timer backoff (cf., Phil Karn's retransmit alg.). 2425 * Recompute the initial retransmit timer. 2426 */ 2427 if (ts_rtt) 2428 tcp_xmit_timer(tp, ts_rtt); 2429 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2430 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2431 2432 /* 2433 * If all outstanding data is acked, stop retransmit 2434 * timer and remember to restart (more output or persist). 2435 * If there is more data to be acked, restart retransmit 2436 * timer, using current (possibly backed-off) value. 2437 */ 2438 if (th->th_ack == tp->snd_max) { 2439 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2440 needoutput = 1; 2441 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2442 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2443 2444 /* 2445 * New data has been acked, adjust the congestion window. 2446 */ 2447 tp->t_congctl->newack(tp, th); 2448 2449 nd6_hint(tp); 2450 if (acked > so->so_snd.sb_cc) { 2451 tp->snd_wnd -= so->so_snd.sb_cc; 2452 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2453 ourfinisacked = 1; 2454 } else { 2455 if (acked > (tp->t_lastoff - tp->t_inoff)) 2456 tp->t_lastm = NULL; 2457 sbdrop(&so->so_snd, acked); 2458 tp->t_lastoff -= acked; 2459 tp->snd_wnd -= acked; 2460 ourfinisacked = 0; 2461 } 2462 sowwakeup(so); 2463 2464 icmp_check(tp, th, acked); 2465 2466 tp->snd_una = th->th_ack; 2467 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2468 tp->snd_fack = tp->snd_una; 2469 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2470 tp->snd_nxt = tp->snd_una; 2471 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2472 tp->snd_high = tp->snd_una; 2473 2474 switch (tp->t_state) { 2475 2476 /* 2477 * In FIN_WAIT_1 STATE in addition to the processing 2478 * for the ESTABLISHED state if our FIN is now acknowledged 2479 * then enter FIN_WAIT_2. 2480 */ 2481 case TCPS_FIN_WAIT_1: 2482 if (ourfinisacked) { 2483 /* 2484 * If we can't receive any more 2485 * data, then closing user can proceed. 2486 * Starting the timer is contrary to the 2487 * specification, but if we don't get a FIN 2488 * we'll hang forever. 2489 */ 2490 if (so->so_state & SS_CANTRCVMORE) { 2491 soisdisconnected(so); 2492 if (tp->t_maxidle > 0) 2493 TCP_TIMER_ARM(tp, TCPT_2MSL, 2494 tp->t_maxidle); 2495 } 2496 tp->t_state = TCPS_FIN_WAIT_2; 2497 } 2498 break; 2499 2500 /* 2501 * In CLOSING STATE in addition to the processing for 2502 * the ESTABLISHED state if the ACK acknowledges our FIN 2503 * then enter the TIME-WAIT state, otherwise ignore 2504 * the segment. 2505 */ 2506 case TCPS_CLOSING: 2507 if (ourfinisacked) { 2508 tp->t_state = TCPS_TIME_WAIT; 2509 tcp_canceltimers(tp); 2510 TCP_TIMER_ARM(tp, TCPT_2MSL, 2511 2 * PR_SLOWHZ * tcp_msl); 2512 soisdisconnected(so); 2513 } 2514 break; 2515 2516 /* 2517 * In LAST_ACK, we may still be waiting for data to drain 2518 * and/or to be acked, as well as for the ack of our FIN. 2519 * If our FIN is now acknowledged, delete the TCB, 2520 * enter the closed state and return. 2521 */ 2522 case TCPS_LAST_ACK: 2523 if (ourfinisacked) { 2524 tp = tcp_close(tp); 2525 goto drop; 2526 } 2527 break; 2528 2529 /* 2530 * In TIME_WAIT state the only thing that should arrive 2531 * is a retransmission of the remote FIN. Acknowledge 2532 * it and restart the finack timer. 2533 */ 2534 case TCPS_TIME_WAIT: 2535 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * PR_SLOWHZ * tcp_msl); 2536 goto dropafterack; 2537 } 2538 } 2539 2540 step6: 2541 /* 2542 * Update window information. 2543 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2544 */ 2545 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2546 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2547 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2548 /* keep track of pure window updates */ 2549 if (tlen == 0 && 2550 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2551 TCP_STATINC(TCP_STAT_RCVWINUPD); 2552 tp->snd_wnd = tiwin; 2553 tp->snd_wl1 = th->th_seq; 2554 tp->snd_wl2 = th->th_ack; 2555 if (tp->snd_wnd > tp->max_sndwnd) 2556 tp->max_sndwnd = tp->snd_wnd; 2557 needoutput = 1; 2558 } 2559 2560 /* 2561 * Process segments with URG. 2562 */ 2563 if ((tiflags & TH_URG) && th->th_urp && 2564 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2565 /* 2566 * This is a kludge, but if we receive and accept 2567 * random urgent pointers, we'll crash in 2568 * soreceive. It's hard to imagine someone 2569 * actually wanting to send this much urgent data. 2570 */ 2571 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2572 th->th_urp = 0; /* XXX */ 2573 tiflags &= ~TH_URG; /* XXX */ 2574 goto dodata; /* XXX */ 2575 } 2576 /* 2577 * If this segment advances the known urgent pointer, 2578 * then mark the data stream. This should not happen 2579 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2580 * a FIN has been received from the remote side. 2581 * In these states we ignore the URG. 2582 * 2583 * According to RFC961 (Assigned Protocols), 2584 * the urgent pointer points to the last octet 2585 * of urgent data. We continue, however, 2586 * to consider it to indicate the first octet 2587 * of data past the urgent section as the original 2588 * spec states (in one of two places). 2589 */ 2590 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2591 tp->rcv_up = th->th_seq + th->th_urp; 2592 so->so_oobmark = so->so_rcv.sb_cc + 2593 (tp->rcv_up - tp->rcv_nxt) - 1; 2594 if (so->so_oobmark == 0) 2595 so->so_state |= SS_RCVATMARK; 2596 sohasoutofband(so); 2597 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2598 } 2599 /* 2600 * Remove out of band data so doesn't get presented to user. 2601 * This can happen independent of advancing the URG pointer, 2602 * but if two URG's are pending at once, some out-of-band 2603 * data may creep in... ick. 2604 */ 2605 if (th->th_urp <= (u_int16_t) tlen 2606 #ifdef SO_OOBINLINE 2607 && (so->so_options & SO_OOBINLINE) == 0 2608 #endif 2609 ) 2610 tcp_pulloutofband(so, th, m, hdroptlen); 2611 } else 2612 /* 2613 * If no out of band data is expected, 2614 * pull receive urgent pointer along 2615 * with the receive window. 2616 */ 2617 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2618 tp->rcv_up = tp->rcv_nxt; 2619 dodata: /* XXX */ 2620 2621 /* 2622 * Process the segment text, merging it into the TCP sequencing queue, 2623 * and arranging for acknowledgement of receipt if necessary. 2624 * This process logically involves adjusting tp->rcv_wnd as data 2625 * is presented to the user (this happens in tcp_usrreq.c, 2626 * case PRU_RCVD). If a FIN has already been received on this 2627 * connection then we just ignore the text. 2628 */ 2629 if ((tlen || (tiflags & TH_FIN)) && 2630 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2631 /* 2632 * Insert segment ti into reassembly queue of tcp with 2633 * control block tp. Return TH_FIN if reassembly now includes 2634 * a segment with FIN. The macro form does the common case 2635 * inline (segment is the next to be received on an 2636 * established connection, and the queue is empty), 2637 * avoiding linkage into and removal from the queue and 2638 * repetition of various conversions. 2639 * Set DELACK for segments received in order, but ack 2640 * immediately when segments are out of order 2641 * (so fast retransmit can work). 2642 */ 2643 /* NOTE: this was TCP_REASS() macro, but used only once */ 2644 TCP_REASS_LOCK(tp); 2645 if (th->th_seq == tp->rcv_nxt && 2646 TAILQ_FIRST(&tp->segq) == NULL && 2647 tp->t_state == TCPS_ESTABLISHED) { 2648 tcp_setup_ack(tp, th); 2649 tp->rcv_nxt += tlen; 2650 tiflags = th->th_flags & TH_FIN; 2651 tcps = TCP_STAT_GETREF(); 2652 tcps[TCP_STAT_RCVPACK]++; 2653 tcps[TCP_STAT_RCVBYTE] += tlen; 2654 TCP_STAT_PUTREF(); 2655 nd6_hint(tp); 2656 if (so->so_state & SS_CANTRCVMORE) 2657 m_freem(m); 2658 else { 2659 m_adj(m, hdroptlen); 2660 sbappendstream(&(so)->so_rcv, m); 2661 } 2662 TCP_REASS_UNLOCK(tp); 2663 sorwakeup(so); 2664 } else { 2665 m_adj(m, hdroptlen); 2666 tiflags = tcp_reass(tp, th, m, &tlen); 2667 tp->t_flags |= TF_ACKNOW; 2668 TCP_REASS_UNLOCK(tp); 2669 } 2670 2671 /* 2672 * Note the amount of data that peer has sent into 2673 * our window, in order to estimate the sender's 2674 * buffer size. 2675 */ 2676 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2677 } else { 2678 m_freem(m); 2679 m = NULL; 2680 tiflags &= ~TH_FIN; 2681 } 2682 2683 /* 2684 * If FIN is received ACK the FIN and let the user know 2685 * that the connection is closing. Ignore a FIN received before 2686 * the connection is fully established. 2687 */ 2688 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2689 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2690 socantrcvmore(so); 2691 tp->t_flags |= TF_ACKNOW; 2692 tp->rcv_nxt++; 2693 } 2694 switch (tp->t_state) { 2695 2696 /* 2697 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2698 */ 2699 case TCPS_ESTABLISHED: 2700 tp->t_state = TCPS_CLOSE_WAIT; 2701 break; 2702 2703 /* 2704 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2705 * enter the CLOSING state. 2706 */ 2707 case TCPS_FIN_WAIT_1: 2708 tp->t_state = TCPS_CLOSING; 2709 break; 2710 2711 /* 2712 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2713 * starting the time-wait timer, turning off the other 2714 * standard timers. 2715 */ 2716 case TCPS_FIN_WAIT_2: 2717 tp->t_state = TCPS_TIME_WAIT; 2718 tcp_canceltimers(tp); 2719 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * PR_SLOWHZ * tcp_msl); 2720 soisdisconnected(so); 2721 break; 2722 2723 /* 2724 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2725 */ 2726 case TCPS_TIME_WAIT: 2727 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * PR_SLOWHZ * tcp_msl); 2728 break; 2729 } 2730 } 2731 #ifdef TCP_DEBUG 2732 if (so->so_options & SO_DEBUG) 2733 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2734 #endif 2735 2736 /* 2737 * Return any desired output. 2738 */ 2739 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2740 KERNEL_LOCK(1, NULL); 2741 (void) tcp_output(tp); 2742 KERNEL_UNLOCK_ONE(NULL); 2743 } 2744 if (tcp_saveti) 2745 m_freem(tcp_saveti); 2746 return; 2747 2748 badsyn: 2749 /* 2750 * Received a bad SYN. Increment counters and dropwithreset. 2751 */ 2752 TCP_STATINC(TCP_STAT_BADSYN); 2753 tp = NULL; 2754 goto dropwithreset; 2755 2756 dropafterack: 2757 /* 2758 * Generate an ACK dropping incoming segment if it occupies 2759 * sequence space, where the ACK reflects our state. 2760 */ 2761 if (tiflags & TH_RST) 2762 goto drop; 2763 goto dropafterack2; 2764 2765 dropafterack_ratelim: 2766 /* 2767 * We may want to rate-limit ACKs against SYN/RST attack. 2768 */ 2769 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2770 tcp_ackdrop_ppslim) == 0) { 2771 /* XXX stat */ 2772 goto drop; 2773 } 2774 /* ...fall into dropafterack2... */ 2775 2776 dropafterack2: 2777 m_freem(m); 2778 tp->t_flags |= TF_ACKNOW; 2779 KERNEL_LOCK(1, NULL); 2780 (void) tcp_output(tp); 2781 KERNEL_UNLOCK_ONE(NULL); 2782 if (tcp_saveti) 2783 m_freem(tcp_saveti); 2784 return; 2785 2786 dropwithreset_ratelim: 2787 /* 2788 * We may want to rate-limit RSTs in certain situations, 2789 * particularly if we are sending an RST in response to 2790 * an attempt to connect to or otherwise communicate with 2791 * a port for which we have no socket. 2792 */ 2793 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2794 tcp_rst_ppslim) == 0) { 2795 /* XXX stat */ 2796 goto drop; 2797 } 2798 /* ...fall into dropwithreset... */ 2799 2800 dropwithreset: 2801 /* 2802 * Generate a RST, dropping incoming segment. 2803 * Make ACK acceptable to originator of segment. 2804 */ 2805 if (tiflags & TH_RST) 2806 goto drop; 2807 2808 switch (af) { 2809 #ifdef INET6 2810 case AF_INET6: 2811 /* For following calls to tcp_respond */ 2812 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2813 goto drop; 2814 break; 2815 #endif /* INET6 */ 2816 case AF_INET: 2817 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2818 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2819 goto drop; 2820 } 2821 2822 if (tiflags & TH_ACK) 2823 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2824 else { 2825 if (tiflags & TH_SYN) 2826 tlen++; 2827 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2828 TH_RST|TH_ACK); 2829 } 2830 if (tcp_saveti) 2831 m_freem(tcp_saveti); 2832 return; 2833 2834 badcsum: 2835 drop: 2836 /* 2837 * Drop space held by incoming segment and return. 2838 */ 2839 if (tp) { 2840 if (tp->t_inpcb) 2841 so = tp->t_inpcb->inp_socket; 2842 #ifdef INET6 2843 else if (tp->t_in6pcb) 2844 so = tp->t_in6pcb->in6p_socket; 2845 #endif 2846 else 2847 so = NULL; 2848 #ifdef TCP_DEBUG 2849 if (so && (so->so_options & SO_DEBUG) != 0) 2850 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2851 #endif 2852 } 2853 if (tcp_saveti) 2854 m_freem(tcp_saveti); 2855 m_freem(m); 2856 return; 2857 } 2858 2859 #ifdef TCP_SIGNATURE 2860 int 2861 tcp_signature_apply(void *fstate, void *data, u_int len) 2862 { 2863 2864 MD5Update(fstate, (u_char *)data, len); 2865 return (0); 2866 } 2867 2868 struct secasvar * 2869 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2870 { 2871 struct secasvar *sav; 2872 #ifdef FAST_IPSEC 2873 union sockaddr_union dst; 2874 #endif 2875 struct ip *ip; 2876 struct ip6_hdr *ip6; 2877 2878 ip = mtod(m, struct ip *); 2879 switch (ip->ip_v) { 2880 case 4: 2881 ip = mtod(m, struct ip *); 2882 ip6 = NULL; 2883 break; 2884 case 6: 2885 ip = NULL; 2886 ip6 = mtod(m, struct ip6_hdr *); 2887 break; 2888 default: 2889 return (NULL); 2890 } 2891 2892 #ifdef FAST_IPSEC 2893 /* Extract the destination from the IP header in the mbuf. */ 2894 memset(&dst, 0, sizeof(union sockaddr_union)); 2895 if (ip !=NULL) { 2896 dst.sa.sa_len = sizeof(struct sockaddr_in); 2897 dst.sa.sa_family = AF_INET; 2898 dst.sin.sin_addr = ip->ip_dst; 2899 } else { 2900 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2901 dst.sa.sa_family = AF_INET6; 2902 dst.sin6.sin6_addr = ip6->ip6_dst; 2903 } 2904 2905 /* 2906 * Look up an SADB entry which matches the address of the peer. 2907 */ 2908 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2909 #else 2910 if (ip) 2911 sav = key_allocsa(AF_INET, (void *)&ip->ip_src, 2912 (void *)&ip->ip_dst, IPPROTO_TCP, 2913 htonl(TCP_SIG_SPI), 0, 0); 2914 else 2915 sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src, 2916 (void *)&ip6->ip6_dst, IPPROTO_TCP, 2917 htonl(TCP_SIG_SPI), 0, 0); 2918 #endif 2919 2920 return (sav); /* freesav must be performed by caller */ 2921 } 2922 2923 int 2924 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2925 struct secasvar *sav, char *sig) 2926 { 2927 MD5_CTX ctx; 2928 struct ip *ip; 2929 struct ipovly *ipovly; 2930 struct ip6_hdr *ip6; 2931 struct ippseudo ippseudo; 2932 struct ip6_hdr_pseudo ip6pseudo; 2933 struct tcphdr th0; 2934 int l, tcphdrlen; 2935 2936 if (sav == NULL) 2937 return (-1); 2938 2939 tcphdrlen = th->th_off * 4; 2940 2941 switch (mtod(m, struct ip *)->ip_v) { 2942 case 4: 2943 ip = mtod(m, struct ip *); 2944 ip6 = NULL; 2945 break; 2946 case 6: 2947 ip = NULL; 2948 ip6 = mtod(m, struct ip6_hdr *); 2949 break; 2950 default: 2951 return (-1); 2952 } 2953 2954 MD5Init(&ctx); 2955 2956 if (ip) { 2957 memset(&ippseudo, 0, sizeof(ippseudo)); 2958 ipovly = (struct ipovly *)ip; 2959 ippseudo.ippseudo_src = ipovly->ih_src; 2960 ippseudo.ippseudo_dst = ipovly->ih_dst; 2961 ippseudo.ippseudo_pad = 0; 2962 ippseudo.ippseudo_p = IPPROTO_TCP; 2963 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2964 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2965 } else { 2966 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2967 ip6pseudo.ip6ph_src = ip6->ip6_src; 2968 in6_clearscope(&ip6pseudo.ip6ph_src); 2969 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2970 in6_clearscope(&ip6pseudo.ip6ph_dst); 2971 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2972 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2973 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2974 } 2975 2976 th0 = *th; 2977 th0.th_sum = 0; 2978 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2979 2980 l = m->m_pkthdr.len - thoff - tcphdrlen; 2981 if (l > 0) 2982 m_apply(m, thoff + tcphdrlen, 2983 m->m_pkthdr.len - thoff - tcphdrlen, 2984 tcp_signature_apply, &ctx); 2985 2986 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2987 MD5Final(sig, &ctx); 2988 2989 return (0); 2990 } 2991 #endif 2992 2993 static int 2994 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt, 2995 struct tcphdr *th, 2996 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2997 { 2998 u_int16_t mss; 2999 int opt, optlen = 0; 3000 #ifdef TCP_SIGNATURE 3001 void *sigp = NULL; 3002 char sigbuf[TCP_SIGLEN]; 3003 struct secasvar *sav = NULL; 3004 #endif 3005 3006 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 3007 opt = cp[0]; 3008 if (opt == TCPOPT_EOL) 3009 break; 3010 if (opt == TCPOPT_NOP) 3011 optlen = 1; 3012 else { 3013 if (cnt < 2) 3014 break; 3015 optlen = cp[1]; 3016 if (optlen < 2 || optlen > cnt) 3017 break; 3018 } 3019 switch (opt) { 3020 3021 default: 3022 continue; 3023 3024 case TCPOPT_MAXSEG: 3025 if (optlen != TCPOLEN_MAXSEG) 3026 continue; 3027 if (!(th->th_flags & TH_SYN)) 3028 continue; 3029 if (TCPS_HAVERCVDSYN(tp->t_state)) 3030 continue; 3031 bcopy(cp + 2, &mss, sizeof(mss)); 3032 oi->maxseg = ntohs(mss); 3033 break; 3034 3035 case TCPOPT_WINDOW: 3036 if (optlen != TCPOLEN_WINDOW) 3037 continue; 3038 if (!(th->th_flags & TH_SYN)) 3039 continue; 3040 if (TCPS_HAVERCVDSYN(tp->t_state)) 3041 continue; 3042 tp->t_flags |= TF_RCVD_SCALE; 3043 tp->requested_s_scale = cp[2]; 3044 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 3045 #if 0 /*XXX*/ 3046 char *p; 3047 3048 if (ip) 3049 p = ntohl(ip->ip_src); 3050 #ifdef INET6 3051 else if (ip6) 3052 p = ip6_sprintf(&ip6->ip6_src); 3053 #endif 3054 else 3055 p = "(unknown)"; 3056 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 3057 "assuming %d\n", 3058 tp->requested_s_scale, p, 3059 TCP_MAX_WINSHIFT); 3060 #else 3061 log(LOG_ERR, "TCP: invalid wscale %d, " 3062 "assuming %d\n", 3063 tp->requested_s_scale, 3064 TCP_MAX_WINSHIFT); 3065 #endif 3066 tp->requested_s_scale = TCP_MAX_WINSHIFT; 3067 } 3068 break; 3069 3070 case TCPOPT_TIMESTAMP: 3071 if (optlen != TCPOLEN_TIMESTAMP) 3072 continue; 3073 oi->ts_present = 1; 3074 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 3075 NTOHL(oi->ts_val); 3076 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 3077 NTOHL(oi->ts_ecr); 3078 3079 if (!(th->th_flags & TH_SYN)) 3080 continue; 3081 if (TCPS_HAVERCVDSYN(tp->t_state)) 3082 continue; 3083 /* 3084 * A timestamp received in a SYN makes 3085 * it ok to send timestamp requests and replies. 3086 */ 3087 tp->t_flags |= TF_RCVD_TSTMP; 3088 tp->ts_recent = oi->ts_val; 3089 tp->ts_recent_age = tcp_now; 3090 break; 3091 3092 case TCPOPT_SACK_PERMITTED: 3093 if (optlen != TCPOLEN_SACK_PERMITTED) 3094 continue; 3095 if (!(th->th_flags & TH_SYN)) 3096 continue; 3097 if (TCPS_HAVERCVDSYN(tp->t_state)) 3098 continue; 3099 if (tcp_do_sack) { 3100 tp->t_flags |= TF_SACK_PERMIT; 3101 tp->t_flags |= TF_WILL_SACK; 3102 } 3103 break; 3104 3105 case TCPOPT_SACK: 3106 tcp_sack_option(tp, th, cp, optlen); 3107 break; 3108 #ifdef TCP_SIGNATURE 3109 case TCPOPT_SIGNATURE: 3110 if (optlen != TCPOLEN_SIGNATURE) 3111 continue; 3112 if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN)) 3113 return (-1); 3114 3115 sigp = sigbuf; 3116 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 3117 tp->t_flags |= TF_SIGNATURE; 3118 break; 3119 #endif 3120 } 3121 } 3122 3123 #ifdef TCP_SIGNATURE 3124 if (tp->t_flags & TF_SIGNATURE) { 3125 3126 sav = tcp_signature_getsav(m, th); 3127 3128 if (sav == NULL && tp->t_state == TCPS_LISTEN) 3129 return (-1); 3130 } 3131 3132 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 3133 if (sav == NULL) 3134 return (-1); 3135 #ifdef FAST_IPSEC 3136 KEY_FREESAV(&sav); 3137 #else 3138 key_freesav(sav); 3139 #endif 3140 return (-1); 3141 } 3142 3143 if (sigp) { 3144 char sig[TCP_SIGLEN]; 3145 3146 tcp_fields_to_net(th); 3147 if (tcp_signature(m, th, toff, sav, sig) < 0) { 3148 tcp_fields_to_host(th); 3149 if (sav == NULL) 3150 return (-1); 3151 #ifdef FAST_IPSEC 3152 KEY_FREESAV(&sav); 3153 #else 3154 key_freesav(sav); 3155 #endif 3156 return (-1); 3157 } 3158 tcp_fields_to_host(th); 3159 3160 if (memcmp(sig, sigp, TCP_SIGLEN)) { 3161 TCP_STATINC(TCP_STAT_BADSIG); 3162 if (sav == NULL) 3163 return (-1); 3164 #ifdef FAST_IPSEC 3165 KEY_FREESAV(&sav); 3166 #else 3167 key_freesav(sav); 3168 #endif 3169 return (-1); 3170 } else 3171 TCP_STATINC(TCP_STAT_GOODSIG); 3172 3173 key_sa_recordxfer(sav, m); 3174 #ifdef FAST_IPSEC 3175 KEY_FREESAV(&sav); 3176 #else 3177 key_freesav(sav); 3178 #endif 3179 } 3180 #endif 3181 3182 return (0); 3183 } 3184 3185 /* 3186 * Pull out of band byte out of a segment so 3187 * it doesn't appear in the user's data queue. 3188 * It is still reflected in the segment length for 3189 * sequencing purposes. 3190 */ 3191 void 3192 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 3193 struct mbuf *m, int off) 3194 { 3195 int cnt = off + th->th_urp - 1; 3196 3197 while (cnt >= 0) { 3198 if (m->m_len > cnt) { 3199 char *cp = mtod(m, char *) + cnt; 3200 struct tcpcb *tp = sototcpcb(so); 3201 3202 tp->t_iobc = *cp; 3203 tp->t_oobflags |= TCPOOB_HAVEDATA; 3204 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3205 m->m_len--; 3206 return; 3207 } 3208 cnt -= m->m_len; 3209 m = m->m_next; 3210 if (m == 0) 3211 break; 3212 } 3213 panic("tcp_pulloutofband"); 3214 } 3215 3216 /* 3217 * Collect new round-trip time estimate 3218 * and update averages and current timeout. 3219 */ 3220 void 3221 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3222 { 3223 int32_t delta; 3224 3225 TCP_STATINC(TCP_STAT_RTTUPDATED); 3226 if (tp->t_srtt != 0) { 3227 /* 3228 * srtt is stored as fixed point with 3 bits after the 3229 * binary point (i.e., scaled by 8). The following magic 3230 * is equivalent to the smoothing algorithm in rfc793 with 3231 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3232 * point). Adjust rtt to origin 0. 3233 */ 3234 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3235 if ((tp->t_srtt += delta) <= 0) 3236 tp->t_srtt = 1 << 2; 3237 /* 3238 * We accumulate a smoothed rtt variance (actually, a 3239 * smoothed mean difference), then set the retransmit 3240 * timer to smoothed rtt + 4 times the smoothed variance. 3241 * rttvar is stored as fixed point with 2 bits after the 3242 * binary point (scaled by 4). The following is 3243 * equivalent to rfc793 smoothing with an alpha of .75 3244 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3245 * rfc793's wired-in beta. 3246 */ 3247 if (delta < 0) 3248 delta = -delta; 3249 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3250 if ((tp->t_rttvar += delta) <= 0) 3251 tp->t_rttvar = 1 << 2; 3252 } else { 3253 /* 3254 * No rtt measurement yet - use the unsmoothed rtt. 3255 * Set the variance to half the rtt (so our first 3256 * retransmit happens at 3*rtt). 3257 */ 3258 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3259 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3260 } 3261 tp->t_rtttime = 0; 3262 tp->t_rxtshift = 0; 3263 3264 /* 3265 * the retransmit should happen at rtt + 4 * rttvar. 3266 * Because of the way we do the smoothing, srtt and rttvar 3267 * will each average +1/2 tick of bias. When we compute 3268 * the retransmit timer, we want 1/2 tick of rounding and 3269 * 1 extra tick because of +-1/2 tick uncertainty in the 3270 * firing of the timer. The bias will give us exactly the 3271 * 1.5 tick we need. But, because the bias is 3272 * statistical, we have to test that we don't drop below 3273 * the minimum feasible timer (which is 2 ticks). 3274 */ 3275 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3276 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3277 3278 /* 3279 * We received an ack for a packet that wasn't retransmitted; 3280 * it is probably safe to discard any error indications we've 3281 * received recently. This isn't quite right, but close enough 3282 * for now (a route might have failed after we sent a segment, 3283 * and the return path might not be symmetrical). 3284 */ 3285 tp->t_softerror = 0; 3286 } 3287 3288 3289 /* 3290 * TCP compressed state engine. Currently used to hold compressed 3291 * state for SYN_RECEIVED. 3292 */ 3293 3294 u_long syn_cache_count; 3295 u_int32_t syn_hash1, syn_hash2; 3296 3297 #define SYN_HASH(sa, sp, dp) \ 3298 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3299 ((u_int32_t)(sp)))^syn_hash2))) 3300 #ifndef INET6 3301 #define SYN_HASHALL(hash, src, dst) \ 3302 do { \ 3303 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3304 ((const struct sockaddr_in *)(src))->sin_port, \ 3305 ((const struct sockaddr_in *)(dst))->sin_port); \ 3306 } while (/*CONSTCOND*/ 0) 3307 #else 3308 #define SYN_HASH6(sa, sp, dp) \ 3309 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3310 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3311 & 0x7fffffff) 3312 3313 #define SYN_HASHALL(hash, src, dst) \ 3314 do { \ 3315 switch ((src)->sa_family) { \ 3316 case AF_INET: \ 3317 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3318 ((const struct sockaddr_in *)(src))->sin_port, \ 3319 ((const struct sockaddr_in *)(dst))->sin_port); \ 3320 break; \ 3321 case AF_INET6: \ 3322 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3323 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3324 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3325 break; \ 3326 default: \ 3327 hash = 0; \ 3328 } \ 3329 } while (/*CONSTCOND*/0) 3330 #endif /* INET6 */ 3331 3332 static struct pool syn_cache_pool; 3333 3334 /* 3335 * We don't estimate RTT with SYNs, so each packet starts with the default 3336 * RTT and each timer step has a fixed timeout value. 3337 */ 3338 #define SYN_CACHE_TIMER_ARM(sc) \ 3339 do { \ 3340 TCPT_RANGESET((sc)->sc_rxtcur, \ 3341 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3342 TCPTV_REXMTMAX); \ 3343 callout_reset(&(sc)->sc_timer, \ 3344 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3345 } while (/*CONSTCOND*/0) 3346 3347 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3348 3349 static inline void 3350 syn_cache_rm(struct syn_cache *sc) 3351 { 3352 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3353 sc, sc_bucketq); 3354 sc->sc_tp = NULL; 3355 LIST_REMOVE(sc, sc_tpq); 3356 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3357 callout_stop(&sc->sc_timer); 3358 syn_cache_count--; 3359 } 3360 3361 static inline void 3362 syn_cache_put(struct syn_cache *sc) 3363 { 3364 if (sc->sc_ipopts) 3365 (void) m_free(sc->sc_ipopts); 3366 rtcache_free(&sc->sc_route); 3367 sc->sc_flags |= SCF_DEAD; 3368 if (!callout_invoking(&sc->sc_timer)) 3369 callout_schedule(&(sc)->sc_timer, 1); 3370 } 3371 3372 void 3373 syn_cache_init(void) 3374 { 3375 int i; 3376 3377 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3378 "synpl", NULL, IPL_SOFTNET); 3379 3380 /* Initialize the hash buckets. */ 3381 for (i = 0; i < tcp_syn_cache_size; i++) 3382 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3383 } 3384 3385 void 3386 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3387 { 3388 struct syn_cache_head *scp; 3389 struct syn_cache *sc2; 3390 int s; 3391 3392 /* 3393 * If there are no entries in the hash table, reinitialize 3394 * the hash secrets. 3395 */ 3396 if (syn_cache_count == 0) { 3397 syn_hash1 = arc4random(); 3398 syn_hash2 = arc4random(); 3399 } 3400 3401 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3402 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3403 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3404 3405 /* 3406 * Make sure that we don't overflow the per-bucket 3407 * limit or the total cache size limit. 3408 */ 3409 s = splsoftnet(); 3410 if (scp->sch_length >= tcp_syn_bucket_limit) { 3411 TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW); 3412 /* 3413 * The bucket is full. Toss the oldest element in the 3414 * bucket. This will be the first entry in the bucket. 3415 */ 3416 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3417 #ifdef DIAGNOSTIC 3418 /* 3419 * This should never happen; we should always find an 3420 * entry in our bucket. 3421 */ 3422 if (sc2 == NULL) 3423 panic("syn_cache_insert: bucketoverflow: impossible"); 3424 #endif 3425 syn_cache_rm(sc2); 3426 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3427 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3428 struct syn_cache_head *scp2, *sce; 3429 3430 TCP_STATINC(TCP_STAT_SC_OVERFLOWED); 3431 /* 3432 * The cache is full. Toss the oldest entry in the 3433 * first non-empty bucket we can find. 3434 * 3435 * XXX We would really like to toss the oldest 3436 * entry in the cache, but we hope that this 3437 * condition doesn't happen very often. 3438 */ 3439 scp2 = scp; 3440 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3441 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3442 for (++scp2; scp2 != scp; scp2++) { 3443 if (scp2 >= sce) 3444 scp2 = &tcp_syn_cache[0]; 3445 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3446 break; 3447 } 3448 #ifdef DIAGNOSTIC 3449 /* 3450 * This should never happen; we should always find a 3451 * non-empty bucket. 3452 */ 3453 if (scp2 == scp) 3454 panic("syn_cache_insert: cacheoverflow: " 3455 "impossible"); 3456 #endif 3457 } 3458 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3459 syn_cache_rm(sc2); 3460 syn_cache_put(sc2); /* calls pool_put but see spl above */ 3461 } 3462 3463 /* 3464 * Initialize the entry's timer. 3465 */ 3466 sc->sc_rxttot = 0; 3467 sc->sc_rxtshift = 0; 3468 SYN_CACHE_TIMER_ARM(sc); 3469 3470 /* Link it from tcpcb entry */ 3471 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3472 3473 /* Put it into the bucket. */ 3474 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3475 scp->sch_length++; 3476 syn_cache_count++; 3477 3478 TCP_STATINC(TCP_STAT_SC_ADDED); 3479 splx(s); 3480 } 3481 3482 /* 3483 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3484 * If we have retransmitted an entry the maximum number of times, expire 3485 * that entry. 3486 */ 3487 void 3488 syn_cache_timer(void *arg) 3489 { 3490 struct syn_cache *sc = arg; 3491 3492 mutex_enter(softnet_lock); 3493 KERNEL_LOCK(1, NULL); 3494 callout_ack(&sc->sc_timer); 3495 3496 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3497 TCP_STATINC(TCP_STAT_SC_DELAYED_FREE); 3498 callout_destroy(&sc->sc_timer); 3499 pool_put(&syn_cache_pool, sc); 3500 KERNEL_UNLOCK_ONE(NULL); 3501 mutex_exit(softnet_lock); 3502 return; 3503 } 3504 3505 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3506 /* Drop it -- too many retransmissions. */ 3507 goto dropit; 3508 } 3509 3510 /* 3511 * Compute the total amount of time this entry has 3512 * been on a queue. If this entry has been on longer 3513 * than the keep alive timer would allow, expire it. 3514 */ 3515 sc->sc_rxttot += sc->sc_rxtcur; 3516 if (sc->sc_rxttot >= tcp_keepinit) 3517 goto dropit; 3518 3519 TCP_STATINC(TCP_STAT_SC_RETRANSMITTED); 3520 (void) syn_cache_respond(sc, NULL); 3521 3522 /* Advance the timer back-off. */ 3523 sc->sc_rxtshift++; 3524 SYN_CACHE_TIMER_ARM(sc); 3525 3526 KERNEL_UNLOCK_ONE(NULL); 3527 mutex_exit(softnet_lock); 3528 return; 3529 3530 dropit: 3531 TCP_STATINC(TCP_STAT_SC_TIMED_OUT); 3532 syn_cache_rm(sc); 3533 if (sc->sc_ipopts) 3534 (void) m_free(sc->sc_ipopts); 3535 rtcache_free(&sc->sc_route); 3536 callout_destroy(&sc->sc_timer); 3537 pool_put(&syn_cache_pool, sc); 3538 KERNEL_UNLOCK_ONE(NULL); 3539 mutex_exit(softnet_lock); 3540 } 3541 3542 /* 3543 * Remove syn cache created by the specified tcb entry, 3544 * because this does not make sense to keep them 3545 * (if there's no tcb entry, syn cache entry will never be used) 3546 */ 3547 void 3548 syn_cache_cleanup(struct tcpcb *tp) 3549 { 3550 struct syn_cache *sc, *nsc; 3551 int s; 3552 3553 s = splsoftnet(); 3554 3555 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3556 nsc = LIST_NEXT(sc, sc_tpq); 3557 3558 #ifdef DIAGNOSTIC 3559 if (sc->sc_tp != tp) 3560 panic("invalid sc_tp in syn_cache_cleanup"); 3561 #endif 3562 syn_cache_rm(sc); 3563 syn_cache_put(sc); /* calls pool_put but see spl above */ 3564 } 3565 /* just for safety */ 3566 LIST_INIT(&tp->t_sc); 3567 3568 splx(s); 3569 } 3570 3571 /* 3572 * Find an entry in the syn cache. 3573 */ 3574 struct syn_cache * 3575 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3576 struct syn_cache_head **headp) 3577 { 3578 struct syn_cache *sc; 3579 struct syn_cache_head *scp; 3580 u_int32_t hash; 3581 int s; 3582 3583 SYN_HASHALL(hash, src, dst); 3584 3585 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3586 *headp = scp; 3587 s = splsoftnet(); 3588 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3589 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3590 if (sc->sc_hash != hash) 3591 continue; 3592 if (!memcmp(&sc->sc_src, src, src->sa_len) && 3593 !memcmp(&sc->sc_dst, dst, dst->sa_len)) { 3594 splx(s); 3595 return (sc); 3596 } 3597 } 3598 splx(s); 3599 return (NULL); 3600 } 3601 3602 /* 3603 * This function gets called when we receive an ACK for a 3604 * socket in the LISTEN state. We look up the connection 3605 * in the syn cache, and if its there, we pull it out of 3606 * the cache and turn it into a full-blown connection in 3607 * the SYN-RECEIVED state. 3608 * 3609 * The return values may not be immediately obvious, and their effects 3610 * can be subtle, so here they are: 3611 * 3612 * NULL SYN was not found in cache; caller should drop the 3613 * packet and send an RST. 3614 * 3615 * -1 We were unable to create the new connection, and are 3616 * aborting it. An ACK,RST is being sent to the peer 3617 * (unless we got screwey sequence numbners; see below), 3618 * because the 3-way handshake has been completed. Caller 3619 * should not free the mbuf, since we may be using it. If 3620 * we are not, we will free it. 3621 * 3622 * Otherwise, the return value is a pointer to the new socket 3623 * associated with the connection. 3624 */ 3625 struct socket * 3626 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3627 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3628 struct socket *so, struct mbuf *m) 3629 { 3630 struct syn_cache *sc; 3631 struct syn_cache_head *scp; 3632 struct inpcb *inp = NULL; 3633 #ifdef INET6 3634 struct in6pcb *in6p = NULL; 3635 #endif 3636 struct tcpcb *tp = 0; 3637 struct mbuf *am; 3638 int s; 3639 struct socket *oso; 3640 3641 s = splsoftnet(); 3642 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3643 splx(s); 3644 return (NULL); 3645 } 3646 3647 /* 3648 * Verify the sequence and ack numbers. Try getting the correct 3649 * response again. 3650 */ 3651 if ((th->th_ack != sc->sc_iss + 1) || 3652 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3653 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3654 (void) syn_cache_respond(sc, m); 3655 splx(s); 3656 return ((struct socket *)(-1)); 3657 } 3658 3659 /* Remove this cache entry */ 3660 syn_cache_rm(sc); 3661 splx(s); 3662 3663 /* 3664 * Ok, create the full blown connection, and set things up 3665 * as they would have been set up if we had created the 3666 * connection when the SYN arrived. If we can't create 3667 * the connection, abort it. 3668 */ 3669 /* 3670 * inp still has the OLD in_pcb stuff, set the 3671 * v6-related flags on the new guy, too. This is 3672 * done particularly for the case where an AF_INET6 3673 * socket is bound only to a port, and a v4 connection 3674 * comes in on that port. 3675 * we also copy the flowinfo from the original pcb 3676 * to the new one. 3677 */ 3678 oso = so; 3679 so = sonewconn(so, SS_ISCONNECTED); 3680 if (so == NULL) 3681 goto resetandabort; 3682 3683 switch (so->so_proto->pr_domain->dom_family) { 3684 #ifdef INET 3685 case AF_INET: 3686 inp = sotoinpcb(so); 3687 break; 3688 #endif 3689 #ifdef INET6 3690 case AF_INET6: 3691 in6p = sotoin6pcb(so); 3692 break; 3693 #endif 3694 } 3695 switch (src->sa_family) { 3696 #ifdef INET 3697 case AF_INET: 3698 if (inp) { 3699 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3700 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3701 inp->inp_options = ip_srcroute(); 3702 in_pcbstate(inp, INP_BOUND); 3703 if (inp->inp_options == NULL) { 3704 inp->inp_options = sc->sc_ipopts; 3705 sc->sc_ipopts = NULL; 3706 } 3707 } 3708 #ifdef INET6 3709 else if (in6p) { 3710 /* IPv4 packet to AF_INET6 socket */ 3711 memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr)); 3712 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3713 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3714 &in6p->in6p_laddr.s6_addr32[3], 3715 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3716 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3717 in6totcpcb(in6p)->t_family = AF_INET; 3718 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3719 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3720 else 3721 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3722 in6_pcbstate(in6p, IN6P_BOUND); 3723 } 3724 #endif 3725 break; 3726 #endif 3727 #ifdef INET6 3728 case AF_INET6: 3729 if (in6p) { 3730 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3731 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3732 in6_pcbstate(in6p, IN6P_BOUND); 3733 } 3734 break; 3735 #endif 3736 } 3737 #ifdef INET6 3738 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3739 struct in6pcb *oin6p = sotoin6pcb(oso); 3740 /* inherit socket options from the listening socket */ 3741 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3742 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3743 m_freem(in6p->in6p_options); 3744 in6p->in6p_options = 0; 3745 } 3746 ip6_savecontrol(in6p, &in6p->in6p_options, 3747 mtod(m, struct ip6_hdr *), m); 3748 } 3749 #endif 3750 3751 #if defined(IPSEC) || defined(FAST_IPSEC) 3752 /* 3753 * we make a copy of policy, instead of sharing the policy, 3754 * for better behavior in terms of SA lookup and dead SA removal. 3755 */ 3756 if (inp) { 3757 /* copy old policy into new socket's */ 3758 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3759 printf("tcp_input: could not copy policy\n"); 3760 } 3761 #ifdef INET6 3762 else if (in6p) { 3763 /* copy old policy into new socket's */ 3764 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3765 in6p->in6p_sp)) 3766 printf("tcp_input: could not copy policy\n"); 3767 } 3768 #endif 3769 #endif 3770 3771 /* 3772 * Give the new socket our cached route reference. 3773 */ 3774 if (inp) { 3775 rtcache_copy(&inp->inp_route, &sc->sc_route); 3776 rtcache_free(&sc->sc_route); 3777 } 3778 #ifdef INET6 3779 else { 3780 rtcache_copy(&in6p->in6p_route, &sc->sc_route); 3781 rtcache_free(&sc->sc_route); 3782 } 3783 #endif 3784 3785 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3786 if (am == NULL) 3787 goto resetandabort; 3788 MCLAIM(am, &tcp_mowner); 3789 am->m_len = src->sa_len; 3790 bcopy(src, mtod(am, void *), src->sa_len); 3791 if (inp) { 3792 if (in_pcbconnect(inp, am, &lwp0)) { 3793 (void) m_free(am); 3794 goto resetandabort; 3795 } 3796 } 3797 #ifdef INET6 3798 else if (in6p) { 3799 if (src->sa_family == AF_INET) { 3800 /* IPv4 packet to AF_INET6 socket */ 3801 struct sockaddr_in6 *sin6; 3802 sin6 = mtod(am, struct sockaddr_in6 *); 3803 am->m_len = sizeof(*sin6); 3804 memset(sin6, 0, sizeof(*sin6)); 3805 sin6->sin6_family = AF_INET6; 3806 sin6->sin6_len = sizeof(*sin6); 3807 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3808 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3809 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3810 &sin6->sin6_addr.s6_addr32[3], 3811 sizeof(sin6->sin6_addr.s6_addr32[3])); 3812 } 3813 if (in6_pcbconnect(in6p, am, NULL)) { 3814 (void) m_free(am); 3815 goto resetandabort; 3816 } 3817 } 3818 #endif 3819 else { 3820 (void) m_free(am); 3821 goto resetandabort; 3822 } 3823 (void) m_free(am); 3824 3825 if (inp) 3826 tp = intotcpcb(inp); 3827 #ifdef INET6 3828 else if (in6p) 3829 tp = in6totcpcb(in6p); 3830 #endif 3831 else 3832 tp = NULL; 3833 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3834 if (sc->sc_request_r_scale != 15) { 3835 tp->requested_s_scale = sc->sc_requested_s_scale; 3836 tp->request_r_scale = sc->sc_request_r_scale; 3837 tp->snd_scale = sc->sc_requested_s_scale; 3838 tp->rcv_scale = sc->sc_request_r_scale; 3839 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3840 } 3841 if (sc->sc_flags & SCF_TIMESTAMP) 3842 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3843 tp->ts_timebase = sc->sc_timebase; 3844 3845 tp->t_template = tcp_template(tp); 3846 if (tp->t_template == 0) { 3847 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3848 so = NULL; 3849 m_freem(m); 3850 goto abort; 3851 } 3852 3853 tp->iss = sc->sc_iss; 3854 tp->irs = sc->sc_irs; 3855 tcp_sendseqinit(tp); 3856 tcp_rcvseqinit(tp); 3857 tp->t_state = TCPS_SYN_RECEIVED; 3858 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit); 3859 TCP_STATINC(TCP_STAT_ACCEPTS); 3860 3861 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3862 tp->t_flags |= TF_WILL_SACK; 3863 3864 if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn) 3865 tp->t_flags |= TF_ECN_PERMIT; 3866 3867 #ifdef TCP_SIGNATURE 3868 if (sc->sc_flags & SCF_SIGNATURE) 3869 tp->t_flags |= TF_SIGNATURE; 3870 #endif 3871 3872 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3873 tp->t_ourmss = sc->sc_ourmaxseg; 3874 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3875 3876 /* 3877 * Initialize the initial congestion window. If we 3878 * had to retransmit the SYN,ACK, we must initialize cwnd 3879 * to 1 segment (i.e. the Loss Window). 3880 */ 3881 if (sc->sc_rxtshift) 3882 tp->snd_cwnd = tp->t_peermss; 3883 else { 3884 int ss = tcp_init_win; 3885 #ifdef INET 3886 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3887 ss = tcp_init_win_local; 3888 #endif 3889 #ifdef INET6 3890 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3891 ss = tcp_init_win_local; 3892 #endif 3893 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3894 } 3895 3896 tcp_rmx_rtt(tp); 3897 tp->snd_wl1 = sc->sc_irs; 3898 tp->rcv_up = sc->sc_irs + 1; 3899 3900 /* 3901 * This is what whould have happened in tcp_output() when 3902 * the SYN,ACK was sent. 3903 */ 3904 tp->snd_up = tp->snd_una; 3905 tp->snd_max = tp->snd_nxt = tp->iss+1; 3906 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3907 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3908 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3909 tp->last_ack_sent = tp->rcv_nxt; 3910 tp->t_partialacks = -1; 3911 tp->t_dupacks = 0; 3912 3913 TCP_STATINC(TCP_STAT_SC_COMPLETED); 3914 s = splsoftnet(); 3915 syn_cache_put(sc); 3916 splx(s); 3917 return (so); 3918 3919 resetandabort: 3920 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3921 abort: 3922 if (so != NULL) { 3923 (void) soqremque(so, 1); 3924 (void) soabort(so); 3925 mutex_enter(softnet_lock); 3926 } 3927 s = splsoftnet(); 3928 syn_cache_put(sc); 3929 splx(s); 3930 TCP_STATINC(TCP_STAT_SC_ABORTED); 3931 return ((struct socket *)(-1)); 3932 } 3933 3934 /* 3935 * This function is called when we get a RST for a 3936 * non-existent connection, so that we can see if the 3937 * connection is in the syn cache. If it is, zap it. 3938 */ 3939 3940 void 3941 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3942 { 3943 struct syn_cache *sc; 3944 struct syn_cache_head *scp; 3945 int s = splsoftnet(); 3946 3947 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3948 splx(s); 3949 return; 3950 } 3951 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3952 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3953 splx(s); 3954 return; 3955 } 3956 syn_cache_rm(sc); 3957 TCP_STATINC(TCP_STAT_SC_RESET); 3958 syn_cache_put(sc); /* calls pool_put but see spl above */ 3959 splx(s); 3960 } 3961 3962 void 3963 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3964 struct tcphdr *th) 3965 { 3966 struct syn_cache *sc; 3967 struct syn_cache_head *scp; 3968 int s; 3969 3970 s = splsoftnet(); 3971 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3972 splx(s); 3973 return; 3974 } 3975 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3976 if (ntohl (th->th_seq) != sc->sc_iss) { 3977 splx(s); 3978 return; 3979 } 3980 3981 /* 3982 * If we've retransmitted 3 times and this is our second error, 3983 * we remove the entry. Otherwise, we allow it to continue on. 3984 * This prevents us from incorrectly nuking an entry during a 3985 * spurious network outage. 3986 * 3987 * See tcp_notify(). 3988 */ 3989 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3990 sc->sc_flags |= SCF_UNREACH; 3991 splx(s); 3992 return; 3993 } 3994 3995 syn_cache_rm(sc); 3996 TCP_STATINC(TCP_STAT_SC_UNREACH); 3997 syn_cache_put(sc); /* calls pool_put but see spl above */ 3998 splx(s); 3999 } 4000 4001 /* 4002 * Given a LISTEN socket and an inbound SYN request, add 4003 * this to the syn cache, and send back a segment: 4004 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 4005 * to the source. 4006 * 4007 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 4008 * Doing so would require that we hold onto the data and deliver it 4009 * to the application. However, if we are the target of a SYN-flood 4010 * DoS attack, an attacker could send data which would eventually 4011 * consume all available buffer space if it were ACKed. By not ACKing 4012 * the data, we avoid this DoS scenario. 4013 */ 4014 4015 int 4016 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 4017 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 4018 int optlen, struct tcp_opt_info *oi) 4019 { 4020 struct tcpcb tb, *tp; 4021 long win; 4022 struct syn_cache *sc; 4023 struct syn_cache_head *scp; 4024 struct mbuf *ipopts; 4025 struct tcp_opt_info opti; 4026 int s; 4027 4028 tp = sototcpcb(so); 4029 4030 memset(&opti, 0, sizeof(opti)); 4031 4032 /* 4033 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 4034 * 4035 * Note this check is performed in tcp_input() very early on. 4036 */ 4037 4038 /* 4039 * Initialize some local state. 4040 */ 4041 win = sbspace(&so->so_rcv); 4042 if (win > TCP_MAXWIN) 4043 win = TCP_MAXWIN; 4044 4045 switch (src->sa_family) { 4046 #ifdef INET 4047 case AF_INET: 4048 /* 4049 * Remember the IP options, if any. 4050 */ 4051 ipopts = ip_srcroute(); 4052 break; 4053 #endif 4054 default: 4055 ipopts = NULL; 4056 } 4057 4058 #ifdef TCP_SIGNATURE 4059 if (optp || (tp->t_flags & TF_SIGNATURE)) 4060 #else 4061 if (optp) 4062 #endif 4063 { 4064 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 4065 #ifdef TCP_SIGNATURE 4066 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 4067 #endif 4068 tb.t_state = TCPS_LISTEN; 4069 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 4070 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 4071 return (0); 4072 } else 4073 tb.t_flags = 0; 4074 4075 /* 4076 * See if we already have an entry for this connection. 4077 * If we do, resend the SYN,ACK. We do not count this 4078 * as a retransmission (XXX though maybe we should). 4079 */ 4080 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 4081 TCP_STATINC(TCP_STAT_SC_DUPESYN); 4082 if (ipopts) { 4083 /* 4084 * If we were remembering a previous source route, 4085 * forget it and use the new one we've been given. 4086 */ 4087 if (sc->sc_ipopts) 4088 (void) m_free(sc->sc_ipopts); 4089 sc->sc_ipopts = ipopts; 4090 } 4091 sc->sc_timestamp = tb.ts_recent; 4092 if (syn_cache_respond(sc, m) == 0) { 4093 uint64_t *tcps = TCP_STAT_GETREF(); 4094 tcps[TCP_STAT_SNDACKS]++; 4095 tcps[TCP_STAT_SNDTOTAL]++; 4096 TCP_STAT_PUTREF(); 4097 } 4098 return (1); 4099 } 4100 4101 s = splsoftnet(); 4102 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 4103 splx(s); 4104 if (sc == NULL) { 4105 if (ipopts) 4106 (void) m_free(ipopts); 4107 return (0); 4108 } 4109 4110 /* 4111 * Fill in the cache, and put the necessary IP and TCP 4112 * options into the reply. 4113 */ 4114 memset(sc, 0, sizeof(struct syn_cache)); 4115 callout_init(&sc->sc_timer, CALLOUT_MPSAFE); 4116 bcopy(src, &sc->sc_src, src->sa_len); 4117 bcopy(dst, &sc->sc_dst, dst->sa_len); 4118 sc->sc_flags = 0; 4119 sc->sc_ipopts = ipopts; 4120 sc->sc_irs = th->th_seq; 4121 switch (src->sa_family) { 4122 #ifdef INET 4123 case AF_INET: 4124 { 4125 struct sockaddr_in *srcin = (void *) src; 4126 struct sockaddr_in *dstin = (void *) dst; 4127 4128 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 4129 &srcin->sin_addr, dstin->sin_port, 4130 srcin->sin_port, sizeof(dstin->sin_addr), 0); 4131 break; 4132 } 4133 #endif /* INET */ 4134 #ifdef INET6 4135 case AF_INET6: 4136 { 4137 struct sockaddr_in6 *srcin6 = (void *) src; 4138 struct sockaddr_in6 *dstin6 = (void *) dst; 4139 4140 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 4141 &srcin6->sin6_addr, dstin6->sin6_port, 4142 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 4143 break; 4144 } 4145 #endif /* INET6 */ 4146 } 4147 sc->sc_peermaxseg = oi->maxseg; 4148 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 4149 m->m_pkthdr.rcvif : NULL, 4150 sc->sc_src.sa.sa_family); 4151 sc->sc_win = win; 4152 sc->sc_timebase = tcp_now - 1; /* see tcp_newtcpcb() */ 4153 sc->sc_timestamp = tb.ts_recent; 4154 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4155 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4156 sc->sc_flags |= SCF_TIMESTAMP; 4157 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4158 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4159 sc->sc_requested_s_scale = tb.requested_s_scale; 4160 sc->sc_request_r_scale = 0; 4161 /* 4162 * Pick the smallest possible scaling factor that 4163 * will still allow us to scale up to sb_max. 4164 * 4165 * We do this because there are broken firewalls that 4166 * will corrupt the window scale option, leading to 4167 * the other endpoint believing that our advertised 4168 * window is unscaled. At scale factors larger than 4169 * 5 the unscaled window will drop below 1500 bytes, 4170 * leading to serious problems when traversing these 4171 * broken firewalls. 4172 * 4173 * With the default sbmax of 256K, a scale factor 4174 * of 3 will be chosen by this algorithm. Those who 4175 * choose a larger sbmax should watch out 4176 * for the compatiblity problems mentioned above. 4177 * 4178 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4179 * or <SYN,ACK>) segment itself is never scaled. 4180 */ 4181 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4182 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4183 sc->sc_request_r_scale++; 4184 } else { 4185 sc->sc_requested_s_scale = 15; 4186 sc->sc_request_r_scale = 15; 4187 } 4188 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4189 sc->sc_flags |= SCF_SACK_PERMIT; 4190 4191 /* 4192 * ECN setup packet recieved. 4193 */ 4194 if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn) 4195 sc->sc_flags |= SCF_ECN_PERMIT; 4196 4197 #ifdef TCP_SIGNATURE 4198 if (tb.t_flags & TF_SIGNATURE) 4199 sc->sc_flags |= SCF_SIGNATURE; 4200 #endif 4201 sc->sc_tp = tp; 4202 if (syn_cache_respond(sc, m) == 0) { 4203 uint64_t *tcps = TCP_STAT_GETREF(); 4204 tcps[TCP_STAT_SNDACKS]++; 4205 tcps[TCP_STAT_SNDTOTAL]++; 4206 TCP_STAT_PUTREF(); 4207 syn_cache_insert(sc, tp); 4208 } else { 4209 s = splsoftnet(); 4210 /* 4211 * syn_cache_put() will try to schedule the timer, so 4212 * we need to initialize it 4213 */ 4214 SYN_CACHE_TIMER_ARM(sc); 4215 syn_cache_put(sc); 4216 splx(s); 4217 TCP_STATINC(TCP_STAT_SC_DROPPED); 4218 } 4219 return (1); 4220 } 4221 4222 int 4223 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4224 { 4225 #ifdef INET6 4226 struct rtentry *rt; 4227 #endif 4228 struct route *ro; 4229 u_int8_t *optp; 4230 int optlen, error; 4231 u_int16_t tlen; 4232 struct ip *ip = NULL; 4233 #ifdef INET6 4234 struct ip6_hdr *ip6 = NULL; 4235 #endif 4236 struct tcpcb *tp = NULL; 4237 struct tcphdr *th; 4238 u_int hlen; 4239 struct socket *so; 4240 4241 ro = &sc->sc_route; 4242 switch (sc->sc_src.sa.sa_family) { 4243 case AF_INET: 4244 hlen = sizeof(struct ip); 4245 break; 4246 #ifdef INET6 4247 case AF_INET6: 4248 hlen = sizeof(struct ip6_hdr); 4249 break; 4250 #endif 4251 default: 4252 if (m) 4253 m_freem(m); 4254 return (EAFNOSUPPORT); 4255 } 4256 4257 /* Compute the size of the TCP options. */ 4258 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4259 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4260 #ifdef TCP_SIGNATURE 4261 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4262 #endif 4263 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4264 4265 tlen = hlen + sizeof(struct tcphdr) + optlen; 4266 4267 /* 4268 * Create the IP+TCP header from scratch. 4269 */ 4270 if (m) 4271 m_freem(m); 4272 #ifdef DIAGNOSTIC 4273 if (max_linkhdr + tlen > MCLBYTES) 4274 return (ENOBUFS); 4275 #endif 4276 MGETHDR(m, M_DONTWAIT, MT_DATA); 4277 if (m && (max_linkhdr + tlen) > MHLEN) { 4278 MCLGET(m, M_DONTWAIT); 4279 if ((m->m_flags & M_EXT) == 0) { 4280 m_freem(m); 4281 m = NULL; 4282 } 4283 } 4284 if (m == NULL) 4285 return (ENOBUFS); 4286 MCLAIM(m, &tcp_tx_mowner); 4287 4288 /* Fixup the mbuf. */ 4289 m->m_data += max_linkhdr; 4290 m->m_len = m->m_pkthdr.len = tlen; 4291 if (sc->sc_tp) { 4292 tp = sc->sc_tp; 4293 if (tp->t_inpcb) 4294 so = tp->t_inpcb->inp_socket; 4295 #ifdef INET6 4296 else if (tp->t_in6pcb) 4297 so = tp->t_in6pcb->in6p_socket; 4298 #endif 4299 else 4300 so = NULL; 4301 } else 4302 so = NULL; 4303 m->m_pkthdr.rcvif = NULL; 4304 memset(mtod(m, u_char *), 0, tlen); 4305 4306 switch (sc->sc_src.sa.sa_family) { 4307 case AF_INET: 4308 ip = mtod(m, struct ip *); 4309 ip->ip_v = 4; 4310 ip->ip_dst = sc->sc_src.sin.sin_addr; 4311 ip->ip_src = sc->sc_dst.sin.sin_addr; 4312 ip->ip_p = IPPROTO_TCP; 4313 th = (struct tcphdr *)(ip + 1); 4314 th->th_dport = sc->sc_src.sin.sin_port; 4315 th->th_sport = sc->sc_dst.sin.sin_port; 4316 break; 4317 #ifdef INET6 4318 case AF_INET6: 4319 ip6 = mtod(m, struct ip6_hdr *); 4320 ip6->ip6_vfc = IPV6_VERSION; 4321 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4322 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4323 ip6->ip6_nxt = IPPROTO_TCP; 4324 /* ip6_plen will be updated in ip6_output() */ 4325 th = (struct tcphdr *)(ip6 + 1); 4326 th->th_dport = sc->sc_src.sin6.sin6_port; 4327 th->th_sport = sc->sc_dst.sin6.sin6_port; 4328 break; 4329 #endif 4330 default: 4331 th = NULL; 4332 } 4333 4334 th->th_seq = htonl(sc->sc_iss); 4335 th->th_ack = htonl(sc->sc_irs + 1); 4336 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4337 th->th_flags = TH_SYN|TH_ACK; 4338 th->th_win = htons(sc->sc_win); 4339 /* th_sum already 0 */ 4340 /* th_urp already 0 */ 4341 4342 /* Tack on the TCP options. */ 4343 optp = (u_int8_t *)(th + 1); 4344 *optp++ = TCPOPT_MAXSEG; 4345 *optp++ = 4; 4346 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4347 *optp++ = sc->sc_ourmaxseg & 0xff; 4348 4349 if (sc->sc_request_r_scale != 15) { 4350 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4351 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4352 sc->sc_request_r_scale); 4353 optp += 4; 4354 } 4355 4356 if (sc->sc_flags & SCF_TIMESTAMP) { 4357 u_int32_t *lp = (u_int32_t *)(optp); 4358 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4359 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4360 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4361 *lp = htonl(sc->sc_timestamp); 4362 optp += TCPOLEN_TSTAMP_APPA; 4363 } 4364 4365 if (sc->sc_flags & SCF_SACK_PERMIT) { 4366 u_int8_t *p = optp; 4367 4368 /* Let the peer know that we will SACK. */ 4369 p[0] = TCPOPT_SACK_PERMITTED; 4370 p[1] = 2; 4371 p[2] = TCPOPT_NOP; 4372 p[3] = TCPOPT_NOP; 4373 optp += 4; 4374 } 4375 4376 /* 4377 * Send ECN SYN-ACK setup packet. 4378 * Routes can be asymetric, so, even if we receive a packet 4379 * with ECE and CWR set, we must not assume no one will block 4380 * the ECE packet we are about to send. 4381 */ 4382 if ((sc->sc_flags & SCF_ECN_PERMIT) && tp && 4383 SEQ_GEQ(tp->snd_nxt, tp->snd_max)) { 4384 th->th_flags |= TH_ECE; 4385 TCP_STATINC(TCP_STAT_ECN_SHS); 4386 4387 /* 4388 * draft-ietf-tcpm-ecnsyn-00.txt 4389 * 4390 * "[...] a TCP node MAY respond to an ECN-setup 4391 * SYN packet by setting ECT in the responding 4392 * ECN-setup SYN/ACK packet, indicating to routers 4393 * that the SYN/ACK packet is ECN-Capable. 4394 * This allows a congested router along the path 4395 * to mark the packet instead of dropping the 4396 * packet as an indication of congestion." 4397 * 4398 * "[...] There can be a great benefit in setting 4399 * an ECN-capable codepoint in SYN/ACK packets [...] 4400 * Congestion is most likely to occur in 4401 * the server-to-client direction. As a result, 4402 * setting an ECN-capable codepoint in SYN/ACK 4403 * packets can reduce the occurence of three-second 4404 * retransmit timeouts resulting from the drop 4405 * of SYN/ACK packets." 4406 * 4407 * Page 4 and 6, January 2006. 4408 */ 4409 4410 switch (sc->sc_src.sa.sa_family) { 4411 #ifdef INET 4412 case AF_INET: 4413 ip->ip_tos |= IPTOS_ECN_ECT0; 4414 break; 4415 #endif 4416 #ifdef INET6 4417 case AF_INET6: 4418 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); 4419 break; 4420 #endif 4421 } 4422 TCP_STATINC(TCP_STAT_ECN_ECT); 4423 } 4424 4425 #ifdef TCP_SIGNATURE 4426 if (sc->sc_flags & SCF_SIGNATURE) { 4427 struct secasvar *sav; 4428 u_int8_t *sigp; 4429 4430 sav = tcp_signature_getsav(m, th); 4431 4432 if (sav == NULL) { 4433 if (m) 4434 m_freem(m); 4435 return (EPERM); 4436 } 4437 4438 *optp++ = TCPOPT_SIGNATURE; 4439 *optp++ = TCPOLEN_SIGNATURE; 4440 sigp = optp; 4441 memset(optp, 0, TCP_SIGLEN); 4442 optp += TCP_SIGLEN; 4443 *optp++ = TCPOPT_NOP; 4444 *optp++ = TCPOPT_EOL; 4445 4446 (void)tcp_signature(m, th, hlen, sav, sigp); 4447 4448 key_sa_recordxfer(sav, m); 4449 #ifdef FAST_IPSEC 4450 KEY_FREESAV(&sav); 4451 #else 4452 key_freesav(sav); 4453 #endif 4454 } 4455 #endif 4456 4457 /* Compute the packet's checksum. */ 4458 switch (sc->sc_src.sa.sa_family) { 4459 case AF_INET: 4460 ip->ip_len = htons(tlen - hlen); 4461 th->th_sum = 0; 4462 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4463 break; 4464 #ifdef INET6 4465 case AF_INET6: 4466 ip6->ip6_plen = htons(tlen - hlen); 4467 th->th_sum = 0; 4468 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4469 break; 4470 #endif 4471 } 4472 4473 /* 4474 * Fill in some straggling IP bits. Note the stack expects 4475 * ip_len to be in host order, for convenience. 4476 */ 4477 switch (sc->sc_src.sa.sa_family) { 4478 #ifdef INET 4479 case AF_INET: 4480 ip->ip_len = htons(tlen); 4481 ip->ip_ttl = ip_defttl; 4482 /* XXX tos? */ 4483 break; 4484 #endif 4485 #ifdef INET6 4486 case AF_INET6: 4487 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4488 ip6->ip6_vfc |= IPV6_VERSION; 4489 ip6->ip6_plen = htons(tlen - hlen); 4490 /* ip6_hlim will be initialized afterwards */ 4491 /* XXX flowlabel? */ 4492 break; 4493 #endif 4494 } 4495 4496 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4497 tp = sc->sc_tp; 4498 4499 switch (sc->sc_src.sa.sa_family) { 4500 #ifdef INET 4501 case AF_INET: 4502 error = ip_output(m, sc->sc_ipopts, ro, 4503 (ip_mtudisc ? IP_MTUDISC : 0), 4504 (struct ip_moptions *)NULL, so); 4505 break; 4506 #endif 4507 #ifdef INET6 4508 case AF_INET6: 4509 ip6->ip6_hlim = in6_selecthlim(NULL, 4510 (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp 4511 : NULL); 4512 4513 error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL); 4514 break; 4515 #endif 4516 default: 4517 error = EAFNOSUPPORT; 4518 break; 4519 } 4520 return (error); 4521 } 4522