tcp_input.c revision 1.103
1 /* $NetBSD: tcp_input.c,v 1.103 2000/02/12 17:19:34 thorpej Exp $ */ 2 3 /* 4 %%% portions-copyright-nrl-95 5 Portions of this software are Copyright 1995-1998 by Randall Atkinson, 6 Ronald Lee, Daniel McDonald, Bao Phan, and Chris Winters. All Rights 7 Reserved. All rights under this copyright have been assigned to the US 8 Naval Research Laboratory (NRL). The NRL Copyright Notice and License 9 Agreement Version 1.1 (January 17, 1995) applies to these portions of the 10 software. 11 You should have received a copy of the license with this software. If you 12 didn't get a copy, you may request one from <license (at) ipv6.nrl.navy.mil>. 13 14 */ 15 16 /* 17 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 18 * All rights reserved. 19 * 20 * Redistribution and use in source and binary forms, with or without 21 * modification, are permitted provided that the following conditions 22 * are met: 23 * 1. Redistributions of source code must retain the above copyright 24 * notice, this list of conditions and the following disclaimer. 25 * 2. Redistributions in binary form must reproduce the above copyright 26 * notice, this list of conditions and the following disclaimer in the 27 * documentation and/or other materials provided with the distribution. 28 * 3. Neither the name of the project nor the names of its contributors 29 * may be used to endorse or promote products derived from this software 30 * without specific prior written permission. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 33 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 34 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 35 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 36 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 37 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 38 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 39 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 40 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 41 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 42 * SUCH DAMAGE. 43 */ 44 45 /*- 46 * Copyright (c) 1997, 1998, 1999 The NetBSD Foundation, Inc. 47 * All rights reserved. 48 * 49 * This code is derived from software contributed to The NetBSD Foundation 50 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 51 * Facility, NASA Ames Research Center. 52 * 53 * Redistribution and use in source and binary forms, with or without 54 * modification, are permitted provided that the following conditions 55 * are met: 56 * 1. Redistributions of source code must retain the above copyright 57 * notice, this list of conditions and the following disclaimer. 58 * 2. Redistributions in binary form must reproduce the above copyright 59 * notice, this list of conditions and the following disclaimer in the 60 * documentation and/or other materials provided with the distribution. 61 * 3. All advertising materials mentioning features or use of this software 62 * must display the following acknowledgement: 63 * This product includes software developed by the NetBSD 64 * Foundation, Inc. and its contributors. 65 * 4. Neither the name of The NetBSD Foundation nor the names of its 66 * contributors may be used to endorse or promote products derived 67 * from this software without specific prior written permission. 68 * 69 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 70 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 71 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 72 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 73 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 74 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 75 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 76 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 77 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 78 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 79 * POSSIBILITY OF SUCH DAMAGE. 80 */ 81 82 /* 83 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 84 * The Regents of the University of California. All rights reserved. 85 * 86 * Redistribution and use in source and binary forms, with or without 87 * modification, are permitted provided that the following conditions 88 * are met: 89 * 1. Redistributions of source code must retain the above copyright 90 * notice, this list of conditions and the following disclaimer. 91 * 2. Redistributions in binary form must reproduce the above copyright 92 * notice, this list of conditions and the following disclaimer in the 93 * documentation and/or other materials provided with the distribution. 94 * 3. All advertising materials mentioning features or use of this software 95 * must display the following acknowledgement: 96 * This product includes software developed by the University of 97 * California, Berkeley and its contributors. 98 * 4. Neither the name of the University nor the names of its contributors 99 * may be used to endorse or promote products derived from this software 100 * without specific prior written permission. 101 * 102 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 103 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 104 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 105 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 106 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 107 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 108 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 109 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 110 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 111 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 112 * SUCH DAMAGE. 113 * 114 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 115 */ 116 117 /* 118 * TODO list for SYN cache stuff: 119 * 120 * Find room for a "state" field, which is needed to keep a 121 * compressed state for TIME_WAIT TCBs. It's been noted already 122 * that this is fairly important for very high-volume web and 123 * mail servers, which use a large number of short-lived 124 * connections. 125 */ 126 127 #include "opt_inet.h" 128 #include "opt_ipsec.h" 129 130 #include <sys/param.h> 131 #include <sys/systm.h> 132 #include <sys/malloc.h> 133 #include <sys/mbuf.h> 134 #include <sys/protosw.h> 135 #include <sys/socket.h> 136 #include <sys/socketvar.h> 137 #include <sys/errno.h> 138 #include <sys/syslog.h> 139 #include <sys/pool.h> 140 #include <sys/domain.h> 141 142 #include <net/if.h> 143 #include <net/route.h> 144 #include <net/if_types.h> 145 146 #include <netinet/in.h> 147 #include <netinet/in_systm.h> 148 #include <netinet/ip.h> 149 #include <netinet/in_pcb.h> 150 #include <netinet/ip_var.h> 151 152 #ifdef INET6 153 #ifndef INET 154 #include <netinet/in.h> 155 #endif 156 #include <netinet/ip6.h> 157 #include <netinet6/in6_pcb.h> 158 #include <netinet6/ip6_var.h> 159 #include <netinet6/in6_var.h> 160 #include <netinet/icmp6.h> 161 #include <netinet6/nd6.h> 162 #endif 163 164 #ifdef PULLDOWN_TEST 165 #ifndef INET6 166 /* always need ip6.h for IP6_EXTHDR_GET */ 167 #include <netinet/ip6.h> 168 #endif 169 #endif 170 171 #include <netinet/tcp.h> 172 #include <netinet/tcp_fsm.h> 173 #include <netinet/tcp_seq.h> 174 #include <netinet/tcp_timer.h> 175 #include <netinet/tcp_var.h> 176 #include <netinet/tcpip.h> 177 #include <netinet/tcp_debug.h> 178 179 #include <machine/stdarg.h> 180 181 #ifdef IPSEC 182 #include <netinet6/ipsec.h> 183 #include <netkey/key.h> 184 #include <netkey/key_debug.h> 185 #endif /*IPSEC*/ 186 #ifdef INET6 187 #include "faith.h" 188 #endif 189 190 int tcprexmtthresh = 3; 191 int tcp_log_refused; 192 193 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 194 195 /* for modulo comparisons of timestamps */ 196 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 197 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 198 199 /* 200 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 201 */ 202 #ifdef INET6 203 #define ND6_HINT(tp) \ 204 do { \ 205 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 206 && tp->t_in6pcb->in6p_route.ro_rt) { \ 207 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL); \ 208 } \ 209 } while (0) 210 #else 211 #define ND6_HINT(tp) 212 #endif 213 214 /* 215 * Macro to compute ACK transmission behavior. Delay the ACK unless 216 * we have already delayed an ACK (must send an ACK every two segments). 217 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 218 * option is enabled. 219 */ 220 #define TCP_SETUP_ACK(tp, th) \ 221 do { \ 222 if ((tp)->t_flags & TF_DELACK || \ 223 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 224 tp->t_flags |= TF_ACKNOW; \ 225 else \ 226 TCP_SET_DELACK(tp); \ 227 } while (0) 228 229 /* 230 * Convert TCP protocol fields to host order for easier processing. 231 */ 232 #define TCP_FIELDS_TO_HOST(th) \ 233 do { \ 234 NTOHL((th)->th_seq); \ 235 NTOHL((th)->th_ack); \ 236 NTOHS((th)->th_win); \ 237 NTOHS((th)->th_urp); \ 238 } while (0) 239 240 int 241 tcp_reass(tp, th, m, tlen) 242 register struct tcpcb *tp; 243 register struct tcphdr *th; 244 struct mbuf *m; 245 int *tlen; 246 { 247 register struct ipqent *p, *q, *nq, *tiqe = NULL; 248 struct socket *so = NULL; 249 int pkt_flags; 250 tcp_seq pkt_seq; 251 unsigned pkt_len; 252 u_long rcvpartdupbyte = 0; 253 u_long rcvoobyte; 254 255 if (tp->t_inpcb) 256 so = tp->t_inpcb->inp_socket; 257 #ifdef INET6 258 else if (tp->t_in6pcb) 259 so = tp->t_in6pcb->in6p_socket; 260 #endif 261 262 TCP_REASS_LOCK_CHECK(tp); 263 264 /* 265 * Call with th==0 after become established to 266 * force pre-ESTABLISHED data up to user socket. 267 */ 268 if (th == 0) 269 goto present; 270 271 rcvoobyte = *tlen; 272 /* 273 * Copy these to local variables because the tcpiphdr 274 * gets munged while we are collapsing mbufs. 275 */ 276 pkt_seq = th->th_seq; 277 pkt_len = *tlen; 278 pkt_flags = th->th_flags; 279 /* 280 * Find a segment which begins after this one does. 281 */ 282 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 283 nq = q->ipqe_q.le_next; 284 /* 285 * If the received segment is just right after this 286 * fragment, merge the two together and then check 287 * for further overlaps. 288 */ 289 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 290 #ifdef TCPREASS_DEBUG 291 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 292 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 293 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 294 #endif 295 pkt_len += q->ipqe_len; 296 pkt_flags |= q->ipqe_flags; 297 pkt_seq = q->ipqe_seq; 298 m_cat(q->ipqe_m, m); 299 m = q->ipqe_m; 300 goto free_ipqe; 301 } 302 /* 303 * If the received segment is completely past this 304 * fragment, we need to go the next fragment. 305 */ 306 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 307 p = q; 308 continue; 309 } 310 /* 311 * If the fragment is past the received segment, 312 * it (or any following) can't be concatenated. 313 */ 314 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 315 break; 316 /* 317 * We've received all the data in this segment before. 318 * mark it as a duplicate and return. 319 */ 320 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 321 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 322 tcpstat.tcps_rcvduppack++; 323 tcpstat.tcps_rcvdupbyte += pkt_len; 324 m_freem(m); 325 if (tiqe != NULL) 326 pool_put(&ipqent_pool, tiqe); 327 return (0); 328 } 329 /* 330 * Received segment completely overlaps this fragment 331 * so we drop the fragment (this keeps the temporal 332 * ordering of segments correct). 333 */ 334 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 335 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 336 rcvpartdupbyte += q->ipqe_len; 337 m_freem(q->ipqe_m); 338 goto free_ipqe; 339 } 340 /* 341 * RX'ed segment extends past the end of the 342 * fragment. Drop the overlapping bytes. Then 343 * merge the fragment and segment then treat as 344 * a longer received packet. 345 */ 346 if (SEQ_LT(q->ipqe_seq, pkt_seq) 347 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 348 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 349 #ifdef TCPREASS_DEBUG 350 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 351 tp, overlap, 352 pkt_seq, pkt_seq + pkt_len, pkt_len); 353 #endif 354 m_adj(m, overlap); 355 rcvpartdupbyte += overlap; 356 m_cat(q->ipqe_m, m); 357 m = q->ipqe_m; 358 pkt_seq = q->ipqe_seq; 359 pkt_len += q->ipqe_len - overlap; 360 rcvoobyte -= overlap; 361 goto free_ipqe; 362 } 363 /* 364 * RX'ed segment extends past the front of the 365 * fragment. Drop the overlapping bytes on the 366 * received packet. The packet will then be 367 * contatentated with this fragment a bit later. 368 */ 369 if (SEQ_GT(q->ipqe_seq, pkt_seq) 370 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 371 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 372 #ifdef TCPREASS_DEBUG 373 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 374 tp, overlap, 375 pkt_seq, pkt_seq + pkt_len, pkt_len); 376 #endif 377 m_adj(m, -overlap); 378 pkt_len -= overlap; 379 rcvpartdupbyte += overlap; 380 rcvoobyte -= overlap; 381 } 382 /* 383 * If the received segment immediates precedes this 384 * fragment then tack the fragment onto this segment 385 * and reinsert the data. 386 */ 387 if (q->ipqe_seq == pkt_seq + pkt_len) { 388 #ifdef TCPREASS_DEBUG 389 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 390 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 391 pkt_seq, pkt_seq + pkt_len, pkt_len); 392 #endif 393 pkt_len += q->ipqe_len; 394 pkt_flags |= q->ipqe_flags; 395 m_cat(m, q->ipqe_m); 396 LIST_REMOVE(q, ipqe_q); 397 LIST_REMOVE(q, ipqe_timeq); 398 if (tiqe == NULL) { 399 tiqe = q; 400 } else { 401 pool_put(&ipqent_pool, q); 402 } 403 break; 404 } 405 /* 406 * If the fragment is before the segment, remember it. 407 * When this loop is terminated, p will contain the 408 * pointer to fragment that is right before the received 409 * segment. 410 */ 411 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 412 p = q; 413 414 continue; 415 416 /* 417 * This is a common operation. It also will allow 418 * to save doing a malloc/free in most instances. 419 */ 420 free_ipqe: 421 LIST_REMOVE(q, ipqe_q); 422 LIST_REMOVE(q, ipqe_timeq); 423 if (tiqe == NULL) { 424 tiqe = q; 425 } else { 426 pool_put(&ipqent_pool, q); 427 } 428 } 429 430 /* 431 * Allocate a new queue entry since the received segment did not 432 * collapse onto any other out-of-order block; thus we are allocating 433 * a new block. If it had collapsed, tiqe would not be NULL and 434 * we would be reusing it. 435 * XXX If we can't, just drop the packet. XXX 436 */ 437 if (tiqe == NULL) { 438 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 439 if (tiqe == NULL) { 440 tcpstat.tcps_rcvmemdrop++; 441 m_freem(m); 442 return (0); 443 } 444 } 445 446 /* 447 * Update the counters. 448 */ 449 tcpstat.tcps_rcvoopack++; 450 tcpstat.tcps_rcvoobyte += rcvoobyte; 451 if (rcvpartdupbyte) { 452 tcpstat.tcps_rcvpartduppack++; 453 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 454 } 455 456 /* 457 * Insert the new fragment queue entry into both queues. 458 */ 459 tiqe->ipqe_m = m; 460 tiqe->ipqe_seq = pkt_seq; 461 tiqe->ipqe_len = pkt_len; 462 tiqe->ipqe_flags = pkt_flags; 463 if (p == NULL) { 464 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 465 #ifdef TCPREASS_DEBUG 466 if (tiqe->ipqe_seq != tp->rcv_nxt) 467 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 468 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 469 #endif 470 } else { 471 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 472 #ifdef TCPREASS_DEBUG 473 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 474 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 475 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 476 #endif 477 } 478 479 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 480 481 present: 482 /* 483 * Present data to user, advancing rcv_nxt through 484 * completed sequence space. 485 */ 486 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 487 return (0); 488 q = tp->segq.lh_first; 489 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 490 return (0); 491 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 492 return (0); 493 494 tp->rcv_nxt += q->ipqe_len; 495 pkt_flags = q->ipqe_flags & TH_FIN; 496 ND6_HINT(tp); 497 498 LIST_REMOVE(q, ipqe_q); 499 LIST_REMOVE(q, ipqe_timeq); 500 if (so->so_state & SS_CANTRCVMORE) 501 m_freem(q->ipqe_m); 502 else 503 sbappend(&so->so_rcv, q->ipqe_m); 504 pool_put(&ipqent_pool, q); 505 sorwakeup(so); 506 return (pkt_flags); 507 } 508 509 #if defined(INET6) && !defined(TCP6) 510 int 511 tcp6_input(mp, offp, proto) 512 struct mbuf **mp; 513 int *offp, proto; 514 { 515 struct mbuf *m = *mp; 516 517 #if defined(NFAITH) && 0 < NFAITH 518 if (m->m_pkthdr.rcvif) { 519 if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 520 /* XXX send icmp6 host/port unreach? */ 521 m_freem(m); 522 return IPPROTO_DONE; 523 } 524 } 525 #endif 526 527 /* 528 * draft-itojun-ipv6-tcp-to-anycast 529 * better place to put this in? 530 */ 531 if (m->m_flags & M_ANYCAST6) { 532 struct ip6_hdr *ip6; 533 if (m->m_len < sizeof(struct ip6_hdr)) { 534 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 535 tcpstat.tcps_rcvshort++; 536 return IPPROTO_DONE; 537 } 538 } 539 ip6 = mtod(m, struct ip6_hdr *); 540 icmp6_error(m, ICMP6_DST_UNREACH, 541 ICMP6_DST_UNREACH_ADDR, 542 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 543 return IPPROTO_DONE; 544 } 545 546 tcp_input(m, *offp, proto); 547 return IPPROTO_DONE; 548 } 549 #endif 550 551 /* 552 * TCP input routine, follows pages 65-76 of the 553 * protocol specification dated September, 1981 very closely. 554 */ 555 void 556 #if __STDC__ 557 tcp_input(struct mbuf *m, ...) 558 #else 559 tcp_input(m, va_alist) 560 register struct mbuf *m; 561 #endif 562 { 563 int proto; 564 register struct tcphdr *th; 565 struct ip *ip; 566 register struct inpcb *inp; 567 #ifdef INET6 568 struct ip6_hdr *ip6; 569 register struct in6pcb *in6p; 570 #endif 571 caddr_t optp = NULL; 572 int optlen = 0; 573 int len, tlen, toff, hdroptlen = 0; 574 register struct tcpcb *tp = 0; 575 register int tiflags; 576 struct socket *so = NULL; 577 int todrop, acked, ourfinisacked, needoutput = 0; 578 short ostate = 0; 579 int iss = 0; 580 u_long tiwin; 581 struct tcp_opt_info opti; 582 int off, iphlen; 583 va_list ap; 584 int af; /* af on the wire */ 585 struct mbuf *tcp_saveti = NULL; 586 587 va_start(ap, m); 588 toff = va_arg(ap, int); 589 proto = va_arg(ap, int); 590 va_end(ap); 591 592 tcpstat.tcps_rcvtotal++; 593 594 bzero(&opti, sizeof(opti)); 595 opti.ts_present = 0; 596 opti.maxseg = 0; 597 598 /* 599 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 600 * 601 * TCP is, by definition, unicast, so we reject all 602 * multicast outright. 603 * 604 * Note, there are additional src/dst address checks in 605 * the AF-specific code below. 606 */ 607 if (m->m_flags & (M_BCAST|M_MCAST)) { 608 /* XXX stat */ 609 goto drop; 610 } 611 #ifdef INET6 612 if (m->m_flags & M_ANYCAST6) { 613 /* XXX stat */ 614 goto drop; 615 } 616 #endif 617 618 /* 619 * Get IP and TCP header together in first mbuf. 620 * Note: IP leaves IP header in first mbuf. 621 */ 622 ip = mtod(m, struct ip *); 623 #ifdef INET6 624 ip6 = NULL; 625 #endif 626 switch (ip->ip_v) { 627 case 4: 628 af = AF_INET; 629 iphlen = sizeof(struct ip); 630 #ifndef PULLDOWN_TEST 631 /* would like to get rid of this... */ 632 if (toff > sizeof (struct ip)) { 633 ip_stripoptions(m, (struct mbuf *)0); 634 toff = sizeof(struct ip); 635 } 636 if (m->m_len < toff + sizeof (struct tcphdr)) { 637 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 638 tcpstat.tcps_rcvshort++; 639 return; 640 } 641 } 642 ip = mtod(m, struct ip *); 643 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 644 #else 645 ip = mtod(m, struct ip *); 646 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 647 sizeof(struct tcphdr)); 648 if (th == NULL) { 649 tcpstat.tcps_rcvshort++; 650 return; 651 } 652 #endif 653 654 /* 655 * Make sure destination address is not multicast. 656 * Source address checked in ip_input(). 657 */ 658 if (IN_MULTICAST(ip->ip_dst.s_addr)) { 659 /* XXX stat */ 660 goto drop; 661 } 662 663 /* We do the checksum after PCB lookup... */ 664 len = ip->ip_len; 665 tlen = len - toff; 666 break; 667 #ifdef INET6 668 case 6: 669 ip = NULL; 670 iphlen = sizeof(struct ip6_hdr); 671 af = AF_INET6; 672 #ifndef PULLDOWN_TEST 673 if (m->m_len < toff + sizeof(struct tcphdr)) { 674 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 675 if (m == NULL) { 676 tcpstat.tcps_rcvshort++; 677 return; 678 } 679 } 680 ip6 = mtod(m, struct ip6_hdr *); 681 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 682 #else 683 ip6 = mtod(m, struct ip6_hdr *); 684 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 685 sizeof(struct tcphdr)); 686 if (th == NULL) { 687 tcpstat.tcps_rcvshort++; 688 return; 689 } 690 #endif 691 692 /* Be proactive about malicious use of IPv4 mapped address */ 693 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 694 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 695 /* XXX stat */ 696 goto drop; 697 } 698 699 /* 700 * Make sure destination address is not multicast. 701 * Source address checked in ip6_input(). 702 */ 703 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 704 /* XXX stat */ 705 goto drop; 706 } 707 708 /* We do the checksum after PCB lookup... */ 709 len = m->m_pkthdr.len; 710 tlen = len - toff; 711 break; 712 #endif 713 default: 714 m_freem(m); 715 return; 716 } 717 718 /* 719 * Check that TCP offset makes sense, 720 * pull out TCP options and adjust length. XXX 721 */ 722 off = th->th_off << 2; 723 if (off < sizeof (struct tcphdr) || off > tlen) { 724 tcpstat.tcps_rcvbadoff++; 725 goto drop; 726 } 727 tlen -= off; 728 729 /* 730 * tcp_input() has been modified to use tlen to mean the TCP data 731 * length throughout the function. Other functions can use 732 * m->m_pkthdr.len as the basis for calculating the TCP data length. 733 * rja 734 */ 735 736 if (off > sizeof (struct tcphdr)) { 737 #ifndef PULLDOWN_TEST 738 if (m->m_len < toff + off) { 739 if ((m = m_pullup(m, toff + off)) == 0) { 740 tcpstat.tcps_rcvshort++; 741 return; 742 } 743 switch (af) { 744 case AF_INET: 745 ip = mtod(m, struct ip *); 746 break; 747 #ifdef INET6 748 case AF_INET6: 749 ip6 = mtod(m, struct ip6_hdr *); 750 break; 751 #endif 752 } 753 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 754 } 755 #else 756 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 757 if (th == NULL) { 758 tcpstat.tcps_rcvshort++; 759 return; 760 } 761 /* 762 * NOTE: ip/ip6 will not be affected by m_pulldown() 763 * (as they're before toff) and we don't need to update those. 764 */ 765 #endif 766 optlen = off - sizeof (struct tcphdr); 767 optp = ((caddr_t)th) + sizeof(struct tcphdr); 768 /* 769 * Do quick retrieval of timestamp options ("options 770 * prediction?"). If timestamp is the only option and it's 771 * formatted as recommended in RFC 1323 appendix A, we 772 * quickly get the values now and not bother calling 773 * tcp_dooptions(), etc. 774 */ 775 if ((optlen == TCPOLEN_TSTAMP_APPA || 776 (optlen > TCPOLEN_TSTAMP_APPA && 777 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 778 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 779 (th->th_flags & TH_SYN) == 0) { 780 opti.ts_present = 1; 781 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 782 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 783 optp = NULL; /* we've parsed the options */ 784 } 785 } 786 tiflags = th->th_flags; 787 788 /* 789 * Locate pcb for segment. 790 */ 791 findpcb: 792 inp = NULL; 793 #ifdef INET6 794 in6p = NULL; 795 #endif 796 switch (af) { 797 case AF_INET: 798 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 799 ip->ip_dst, th->th_dport); 800 if (inp == 0) { 801 ++tcpstat.tcps_pcbhashmiss; 802 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 803 } 804 #if defined(INET6) && !defined(TCP6) 805 if (inp == 0) { 806 struct in6_addr s, d; 807 808 /* mapped addr case */ 809 bzero(&s, sizeof(s)); 810 s.s6_addr16[5] = htons(0xffff); 811 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 812 bzero(&d, sizeof(d)); 813 d.s6_addr16[5] = htons(0xffff); 814 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 815 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 816 &d, th->th_dport, 0); 817 if (in6p == 0) { 818 ++tcpstat.tcps_pcbhashmiss; 819 in6p = in6_pcblookup_bind(&tcb6, &d, 820 th->th_dport, 0); 821 } 822 } 823 #endif 824 #ifndef INET6 825 if (inp == 0) 826 #else 827 if (inp == 0 && in6p == 0) 828 #endif 829 { 830 ++tcpstat.tcps_noport; 831 if (tcp_log_refused && (tiflags & TH_SYN)) { 832 #ifndef INET6 833 char src[4*sizeof "123"]; 834 char dst[4*sizeof "123"]; 835 #else 836 char src[INET6_ADDRSTRLEN]; 837 char dst[INET6_ADDRSTRLEN]; 838 #endif 839 if (ip) { 840 strcpy(src, inet_ntoa(ip->ip_src)); 841 strcpy(dst, inet_ntoa(ip->ip_dst)); 842 } 843 #ifdef INET6 844 else if (ip6) { 845 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 846 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 847 } 848 #endif 849 else { 850 strcpy(src, "(unknown)"); 851 strcpy(dst, "(unknown)"); 852 } 853 log(LOG_INFO, 854 "Connection attempt to TCP %s:%d from %s:%d\n", 855 dst, ntohs(th->th_dport), 856 src, ntohs(th->th_sport)); 857 } 858 TCP_FIELDS_TO_HOST(th); 859 goto dropwithreset; 860 } 861 #ifdef IPSEC 862 if (inp && ipsec4_in_reject(m, inp)) { 863 ipsecstat.in_polvio++; 864 goto drop; 865 } 866 #ifdef INET6 867 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 868 ipsecstat.in_polvio++; 869 goto drop; 870 } 871 #endif 872 #endif /*IPSEC*/ 873 break; 874 #if defined(INET6) && !defined(TCP6) 875 case AF_INET6: 876 { 877 int faith; 878 879 #if defined(NFAITH) && NFAITH > 0 880 if (m->m_pkthdr.rcvif 881 && m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 882 faith = 1; 883 } else 884 faith = 0; 885 #else 886 faith = 0; 887 #endif 888 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 889 &ip6->ip6_dst, th->th_dport, faith); 890 if (in6p == NULL) { 891 ++tcpstat.tcps_pcbhashmiss; 892 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 893 th->th_dport, faith); 894 } 895 if (in6p == NULL) { 896 ++tcpstat.tcps_noport; 897 TCP_FIELDS_TO_HOST(th); 898 goto dropwithreset; 899 } 900 #ifdef IPSEC 901 if (ipsec6_in_reject(m, in6p)) { 902 ipsec6stat.in_polvio++; 903 goto drop; 904 } 905 #endif /*IPSEC*/ 906 break; 907 } 908 #endif 909 } 910 911 /* 912 * If the state is CLOSED (i.e., TCB does not exist) then 913 * all data in the incoming segment is discarded. 914 * If the TCB exists but is in CLOSED state, it is embryonic, 915 * but should either do a listen or a connect soon. 916 */ 917 tp = NULL; 918 so = NULL; 919 if (inp) { 920 tp = intotcpcb(inp); 921 so = inp->inp_socket; 922 } 923 #ifdef INET6 924 else if (in6p) { 925 tp = in6totcpcb(in6p); 926 so = in6p->in6p_socket; 927 } 928 #endif 929 if (tp == 0) { 930 TCP_FIELDS_TO_HOST(th); 931 goto dropwithreset; 932 } 933 if (tp->t_state == TCPS_CLOSED) 934 goto drop; 935 936 /* 937 * Checksum extended TCP header and data. 938 */ 939 switch (af) { 940 case AF_INET: 941 #ifndef PULLDOWN_TEST 942 { 943 struct ipovly *ipov; 944 ipov = (struct ipovly *)ip; 945 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 946 ipov->ih_len = htons(tlen + off); 947 948 if (in_cksum(m, len) != 0) { 949 tcpstat.tcps_rcvbadsum++; 950 goto drop; 951 } 952 } 953 #else 954 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) { 955 tcpstat.tcps_rcvbadsum++; 956 goto drop; 957 } 958 #endif 959 break; 960 961 #ifdef INET6 962 case AF_INET6: 963 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) { 964 tcpstat.tcps_rcvbadsum++; 965 goto drop; 966 } 967 break; 968 #endif 969 } 970 971 TCP_FIELDS_TO_HOST(th); 972 973 /* Unscale the window into a 32-bit value. */ 974 if ((tiflags & TH_SYN) == 0) 975 tiwin = th->th_win << tp->snd_scale; 976 else 977 tiwin = th->th_win; 978 979 #ifdef INET6 980 /* save packet options if user wanted */ 981 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 982 if (in6p->in6p_options) { 983 m_freem(in6p->in6p_options); 984 in6p->in6p_options = 0; 985 } 986 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 987 } 988 #endif 989 990 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 991 union syn_cache_sa src; 992 union syn_cache_sa dst; 993 994 bzero(&src, sizeof(src)); 995 bzero(&dst, sizeof(dst)); 996 switch (af) { 997 case AF_INET: 998 src.sin.sin_len = sizeof(struct sockaddr_in); 999 src.sin.sin_family = AF_INET; 1000 src.sin.sin_addr = ip->ip_src; 1001 src.sin.sin_port = th->th_sport; 1002 1003 dst.sin.sin_len = sizeof(struct sockaddr_in); 1004 dst.sin.sin_family = AF_INET; 1005 dst.sin.sin_addr = ip->ip_dst; 1006 dst.sin.sin_port = th->th_dport; 1007 break; 1008 #ifdef INET6 1009 case AF_INET6: 1010 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1011 src.sin6.sin6_family = AF_INET6; 1012 src.sin6.sin6_addr = ip6->ip6_src; 1013 src.sin6.sin6_port = th->th_sport; 1014 1015 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1016 dst.sin6.sin6_family = AF_INET6; 1017 dst.sin6.sin6_addr = ip6->ip6_dst; 1018 dst.sin6.sin6_port = th->th_dport; 1019 break; 1020 #endif /* INET6 */ 1021 default: 1022 goto badsyn; /*sanity*/ 1023 } 1024 1025 if (so->so_options & SO_DEBUG) { 1026 ostate = tp->t_state; 1027 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1028 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1029 m_freem(tcp_saveti); 1030 tcp_saveti = NULL; 1031 } else { 1032 tcp_saveti->m_len += sizeof(struct tcphdr); 1033 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1034 sizeof(struct tcphdr)); 1035 } 1036 if (tcp_saveti) { 1037 /* 1038 * need to recover version # field, which was 1039 * overwritten on ip_cksum computation. 1040 */ 1041 struct ip *sip; 1042 sip = mtod(tcp_saveti, struct ip *); 1043 switch (af) { 1044 case AF_INET: 1045 sip->ip_v = 4; 1046 break; 1047 #ifdef INET6 1048 case AF_INET6: 1049 sip->ip_v = 6; 1050 break; 1051 #endif 1052 } 1053 } 1054 } 1055 if (so->so_options & SO_ACCEPTCONN) { 1056 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1057 if (tiflags & TH_RST) { 1058 syn_cache_reset(&src.sa, &dst.sa, th); 1059 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1060 (TH_ACK|TH_SYN)) { 1061 /* 1062 * Received a SYN,ACK. This should 1063 * never happen while we are in 1064 * LISTEN. Send an RST. 1065 */ 1066 goto badsyn; 1067 } else if (tiflags & TH_ACK) { 1068 so = syn_cache_get(&src.sa, &dst.sa, 1069 th, toff, tlen, so, m); 1070 if (so == NULL) { 1071 /* 1072 * We don't have a SYN for 1073 * this ACK; send an RST. 1074 */ 1075 goto badsyn; 1076 } else if (so == 1077 (struct socket *)(-1)) { 1078 /* 1079 * We were unable to create 1080 * the connection. If the 1081 * 3-way handshake was 1082 * completed, and RST has 1083 * been sent to the peer. 1084 * Since the mbuf might be 1085 * in use for the reply, 1086 * do not free it. 1087 */ 1088 m = NULL; 1089 } else { 1090 /* 1091 * We have created a 1092 * full-blown connection. 1093 */ 1094 tp = NULL; 1095 inp = NULL; 1096 #ifdef INET6 1097 in6p = NULL; 1098 #endif 1099 switch (so->so_proto->pr_domain->dom_family) { 1100 case AF_INET: 1101 inp = sotoinpcb(so); 1102 tp = intotcpcb(inp); 1103 break; 1104 #ifdef INET6 1105 case AF_INET6: 1106 in6p = sotoin6pcb(so); 1107 tp = in6totcpcb(in6p); 1108 break; 1109 #endif 1110 } 1111 if (tp == NULL) 1112 goto badsyn; /*XXX*/ 1113 tiwin <<= tp->snd_scale; 1114 goto after_listen; 1115 } 1116 } else { 1117 /* 1118 * None of RST, SYN or ACK was set. 1119 * This is an invalid packet for a 1120 * TCB in LISTEN state. Send a RST. 1121 */ 1122 goto badsyn; 1123 } 1124 } else { 1125 /* 1126 * Received a SYN. 1127 */ 1128 1129 /* 1130 * LISTEN socket received a SYN 1131 * from itself? This can't possibly 1132 * be valid; drop the packet. 1133 */ 1134 if (th->th_sport == th->th_dport) { 1135 int i; 1136 1137 switch (af) { 1138 case AF_INET: 1139 i = in_hosteq(ip->ip_src, ip->ip_dst); 1140 break; 1141 #ifdef INET6 1142 case AF_INET6: 1143 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1144 break; 1145 #endif 1146 default: 1147 i = 1; 1148 } 1149 if (i) { 1150 tcpstat.tcps_badsyn++; 1151 goto drop; 1152 } 1153 } 1154 1155 /* 1156 * SYN looks ok; create compressed TCP 1157 * state for it. 1158 */ 1159 if (so->so_qlen <= so->so_qlimit && 1160 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1161 so, m, optp, optlen, &opti)) 1162 m = NULL; 1163 } 1164 goto drop; 1165 } 1166 } 1167 1168 after_listen: 1169 #ifdef DIAGNOSTIC 1170 /* 1171 * Should not happen now that all embryonic connections 1172 * are handled with compressed state. 1173 */ 1174 if (tp->t_state == TCPS_LISTEN) 1175 panic("tcp_input: TCPS_LISTEN"); 1176 #endif 1177 1178 /* 1179 * Segment received on connection. 1180 * Reset idle time and keep-alive timer. 1181 */ 1182 tp->t_idle = 0; 1183 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1184 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1185 1186 /* 1187 * Process options. 1188 */ 1189 if (optp) 1190 tcp_dooptions(tp, optp, optlen, th, &opti); 1191 1192 /* 1193 * Header prediction: check for the two common cases 1194 * of a uni-directional data xfer. If the packet has 1195 * no control flags, is in-sequence, the window didn't 1196 * change and we're not retransmitting, it's a 1197 * candidate. If the length is zero and the ack moved 1198 * forward, we're the sender side of the xfer. Just 1199 * free the data acked & wake any higher level process 1200 * that was blocked waiting for space. If the length 1201 * is non-zero and the ack didn't move, we're the 1202 * receiver side. If we're getting packets in-order 1203 * (the reassembly queue is empty), add the data to 1204 * the socket buffer and note that we need a delayed ack. 1205 */ 1206 if (tp->t_state == TCPS_ESTABLISHED && 1207 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1208 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1209 th->th_seq == tp->rcv_nxt && 1210 tiwin && tiwin == tp->snd_wnd && 1211 tp->snd_nxt == tp->snd_max) { 1212 1213 /* 1214 * If last ACK falls within this segment's sequence numbers, 1215 * record the timestamp. 1216 */ 1217 if (opti.ts_present && 1218 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1219 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1220 tp->ts_recent_age = tcp_now; 1221 tp->ts_recent = opti.ts_val; 1222 } 1223 1224 if (tlen == 0) { 1225 if (SEQ_GT(th->th_ack, tp->snd_una) && 1226 SEQ_LEQ(th->th_ack, tp->snd_max) && 1227 tp->snd_cwnd >= tp->snd_wnd && 1228 tp->t_dupacks < tcprexmtthresh) { 1229 /* 1230 * this is a pure ack for outstanding data. 1231 */ 1232 ++tcpstat.tcps_predack; 1233 if (opti.ts_present && opti.ts_ecr) 1234 tcp_xmit_timer(tp, 1235 tcp_now - opti.ts_ecr + 1); 1236 else if (tp->t_rtt && 1237 SEQ_GT(th->th_ack, tp->t_rtseq)) 1238 tcp_xmit_timer(tp, tp->t_rtt); 1239 acked = th->th_ack - tp->snd_una; 1240 tcpstat.tcps_rcvackpack++; 1241 tcpstat.tcps_rcvackbyte += acked; 1242 ND6_HINT(tp); 1243 sbdrop(&so->so_snd, acked); 1244 /* 1245 * We want snd_recover to track snd_una to 1246 * avoid sequence wraparound problems for 1247 * very large transfers. 1248 */ 1249 tp->snd_una = tp->snd_recover = th->th_ack; 1250 m_freem(m); 1251 1252 /* 1253 * If all outstanding data are acked, stop 1254 * retransmit timer, otherwise restart timer 1255 * using current (possibly backed-off) value. 1256 * If process is waiting for space, 1257 * wakeup/selwakeup/signal. If data 1258 * are ready to send, let tcp_output 1259 * decide between more output or persist. 1260 */ 1261 if (tp->snd_una == tp->snd_max) 1262 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1263 else if (TCP_TIMER_ISARMED(tp, 1264 TCPT_PERSIST) == 0) 1265 TCP_TIMER_ARM(tp, TCPT_REXMT, 1266 tp->t_rxtcur); 1267 1268 sowwakeup(so); 1269 if (so->so_snd.sb_cc) 1270 (void) tcp_output(tp); 1271 if (tcp_saveti) 1272 m_freem(tcp_saveti); 1273 return; 1274 } 1275 } else if (th->th_ack == tp->snd_una && 1276 tp->segq.lh_first == NULL && 1277 tlen <= sbspace(&so->so_rcv)) { 1278 /* 1279 * this is a pure, in-sequence data packet 1280 * with nothing on the reassembly queue and 1281 * we have enough buffer space to take it. 1282 */ 1283 ++tcpstat.tcps_preddat; 1284 tp->rcv_nxt += tlen; 1285 tcpstat.tcps_rcvpack++; 1286 tcpstat.tcps_rcvbyte += tlen; 1287 ND6_HINT(tp); 1288 /* 1289 * Drop TCP, IP headers and TCP options then add data 1290 * to socket buffer. 1291 */ 1292 m_adj(m, toff + off); 1293 sbappend(&so->so_rcv, m); 1294 sorwakeup(so); 1295 TCP_SETUP_ACK(tp, th); 1296 if (tp->t_flags & TF_ACKNOW) 1297 (void) tcp_output(tp); 1298 if (tcp_saveti) 1299 m_freem(tcp_saveti); 1300 return; 1301 } 1302 } 1303 1304 /* 1305 * Compute mbuf offset to TCP data segment. 1306 */ 1307 hdroptlen = toff + off; 1308 1309 /* 1310 * Calculate amount of space in receive window, 1311 * and then do TCP input processing. 1312 * Receive window is amount of space in rcv queue, 1313 * but not less than advertised window. 1314 */ 1315 { int win; 1316 1317 win = sbspace(&so->so_rcv); 1318 if (win < 0) 1319 win = 0; 1320 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1321 } 1322 1323 switch (tp->t_state) { 1324 1325 /* 1326 * If the state is SYN_SENT: 1327 * if seg contains an ACK, but not for our SYN, drop the input. 1328 * if seg contains a RST, then drop the connection. 1329 * if seg does not contain SYN, then drop it. 1330 * Otherwise this is an acceptable SYN segment 1331 * initialize tp->rcv_nxt and tp->irs 1332 * if seg contains ack then advance tp->snd_una 1333 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1334 * arrange for segment to be acked (eventually) 1335 * continue processing rest of data/controls, beginning with URG 1336 */ 1337 case TCPS_SYN_SENT: 1338 if ((tiflags & TH_ACK) && 1339 (SEQ_LEQ(th->th_ack, tp->iss) || 1340 SEQ_GT(th->th_ack, tp->snd_max))) 1341 goto dropwithreset; 1342 if (tiflags & TH_RST) { 1343 if (tiflags & TH_ACK) 1344 tp = tcp_drop(tp, ECONNREFUSED); 1345 goto drop; 1346 } 1347 if ((tiflags & TH_SYN) == 0) 1348 goto drop; 1349 if (tiflags & TH_ACK) { 1350 tp->snd_una = tp->snd_recover = th->th_ack; 1351 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1352 tp->snd_nxt = tp->snd_una; 1353 } 1354 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1355 tp->irs = th->th_seq; 1356 tcp_rcvseqinit(tp); 1357 tp->t_flags |= TF_ACKNOW; 1358 tcp_mss_from_peer(tp, opti.maxseg); 1359 1360 /* 1361 * Initialize the initial congestion window. If we 1362 * had to retransmit the SYN, we must initialize cwnd 1363 * to 1 segment (i.e. the Loss Window). 1364 */ 1365 if (tp->t_flags & TF_SYN_REXMT) 1366 tp->snd_cwnd = tp->t_peermss; 1367 else 1368 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1369 tp->t_peermss); 1370 1371 tcp_rmx_rtt(tp); 1372 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1373 tcpstat.tcps_connects++; 1374 soisconnected(so); 1375 tcp_established(tp); 1376 /* Do window scaling on this connection? */ 1377 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1378 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1379 tp->snd_scale = tp->requested_s_scale; 1380 tp->rcv_scale = tp->request_r_scale; 1381 } 1382 TCP_REASS_LOCK(tp); 1383 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1384 TCP_REASS_UNLOCK(tp); 1385 /* 1386 * if we didn't have to retransmit the SYN, 1387 * use its rtt as our initial srtt & rtt var. 1388 */ 1389 if (tp->t_rtt) 1390 tcp_xmit_timer(tp, tp->t_rtt); 1391 } else 1392 tp->t_state = TCPS_SYN_RECEIVED; 1393 1394 /* 1395 * Advance th->th_seq to correspond to first data byte. 1396 * If data, trim to stay within window, 1397 * dropping FIN if necessary. 1398 */ 1399 th->th_seq++; 1400 if (tlen > tp->rcv_wnd) { 1401 todrop = tlen - tp->rcv_wnd; 1402 m_adj(m, -todrop); 1403 tlen = tp->rcv_wnd; 1404 tiflags &= ~TH_FIN; 1405 tcpstat.tcps_rcvpackafterwin++; 1406 tcpstat.tcps_rcvbyteafterwin += todrop; 1407 } 1408 tp->snd_wl1 = th->th_seq - 1; 1409 tp->rcv_up = th->th_seq; 1410 goto step6; 1411 1412 /* 1413 * If the state is SYN_RECEIVED: 1414 * If seg contains an ACK, but not for our SYN, drop the input 1415 * and generate an RST. See page 36, rfc793 1416 */ 1417 case TCPS_SYN_RECEIVED: 1418 if ((tiflags & TH_ACK) && 1419 (SEQ_LEQ(th->th_ack, tp->iss) || 1420 SEQ_GT(th->th_ack, tp->snd_max))) 1421 goto dropwithreset; 1422 break; 1423 } 1424 1425 /* 1426 * States other than LISTEN or SYN_SENT. 1427 * First check timestamp, if present. 1428 * Then check that at least some bytes of segment are within 1429 * receive window. If segment begins before rcv_nxt, 1430 * drop leading data (and SYN); if nothing left, just ack. 1431 * 1432 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1433 * and it's less than ts_recent, drop it. 1434 */ 1435 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1436 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1437 1438 /* Check to see if ts_recent is over 24 days old. */ 1439 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1440 /* 1441 * Invalidate ts_recent. If this segment updates 1442 * ts_recent, the age will be reset later and ts_recent 1443 * will get a valid value. If it does not, setting 1444 * ts_recent to zero will at least satisfy the 1445 * requirement that zero be placed in the timestamp 1446 * echo reply when ts_recent isn't valid. The 1447 * age isn't reset until we get a valid ts_recent 1448 * because we don't want out-of-order segments to be 1449 * dropped when ts_recent is old. 1450 */ 1451 tp->ts_recent = 0; 1452 } else { 1453 tcpstat.tcps_rcvduppack++; 1454 tcpstat.tcps_rcvdupbyte += tlen; 1455 tcpstat.tcps_pawsdrop++; 1456 goto dropafterack; 1457 } 1458 } 1459 1460 todrop = tp->rcv_nxt - th->th_seq; 1461 if (todrop > 0) { 1462 if (tiflags & TH_SYN) { 1463 tiflags &= ~TH_SYN; 1464 th->th_seq++; 1465 if (th->th_urp > 1) 1466 th->th_urp--; 1467 else { 1468 tiflags &= ~TH_URG; 1469 th->th_urp = 0; 1470 } 1471 todrop--; 1472 } 1473 if (todrop > tlen || 1474 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1475 /* 1476 * Any valid FIN must be to the left of the window. 1477 * At this point the FIN must be a duplicate or 1478 * out of sequence; drop it. 1479 */ 1480 tiflags &= ~TH_FIN; 1481 /* 1482 * Send an ACK to resynchronize and drop any data. 1483 * But keep on processing for RST or ACK. 1484 */ 1485 tp->t_flags |= TF_ACKNOW; 1486 todrop = tlen; 1487 tcpstat.tcps_rcvdupbyte += todrop; 1488 tcpstat.tcps_rcvduppack++; 1489 } else { 1490 tcpstat.tcps_rcvpartduppack++; 1491 tcpstat.tcps_rcvpartdupbyte += todrop; 1492 } 1493 hdroptlen += todrop; /*drop from head afterwards*/ 1494 th->th_seq += todrop; 1495 tlen -= todrop; 1496 if (th->th_urp > todrop) 1497 th->th_urp -= todrop; 1498 else { 1499 tiflags &= ~TH_URG; 1500 th->th_urp = 0; 1501 } 1502 } 1503 1504 /* 1505 * If new data are received on a connection after the 1506 * user processes are gone, then RST the other end. 1507 */ 1508 if ((so->so_state & SS_NOFDREF) && 1509 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1510 tp = tcp_close(tp); 1511 tcpstat.tcps_rcvafterclose++; 1512 goto dropwithreset; 1513 } 1514 1515 /* 1516 * If segment ends after window, drop trailing data 1517 * (and PUSH and FIN); if nothing left, just ACK. 1518 */ 1519 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1520 if (todrop > 0) { 1521 tcpstat.tcps_rcvpackafterwin++; 1522 if (todrop >= tlen) { 1523 tcpstat.tcps_rcvbyteafterwin += tlen; 1524 /* 1525 * If a new connection request is received 1526 * while in TIME_WAIT, drop the old connection 1527 * and start over if the sequence numbers 1528 * are above the previous ones. 1529 */ 1530 if (tiflags & TH_SYN && 1531 tp->t_state == TCPS_TIME_WAIT && 1532 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1533 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1534 tp->snd_nxt); 1535 tp = tcp_close(tp); 1536 goto findpcb; 1537 } 1538 /* 1539 * If window is closed can only take segments at 1540 * window edge, and have to drop data and PUSH from 1541 * incoming segments. Continue processing, but 1542 * remember to ack. Otherwise, drop segment 1543 * and ack. 1544 */ 1545 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1546 tp->t_flags |= TF_ACKNOW; 1547 tcpstat.tcps_rcvwinprobe++; 1548 } else 1549 goto dropafterack; 1550 } else 1551 tcpstat.tcps_rcvbyteafterwin += todrop; 1552 m_adj(m, -todrop); 1553 tlen -= todrop; 1554 tiflags &= ~(TH_PUSH|TH_FIN); 1555 } 1556 1557 /* 1558 * If last ACK falls within this segment's sequence numbers, 1559 * and the timestamp is newer, record it. 1560 */ 1561 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1562 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1563 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1564 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1565 tp->ts_recent_age = tcp_now; 1566 tp->ts_recent = opti.ts_val; 1567 } 1568 1569 /* 1570 * If the RST bit is set examine the state: 1571 * SYN_RECEIVED STATE: 1572 * If passive open, return to LISTEN state. 1573 * If active open, inform user that connection was refused. 1574 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1575 * Inform user that connection was reset, and close tcb. 1576 * CLOSING, LAST_ACK, TIME_WAIT STATES 1577 * Close the tcb. 1578 */ 1579 if (tiflags&TH_RST) switch (tp->t_state) { 1580 1581 case TCPS_SYN_RECEIVED: 1582 so->so_error = ECONNREFUSED; 1583 goto close; 1584 1585 case TCPS_ESTABLISHED: 1586 case TCPS_FIN_WAIT_1: 1587 case TCPS_FIN_WAIT_2: 1588 case TCPS_CLOSE_WAIT: 1589 so->so_error = ECONNRESET; 1590 close: 1591 tp->t_state = TCPS_CLOSED; 1592 tcpstat.tcps_drops++; 1593 tp = tcp_close(tp); 1594 goto drop; 1595 1596 case TCPS_CLOSING: 1597 case TCPS_LAST_ACK: 1598 case TCPS_TIME_WAIT: 1599 tp = tcp_close(tp); 1600 goto drop; 1601 } 1602 1603 /* 1604 * If a SYN is in the window, then this is an 1605 * error and we send an RST and drop the connection. 1606 */ 1607 if (tiflags & TH_SYN) { 1608 tp = tcp_drop(tp, ECONNRESET); 1609 goto dropwithreset; 1610 } 1611 1612 /* 1613 * If the ACK bit is off we drop the segment and return. 1614 */ 1615 if ((tiflags & TH_ACK) == 0) { 1616 if (tp->t_flags & TF_ACKNOW) 1617 goto dropafterack; 1618 else 1619 goto drop; 1620 } 1621 1622 /* 1623 * Ack processing. 1624 */ 1625 switch (tp->t_state) { 1626 1627 /* 1628 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1629 * ESTABLISHED state and continue processing, otherwise 1630 * send an RST. 1631 */ 1632 case TCPS_SYN_RECEIVED: 1633 if (SEQ_GT(tp->snd_una, th->th_ack) || 1634 SEQ_GT(th->th_ack, tp->snd_max)) 1635 goto dropwithreset; 1636 tcpstat.tcps_connects++; 1637 soisconnected(so); 1638 tcp_established(tp); 1639 /* Do window scaling? */ 1640 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1641 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1642 tp->snd_scale = tp->requested_s_scale; 1643 tp->rcv_scale = tp->request_r_scale; 1644 } 1645 TCP_REASS_LOCK(tp); 1646 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1647 TCP_REASS_UNLOCK(tp); 1648 tp->snd_wl1 = th->th_seq - 1; 1649 /* fall into ... */ 1650 1651 /* 1652 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1653 * ACKs. If the ack is in the range 1654 * tp->snd_una < th->th_ack <= tp->snd_max 1655 * then advance tp->snd_una to th->th_ack and drop 1656 * data from the retransmission queue. If this ACK reflects 1657 * more up to date window information we update our window information. 1658 */ 1659 case TCPS_ESTABLISHED: 1660 case TCPS_FIN_WAIT_1: 1661 case TCPS_FIN_WAIT_2: 1662 case TCPS_CLOSE_WAIT: 1663 case TCPS_CLOSING: 1664 case TCPS_LAST_ACK: 1665 case TCPS_TIME_WAIT: 1666 1667 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1668 if (tlen == 0 && tiwin == tp->snd_wnd) { 1669 tcpstat.tcps_rcvdupack++; 1670 /* 1671 * If we have outstanding data (other than 1672 * a window probe), this is a completely 1673 * duplicate ack (ie, window info didn't 1674 * change), the ack is the biggest we've 1675 * seen and we've seen exactly our rexmt 1676 * threshhold of them, assume a packet 1677 * has been dropped and retransmit it. 1678 * Kludge snd_nxt & the congestion 1679 * window so we send only this one 1680 * packet. 1681 * 1682 * We know we're losing at the current 1683 * window size so do congestion avoidance 1684 * (set ssthresh to half the current window 1685 * and pull our congestion window back to 1686 * the new ssthresh). 1687 * 1688 * Dup acks mean that packets have left the 1689 * network (they're now cached at the receiver) 1690 * so bump cwnd by the amount in the receiver 1691 * to keep a constant cwnd packets in the 1692 * network. 1693 */ 1694 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1695 th->th_ack != tp->snd_una) 1696 tp->t_dupacks = 0; 1697 else if (++tp->t_dupacks == tcprexmtthresh) { 1698 tcp_seq onxt = tp->snd_nxt; 1699 u_int win = 1700 min(tp->snd_wnd, tp->snd_cwnd) / 1701 2 / tp->t_segsz; 1702 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1703 tp->snd_recover)) { 1704 /* 1705 * False fast retransmit after 1706 * timeout. Do not cut window. 1707 */ 1708 tp->snd_cwnd += tp->t_segsz; 1709 tp->t_dupacks = 0; 1710 (void) tcp_output(tp); 1711 goto drop; 1712 } 1713 1714 if (win < 2) 1715 win = 2; 1716 tp->snd_ssthresh = win * tp->t_segsz; 1717 tp->snd_recover = tp->snd_max; 1718 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1719 tp->t_rtt = 0; 1720 tp->snd_nxt = th->th_ack; 1721 tp->snd_cwnd = tp->t_segsz; 1722 (void) tcp_output(tp); 1723 tp->snd_cwnd = tp->snd_ssthresh + 1724 tp->t_segsz * tp->t_dupacks; 1725 if (SEQ_GT(onxt, tp->snd_nxt)) 1726 tp->snd_nxt = onxt; 1727 goto drop; 1728 } else if (tp->t_dupacks > tcprexmtthresh) { 1729 tp->snd_cwnd += tp->t_segsz; 1730 (void) tcp_output(tp); 1731 goto drop; 1732 } 1733 } else 1734 tp->t_dupacks = 0; 1735 break; 1736 } 1737 /* 1738 * If the congestion window was inflated to account 1739 * for the other side's cached packets, retract it. 1740 */ 1741 if (tcp_do_newreno == 0) { 1742 if (tp->t_dupacks >= tcprexmtthresh && 1743 tp->snd_cwnd > tp->snd_ssthresh) 1744 tp->snd_cwnd = tp->snd_ssthresh; 1745 tp->t_dupacks = 0; 1746 } else if (tp->t_dupacks >= tcprexmtthresh && 1747 tcp_newreno(tp, th) == 0) { 1748 tp->snd_cwnd = tp->snd_ssthresh; 1749 /* 1750 * Window inflation should have left us with approx. 1751 * snd_ssthresh outstanding data. But in case we 1752 * would be inclined to send a burst, better to do 1753 * it via the slow start mechanism. 1754 */ 1755 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1756 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1757 + tp->t_segsz; 1758 tp->t_dupacks = 0; 1759 } 1760 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1761 tcpstat.tcps_rcvacktoomuch++; 1762 goto dropafterack; 1763 } 1764 acked = th->th_ack - tp->snd_una; 1765 tcpstat.tcps_rcvackpack++; 1766 tcpstat.tcps_rcvackbyte += acked; 1767 1768 /* 1769 * If we have a timestamp reply, update smoothed 1770 * round trip time. If no timestamp is present but 1771 * transmit timer is running and timed sequence 1772 * number was acked, update smoothed round trip time. 1773 * Since we now have an rtt measurement, cancel the 1774 * timer backoff (cf., Phil Karn's retransmit alg.). 1775 * Recompute the initial retransmit timer. 1776 */ 1777 if (opti.ts_present && opti.ts_ecr) 1778 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1779 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1780 tcp_xmit_timer(tp,tp->t_rtt); 1781 1782 /* 1783 * If all outstanding data is acked, stop retransmit 1784 * timer and remember to restart (more output or persist). 1785 * If there is more data to be acked, restart retransmit 1786 * timer, using current (possibly backed-off) value. 1787 */ 1788 if (th->th_ack == tp->snd_max) { 1789 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1790 needoutput = 1; 1791 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1792 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1793 /* 1794 * When new data is acked, open the congestion window. 1795 * If the window gives us less than ssthresh packets 1796 * in flight, open exponentially (segsz per packet). 1797 * Otherwise open linearly: segsz per window 1798 * (segsz^2 / cwnd per packet), plus a constant 1799 * fraction of a packet (segsz/8) to help larger windows 1800 * open quickly enough. 1801 */ 1802 { 1803 register u_int cw = tp->snd_cwnd; 1804 register u_int incr = tp->t_segsz; 1805 1806 if (cw > tp->snd_ssthresh) 1807 incr = incr * incr / cw; 1808 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1809 tp->snd_cwnd = min(cw + incr, 1810 TCP_MAXWIN << tp->snd_scale); 1811 } 1812 ND6_HINT(tp); 1813 if (acked > so->so_snd.sb_cc) { 1814 tp->snd_wnd -= so->so_snd.sb_cc; 1815 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1816 ourfinisacked = 1; 1817 } else { 1818 sbdrop(&so->so_snd, acked); 1819 tp->snd_wnd -= acked; 1820 ourfinisacked = 0; 1821 } 1822 sowwakeup(so); 1823 /* 1824 * We want snd_recover to track snd_una to 1825 * avoid sequence wraparound problems for 1826 * very large transfers. 1827 */ 1828 tp->snd_una = tp->snd_recover = th->th_ack; 1829 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1830 tp->snd_nxt = tp->snd_una; 1831 1832 switch (tp->t_state) { 1833 1834 /* 1835 * In FIN_WAIT_1 STATE in addition to the processing 1836 * for the ESTABLISHED state if our FIN is now acknowledged 1837 * then enter FIN_WAIT_2. 1838 */ 1839 case TCPS_FIN_WAIT_1: 1840 if (ourfinisacked) { 1841 /* 1842 * If we can't receive any more 1843 * data, then closing user can proceed. 1844 * Starting the timer is contrary to the 1845 * specification, but if we don't get a FIN 1846 * we'll hang forever. 1847 */ 1848 if (so->so_state & SS_CANTRCVMORE) { 1849 soisdisconnected(so); 1850 if (tcp_maxidle > 0) 1851 TCP_TIMER_ARM(tp, TCPT_2MSL, 1852 tcp_maxidle); 1853 } 1854 tp->t_state = TCPS_FIN_WAIT_2; 1855 } 1856 break; 1857 1858 /* 1859 * In CLOSING STATE in addition to the processing for 1860 * the ESTABLISHED state if the ACK acknowledges our FIN 1861 * then enter the TIME-WAIT state, otherwise ignore 1862 * the segment. 1863 */ 1864 case TCPS_CLOSING: 1865 if (ourfinisacked) { 1866 tp->t_state = TCPS_TIME_WAIT; 1867 tcp_canceltimers(tp); 1868 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1869 soisdisconnected(so); 1870 } 1871 break; 1872 1873 /* 1874 * In LAST_ACK, we may still be waiting for data to drain 1875 * and/or to be acked, as well as for the ack of our FIN. 1876 * If our FIN is now acknowledged, delete the TCB, 1877 * enter the closed state and return. 1878 */ 1879 case TCPS_LAST_ACK: 1880 if (ourfinisacked) { 1881 tp = tcp_close(tp); 1882 goto drop; 1883 } 1884 break; 1885 1886 /* 1887 * In TIME_WAIT state the only thing that should arrive 1888 * is a retransmission of the remote FIN. Acknowledge 1889 * it and restart the finack timer. 1890 */ 1891 case TCPS_TIME_WAIT: 1892 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1893 goto dropafterack; 1894 } 1895 } 1896 1897 step6: 1898 /* 1899 * Update window information. 1900 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1901 */ 1902 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1903 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1904 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1905 /* keep track of pure window updates */ 1906 if (tlen == 0 && 1907 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1908 tcpstat.tcps_rcvwinupd++; 1909 tp->snd_wnd = tiwin; 1910 tp->snd_wl1 = th->th_seq; 1911 tp->snd_wl2 = th->th_ack; 1912 if (tp->snd_wnd > tp->max_sndwnd) 1913 tp->max_sndwnd = tp->snd_wnd; 1914 needoutput = 1; 1915 } 1916 1917 /* 1918 * Process segments with URG. 1919 */ 1920 if ((tiflags & TH_URG) && th->th_urp && 1921 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1922 /* 1923 * This is a kludge, but if we receive and accept 1924 * random urgent pointers, we'll crash in 1925 * soreceive. It's hard to imagine someone 1926 * actually wanting to send this much urgent data. 1927 */ 1928 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1929 th->th_urp = 0; /* XXX */ 1930 tiflags &= ~TH_URG; /* XXX */ 1931 goto dodata; /* XXX */ 1932 } 1933 /* 1934 * If this segment advances the known urgent pointer, 1935 * then mark the data stream. This should not happen 1936 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1937 * a FIN has been received from the remote side. 1938 * In these states we ignore the URG. 1939 * 1940 * According to RFC961 (Assigned Protocols), 1941 * the urgent pointer points to the last octet 1942 * of urgent data. We continue, however, 1943 * to consider it to indicate the first octet 1944 * of data past the urgent section as the original 1945 * spec states (in one of two places). 1946 */ 1947 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1948 tp->rcv_up = th->th_seq + th->th_urp; 1949 so->so_oobmark = so->so_rcv.sb_cc + 1950 (tp->rcv_up - tp->rcv_nxt) - 1; 1951 if (so->so_oobmark == 0) 1952 so->so_state |= SS_RCVATMARK; 1953 sohasoutofband(so); 1954 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1955 } 1956 /* 1957 * Remove out of band data so doesn't get presented to user. 1958 * This can happen independent of advancing the URG pointer, 1959 * but if two URG's are pending at once, some out-of-band 1960 * data may creep in... ick. 1961 */ 1962 if (th->th_urp <= (u_int16_t) tlen 1963 #ifdef SO_OOBINLINE 1964 && (so->so_options & SO_OOBINLINE) == 0 1965 #endif 1966 ) 1967 tcp_pulloutofband(so, th, m, hdroptlen); 1968 } else 1969 /* 1970 * If no out of band data is expected, 1971 * pull receive urgent pointer along 1972 * with the receive window. 1973 */ 1974 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1975 tp->rcv_up = tp->rcv_nxt; 1976 dodata: /* XXX */ 1977 1978 /* 1979 * Process the segment text, merging it into the TCP sequencing queue, 1980 * and arranging for acknowledgement of receipt if necessary. 1981 * This process logically involves adjusting tp->rcv_wnd as data 1982 * is presented to the user (this happens in tcp_usrreq.c, 1983 * case PRU_RCVD). If a FIN has already been received on this 1984 * connection then we just ignore the text. 1985 */ 1986 if ((tlen || (tiflags & TH_FIN)) && 1987 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1988 /* 1989 * Insert segment ti into reassembly queue of tcp with 1990 * control block tp. Return TH_FIN if reassembly now includes 1991 * a segment with FIN. The macro form does the common case 1992 * inline (segment is the next to be received on an 1993 * established connection, and the queue is empty), 1994 * avoiding linkage into and removal from the queue and 1995 * repetition of various conversions. 1996 * Set DELACK for segments received in order, but ack 1997 * immediately when segments are out of order 1998 * (so fast retransmit can work). 1999 */ 2000 /* NOTE: this was TCP_REASS() macro, but used only once */ 2001 TCP_REASS_LOCK(tp); 2002 if (th->th_seq == tp->rcv_nxt && 2003 tp->segq.lh_first == NULL && 2004 tp->t_state == TCPS_ESTABLISHED) { 2005 TCP_SETUP_ACK(tp, th); 2006 tp->rcv_nxt += tlen; 2007 tiflags = th->th_flags & TH_FIN; 2008 tcpstat.tcps_rcvpack++; 2009 tcpstat.tcps_rcvbyte += tlen; 2010 ND6_HINT(tp); 2011 m_adj(m, hdroptlen); 2012 sbappend(&(so)->so_rcv, m); 2013 sorwakeup(so); 2014 } else { 2015 m_adj(m, hdroptlen); 2016 tiflags = tcp_reass(tp, th, m, &tlen); 2017 tp->t_flags |= TF_ACKNOW; 2018 } 2019 TCP_REASS_UNLOCK(tp); 2020 2021 /* 2022 * Note the amount of data that peer has sent into 2023 * our window, in order to estimate the sender's 2024 * buffer size. 2025 */ 2026 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2027 } else { 2028 m_freem(m); 2029 m = NULL; 2030 tiflags &= ~TH_FIN; 2031 } 2032 2033 /* 2034 * If FIN is received ACK the FIN and let the user know 2035 * that the connection is closing. Ignore a FIN received before 2036 * the connection is fully established. 2037 */ 2038 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2039 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2040 socantrcvmore(so); 2041 tp->t_flags |= TF_ACKNOW; 2042 tp->rcv_nxt++; 2043 } 2044 switch (tp->t_state) { 2045 2046 /* 2047 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2048 */ 2049 case TCPS_ESTABLISHED: 2050 tp->t_state = TCPS_CLOSE_WAIT; 2051 break; 2052 2053 /* 2054 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2055 * enter the CLOSING state. 2056 */ 2057 case TCPS_FIN_WAIT_1: 2058 tp->t_state = TCPS_CLOSING; 2059 break; 2060 2061 /* 2062 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2063 * starting the time-wait timer, turning off the other 2064 * standard timers. 2065 */ 2066 case TCPS_FIN_WAIT_2: 2067 tp->t_state = TCPS_TIME_WAIT; 2068 tcp_canceltimers(tp); 2069 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2070 soisdisconnected(so); 2071 break; 2072 2073 /* 2074 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2075 */ 2076 case TCPS_TIME_WAIT: 2077 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2078 break; 2079 } 2080 } 2081 if (so->so_options & SO_DEBUG) { 2082 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2083 } 2084 2085 /* 2086 * Return any desired output. 2087 */ 2088 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2089 (void) tcp_output(tp); 2090 if (tcp_saveti) 2091 m_freem(tcp_saveti); 2092 return; 2093 2094 badsyn: 2095 /* 2096 * Received a bad SYN. Increment counters and dropwithreset. 2097 */ 2098 tcpstat.tcps_badsyn++; 2099 tp = NULL; 2100 goto dropwithreset; 2101 2102 dropafterack: 2103 /* 2104 * Generate an ACK dropping incoming segment if it occupies 2105 * sequence space, where the ACK reflects our state. 2106 */ 2107 if (tiflags & TH_RST) 2108 goto drop; 2109 m_freem(m); 2110 tp->t_flags |= TF_ACKNOW; 2111 (void) tcp_output(tp); 2112 if (tcp_saveti) 2113 m_freem(tcp_saveti); 2114 return; 2115 2116 dropwithreset: 2117 /* 2118 * Generate a RST, dropping incoming segment. 2119 * Make ACK acceptable to originator of segment. 2120 * Don't bother to respond if destination was broadcast/multicast. 2121 */ 2122 if (tiflags & TH_RST) 2123 goto drop; 2124 { 2125 /* 2126 * need to recover version # field, which was overwritten on 2127 * ip_cksum computation. 2128 */ 2129 struct ip *sip; 2130 sip = mtod(m, struct ip *); 2131 switch (af) { 2132 case AF_INET: 2133 sip->ip_v = 4; 2134 break; 2135 #ifdef INET6 2136 case AF_INET6: 2137 sip->ip_v = 6; 2138 break; 2139 #endif 2140 } 2141 } 2142 if (tiflags & TH_ACK) 2143 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2144 else { 2145 if (tiflags & TH_SYN) 2146 tlen++; 2147 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2148 TH_RST|TH_ACK); 2149 } 2150 if (tcp_saveti) 2151 m_freem(tcp_saveti); 2152 return; 2153 2154 drop: 2155 /* 2156 * Drop space held by incoming segment and return. 2157 */ 2158 if (tp) { 2159 if (tp->t_inpcb) 2160 so = tp->t_inpcb->inp_socket; 2161 #ifdef INET6 2162 else if (tp->t_in6pcb) 2163 so = tp->t_in6pcb->in6p_socket; 2164 #endif 2165 else 2166 so = NULL; 2167 if (so && (so->so_options & SO_DEBUG) != 0) 2168 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2169 } 2170 if (tcp_saveti) 2171 m_freem(tcp_saveti); 2172 m_freem(m); 2173 return; 2174 } 2175 2176 void 2177 tcp_dooptions(tp, cp, cnt, th, oi) 2178 struct tcpcb *tp; 2179 u_char *cp; 2180 int cnt; 2181 struct tcphdr *th; 2182 struct tcp_opt_info *oi; 2183 { 2184 u_int16_t mss; 2185 int opt, optlen; 2186 2187 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2188 opt = cp[0]; 2189 if (opt == TCPOPT_EOL) 2190 break; 2191 if (opt == TCPOPT_NOP) 2192 optlen = 1; 2193 else { 2194 optlen = cp[1]; 2195 if (optlen <= 0) 2196 break; 2197 } 2198 switch (opt) { 2199 2200 default: 2201 continue; 2202 2203 case TCPOPT_MAXSEG: 2204 if (optlen != TCPOLEN_MAXSEG) 2205 continue; 2206 if (!(th->th_flags & TH_SYN)) 2207 continue; 2208 bcopy(cp + 2, &mss, sizeof(mss)); 2209 oi->maxseg = ntohs(mss); 2210 break; 2211 2212 case TCPOPT_WINDOW: 2213 if (optlen != TCPOLEN_WINDOW) 2214 continue; 2215 if (!(th->th_flags & TH_SYN)) 2216 continue; 2217 tp->t_flags |= TF_RCVD_SCALE; 2218 tp->requested_s_scale = cp[2]; 2219 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2220 #if 0 /*XXX*/ 2221 char *p; 2222 2223 if (ip) 2224 p = ntohl(ip->ip_src); 2225 #ifdef INET6 2226 else if (ip6) 2227 p = ip6_sprintf(&ip6->ip6_src); 2228 #endif 2229 else 2230 p = "(unknown)"; 2231 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2232 "assuming %d\n", 2233 tp->requested_s_scale, p, 2234 TCP_MAX_WINSHIFT); 2235 #else 2236 log(LOG_ERR, "TCP: invalid wscale %d, " 2237 "assuming %d\n", 2238 tp->requested_s_scale, 2239 TCP_MAX_WINSHIFT); 2240 #endif 2241 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2242 } 2243 break; 2244 2245 case TCPOPT_TIMESTAMP: 2246 if (optlen != TCPOLEN_TIMESTAMP) 2247 continue; 2248 oi->ts_present = 1; 2249 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2250 NTOHL(oi->ts_val); 2251 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2252 NTOHL(oi->ts_ecr); 2253 2254 /* 2255 * A timestamp received in a SYN makes 2256 * it ok to send timestamp requests and replies. 2257 */ 2258 if (th->th_flags & TH_SYN) { 2259 tp->t_flags |= TF_RCVD_TSTMP; 2260 tp->ts_recent = oi->ts_val; 2261 tp->ts_recent_age = tcp_now; 2262 } 2263 break; 2264 case TCPOPT_SACK_PERMITTED: 2265 if (optlen != TCPOLEN_SACK_PERMITTED) 2266 continue; 2267 if (!(th->th_flags & TH_SYN)) 2268 continue; 2269 tp->t_flags &= ~TF_CANT_TXSACK; 2270 break; 2271 2272 case TCPOPT_SACK: 2273 if (tp->t_flags & TF_IGNR_RXSACK) 2274 continue; 2275 if (optlen % 8 != 2 || optlen < 10) 2276 continue; 2277 cp += 2; 2278 optlen -= 2; 2279 for (; optlen > 0; cp -= 8, optlen -= 8) { 2280 tcp_seq lwe, rwe; 2281 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2282 NTOHL(lwe); 2283 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2284 NTOHL(rwe); 2285 /* tcp_mark_sacked(tp, lwe, rwe); */ 2286 } 2287 break; 2288 } 2289 } 2290 } 2291 2292 /* 2293 * Pull out of band byte out of a segment so 2294 * it doesn't appear in the user's data queue. 2295 * It is still reflected in the segment length for 2296 * sequencing purposes. 2297 */ 2298 void 2299 tcp_pulloutofband(so, th, m, off) 2300 struct socket *so; 2301 struct tcphdr *th; 2302 register struct mbuf *m; 2303 int off; 2304 { 2305 int cnt = off + th->th_urp - 1; 2306 2307 while (cnt >= 0) { 2308 if (m->m_len > cnt) { 2309 char *cp = mtod(m, caddr_t) + cnt; 2310 struct tcpcb *tp = sototcpcb(so); 2311 2312 tp->t_iobc = *cp; 2313 tp->t_oobflags |= TCPOOB_HAVEDATA; 2314 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2315 m->m_len--; 2316 return; 2317 } 2318 cnt -= m->m_len; 2319 m = m->m_next; 2320 if (m == 0) 2321 break; 2322 } 2323 panic("tcp_pulloutofband"); 2324 } 2325 2326 /* 2327 * Collect new round-trip time estimate 2328 * and update averages and current timeout. 2329 */ 2330 void 2331 tcp_xmit_timer(tp, rtt) 2332 register struct tcpcb *tp; 2333 short rtt; 2334 { 2335 register short delta; 2336 short rttmin; 2337 2338 tcpstat.tcps_rttupdated++; 2339 --rtt; 2340 if (tp->t_srtt != 0) { 2341 /* 2342 * srtt is stored as fixed point with 3 bits after the 2343 * binary point (i.e., scaled by 8). The following magic 2344 * is equivalent to the smoothing algorithm in rfc793 with 2345 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2346 * point). Adjust rtt to origin 0. 2347 */ 2348 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2349 if ((tp->t_srtt += delta) <= 0) 2350 tp->t_srtt = 1 << 2; 2351 /* 2352 * We accumulate a smoothed rtt variance (actually, a 2353 * smoothed mean difference), then set the retransmit 2354 * timer to smoothed rtt + 4 times the smoothed variance. 2355 * rttvar is stored as fixed point with 2 bits after the 2356 * binary point (scaled by 4). The following is 2357 * equivalent to rfc793 smoothing with an alpha of .75 2358 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2359 * rfc793's wired-in beta. 2360 */ 2361 if (delta < 0) 2362 delta = -delta; 2363 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2364 if ((tp->t_rttvar += delta) <= 0) 2365 tp->t_rttvar = 1 << 2; 2366 } else { 2367 /* 2368 * No rtt measurement yet - use the unsmoothed rtt. 2369 * Set the variance to half the rtt (so our first 2370 * retransmit happens at 3*rtt). 2371 */ 2372 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2373 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2374 } 2375 tp->t_rtt = 0; 2376 tp->t_rxtshift = 0; 2377 2378 /* 2379 * the retransmit should happen at rtt + 4 * rttvar. 2380 * Because of the way we do the smoothing, srtt and rttvar 2381 * will each average +1/2 tick of bias. When we compute 2382 * the retransmit timer, we want 1/2 tick of rounding and 2383 * 1 extra tick because of +-1/2 tick uncertainty in the 2384 * firing of the timer. The bias will give us exactly the 2385 * 1.5 tick we need. But, because the bias is 2386 * statistical, we have to test that we don't drop below 2387 * the minimum feasible timer (which is 2 ticks). 2388 */ 2389 if (tp->t_rttmin > rtt + 2) 2390 rttmin = tp->t_rttmin; 2391 else 2392 rttmin = rtt + 2; 2393 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2394 2395 /* 2396 * We received an ack for a packet that wasn't retransmitted; 2397 * it is probably safe to discard any error indications we've 2398 * received recently. This isn't quite right, but close enough 2399 * for now (a route might have failed after we sent a segment, 2400 * and the return path might not be symmetrical). 2401 */ 2402 tp->t_softerror = 0; 2403 } 2404 2405 /* 2406 * Checks for partial ack. If partial ack arrives, force the retransmission 2407 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2408 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2409 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2410 */ 2411 int 2412 tcp_newreno(tp, th) 2413 struct tcpcb *tp; 2414 struct tcphdr *th; 2415 { 2416 tcp_seq onxt = tp->snd_nxt; 2417 u_long ocwnd = tp->snd_cwnd; 2418 2419 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2420 /* 2421 * snd_una has not yet been updated and the socket's send 2422 * buffer has not yet drained off the ACK'd data, so we 2423 * have to leave snd_una as it was to get the correct data 2424 * offset in tcp_output(). 2425 */ 2426 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2427 tp->t_rtt = 0; 2428 tp->snd_nxt = th->th_ack; 2429 /* 2430 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2431 * is not yet updated when we're called. 2432 */ 2433 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2434 (void) tcp_output(tp); 2435 tp->snd_cwnd = ocwnd; 2436 if (SEQ_GT(onxt, tp->snd_nxt)) 2437 tp->snd_nxt = onxt; 2438 /* 2439 * Partial window deflation. Relies on fact that tp->snd_una 2440 * not updated yet. 2441 */ 2442 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2443 return 1; 2444 } 2445 return 0; 2446 } 2447 2448 2449 /* 2450 * TCP compressed state engine. Currently used to hold compressed 2451 * state for SYN_RECEIVED. 2452 */ 2453 2454 u_long syn_cache_count; 2455 u_int32_t syn_hash1, syn_hash2; 2456 2457 #define SYN_HASH(sa, sp, dp) \ 2458 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2459 ((u_int32_t)(sp)))^syn_hash2))) 2460 #ifndef INET6 2461 #define SYN_HASHALL(hash, src, dst) \ 2462 do { \ 2463 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2464 ((struct sockaddr_in *)(src))->sin_port, \ 2465 ((struct sockaddr_in *)(dst))->sin_port); \ 2466 } while (0) 2467 #else 2468 #define SYN_HASH6(sa, sp, dp) \ 2469 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2470 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2471 & 0x7fffffff) 2472 2473 #define SYN_HASHALL(hash, src, dst) \ 2474 do { \ 2475 switch ((src)->sa_family) { \ 2476 case AF_INET: \ 2477 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2478 ((struct sockaddr_in *)(src))->sin_port, \ 2479 ((struct sockaddr_in *)(dst))->sin_port); \ 2480 break; \ 2481 case AF_INET6: \ 2482 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2483 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2484 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2485 break; \ 2486 default: \ 2487 hash = 0; \ 2488 } \ 2489 } while (0) 2490 #endif /* INET6 */ 2491 2492 #define SYN_CACHE_RM(sc) \ 2493 do { \ 2494 LIST_REMOVE((sc), sc_bucketq); \ 2495 (sc)->sc_tp = NULL; \ 2496 LIST_REMOVE((sc), sc_tpq); \ 2497 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2498 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2499 syn_cache_count--; \ 2500 } while (0) 2501 2502 #define SYN_CACHE_PUT(sc) \ 2503 do { \ 2504 if ((sc)->sc_ipopts) \ 2505 (void) m_free((sc)->sc_ipopts); \ 2506 if ((sc)->sc_route4.ro_rt != NULL) \ 2507 RTFREE((sc)->sc_route4.ro_rt); \ 2508 pool_put(&syn_cache_pool, (sc)); \ 2509 } while (0) 2510 2511 struct pool syn_cache_pool; 2512 2513 /* 2514 * We don't estimate RTT with SYNs, so each packet starts with the default 2515 * RTT and each timer queue has a fixed timeout value. This allows us to 2516 * optimize the timer queues somewhat. 2517 */ 2518 #define SYN_CACHE_TIMER_ARM(sc) \ 2519 do { \ 2520 TCPT_RANGESET((sc)->sc_rxtcur, \ 2521 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2522 TCPTV_REXMTMAX); \ 2523 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2524 } while (0) 2525 2526 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2527 2528 void 2529 syn_cache_init() 2530 { 2531 int i; 2532 2533 /* Initialize the hash buckets. */ 2534 for (i = 0; i < tcp_syn_cache_size; i++) 2535 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2536 2537 /* Initialize the timer queues. */ 2538 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2539 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2540 2541 /* Initialize the syn cache pool. */ 2542 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2543 "synpl", 0, NULL, NULL, M_PCB); 2544 } 2545 2546 void 2547 syn_cache_insert(sc, tp) 2548 struct syn_cache *sc; 2549 struct tcpcb *tp; 2550 { 2551 struct syn_cache_head *scp; 2552 struct syn_cache *sc2; 2553 int s, i; 2554 2555 /* 2556 * If there are no entries in the hash table, reinitialize 2557 * the hash secrets. 2558 */ 2559 if (syn_cache_count == 0) { 2560 struct timeval tv; 2561 microtime(&tv); 2562 syn_hash1 = random() ^ (u_long)≻ 2563 syn_hash2 = random() ^ tv.tv_usec; 2564 } 2565 2566 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2567 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2568 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2569 2570 /* 2571 * Make sure that we don't overflow the per-bucket 2572 * limit or the total cache size limit. 2573 */ 2574 s = splsoftnet(); 2575 if (scp->sch_length >= tcp_syn_bucket_limit) { 2576 tcpstat.tcps_sc_bucketoverflow++; 2577 /* 2578 * The bucket is full. Toss the oldest element in the 2579 * bucket. This will be the entry with our bucket 2580 * index closest to the front of the timer queue with 2581 * the largest timeout value. 2582 * 2583 * Note: This timer queue traversal may be expensive, so 2584 * we hope that this doesn't happen very often. It is 2585 * much more likely that we'll overflow the entire 2586 * cache, which is much easier to handle; see below. 2587 */ 2588 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2589 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2590 sc2 != NULL; 2591 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2592 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2593 SYN_CACHE_RM(sc2); 2594 SYN_CACHE_PUT(sc2); 2595 goto insert; /* 2 level break */ 2596 } 2597 } 2598 } 2599 #ifdef DIAGNOSTIC 2600 /* 2601 * This should never happen; we should always find an 2602 * entry in our bucket. 2603 */ 2604 panic("syn_cache_insert: bucketoverflow: impossible"); 2605 #endif 2606 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2607 tcpstat.tcps_sc_overflowed++; 2608 /* 2609 * The cache is full. Toss the oldest entry in the 2610 * entire cache. This is the front entry in the 2611 * first non-empty timer queue with the largest 2612 * timeout value. 2613 */ 2614 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2615 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2616 if (sc2 == NULL) 2617 continue; 2618 SYN_CACHE_RM(sc2); 2619 SYN_CACHE_PUT(sc2); 2620 goto insert; /* symmetry with above */ 2621 } 2622 #ifdef DIAGNOSTIC 2623 /* 2624 * This should never happen; we should always find an 2625 * entry in the cache. 2626 */ 2627 panic("syn_cache_insert: cache overflow: impossible"); 2628 #endif 2629 } 2630 2631 insert: 2632 /* 2633 * Initialize the entry's timer. 2634 */ 2635 sc->sc_rxttot = 0; 2636 sc->sc_rxtshift = 0; 2637 SYN_CACHE_TIMER_ARM(sc); 2638 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2639 2640 /* Link it from tcpcb entry */ 2641 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2642 2643 /* Put it into the bucket. */ 2644 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2645 scp->sch_length++; 2646 syn_cache_count++; 2647 2648 tcpstat.tcps_sc_added++; 2649 splx(s); 2650 } 2651 2652 /* 2653 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2654 * If we have retransmitted an entry the maximum number of times, expire 2655 * that entry. 2656 */ 2657 void 2658 syn_cache_timer() 2659 { 2660 struct syn_cache *sc, *nsc; 2661 int i, s; 2662 2663 s = splsoftnet(); 2664 2665 /* 2666 * First, get all the entries that need to be retransmitted, or 2667 * must be expired due to exceeding the initial keepalive time. 2668 */ 2669 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2670 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2671 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2672 sc = nsc) { 2673 nsc = TAILQ_NEXT(sc, sc_timeq); 2674 2675 /* 2676 * Compute the total amount of time this entry has 2677 * been on a queue. If this entry has been on longer 2678 * than the keep alive timer would allow, expire it. 2679 */ 2680 sc->sc_rxttot += sc->sc_rxtcur; 2681 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2682 tcpstat.tcps_sc_timed_out++; 2683 SYN_CACHE_RM(sc); 2684 SYN_CACHE_PUT(sc); 2685 continue; 2686 } 2687 2688 tcpstat.tcps_sc_retransmitted++; 2689 (void) syn_cache_respond(sc, NULL); 2690 2691 /* Advance this entry onto the next timer queue. */ 2692 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2693 sc->sc_rxtshift = i + 1; 2694 SYN_CACHE_TIMER_ARM(sc); 2695 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2696 sc, sc_timeq); 2697 } 2698 } 2699 2700 /* 2701 * Now get all the entries that are expired due to too many 2702 * retransmissions. 2703 */ 2704 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2705 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2706 sc = nsc) { 2707 nsc = TAILQ_NEXT(sc, sc_timeq); 2708 tcpstat.tcps_sc_timed_out++; 2709 SYN_CACHE_RM(sc); 2710 SYN_CACHE_PUT(sc); 2711 } 2712 splx(s); 2713 } 2714 2715 /* 2716 * Remove syn cache created by the specified tcb entry, 2717 * because this does not make sense to keep them 2718 * (if there's no tcb entry, syn cache entry will never be used) 2719 */ 2720 void 2721 syn_cache_cleanup(tp) 2722 struct tcpcb *tp; 2723 { 2724 struct syn_cache *sc, *nsc; 2725 int s; 2726 2727 s = splsoftnet(); 2728 2729 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2730 nsc = LIST_NEXT(sc, sc_tpq); 2731 2732 #ifdef DIAGNOSTIC 2733 if (sc->sc_tp != tp) 2734 panic("invalid sc_tp in syn_cache_cleanup"); 2735 #endif 2736 SYN_CACHE_RM(sc); 2737 SYN_CACHE_PUT(sc); 2738 } 2739 /* just for safety */ 2740 LIST_INIT(&tp->t_sc); 2741 2742 splx(s); 2743 } 2744 2745 /* 2746 * Find an entry in the syn cache. 2747 */ 2748 struct syn_cache * 2749 syn_cache_lookup(src, dst, headp) 2750 struct sockaddr *src; 2751 struct sockaddr *dst; 2752 struct syn_cache_head **headp; 2753 { 2754 struct syn_cache *sc; 2755 struct syn_cache_head *scp; 2756 u_int32_t hash; 2757 int s; 2758 2759 SYN_HASHALL(hash, src, dst); 2760 2761 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2762 *headp = scp; 2763 s = splsoftnet(); 2764 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2765 sc = LIST_NEXT(sc, sc_bucketq)) { 2766 if (sc->sc_hash != hash) 2767 continue; 2768 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2769 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2770 splx(s); 2771 return (sc); 2772 } 2773 } 2774 splx(s); 2775 return (NULL); 2776 } 2777 2778 /* 2779 * This function gets called when we receive an ACK for a 2780 * socket in the LISTEN state. We look up the connection 2781 * in the syn cache, and if its there, we pull it out of 2782 * the cache and turn it into a full-blown connection in 2783 * the SYN-RECEIVED state. 2784 * 2785 * The return values may not be immediately obvious, and their effects 2786 * can be subtle, so here they are: 2787 * 2788 * NULL SYN was not found in cache; caller should drop the 2789 * packet and send an RST. 2790 * 2791 * -1 We were unable to create the new connection, and are 2792 * aborting it. An ACK,RST is being sent to the peer 2793 * (unless we got screwey sequence numbners; see below), 2794 * because the 3-way handshake has been completed. Caller 2795 * should not free the mbuf, since we may be using it. If 2796 * we are not, we will free it. 2797 * 2798 * Otherwise, the return value is a pointer to the new socket 2799 * associated with the connection. 2800 */ 2801 struct socket * 2802 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2803 struct sockaddr *src; 2804 struct sockaddr *dst; 2805 struct tcphdr *th; 2806 unsigned int hlen, tlen; 2807 struct socket *so; 2808 struct mbuf *m; 2809 { 2810 struct syn_cache *sc; 2811 struct syn_cache_head *scp; 2812 register struct inpcb *inp = NULL; 2813 #ifdef INET6 2814 register struct in6pcb *in6p = NULL; 2815 #endif 2816 register struct tcpcb *tp = 0; 2817 struct mbuf *am; 2818 int s; 2819 struct socket *oso; 2820 2821 s = splsoftnet(); 2822 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2823 splx(s); 2824 return (NULL); 2825 } 2826 2827 /* 2828 * Verify the sequence and ack numbers. Try getting the correct 2829 * response again. 2830 */ 2831 if ((th->th_ack != sc->sc_iss + 1) || 2832 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2833 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2834 (void) syn_cache_respond(sc, m); 2835 splx(s); 2836 return ((struct socket *)(-1)); 2837 } 2838 2839 /* Remove this cache entry */ 2840 SYN_CACHE_RM(sc); 2841 splx(s); 2842 2843 /* 2844 * Ok, create the full blown connection, and set things up 2845 * as they would have been set up if we had created the 2846 * connection when the SYN arrived. If we can't create 2847 * the connection, abort it. 2848 */ 2849 /* 2850 * inp still has the OLD in_pcb stuff, set the 2851 * v6-related flags on the new guy, too. This is 2852 * done particularly for the case where an AF_INET6 2853 * socket is bound only to a port, and a v4 connection 2854 * comes in on that port. 2855 * we also copy the flowinfo from the original pcb 2856 * to the new one. 2857 */ 2858 { 2859 struct inpcb *parentinpcb; 2860 2861 parentinpcb = (struct inpcb *)so->so_pcb; 2862 2863 oso = so; 2864 so = sonewconn(so, SS_ISCONNECTED); 2865 if (so == NULL) 2866 goto resetandabort; 2867 2868 switch (so->so_proto->pr_domain->dom_family) { 2869 case AF_INET: 2870 inp = sotoinpcb(so); 2871 break; 2872 #ifdef INET6 2873 case AF_INET6: 2874 in6p = sotoin6pcb(so); 2875 #if 0 /*def INET6*/ 2876 inp->inp_flags |= (parentinpcb->inp_flags & 2877 (INP_IPV6 | INP_IPV6_UNDEC | INP_IPV6_MAPPED)); 2878 if ((inp->inp_flags & INP_IPV6) && 2879 !(inp->inp_flags & INP_IPV6_MAPPED)) { 2880 inp->inp_ipv6.ip6_hlim = parentinpcb->inp_ipv6.ip6_hlim; 2881 inp->inp_ipv6.ip6_vfc = parentinpcb->inp_ipv6.ip6_vfc; 2882 } 2883 #endif 2884 break; 2885 #endif 2886 } 2887 } 2888 switch (src->sa_family) { 2889 case AF_INET: 2890 if (inp) { 2891 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2892 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2893 inp->inp_options = ip_srcroute(); 2894 in_pcbstate(inp, INP_BOUND); 2895 if (inp->inp_options == NULL) { 2896 inp->inp_options = sc->sc_ipopts; 2897 sc->sc_ipopts = NULL; 2898 } 2899 } 2900 #ifdef INET6 2901 else if (in6p) { 2902 /* IPv4 packet to AF_INET6 socket */ 2903 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2904 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2905 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2906 &in6p->in6p_laddr.s6_addr32[3], 2907 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2908 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2909 in6totcpcb(in6p)->t_family = AF_INET; 2910 } 2911 #endif 2912 break; 2913 #ifdef INET6 2914 case AF_INET6: 2915 if (in6p) { 2916 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 2917 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 2918 #if 0 2919 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 2920 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 2921 #endif 2922 } 2923 break; 2924 #endif 2925 } 2926 #ifdef INET6 2927 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 2928 struct in6pcb *oin6p = sotoin6pcb(oso); 2929 /* inherit socket options from the listening socket */ 2930 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 2931 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 2932 m_freem(in6p->in6p_options); 2933 in6p->in6p_options = 0; 2934 } 2935 ip6_savecontrol(in6p, &in6p->in6p_options, 2936 mtod(m, struct ip6_hdr *), m); 2937 } 2938 #endif 2939 2940 #ifdef IPSEC 2941 /* 2942 * we make a copy of policy, instead of sharing the policy, 2943 * for better behavior in terms of SA lookup and dead SA removal. 2944 */ 2945 if (inp) { 2946 /* copy old policy into new socket's */ 2947 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 2948 printf("tcp_input: could not copy policy\n"); 2949 } 2950 #ifdef INET6 2951 else if (in6p) { 2952 /* copy old policy into new socket's */ 2953 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 2954 printf("tcp_input: could not copy policy\n"); 2955 } 2956 #endif 2957 #endif 2958 2959 /* 2960 * Give the new socket our cached route reference. 2961 */ 2962 if (inp) 2963 inp->inp_route = sc->sc_route4; /* struct assignment */ 2964 #ifdef INET6 2965 else 2966 in6p->in6p_route = sc->sc_route6; 2967 #endif 2968 sc->sc_route4.ro_rt = NULL; 2969 2970 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 2971 if (am == NULL) 2972 goto resetandabort; 2973 am->m_len = src->sa_len; 2974 bcopy(src, mtod(am, caddr_t), src->sa_len); 2975 if (inp) { 2976 if (in_pcbconnect(inp, am)) { 2977 (void) m_free(am); 2978 goto resetandabort; 2979 } 2980 } 2981 #ifdef INET6 2982 else if (in6p) { 2983 if (src->sa_family == AF_INET) { 2984 /* IPv4 packet to AF_INET6 socket */ 2985 struct sockaddr_in6 *sin6; 2986 sin6 = mtod(am, struct sockaddr_in6 *); 2987 am->m_len = sizeof(*sin6); 2988 bzero(sin6, sizeof(*sin6)); 2989 sin6->sin6_family = AF_INET6; 2990 sin6->sin6_len = sizeof(*sin6); 2991 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 2992 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 2993 bcopy(&((struct sockaddr_in *)src)->sin_addr, 2994 &sin6->sin6_addr.s6_addr32[3], 2995 sizeof(sin6->sin6_addr.s6_addr32[3])); 2996 } 2997 if (in6_pcbconnect(in6p, am)) { 2998 (void) m_free(am); 2999 goto resetandabort; 3000 } 3001 } 3002 #endif 3003 else { 3004 (void) m_free(am); 3005 goto resetandabort; 3006 } 3007 (void) m_free(am); 3008 3009 if (inp) 3010 tp = intotcpcb(inp); 3011 #ifdef INET6 3012 else if (in6p) 3013 tp = in6totcpcb(in6p); 3014 #endif 3015 else 3016 tp = NULL; 3017 if (sc->sc_request_r_scale != 15) { 3018 tp->requested_s_scale = sc->sc_requested_s_scale; 3019 tp->request_r_scale = sc->sc_request_r_scale; 3020 tp->snd_scale = sc->sc_requested_s_scale; 3021 tp->rcv_scale = sc->sc_request_r_scale; 3022 tp->t_flags |= TF_RCVD_SCALE; 3023 } 3024 if (sc->sc_flags & SCF_TIMESTAMP) 3025 tp->t_flags |= TF_RCVD_TSTMP; 3026 3027 tp->t_template = tcp_template(tp); 3028 if (tp->t_template == 0) { 3029 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3030 so = NULL; 3031 m_freem(m); 3032 goto abort; 3033 } 3034 3035 tp->iss = sc->sc_iss; 3036 tp->irs = sc->sc_irs; 3037 tcp_sendseqinit(tp); 3038 tcp_rcvseqinit(tp); 3039 tp->t_state = TCPS_SYN_RECEIVED; 3040 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3041 tcpstat.tcps_accepts++; 3042 3043 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3044 tp->t_ourmss = sc->sc_ourmaxseg; 3045 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3046 3047 /* 3048 * Initialize the initial congestion window. If we 3049 * had to retransmit the SYN,ACK, we must initialize cwnd 3050 * to 1 segment (i.e. the Loss Window). 3051 */ 3052 if (sc->sc_rxtshift) 3053 tp->snd_cwnd = tp->t_peermss; 3054 else 3055 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3056 3057 tcp_rmx_rtt(tp); 3058 tp->snd_wl1 = sc->sc_irs; 3059 tp->rcv_up = sc->sc_irs + 1; 3060 3061 /* 3062 * This is what whould have happened in tcp_ouput() when 3063 * the SYN,ACK was sent. 3064 */ 3065 tp->snd_up = tp->snd_una; 3066 tp->snd_max = tp->snd_nxt = tp->iss+1; 3067 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3068 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3069 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3070 tp->last_ack_sent = tp->rcv_nxt; 3071 3072 tcpstat.tcps_sc_completed++; 3073 SYN_CACHE_PUT(sc); 3074 return (so); 3075 3076 resetandabort: 3077 (void) tcp_respond(NULL, m, m, th, 3078 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3079 abort: 3080 if (so != NULL) 3081 (void) soabort(so); 3082 SYN_CACHE_PUT(sc); 3083 tcpstat.tcps_sc_aborted++; 3084 return ((struct socket *)(-1)); 3085 } 3086 3087 /* 3088 * This function is called when we get a RST for a 3089 * non-existant connection, so that we can see if the 3090 * connection is in the syn cache. If it is, zap it. 3091 */ 3092 3093 void 3094 syn_cache_reset(src, dst, th) 3095 struct sockaddr *src; 3096 struct sockaddr *dst; 3097 struct tcphdr *th; 3098 { 3099 struct syn_cache *sc; 3100 struct syn_cache_head *scp; 3101 int s = splsoftnet(); 3102 3103 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3104 splx(s); 3105 return; 3106 } 3107 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3108 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3109 splx(s); 3110 return; 3111 } 3112 SYN_CACHE_RM(sc); 3113 splx(s); 3114 tcpstat.tcps_sc_reset++; 3115 SYN_CACHE_PUT(sc); 3116 } 3117 3118 void 3119 syn_cache_unreach(src, dst, th) 3120 struct sockaddr *src; 3121 struct sockaddr *dst; 3122 struct tcphdr *th; 3123 { 3124 struct syn_cache *sc; 3125 struct syn_cache_head *scp; 3126 int s; 3127 3128 s = splsoftnet(); 3129 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3130 splx(s); 3131 return; 3132 } 3133 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3134 if (ntohl (th->th_seq) != sc->sc_iss) { 3135 splx(s); 3136 return; 3137 } 3138 3139 /* 3140 * If we've rertransmitted 3 times and this is our second error, 3141 * we remove the entry. Otherwise, we allow it to continue on. 3142 * This prevents us from incorrectly nuking an entry during a 3143 * spurious network outage. 3144 * 3145 * See tcp_notify(). 3146 */ 3147 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3148 sc->sc_flags |= SCF_UNREACH; 3149 splx(s); 3150 return; 3151 } 3152 3153 SYN_CACHE_RM(sc); 3154 splx(s); 3155 tcpstat.tcps_sc_unreach++; 3156 SYN_CACHE_PUT(sc); 3157 } 3158 3159 /* 3160 * Given a LISTEN socket and an inbound SYN request, add 3161 * this to the syn cache, and send back a segment: 3162 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3163 * to the source. 3164 * 3165 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3166 * Doing so would require that we hold onto the data and deliver it 3167 * to the application. However, if we are the target of a SYN-flood 3168 * DoS attack, an attacker could send data which would eventually 3169 * consume all available buffer space if it were ACKed. By not ACKing 3170 * the data, we avoid this DoS scenario. 3171 */ 3172 3173 int 3174 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3175 struct sockaddr *src; 3176 struct sockaddr *dst; 3177 struct tcphdr *th; 3178 unsigned int hlen; 3179 struct socket *so; 3180 struct mbuf *m; 3181 u_char *optp; 3182 int optlen; 3183 struct tcp_opt_info *oi; 3184 { 3185 struct tcpcb tb, *tp; 3186 long win; 3187 struct syn_cache *sc; 3188 struct syn_cache_head *scp; 3189 struct mbuf *ipopts; 3190 3191 tp = sototcpcb(so); 3192 3193 /* 3194 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3195 * 3196 * Note this check is performed in tcp_input() very early on. 3197 */ 3198 3199 /* 3200 * Initialize some local state. 3201 */ 3202 win = sbspace(&so->so_rcv); 3203 if (win > TCP_MAXWIN) 3204 win = TCP_MAXWIN; 3205 3206 if (src->sa_family == AF_INET) { 3207 /* 3208 * Remember the IP options, if any. 3209 */ 3210 ipopts = ip_srcroute(); 3211 } else 3212 ipopts = NULL; 3213 3214 if (optp) { 3215 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3216 tcp_dooptions(&tb, optp, optlen, th, oi); 3217 } else 3218 tb.t_flags = 0; 3219 3220 /* 3221 * See if we already have an entry for this connection. 3222 * If we do, resend the SYN,ACK. We do not count this 3223 * as a retransmission (XXX though maybe we should). 3224 */ 3225 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3226 tcpstat.tcps_sc_dupesyn++; 3227 if (ipopts) { 3228 /* 3229 * If we were remembering a previous source route, 3230 * forget it and use the new one we've been given. 3231 */ 3232 if (sc->sc_ipopts) 3233 (void) m_free(sc->sc_ipopts); 3234 sc->sc_ipopts = ipopts; 3235 } 3236 sc->sc_timestamp = tb.ts_recent; 3237 if (syn_cache_respond(sc, m) == 0) { 3238 tcpstat.tcps_sndacks++; 3239 tcpstat.tcps_sndtotal++; 3240 } 3241 return (1); 3242 } 3243 3244 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3245 if (sc == NULL) { 3246 if (ipopts) 3247 (void) m_free(ipopts); 3248 return (0); 3249 } 3250 3251 /* 3252 * Fill in the cache, and put the necessary IP and TCP 3253 * options into the reply. 3254 */ 3255 bzero(sc, sizeof(struct syn_cache)); 3256 bcopy(src, &sc->sc_src, src->sa_len); 3257 bcopy(dst, &sc->sc_dst, dst->sa_len); 3258 sc->sc_flags = 0; 3259 sc->sc_ipopts = ipopts; 3260 sc->sc_irs = th->th_seq; 3261 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 3262 sc->sc_peermaxseg = oi->maxseg; 3263 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3264 m->m_pkthdr.rcvif : NULL, 3265 sc->sc_src.sa.sa_family); 3266 sc->sc_win = win; 3267 sc->sc_timestamp = tb.ts_recent; 3268 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3269 sc->sc_flags |= SCF_TIMESTAMP; 3270 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3271 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3272 sc->sc_requested_s_scale = tb.requested_s_scale; 3273 sc->sc_request_r_scale = 0; 3274 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3275 TCP_MAXWIN << sc->sc_request_r_scale < 3276 so->so_rcv.sb_hiwat) 3277 sc->sc_request_r_scale++; 3278 } else { 3279 sc->sc_requested_s_scale = 15; 3280 sc->sc_request_r_scale = 15; 3281 } 3282 sc->sc_tp = tp; 3283 if (syn_cache_respond(sc, m) == 0) { 3284 syn_cache_insert(sc, tp); 3285 tcpstat.tcps_sndacks++; 3286 tcpstat.tcps_sndtotal++; 3287 } else { 3288 SYN_CACHE_PUT(sc); 3289 tcpstat.tcps_sc_dropped++; 3290 } 3291 return (1); 3292 } 3293 3294 int 3295 syn_cache_respond(sc, m) 3296 struct syn_cache *sc; 3297 struct mbuf *m; 3298 { 3299 struct route *ro; 3300 struct rtentry *rt; 3301 u_int8_t *optp; 3302 int optlen, error; 3303 u_int16_t tlen; 3304 struct ip *ip = NULL; 3305 #ifdef INET6 3306 struct ip6_hdr *ip6 = NULL; 3307 #endif 3308 struct tcphdr *th; 3309 u_int hlen; 3310 3311 switch (sc->sc_src.sa.sa_family) { 3312 case AF_INET: 3313 hlen = sizeof(struct ip); 3314 ro = &sc->sc_route4; 3315 break; 3316 #ifdef INET6 3317 case AF_INET6: 3318 hlen = sizeof(struct ip6_hdr); 3319 ro = (struct route *)&sc->sc_route6; 3320 break; 3321 #endif 3322 default: 3323 if (m) 3324 m_freem(m); 3325 return EAFNOSUPPORT; 3326 } 3327 3328 /* Compute the size of the TCP options. */ 3329 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3330 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3331 3332 tlen = hlen + sizeof(struct tcphdr) + optlen; 3333 3334 /* 3335 * Create the IP+TCP header from scratch. Reuse the received mbuf 3336 * if possible. 3337 */ 3338 if (m != NULL) { 3339 m_freem(m->m_next); 3340 m->m_next = NULL; 3341 MRESETDATA(m); 3342 } else { 3343 MGETHDR(m, M_DONTWAIT, MT_DATA); 3344 if (m == NULL) 3345 return (ENOBUFS); 3346 } 3347 3348 /* Fixup the mbuf. */ 3349 m->m_data += max_linkhdr; 3350 m->m_len = m->m_pkthdr.len = tlen; 3351 #ifdef IPSEC 3352 if (sc->sc_tp) { 3353 struct tcpcb *tp; 3354 struct socket *so; 3355 3356 tp = sc->sc_tp; 3357 if (tp->t_inpcb) 3358 so = tp->t_inpcb->inp_socket; 3359 #ifdef INET6 3360 else if (tp->t_in6pcb) 3361 so = tp->t_in6pcb->in6p_socket; 3362 #endif 3363 else 3364 so = NULL; 3365 /* use IPsec policy on listening socket, on SYN ACK */ 3366 m->m_pkthdr.rcvif = (struct ifnet *)so; 3367 } 3368 #else 3369 m->m_pkthdr.rcvif = NULL; 3370 #endif 3371 memset(mtod(m, u_char *), 0, tlen); 3372 3373 switch (sc->sc_src.sa.sa_family) { 3374 case AF_INET: 3375 ip = mtod(m, struct ip *); 3376 ip->ip_dst = sc->sc_src.sin.sin_addr; 3377 ip->ip_src = sc->sc_dst.sin.sin_addr; 3378 ip->ip_p = IPPROTO_TCP; 3379 th = (struct tcphdr *)(ip + 1); 3380 th->th_dport = sc->sc_src.sin.sin_port; 3381 th->th_sport = sc->sc_dst.sin.sin_port; 3382 break; 3383 #ifdef INET6 3384 case AF_INET6: 3385 ip6 = mtod(m, struct ip6_hdr *); 3386 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3387 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3388 ip6->ip6_nxt = IPPROTO_TCP; 3389 /* ip6_plen will be updated in ip6_output() */ 3390 th = (struct tcphdr *)(ip6 + 1); 3391 th->th_dport = sc->sc_src.sin6.sin6_port; 3392 th->th_sport = sc->sc_dst.sin6.sin6_port; 3393 break; 3394 #endif 3395 default: 3396 th = NULL; 3397 } 3398 3399 th->th_seq = htonl(sc->sc_iss); 3400 th->th_ack = htonl(sc->sc_irs + 1); 3401 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3402 th->th_flags = TH_SYN|TH_ACK; 3403 th->th_win = htons(sc->sc_win); 3404 /* th_sum already 0 */ 3405 /* th_urp already 0 */ 3406 3407 /* Tack on the TCP options. */ 3408 optp = (u_int8_t *)(th + 1); 3409 *optp++ = TCPOPT_MAXSEG; 3410 *optp++ = 4; 3411 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3412 *optp++ = sc->sc_ourmaxseg & 0xff; 3413 3414 if (sc->sc_request_r_scale != 15) { 3415 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3416 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3417 sc->sc_request_r_scale); 3418 optp += 4; 3419 } 3420 3421 if (sc->sc_flags & SCF_TIMESTAMP) { 3422 u_int32_t *lp = (u_int32_t *)(optp); 3423 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3424 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3425 *lp++ = htonl(tcp_now); 3426 *lp = htonl(sc->sc_timestamp); 3427 optp += TCPOLEN_TSTAMP_APPA; 3428 } 3429 3430 /* Compute the packet's checksum. */ 3431 switch (sc->sc_src.sa.sa_family) { 3432 case AF_INET: 3433 ip->ip_len = htons(tlen - hlen); 3434 th->th_sum = 0; 3435 th->th_sum = in_cksum(m, tlen); 3436 break; 3437 #ifdef INET6 3438 case AF_INET6: 3439 ip6->ip6_plen = htons(tlen - hlen); 3440 th->th_sum = 0; 3441 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3442 break; 3443 #endif 3444 } 3445 3446 /* 3447 * Fill in some straggling IP bits. Note the stack expects 3448 * ip_len to be in host order, for convenience. 3449 */ 3450 switch (sc->sc_src.sa.sa_family) { 3451 case AF_INET: 3452 ip->ip_len = tlen; 3453 ip->ip_ttl = ip_defttl; 3454 /* XXX tos? */ 3455 break; 3456 #ifdef INET6 3457 case AF_INET6: 3458 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3459 ip6->ip6_vfc |= IPV6_VERSION; 3460 ip6->ip6_plen = htons(tlen - hlen); 3461 /* ip6_hlim will be initialized afterwards */ 3462 /* XXX flowlabel? */ 3463 break; 3464 #endif 3465 } 3466 3467 /* 3468 * If we're doing Path MTU discovery, we need to set DF unless 3469 * the route's MTU is locked. If we don't yet know the route, 3470 * look it up now. We will copy this reference to the inpcb 3471 * when we finish creating the connection. 3472 */ 3473 if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) { 3474 if (ro->ro_rt != NULL) { 3475 RTFREE(ro->ro_rt); 3476 ro->ro_rt = NULL; 3477 } 3478 bcopy(&sc->sc_src, &ro->ro_dst, sc->sc_src.sa.sa_len); 3479 rtalloc(ro); 3480 if ((rt = ro->ro_rt) == NULL) { 3481 m_freem(m); 3482 switch (sc->sc_src.sa.sa_family) { 3483 case AF_INET: 3484 ipstat.ips_noroute++; 3485 break; 3486 #ifdef INET6 3487 case AF_INET6: 3488 ip6stat.ip6s_noroute++; 3489 break; 3490 #endif 3491 } 3492 return (EHOSTUNREACH); 3493 } 3494 } 3495 3496 switch (sc->sc_src.sa.sa_family) { 3497 case AF_INET: 3498 if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0) 3499 ip->ip_off |= IP_DF; 3500 3501 /* ...and send it off! */ 3502 error = ip_output(m, sc->sc_ipopts, ro, 0, NULL); 3503 break; 3504 #ifdef INET6 3505 case AF_INET6: 3506 ip6->ip6_hlim = in6_selecthlim(NULL, 3507 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3508 3509 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3510 0, NULL, NULL); 3511 break; 3512 #endif 3513 default: 3514 error = EAFNOSUPPORT; 3515 break; 3516 } 3517 return (error); 3518 } 3519
Indexes created Thu Sep 25 14:09:45 GMT 2025