tcp_input.c revision 1.90
1 /* $NetBSD: tcp_input.c,v 1.90 1999/08/11 03:02:18 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 #endif 162 163 #include <netinet/tcp.h> 164 #include <netinet/tcp_fsm.h> 165 #include <netinet/tcp_seq.h> 166 #include <netinet/tcp_timer.h> 167 #include <netinet/tcp_var.h> 168 #include <netinet/tcpip.h> 169 #include <netinet/tcp_debug.h> 170 171 #include <machine/stdarg.h> 172 173 #ifdef IPSEC 174 #include <netinet6/ipsec.h> 175 #include <netkey/key.h> 176 #include <netkey/key_debug.h> 177 #endif /*IPSEC*/ 178 #ifdef INET6 179 #include "faith.h" 180 #endif 181 182 int tcprexmtthresh = 3; 183 int tcp_log_refused; 184 185 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 186 187 /* for modulo comparisons of timestamps */ 188 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 189 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 190 191 /* 192 * Macro to compute ACK transmission behavior. Delay the ACK unless 193 * we have already delayed an ACK (must send an ACK every two segments). 194 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 195 * option is enabled. 196 */ 197 #define TCP_SETUP_ACK(tp, th) \ 198 do { \ 199 if ((tp)->t_flags & TF_DELACK || \ 200 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 201 tp->t_flags |= TF_ACKNOW; \ 202 else \ 203 TCP_SET_DELACK(tp); \ 204 } while (0) 205 206 int 207 tcp_reass(tp, th, m, tlen) 208 register struct tcpcb *tp; 209 register struct tcphdr *th; 210 struct mbuf *m; 211 int *tlen; 212 { 213 register struct ipqent *p, *q, *nq, *tiqe = NULL; 214 struct socket *so = NULL; 215 int pkt_flags; 216 tcp_seq pkt_seq; 217 unsigned pkt_len; 218 u_long rcvpartdupbyte = 0; 219 u_long rcvoobyte; 220 221 if (tp->t_inpcb) 222 so = tp->t_inpcb->inp_socket; 223 #ifdef INET6 224 else if (tp->t_in6pcb) 225 so = tp->t_in6pcb->in6p_socket; 226 #endif 227 228 TCP_REASS_LOCK_CHECK(tp); 229 230 /* 231 * Call with th==0 after become established to 232 * force pre-ESTABLISHED data up to user socket. 233 */ 234 if (th == 0) 235 goto present; 236 237 rcvoobyte = *tlen; 238 /* 239 * Copy these to local variables because the tcpiphdr 240 * gets munged while we are collapsing mbufs. 241 */ 242 pkt_seq = th->th_seq; 243 pkt_len = *tlen; 244 pkt_flags = th->th_flags; 245 /* 246 * Find a segment which begins after this one does. 247 */ 248 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 249 nq = q->ipqe_q.le_next; 250 /* 251 * If the received segment is just right after this 252 * fragment, merge the two together and then check 253 * for further overlaps. 254 */ 255 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 256 #ifdef TCPREASS_DEBUG 257 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 258 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 259 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 260 #endif 261 pkt_len += q->ipqe_len; 262 pkt_flags |= q->ipqe_flags; 263 pkt_seq = q->ipqe_seq; 264 m_cat(q->ipqe_m, m); 265 m = q->ipqe_m; 266 goto free_ipqe; 267 } 268 /* 269 * If the received segment is completely past this 270 * fragment, we need to go the next fragment. 271 */ 272 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 273 p = q; 274 continue; 275 } 276 /* 277 * If the fragment is past the received segment, 278 * it (or any following) can't be concatenated. 279 */ 280 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 281 break; 282 /* 283 * We've received all the data in this segment before. 284 * mark it as a duplicate and return. 285 */ 286 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 287 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 288 tcpstat.tcps_rcvduppack++; 289 tcpstat.tcps_rcvdupbyte += pkt_len; 290 m_freem(m); 291 if (tiqe != NULL) 292 pool_put(&ipqent_pool, tiqe); 293 return (0); 294 } 295 /* 296 * Received segment completely overlaps this fragment 297 * so we drop the fragment (this keeps the temporal 298 * ordering of segments correct). 299 */ 300 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 301 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 302 rcvpartdupbyte += q->ipqe_len; 303 m_freem(q->ipqe_m); 304 goto free_ipqe; 305 } 306 /* 307 * RX'ed segment extends past the end of the 308 * fragment. Drop the overlapping bytes. Then 309 * merge the fragment and segment then treat as 310 * a longer received packet. 311 */ 312 if (SEQ_LT(q->ipqe_seq, pkt_seq) 313 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 314 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 315 #ifdef TCPREASS_DEBUG 316 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 317 tp, overlap, 318 pkt_seq, pkt_seq + pkt_len, pkt_len); 319 #endif 320 m_adj(m, overlap); 321 rcvpartdupbyte += overlap; 322 m_cat(q->ipqe_m, m); 323 m = q->ipqe_m; 324 pkt_seq = q->ipqe_seq; 325 pkt_len += q->ipqe_len - overlap; 326 rcvoobyte -= overlap; 327 goto free_ipqe; 328 } 329 /* 330 * RX'ed segment extends past the front of the 331 * fragment. Drop the overlapping bytes on the 332 * received packet. The packet will then be 333 * contatentated with this fragment a bit later. 334 */ 335 if (SEQ_GT(q->ipqe_seq, pkt_seq) 336 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 337 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 338 #ifdef TCPREASS_DEBUG 339 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 340 tp, overlap, 341 pkt_seq, pkt_seq + pkt_len, pkt_len); 342 #endif 343 m_adj(m, -overlap); 344 pkt_len -= overlap; 345 rcvpartdupbyte += overlap; 346 rcvoobyte -= overlap; 347 } 348 /* 349 * If the received segment immediates precedes this 350 * fragment then tack the fragment onto this segment 351 * and reinsert the data. 352 */ 353 if (q->ipqe_seq == pkt_seq + pkt_len) { 354 #ifdef TCPREASS_DEBUG 355 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 356 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 357 pkt_seq, pkt_seq + pkt_len, pkt_len); 358 #endif 359 pkt_len += q->ipqe_len; 360 pkt_flags |= q->ipqe_flags; 361 m_cat(m, q->ipqe_m); 362 LIST_REMOVE(q, ipqe_q); 363 LIST_REMOVE(q, ipqe_timeq); 364 if (tiqe == NULL) { 365 tiqe = q; 366 } else { 367 pool_put(&ipqent_pool, q); 368 } 369 break; 370 } 371 /* 372 * If the fragment is before the segment, remember it. 373 * When this loop is terminated, p will contain the 374 * pointer to fragment that is right before the received 375 * segment. 376 */ 377 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 378 p = q; 379 380 continue; 381 382 /* 383 * This is a common operation. It also will allow 384 * to save doing a malloc/free in most instances. 385 */ 386 free_ipqe: 387 LIST_REMOVE(q, ipqe_q); 388 LIST_REMOVE(q, ipqe_timeq); 389 if (tiqe == NULL) { 390 tiqe = q; 391 } else { 392 pool_put(&ipqent_pool, q); 393 } 394 } 395 396 /* 397 * Allocate a new queue entry since the received segment did not 398 * collapse onto any other out-of-order block; thus we are allocating 399 * a new block. If it had collapsed, tiqe would not be NULL and 400 * we would be reusing it. 401 * XXX If we can't, just drop the packet. XXX 402 */ 403 if (tiqe == NULL) { 404 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 405 if (tiqe == NULL) { 406 tcpstat.tcps_rcvmemdrop++; 407 m_freem(m); 408 return (0); 409 } 410 } 411 412 /* 413 * Update the counters. 414 */ 415 tcpstat.tcps_rcvoopack++; 416 tcpstat.tcps_rcvoobyte += rcvoobyte; 417 if (rcvpartdupbyte) { 418 tcpstat.tcps_rcvpartduppack++; 419 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 420 } 421 422 /* 423 * Insert the new fragment queue entry into both queues. 424 */ 425 tiqe->ipqe_m = m; 426 tiqe->ipqe_seq = pkt_seq; 427 tiqe->ipqe_len = pkt_len; 428 tiqe->ipqe_flags = pkt_flags; 429 if (p == NULL) { 430 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 431 #ifdef TCPREASS_DEBUG 432 if (tiqe->ipqe_seq != tp->rcv_nxt) 433 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 434 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 435 #endif 436 } else { 437 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 438 #ifdef TCPREASS_DEBUG 439 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 440 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 441 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 442 #endif 443 } 444 445 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 446 447 present: 448 /* 449 * Present data to user, advancing rcv_nxt through 450 * completed sequence space. 451 */ 452 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 453 return (0); 454 q = tp->segq.lh_first; 455 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 456 return (0); 457 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 458 return (0); 459 460 tp->rcv_nxt += q->ipqe_len; 461 pkt_flags = q->ipqe_flags & TH_FIN; 462 463 LIST_REMOVE(q, ipqe_q); 464 LIST_REMOVE(q, ipqe_timeq); 465 if (so->so_state & SS_CANTRCVMORE) 466 m_freem(q->ipqe_m); 467 else 468 sbappend(&so->so_rcv, q->ipqe_m); 469 pool_put(&ipqent_pool, q); 470 sorwakeup(so); 471 return (pkt_flags); 472 } 473 474 #if defined(INET6) && !defined(TCP6) 475 int 476 tcp6_input(mp, offp, proto) 477 struct mbuf **mp; 478 int *offp, proto; 479 { 480 struct mbuf *m = *mp; 481 482 #if defined(NFAITH) && 0 < NFAITH 483 if (m->m_pkthdr.rcvif) { 484 if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 485 /* XXX send icmp6 host/port unreach? */ 486 m_freem(m); 487 return IPPROTO_DONE; 488 } 489 } 490 #endif 491 492 /* 493 * draft-itojun-ipv6-tcp-to-anycast 494 * better place to put this in? 495 */ 496 if (m->m_flags & M_ANYCAST6) { 497 if (m->m_len >= sizeof(struct ip6_hdr)) { 498 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 499 icmp6_error(m, ICMP6_DST_UNREACH, 500 ICMP6_DST_UNREACH_ADDR, 501 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 502 } else 503 m_freem(m); 504 return IPPROTO_DONE; 505 } 506 507 tcp_input(m, *offp, proto); 508 return IPPROTO_DONE; 509 } 510 #endif 511 512 /* 513 * TCP input routine, follows pages 65-76 of the 514 * protocol specification dated September, 1981 very closely. 515 */ 516 void 517 #if __STDC__ 518 tcp_input(struct mbuf *m, ...) 519 #else 520 tcp_input(m, va_alist) 521 register struct mbuf *m; 522 #endif 523 { 524 int proto; 525 register struct tcphdr *th; 526 struct ip *ip; 527 register struct inpcb *inp; 528 #ifdef INET6 529 struct ip6_hdr *ip6; 530 register struct in6pcb *in6p; 531 #endif 532 caddr_t optp = NULL; 533 int optlen = 0; 534 int len, tlen, toff, hdroptlen = 0; 535 register struct tcpcb *tp = 0; 536 register int tiflags; 537 struct socket *so = NULL; 538 int todrop, acked, ourfinisacked, needoutput = 0; 539 short ostate = 0; 540 int iss = 0; 541 u_long tiwin; 542 struct tcp_opt_info opti; 543 int off, iphlen; 544 va_list ap; 545 int af; /* af on the wire */ 546 struct mbuf *tcp_saveti = NULL; 547 548 va_start(ap, m); 549 toff = va_arg(ap, int); 550 proto = va_arg(ap, int); 551 va_end(ap); 552 553 tcpstat.tcps_rcvtotal++; 554 555 bzero(&opti, sizeof(opti)); 556 opti.ts_present = 0; 557 opti.maxseg = 0; 558 559 /* 560 * Get IP and TCP header together in first mbuf. 561 * Note: IP leaves IP header in first mbuf. 562 */ 563 ip = mtod(m, struct ip *); 564 #ifdef INET6 565 ip6 = NULL; 566 #endif 567 switch (ip->ip_v) { 568 case 4: 569 af = AF_INET; 570 iphlen = sizeof(struct ip); 571 /* would like to get rid of this... */ 572 if (toff > sizeof (struct ip)) { 573 ip_stripoptions(m, (struct mbuf *)0); 574 toff = sizeof(struct ip); 575 } 576 if (m->m_len < toff + sizeof (struct tcphdr)) { 577 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 578 tcpstat.tcps_rcvshort++; 579 return; 580 } 581 } 582 ip = mtod(m, struct ip *); 583 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 584 585 /* 586 * Checksum extended TCP header and data. 587 */ 588 len = ip->ip_len; 589 tlen = len - toff; 590 { 591 struct ipovly *ipov; 592 ipov = (struct ipovly *)ip; 593 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 594 ipov->ih_len = htons(tlen); 595 } 596 if (in_cksum(m, len) != 0) { 597 tcpstat.tcps_rcvbadsum++; 598 goto drop; 599 } 600 break; 601 #ifdef INET6 602 case 6: 603 ip = NULL; 604 iphlen = sizeof(struct ip6_hdr); 605 af = AF_INET6; 606 if (m->m_len < toff + sizeof(struct tcphdr)) { 607 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 608 if (m == NULL) { 609 tcpstat.tcps_rcvshort++; 610 return; 611 } 612 } 613 ip6 = mtod(m, struct ip6_hdr *); 614 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 615 616 /* 617 * Checksum extended TCP header and data. 618 */ 619 len = m->m_pkthdr.len; 620 tlen = len - toff; 621 if (in6_cksum(m, IPPROTO_TCP, toff, tlen)) { 622 tcpstat.tcps_rcvbadsum++; 623 goto drop; 624 } 625 break; 626 #endif 627 default: 628 m_freem(m); 629 return; 630 } 631 632 /* 633 * Check that TCP offset makes sense, 634 * pull out TCP options and adjust length. XXX 635 */ 636 off = th->th_off << 2; 637 if (off < sizeof (struct tcphdr) || off > tlen) { 638 tcpstat.tcps_rcvbadoff++; 639 goto drop; 640 } 641 tlen -= off; 642 643 /* 644 * tcp_input() has been modified to use tlen to mean the TCP data 645 * length throughout the function. Other functions can use 646 * m->m_pkthdr.len as the basis for calculating the TCP data length. 647 * rja 648 */ 649 650 if (off > sizeof (struct tcphdr)) { 651 if (m->m_len < toff + off) { 652 if ((m = m_pullup(m, toff + off)) == 0) { 653 tcpstat.tcps_rcvshort++; 654 return; 655 } 656 switch (af) { 657 case AF_INET: 658 ip = mtod(m, struct ip *); 659 break; 660 #ifdef INET6 661 case AF_INET6: 662 ip6 = mtod(m, struct ip6_hdr *); 663 break; 664 #endif 665 } 666 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 667 } 668 optlen = off - sizeof (struct tcphdr); 669 optp = mtod(m, caddr_t) + toff + sizeof (struct tcphdr); 670 /* 671 * Do quick retrieval of timestamp options ("options 672 * prediction?"). If timestamp is the only option and it's 673 * formatted as recommended in RFC 1323 appendix A, we 674 * quickly get the values now and not bother calling 675 * tcp_dooptions(), etc. 676 */ 677 if ((optlen == TCPOLEN_TSTAMP_APPA || 678 (optlen > TCPOLEN_TSTAMP_APPA && 679 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 680 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 681 (th->th_flags & TH_SYN) == 0) { 682 opti.ts_present = 1; 683 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 684 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 685 optp = NULL; /* we've parsed the options */ 686 } 687 } 688 tiflags = th->th_flags; 689 690 /* 691 * Convert TCP protocol specific fields to host format. 692 */ 693 NTOHL(th->th_seq); 694 NTOHL(th->th_ack); 695 NTOHS(th->th_win); 696 NTOHS(th->th_urp); 697 698 /* 699 * Locate pcb for segment. 700 */ 701 findpcb: 702 inp = NULL; 703 #ifdef INET6 704 in6p = NULL; 705 #endif 706 switch (af) { 707 case AF_INET: 708 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 709 ip->ip_dst, th->th_dport); 710 if (inp == 0) { 711 ++tcpstat.tcps_pcbhashmiss; 712 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 713 } 714 #if defined(INET6) && !defined(TCP6) 715 if (inp == 0) { 716 struct in6_addr s, d; 717 718 /* mapped addr case */ 719 bzero(&s, sizeof(s)); 720 s.s6_addr16[5] = htons(0xffff); 721 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 722 bzero(&d, sizeof(d)); 723 d.s6_addr16[5] = htons(0xffff); 724 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 725 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 726 &d, th->th_dport, 0); 727 if (in6p == 0) { 728 ++tcpstat.tcps_pcbhashmiss; 729 in6p = in6_pcblookup_bind(&tcb6, &d, 730 th->th_dport, 0); 731 } 732 } 733 #endif 734 #ifndef INET6 735 if (inp == 0) 736 #else 737 if (inp == 0 && in6p == 0) 738 #endif 739 { 740 ++tcpstat.tcps_noport; 741 if (tcp_log_refused && (tiflags & TH_SYN)) { 742 #ifndef INET6 743 char src[4*sizeof "123"]; 744 char dst[4*sizeof "123"]; 745 #else 746 char src[INET6_ADDRSTRLEN]; 747 char dst[INET6_ADDRSTRLEN]; 748 #endif 749 if (ip) { 750 strcpy(src, inet_ntoa(ip->ip_dst)); 751 strcpy(dst, inet_ntoa(ip->ip_dst)); 752 } 753 #ifdef INET6 754 else if (ip6) { 755 strcpy(src, ip6_sprintf(&ip6->ip6_dst)); 756 strcpy(dst, ip6_sprintf(&ip6->ip6_src)); 757 } 758 #endif 759 else { 760 strcpy(src, "(unknown)"); 761 strcpy(dst, "(unknown)"); 762 } 763 log(LOG_INFO, 764 "Connection attempt to TCP %s:%d from %s:%d\n", 765 src, ntohs(th->th_dport), 766 dst, ntohs(th->th_sport)); 767 } 768 goto dropwithreset; 769 } 770 #ifdef IPSEC 771 if (inp && ipsec4_in_reject(m, inp)) { 772 ipsecstat.in_polvio++; 773 goto drop; 774 } 775 #ifdef INET6 776 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 777 ipsecstat.in_polvio++; 778 goto drop; 779 } 780 #endif 781 #endif /*IPSEC*/ 782 break; 783 #if defined(INET6) && !defined(TCP6) 784 case AF_INET6: 785 { 786 int faith; 787 788 #if defined(NFAITH) && NFAITH > 0 789 if (m->m_pkthdr.rcvif 790 && m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 791 faith = 1; 792 } else 793 faith = 0; 794 #else 795 faith = 0; 796 #endif 797 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 798 &ip6->ip6_dst, th->th_dport, faith); 799 if (in6p == NULL) { 800 ++tcpstat.tcps_pcbhashmiss; 801 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 802 th->th_dport, faith); 803 } 804 if (in6p == NULL) { 805 ++tcpstat.tcps_noport; 806 goto dropwithreset; 807 } 808 #ifdef IPSEC 809 if (ipsec6_in_reject(m, in6p)) { 810 ipsec6stat.in_polvio++; 811 goto drop; 812 } 813 #endif /*IPSEC*/ 814 break; 815 } 816 #endif 817 } 818 819 /* 820 * If the state is CLOSED (i.e., TCB does not exist) then 821 * all data in the incoming segment is discarded. 822 * If the TCB exists but is in CLOSED state, it is embryonic, 823 * but should either do a listen or a connect soon. 824 */ 825 tp = NULL; 826 so = NULL; 827 if (inp) { 828 tp = intotcpcb(inp); 829 so = inp->inp_socket; 830 } 831 #ifdef INET6 832 else if (in6p) { 833 tp = in6totcpcb(in6p); 834 so = in6p->in6p_socket; 835 } 836 #endif 837 if (tp == 0) { 838 goto dropwithreset; 839 } 840 if (tp->t_state == TCPS_CLOSED) 841 goto drop; 842 843 /* Unscale the window into a 32-bit value. */ 844 if ((tiflags & TH_SYN) == 0) 845 tiwin = th->th_win << tp->snd_scale; 846 else 847 tiwin = th->th_win; 848 849 #ifdef INET6 850 /* save packet options if user wanted */ 851 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 852 if (in6p->in6p_options) { 853 m_freem(in6p->in6p_options); 854 in6p->in6p_options = 0; 855 } 856 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 857 } 858 #endif 859 860 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 861 union syn_cache_sa src; 862 union syn_cache_sa dst; 863 864 bzero(&src, sizeof(src)); 865 bzero(&dst, sizeof(dst)); 866 switch (af) { 867 case AF_INET: 868 src.sin.sin_len = sizeof(struct sockaddr_in); 869 src.sin.sin_family = AF_INET; 870 src.sin.sin_addr = ip->ip_src; 871 src.sin.sin_port = th->th_sport; 872 873 dst.sin.sin_len = sizeof(struct sockaddr_in); 874 dst.sin.sin_family = AF_INET; 875 dst.sin.sin_addr = ip->ip_dst; 876 dst.sin.sin_port = th->th_dport; 877 break; 878 #ifdef INET6 879 case AF_INET6: 880 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 881 src.sin6.sin6_family = AF_INET6; 882 src.sin6.sin6_addr = ip6->ip6_src; 883 src.sin6.sin6_port = th->th_sport; 884 885 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 886 dst.sin6.sin6_family = AF_INET6; 887 dst.sin6.sin6_addr = ip6->ip6_dst; 888 dst.sin6.sin6_port = th->th_dport; 889 break; 890 #endif /* INET6 */ 891 default: 892 goto badsyn; /*sanity*/ 893 } 894 895 if (so->so_options & SO_DEBUG) { 896 ostate = tp->t_state; 897 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 898 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 899 m_freem(tcp_saveti); 900 tcp_saveti = NULL; 901 } else { 902 tcp_saveti->m_len += sizeof(struct tcphdr); 903 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 904 sizeof(struct tcphdr)); 905 } 906 if (tcp_saveti) { 907 /* 908 * need to recover version # field, which was 909 * overwritten on ip_cksum computation. 910 */ 911 struct ip *sip; 912 sip = mtod(tcp_saveti, struct ip *); 913 switch (af) { 914 case AF_INET: 915 sip->ip_v = 4; 916 break; 917 #ifdef INET6 918 case AF_INET6: 919 sip->ip_v = 6; 920 break; 921 #endif 922 } 923 } 924 } 925 if (so->so_options & SO_ACCEPTCONN) { 926 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 927 if (tiflags & TH_RST) { 928 syn_cache_reset(&src.sa, &dst.sa, th); 929 } else if ((tiflags & (TH_ACK|TH_SYN)) == 930 (TH_ACK|TH_SYN)) { 931 /* 932 * Received a SYN,ACK. This should 933 * never happen while we are in 934 * LISTEN. Send an RST. 935 */ 936 goto badsyn; 937 } else if (tiflags & TH_ACK) { 938 so = syn_cache_get(&src.sa, &dst.sa, 939 th, toff, tlen, so, m); 940 if (so == NULL) { 941 /* 942 * We don't have a SYN for 943 * this ACK; send an RST. 944 */ 945 goto badsyn; 946 } else if (so == 947 (struct socket *)(-1)) { 948 /* 949 * We were unable to create 950 * the connection. If the 951 * 3-way handshake was 952 * completed, and RST has 953 * been sent to the peer. 954 * Since the mbuf might be 955 * in use for the reply, 956 * do not free it. 957 */ 958 m = NULL; 959 } else { 960 /* 961 * We have created a 962 * full-blown connection. 963 */ 964 tp = NULL; 965 inp = NULL; 966 #ifdef INET6 967 in6p = NULL; 968 #endif 969 switch (so->so_proto->pr_domain->dom_family) { 970 case AF_INET: 971 inp = sotoinpcb(so); 972 tp = intotcpcb(inp); 973 break; 974 #ifdef INET6 975 case AF_INET6: 976 in6p = sotoin6pcb(so); 977 tp = in6totcpcb(in6p); 978 break; 979 #endif 980 } 981 if (tp == NULL) 982 goto badsyn; /*XXX*/ 983 tiwin <<= tp->snd_scale; 984 goto after_listen; 985 } 986 } else { 987 /* 988 * None of RST, SYN or ACK was set. 989 * This is an invalid packet for a 990 * TCB in LISTEN state. Send a RST. 991 */ 992 goto badsyn; 993 } 994 } else { 995 /* 996 * Received a SYN. 997 */ 998 999 /* 1000 * LISTEN socket received a SYN 1001 * from itself? This can't possibly 1002 * be valid; drop the packet. 1003 */ 1004 if (th->th_sport == th->th_dport) { 1005 int i; 1006 1007 switch (af) { 1008 case AF_INET: 1009 i = in_hosteq(ip->ip_src, ip->ip_dst); 1010 break; 1011 #ifdef INET6 1012 case AF_INET6: 1013 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1014 break; 1015 #endif 1016 default: 1017 i = 1; 1018 } 1019 if (i) { 1020 tcpstat.tcps_badsyn++; 1021 goto drop; 1022 } 1023 } 1024 1025 /* 1026 * SYN looks ok; create compressed TCP 1027 * state for it. 1028 */ 1029 if (so->so_qlen <= so->so_qlimit && 1030 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1031 so, m, optp, optlen, &opti)) 1032 m = NULL; 1033 } 1034 goto drop; 1035 } 1036 } 1037 1038 after_listen: 1039 #ifdef DIAGNOSTIC 1040 /* 1041 * Should not happen now that all embryonic connections 1042 * are handled with compressed state. 1043 */ 1044 if (tp->t_state == TCPS_LISTEN) 1045 panic("tcp_input: TCPS_LISTEN"); 1046 #endif 1047 1048 /* 1049 * Segment received on connection. 1050 * Reset idle time and keep-alive timer. 1051 */ 1052 tp->t_idle = 0; 1053 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1054 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1055 1056 /* 1057 * Process options. 1058 */ 1059 if (optp) 1060 tcp_dooptions(tp, optp, optlen, th, &opti); 1061 1062 /* 1063 * Header prediction: check for the two common cases 1064 * of a uni-directional data xfer. If the packet has 1065 * no control flags, is in-sequence, the window didn't 1066 * change and we're not retransmitting, it's a 1067 * candidate. If the length is zero and the ack moved 1068 * forward, we're the sender side of the xfer. Just 1069 * free the data acked & wake any higher level process 1070 * that was blocked waiting for space. If the length 1071 * is non-zero and the ack didn't move, we're the 1072 * receiver side. If we're getting packets in-order 1073 * (the reassembly queue is empty), add the data to 1074 * the socket buffer and note that we need a delayed ack. 1075 */ 1076 if (tp->t_state == TCPS_ESTABLISHED && 1077 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1078 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1079 th->th_seq == tp->rcv_nxt && 1080 tiwin && tiwin == tp->snd_wnd && 1081 tp->snd_nxt == tp->snd_max) { 1082 1083 /* 1084 * If last ACK falls within this segment's sequence numbers, 1085 * record the timestamp. 1086 */ 1087 if (opti.ts_present && 1088 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1089 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1090 tp->ts_recent_age = tcp_now; 1091 tp->ts_recent = opti.ts_val; 1092 } 1093 1094 if (tlen == 0) { 1095 if (SEQ_GT(th->th_ack, tp->snd_una) && 1096 SEQ_LEQ(th->th_ack, tp->snd_max) && 1097 tp->snd_cwnd >= tp->snd_wnd && 1098 tp->t_dupacks < tcprexmtthresh) { 1099 /* 1100 * this is a pure ack for outstanding data. 1101 */ 1102 ++tcpstat.tcps_predack; 1103 if (opti.ts_present && opti.ts_ecr) 1104 tcp_xmit_timer(tp, 1105 tcp_now - opti.ts_ecr + 1); 1106 else if (tp->t_rtt && 1107 SEQ_GT(th->th_ack, tp->t_rtseq)) 1108 tcp_xmit_timer(tp, tp->t_rtt); 1109 acked = th->th_ack - tp->snd_una; 1110 tcpstat.tcps_rcvackpack++; 1111 tcpstat.tcps_rcvackbyte += acked; 1112 sbdrop(&so->so_snd, acked); 1113 tp->snd_una = th->th_ack; 1114 m_freem(m); 1115 1116 /* 1117 * If all outstanding data are acked, stop 1118 * retransmit timer, otherwise restart timer 1119 * using current (possibly backed-off) value. 1120 * If process is waiting for space, 1121 * wakeup/selwakeup/signal. If data 1122 * are ready to send, let tcp_output 1123 * decide between more output or persist. 1124 */ 1125 if (tp->snd_una == tp->snd_max) 1126 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1127 else if (TCP_TIMER_ISARMED(tp, 1128 TCPT_PERSIST) == 0) 1129 TCP_TIMER_ARM(tp, TCPT_REXMT, 1130 tp->t_rxtcur); 1131 1132 sowwakeup(so); 1133 if (so->so_snd.sb_cc) 1134 (void) tcp_output(tp); 1135 if (tcp_saveti) 1136 m_freem(tcp_saveti); 1137 return; 1138 } 1139 } else if (th->th_ack == tp->snd_una && 1140 tp->segq.lh_first == NULL && 1141 tlen <= sbspace(&so->so_rcv)) { 1142 /* 1143 * this is a pure, in-sequence data packet 1144 * with nothing on the reassembly queue and 1145 * we have enough buffer space to take it. 1146 */ 1147 ++tcpstat.tcps_preddat; 1148 tp->rcv_nxt += tlen; 1149 tcpstat.tcps_rcvpack++; 1150 tcpstat.tcps_rcvbyte += tlen; 1151 /* 1152 * Drop TCP, IP headers and TCP options then add data 1153 * to socket buffer. 1154 */ 1155 m->m_data += toff + off; 1156 m->m_len -= (toff + off); 1157 sbappend(&so->so_rcv, m); 1158 sorwakeup(so); 1159 TCP_SETUP_ACK(tp, th); 1160 if (tp->t_flags & TF_ACKNOW) 1161 (void) tcp_output(tp); 1162 if (tcp_saveti) 1163 m_freem(tcp_saveti); 1164 return; 1165 } 1166 } 1167 1168 /* 1169 * Drop TCP, IP headers and TCP options. 1170 */ 1171 hdroptlen = toff + off; 1172 m->m_data += hdroptlen; 1173 m->m_len -= hdroptlen; 1174 1175 /* 1176 * Calculate amount of space in receive window, 1177 * and then do TCP input processing. 1178 * Receive window is amount of space in rcv queue, 1179 * but not less than advertised window. 1180 */ 1181 { int win; 1182 1183 win = sbspace(&so->so_rcv); 1184 if (win < 0) 1185 win = 0; 1186 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1187 } 1188 1189 switch (tp->t_state) { 1190 1191 /* 1192 * If the state is SYN_SENT: 1193 * if seg contains an ACK, but not for our SYN, drop the input. 1194 * if seg contains a RST, then drop the connection. 1195 * if seg does not contain SYN, then drop it. 1196 * Otherwise this is an acceptable SYN segment 1197 * initialize tp->rcv_nxt and tp->irs 1198 * if seg contains ack then advance tp->snd_una 1199 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1200 * arrange for segment to be acked (eventually) 1201 * continue processing rest of data/controls, beginning with URG 1202 */ 1203 case TCPS_SYN_SENT: 1204 if ((tiflags & TH_ACK) && 1205 (SEQ_LEQ(th->th_ack, tp->iss) || 1206 SEQ_GT(th->th_ack, tp->snd_max))) 1207 goto dropwithreset; 1208 if (tiflags & TH_RST) { 1209 if (tiflags & TH_ACK) 1210 tp = tcp_drop(tp, ECONNREFUSED); 1211 goto drop; 1212 } 1213 if ((tiflags & TH_SYN) == 0) 1214 goto drop; 1215 if (tiflags & TH_ACK) { 1216 tp->snd_una = th->th_ack; 1217 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1218 tp->snd_nxt = tp->snd_una; 1219 } 1220 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1221 tp->irs = th->th_seq; 1222 tcp_rcvseqinit(tp); 1223 tp->t_flags |= TF_ACKNOW; 1224 tcp_mss_from_peer(tp, opti.maxseg); 1225 1226 /* 1227 * Initialize the initial congestion window. If we 1228 * had to retransmit the SYN, we must initialize cwnd 1229 * to 1 segment (i.e. the Loss Window). 1230 */ 1231 if (tp->t_flags & TF_SYN_REXMT) 1232 tp->snd_cwnd = tp->t_peermss; 1233 else 1234 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1235 tp->t_peermss); 1236 1237 tcp_rmx_rtt(tp); 1238 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1239 tcpstat.tcps_connects++; 1240 soisconnected(so); 1241 tcp_established(tp); 1242 /* Do window scaling on this connection? */ 1243 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1244 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1245 tp->snd_scale = tp->requested_s_scale; 1246 tp->rcv_scale = tp->request_r_scale; 1247 } 1248 TCP_REASS_LOCK(tp); 1249 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1250 TCP_REASS_UNLOCK(tp); 1251 /* 1252 * if we didn't have to retransmit the SYN, 1253 * use its rtt as our initial srtt & rtt var. 1254 */ 1255 if (tp->t_rtt) 1256 tcp_xmit_timer(tp, tp->t_rtt); 1257 } else 1258 tp->t_state = TCPS_SYN_RECEIVED; 1259 1260 /* 1261 * Advance th->th_seq to correspond to first data byte. 1262 * If data, trim to stay within window, 1263 * dropping FIN if necessary. 1264 */ 1265 th->th_seq++; 1266 if (tlen > tp->rcv_wnd) { 1267 todrop = tlen - tp->rcv_wnd; 1268 m_adj(m, -todrop); 1269 tlen = tp->rcv_wnd; 1270 tiflags &= ~TH_FIN; 1271 tcpstat.tcps_rcvpackafterwin++; 1272 tcpstat.tcps_rcvbyteafterwin += todrop; 1273 } 1274 tp->snd_wl1 = th->th_seq - 1; 1275 tp->rcv_up = th->th_seq; 1276 goto step6; 1277 1278 /* 1279 * If the state is SYN_RECEIVED: 1280 * If seg contains an ACK, but not for our SYN, drop the input 1281 * and generate an RST. See page 36, rfc793 1282 */ 1283 case TCPS_SYN_RECEIVED: 1284 if ((tiflags & TH_ACK) && 1285 (SEQ_LEQ(th->th_ack, tp->iss) || 1286 SEQ_GT(th->th_ack, tp->snd_max))) 1287 goto dropwithreset; 1288 break; 1289 } 1290 1291 /* 1292 * States other than LISTEN or SYN_SENT. 1293 * First check timestamp, if present. 1294 * Then check that at least some bytes of segment are within 1295 * receive window. If segment begins before rcv_nxt, 1296 * drop leading data (and SYN); if nothing left, just ack. 1297 * 1298 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1299 * and it's less than ts_recent, drop it. 1300 */ 1301 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1302 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1303 1304 /* Check to see if ts_recent is over 24 days old. */ 1305 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1306 /* 1307 * Invalidate ts_recent. If this segment updates 1308 * ts_recent, the age will be reset later and ts_recent 1309 * will get a valid value. If it does not, setting 1310 * ts_recent to zero will at least satisfy the 1311 * requirement that zero be placed in the timestamp 1312 * echo reply when ts_recent isn't valid. The 1313 * age isn't reset until we get a valid ts_recent 1314 * because we don't want out-of-order segments to be 1315 * dropped when ts_recent is old. 1316 */ 1317 tp->ts_recent = 0; 1318 } else { 1319 tcpstat.tcps_rcvduppack++; 1320 tcpstat.tcps_rcvdupbyte += tlen; 1321 tcpstat.tcps_pawsdrop++; 1322 goto dropafterack; 1323 } 1324 } 1325 1326 todrop = tp->rcv_nxt - th->th_seq; 1327 if (todrop > 0) { 1328 if (tiflags & TH_SYN) { 1329 tiflags &= ~TH_SYN; 1330 th->th_seq++; 1331 if (th->th_urp > 1) 1332 th->th_urp--; 1333 else { 1334 tiflags &= ~TH_URG; 1335 th->th_urp = 0; 1336 } 1337 todrop--; 1338 } 1339 if (todrop > tlen || 1340 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1341 /* 1342 * Any valid FIN must be to the left of the window. 1343 * At this point the FIN must be a duplicate or 1344 * out of sequence; drop it. 1345 */ 1346 tiflags &= ~TH_FIN; 1347 /* 1348 * Send an ACK to resynchronize and drop any data. 1349 * But keep on processing for RST or ACK. 1350 */ 1351 tp->t_flags |= TF_ACKNOW; 1352 todrop = tlen; 1353 tcpstat.tcps_rcvdupbyte += todrop; 1354 tcpstat.tcps_rcvduppack++; 1355 } else { 1356 tcpstat.tcps_rcvpartduppack++; 1357 tcpstat.tcps_rcvpartdupbyte += todrop; 1358 } 1359 m_adj(m, todrop); 1360 th->th_seq += todrop; 1361 tlen -= todrop; 1362 if (th->th_urp > todrop) 1363 th->th_urp -= todrop; 1364 else { 1365 tiflags &= ~TH_URG; 1366 th->th_urp = 0; 1367 } 1368 } 1369 1370 /* 1371 * If new data are received on a connection after the 1372 * user processes are gone, then RST the other end. 1373 */ 1374 if ((so->so_state & SS_NOFDREF) && 1375 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1376 tp = tcp_close(tp); 1377 tcpstat.tcps_rcvafterclose++; 1378 goto dropwithreset; 1379 } 1380 1381 /* 1382 * If segment ends after window, drop trailing data 1383 * (and PUSH and FIN); if nothing left, just ACK. 1384 */ 1385 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1386 if (todrop > 0) { 1387 tcpstat.tcps_rcvpackafterwin++; 1388 if (todrop >= tlen) { 1389 tcpstat.tcps_rcvbyteafterwin += tlen; 1390 /* 1391 * If a new connection request is received 1392 * while in TIME_WAIT, drop the old connection 1393 * and start over if the sequence numbers 1394 * are above the previous ones. 1395 */ 1396 if (tiflags & TH_SYN && 1397 tp->t_state == TCPS_TIME_WAIT && 1398 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1399 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1400 tp->snd_nxt); 1401 tp = tcp_close(tp); 1402 /* 1403 * We have already advanced the mbuf 1404 * pointers past the IP+TCP headers and 1405 * options. Restore those pointers before 1406 * attempting to use the TCP header again. 1407 */ 1408 m->m_data -= hdroptlen; 1409 m->m_len += hdroptlen; 1410 hdroptlen = 0; 1411 goto findpcb; 1412 } 1413 /* 1414 * If window is closed can only take segments at 1415 * window edge, and have to drop data and PUSH from 1416 * incoming segments. Continue processing, but 1417 * remember to ack. Otherwise, drop segment 1418 * and ack. 1419 */ 1420 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1421 tp->t_flags |= TF_ACKNOW; 1422 tcpstat.tcps_rcvwinprobe++; 1423 } else 1424 goto dropafterack; 1425 } else 1426 tcpstat.tcps_rcvbyteafterwin += todrop; 1427 m_adj(m, -todrop); 1428 tlen -= todrop; 1429 tiflags &= ~(TH_PUSH|TH_FIN); 1430 } 1431 1432 /* 1433 * If last ACK falls within this segment's sequence numbers, 1434 * and the timestamp is newer, record it. 1435 */ 1436 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1437 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1438 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1439 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1440 tp->ts_recent_age = tcp_now; 1441 tp->ts_recent = opti.ts_val; 1442 } 1443 1444 /* 1445 * If the RST bit is set examine the state: 1446 * SYN_RECEIVED STATE: 1447 * If passive open, return to LISTEN state. 1448 * If active open, inform user that connection was refused. 1449 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1450 * Inform user that connection was reset, and close tcb. 1451 * CLOSING, LAST_ACK, TIME_WAIT STATES 1452 * Close the tcb. 1453 */ 1454 if (tiflags&TH_RST) switch (tp->t_state) { 1455 1456 case TCPS_SYN_RECEIVED: 1457 so->so_error = ECONNREFUSED; 1458 goto close; 1459 1460 case TCPS_ESTABLISHED: 1461 case TCPS_FIN_WAIT_1: 1462 case TCPS_FIN_WAIT_2: 1463 case TCPS_CLOSE_WAIT: 1464 so->so_error = ECONNRESET; 1465 close: 1466 tp->t_state = TCPS_CLOSED; 1467 tcpstat.tcps_drops++; 1468 tp = tcp_close(tp); 1469 goto drop; 1470 1471 case TCPS_CLOSING: 1472 case TCPS_LAST_ACK: 1473 case TCPS_TIME_WAIT: 1474 tp = tcp_close(tp); 1475 goto drop; 1476 } 1477 1478 /* 1479 * If a SYN is in the window, then this is an 1480 * error and we send an RST and drop the connection. 1481 */ 1482 if (tiflags & TH_SYN) { 1483 tp = tcp_drop(tp, ECONNRESET); 1484 goto dropwithreset; 1485 } 1486 1487 /* 1488 * If the ACK bit is off we drop the segment and return. 1489 */ 1490 if ((tiflags & TH_ACK) == 0) { 1491 if (tp->t_flags & TF_ACKNOW) 1492 goto dropafterack; 1493 else 1494 goto drop; 1495 } 1496 1497 /* 1498 * Ack processing. 1499 */ 1500 switch (tp->t_state) { 1501 1502 /* 1503 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1504 * ESTABLISHED state and continue processing, otherwise 1505 * send an RST. 1506 */ 1507 case TCPS_SYN_RECEIVED: 1508 if (SEQ_GT(tp->snd_una, th->th_ack) || 1509 SEQ_GT(th->th_ack, tp->snd_max)) 1510 goto dropwithreset; 1511 tcpstat.tcps_connects++; 1512 soisconnected(so); 1513 tcp_established(tp); 1514 /* Do window scaling? */ 1515 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1516 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1517 tp->snd_scale = tp->requested_s_scale; 1518 tp->rcv_scale = tp->request_r_scale; 1519 } 1520 TCP_REASS_LOCK(tp); 1521 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1522 TCP_REASS_UNLOCK(tp); 1523 tp->snd_wl1 = th->th_seq - 1; 1524 /* fall into ... */ 1525 1526 /* 1527 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1528 * ACKs. If the ack is in the range 1529 * tp->snd_una < th->th_ack <= tp->snd_max 1530 * then advance tp->snd_una to th->th_ack and drop 1531 * data from the retransmission queue. If this ACK reflects 1532 * more up to date window information we update our window information. 1533 */ 1534 case TCPS_ESTABLISHED: 1535 case TCPS_FIN_WAIT_1: 1536 case TCPS_FIN_WAIT_2: 1537 case TCPS_CLOSE_WAIT: 1538 case TCPS_CLOSING: 1539 case TCPS_LAST_ACK: 1540 case TCPS_TIME_WAIT: 1541 1542 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1543 if (tlen == 0 && tiwin == tp->snd_wnd) { 1544 tcpstat.tcps_rcvdupack++; 1545 /* 1546 * If we have outstanding data (other than 1547 * a window probe), this is a completely 1548 * duplicate ack (ie, window info didn't 1549 * change), the ack is the biggest we've 1550 * seen and we've seen exactly our rexmt 1551 * threshhold of them, assume a packet 1552 * has been dropped and retransmit it. 1553 * Kludge snd_nxt & the congestion 1554 * window so we send only this one 1555 * packet. 1556 * 1557 * We know we're losing at the current 1558 * window size so do congestion avoidance 1559 * (set ssthresh to half the current window 1560 * and pull our congestion window back to 1561 * the new ssthresh). 1562 * 1563 * Dup acks mean that packets have left the 1564 * network (they're now cached at the receiver) 1565 * so bump cwnd by the amount in the receiver 1566 * to keep a constant cwnd packets in the 1567 * network. 1568 */ 1569 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1570 th->th_ack != tp->snd_una) 1571 tp->t_dupacks = 0; 1572 else if (++tp->t_dupacks == tcprexmtthresh) { 1573 tcp_seq onxt = tp->snd_nxt; 1574 u_int win = 1575 min(tp->snd_wnd, tp->snd_cwnd) / 1576 2 / tp->t_segsz; 1577 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1578 /* 1579 * False fast retransmit after 1580 * timeout. Do not cut window. 1581 */ 1582 tp->snd_cwnd += tp->t_segsz; 1583 tp->t_dupacks = 0; 1584 (void) tcp_output(tp); 1585 goto drop; 1586 } 1587 1588 if (win < 2) 1589 win = 2; 1590 tp->snd_ssthresh = win * tp->t_segsz; 1591 tp->snd_recover = tp->snd_max; 1592 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1593 tp->t_rtt = 0; 1594 tp->snd_nxt = th->th_ack; 1595 tp->snd_cwnd = tp->t_segsz; 1596 (void) tcp_output(tp); 1597 tp->snd_cwnd = tp->snd_ssthresh + 1598 tp->t_segsz * tp->t_dupacks; 1599 if (SEQ_GT(onxt, tp->snd_nxt)) 1600 tp->snd_nxt = onxt; 1601 goto drop; 1602 } else if (tp->t_dupacks > tcprexmtthresh) { 1603 tp->snd_cwnd += tp->t_segsz; 1604 (void) tcp_output(tp); 1605 goto drop; 1606 } 1607 } else 1608 tp->t_dupacks = 0; 1609 break; 1610 } 1611 /* 1612 * If the congestion window was inflated to account 1613 * for the other side's cached packets, retract it. 1614 */ 1615 if (!tcp_do_newreno) { 1616 if (tp->t_dupacks >= tcprexmtthresh && 1617 tp->snd_cwnd > tp->snd_ssthresh) 1618 tp->snd_cwnd = tp->snd_ssthresh; 1619 tp->t_dupacks = 0; 1620 } else if (tp->t_dupacks >= tcprexmtthresh 1621 && !tcp_newreno(tp, th)) { 1622 tp->snd_cwnd = tp->snd_ssthresh; 1623 /* 1624 * Window inflation should have left us with approx. 1625 * snd_ssthresh outstanding data. But in case we 1626 * would be inclined to send a burst, better to do 1627 * it via the slow start mechanism. 1628 */ 1629 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1630 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1631 + tp->t_segsz; 1632 tp->t_dupacks = 0; 1633 } 1634 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1635 tcpstat.tcps_rcvacktoomuch++; 1636 goto dropafterack; 1637 } 1638 acked = th->th_ack - tp->snd_una; 1639 tcpstat.tcps_rcvackpack++; 1640 tcpstat.tcps_rcvackbyte += acked; 1641 1642 /* 1643 * If we have a timestamp reply, update smoothed 1644 * round trip time. If no timestamp is present but 1645 * transmit timer is running and timed sequence 1646 * number was acked, update smoothed round trip time. 1647 * Since we now have an rtt measurement, cancel the 1648 * timer backoff (cf., Phil Karn's retransmit alg.). 1649 * Recompute the initial retransmit timer. 1650 */ 1651 if (opti.ts_present && opti.ts_ecr) 1652 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1653 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1654 tcp_xmit_timer(tp,tp->t_rtt); 1655 1656 /* 1657 * If all outstanding data is acked, stop retransmit 1658 * timer and remember to restart (more output or persist). 1659 * If there is more data to be acked, restart retransmit 1660 * timer, using current (possibly backed-off) value. 1661 */ 1662 if (th->th_ack == tp->snd_max) { 1663 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1664 needoutput = 1; 1665 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1666 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1667 /* 1668 * When new data is acked, open the congestion window. 1669 * If the window gives us less than ssthresh packets 1670 * in flight, open exponentially (segsz per packet). 1671 * Otherwise open linearly: segsz per window 1672 * (segsz^2 / cwnd per packet), plus a constant 1673 * fraction of a packet (segsz/8) to help larger windows 1674 * open quickly enough. 1675 */ 1676 { 1677 register u_int cw = tp->snd_cwnd; 1678 register u_int incr = tp->t_segsz; 1679 1680 if (cw > tp->snd_ssthresh) 1681 incr = incr * incr / cw; 1682 if (!tcp_do_newreno || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1683 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1684 } 1685 if (acked > so->so_snd.sb_cc) { 1686 tp->snd_wnd -= so->so_snd.sb_cc; 1687 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1688 ourfinisacked = 1; 1689 } else { 1690 sbdrop(&so->so_snd, acked); 1691 tp->snd_wnd -= acked; 1692 ourfinisacked = 0; 1693 } 1694 sowwakeup(so); 1695 tp->snd_una = th->th_ack; 1696 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1697 tp->snd_nxt = tp->snd_una; 1698 1699 switch (tp->t_state) { 1700 1701 /* 1702 * In FIN_WAIT_1 STATE in addition to the processing 1703 * for the ESTABLISHED state if our FIN is now acknowledged 1704 * then enter FIN_WAIT_2. 1705 */ 1706 case TCPS_FIN_WAIT_1: 1707 if (ourfinisacked) { 1708 /* 1709 * If we can't receive any more 1710 * data, then closing user can proceed. 1711 * Starting the timer is contrary to the 1712 * specification, but if we don't get a FIN 1713 * we'll hang forever. 1714 */ 1715 if (so->so_state & SS_CANTRCVMORE) { 1716 soisdisconnected(so); 1717 if (tcp_maxidle > 0) 1718 TCP_TIMER_ARM(tp, TCPT_2MSL, 1719 tcp_maxidle); 1720 } 1721 tp->t_state = TCPS_FIN_WAIT_2; 1722 } 1723 break; 1724 1725 /* 1726 * In CLOSING STATE in addition to the processing for 1727 * the ESTABLISHED state if the ACK acknowledges our FIN 1728 * then enter the TIME-WAIT state, otherwise ignore 1729 * the segment. 1730 */ 1731 case TCPS_CLOSING: 1732 if (ourfinisacked) { 1733 tp->t_state = TCPS_TIME_WAIT; 1734 tcp_canceltimers(tp); 1735 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1736 soisdisconnected(so); 1737 } 1738 break; 1739 1740 /* 1741 * In LAST_ACK, we may still be waiting for data to drain 1742 * and/or to be acked, as well as for the ack of our FIN. 1743 * If our FIN is now acknowledged, delete the TCB, 1744 * enter the closed state and return. 1745 */ 1746 case TCPS_LAST_ACK: 1747 if (ourfinisacked) { 1748 tp = tcp_close(tp); 1749 goto drop; 1750 } 1751 break; 1752 1753 /* 1754 * In TIME_WAIT state the only thing that should arrive 1755 * is a retransmission of the remote FIN. Acknowledge 1756 * it and restart the finack timer. 1757 */ 1758 case TCPS_TIME_WAIT: 1759 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1760 goto dropafterack; 1761 } 1762 } 1763 1764 step6: 1765 /* 1766 * Update window information. 1767 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1768 */ 1769 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1770 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1771 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1772 /* keep track of pure window updates */ 1773 if (tlen == 0 && 1774 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1775 tcpstat.tcps_rcvwinupd++; 1776 tp->snd_wnd = tiwin; 1777 tp->snd_wl1 = th->th_seq; 1778 tp->snd_wl2 = th->th_ack; 1779 if (tp->snd_wnd > tp->max_sndwnd) 1780 tp->max_sndwnd = tp->snd_wnd; 1781 needoutput = 1; 1782 } 1783 1784 /* 1785 * Process segments with URG. 1786 */ 1787 if ((tiflags & TH_URG) && th->th_urp && 1788 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1789 /* 1790 * This is a kludge, but if we receive and accept 1791 * random urgent pointers, we'll crash in 1792 * soreceive. It's hard to imagine someone 1793 * actually wanting to send this much urgent data. 1794 */ 1795 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1796 th->th_urp = 0; /* XXX */ 1797 tiflags &= ~TH_URG; /* XXX */ 1798 goto dodata; /* XXX */ 1799 } 1800 /* 1801 * If this segment advances the known urgent pointer, 1802 * then mark the data stream. This should not happen 1803 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1804 * a FIN has been received from the remote side. 1805 * In these states we ignore the URG. 1806 * 1807 * According to RFC961 (Assigned Protocols), 1808 * the urgent pointer points to the last octet 1809 * of urgent data. We continue, however, 1810 * to consider it to indicate the first octet 1811 * of data past the urgent section as the original 1812 * spec states (in one of two places). 1813 */ 1814 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1815 tp->rcv_up = th->th_seq + th->th_urp; 1816 so->so_oobmark = so->so_rcv.sb_cc + 1817 (tp->rcv_up - tp->rcv_nxt) - 1; 1818 if (so->so_oobmark == 0) 1819 so->so_state |= SS_RCVATMARK; 1820 sohasoutofband(so); 1821 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1822 } 1823 /* 1824 * Remove out of band data so doesn't get presented to user. 1825 * This can happen independent of advancing the URG pointer, 1826 * but if two URG's are pending at once, some out-of-band 1827 * data may creep in... ick. 1828 */ 1829 if (th->th_urp <= (u_int16_t) tlen 1830 #ifdef SO_OOBINLINE 1831 && (so->so_options & SO_OOBINLINE) == 0 1832 #endif 1833 ) 1834 tcp_pulloutofband(so, th, m); 1835 } else 1836 /* 1837 * If no out of band data is expected, 1838 * pull receive urgent pointer along 1839 * with the receive window. 1840 */ 1841 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1842 tp->rcv_up = tp->rcv_nxt; 1843 dodata: /* XXX */ 1844 1845 /* 1846 * Process the segment text, merging it into the TCP sequencing queue, 1847 * and arranging for acknowledgment of receipt if necessary. 1848 * This process logically involves adjusting tp->rcv_wnd as data 1849 * is presented to the user (this happens in tcp_usrreq.c, 1850 * case PRU_RCVD). If a FIN has already been received on this 1851 * connection then we just ignore the text. 1852 */ 1853 if ((tlen || (tiflags & TH_FIN)) && 1854 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1855 /* 1856 * Insert segment ti into reassembly queue of tcp with 1857 * control block tp. Return TH_FIN if reassembly now includes 1858 * a segment with FIN. The macro form does the common case 1859 * inline (segment is the next to be received on an 1860 * established connection, and the queue is empty), 1861 * avoiding linkage into and removal from the queue and 1862 * repetition of various conversions. 1863 * Set DELACK for segments received in order, but ack 1864 * immediately when segments are out of order 1865 * (so fast retransmit can work). 1866 */ 1867 /* NOTE: this was TCP_REASS() macro, but used only once */ 1868 TCP_REASS_LOCK(tp); 1869 if (th->th_seq == tp->rcv_nxt && 1870 tp->segq.lh_first == NULL && 1871 tp->t_state == TCPS_ESTABLISHED) { 1872 TCP_SETUP_ACK(tp, th); 1873 tp->rcv_nxt += tlen; 1874 tiflags = th->th_flags & TH_FIN; 1875 tcpstat.tcps_rcvpack++;\ 1876 tcpstat.tcps_rcvbyte += tlen;\ 1877 sbappend(&(so)->so_rcv, m); 1878 sorwakeup(so); 1879 } else { 1880 tiflags = tcp_reass(tp, th, m, &tlen); 1881 tp->t_flags |= TF_ACKNOW; 1882 } 1883 TCP_REASS_UNLOCK(tp); 1884 1885 /* 1886 * Note the amount of data that peer has sent into 1887 * our window, in order to estimate the sender's 1888 * buffer size. 1889 */ 1890 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1891 } else { 1892 m_freem(m); 1893 m = NULL; 1894 tiflags &= ~TH_FIN; 1895 } 1896 1897 /* 1898 * If FIN is received ACK the FIN and let the user know 1899 * that the connection is closing. Ignore a FIN received before 1900 * the connection is fully established. 1901 */ 1902 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 1903 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1904 socantrcvmore(so); 1905 tp->t_flags |= TF_ACKNOW; 1906 tp->rcv_nxt++; 1907 } 1908 switch (tp->t_state) { 1909 1910 /* 1911 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 1912 */ 1913 case TCPS_ESTABLISHED: 1914 tp->t_state = TCPS_CLOSE_WAIT; 1915 break; 1916 1917 /* 1918 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1919 * enter the CLOSING state. 1920 */ 1921 case TCPS_FIN_WAIT_1: 1922 tp->t_state = TCPS_CLOSING; 1923 break; 1924 1925 /* 1926 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1927 * starting the time-wait timer, turning off the other 1928 * standard timers. 1929 */ 1930 case TCPS_FIN_WAIT_2: 1931 tp->t_state = TCPS_TIME_WAIT; 1932 tcp_canceltimers(tp); 1933 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1934 soisdisconnected(so); 1935 break; 1936 1937 /* 1938 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1939 */ 1940 case TCPS_TIME_WAIT: 1941 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1942 break; 1943 } 1944 } 1945 if (so->so_options & SO_DEBUG) { 1946 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 1947 } 1948 1949 /* 1950 * Return any desired output. 1951 */ 1952 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1953 (void) tcp_output(tp); 1954 if (tcp_saveti) 1955 m_freem(tcp_saveti); 1956 return; 1957 1958 badsyn: 1959 /* 1960 * Received a bad SYN. Increment counters and dropwithreset. 1961 */ 1962 tcpstat.tcps_badsyn++; 1963 tp = NULL; 1964 goto dropwithreset; 1965 1966 dropafterack: 1967 /* 1968 * Generate an ACK dropping incoming segment if it occupies 1969 * sequence space, where the ACK reflects our state. 1970 */ 1971 if (tiflags & TH_RST) 1972 goto drop; 1973 m_freem(m); 1974 tp->t_flags |= TF_ACKNOW; 1975 (void) tcp_output(tp); 1976 if (tcp_saveti) 1977 m_freem(tcp_saveti); 1978 return; 1979 1980 dropwithreset: 1981 /* 1982 * Generate a RST, dropping incoming segment. 1983 * Make ACK acceptable to originator of segment. 1984 * Don't bother to respond if destination was broadcast/multicast. 1985 */ 1986 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 1987 goto drop; 1988 if (ip && IN_MULTICAST(ip->ip_dst.s_addr)) 1989 goto drop; 1990 #ifdef INET6 1991 if (m->m_flags & M_ANYCAST6) 1992 goto drop; 1993 else if (ip6 && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1994 goto drop; 1995 #endif 1996 /* recover the header if dropped. */ 1997 m->m_data -= hdroptlen; 1998 m->m_len += hdroptlen; 1999 { 2000 /* 2001 * need to recover version # field, which was overwritten on 2002 * ip_cksum computation. 2003 */ 2004 struct ip *sip; 2005 sip = mtod(m, struct ip *); 2006 switch (af) { 2007 case AF_INET: 2008 sip->ip_v = 4; 2009 break; 2010 #ifdef INET6 2011 case AF_INET6: 2012 sip->ip_v = 6; 2013 break; 2014 #endif 2015 } 2016 } 2017 if (tiflags & TH_ACK) 2018 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2019 else { 2020 if (tiflags & TH_SYN) 2021 tlen++; 2022 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2023 TH_RST|TH_ACK); 2024 } 2025 if (tcp_saveti) 2026 m_freem(tcp_saveti); 2027 return; 2028 2029 drop: 2030 /* 2031 * Drop space held by incoming segment and return. 2032 */ 2033 if (tp) { 2034 if (tp->t_inpcb) 2035 so = tp->t_inpcb->inp_socket; 2036 #ifdef INET6 2037 else if (tp->t_in6pcb) 2038 so = tp->t_in6pcb->in6p_socket; 2039 #endif 2040 else 2041 so = NULL; 2042 if (so && (so->so_options & SO_DEBUG) != 0) 2043 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2044 } 2045 if (tcp_saveti) 2046 m_freem(tcp_saveti); 2047 m_freem(m); 2048 return; 2049 } 2050 2051 void 2052 tcp_dooptions(tp, cp, cnt, th, oi) 2053 struct tcpcb *tp; 2054 u_char *cp; 2055 int cnt; 2056 struct tcphdr *th; 2057 struct tcp_opt_info *oi; 2058 { 2059 u_int16_t mss; 2060 int opt, optlen; 2061 2062 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2063 opt = cp[0]; 2064 if (opt == TCPOPT_EOL) 2065 break; 2066 if (opt == TCPOPT_NOP) 2067 optlen = 1; 2068 else { 2069 optlen = cp[1]; 2070 if (optlen <= 0) 2071 break; 2072 } 2073 switch (opt) { 2074 2075 default: 2076 continue; 2077 2078 case TCPOPT_MAXSEG: 2079 if (optlen != TCPOLEN_MAXSEG) 2080 continue; 2081 if (!(th->th_flags & TH_SYN)) 2082 continue; 2083 bcopy(cp + 2, &mss, sizeof(mss)); 2084 oi->maxseg = ntohs(mss); 2085 break; 2086 2087 case TCPOPT_WINDOW: 2088 if (optlen != TCPOLEN_WINDOW) 2089 continue; 2090 if (!(th->th_flags & TH_SYN)) 2091 continue; 2092 tp->t_flags |= TF_RCVD_SCALE; 2093 tp->requested_s_scale = cp[2]; 2094 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2095 #if 0 /*XXX*/ 2096 char *p; 2097 2098 if (ip) 2099 p = ntohl(ip->ip_src); 2100 #ifdef INET6 2101 else if (ip6) 2102 p = ip6_sprintf(&ip6->ip6_src); 2103 #endif 2104 else 2105 p = "(unknown)"; 2106 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2107 "assuming %d\n", 2108 tp->requested_s_scale, p, 2109 TCP_MAX_WINSHIFT); 2110 #else 2111 log(LOG_ERR, "TCP: invalid wscale %d, " 2112 "assuming %d\n", 2113 tp->requested_s_scale, 2114 TCP_MAX_WINSHIFT); 2115 #endif 2116 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2117 } 2118 break; 2119 2120 case TCPOPT_TIMESTAMP: 2121 if (optlen != TCPOLEN_TIMESTAMP) 2122 continue; 2123 oi->ts_present = 1; 2124 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2125 NTOHL(oi->ts_val); 2126 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2127 NTOHL(oi->ts_ecr); 2128 2129 /* 2130 * A timestamp received in a SYN makes 2131 * it ok to send timestamp requests and replies. 2132 */ 2133 if (th->th_flags & TH_SYN) { 2134 tp->t_flags |= TF_RCVD_TSTMP; 2135 tp->ts_recent = oi->ts_val; 2136 tp->ts_recent_age = tcp_now; 2137 } 2138 break; 2139 case TCPOPT_SACK_PERMITTED: 2140 if (optlen != TCPOLEN_SACK_PERMITTED) 2141 continue; 2142 if (!(th->th_flags & TH_SYN)) 2143 continue; 2144 tp->t_flags &= ~TF_CANT_TXSACK; 2145 break; 2146 2147 case TCPOPT_SACK: 2148 if (tp->t_flags & TF_IGNR_RXSACK) 2149 continue; 2150 if (optlen % 8 != 2 || optlen < 10) 2151 continue; 2152 cp += 2; 2153 optlen -= 2; 2154 for (; optlen > 0; cp -= 8, optlen -= 8) { 2155 tcp_seq lwe, rwe; 2156 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2157 NTOHL(lwe); 2158 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2159 NTOHL(rwe); 2160 /* tcp_mark_sacked(tp, lwe, rwe); */ 2161 } 2162 break; 2163 } 2164 } 2165 } 2166 2167 /* 2168 * Pull out of band byte out of a segment so 2169 * it doesn't appear in the user's data queue. 2170 * It is still reflected in the segment length for 2171 * sequencing purposes. 2172 */ 2173 void 2174 tcp_pulloutofband(so, th, m) 2175 struct socket *so; 2176 struct tcphdr *th; 2177 register struct mbuf *m; 2178 { 2179 int cnt = th->th_urp - 1; 2180 2181 while (cnt >= 0) { 2182 if (m->m_len > cnt) { 2183 char *cp = mtod(m, caddr_t) + cnt; 2184 struct tcpcb *tp = sototcpcb(so); 2185 2186 tp->t_iobc = *cp; 2187 tp->t_oobflags |= TCPOOB_HAVEDATA; 2188 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2189 m->m_len--; 2190 return; 2191 } 2192 cnt -= m->m_len; 2193 m = m->m_next; 2194 if (m == 0) 2195 break; 2196 } 2197 panic("tcp_pulloutofband"); 2198 } 2199 2200 /* 2201 * Collect new round-trip time estimate 2202 * and update averages and current timeout. 2203 */ 2204 void 2205 tcp_xmit_timer(tp, rtt) 2206 register struct tcpcb *tp; 2207 short rtt; 2208 { 2209 register short delta; 2210 short rttmin; 2211 2212 tcpstat.tcps_rttupdated++; 2213 --rtt; 2214 if (tp->t_srtt != 0) { 2215 /* 2216 * srtt is stored as fixed point with 3 bits after the 2217 * binary point (i.e., scaled by 8). The following magic 2218 * is equivalent to the smoothing algorithm in rfc793 with 2219 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2220 * point). Adjust rtt to origin 0. 2221 */ 2222 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2223 if ((tp->t_srtt += delta) <= 0) 2224 tp->t_srtt = 1 << 2; 2225 /* 2226 * We accumulate a smoothed rtt variance (actually, a 2227 * smoothed mean difference), then set the retransmit 2228 * timer to smoothed rtt + 4 times the smoothed variance. 2229 * rttvar is stored as fixed point with 2 bits after the 2230 * binary point (scaled by 4). The following is 2231 * equivalent to rfc793 smoothing with an alpha of .75 2232 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2233 * rfc793's wired-in beta. 2234 */ 2235 if (delta < 0) 2236 delta = -delta; 2237 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2238 if ((tp->t_rttvar += delta) <= 0) 2239 tp->t_rttvar = 1 << 2; 2240 } else { 2241 /* 2242 * No rtt measurement yet - use the unsmoothed rtt. 2243 * Set the variance to half the rtt (so our first 2244 * retransmit happens at 3*rtt). 2245 */ 2246 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2247 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2248 } 2249 tp->t_rtt = 0; 2250 tp->t_rxtshift = 0; 2251 2252 /* 2253 * the retransmit should happen at rtt + 4 * rttvar. 2254 * Because of the way we do the smoothing, srtt and rttvar 2255 * will each average +1/2 tick of bias. When we compute 2256 * the retransmit timer, we want 1/2 tick of rounding and 2257 * 1 extra tick because of +-1/2 tick uncertainty in the 2258 * firing of the timer. The bias will give us exactly the 2259 * 1.5 tick we need. But, because the bias is 2260 * statistical, we have to test that we don't drop below 2261 * the minimum feasible timer (which is 2 ticks). 2262 */ 2263 if (tp->t_rttmin > rtt + 2) 2264 rttmin = tp->t_rttmin; 2265 else 2266 rttmin = rtt + 2; 2267 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2268 2269 /* 2270 * We received an ack for a packet that wasn't retransmitted; 2271 * it is probably safe to discard any error indications we've 2272 * received recently. This isn't quite right, but close enough 2273 * for now (a route might have failed after we sent a segment, 2274 * and the return path might not be symmetrical). 2275 */ 2276 tp->t_softerror = 0; 2277 } 2278 2279 /* 2280 * Checks for partial ack. If partial ack arrives, force the retransmission 2281 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2282 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2283 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2284 */ 2285 int 2286 tcp_newreno(tp, th) 2287 struct tcpcb *tp; 2288 struct tcphdr *th; 2289 { 2290 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2291 tcp_seq onxt = tp->snd_nxt; 2292 tcp_seq ouna = tp->snd_una; /* Haven't updated snd_una yet*/ 2293 u_long ocwnd = tp->snd_cwnd; 2294 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2295 tp->t_rtt = 0; 2296 tp->snd_nxt = th->th_ack; 2297 tp->snd_cwnd = tp->t_segsz; 2298 tp->snd_una = th->th_ack; 2299 (void) tcp_output(tp); 2300 tp->snd_cwnd = ocwnd; 2301 tp->snd_una = ouna; 2302 if (SEQ_GT(onxt, tp->snd_nxt)) 2303 tp->snd_nxt = onxt; 2304 /* 2305 * Partial window deflation. Relies on fact that tp->snd_una 2306 * not updated yet. 2307 */ 2308 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2309 return 1; 2310 } 2311 return 0; 2312 } 2313 2314 2315 /* 2316 * TCP compressed state engine. Currently used to hold compressed 2317 * state for SYN_RECEIVED. 2318 */ 2319 2320 u_long syn_cache_count; 2321 u_int32_t syn_hash1, syn_hash2; 2322 2323 #define SYN_HASH(sa, sp, dp) \ 2324 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2325 ((u_int32_t)(sp)))^syn_hash2))) 2326 #ifndef INET6 2327 #define SYN_HASHALL(hash, src, dst) \ 2328 do { \ 2329 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2330 ((struct sockaddr_in *)(src))->sin_port, \ 2331 ((struct sockaddr_in *)(dst))->sin_port); \ 2332 } while (0) 2333 #else 2334 #define SYN_HASH6(sa, sp, dp) \ 2335 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2336 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2337 & 0x7fffffff) 2338 2339 #define SYN_HASHALL(hash, src, dst) \ 2340 do { \ 2341 switch ((src)->sa_family) { \ 2342 case AF_INET: \ 2343 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2344 ((struct sockaddr_in *)(src))->sin_port, \ 2345 ((struct sockaddr_in *)(dst))->sin_port); \ 2346 break; \ 2347 case AF_INET6: \ 2348 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2349 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2350 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2351 break; \ 2352 default: \ 2353 hash = 0; \ 2354 } \ 2355 } while (0) 2356 #endif /* INET6 */ 2357 2358 #define SYN_CACHE_RM(sc) \ 2359 do { \ 2360 LIST_REMOVE((sc), sc_bucketq); \ 2361 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2362 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2363 syn_cache_count--; \ 2364 } while (0) 2365 2366 #define SYN_CACHE_PUT(sc) \ 2367 do { \ 2368 if ((sc)->sc_ipopts) \ 2369 (void) m_free((sc)->sc_ipopts); \ 2370 if ((sc)->sc_route4.ro_rt != NULL) \ 2371 RTFREE((sc)->sc_route4.ro_rt); \ 2372 pool_put(&syn_cache_pool, (sc)); \ 2373 } while (0) 2374 2375 struct pool syn_cache_pool; 2376 2377 /* 2378 * We don't estimate RTT with SYNs, so each packet starts with the default 2379 * RTT and each timer queue has a fixed timeout value. This allows us to 2380 * optimize the timer queues somewhat. 2381 */ 2382 #define SYN_CACHE_TIMER_ARM(sc) \ 2383 do { \ 2384 TCPT_RANGESET((sc)->sc_rxtcur, \ 2385 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2386 TCPTV_REXMTMAX); \ 2387 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2388 } while (0) 2389 2390 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2391 2392 void 2393 syn_cache_init() 2394 { 2395 int i; 2396 2397 /* Initialize the hash buckets. */ 2398 for (i = 0; i < tcp_syn_cache_size; i++) 2399 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2400 2401 /* Initialize the timer queues. */ 2402 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2403 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2404 2405 /* Initialize the syn cache pool. */ 2406 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2407 "synpl", 0, NULL, NULL, M_PCB); 2408 } 2409 2410 void 2411 syn_cache_insert(sc) 2412 struct syn_cache *sc; 2413 { 2414 struct syn_cache_head *scp; 2415 struct syn_cache *sc2; 2416 int s, i; 2417 2418 /* 2419 * If there are no entries in the hash table, reinitialize 2420 * the hash secrets. 2421 */ 2422 if (syn_cache_count == 0) { 2423 struct timeval tv; 2424 microtime(&tv); 2425 syn_hash1 = random() ^ (u_long)≻ 2426 syn_hash2 = random() ^ tv.tv_usec; 2427 } 2428 2429 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2430 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2431 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2432 2433 /* 2434 * Make sure that we don't overflow the per-bucket 2435 * limit or the total cache size limit. 2436 */ 2437 s = splsoftnet(); 2438 if (scp->sch_length >= tcp_syn_bucket_limit) { 2439 tcpstat.tcps_sc_bucketoverflow++; 2440 /* 2441 * The bucket is full. Toss the oldest element in the 2442 * bucket. This will be the entry with our bucket 2443 * index closest to the front of the timer queue with 2444 * the largest timeout value. 2445 * 2446 * Note: This timer queue traversal may be expensive, so 2447 * we hope that this doesn't happen very often. It is 2448 * much more likely that we'll overflow the entire 2449 * cache, which is much easier to handle; see below. 2450 */ 2451 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2452 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2453 sc2 != NULL; 2454 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2455 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2456 SYN_CACHE_RM(sc2); 2457 SYN_CACHE_PUT(sc2); 2458 goto insert; /* 2 level break */ 2459 } 2460 } 2461 } 2462 #ifdef DIAGNOSTIC 2463 /* 2464 * This should never happen; we should always find an 2465 * entry in our bucket. 2466 */ 2467 panic("syn_cache_insert: bucketoverflow: impossible"); 2468 #endif 2469 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2470 tcpstat.tcps_sc_overflowed++; 2471 /* 2472 * The cache is full. Toss the oldest entry in the 2473 * entire cache. This is the front entry in the 2474 * first non-empty timer queue with the largest 2475 * timeout value. 2476 */ 2477 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2478 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2479 if (sc2 == NULL) 2480 continue; 2481 SYN_CACHE_RM(sc2); 2482 SYN_CACHE_PUT(sc2); 2483 goto insert; /* symmetry with above */ 2484 } 2485 #ifdef DIAGNOSTIC 2486 /* 2487 * This should never happen; we should always find an 2488 * entry in the cache. 2489 */ 2490 panic("syn_cache_insert: cache overflow: impossible"); 2491 #endif 2492 } 2493 2494 insert: 2495 /* 2496 * Initialize the entry's timer. 2497 */ 2498 sc->sc_rxttot = 0; 2499 sc->sc_rxtshift = 0; 2500 SYN_CACHE_TIMER_ARM(sc); 2501 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2502 2503 /* Put it into the bucket. */ 2504 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2505 scp->sch_length++; 2506 syn_cache_count++; 2507 2508 tcpstat.tcps_sc_added++; 2509 splx(s); 2510 } 2511 2512 /* 2513 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2514 * If we have retransmitted an entry the maximum number of times, expire 2515 * that entry. 2516 */ 2517 void 2518 syn_cache_timer() 2519 { 2520 struct syn_cache *sc, *nsc; 2521 int i, s; 2522 2523 s = splsoftnet(); 2524 2525 /* 2526 * First, get all the entries that need to be retransmitted, or 2527 * must be expired due to exceeding the initial keepalive time. 2528 */ 2529 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2530 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2531 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2532 sc = nsc) { 2533 nsc = TAILQ_NEXT(sc, sc_timeq); 2534 2535 /* 2536 * Compute the total amount of time this entry has 2537 * been on a queue. If this entry has been on longer 2538 * than the keep alive timer would allow, expire it. 2539 */ 2540 sc->sc_rxttot += sc->sc_rxtcur; 2541 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2542 tcpstat.tcps_sc_timed_out++; 2543 SYN_CACHE_RM(sc); 2544 SYN_CACHE_PUT(sc); 2545 continue; 2546 } 2547 2548 tcpstat.tcps_sc_retransmitted++; 2549 (void) syn_cache_respond(sc, NULL); 2550 2551 /* Advance this entry onto the next timer queue. */ 2552 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2553 sc->sc_rxtshift = i + 1; 2554 SYN_CACHE_TIMER_ARM(sc); 2555 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2556 sc, sc_timeq); 2557 } 2558 } 2559 2560 /* 2561 * Now get all the entries that are expired due to too many 2562 * retransmissions. 2563 */ 2564 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2565 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2566 sc = nsc) { 2567 nsc = TAILQ_NEXT(sc, sc_timeq); 2568 tcpstat.tcps_sc_timed_out++; 2569 SYN_CACHE_RM(sc); 2570 SYN_CACHE_PUT(sc); 2571 } 2572 splx(s); 2573 } 2574 2575 /* 2576 * Find an entry in the syn cache. 2577 */ 2578 struct syn_cache * 2579 syn_cache_lookup(src, dst, headp) 2580 struct sockaddr *src; 2581 struct sockaddr *dst; 2582 struct syn_cache_head **headp; 2583 { 2584 struct syn_cache *sc; 2585 struct syn_cache_head *scp; 2586 u_int32_t hash; 2587 int s; 2588 2589 SYN_HASHALL(hash, src, dst); 2590 2591 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2592 *headp = scp; 2593 s = splsoftnet(); 2594 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2595 sc = LIST_NEXT(sc, sc_bucketq)) { 2596 if (sc->sc_hash != hash) 2597 continue; 2598 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2599 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2600 splx(s); 2601 return (sc); 2602 } 2603 } 2604 splx(s); 2605 return (NULL); 2606 } 2607 2608 /* 2609 * This function gets called when we receive an ACK for a 2610 * socket in the LISTEN state. We look up the connection 2611 * in the syn cache, and if its there, we pull it out of 2612 * the cache and turn it into a full-blown connection in 2613 * the SYN-RECEIVED state. 2614 * 2615 * The return values may not be immediately obvious, and their effects 2616 * can be subtle, so here they are: 2617 * 2618 * NULL SYN was not found in cache; caller should drop the 2619 * packet and send an RST. 2620 * 2621 * -1 We were unable to create the new connection, and are 2622 * aborting it. An ACK,RST is being sent to the peer 2623 * (unless we got screwey sequence numbners; see below), 2624 * because the 3-way handshake has been completed. Caller 2625 * should not free the mbuf, since we may be using it. If 2626 * we are not, we will free it. 2627 * 2628 * Otherwise, the return value is a pointer to the new socket 2629 * associated with the connection. 2630 */ 2631 struct socket * 2632 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2633 struct sockaddr *src; 2634 struct sockaddr *dst; 2635 struct tcphdr *th; 2636 unsigned int hlen, tlen; 2637 struct socket *so; 2638 struct mbuf *m; 2639 { 2640 struct syn_cache *sc; 2641 struct syn_cache_head *scp; 2642 register struct inpcb *inp = NULL; 2643 #ifdef INET6 2644 register struct in6pcb *in6p = NULL; 2645 #endif 2646 register struct tcpcb *tp = 0; 2647 struct mbuf *am; 2648 int s; 2649 struct socket *oso; 2650 2651 s = splsoftnet(); 2652 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2653 splx(s); 2654 return (NULL); 2655 } 2656 2657 /* 2658 * Verify the sequence and ack numbers. Try getting the correct 2659 * response again. 2660 */ 2661 if ((th->th_ack != sc->sc_iss + 1) || 2662 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2663 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2664 (void) syn_cache_respond(sc, m); 2665 splx(s); 2666 return ((struct socket *)(-1)); 2667 } 2668 2669 /* Remove this cache entry */ 2670 SYN_CACHE_RM(sc); 2671 splx(s); 2672 2673 /* 2674 * Ok, create the full blown connection, and set things up 2675 * as they would have been set up if we had created the 2676 * connection when the SYN arrived. If we can't create 2677 * the connection, abort it. 2678 */ 2679 /* 2680 * inp still has the OLD in_pcb stuff, set the 2681 * v6-related flags on the new guy, too. This is 2682 * done particularly for the case where an AF_INET6 2683 * socket is bound only to a port, and a v4 connection 2684 * comes in on that port. 2685 * we also copy the flowinfo from the original pcb 2686 * to the new one. 2687 */ 2688 { 2689 struct inpcb *parentinpcb; 2690 2691 parentinpcb = (struct inpcb *)so->so_pcb; 2692 2693 oso = so; 2694 so = sonewconn(so, SS_ISCONNECTED); 2695 if (so == NULL) 2696 goto resetandabort; 2697 2698 switch (so->so_proto->pr_domain->dom_family) { 2699 case AF_INET: 2700 inp = sotoinpcb(so); 2701 break; 2702 #ifdef INET6 2703 case AF_INET6: 2704 in6p = sotoin6pcb(so); 2705 #if 0 /*def INET6*/ 2706 inp->inp_flags |= (parentinpcb->inp_flags & 2707 (INP_IPV6 | INP_IPV6_UNDEC | INP_IPV6_MAPPED)); 2708 if ((inp->inp_flags & INP_IPV6) && 2709 !(inp->inp_flags & INP_IPV6_MAPPED)) { 2710 inp->inp_ipv6.ip6_hlim = parentinpcb->inp_ipv6.ip6_hlim; 2711 inp->inp_ipv6.ip6_vfc = parentinpcb->inp_ipv6.ip6_vfc; 2712 } 2713 #endif 2714 break; 2715 #endif 2716 } 2717 } 2718 switch (src->sa_family) { 2719 case AF_INET: 2720 if (inp) { 2721 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2722 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2723 inp->inp_options = ip_srcroute(); 2724 in_pcbstate(inp, INP_BOUND); 2725 if (inp->inp_options == NULL) { 2726 inp->inp_options = sc->sc_ipopts; 2727 sc->sc_ipopts = NULL; 2728 } 2729 } 2730 #ifdef INET6 2731 else if (in6p) { 2732 /* IPv4 packet to AF_INET6 socket */ 2733 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2734 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2735 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2736 &in6p->in6p_laddr.s6_addr32[3], 2737 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2738 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2739 in6totcpcb(in6p)->t_family = AF_INET; 2740 } 2741 #endif 2742 break; 2743 #ifdef INET6 2744 case AF_INET6: 2745 if (in6p) { 2746 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 2747 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 2748 #if 0 2749 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 2750 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 2751 #endif 2752 } 2753 break; 2754 #endif 2755 } 2756 #ifdef INET6 2757 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 2758 struct in6pcb *oin6p = sotoin6pcb(oso); 2759 /* inherit socket options from the listening socket */ 2760 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 2761 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 2762 m_freem(in6p->in6p_options); 2763 in6p->in6p_options = 0; 2764 } 2765 ip6_savecontrol(in6p, &in6p->in6p_options, 2766 mtod(m, struct ip6_hdr *), m); 2767 } 2768 #endif 2769 2770 #ifdef IPSEC 2771 { 2772 struct secpolicy *sp; 2773 if (inp) { 2774 sp = ipsec_copy_policy(sotoinpcb(oso)->inp_sp); 2775 if (sp) { 2776 key_freesp(inp->inp_sp); 2777 inp->inp_sp = sp; 2778 } else 2779 printf("tcp_input: could not copy policy\n"); 2780 } 2781 #ifdef INET6 2782 else if (in6p) { 2783 sp = ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp); 2784 if (sp) { 2785 key_freesp(in6p->in6p_sp); 2786 in6p->in6p_sp = sp; 2787 } else 2788 printf("tcp_input: could not copy policy\n"); 2789 } 2790 #endif 2791 } 2792 #endif 2793 2794 /* 2795 * Give the new socket our cached route reference. 2796 */ 2797 if (inp) 2798 inp->inp_route = sc->sc_route4; /* struct assignment */ 2799 #ifdef INET6 2800 else 2801 in6p->in6p_route = sc->sc_route6; 2802 #endif 2803 sc->sc_route4.ro_rt = NULL; 2804 2805 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 2806 if (am == NULL) 2807 goto resetandabort; 2808 am->m_len = src->sa_len; 2809 bcopy(src, mtod(am, caddr_t), src->sa_len); 2810 if (inp) { 2811 if (in_pcbconnect(inp, am)) { 2812 (void) m_free(am); 2813 goto resetandabort; 2814 } 2815 } 2816 #ifdef INET6 2817 else if (in6p) { 2818 if (src->sa_family == AF_INET) { 2819 /* IPv4 packet to AF_INET6 socket */ 2820 struct sockaddr_in6 *sin6; 2821 sin6 = mtod(am, struct sockaddr_in6 *); 2822 am->m_len = sizeof(*sin6); 2823 bzero(sin6, sizeof(*sin6)); 2824 sin6->sin6_family = AF_INET6; 2825 sin6->sin6_len = sizeof(*sin6); 2826 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 2827 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 2828 bcopy(&((struct sockaddr_in *)src)->sin_addr, 2829 &sin6->sin6_addr.s6_addr32[3], 2830 sizeof(sin6->sin6_addr.s6_addr32[3])); 2831 } 2832 if (in6_pcbconnect(in6p, am)) { 2833 (void) m_free(am); 2834 goto resetandabort; 2835 } 2836 } 2837 #endif 2838 else { 2839 (void) m_free(am); 2840 goto resetandabort; 2841 } 2842 (void) m_free(am); 2843 2844 if (inp) 2845 tp = intotcpcb(inp); 2846 #ifdef INET6 2847 else if (in6p) 2848 tp = in6totcpcb(in6p); 2849 #endif 2850 else 2851 tp = NULL; 2852 if (sc->sc_request_r_scale != 15) { 2853 tp->requested_s_scale = sc->sc_requested_s_scale; 2854 tp->request_r_scale = sc->sc_request_r_scale; 2855 tp->snd_scale = sc->sc_requested_s_scale; 2856 tp->rcv_scale = sc->sc_request_r_scale; 2857 tp->t_flags |= TF_RCVD_SCALE; 2858 } 2859 if (sc->sc_flags & SCF_TIMESTAMP) 2860 tp->t_flags |= TF_RCVD_TSTMP; 2861 2862 tp->t_template = tcp_template(tp); 2863 if (tp->t_template == 0) { 2864 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 2865 so = NULL; 2866 m_freem(m); 2867 goto abort; 2868 } 2869 2870 tp->iss = sc->sc_iss; 2871 tp->irs = sc->sc_irs; 2872 tcp_sendseqinit(tp); 2873 tcp_rcvseqinit(tp); 2874 tp->t_state = TCPS_SYN_RECEIVED; 2875 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 2876 tcpstat.tcps_accepts++; 2877 2878 /* Initialize tp->t_ourmss before we deal with the peer's! */ 2879 tp->t_ourmss = sc->sc_ourmaxseg; 2880 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 2881 2882 /* 2883 * Initialize the initial congestion window. If we 2884 * had to retransmit the SYN,ACK, we must initialize cwnd 2885 * to 1 segment (i.e. the Loss Window). 2886 */ 2887 if (sc->sc_rxtshift) 2888 tp->snd_cwnd = tp->t_peermss; 2889 else 2890 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 2891 2892 tcp_rmx_rtt(tp); 2893 tp->snd_wl1 = sc->sc_irs; 2894 tp->rcv_up = sc->sc_irs + 1; 2895 2896 /* 2897 * This is what whould have happened in tcp_ouput() when 2898 * the SYN,ACK was sent. 2899 */ 2900 tp->snd_up = tp->snd_una; 2901 tp->snd_max = tp->snd_nxt = tp->iss+1; 2902 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2903 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 2904 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 2905 tp->last_ack_sent = tp->rcv_nxt; 2906 2907 tcpstat.tcps_sc_completed++; 2908 SYN_CACHE_PUT(sc); 2909 return (so); 2910 2911 resetandabort: 2912 (void) tcp_respond(NULL, m, m, th, 2913 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 2914 abort: 2915 if (so != NULL) 2916 (void) soabort(so); 2917 SYN_CACHE_PUT(sc); 2918 tcpstat.tcps_sc_aborted++; 2919 return ((struct socket *)(-1)); 2920 } 2921 2922 /* 2923 * This function is called when we get a RST for a 2924 * non-existant connection, so that we can see if the 2925 * connection is in the syn cache. If it is, zap it. 2926 */ 2927 2928 void 2929 syn_cache_reset(src, dst, th) 2930 struct sockaddr *src; 2931 struct sockaddr *dst; 2932 struct tcphdr *th; 2933 { 2934 struct syn_cache *sc; 2935 struct syn_cache_head *scp; 2936 int s = splsoftnet(); 2937 2938 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2939 splx(s); 2940 return; 2941 } 2942 if (SEQ_LT(th->th_seq, sc->sc_irs) || 2943 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 2944 splx(s); 2945 return; 2946 } 2947 SYN_CACHE_RM(sc); 2948 splx(s); 2949 tcpstat.tcps_sc_reset++; 2950 SYN_CACHE_PUT(sc); 2951 } 2952 2953 void 2954 syn_cache_unreach(src, dst, th) 2955 struct sockaddr *src; 2956 struct sockaddr *dst; 2957 struct tcphdr *th; 2958 { 2959 struct syn_cache *sc; 2960 struct syn_cache_head *scp; 2961 int s; 2962 2963 s = splsoftnet(); 2964 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2965 splx(s); 2966 return; 2967 } 2968 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 2969 if (ntohl (th->th_seq) != sc->sc_iss) { 2970 splx(s); 2971 return; 2972 } 2973 2974 /* 2975 * If we've rertransmitted 3 times and this is our second error, 2976 * we remove the entry. Otherwise, we allow it to continue on. 2977 * This prevents us from incorrectly nuking an entry during a 2978 * spurious network outage. 2979 * 2980 * See tcp_notify(). 2981 */ 2982 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 2983 sc->sc_flags |= SCF_UNREACH; 2984 splx(s); 2985 return; 2986 } 2987 2988 SYN_CACHE_RM(sc); 2989 splx(s); 2990 tcpstat.tcps_sc_unreach++; 2991 SYN_CACHE_PUT(sc); 2992 } 2993 2994 /* 2995 * Given a LISTEN socket and an inbound SYN request, add 2996 * this to the syn cache, and send back a segment: 2997 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 2998 * to the source. 2999 * 3000 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3001 * Doing so would require that we hold onto the data and deliver it 3002 * to the application. However, if we are the target of a SYN-flood 3003 * DoS attack, an attacker could send data which would eventually 3004 * consume all available buffer space if it were ACKed. By not ACKing 3005 * the data, we avoid this DoS scenario. 3006 */ 3007 3008 int 3009 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3010 struct sockaddr *src; 3011 struct sockaddr *dst; 3012 struct tcphdr *th; 3013 unsigned int hlen; 3014 struct socket *so; 3015 struct mbuf *m; 3016 u_char *optp; 3017 int optlen; 3018 struct tcp_opt_info *oi; 3019 { 3020 struct tcpcb tb, *tp; 3021 long win; 3022 struct syn_cache *sc; 3023 struct syn_cache_head *scp; 3024 struct mbuf *ipopts; 3025 #ifdef IPSEC 3026 size_t ipsechdrsiz; 3027 #endif 3028 3029 tp = sototcpcb(so); 3030 3031 /* 3032 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3033 * in_broadcast() should never return true on a received 3034 * packet with M_BCAST not set. 3035 */ 3036 if (m->m_flags & (M_BCAST|M_MCAST)) 3037 return 0; 3038 #ifdef INET6 3039 if (m->m_flags & M_ANYCAST6) 3040 return 0; 3041 #endif 3042 3043 switch (src->sa_family) { 3044 case AF_INET: 3045 if (IN_MULTICAST(((struct sockaddr_in *)src)->sin_addr.s_addr) 3046 || IN_MULTICAST(((struct sockaddr_in *)dst)->sin_addr.s_addr)) 3047 return 0; 3048 #ifdef IPSEC 3049 ipsechdrsiz = ipsec4_hdrsiz_tcp(tp); 3050 #endif 3051 break; 3052 #ifdef INET6 3053 case AF_INET6: 3054 if (IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)src)->sin6_addr) 3055 || IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)dst)->sin6_addr)) 3056 return 0; 3057 #if defined(IPSEC) && !defined(TCP6) 3058 ipsechdrsiz = ipsec6_hdrsiz_tcp(tp); 3059 #endif 3060 break; 3061 #endif 3062 } 3063 3064 /* 3065 * Initialize some local state. 3066 */ 3067 win = sbspace(&so->so_rcv); 3068 if (win > TCP_MAXWIN) 3069 win = TCP_MAXWIN; 3070 3071 if (src->sa_family == AF_INET) { 3072 /* 3073 * Remember the IP options, if any. 3074 */ 3075 ipopts = ip_srcroute(); 3076 } else 3077 ipopts = NULL; 3078 3079 if (optp) { 3080 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3081 tcp_dooptions(&tb, optp, optlen, th, oi); 3082 } else 3083 tb.t_flags = 0; 3084 3085 /* 3086 * See if we already have an entry for this connection. 3087 * If we do, resend the SYN,ACK. We do not count this 3088 * as a retransmission (XXX though maybe we should). 3089 */ 3090 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3091 tcpstat.tcps_sc_dupesyn++; 3092 if (ipopts) { 3093 /* 3094 * If we were remembering a previous source route, 3095 * forget it and use the new one we've been given. 3096 */ 3097 if (sc->sc_ipopts) 3098 (void) m_free(sc->sc_ipopts); 3099 sc->sc_ipopts = ipopts; 3100 } 3101 sc->sc_timestamp = tb.ts_recent; 3102 if (syn_cache_respond(sc, m) == 0) { 3103 tcpstat.tcps_sndacks++; 3104 tcpstat.tcps_sndtotal++; 3105 } 3106 return (1); 3107 } 3108 3109 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3110 if (sc == NULL) { 3111 if (ipopts) 3112 (void) m_free(ipopts); 3113 return (0); 3114 } 3115 3116 /* 3117 * Fill in the cache, and put the necessary IP and TCP 3118 * options into the reply. 3119 */ 3120 bzero(sc, sizeof(struct syn_cache)); 3121 bcopy(src, &sc->sc_src, src->sa_len); 3122 bcopy(dst, &sc->sc_dst, dst->sa_len); 3123 sc->sc_flags = 0; 3124 sc->sc_ipopts = ipopts; 3125 sc->sc_irs = th->th_seq; 3126 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 3127 sc->sc_peermaxseg = oi->maxseg; 3128 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3129 m->m_pkthdr.rcvif : NULL); 3130 #ifdef IPSEC 3131 if (ipsechdrsiz < sc->sc_ourmaxseg) 3132 sc->sc_ourmaxseg -= ipsechdrsiz; 3133 #endif 3134 sc->sc_win = win; 3135 sc->sc_timestamp = tb.ts_recent; 3136 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3137 sc->sc_flags |= SCF_TIMESTAMP; 3138 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3139 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3140 sc->sc_requested_s_scale = tb.requested_s_scale; 3141 sc->sc_request_r_scale = 0; 3142 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3143 TCP_MAXWIN << sc->sc_request_r_scale < 3144 so->so_rcv.sb_hiwat) 3145 sc->sc_request_r_scale++; 3146 } else { 3147 sc->sc_requested_s_scale = 15; 3148 sc->sc_request_r_scale = 15; 3149 } 3150 sc->sc_so = so; 3151 if (syn_cache_respond(sc, m) == 0) { 3152 syn_cache_insert(sc); 3153 tcpstat.tcps_sndacks++; 3154 tcpstat.tcps_sndtotal++; 3155 } else { 3156 SYN_CACHE_PUT(sc); 3157 tcpstat.tcps_sc_dropped++; 3158 } 3159 return (1); 3160 } 3161 3162 int 3163 syn_cache_respond(sc, m) 3164 struct syn_cache *sc; 3165 struct mbuf *m; 3166 { 3167 struct route *ro = &sc->sc_route4; 3168 struct rtentry *rt; 3169 u_int8_t *optp; 3170 int optlen, error; 3171 u_int16_t tlen; 3172 struct ip *ip = NULL; 3173 #ifdef INET6 3174 struct ip6_hdr *ip6 = NULL; 3175 #endif 3176 struct tcphdr *th; 3177 u_int hlen; 3178 3179 switch (sc->sc_src.sa.sa_family) { 3180 case AF_INET: 3181 hlen = sizeof(struct ip); 3182 break; 3183 #ifdef INET6 3184 case AF_INET6: 3185 hlen = sizeof(struct ip6_hdr); 3186 break; 3187 #endif 3188 default: 3189 if (m) 3190 m_freem(m); 3191 return EAFNOSUPPORT; 3192 } 3193 3194 /* Compute the size of the TCP options. */ 3195 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3196 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3197 3198 tlen = hlen + sizeof(struct tcphdr) + optlen; 3199 3200 /* 3201 * Create the IP+TCP header from scratch. Reuse the received mbuf 3202 * if possible. 3203 */ 3204 if (m != NULL) { 3205 m_freem(m->m_next); 3206 m->m_next = NULL; 3207 MRESETDATA(m); 3208 } else { 3209 MGETHDR(m, M_DONTWAIT, MT_DATA); 3210 if (m == NULL) 3211 return (ENOBUFS); 3212 } 3213 3214 /* Fixup the mbuf. */ 3215 m->m_data += max_linkhdr; 3216 m->m_len = m->m_pkthdr.len = tlen; 3217 #ifdef IPSEC 3218 /* use IPsec policy on listening socket, on SYN ACK */ 3219 m->m_pkthdr.rcvif = (struct ifnet *)sc->sc_so; 3220 #else 3221 m->m_pkthdr.rcvif = NULL; 3222 #endif 3223 memset(mtod(m, u_char *), 0, tlen); 3224 3225 switch (sc->sc_src.sa.sa_family) { 3226 case AF_INET: 3227 ip = mtod(m, struct ip *); 3228 ip->ip_dst = sc->sc_src.sin.sin_addr; 3229 ip->ip_src = sc->sc_dst.sin.sin_addr; 3230 ip->ip_p = IPPROTO_TCP; 3231 th = (struct tcphdr *)(ip + 1); 3232 th->th_dport = sc->sc_src.sin.sin_port; 3233 th->th_sport = sc->sc_dst.sin.sin_port; 3234 break; 3235 #ifdef INET6 3236 case AF_INET6: 3237 ip6 = mtod(m, struct ip6_hdr *); 3238 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3239 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3240 ip6->ip6_nxt = IPPROTO_TCP; 3241 /* ip6_plen will be updated in ip6_output() */ 3242 th = (struct tcphdr *)(ip6 + 1); 3243 th->th_dport = sc->sc_src.sin6.sin6_port; 3244 th->th_sport = sc->sc_dst.sin6.sin6_port; 3245 break; 3246 #endif 3247 default: 3248 th = NULL; 3249 } 3250 3251 th->th_seq = htonl(sc->sc_iss); 3252 th->th_ack = htonl(sc->sc_irs + 1); 3253 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3254 th->th_flags = TH_SYN|TH_ACK; 3255 th->th_win = htons(sc->sc_win); 3256 /* th_sum already 0 */ 3257 /* th_urp already 0 */ 3258 3259 /* Tack on the TCP options. */ 3260 optp = (u_int8_t *)(th + 1); 3261 *optp++ = TCPOPT_MAXSEG; 3262 *optp++ = 4; 3263 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3264 *optp++ = sc->sc_ourmaxseg & 0xff; 3265 3266 if (sc->sc_request_r_scale != 15) { 3267 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3268 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3269 sc->sc_request_r_scale); 3270 optp += 4; 3271 } 3272 3273 if (sc->sc_flags & SCF_TIMESTAMP) { 3274 u_int32_t *lp = (u_int32_t *)(optp); 3275 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3276 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3277 *lp++ = htonl(tcp_now); 3278 *lp = htonl(sc->sc_timestamp); 3279 optp += TCPOLEN_TSTAMP_APPA; 3280 } 3281 3282 /* Compute the packet's checksum. */ 3283 switch (sc->sc_src.sa.sa_family) { 3284 case AF_INET: 3285 ip->ip_len = htons(tlen - hlen); 3286 th->th_sum = 0; 3287 th->th_sum = in_cksum(m, tlen); 3288 break; 3289 #ifdef INET6 3290 case AF_INET6: 3291 ip6->ip6_plen = htons(tlen - hlen); 3292 th->th_sum = 0; 3293 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3294 break; 3295 #endif 3296 } 3297 3298 /* 3299 * Fill in some straggling IP bits. Note the stack expects 3300 * ip_len to be in host order, for convenience. 3301 */ 3302 switch (sc->sc_src.sa.sa_family) { 3303 case AF_INET: 3304 ip->ip_len = tlen; 3305 ip->ip_ttl = ip_defttl; 3306 /* XXX tos? */ 3307 break; 3308 #ifdef INET6 3309 case AF_INET6: 3310 ip6->ip6_vfc = IPV6_VERSION; 3311 ip6->ip6_plen = htons(tlen - hlen); 3312 ip6->ip6_hlim = ip6_defhlim; 3313 /* XXX flowlabel? */ 3314 break; 3315 #endif 3316 } 3317 3318 /* 3319 * If we're doing Path MTU discovery, we need to set DF unless 3320 * the route's MTU is locked. If we don't yet know the route, 3321 * look it up now. We will copy this reference to the inpcb 3322 * when we finish creating the connection. 3323 */ 3324 if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) { 3325 if (ro->ro_rt != NULL) { 3326 RTFREE(ro->ro_rt); 3327 ro->ro_rt = NULL; 3328 } 3329 bcopy(&sc->sc_src, &ro->ro_dst, sc->sc_src.sa.sa_len); 3330 rtalloc(ro); 3331 if ((rt = ro->ro_rt) == NULL) { 3332 m_freem(m); 3333 switch (sc->sc_src.sa.sa_family) { 3334 case AF_INET: 3335 ipstat.ips_noroute++; 3336 break; 3337 #ifdef INET6 3338 case AF_INET6: 3339 ip6stat.ip6s_noroute++; 3340 break; 3341 #endif 3342 } 3343 return (EHOSTUNREACH); 3344 } 3345 } 3346 3347 switch (sc->sc_src.sa.sa_family) { 3348 case AF_INET: 3349 if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0) 3350 ip->ip_off |= IP_DF; 3351 3352 /* ...and send it off! */ 3353 error = ip_output(m, sc->sc_ipopts, ro, 0, NULL); 3354 break; 3355 #ifdef INET6 3356 case AF_INET6: 3357 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3358 0, NULL); 3359 break; 3360 #endif 3361 default: 3362 error = EAFNOSUPPORT; 3363 break; 3364 } 3365 return (error); 3366 } 3367
Indexes created Tue Sep 23 01:09:51 GMT 2025