1 /* $NetBSD: tcp_input.c,v 1.96 1999/09/23 02:21:30 itojun 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@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_src)); 751 strcpy(dst, inet_ntoa(ip->ip_dst)); 752 } 753 #ifdef INET6 754 else if (ip6) { 755 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 756 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 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 dst, ntohs(th->th_dport), 766 src, 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 /* 1114 * We want snd_recover to track snd_una to 1115 * avoid sequence wraparound problems for 1116 * very large transfers. 1117 */ 1118 tp->snd_una = tp->snd_recover = th->th_ack; 1119 m_freem(m); 1120 1121 /* 1122 * If all outstanding data are acked, stop 1123 * retransmit timer, otherwise restart timer 1124 * using current (possibly backed-off) value. 1125 * If process is waiting for space, 1126 * wakeup/selwakeup/signal. If data 1127 * are ready to send, let tcp_output 1128 * decide between more output or persist. 1129 */ 1130 if (tp->snd_una == tp->snd_max) 1131 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1132 else if (TCP_TIMER_ISARMED(tp, 1133 TCPT_PERSIST) == 0) 1134 TCP_TIMER_ARM(tp, TCPT_REXMT, 1135 tp->t_rxtcur); 1136 1137 sowwakeup(so); 1138 if (so->so_snd.sb_cc) 1139 (void) tcp_output(tp); 1140 if (tcp_saveti) 1141 m_freem(tcp_saveti); 1142 return; 1143 } 1144 } else if (th->th_ack == tp->snd_una && 1145 tp->segq.lh_first == NULL && 1146 tlen <= sbspace(&so->so_rcv)) { 1147 /* 1148 * this is a pure, in-sequence data packet 1149 * with nothing on the reassembly queue and 1150 * we have enough buffer space to take it. 1151 */ 1152 ++tcpstat.tcps_preddat; 1153 tp->rcv_nxt += tlen; 1154 tcpstat.tcps_rcvpack++; 1155 tcpstat.tcps_rcvbyte += tlen; 1156 /* 1157 * Drop TCP, IP headers and TCP options then add data 1158 * to socket buffer. 1159 */ 1160 m->m_data += toff + off; 1161 m->m_len -= (toff + off); 1162 sbappend(&so->so_rcv, m); 1163 sorwakeup(so); 1164 TCP_SETUP_ACK(tp, th); 1165 if (tp->t_flags & TF_ACKNOW) 1166 (void) tcp_output(tp); 1167 if (tcp_saveti) 1168 m_freem(tcp_saveti); 1169 return; 1170 } 1171 } 1172 1173 /* 1174 * Drop TCP, IP headers and TCP options. 1175 */ 1176 hdroptlen = toff + off; 1177 m->m_data += hdroptlen; 1178 m->m_len -= hdroptlen; 1179 1180 /* 1181 * Calculate amount of space in receive window, 1182 * and then do TCP input processing. 1183 * Receive window is amount of space in rcv queue, 1184 * but not less than advertised window. 1185 */ 1186 { int win; 1187 1188 win = sbspace(&so->so_rcv); 1189 if (win < 0) 1190 win = 0; 1191 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1192 } 1193 1194 switch (tp->t_state) { 1195 1196 /* 1197 * If the state is SYN_SENT: 1198 * if seg contains an ACK, but not for our SYN, drop the input. 1199 * if seg contains a RST, then drop the connection. 1200 * if seg does not contain SYN, then drop it. 1201 * Otherwise this is an acceptable SYN segment 1202 * initialize tp->rcv_nxt and tp->irs 1203 * if seg contains ack then advance tp->snd_una 1204 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1205 * arrange for segment to be acked (eventually) 1206 * continue processing rest of data/controls, beginning with URG 1207 */ 1208 case TCPS_SYN_SENT: 1209 if ((tiflags & TH_ACK) && 1210 (SEQ_LEQ(th->th_ack, tp->iss) || 1211 SEQ_GT(th->th_ack, tp->snd_max))) 1212 goto dropwithreset; 1213 if (tiflags & TH_RST) { 1214 if (tiflags & TH_ACK) 1215 tp = tcp_drop(tp, ECONNREFUSED); 1216 goto drop; 1217 } 1218 if ((tiflags & TH_SYN) == 0) 1219 goto drop; 1220 if (tiflags & TH_ACK) { 1221 tp->snd_una = tp->snd_recover = th->th_ack; 1222 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1223 tp->snd_nxt = tp->snd_una; 1224 } 1225 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1226 tp->irs = th->th_seq; 1227 tcp_rcvseqinit(tp); 1228 tp->t_flags |= TF_ACKNOW; 1229 tcp_mss_from_peer(tp, opti.maxseg); 1230 1231 /* 1232 * Initialize the initial congestion window. If we 1233 * had to retransmit the SYN, we must initialize cwnd 1234 * to 1 segment (i.e. the Loss Window). 1235 */ 1236 if (tp->t_flags & TF_SYN_REXMT) 1237 tp->snd_cwnd = tp->t_peermss; 1238 else 1239 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1240 tp->t_peermss); 1241 1242 tcp_rmx_rtt(tp); 1243 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1244 tcpstat.tcps_connects++; 1245 soisconnected(so); 1246 tcp_established(tp); 1247 /* Do window scaling on this connection? */ 1248 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1249 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1250 tp->snd_scale = tp->requested_s_scale; 1251 tp->rcv_scale = tp->request_r_scale; 1252 } 1253 TCP_REASS_LOCK(tp); 1254 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1255 TCP_REASS_UNLOCK(tp); 1256 /* 1257 * if we didn't have to retransmit the SYN, 1258 * use its rtt as our initial srtt & rtt var. 1259 */ 1260 if (tp->t_rtt) 1261 tcp_xmit_timer(tp, tp->t_rtt); 1262 } else 1263 tp->t_state = TCPS_SYN_RECEIVED; 1264 1265 /* 1266 * Advance th->th_seq to correspond to first data byte. 1267 * If data, trim to stay within window, 1268 * dropping FIN if necessary. 1269 */ 1270 th->th_seq++; 1271 if (tlen > tp->rcv_wnd) { 1272 todrop = tlen - tp->rcv_wnd; 1273 m_adj(m, -todrop); 1274 tlen = tp->rcv_wnd; 1275 tiflags &= ~TH_FIN; 1276 tcpstat.tcps_rcvpackafterwin++; 1277 tcpstat.tcps_rcvbyteafterwin += todrop; 1278 } 1279 tp->snd_wl1 = th->th_seq - 1; 1280 tp->rcv_up = th->th_seq; 1281 goto step6; 1282 1283 /* 1284 * If the state is SYN_RECEIVED: 1285 * If seg contains an ACK, but not for our SYN, drop the input 1286 * and generate an RST. See page 36, rfc793 1287 */ 1288 case TCPS_SYN_RECEIVED: 1289 if ((tiflags & TH_ACK) && 1290 (SEQ_LEQ(th->th_ack, tp->iss) || 1291 SEQ_GT(th->th_ack, tp->snd_max))) 1292 goto dropwithreset; 1293 break; 1294 } 1295 1296 /* 1297 * States other than LISTEN or SYN_SENT. 1298 * First check timestamp, if present. 1299 * Then check that at least some bytes of segment are within 1300 * receive window. If segment begins before rcv_nxt, 1301 * drop leading data (and SYN); if nothing left, just ack. 1302 * 1303 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1304 * and it's less than ts_recent, drop it. 1305 */ 1306 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1307 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1308 1309 /* Check to see if ts_recent is over 24 days old. */ 1310 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1311 /* 1312 * Invalidate ts_recent. If this segment updates 1313 * ts_recent, the age will be reset later and ts_recent 1314 * will get a valid value. If it does not, setting 1315 * ts_recent to zero will at least satisfy the 1316 * requirement that zero be placed in the timestamp 1317 * echo reply when ts_recent isn't valid. The 1318 * age isn't reset until we get a valid ts_recent 1319 * because we don't want out-of-order segments to be 1320 * dropped when ts_recent is old. 1321 */ 1322 tp->ts_recent = 0; 1323 } else { 1324 tcpstat.tcps_rcvduppack++; 1325 tcpstat.tcps_rcvdupbyte += tlen; 1326 tcpstat.tcps_pawsdrop++; 1327 goto dropafterack; 1328 } 1329 } 1330 1331 todrop = tp->rcv_nxt - th->th_seq; 1332 if (todrop > 0) { 1333 if (tiflags & TH_SYN) { 1334 tiflags &= ~TH_SYN; 1335 th->th_seq++; 1336 if (th->th_urp > 1) 1337 th->th_urp--; 1338 else { 1339 tiflags &= ~TH_URG; 1340 th->th_urp = 0; 1341 } 1342 todrop--; 1343 } 1344 if (todrop > tlen || 1345 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1346 /* 1347 * Any valid FIN must be to the left of the window. 1348 * At this point the FIN must be a duplicate or 1349 * out of sequence; drop it. 1350 */ 1351 tiflags &= ~TH_FIN; 1352 /* 1353 * Send an ACK to resynchronize and drop any data. 1354 * But keep on processing for RST or ACK. 1355 */ 1356 tp->t_flags |= TF_ACKNOW; 1357 todrop = tlen; 1358 tcpstat.tcps_rcvdupbyte += todrop; 1359 tcpstat.tcps_rcvduppack++; 1360 } else { 1361 tcpstat.tcps_rcvpartduppack++; 1362 tcpstat.tcps_rcvpartdupbyte += todrop; 1363 } 1364 m_adj(m, todrop); 1365 th->th_seq += todrop; 1366 tlen -= todrop; 1367 if (th->th_urp > todrop) 1368 th->th_urp -= todrop; 1369 else { 1370 tiflags &= ~TH_URG; 1371 th->th_urp = 0; 1372 } 1373 } 1374 1375 /* 1376 * If new data are received on a connection after the 1377 * user processes are gone, then RST the other end. 1378 */ 1379 if ((so->so_state & SS_NOFDREF) && 1380 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1381 tp = tcp_close(tp); 1382 tcpstat.tcps_rcvafterclose++; 1383 goto dropwithreset; 1384 } 1385 1386 /* 1387 * If segment ends after window, drop trailing data 1388 * (and PUSH and FIN); if nothing left, just ACK. 1389 */ 1390 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1391 if (todrop > 0) { 1392 tcpstat.tcps_rcvpackafterwin++; 1393 if (todrop >= tlen) { 1394 tcpstat.tcps_rcvbyteafterwin += tlen; 1395 /* 1396 * If a new connection request is received 1397 * while in TIME_WAIT, drop the old connection 1398 * and start over if the sequence numbers 1399 * are above the previous ones. 1400 */ 1401 if (tiflags & TH_SYN && 1402 tp->t_state == TCPS_TIME_WAIT && 1403 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1404 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1405 tp->snd_nxt); 1406 tp = tcp_close(tp); 1407 /* 1408 * We have already advanced the mbuf 1409 * pointers past the IP+TCP headers and 1410 * options. Restore those pointers before 1411 * attempting to use the TCP header again. 1412 */ 1413 m->m_data -= hdroptlen; 1414 m->m_len += hdroptlen; 1415 hdroptlen = 0; 1416 goto findpcb; 1417 } 1418 /* 1419 * If window is closed can only take segments at 1420 * window edge, and have to drop data and PUSH from 1421 * incoming segments. Continue processing, but 1422 * remember to ack. Otherwise, drop segment 1423 * and ack. 1424 */ 1425 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1426 tp->t_flags |= TF_ACKNOW; 1427 tcpstat.tcps_rcvwinprobe++; 1428 } else 1429 goto dropafterack; 1430 } else 1431 tcpstat.tcps_rcvbyteafterwin += todrop; 1432 m_adj(m, -todrop); 1433 tlen -= todrop; 1434 tiflags &= ~(TH_PUSH|TH_FIN); 1435 } 1436 1437 /* 1438 * If last ACK falls within this segment's sequence numbers, 1439 * and the timestamp is newer, record it. 1440 */ 1441 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1442 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1443 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1444 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1445 tp->ts_recent_age = tcp_now; 1446 tp->ts_recent = opti.ts_val; 1447 } 1448 1449 /* 1450 * If the RST bit is set examine the state: 1451 * SYN_RECEIVED STATE: 1452 * If passive open, return to LISTEN state. 1453 * If active open, inform user that connection was refused. 1454 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1455 * Inform user that connection was reset, and close tcb. 1456 * CLOSING, LAST_ACK, TIME_WAIT STATES 1457 * Close the tcb. 1458 */ 1459 if (tiflags&TH_RST) switch (tp->t_state) { 1460 1461 case TCPS_SYN_RECEIVED: 1462 so->so_error = ECONNREFUSED; 1463 goto close; 1464 1465 case TCPS_ESTABLISHED: 1466 case TCPS_FIN_WAIT_1: 1467 case TCPS_FIN_WAIT_2: 1468 case TCPS_CLOSE_WAIT: 1469 so->so_error = ECONNRESET; 1470 close: 1471 tp->t_state = TCPS_CLOSED; 1472 tcpstat.tcps_drops++; 1473 tp = tcp_close(tp); 1474 goto drop; 1475 1476 case TCPS_CLOSING: 1477 case TCPS_LAST_ACK: 1478 case TCPS_TIME_WAIT: 1479 tp = tcp_close(tp); 1480 goto drop; 1481 } 1482 1483 /* 1484 * If a SYN is in the window, then this is an 1485 * error and we send an RST and drop the connection. 1486 */ 1487 if (tiflags & TH_SYN) { 1488 tp = tcp_drop(tp, ECONNRESET); 1489 goto dropwithreset; 1490 } 1491 1492 /* 1493 * If the ACK bit is off we drop the segment and return. 1494 */ 1495 if ((tiflags & TH_ACK) == 0) { 1496 if (tp->t_flags & TF_ACKNOW) 1497 goto dropafterack; 1498 else 1499 goto drop; 1500 } 1501 1502 /* 1503 * Ack processing. 1504 */ 1505 switch (tp->t_state) { 1506 1507 /* 1508 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1509 * ESTABLISHED state and continue processing, otherwise 1510 * send an RST. 1511 */ 1512 case TCPS_SYN_RECEIVED: 1513 if (SEQ_GT(tp->snd_una, th->th_ack) || 1514 SEQ_GT(th->th_ack, tp->snd_max)) 1515 goto dropwithreset; 1516 tcpstat.tcps_connects++; 1517 soisconnected(so); 1518 tcp_established(tp); 1519 /* Do window scaling? */ 1520 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1521 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1522 tp->snd_scale = tp->requested_s_scale; 1523 tp->rcv_scale = tp->request_r_scale; 1524 } 1525 TCP_REASS_LOCK(tp); 1526 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1527 TCP_REASS_UNLOCK(tp); 1528 tp->snd_wl1 = th->th_seq - 1; 1529 /* fall into ... */ 1530 1531 /* 1532 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1533 * ACKs. If the ack is in the range 1534 * tp->snd_una < th->th_ack <= tp->snd_max 1535 * then advance tp->snd_una to th->th_ack and drop 1536 * data from the retransmission queue. If this ACK reflects 1537 * more up to date window information we update our window information. 1538 */ 1539 case TCPS_ESTABLISHED: 1540 case TCPS_FIN_WAIT_1: 1541 case TCPS_FIN_WAIT_2: 1542 case TCPS_CLOSE_WAIT: 1543 case TCPS_CLOSING: 1544 case TCPS_LAST_ACK: 1545 case TCPS_TIME_WAIT: 1546 1547 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1548 if (tlen == 0 && tiwin == tp->snd_wnd) { 1549 tcpstat.tcps_rcvdupack++; 1550 /* 1551 * If we have outstanding data (other than 1552 * a window probe), this is a completely 1553 * duplicate ack (ie, window info didn't 1554 * change), the ack is the biggest we've 1555 * seen and we've seen exactly our rexmt 1556 * threshhold of them, assume a packet 1557 * has been dropped and retransmit it. 1558 * Kludge snd_nxt & the congestion 1559 * window so we send only this one 1560 * packet. 1561 * 1562 * We know we're losing at the current 1563 * window size so do congestion avoidance 1564 * (set ssthresh to half the current window 1565 * and pull our congestion window back to 1566 * the new ssthresh). 1567 * 1568 * Dup acks mean that packets have left the 1569 * network (they're now cached at the receiver) 1570 * so bump cwnd by the amount in the receiver 1571 * to keep a constant cwnd packets in the 1572 * network. 1573 */ 1574 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1575 th->th_ack != tp->snd_una) 1576 tp->t_dupacks = 0; 1577 else if (++tp->t_dupacks == tcprexmtthresh) { 1578 tcp_seq onxt = tp->snd_nxt; 1579 u_int win = 1580 min(tp->snd_wnd, tp->snd_cwnd) / 1581 2 / tp->t_segsz; 1582 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1583 tp->snd_recover)) { 1584 /* 1585 * False fast retransmit after 1586 * timeout. Do not cut window. 1587 */ 1588 tp->snd_cwnd += tp->t_segsz; 1589 tp->t_dupacks = 0; 1590 (void) tcp_output(tp); 1591 goto drop; 1592 } 1593 1594 if (win < 2) 1595 win = 2; 1596 tp->snd_ssthresh = win * tp->t_segsz; 1597 tp->snd_recover = tp->snd_max; 1598 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1599 tp->t_rtt = 0; 1600 tp->snd_nxt = th->th_ack; 1601 tp->snd_cwnd = tp->t_segsz; 1602 (void) tcp_output(tp); 1603 tp->snd_cwnd = tp->snd_ssthresh + 1604 tp->t_segsz * tp->t_dupacks; 1605 if (SEQ_GT(onxt, tp->snd_nxt)) 1606 tp->snd_nxt = onxt; 1607 goto drop; 1608 } else if (tp->t_dupacks > tcprexmtthresh) { 1609 tp->snd_cwnd += tp->t_segsz; 1610 (void) tcp_output(tp); 1611 goto drop; 1612 } 1613 } else 1614 tp->t_dupacks = 0; 1615 break; 1616 } 1617 /* 1618 * If the congestion window was inflated to account 1619 * for the other side's cached packets, retract it. 1620 */ 1621 if (tcp_do_newreno == 0) { 1622 if (tp->t_dupacks >= tcprexmtthresh && 1623 tp->snd_cwnd > tp->snd_ssthresh) 1624 tp->snd_cwnd = tp->snd_ssthresh; 1625 tp->t_dupacks = 0; 1626 } else if (tp->t_dupacks >= tcprexmtthresh && 1627 tcp_newreno(tp, th) == 0) { 1628 tp->snd_cwnd = tp->snd_ssthresh; 1629 /* 1630 * Window inflation should have left us with approx. 1631 * snd_ssthresh outstanding data. But in case we 1632 * would be inclined to send a burst, better to do 1633 * it via the slow start mechanism. 1634 */ 1635 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1636 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1637 + tp->t_segsz; 1638 tp->t_dupacks = 0; 1639 } 1640 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1641 tcpstat.tcps_rcvacktoomuch++; 1642 goto dropafterack; 1643 } 1644 acked = th->th_ack - tp->snd_una; 1645 tcpstat.tcps_rcvackpack++; 1646 tcpstat.tcps_rcvackbyte += acked; 1647 1648 /* 1649 * If we have a timestamp reply, update smoothed 1650 * round trip time. If no timestamp is present but 1651 * transmit timer is running and timed sequence 1652 * number was acked, update smoothed round trip time. 1653 * Since we now have an rtt measurement, cancel the 1654 * timer backoff (cf., Phil Karn's retransmit alg.). 1655 * Recompute the initial retransmit timer. 1656 */ 1657 if (opti.ts_present && opti.ts_ecr) 1658 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1659 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1660 tcp_xmit_timer(tp,tp->t_rtt); 1661 1662 /* 1663 * If all outstanding data is acked, stop retransmit 1664 * timer and remember to restart (more output or persist). 1665 * If there is more data to be acked, restart retransmit 1666 * timer, using current (possibly backed-off) value. 1667 */ 1668 if (th->th_ack == tp->snd_max) { 1669 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1670 needoutput = 1; 1671 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1672 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1673 /* 1674 * When new data is acked, open the congestion window. 1675 * If the window gives us less than ssthresh packets 1676 * in flight, open exponentially (segsz per packet). 1677 * Otherwise open linearly: segsz per window 1678 * (segsz^2 / cwnd per packet), plus a constant 1679 * fraction of a packet (segsz/8) to help larger windows 1680 * open quickly enough. 1681 */ 1682 { 1683 register u_int cw = tp->snd_cwnd; 1684 register u_int incr = tp->t_segsz; 1685 1686 if (cw > tp->snd_ssthresh) 1687 incr = incr * incr / cw; 1688 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1689 tp->snd_cwnd = min(cw + incr, 1690 TCP_MAXWIN << tp->snd_scale); 1691 } 1692 if (acked > so->so_snd.sb_cc) { 1693 tp->snd_wnd -= so->so_snd.sb_cc; 1694 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1695 ourfinisacked = 1; 1696 } else { 1697 sbdrop(&so->so_snd, acked); 1698 tp->snd_wnd -= acked; 1699 ourfinisacked = 0; 1700 } 1701 sowwakeup(so); 1702 /* 1703 * We want snd_recover to track snd_una to 1704 * avoid sequence wraparound problems for 1705 * very large transfers. 1706 */ 1707 tp->snd_una = tp->snd_recover = th->th_ack; 1708 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1709 tp->snd_nxt = tp->snd_una; 1710 1711 switch (tp->t_state) { 1712 1713 /* 1714 * In FIN_WAIT_1 STATE in addition to the processing 1715 * for the ESTABLISHED state if our FIN is now acknowledged 1716 * then enter FIN_WAIT_2. 1717 */ 1718 case TCPS_FIN_WAIT_1: 1719 if (ourfinisacked) { 1720 /* 1721 * If we can't receive any more 1722 * data, then closing user can proceed. 1723 * Starting the timer is contrary to the 1724 * specification, but if we don't get a FIN 1725 * we'll hang forever. 1726 */ 1727 if (so->so_state & SS_CANTRCVMORE) { 1728 soisdisconnected(so); 1729 if (tcp_maxidle > 0) 1730 TCP_TIMER_ARM(tp, TCPT_2MSL, 1731 tcp_maxidle); 1732 } 1733 tp->t_state = TCPS_FIN_WAIT_2; 1734 } 1735 break; 1736 1737 /* 1738 * In CLOSING STATE in addition to the processing for 1739 * the ESTABLISHED state if the ACK acknowledges our FIN 1740 * then enter the TIME-WAIT state, otherwise ignore 1741 * the segment. 1742 */ 1743 case TCPS_CLOSING: 1744 if (ourfinisacked) { 1745 tp->t_state = TCPS_TIME_WAIT; 1746 tcp_canceltimers(tp); 1747 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1748 soisdisconnected(so); 1749 } 1750 break; 1751 1752 /* 1753 * In LAST_ACK, we may still be waiting for data to drain 1754 * and/or to be acked, as well as for the ack of our FIN. 1755 * If our FIN is now acknowledged, delete the TCB, 1756 * enter the closed state and return. 1757 */ 1758 case TCPS_LAST_ACK: 1759 if (ourfinisacked) { 1760 tp = tcp_close(tp); 1761 goto drop; 1762 } 1763 break; 1764 1765 /* 1766 * In TIME_WAIT state the only thing that should arrive 1767 * is a retransmission of the remote FIN. Acknowledge 1768 * it and restart the finack timer. 1769 */ 1770 case TCPS_TIME_WAIT: 1771 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1772 goto dropafterack; 1773 } 1774 } 1775 1776 step6: 1777 /* 1778 * Update window information. 1779 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1780 */ 1781 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1782 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1783 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1784 /* keep track of pure window updates */ 1785 if (tlen == 0 && 1786 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1787 tcpstat.tcps_rcvwinupd++; 1788 tp->snd_wnd = tiwin; 1789 tp->snd_wl1 = th->th_seq; 1790 tp->snd_wl2 = th->th_ack; 1791 if (tp->snd_wnd > tp->max_sndwnd) 1792 tp->max_sndwnd = tp->snd_wnd; 1793 needoutput = 1; 1794 } 1795 1796 /* 1797 * Process segments with URG. 1798 */ 1799 if ((tiflags & TH_URG) && th->th_urp && 1800 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1801 /* 1802 * This is a kludge, but if we receive and accept 1803 * random urgent pointers, we'll crash in 1804 * soreceive. It's hard to imagine someone 1805 * actually wanting to send this much urgent data. 1806 */ 1807 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1808 th->th_urp = 0; /* XXX */ 1809 tiflags &= ~TH_URG; /* XXX */ 1810 goto dodata; /* XXX */ 1811 } 1812 /* 1813 * If this segment advances the known urgent pointer, 1814 * then mark the data stream. This should not happen 1815 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1816 * a FIN has been received from the remote side. 1817 * In these states we ignore the URG. 1818 * 1819 * According to RFC961 (Assigned Protocols), 1820 * the urgent pointer points to the last octet 1821 * of urgent data. We continue, however, 1822 * to consider it to indicate the first octet 1823 * of data past the urgent section as the original 1824 * spec states (in one of two places). 1825 */ 1826 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1827 tp->rcv_up = th->th_seq + th->th_urp; 1828 so->so_oobmark = so->so_rcv.sb_cc + 1829 (tp->rcv_up - tp->rcv_nxt) - 1; 1830 if (so->so_oobmark == 0) 1831 so->so_state |= SS_RCVATMARK; 1832 sohasoutofband(so); 1833 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1834 } 1835 /* 1836 * Remove out of band data so doesn't get presented to user. 1837 * This can happen independent of advancing the URG pointer, 1838 * but if two URG's are pending at once, some out-of-band 1839 * data may creep in... ick. 1840 */ 1841 if (th->th_urp <= (u_int16_t) tlen 1842 #ifdef SO_OOBINLINE 1843 && (so->so_options & SO_OOBINLINE) == 0 1844 #endif 1845 ) 1846 tcp_pulloutofband(so, th, m); 1847 } else 1848 /* 1849 * If no out of band data is expected, 1850 * pull receive urgent pointer along 1851 * with the receive window. 1852 */ 1853 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1854 tp->rcv_up = tp->rcv_nxt; 1855 dodata: /* XXX */ 1856 1857 /* 1858 * Process the segment text, merging it into the TCP sequencing queue, 1859 * and arranging for acknowledgement of receipt if necessary. 1860 * This process logically involves adjusting tp->rcv_wnd as data 1861 * is presented to the user (this happens in tcp_usrreq.c, 1862 * case PRU_RCVD). If a FIN has already been received on this 1863 * connection then we just ignore the text. 1864 */ 1865 if ((tlen || (tiflags & TH_FIN)) && 1866 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1867 /* 1868 * Insert segment ti into reassembly queue of tcp with 1869 * control block tp. Return TH_FIN if reassembly now includes 1870 * a segment with FIN. The macro form does the common case 1871 * inline (segment is the next to be received on an 1872 * established connection, and the queue is empty), 1873 * avoiding linkage into and removal from the queue and 1874 * repetition of various conversions. 1875 * Set DELACK for segments received in order, but ack 1876 * immediately when segments are out of order 1877 * (so fast retransmit can work). 1878 */ 1879 /* NOTE: this was TCP_REASS() macro, but used only once */ 1880 TCP_REASS_LOCK(tp); 1881 if (th->th_seq == tp->rcv_nxt && 1882 tp->segq.lh_first == NULL && 1883 tp->t_state == TCPS_ESTABLISHED) { 1884 TCP_SETUP_ACK(tp, th); 1885 tp->rcv_nxt += tlen; 1886 tiflags = th->th_flags & TH_FIN; 1887 tcpstat.tcps_rcvpack++;\ 1888 tcpstat.tcps_rcvbyte += tlen;\ 1889 sbappend(&(so)->so_rcv, m); 1890 sorwakeup(so); 1891 } else { 1892 tiflags = tcp_reass(tp, th, m, &tlen); 1893 tp->t_flags |= TF_ACKNOW; 1894 } 1895 TCP_REASS_UNLOCK(tp); 1896 1897 /* 1898 * Note the amount of data that peer has sent into 1899 * our window, in order to estimate the sender's 1900 * buffer size. 1901 */ 1902 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1903 } else { 1904 m_freem(m); 1905 m = NULL; 1906 tiflags &= ~TH_FIN; 1907 } 1908 1909 /* 1910 * If FIN is received ACK the FIN and let the user know 1911 * that the connection is closing. Ignore a FIN received before 1912 * the connection is fully established. 1913 */ 1914 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 1915 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1916 socantrcvmore(so); 1917 tp->t_flags |= TF_ACKNOW; 1918 tp->rcv_nxt++; 1919 } 1920 switch (tp->t_state) { 1921 1922 /* 1923 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 1924 */ 1925 case TCPS_ESTABLISHED: 1926 tp->t_state = TCPS_CLOSE_WAIT; 1927 break; 1928 1929 /* 1930 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1931 * enter the CLOSING state. 1932 */ 1933 case TCPS_FIN_WAIT_1: 1934 tp->t_state = TCPS_CLOSING; 1935 break; 1936 1937 /* 1938 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1939 * starting the time-wait timer, turning off the other 1940 * standard timers. 1941 */ 1942 case TCPS_FIN_WAIT_2: 1943 tp->t_state = TCPS_TIME_WAIT; 1944 tcp_canceltimers(tp); 1945 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1946 soisdisconnected(so); 1947 break; 1948 1949 /* 1950 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1951 */ 1952 case TCPS_TIME_WAIT: 1953 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1954 break; 1955 } 1956 } 1957 if (so->so_options & SO_DEBUG) { 1958 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 1959 } 1960 1961 /* 1962 * Return any desired output. 1963 */ 1964 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1965 (void) tcp_output(tp); 1966 if (tcp_saveti) 1967 m_freem(tcp_saveti); 1968 return; 1969 1970 badsyn: 1971 /* 1972 * Received a bad SYN. Increment counters and dropwithreset. 1973 */ 1974 tcpstat.tcps_badsyn++; 1975 tp = NULL; 1976 goto dropwithreset; 1977 1978 dropafterack: 1979 /* 1980 * Generate an ACK dropping incoming segment if it occupies 1981 * sequence space, where the ACK reflects our state. 1982 */ 1983 if (tiflags & TH_RST) 1984 goto drop; 1985 m_freem(m); 1986 tp->t_flags |= TF_ACKNOW; 1987 (void) tcp_output(tp); 1988 if (tcp_saveti) 1989 m_freem(tcp_saveti); 1990 return; 1991 1992 dropwithreset: 1993 /* 1994 * Generate a RST, dropping incoming segment. 1995 * Make ACK acceptable to originator of segment. 1996 * Don't bother to respond if destination was broadcast/multicast. 1997 */ 1998 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 1999 goto drop; 2000 if (ip && IN_MULTICAST(ip->ip_dst.s_addr)) 2001 goto drop; 2002 #ifdef INET6 2003 if (m->m_flags & M_ANYCAST6) 2004 goto drop; 2005 else if (ip6 && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2006 goto drop; 2007 #endif 2008 /* recover the header if dropped. */ 2009 m->m_data -= hdroptlen; 2010 m->m_len += hdroptlen; 2011 { 2012 /* 2013 * need to recover version # field, which was overwritten on 2014 * ip_cksum computation. 2015 */ 2016 struct ip *sip; 2017 sip = mtod(m, struct ip *); 2018 switch (af) { 2019 case AF_INET: 2020 sip->ip_v = 4; 2021 break; 2022 #ifdef INET6 2023 case AF_INET6: 2024 sip->ip_v = 6; 2025 break; 2026 #endif 2027 } 2028 } 2029 if (tiflags & TH_ACK) 2030 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2031 else { 2032 if (tiflags & TH_SYN) 2033 tlen++; 2034 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2035 TH_RST|TH_ACK); 2036 } 2037 if (tcp_saveti) 2038 m_freem(tcp_saveti); 2039 return; 2040 2041 drop: 2042 /* 2043 * Drop space held by incoming segment and return. 2044 */ 2045 if (tp) { 2046 if (tp->t_inpcb) 2047 so = tp->t_inpcb->inp_socket; 2048 #ifdef INET6 2049 else if (tp->t_in6pcb) 2050 so = tp->t_in6pcb->in6p_socket; 2051 #endif 2052 else 2053 so = NULL; 2054 if (so && (so->so_options & SO_DEBUG) != 0) 2055 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2056 } 2057 if (tcp_saveti) 2058 m_freem(tcp_saveti); 2059 m_freem(m); 2060 return; 2061 } 2062 2063 void 2064 tcp_dooptions(tp, cp, cnt, th, oi) 2065 struct tcpcb *tp; 2066 u_char *cp; 2067 int cnt; 2068 struct tcphdr *th; 2069 struct tcp_opt_info *oi; 2070 { 2071 u_int16_t mss; 2072 int opt, optlen; 2073 2074 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2075 opt = cp[0]; 2076 if (opt == TCPOPT_EOL) 2077 break; 2078 if (opt == TCPOPT_NOP) 2079 optlen = 1; 2080 else { 2081 optlen = cp[1]; 2082 if (optlen <= 0) 2083 break; 2084 } 2085 switch (opt) { 2086 2087 default: 2088 continue; 2089 2090 case TCPOPT_MAXSEG: 2091 if (optlen != TCPOLEN_MAXSEG) 2092 continue; 2093 if (!(th->th_flags & TH_SYN)) 2094 continue; 2095 bcopy(cp + 2, &mss, sizeof(mss)); 2096 oi->maxseg = ntohs(mss); 2097 break; 2098 2099 case TCPOPT_WINDOW: 2100 if (optlen != TCPOLEN_WINDOW) 2101 continue; 2102 if (!(th->th_flags & TH_SYN)) 2103 continue; 2104 tp->t_flags |= TF_RCVD_SCALE; 2105 tp->requested_s_scale = cp[2]; 2106 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2107 #if 0 /*XXX*/ 2108 char *p; 2109 2110 if (ip) 2111 p = ntohl(ip->ip_src); 2112 #ifdef INET6 2113 else if (ip6) 2114 p = ip6_sprintf(&ip6->ip6_src); 2115 #endif 2116 else 2117 p = "(unknown)"; 2118 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2119 "assuming %d\n", 2120 tp->requested_s_scale, p, 2121 TCP_MAX_WINSHIFT); 2122 #else 2123 log(LOG_ERR, "TCP: invalid wscale %d, " 2124 "assuming %d\n", 2125 tp->requested_s_scale, 2126 TCP_MAX_WINSHIFT); 2127 #endif 2128 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2129 } 2130 break; 2131 2132 case TCPOPT_TIMESTAMP: 2133 if (optlen != TCPOLEN_TIMESTAMP) 2134 continue; 2135 oi->ts_present = 1; 2136 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2137 NTOHL(oi->ts_val); 2138 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2139 NTOHL(oi->ts_ecr); 2140 2141 /* 2142 * A timestamp received in a SYN makes 2143 * it ok to send timestamp requests and replies. 2144 */ 2145 if (th->th_flags & TH_SYN) { 2146 tp->t_flags |= TF_RCVD_TSTMP; 2147 tp->ts_recent = oi->ts_val; 2148 tp->ts_recent_age = tcp_now; 2149 } 2150 break; 2151 case TCPOPT_SACK_PERMITTED: 2152 if (optlen != TCPOLEN_SACK_PERMITTED) 2153 continue; 2154 if (!(th->th_flags & TH_SYN)) 2155 continue; 2156 tp->t_flags &= ~TF_CANT_TXSACK; 2157 break; 2158 2159 case TCPOPT_SACK: 2160 if (tp->t_flags & TF_IGNR_RXSACK) 2161 continue; 2162 if (optlen % 8 != 2 || optlen < 10) 2163 continue; 2164 cp += 2; 2165 optlen -= 2; 2166 for (; optlen > 0; cp -= 8, optlen -= 8) { 2167 tcp_seq lwe, rwe; 2168 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2169 NTOHL(lwe); 2170 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2171 NTOHL(rwe); 2172 /* tcp_mark_sacked(tp, lwe, rwe); */ 2173 } 2174 break; 2175 } 2176 } 2177 } 2178 2179 /* 2180 * Pull out of band byte out of a segment so 2181 * it doesn't appear in the user's data queue. 2182 * It is still reflected in the segment length for 2183 * sequencing purposes. 2184 */ 2185 void 2186 tcp_pulloutofband(so, th, m) 2187 struct socket *so; 2188 struct tcphdr *th; 2189 register struct mbuf *m; 2190 { 2191 int cnt = th->th_urp - 1; 2192 2193 while (cnt >= 0) { 2194 if (m->m_len > cnt) { 2195 char *cp = mtod(m, caddr_t) + cnt; 2196 struct tcpcb *tp = sototcpcb(so); 2197 2198 tp->t_iobc = *cp; 2199 tp->t_oobflags |= TCPOOB_HAVEDATA; 2200 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2201 m->m_len--; 2202 return; 2203 } 2204 cnt -= m->m_len; 2205 m = m->m_next; 2206 if (m == 0) 2207 break; 2208 } 2209 panic("tcp_pulloutofband"); 2210 } 2211 2212 /* 2213 * Collect new round-trip time estimate 2214 * and update averages and current timeout. 2215 */ 2216 void 2217 tcp_xmit_timer(tp, rtt) 2218 register struct tcpcb *tp; 2219 short rtt; 2220 { 2221 register short delta; 2222 short rttmin; 2223 2224 tcpstat.tcps_rttupdated++; 2225 --rtt; 2226 if (tp->t_srtt != 0) { 2227 /* 2228 * srtt is stored as fixed point with 3 bits after the 2229 * binary point (i.e., scaled by 8). The following magic 2230 * is equivalent to the smoothing algorithm in rfc793 with 2231 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2232 * point). Adjust rtt to origin 0. 2233 */ 2234 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2235 if ((tp->t_srtt += delta) <= 0) 2236 tp->t_srtt = 1 << 2; 2237 /* 2238 * We accumulate a smoothed rtt variance (actually, a 2239 * smoothed mean difference), then set the retransmit 2240 * timer to smoothed rtt + 4 times the smoothed variance. 2241 * rttvar is stored as fixed point with 2 bits after the 2242 * binary point (scaled by 4). The following is 2243 * equivalent to rfc793 smoothing with an alpha of .75 2244 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2245 * rfc793's wired-in beta. 2246 */ 2247 if (delta < 0) 2248 delta = -delta; 2249 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2250 if ((tp->t_rttvar += delta) <= 0) 2251 tp->t_rttvar = 1 << 2; 2252 } else { 2253 /* 2254 * No rtt measurement yet - use the unsmoothed rtt. 2255 * Set the variance to half the rtt (so our first 2256 * retransmit happens at 3*rtt). 2257 */ 2258 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2259 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2260 } 2261 tp->t_rtt = 0; 2262 tp->t_rxtshift = 0; 2263 2264 /* 2265 * the retransmit should happen at rtt + 4 * rttvar. 2266 * Because of the way we do the smoothing, srtt and rttvar 2267 * will each average +1/2 tick of bias. When we compute 2268 * the retransmit timer, we want 1/2 tick of rounding and 2269 * 1 extra tick because of +-1/2 tick uncertainty in the 2270 * firing of the timer. The bias will give us exactly the 2271 * 1.5 tick we need. But, because the bias is 2272 * statistical, we have to test that we don't drop below 2273 * the minimum feasible timer (which is 2 ticks). 2274 */ 2275 if (tp->t_rttmin > rtt + 2) 2276 rttmin = tp->t_rttmin; 2277 else 2278 rttmin = rtt + 2; 2279 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2280 2281 /* 2282 * We received an ack for a packet that wasn't retransmitted; 2283 * it is probably safe to discard any error indications we've 2284 * received recently. This isn't quite right, but close enough 2285 * for now (a route might have failed after we sent a segment, 2286 * and the return path might not be symmetrical). 2287 */ 2288 tp->t_softerror = 0; 2289 } 2290 2291 /* 2292 * Checks for partial ack. If partial ack arrives, force the retransmission 2293 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2294 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2295 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2296 */ 2297 int 2298 tcp_newreno(tp, th) 2299 struct tcpcb *tp; 2300 struct tcphdr *th; 2301 { 2302 tcp_seq onxt = tp->snd_nxt; 2303 u_long ocwnd = tp->snd_cwnd; 2304 2305 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2306 /* 2307 * snd_una has not yet been updated and the socket's send 2308 * buffer has not yet drained off the ACK'd data, so we 2309 * have to leave snd_una as it was to get the correct data 2310 * offset in tcp_output(). 2311 */ 2312 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2313 tp->t_rtt = 0; 2314 tp->snd_nxt = th->th_ack; 2315 /* 2316 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2317 * is not yet updated when we're called. 2318 */ 2319 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2320 (void) tcp_output(tp); 2321 tp->snd_cwnd = ocwnd; 2322 if (SEQ_GT(onxt, tp->snd_nxt)) 2323 tp->snd_nxt = onxt; 2324 /* 2325 * Partial window deflation. Relies on fact that tp->snd_una 2326 * not updated yet. 2327 */ 2328 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2329 return 1; 2330 } 2331 return 0; 2332 } 2333 2334 2335 /* 2336 * TCP compressed state engine. Currently used to hold compressed 2337 * state for SYN_RECEIVED. 2338 */ 2339 2340 u_long syn_cache_count; 2341 u_int32_t syn_hash1, syn_hash2; 2342 2343 #define SYN_HASH(sa, sp, dp) \ 2344 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2345 ((u_int32_t)(sp)))^syn_hash2))) 2346 #ifndef INET6 2347 #define SYN_HASHALL(hash, src, dst) \ 2348 do { \ 2349 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2350 ((struct sockaddr_in *)(src))->sin_port, \ 2351 ((struct sockaddr_in *)(dst))->sin_port); \ 2352 } while (0) 2353 #else 2354 #define SYN_HASH6(sa, sp, dp) \ 2355 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2356 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2357 & 0x7fffffff) 2358 2359 #define SYN_HASHALL(hash, src, dst) \ 2360 do { \ 2361 switch ((src)->sa_family) { \ 2362 case AF_INET: \ 2363 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2364 ((struct sockaddr_in *)(src))->sin_port, \ 2365 ((struct sockaddr_in *)(dst))->sin_port); \ 2366 break; \ 2367 case AF_INET6: \ 2368 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2369 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2370 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2371 break; \ 2372 default: \ 2373 hash = 0; \ 2374 } \ 2375 } while (0) 2376 #endif /* INET6 */ 2377 2378 #define SYN_CACHE_RM(sc) \ 2379 do { \ 2380 LIST_REMOVE((sc), sc_bucketq); \ 2381 (sc)->sc_tp = NULL; \ 2382 LIST_REMOVE((sc), sc_tpq); \ 2383 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2384 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2385 syn_cache_count--; \ 2386 } while (0) 2387 2388 #define SYN_CACHE_PUT(sc) \ 2389 do { \ 2390 if ((sc)->sc_ipopts) \ 2391 (void) m_free((sc)->sc_ipopts); \ 2392 if ((sc)->sc_route4.ro_rt != NULL) \ 2393 RTFREE((sc)->sc_route4.ro_rt); \ 2394 pool_put(&syn_cache_pool, (sc)); \ 2395 } while (0) 2396 2397 struct pool syn_cache_pool; 2398 2399 /* 2400 * We don't estimate RTT with SYNs, so each packet starts with the default 2401 * RTT and each timer queue has a fixed timeout value. This allows us to 2402 * optimize the timer queues somewhat. 2403 */ 2404 #define SYN_CACHE_TIMER_ARM(sc) \ 2405 do { \ 2406 TCPT_RANGESET((sc)->sc_rxtcur, \ 2407 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2408 TCPTV_REXMTMAX); \ 2409 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2410 } while (0) 2411 2412 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2413 2414 void 2415 syn_cache_init() 2416 { 2417 int i; 2418 2419 /* Initialize the hash buckets. */ 2420 for (i = 0; i < tcp_syn_cache_size; i++) 2421 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2422 2423 /* Initialize the timer queues. */ 2424 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2425 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2426 2427 /* Initialize the syn cache pool. */ 2428 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2429 "synpl", 0, NULL, NULL, M_PCB); 2430 } 2431 2432 void 2433 syn_cache_insert(sc, tp) 2434 struct syn_cache *sc; 2435 struct tcpcb *tp; 2436 { 2437 struct syn_cache_head *scp; 2438 struct syn_cache *sc2; 2439 int s, i; 2440 2441 /* 2442 * If there are no entries in the hash table, reinitialize 2443 * the hash secrets. 2444 */ 2445 if (syn_cache_count == 0) { 2446 struct timeval tv; 2447 microtime(&tv); 2448 syn_hash1 = random() ^ (u_long)≻ 2449 syn_hash2 = random() ^ tv.tv_usec; 2450 } 2451 2452 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2453 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2454 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2455 2456 /* 2457 * Make sure that we don't overflow the per-bucket 2458 * limit or the total cache size limit. 2459 */ 2460 s = splsoftnet(); 2461 if (scp->sch_length >= tcp_syn_bucket_limit) { 2462 tcpstat.tcps_sc_bucketoverflow++; 2463 /* 2464 * The bucket is full. Toss the oldest element in the 2465 * bucket. This will be the entry with our bucket 2466 * index closest to the front of the timer queue with 2467 * the largest timeout value. 2468 * 2469 * Note: This timer queue traversal may be expensive, so 2470 * we hope that this doesn't happen very often. It is 2471 * much more likely that we'll overflow the entire 2472 * cache, which is much easier to handle; see below. 2473 */ 2474 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2475 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2476 sc2 != NULL; 2477 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2478 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2479 SYN_CACHE_RM(sc2); 2480 SYN_CACHE_PUT(sc2); 2481 goto insert; /* 2 level break */ 2482 } 2483 } 2484 } 2485 #ifdef DIAGNOSTIC 2486 /* 2487 * This should never happen; we should always find an 2488 * entry in our bucket. 2489 */ 2490 panic("syn_cache_insert: bucketoverflow: impossible"); 2491 #endif 2492 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2493 tcpstat.tcps_sc_overflowed++; 2494 /* 2495 * The cache is full. Toss the oldest entry in the 2496 * entire cache. This is the front entry in the 2497 * first non-empty timer queue with the largest 2498 * timeout value. 2499 */ 2500 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2501 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2502 if (sc2 == NULL) 2503 continue; 2504 SYN_CACHE_RM(sc2); 2505 SYN_CACHE_PUT(sc2); 2506 goto insert; /* symmetry with above */ 2507 } 2508 #ifdef DIAGNOSTIC 2509 /* 2510 * This should never happen; we should always find an 2511 * entry in the cache. 2512 */ 2513 panic("syn_cache_insert: cache overflow: impossible"); 2514 #endif 2515 } 2516 2517 insert: 2518 /* 2519 * Initialize the entry's timer. 2520 */ 2521 sc->sc_rxttot = 0; 2522 sc->sc_rxtshift = 0; 2523 SYN_CACHE_TIMER_ARM(sc); 2524 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2525 2526 /* Link it from tcpcb entry */ 2527 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2528 2529 /* Put it into the bucket. */ 2530 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2531 scp->sch_length++; 2532 syn_cache_count++; 2533 2534 tcpstat.tcps_sc_added++; 2535 splx(s); 2536 } 2537 2538 /* 2539 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2540 * If we have retransmitted an entry the maximum number of times, expire 2541 * that entry. 2542 */ 2543 void 2544 syn_cache_timer() 2545 { 2546 struct syn_cache *sc, *nsc; 2547 int i, s; 2548 2549 s = splsoftnet(); 2550 2551 /* 2552 * First, get all the entries that need to be retransmitted, or 2553 * must be expired due to exceeding the initial keepalive time. 2554 */ 2555 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2556 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2557 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2558 sc = nsc) { 2559 nsc = TAILQ_NEXT(sc, sc_timeq); 2560 2561 /* 2562 * Compute the total amount of time this entry has 2563 * been on a queue. If this entry has been on longer 2564 * than the keep alive timer would allow, expire it. 2565 */ 2566 sc->sc_rxttot += sc->sc_rxtcur; 2567 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2568 tcpstat.tcps_sc_timed_out++; 2569 SYN_CACHE_RM(sc); 2570 SYN_CACHE_PUT(sc); 2571 continue; 2572 } 2573 2574 tcpstat.tcps_sc_retransmitted++; 2575 (void) syn_cache_respond(sc, NULL); 2576 2577 /* Advance this entry onto the next timer queue. */ 2578 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2579 sc->sc_rxtshift = i + 1; 2580 SYN_CACHE_TIMER_ARM(sc); 2581 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2582 sc, sc_timeq); 2583 } 2584 } 2585 2586 /* 2587 * Now get all the entries that are expired due to too many 2588 * retransmissions. 2589 */ 2590 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2591 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2592 sc = nsc) { 2593 nsc = TAILQ_NEXT(sc, sc_timeq); 2594 tcpstat.tcps_sc_timed_out++; 2595 SYN_CACHE_RM(sc); 2596 SYN_CACHE_PUT(sc); 2597 } 2598 splx(s); 2599 } 2600 2601 /* 2602 * Remove syn cache created by the specified tcb entry, 2603 * because this does not make sense to keep them 2604 * (if there's no tcb entry, syn cache entry will never be used) 2605 */ 2606 void 2607 syn_cache_cleanup(tp) 2608 struct tcpcb *tp; 2609 { 2610 struct syn_cache *sc, *nsc; 2611 int s; 2612 2613 s = splsoftnet(); 2614 2615 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2616 nsc = LIST_NEXT(sc, sc_tpq); 2617 2618 #ifdef DIAGNOSTIC 2619 if (sc->sc_tp != tp) 2620 panic("invalid sc_tp in syn_cache_cleanup"); 2621 #endif 2622 SYN_CACHE_RM(sc); 2623 SYN_CACHE_PUT(sc); 2624 } 2625 /* just for safety */ 2626 LIST_INIT(&tp->t_sc); 2627 2628 splx(s); 2629 } 2630 2631 /* 2632 * Find an entry in the syn cache. 2633 */ 2634 struct syn_cache * 2635 syn_cache_lookup(src, dst, headp) 2636 struct sockaddr *src; 2637 struct sockaddr *dst; 2638 struct syn_cache_head **headp; 2639 { 2640 struct syn_cache *sc; 2641 struct syn_cache_head *scp; 2642 u_int32_t hash; 2643 int s; 2644 2645 SYN_HASHALL(hash, src, dst); 2646 2647 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2648 *headp = scp; 2649 s = splsoftnet(); 2650 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2651 sc = LIST_NEXT(sc, sc_bucketq)) { 2652 if (sc->sc_hash != hash) 2653 continue; 2654 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2655 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2656 splx(s); 2657 return (sc); 2658 } 2659 } 2660 splx(s); 2661 return (NULL); 2662 } 2663 2664 /* 2665 * This function gets called when we receive an ACK for a 2666 * socket in the LISTEN state. We look up the connection 2667 * in the syn cache, and if its there, we pull it out of 2668 * the cache and turn it into a full-blown connection in 2669 * the SYN-RECEIVED state. 2670 * 2671 * The return values may not be immediately obvious, and their effects 2672 * can be subtle, so here they are: 2673 * 2674 * NULL SYN was not found in cache; caller should drop the 2675 * packet and send an RST. 2676 * 2677 * -1 We were unable to create the new connection, and are 2678 * aborting it. An ACK,RST is being sent to the peer 2679 * (unless we got screwey sequence numbners; see below), 2680 * because the 3-way handshake has been completed. Caller 2681 * should not free the mbuf, since we may be using it. If 2682 * we are not, we will free it. 2683 * 2684 * Otherwise, the return value is a pointer to the new socket 2685 * associated with the connection. 2686 */ 2687 struct socket * 2688 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2689 struct sockaddr *src; 2690 struct sockaddr *dst; 2691 struct tcphdr *th; 2692 unsigned int hlen, tlen; 2693 struct socket *so; 2694 struct mbuf *m; 2695 { 2696 struct syn_cache *sc; 2697 struct syn_cache_head *scp; 2698 register struct inpcb *inp = NULL; 2699 #ifdef INET6 2700 register struct in6pcb *in6p = NULL; 2701 #endif 2702 register struct tcpcb *tp = 0; 2703 struct mbuf *am; 2704 int s; 2705 struct socket *oso; 2706 2707 s = splsoftnet(); 2708 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2709 splx(s); 2710 return (NULL); 2711 } 2712 2713 /* 2714 * Verify the sequence and ack numbers. Try getting the correct 2715 * response again. 2716 */ 2717 if ((th->th_ack != sc->sc_iss + 1) || 2718 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2719 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2720 (void) syn_cache_respond(sc, m); 2721 splx(s); 2722 return ((struct socket *)(-1)); 2723 } 2724 2725 /* Remove this cache entry */ 2726 SYN_CACHE_RM(sc); 2727 splx(s); 2728 2729 /* 2730 * Ok, create the full blown connection, and set things up 2731 * as they would have been set up if we had created the 2732 * connection when the SYN arrived. If we can't create 2733 * the connection, abort it. 2734 */ 2735 /* 2736 * inp still has the OLD in_pcb stuff, set the 2737 * v6-related flags on the new guy, too. This is 2738 * done particularly for the case where an AF_INET6 2739 * socket is bound only to a port, and a v4 connection 2740 * comes in on that port. 2741 * we also copy the flowinfo from the original pcb 2742 * to the new one. 2743 */ 2744 { 2745 struct inpcb *parentinpcb; 2746 2747 parentinpcb = (struct inpcb *)so->so_pcb; 2748 2749 oso = so; 2750 so = sonewconn(so, SS_ISCONNECTED); 2751 if (so == NULL) 2752 goto resetandabort; 2753 2754 switch (so->so_proto->pr_domain->dom_family) { 2755 case AF_INET: 2756 inp = sotoinpcb(so); 2757 break; 2758 #ifdef INET6 2759 case AF_INET6: 2760 in6p = sotoin6pcb(so); 2761 #if 0 /*def INET6*/ 2762 inp->inp_flags |= (parentinpcb->inp_flags & 2763 (INP_IPV6 | INP_IPV6_UNDEC | INP_IPV6_MAPPED)); 2764 if ((inp->inp_flags & INP_IPV6) && 2765 !(inp->inp_flags & INP_IPV6_MAPPED)) { 2766 inp->inp_ipv6.ip6_hlim = parentinpcb->inp_ipv6.ip6_hlim; 2767 inp->inp_ipv6.ip6_vfc = parentinpcb->inp_ipv6.ip6_vfc; 2768 } 2769 #endif 2770 break; 2771 #endif 2772 } 2773 } 2774 switch (src->sa_family) { 2775 case AF_INET: 2776 if (inp) { 2777 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2778 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2779 inp->inp_options = ip_srcroute(); 2780 in_pcbstate(inp, INP_BOUND); 2781 if (inp->inp_options == NULL) { 2782 inp->inp_options = sc->sc_ipopts; 2783 sc->sc_ipopts = NULL; 2784 } 2785 } 2786 #ifdef INET6 2787 else if (in6p) { 2788 /* IPv4 packet to AF_INET6 socket */ 2789 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2790 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2791 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2792 &in6p->in6p_laddr.s6_addr32[3], 2793 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2794 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2795 in6totcpcb(in6p)->t_family = AF_INET; 2796 } 2797 #endif 2798 break; 2799 #ifdef INET6 2800 case AF_INET6: 2801 if (in6p) { 2802 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 2803 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 2804 #if 0 2805 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 2806 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 2807 #endif 2808 } 2809 break; 2810 #endif 2811 } 2812 #ifdef INET6 2813 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 2814 struct in6pcb *oin6p = sotoin6pcb(oso); 2815 /* inherit socket options from the listening socket */ 2816 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 2817 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 2818 m_freem(in6p->in6p_options); 2819 in6p->in6p_options = 0; 2820 } 2821 ip6_savecontrol(in6p, &in6p->in6p_options, 2822 mtod(m, struct ip6_hdr *), m); 2823 } 2824 #endif 2825 2826 #ifdef IPSEC 2827 { 2828 struct secpolicy *sp; 2829 if (inp) { 2830 sp = ipsec_copy_policy(sotoinpcb(oso)->inp_sp); 2831 if (sp) { 2832 key_freesp(inp->inp_sp); 2833 inp->inp_sp = sp; 2834 } else 2835 printf("tcp_input: could not copy policy\n"); 2836 } 2837 #ifdef INET6 2838 else if (in6p) { 2839 sp = ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp); 2840 if (sp) { 2841 key_freesp(in6p->in6p_sp); 2842 in6p->in6p_sp = sp; 2843 } else 2844 printf("tcp_input: could not copy policy\n"); 2845 } 2846 #endif 2847 } 2848 #endif 2849 2850 /* 2851 * Give the new socket our cached route reference. 2852 */ 2853 if (inp) 2854 inp->inp_route = sc->sc_route4; /* struct assignment */ 2855 #ifdef INET6 2856 else 2857 in6p->in6p_route = sc->sc_route6; 2858 #endif 2859 sc->sc_route4.ro_rt = NULL; 2860 2861 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 2862 if (am == NULL) 2863 goto resetandabort; 2864 am->m_len = src->sa_len; 2865 bcopy(src, mtod(am, caddr_t), src->sa_len); 2866 if (inp) { 2867 if (in_pcbconnect(inp, am)) { 2868 (void) m_free(am); 2869 goto resetandabort; 2870 } 2871 } 2872 #ifdef INET6 2873 else if (in6p) { 2874 if (src->sa_family == AF_INET) { 2875 /* IPv4 packet to AF_INET6 socket */ 2876 struct sockaddr_in6 *sin6; 2877 sin6 = mtod(am, struct sockaddr_in6 *); 2878 am->m_len = sizeof(*sin6); 2879 bzero(sin6, sizeof(*sin6)); 2880 sin6->sin6_family = AF_INET6; 2881 sin6->sin6_len = sizeof(*sin6); 2882 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 2883 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 2884 bcopy(&((struct sockaddr_in *)src)->sin_addr, 2885 &sin6->sin6_addr.s6_addr32[3], 2886 sizeof(sin6->sin6_addr.s6_addr32[3])); 2887 } 2888 if (in6_pcbconnect(in6p, am)) { 2889 (void) m_free(am); 2890 goto resetandabort; 2891 } 2892 } 2893 #endif 2894 else { 2895 (void) m_free(am); 2896 goto resetandabort; 2897 } 2898 (void) m_free(am); 2899 2900 if (inp) 2901 tp = intotcpcb(inp); 2902 #ifdef INET6 2903 else if (in6p) 2904 tp = in6totcpcb(in6p); 2905 #endif 2906 else 2907 tp = NULL; 2908 if (sc->sc_request_r_scale != 15) { 2909 tp->requested_s_scale = sc->sc_requested_s_scale; 2910 tp->request_r_scale = sc->sc_request_r_scale; 2911 tp->snd_scale = sc->sc_requested_s_scale; 2912 tp->rcv_scale = sc->sc_request_r_scale; 2913 tp->t_flags |= TF_RCVD_SCALE; 2914 } 2915 if (sc->sc_flags & SCF_TIMESTAMP) 2916 tp->t_flags |= TF_RCVD_TSTMP; 2917 2918 tp->t_template = tcp_template(tp); 2919 if (tp->t_template == 0) { 2920 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 2921 so = NULL; 2922 m_freem(m); 2923 goto abort; 2924 } 2925 2926 tp->iss = sc->sc_iss; 2927 tp->irs = sc->sc_irs; 2928 tcp_sendseqinit(tp); 2929 tcp_rcvseqinit(tp); 2930 tp->t_state = TCPS_SYN_RECEIVED; 2931 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 2932 tcpstat.tcps_accepts++; 2933 2934 /* Initialize tp->t_ourmss before we deal with the peer's! */ 2935 tp->t_ourmss = sc->sc_ourmaxseg; 2936 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 2937 2938 /* 2939 * Initialize the initial congestion window. If we 2940 * had to retransmit the SYN,ACK, we must initialize cwnd 2941 * to 1 segment (i.e. the Loss Window). 2942 */ 2943 if (sc->sc_rxtshift) 2944 tp->snd_cwnd = tp->t_peermss; 2945 else 2946 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 2947 2948 tcp_rmx_rtt(tp); 2949 tp->snd_wl1 = sc->sc_irs; 2950 tp->rcv_up = sc->sc_irs + 1; 2951 2952 /* 2953 * This is what whould have happened in tcp_ouput() when 2954 * the SYN,ACK was sent. 2955 */ 2956 tp->snd_up = tp->snd_una; 2957 tp->snd_max = tp->snd_nxt = tp->iss+1; 2958 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2959 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 2960 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 2961 tp->last_ack_sent = tp->rcv_nxt; 2962 2963 tcpstat.tcps_sc_completed++; 2964 SYN_CACHE_PUT(sc); 2965 return (so); 2966 2967 resetandabort: 2968 (void) tcp_respond(NULL, m, m, th, 2969 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 2970 abort: 2971 if (so != NULL) 2972 (void) soabort(so); 2973 SYN_CACHE_PUT(sc); 2974 tcpstat.tcps_sc_aborted++; 2975 return ((struct socket *)(-1)); 2976 } 2977 2978 /* 2979 * This function is called when we get a RST for a 2980 * non-existant connection, so that we can see if the 2981 * connection is in the syn cache. If it is, zap it. 2982 */ 2983 2984 void 2985 syn_cache_reset(src, dst, th) 2986 struct sockaddr *src; 2987 struct sockaddr *dst; 2988 struct tcphdr *th; 2989 { 2990 struct syn_cache *sc; 2991 struct syn_cache_head *scp; 2992 int s = splsoftnet(); 2993 2994 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2995 splx(s); 2996 return; 2997 } 2998 if (SEQ_LT(th->th_seq, sc->sc_irs) || 2999 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3000 splx(s); 3001 return; 3002 } 3003 SYN_CACHE_RM(sc); 3004 splx(s); 3005 tcpstat.tcps_sc_reset++; 3006 SYN_CACHE_PUT(sc); 3007 } 3008 3009 void 3010 syn_cache_unreach(src, dst, th) 3011 struct sockaddr *src; 3012 struct sockaddr *dst; 3013 struct tcphdr *th; 3014 { 3015 struct syn_cache *sc; 3016 struct syn_cache_head *scp; 3017 int s; 3018 3019 s = splsoftnet(); 3020 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3021 splx(s); 3022 return; 3023 } 3024 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3025 if (ntohl (th->th_seq) != sc->sc_iss) { 3026 splx(s); 3027 return; 3028 } 3029 3030 /* 3031 * If we've rertransmitted 3 times and this is our second error, 3032 * we remove the entry. Otherwise, we allow it to continue on. 3033 * This prevents us from incorrectly nuking an entry during a 3034 * spurious network outage. 3035 * 3036 * See tcp_notify(). 3037 */ 3038 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3039 sc->sc_flags |= SCF_UNREACH; 3040 splx(s); 3041 return; 3042 } 3043 3044 SYN_CACHE_RM(sc); 3045 splx(s); 3046 tcpstat.tcps_sc_unreach++; 3047 SYN_CACHE_PUT(sc); 3048 } 3049 3050 /* 3051 * Given a LISTEN socket and an inbound SYN request, add 3052 * this to the syn cache, and send back a segment: 3053 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3054 * to the source. 3055 * 3056 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3057 * Doing so would require that we hold onto the data and deliver it 3058 * to the application. However, if we are the target of a SYN-flood 3059 * DoS attack, an attacker could send data which would eventually 3060 * consume all available buffer space if it were ACKed. By not ACKing 3061 * the data, we avoid this DoS scenario. 3062 */ 3063 3064 int 3065 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3066 struct sockaddr *src; 3067 struct sockaddr *dst; 3068 struct tcphdr *th; 3069 unsigned int hlen; 3070 struct socket *so; 3071 struct mbuf *m; 3072 u_char *optp; 3073 int optlen; 3074 struct tcp_opt_info *oi; 3075 { 3076 struct tcpcb tb, *tp; 3077 long win; 3078 struct syn_cache *sc; 3079 struct syn_cache_head *scp; 3080 struct mbuf *ipopts; 3081 3082 tp = sototcpcb(so); 3083 3084 /* 3085 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3086 * in_broadcast() should never return true on a received 3087 * packet with M_BCAST not set. 3088 */ 3089 if (m->m_flags & (M_BCAST|M_MCAST)) 3090 return 0; 3091 #ifdef INET6 3092 if (m->m_flags & M_ANYCAST6) 3093 return 0; 3094 #endif 3095 3096 switch (src->sa_family) { 3097 case AF_INET: 3098 if (IN_MULTICAST(((struct sockaddr_in *)src)->sin_addr.s_addr) 3099 || IN_MULTICAST(((struct sockaddr_in *)dst)->sin_addr.s_addr)) 3100 return 0; 3101 break; 3102 #ifdef INET6 3103 case AF_INET6: 3104 if (IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)src)->sin6_addr) 3105 || IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)dst)->sin6_addr)) 3106 return 0; 3107 break; 3108 #endif 3109 default: 3110 return 0; 3111 } 3112 3113 /* 3114 * Initialize some local state. 3115 */ 3116 win = sbspace(&so->so_rcv); 3117 if (win > TCP_MAXWIN) 3118 win = TCP_MAXWIN; 3119 3120 if (src->sa_family == AF_INET) { 3121 /* 3122 * Remember the IP options, if any. 3123 */ 3124 ipopts = ip_srcroute(); 3125 } else 3126 ipopts = NULL; 3127 3128 if (optp) { 3129 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3130 tcp_dooptions(&tb, optp, optlen, th, oi); 3131 } else 3132 tb.t_flags = 0; 3133 3134 /* 3135 * See if we already have an entry for this connection. 3136 * If we do, resend the SYN,ACK. We do not count this 3137 * as a retransmission (XXX though maybe we should). 3138 */ 3139 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3140 tcpstat.tcps_sc_dupesyn++; 3141 if (ipopts) { 3142 /* 3143 * If we were remembering a previous source route, 3144 * forget it and use the new one we've been given. 3145 */ 3146 if (sc->sc_ipopts) 3147 (void) m_free(sc->sc_ipopts); 3148 sc->sc_ipopts = ipopts; 3149 } 3150 sc->sc_timestamp = tb.ts_recent; 3151 if (syn_cache_respond(sc, m) == 0) { 3152 tcpstat.tcps_sndacks++; 3153 tcpstat.tcps_sndtotal++; 3154 } 3155 return (1); 3156 } 3157 3158 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3159 if (sc == NULL) { 3160 if (ipopts) 3161 (void) m_free(ipopts); 3162 return (0); 3163 } 3164 3165 /* 3166 * Fill in the cache, and put the necessary IP and TCP 3167 * options into the reply. 3168 */ 3169 bzero(sc, sizeof(struct syn_cache)); 3170 bcopy(src, &sc->sc_src, src->sa_len); 3171 bcopy(dst, &sc->sc_dst, dst->sa_len); 3172 sc->sc_flags = 0; 3173 sc->sc_ipopts = ipopts; 3174 sc->sc_irs = th->th_seq; 3175 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 3176 sc->sc_peermaxseg = oi->maxseg; 3177 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3178 m->m_pkthdr.rcvif : NULL, 3179 sc->sc_src.sa.sa_family); 3180 sc->sc_win = win; 3181 sc->sc_timestamp = tb.ts_recent; 3182 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3183 sc->sc_flags |= SCF_TIMESTAMP; 3184 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3185 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3186 sc->sc_requested_s_scale = tb.requested_s_scale; 3187 sc->sc_request_r_scale = 0; 3188 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3189 TCP_MAXWIN << sc->sc_request_r_scale < 3190 so->so_rcv.sb_hiwat) 3191 sc->sc_request_r_scale++; 3192 } else { 3193 sc->sc_requested_s_scale = 15; 3194 sc->sc_request_r_scale = 15; 3195 } 3196 sc->sc_tp = tp; 3197 if (syn_cache_respond(sc, m) == 0) { 3198 syn_cache_insert(sc, tp); 3199 tcpstat.tcps_sndacks++; 3200 tcpstat.tcps_sndtotal++; 3201 } else { 3202 SYN_CACHE_PUT(sc); 3203 tcpstat.tcps_sc_dropped++; 3204 } 3205 return (1); 3206 } 3207 3208 int 3209 syn_cache_respond(sc, m) 3210 struct syn_cache *sc; 3211 struct mbuf *m; 3212 { 3213 struct route *ro = &sc->sc_route4; 3214 struct rtentry *rt; 3215 u_int8_t *optp; 3216 int optlen, error; 3217 u_int16_t tlen; 3218 struct ip *ip = NULL; 3219 #ifdef INET6 3220 struct ip6_hdr *ip6 = NULL; 3221 #endif 3222 struct tcphdr *th; 3223 u_int hlen; 3224 3225 switch (sc->sc_src.sa.sa_family) { 3226 case AF_INET: 3227 hlen = sizeof(struct ip); 3228 break; 3229 #ifdef INET6 3230 case AF_INET6: 3231 hlen = sizeof(struct ip6_hdr); 3232 break; 3233 #endif 3234 default: 3235 if (m) 3236 m_freem(m); 3237 return EAFNOSUPPORT; 3238 } 3239 3240 /* Compute the size of the TCP options. */ 3241 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3242 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3243 3244 tlen = hlen + sizeof(struct tcphdr) + optlen; 3245 3246 /* 3247 * Create the IP+TCP header from scratch. Reuse the received mbuf 3248 * if possible. 3249 */ 3250 if (m != NULL) { 3251 m_freem(m->m_next); 3252 m->m_next = NULL; 3253 MRESETDATA(m); 3254 } else { 3255 MGETHDR(m, M_DONTWAIT, MT_DATA); 3256 if (m == NULL) 3257 return (ENOBUFS); 3258 } 3259 3260 /* Fixup the mbuf. */ 3261 m->m_data += max_linkhdr; 3262 m->m_len = m->m_pkthdr.len = tlen; 3263 #ifdef IPSEC 3264 if (sc->sc_tp) { 3265 struct tcpcb *tp; 3266 struct socket *so; 3267 3268 tp = sc->sc_tp; 3269 if (tp->t_inpcb) 3270 so = tp->t_inpcb->inp_socket; 3271 #ifdef INET6 3272 else if (tp->t_in6pcb) 3273 so = tp->t_in6pcb->in6p_socket; 3274 #endif 3275 else 3276 so = NULL; 3277 /* use IPsec policy on listening socket, on SYN ACK */ 3278 m->m_pkthdr.rcvif = (struct ifnet *)so; 3279 } 3280 #else 3281 m->m_pkthdr.rcvif = NULL; 3282 #endif 3283 memset(mtod(m, u_char *), 0, tlen); 3284 3285 switch (sc->sc_src.sa.sa_family) { 3286 case AF_INET: 3287 ip = mtod(m, struct ip *); 3288 ip->ip_dst = sc->sc_src.sin.sin_addr; 3289 ip->ip_src = sc->sc_dst.sin.sin_addr; 3290 ip->ip_p = IPPROTO_TCP; 3291 th = (struct tcphdr *)(ip + 1); 3292 th->th_dport = sc->sc_src.sin.sin_port; 3293 th->th_sport = sc->sc_dst.sin.sin_port; 3294 break; 3295 #ifdef INET6 3296 case AF_INET6: 3297 ip6 = mtod(m, struct ip6_hdr *); 3298 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3299 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3300 ip6->ip6_nxt = IPPROTO_TCP; 3301 /* ip6_plen will be updated in ip6_output() */ 3302 th = (struct tcphdr *)(ip6 + 1); 3303 th->th_dport = sc->sc_src.sin6.sin6_port; 3304 th->th_sport = sc->sc_dst.sin6.sin6_port; 3305 break; 3306 #endif 3307 default: 3308 th = NULL; 3309 } 3310 3311 th->th_seq = htonl(sc->sc_iss); 3312 th->th_ack = htonl(sc->sc_irs + 1); 3313 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3314 th->th_flags = TH_SYN|TH_ACK; 3315 th->th_win = htons(sc->sc_win); 3316 /* th_sum already 0 */ 3317 /* th_urp already 0 */ 3318 3319 /* Tack on the TCP options. */ 3320 optp = (u_int8_t *)(th + 1); 3321 *optp++ = TCPOPT_MAXSEG; 3322 *optp++ = 4; 3323 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3324 *optp++ = sc->sc_ourmaxseg & 0xff; 3325 3326 if (sc->sc_request_r_scale != 15) { 3327 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3328 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3329 sc->sc_request_r_scale); 3330 optp += 4; 3331 } 3332 3333 if (sc->sc_flags & SCF_TIMESTAMP) { 3334 u_int32_t *lp = (u_int32_t *)(optp); 3335 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3336 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3337 *lp++ = htonl(tcp_now); 3338 *lp = htonl(sc->sc_timestamp); 3339 optp += TCPOLEN_TSTAMP_APPA; 3340 } 3341 3342 /* Compute the packet's checksum. */ 3343 switch (sc->sc_src.sa.sa_family) { 3344 case AF_INET: 3345 ip->ip_len = htons(tlen - hlen); 3346 th->th_sum = 0; 3347 th->th_sum = in_cksum(m, tlen); 3348 break; 3349 #ifdef INET6 3350 case AF_INET6: 3351 ip6->ip6_plen = htons(tlen - hlen); 3352 th->th_sum = 0; 3353 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3354 break; 3355 #endif 3356 } 3357 3358 /* 3359 * Fill in some straggling IP bits. Note the stack expects 3360 * ip_len to be in host order, for convenience. 3361 */ 3362 switch (sc->sc_src.sa.sa_family) { 3363 case AF_INET: 3364 ip->ip_len = tlen; 3365 ip->ip_ttl = ip_defttl; 3366 /* XXX tos? */ 3367 break; 3368 #ifdef INET6 3369 case AF_INET6: 3370 ip6->ip6_vfc = IPV6_VERSION; 3371 ip6->ip6_plen = htons(tlen - hlen); 3372 ip6->ip6_hlim = ip6_defhlim; 3373 /* XXX flowlabel? */ 3374 break; 3375 #endif 3376 } 3377 3378 /* 3379 * If we're doing Path MTU discovery, we need to set DF unless 3380 * the route's MTU is locked. If we don't yet know the route, 3381 * look it up now. We will copy this reference to the inpcb 3382 * when we finish creating the connection. 3383 */ 3384 if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) { 3385 if (ro->ro_rt != NULL) { 3386 RTFREE(ro->ro_rt); 3387 ro->ro_rt = NULL; 3388 } 3389 bcopy(&sc->sc_src, &ro->ro_dst, sc->sc_src.sa.sa_len); 3390 rtalloc(ro); 3391 if ((rt = ro->ro_rt) == NULL) { 3392 m_freem(m); 3393 switch (sc->sc_src.sa.sa_family) { 3394 case AF_INET: 3395 ipstat.ips_noroute++; 3396 break; 3397 #ifdef INET6 3398 case AF_INET6: 3399 ip6stat.ip6s_noroute++; 3400 break; 3401 #endif 3402 } 3403 return (EHOSTUNREACH); 3404 } 3405 } 3406 3407 switch (sc->sc_src.sa.sa_family) { 3408 case AF_INET: 3409 if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0) 3410 ip->ip_off |= IP_DF; 3411 3412 /* ...and send it off! */ 3413 error = ip_output(m, sc->sc_ipopts, ro, 0, NULL); 3414 break; 3415 #ifdef INET6 3416 case AF_INET6: 3417 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3418 0, NULL); 3419 break; 3420 #endif 3421 default: 3422 error = EAFNOSUPPORT; 3423 break; 3424 } 3425 return (error); 3426 } 3427