1 /* $NetBSD: tcp_input.c,v 1.126 2001/06/19 13:42:19 wiz 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, 2001 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 #include "opt_inet_csum.h" 130 131 #include <sys/param.h> 132 #include <sys/systm.h> 133 #include <sys/malloc.h> 134 #include <sys/mbuf.h> 135 #include <sys/protosw.h> 136 #include <sys/socket.h> 137 #include <sys/socketvar.h> 138 #include <sys/errno.h> 139 #include <sys/syslog.h> 140 #include <sys/pool.h> 141 #include <sys/domain.h> 142 143 #include <net/if.h> 144 #include <net/route.h> 145 #include <net/if_types.h> 146 147 #include <netinet/in.h> 148 #include <netinet/in_systm.h> 149 #include <netinet/ip.h> 150 #include <netinet/in_pcb.h> 151 #include <netinet/ip_var.h> 152 153 #ifdef INET6 154 #ifndef INET 155 #include <netinet/in.h> 156 #endif 157 #include <netinet/ip6.h> 158 #include <netinet6/ip6_var.h> 159 #include <netinet6/in6_pcb.h> 160 #include <netinet6/ip6_var.h> 161 #include <netinet6/in6_var.h> 162 #include <netinet/icmp6.h> 163 #include <netinet6/nd6.h> 164 #endif 165 166 #ifdef PULLDOWN_TEST 167 #ifndef INET6 168 /* always need ip6.h for IP6_EXTHDR_GET */ 169 #include <netinet/ip6.h> 170 #endif 171 #endif 172 173 #include <netinet/tcp.h> 174 #include <netinet/tcp_fsm.h> 175 #include <netinet/tcp_seq.h> 176 #include <netinet/tcp_timer.h> 177 #include <netinet/tcp_var.h> 178 #include <netinet/tcpip.h> 179 #include <netinet/tcp_debug.h> 180 181 #include <machine/stdarg.h> 182 183 #ifdef IPSEC 184 #include <netinet6/ipsec.h> 185 #include <netkey/key.h> 186 #endif /*IPSEC*/ 187 #ifdef INET6 188 #include "faith.h" 189 #if defined(NFAITH) && NFAITH > 0 190 #include <net/if_faith.h> 191 #endif 192 #endif 193 194 int tcprexmtthresh = 3; 195 int tcp_log_refused; 196 197 static int tcp_rst_ppslim_count = 0; 198 static struct timeval tcp_rst_ppslim_last; 199 200 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 201 202 /* for modulo comparisons of timestamps */ 203 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 204 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 205 206 /* 207 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 208 */ 209 #ifdef INET6 210 #define ND6_HINT(tp) \ 211 do { \ 212 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 213 && tp->t_in6pcb->in6p_route.ro_rt) { \ 214 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 215 } \ 216 } while (0) 217 #else 218 #define ND6_HINT(tp) 219 #endif 220 221 /* 222 * Macro to compute ACK transmission behavior. Delay the ACK unless 223 * we have already delayed an ACK (must send an ACK every two segments). 224 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 225 * option is enabled. 226 */ 227 #define TCP_SETUP_ACK(tp, th) \ 228 do { \ 229 if ((tp)->t_flags & TF_DELACK || \ 230 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 231 tp->t_flags |= TF_ACKNOW; \ 232 else \ 233 TCP_SET_DELACK(tp); \ 234 } while (0) 235 236 /* 237 * Convert TCP protocol fields to host order for easier processing. 238 */ 239 #define TCP_FIELDS_TO_HOST(th) \ 240 do { \ 241 NTOHL((th)->th_seq); \ 242 NTOHL((th)->th_ack); \ 243 NTOHS((th)->th_win); \ 244 NTOHS((th)->th_urp); \ 245 } while (0) 246 247 #ifdef TCP_CSUM_COUNTERS 248 #include <sys/device.h> 249 250 extern struct evcnt tcp_hwcsum_ok; 251 extern struct evcnt tcp_hwcsum_bad; 252 extern struct evcnt tcp_hwcsum_data; 253 extern struct evcnt tcp_swcsum; 254 255 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 256 257 #else 258 259 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 260 261 #endif /* TCP_CSUM_COUNTERS */ 262 263 int 264 tcp_reass(tp, th, m, tlen) 265 struct tcpcb *tp; 266 struct tcphdr *th; 267 struct mbuf *m; 268 int *tlen; 269 { 270 struct ipqent *p, *q, *nq, *tiqe = NULL; 271 struct socket *so = NULL; 272 int pkt_flags; 273 tcp_seq pkt_seq; 274 unsigned pkt_len; 275 u_long rcvpartdupbyte = 0; 276 u_long rcvoobyte; 277 278 if (tp->t_inpcb) 279 so = tp->t_inpcb->inp_socket; 280 #ifdef INET6 281 else if (tp->t_in6pcb) 282 so = tp->t_in6pcb->in6p_socket; 283 #endif 284 285 TCP_REASS_LOCK_CHECK(tp); 286 287 /* 288 * Call with th==0 after become established to 289 * force pre-ESTABLISHED data up to user socket. 290 */ 291 if (th == 0) 292 goto present; 293 294 rcvoobyte = *tlen; 295 /* 296 * Copy these to local variables because the tcpiphdr 297 * gets munged while we are collapsing mbufs. 298 */ 299 pkt_seq = th->th_seq; 300 pkt_len = *tlen; 301 pkt_flags = th->th_flags; 302 /* 303 * Find a segment which begins after this one does. 304 */ 305 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 306 nq = q->ipqe_q.le_next; 307 /* 308 * If the received segment is just right after this 309 * fragment, merge the two together and then check 310 * for further overlaps. 311 */ 312 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 313 #ifdef TCPREASS_DEBUG 314 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 315 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 316 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 317 #endif 318 pkt_len += q->ipqe_len; 319 pkt_flags |= q->ipqe_flags; 320 pkt_seq = q->ipqe_seq; 321 m_cat(q->ipqe_m, m); 322 m = q->ipqe_m; 323 goto free_ipqe; 324 } 325 /* 326 * If the received segment is completely past this 327 * fragment, we need to go the next fragment. 328 */ 329 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 330 p = q; 331 continue; 332 } 333 /* 334 * If the fragment is past the received segment, 335 * it (or any following) can't be concatenated. 336 */ 337 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 338 break; 339 /* 340 * We've received all the data in this segment before. 341 * mark it as a duplicate and return. 342 */ 343 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 344 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 345 tcpstat.tcps_rcvduppack++; 346 tcpstat.tcps_rcvdupbyte += pkt_len; 347 m_freem(m); 348 if (tiqe != NULL) 349 pool_put(&ipqent_pool, tiqe); 350 return (0); 351 } 352 /* 353 * Received segment completely overlaps this fragment 354 * so we drop the fragment (this keeps the temporal 355 * ordering of segments correct). 356 */ 357 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 358 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 359 rcvpartdupbyte += q->ipqe_len; 360 m_freem(q->ipqe_m); 361 goto free_ipqe; 362 } 363 /* 364 * RX'ed segment extends past the end of the 365 * fragment. Drop the overlapping bytes. Then 366 * merge the fragment and segment then treat as 367 * a longer received packet. 368 */ 369 if (SEQ_LT(q->ipqe_seq, pkt_seq) 370 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 371 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 372 #ifdef TCPREASS_DEBUG 373 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 374 tp, overlap, 375 pkt_seq, pkt_seq + pkt_len, pkt_len); 376 #endif 377 m_adj(m, overlap); 378 rcvpartdupbyte += overlap; 379 m_cat(q->ipqe_m, m); 380 m = q->ipqe_m; 381 pkt_seq = q->ipqe_seq; 382 pkt_len += q->ipqe_len - overlap; 383 rcvoobyte -= overlap; 384 goto free_ipqe; 385 } 386 /* 387 * RX'ed segment extends past the front of the 388 * fragment. Drop the overlapping bytes on the 389 * received packet. The packet will then be 390 * contatentated with this fragment a bit later. 391 */ 392 if (SEQ_GT(q->ipqe_seq, pkt_seq) 393 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 394 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 395 #ifdef TCPREASS_DEBUG 396 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 397 tp, overlap, 398 pkt_seq, pkt_seq + pkt_len, pkt_len); 399 #endif 400 m_adj(m, -overlap); 401 pkt_len -= overlap; 402 rcvpartdupbyte += overlap; 403 rcvoobyte -= overlap; 404 } 405 /* 406 * If the received segment immediates precedes this 407 * fragment then tack the fragment onto this segment 408 * and reinsert the data. 409 */ 410 if (q->ipqe_seq == pkt_seq + pkt_len) { 411 #ifdef TCPREASS_DEBUG 412 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 413 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 414 pkt_seq, pkt_seq + pkt_len, pkt_len); 415 #endif 416 pkt_len += q->ipqe_len; 417 pkt_flags |= q->ipqe_flags; 418 m_cat(m, q->ipqe_m); 419 LIST_REMOVE(q, ipqe_q); 420 LIST_REMOVE(q, ipqe_timeq); 421 if (tiqe == NULL) { 422 tiqe = q; 423 } else { 424 pool_put(&ipqent_pool, q); 425 } 426 break; 427 } 428 /* 429 * If the fragment is before the segment, remember it. 430 * When this loop is terminated, p will contain the 431 * pointer to fragment that is right before the received 432 * segment. 433 */ 434 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 435 p = q; 436 437 continue; 438 439 /* 440 * This is a common operation. It also will allow 441 * to save doing a malloc/free in most instances. 442 */ 443 free_ipqe: 444 LIST_REMOVE(q, ipqe_q); 445 LIST_REMOVE(q, ipqe_timeq); 446 if (tiqe == NULL) { 447 tiqe = q; 448 } else { 449 pool_put(&ipqent_pool, q); 450 } 451 } 452 453 /* 454 * Allocate a new queue entry since the received segment did not 455 * collapse onto any other out-of-order block; thus we are allocating 456 * a new block. If it had collapsed, tiqe would not be NULL and 457 * we would be reusing it. 458 * XXX If we can't, just drop the packet. XXX 459 */ 460 if (tiqe == NULL) { 461 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 462 if (tiqe == NULL) { 463 tcpstat.tcps_rcvmemdrop++; 464 m_freem(m); 465 return (0); 466 } 467 } 468 469 /* 470 * Update the counters. 471 */ 472 tcpstat.tcps_rcvoopack++; 473 tcpstat.tcps_rcvoobyte += rcvoobyte; 474 if (rcvpartdupbyte) { 475 tcpstat.tcps_rcvpartduppack++; 476 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 477 } 478 479 /* 480 * Insert the new fragment queue entry into both queues. 481 */ 482 tiqe->ipqe_m = m; 483 tiqe->ipqe_seq = pkt_seq; 484 tiqe->ipqe_len = pkt_len; 485 tiqe->ipqe_flags = pkt_flags; 486 if (p == NULL) { 487 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 488 #ifdef TCPREASS_DEBUG 489 if (tiqe->ipqe_seq != tp->rcv_nxt) 490 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 491 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 492 #endif 493 } else { 494 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 495 #ifdef TCPREASS_DEBUG 496 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 497 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 498 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 499 #endif 500 } 501 502 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 503 504 present: 505 /* 506 * Present data to user, advancing rcv_nxt through 507 * completed sequence space. 508 */ 509 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 510 return (0); 511 q = tp->segq.lh_first; 512 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 513 return (0); 514 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 515 return (0); 516 517 tp->rcv_nxt += q->ipqe_len; 518 pkt_flags = q->ipqe_flags & TH_FIN; 519 ND6_HINT(tp); 520 521 LIST_REMOVE(q, ipqe_q); 522 LIST_REMOVE(q, ipqe_timeq); 523 if (so->so_state & SS_CANTRCVMORE) 524 m_freem(q->ipqe_m); 525 else 526 sbappend(&so->so_rcv, q->ipqe_m); 527 pool_put(&ipqent_pool, q); 528 sorwakeup(so); 529 return (pkt_flags); 530 } 531 532 #ifdef INET6 533 int 534 tcp6_input(mp, offp, proto) 535 struct mbuf **mp; 536 int *offp, proto; 537 { 538 struct mbuf *m = *mp; 539 540 /* 541 * draft-itojun-ipv6-tcp-to-anycast 542 * better place to put this in? 543 */ 544 if (m->m_flags & M_ANYCAST6) { 545 struct ip6_hdr *ip6; 546 if (m->m_len < sizeof(struct ip6_hdr)) { 547 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 548 tcpstat.tcps_rcvshort++; 549 return IPPROTO_DONE; 550 } 551 } 552 ip6 = mtod(m, struct ip6_hdr *); 553 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 554 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 555 return IPPROTO_DONE; 556 } 557 558 tcp_input(m, *offp, proto); 559 return IPPROTO_DONE; 560 } 561 #endif 562 563 /* 564 * TCP input routine, follows pages 65-76 of the 565 * protocol specification dated September, 1981 very closely. 566 */ 567 void 568 #if __STDC__ 569 tcp_input(struct mbuf *m, ...) 570 #else 571 tcp_input(m, va_alist) 572 struct mbuf *m; 573 #endif 574 { 575 int proto; 576 struct tcphdr *th; 577 struct ip *ip; 578 struct inpcb *inp; 579 #ifdef INET6 580 struct ip6_hdr *ip6; 581 struct in6pcb *in6p; 582 #endif 583 caddr_t optp = NULL; 584 int optlen = 0; 585 int len, tlen, toff, hdroptlen = 0; 586 struct tcpcb *tp = 0; 587 int tiflags; 588 struct socket *so = NULL; 589 int todrop, acked, ourfinisacked, needoutput = 0; 590 short ostate = 0; 591 int iss = 0; 592 u_long tiwin; 593 struct tcp_opt_info opti; 594 int off, iphlen; 595 va_list ap; 596 int af; /* af on the wire */ 597 struct mbuf *tcp_saveti = NULL; 598 599 va_start(ap, m); 600 toff = va_arg(ap, int); 601 proto = va_arg(ap, int); 602 va_end(ap); 603 604 tcpstat.tcps_rcvtotal++; 605 606 bzero(&opti, sizeof(opti)); 607 opti.ts_present = 0; 608 opti.maxseg = 0; 609 610 /* 611 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 612 * 613 * TCP is, by definition, unicast, so we reject all 614 * multicast outright. 615 * 616 * Note, there are additional src/dst address checks in 617 * the AF-specific code below. 618 */ 619 if (m->m_flags & (M_BCAST|M_MCAST)) { 620 /* XXX stat */ 621 goto drop; 622 } 623 #ifdef INET6 624 if (m->m_flags & M_ANYCAST6) { 625 /* XXX stat */ 626 goto drop; 627 } 628 #endif 629 630 /* 631 * Get IP and TCP header together in first mbuf. 632 * Note: IP leaves IP header in first mbuf. 633 */ 634 ip = mtod(m, struct ip *); 635 #ifdef INET6 636 ip6 = NULL; 637 #endif 638 switch (ip->ip_v) { 639 #ifdef INET 640 case 4: 641 af = AF_INET; 642 iphlen = sizeof(struct ip); 643 #ifndef PULLDOWN_TEST 644 /* would like to get rid of this... */ 645 if (toff > sizeof (struct ip)) { 646 ip_stripoptions(m, (struct mbuf *)0); 647 toff = sizeof(struct ip); 648 } 649 if (m->m_len < toff + sizeof (struct tcphdr)) { 650 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 651 tcpstat.tcps_rcvshort++; 652 return; 653 } 654 } 655 ip = mtod(m, struct ip *); 656 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 657 #else 658 ip = mtod(m, struct ip *); 659 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 660 sizeof(struct tcphdr)); 661 if (th == NULL) { 662 tcpstat.tcps_rcvshort++; 663 return; 664 } 665 #endif 666 667 /* 668 * Make sure destination address is not multicast. 669 * Source address checked in ip_input(). 670 */ 671 if (IN_MULTICAST(ip->ip_dst.s_addr)) { 672 /* XXX stat */ 673 goto drop; 674 } 675 676 /* We do the checksum after PCB lookup... */ 677 len = ip->ip_len; 678 tlen = len - toff; 679 break; 680 #endif 681 #ifdef INET6 682 case 6: 683 ip = NULL; 684 iphlen = sizeof(struct ip6_hdr); 685 af = AF_INET6; 686 #ifndef PULLDOWN_TEST 687 if (m->m_len < toff + sizeof(struct tcphdr)) { 688 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 689 if (m == NULL) { 690 tcpstat.tcps_rcvshort++; 691 return; 692 } 693 } 694 ip6 = mtod(m, struct ip6_hdr *); 695 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 696 #else 697 ip6 = mtod(m, struct ip6_hdr *); 698 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 699 sizeof(struct tcphdr)); 700 if (th == NULL) { 701 tcpstat.tcps_rcvshort++; 702 return; 703 } 704 #endif 705 706 /* Be proactive about malicious use of IPv4 mapped address */ 707 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 708 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 709 /* XXX stat */ 710 goto drop; 711 } 712 713 /* 714 * Be proactive about unspecified IPv6 address in source. 715 * As we use all-zero to indicate unbounded/unconnected pcb, 716 * unspecified IPv6 address can be used to confuse us. 717 * 718 * Note that packets with unspecified IPv6 destination is 719 * already dropped in ip6_input. 720 */ 721 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 722 /* XXX stat */ 723 goto drop; 724 } 725 726 /* 727 * Make sure destination address is not multicast. 728 * Source address checked in ip6_input(). 729 */ 730 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 731 /* XXX stat */ 732 goto drop; 733 } 734 735 /* We do the checksum after PCB lookup... */ 736 len = m->m_pkthdr.len; 737 tlen = len - toff; 738 break; 739 #endif 740 default: 741 m_freem(m); 742 return; 743 } 744 745 /* 746 * Check that TCP offset makes sense, 747 * pull out TCP options and adjust length. XXX 748 */ 749 off = th->th_off << 2; 750 if (off < sizeof (struct tcphdr) || off > tlen) { 751 tcpstat.tcps_rcvbadoff++; 752 goto drop; 753 } 754 tlen -= off; 755 756 /* 757 * tcp_input() has been modified to use tlen to mean the TCP data 758 * length throughout the function. Other functions can use 759 * m->m_pkthdr.len as the basis for calculating the TCP data length. 760 * rja 761 */ 762 763 if (off > sizeof (struct tcphdr)) { 764 #ifndef PULLDOWN_TEST 765 if (m->m_len < toff + off) { 766 if ((m = m_pullup(m, toff + off)) == 0) { 767 tcpstat.tcps_rcvshort++; 768 return; 769 } 770 switch (af) { 771 #ifdef INET 772 case AF_INET: 773 ip = mtod(m, struct ip *); 774 break; 775 #endif 776 #ifdef INET6 777 case AF_INET6: 778 ip6 = mtod(m, struct ip6_hdr *); 779 break; 780 #endif 781 } 782 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 783 } 784 #else 785 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 786 if (th == NULL) { 787 tcpstat.tcps_rcvshort++; 788 return; 789 } 790 /* 791 * NOTE: ip/ip6 will not be affected by m_pulldown() 792 * (as they're before toff) and we don't need to update those. 793 */ 794 #endif 795 optlen = off - sizeof (struct tcphdr); 796 optp = ((caddr_t)th) + sizeof(struct tcphdr); 797 /* 798 * Do quick retrieval of timestamp options ("options 799 * prediction?"). If timestamp is the only option and it's 800 * formatted as recommended in RFC 1323 appendix A, we 801 * quickly get the values now and not bother calling 802 * tcp_dooptions(), etc. 803 */ 804 if ((optlen == TCPOLEN_TSTAMP_APPA || 805 (optlen > TCPOLEN_TSTAMP_APPA && 806 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 807 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 808 (th->th_flags & TH_SYN) == 0) { 809 opti.ts_present = 1; 810 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 811 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 812 optp = NULL; /* we've parsed the options */ 813 } 814 } 815 tiflags = th->th_flags; 816 817 /* 818 * Locate pcb for segment. 819 */ 820 findpcb: 821 inp = NULL; 822 #ifdef INET6 823 in6p = NULL; 824 #endif 825 switch (af) { 826 #ifdef INET 827 case AF_INET: 828 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 829 ip->ip_dst, th->th_dport); 830 if (inp == 0) { 831 ++tcpstat.tcps_pcbhashmiss; 832 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 833 } 834 #ifdef INET6 835 if (inp == 0) { 836 struct in6_addr s, d; 837 838 /* mapped addr case */ 839 bzero(&s, sizeof(s)); 840 s.s6_addr16[5] = htons(0xffff); 841 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 842 bzero(&d, sizeof(d)); 843 d.s6_addr16[5] = htons(0xffff); 844 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 845 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 846 &d, th->th_dport, 0); 847 if (in6p == 0) { 848 ++tcpstat.tcps_pcbhashmiss; 849 in6p = in6_pcblookup_bind(&tcb6, &d, 850 th->th_dport, 0); 851 } 852 } 853 #endif 854 #ifndef INET6 855 if (inp == 0) 856 #else 857 if (inp == 0 && in6p == 0) 858 #endif 859 { 860 ++tcpstat.tcps_noport; 861 if (tcp_log_refused && (tiflags & TH_SYN)) { 862 #ifndef INET6 863 char src[4*sizeof "123"]; 864 char dst[4*sizeof "123"]; 865 #else 866 char src[INET6_ADDRSTRLEN]; 867 char dst[INET6_ADDRSTRLEN]; 868 #endif 869 if (ip) { 870 strcpy(src, inet_ntoa(ip->ip_src)); 871 strcpy(dst, inet_ntoa(ip->ip_dst)); 872 } 873 #ifdef INET6 874 else if (ip6) { 875 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 876 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 877 } 878 #endif 879 else { 880 strcpy(src, "(unknown)"); 881 strcpy(dst, "(unknown)"); 882 } 883 log(LOG_INFO, 884 "Connection attempt to TCP %s:%d from %s:%d\n", 885 dst, ntohs(th->th_dport), 886 src, ntohs(th->th_sport)); 887 } 888 TCP_FIELDS_TO_HOST(th); 889 goto dropwithreset_ratelim; 890 } 891 #ifdef IPSEC 892 if (inp && ipsec4_in_reject(m, inp)) { 893 ipsecstat.in_polvio++; 894 goto drop; 895 } 896 #ifdef INET6 897 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 898 ipsecstat.in_polvio++; 899 goto drop; 900 } 901 #endif 902 #endif /*IPSEC*/ 903 break; 904 #endif /*INET*/ 905 #ifdef INET6 906 case AF_INET6: 907 { 908 int faith; 909 910 #if defined(NFAITH) && NFAITH > 0 911 faith = faithprefix(&ip6->ip6_dst); 912 #else 913 faith = 0; 914 #endif 915 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 916 &ip6->ip6_dst, th->th_dport, faith); 917 if (in6p == NULL) { 918 ++tcpstat.tcps_pcbhashmiss; 919 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 920 th->th_dport, faith); 921 } 922 if (in6p == NULL) { 923 ++tcpstat.tcps_noport; 924 TCP_FIELDS_TO_HOST(th); 925 goto dropwithreset_ratelim; 926 } 927 #ifdef IPSEC 928 if (ipsec6_in_reject(m, in6p)) { 929 ipsec6stat.in_polvio++; 930 goto drop; 931 } 932 #endif /*IPSEC*/ 933 break; 934 } 935 #endif 936 } 937 938 /* 939 * If the state is CLOSED (i.e., TCB does not exist) then 940 * all data in the incoming segment is discarded. 941 * If the TCB exists but is in CLOSED state, it is embryonic, 942 * but should either do a listen or a connect soon. 943 */ 944 tp = NULL; 945 so = NULL; 946 if (inp) { 947 tp = intotcpcb(inp); 948 so = inp->inp_socket; 949 } 950 #ifdef INET6 951 else if (in6p) { 952 tp = in6totcpcb(in6p); 953 so = in6p->in6p_socket; 954 } 955 #endif 956 if (tp == 0) { 957 TCP_FIELDS_TO_HOST(th); 958 goto dropwithreset_ratelim; 959 } 960 if (tp->t_state == TCPS_CLOSED) 961 goto drop; 962 963 /* 964 * Checksum extended TCP header and data. 965 */ 966 switch (af) { 967 #ifdef INET 968 case AF_INET: 969 switch (m->m_pkthdr.csum_flags & 970 ((m->m_pkthdr.rcvif->if_csum_flags & M_CSUM_TCPv4) | 971 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 972 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 973 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 974 goto badcsum; 975 976 case M_CSUM_TCPv4|M_CSUM_DATA: 977 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 978 if ((m->m_pkthdr.csum_data ^ 0xffff) != 0) 979 goto badcsum; 980 break; 981 982 case M_CSUM_TCPv4: 983 /* Checksum was okay. */ 984 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 985 break; 986 987 default: 988 /* Must compute it ourselves. */ 989 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 990 #ifndef PULLDOWN_TEST 991 { 992 struct ipovly *ipov; 993 ipov = (struct ipovly *)ip; 994 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 995 ipov->ih_len = htons(tlen + off); 996 997 if (in_cksum(m, len) != 0) 998 goto badcsum; 999 } 1000 #else 1001 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1002 goto badcsum; 1003 #endif /* ! PULLDOWN_TEST */ 1004 break; 1005 } 1006 break; 1007 #endif /* INET4 */ 1008 1009 #ifdef INET6 1010 case AF_INET6: 1011 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1012 goto badcsum; 1013 break; 1014 #endif /* INET6 */ 1015 } 1016 1017 TCP_FIELDS_TO_HOST(th); 1018 1019 /* Unscale the window into a 32-bit value. */ 1020 if ((tiflags & TH_SYN) == 0) 1021 tiwin = th->th_win << tp->snd_scale; 1022 else 1023 tiwin = th->th_win; 1024 1025 #ifdef INET6 1026 /* save packet options if user wanted */ 1027 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1028 if (in6p->in6p_options) { 1029 m_freem(in6p->in6p_options); 1030 in6p->in6p_options = 0; 1031 } 1032 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1033 } 1034 #endif 1035 1036 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1037 union syn_cache_sa src; 1038 union syn_cache_sa dst; 1039 1040 bzero(&src, sizeof(src)); 1041 bzero(&dst, sizeof(dst)); 1042 switch (af) { 1043 #ifdef INET 1044 case AF_INET: 1045 src.sin.sin_len = sizeof(struct sockaddr_in); 1046 src.sin.sin_family = AF_INET; 1047 src.sin.sin_addr = ip->ip_src; 1048 src.sin.sin_port = th->th_sport; 1049 1050 dst.sin.sin_len = sizeof(struct sockaddr_in); 1051 dst.sin.sin_family = AF_INET; 1052 dst.sin.sin_addr = ip->ip_dst; 1053 dst.sin.sin_port = th->th_dport; 1054 break; 1055 #endif 1056 #ifdef INET6 1057 case AF_INET6: 1058 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1059 src.sin6.sin6_family = AF_INET6; 1060 src.sin6.sin6_addr = ip6->ip6_src; 1061 src.sin6.sin6_port = th->th_sport; 1062 1063 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1064 dst.sin6.sin6_family = AF_INET6; 1065 dst.sin6.sin6_addr = ip6->ip6_dst; 1066 dst.sin6.sin6_port = th->th_dport; 1067 break; 1068 #endif /* INET6 */ 1069 default: 1070 goto badsyn; /*sanity*/ 1071 } 1072 1073 if (so->so_options & SO_DEBUG) { 1074 ostate = tp->t_state; 1075 1076 tcp_saveti = NULL; 1077 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1078 goto nosave; 1079 1080 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1081 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1082 if (!tcp_saveti) 1083 goto nosave; 1084 } else { 1085 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1086 if (!tcp_saveti) 1087 goto nosave; 1088 tcp_saveti->m_len = iphlen; 1089 m_copydata(m, 0, iphlen, 1090 mtod(tcp_saveti, caddr_t)); 1091 } 1092 1093 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1094 m_freem(tcp_saveti); 1095 tcp_saveti = NULL; 1096 } else { 1097 tcp_saveti->m_len += sizeof(struct tcphdr); 1098 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1099 sizeof(struct tcphdr)); 1100 } 1101 if (tcp_saveti) { 1102 /* 1103 * need to recover version # field, which was 1104 * overwritten on ip_cksum computation. 1105 */ 1106 struct ip *sip; 1107 sip = mtod(tcp_saveti, struct ip *); 1108 switch (af) { 1109 #ifdef INET 1110 case AF_INET: 1111 sip->ip_v = 4; 1112 break; 1113 #endif 1114 #ifdef INET6 1115 case AF_INET6: 1116 sip->ip_v = 6; 1117 break; 1118 #endif 1119 } 1120 } 1121 nosave:; 1122 } 1123 if (so->so_options & SO_ACCEPTCONN) { 1124 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1125 if (tiflags & TH_RST) { 1126 syn_cache_reset(&src.sa, &dst.sa, th); 1127 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1128 (TH_ACK|TH_SYN)) { 1129 /* 1130 * Received a SYN,ACK. This should 1131 * never happen while we are in 1132 * LISTEN. Send an RST. 1133 */ 1134 goto badsyn; 1135 } else if (tiflags & TH_ACK) { 1136 so = syn_cache_get(&src.sa, &dst.sa, 1137 th, toff, tlen, so, m); 1138 if (so == NULL) { 1139 /* 1140 * We don't have a SYN for 1141 * this ACK; send an RST. 1142 */ 1143 goto badsyn; 1144 } else if (so == 1145 (struct socket *)(-1)) { 1146 /* 1147 * We were unable to create 1148 * the connection. If the 1149 * 3-way handshake was 1150 * completed, and RST has 1151 * been sent to the peer. 1152 * Since the mbuf might be 1153 * in use for the reply, 1154 * do not free it. 1155 */ 1156 m = NULL; 1157 } else { 1158 /* 1159 * We have created a 1160 * full-blown connection. 1161 */ 1162 tp = NULL; 1163 inp = NULL; 1164 #ifdef INET6 1165 in6p = NULL; 1166 #endif 1167 switch (so->so_proto->pr_domain->dom_family) { 1168 #ifdef INET 1169 case AF_INET: 1170 inp = sotoinpcb(so); 1171 tp = intotcpcb(inp); 1172 break; 1173 #endif 1174 #ifdef INET6 1175 case AF_INET6: 1176 in6p = sotoin6pcb(so); 1177 tp = in6totcpcb(in6p); 1178 break; 1179 #endif 1180 } 1181 if (tp == NULL) 1182 goto badsyn; /*XXX*/ 1183 tiwin <<= tp->snd_scale; 1184 goto after_listen; 1185 } 1186 } else { 1187 /* 1188 * None of RST, SYN or ACK was set. 1189 * This is an invalid packet for a 1190 * TCB in LISTEN state. Send a RST. 1191 */ 1192 goto badsyn; 1193 } 1194 } else { 1195 /* 1196 * Received a SYN. 1197 */ 1198 1199 /* 1200 * LISTEN socket received a SYN 1201 * from itself? This can't possibly 1202 * be valid; drop the packet. 1203 */ 1204 if (th->th_sport == th->th_dport) { 1205 int i; 1206 1207 switch (af) { 1208 #ifdef INET 1209 case AF_INET: 1210 i = in_hosteq(ip->ip_src, ip->ip_dst); 1211 break; 1212 #endif 1213 #ifdef INET6 1214 case AF_INET6: 1215 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1216 break; 1217 #endif 1218 default: 1219 i = 1; 1220 } 1221 if (i) { 1222 tcpstat.tcps_badsyn++; 1223 goto drop; 1224 } 1225 } 1226 1227 /* 1228 * SYN looks ok; create compressed TCP 1229 * state for it. 1230 */ 1231 if (so->so_qlen <= so->so_qlimit && 1232 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1233 so, m, optp, optlen, &opti)) 1234 m = NULL; 1235 } 1236 goto drop; 1237 } 1238 } 1239 1240 after_listen: 1241 #ifdef DIAGNOSTIC 1242 /* 1243 * Should not happen now that all embryonic connections 1244 * are handled with compressed state. 1245 */ 1246 if (tp->t_state == TCPS_LISTEN) 1247 panic("tcp_input: TCPS_LISTEN"); 1248 #endif 1249 1250 /* 1251 * Segment received on connection. 1252 * Reset idle time and keep-alive timer. 1253 */ 1254 tp->t_idle = 0; 1255 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1256 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1257 1258 /* 1259 * Process options. 1260 */ 1261 if (optp) 1262 tcp_dooptions(tp, optp, optlen, th, &opti); 1263 1264 /* 1265 * Header prediction: check for the two common cases 1266 * of a uni-directional data xfer. If the packet has 1267 * no control flags, is in-sequence, the window didn't 1268 * change and we're not retransmitting, it's a 1269 * candidate. If the length is zero and the ack moved 1270 * forward, we're the sender side of the xfer. Just 1271 * free the data acked & wake any higher level process 1272 * that was blocked waiting for space. If the length 1273 * is non-zero and the ack didn't move, we're the 1274 * receiver side. If we're getting packets in-order 1275 * (the reassembly queue is empty), add the data to 1276 * the socket buffer and note that we need a delayed ack. 1277 */ 1278 if (tp->t_state == TCPS_ESTABLISHED && 1279 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1280 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1281 th->th_seq == tp->rcv_nxt && 1282 tiwin && tiwin == tp->snd_wnd && 1283 tp->snd_nxt == tp->snd_max) { 1284 1285 /* 1286 * If last ACK falls within this segment's sequence numbers, 1287 * record the timestamp. 1288 */ 1289 if (opti.ts_present && 1290 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1291 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1292 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1293 tp->ts_recent = opti.ts_val; 1294 } 1295 1296 if (tlen == 0) { 1297 if (SEQ_GT(th->th_ack, tp->snd_una) && 1298 SEQ_LEQ(th->th_ack, tp->snd_max) && 1299 tp->snd_cwnd >= tp->snd_wnd && 1300 tp->t_dupacks < tcprexmtthresh) { 1301 /* 1302 * this is a pure ack for outstanding data. 1303 */ 1304 ++tcpstat.tcps_predack; 1305 if (opti.ts_present && opti.ts_ecr) 1306 tcp_xmit_timer(tp, 1307 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1308 else if (tp->t_rtt && 1309 SEQ_GT(th->th_ack, tp->t_rtseq)) 1310 tcp_xmit_timer(tp, tp->t_rtt); 1311 acked = th->th_ack - tp->snd_una; 1312 tcpstat.tcps_rcvackpack++; 1313 tcpstat.tcps_rcvackbyte += acked; 1314 ND6_HINT(tp); 1315 sbdrop(&so->so_snd, acked); 1316 /* 1317 * We want snd_recover to track snd_una to 1318 * avoid sequence wraparound problems for 1319 * very large transfers. 1320 */ 1321 tp->snd_una = tp->snd_recover = th->th_ack; 1322 m_freem(m); 1323 1324 /* 1325 * If all outstanding data are acked, stop 1326 * retransmit timer, otherwise restart timer 1327 * using current (possibly backed-off) value. 1328 * If process is waiting for space, 1329 * wakeup/selwakeup/signal. If data 1330 * are ready to send, let tcp_output 1331 * decide between more output or persist. 1332 */ 1333 if (tp->snd_una == tp->snd_max) 1334 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1335 else if (TCP_TIMER_ISARMED(tp, 1336 TCPT_PERSIST) == 0) 1337 TCP_TIMER_ARM(tp, TCPT_REXMT, 1338 tp->t_rxtcur); 1339 1340 sowwakeup(so); 1341 if (so->so_snd.sb_cc) 1342 (void) tcp_output(tp); 1343 if (tcp_saveti) 1344 m_freem(tcp_saveti); 1345 return; 1346 } 1347 } else if (th->th_ack == tp->snd_una && 1348 tp->segq.lh_first == NULL && 1349 tlen <= sbspace(&so->so_rcv)) { 1350 /* 1351 * this is a pure, in-sequence data packet 1352 * with nothing on the reassembly queue and 1353 * we have enough buffer space to take it. 1354 */ 1355 ++tcpstat.tcps_preddat; 1356 tp->rcv_nxt += tlen; 1357 tcpstat.tcps_rcvpack++; 1358 tcpstat.tcps_rcvbyte += tlen; 1359 ND6_HINT(tp); 1360 /* 1361 * Drop TCP, IP headers and TCP options then add data 1362 * to socket buffer. 1363 */ 1364 m_adj(m, toff + off); 1365 sbappend(&so->so_rcv, m); 1366 sorwakeup(so); 1367 TCP_SETUP_ACK(tp, th); 1368 if (tp->t_flags & TF_ACKNOW) 1369 (void) tcp_output(tp); 1370 if (tcp_saveti) 1371 m_freem(tcp_saveti); 1372 return; 1373 } 1374 } 1375 1376 /* 1377 * Compute mbuf offset to TCP data segment. 1378 */ 1379 hdroptlen = toff + off; 1380 1381 /* 1382 * Calculate amount of space in receive window, 1383 * and then do TCP input processing. 1384 * Receive window is amount of space in rcv queue, 1385 * but not less than advertised window. 1386 */ 1387 { int win; 1388 1389 win = sbspace(&so->so_rcv); 1390 if (win < 0) 1391 win = 0; 1392 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1393 } 1394 1395 switch (tp->t_state) { 1396 1397 /* 1398 * If the state is SYN_SENT: 1399 * if seg contains an ACK, but not for our SYN, drop the input. 1400 * if seg contains a RST, then drop the connection. 1401 * if seg does not contain SYN, then drop it. 1402 * Otherwise this is an acceptable SYN segment 1403 * initialize tp->rcv_nxt and tp->irs 1404 * if seg contains ack then advance tp->snd_una 1405 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1406 * arrange for segment to be acked (eventually) 1407 * continue processing rest of data/controls, beginning with URG 1408 */ 1409 case TCPS_SYN_SENT: 1410 if ((tiflags & TH_ACK) && 1411 (SEQ_LEQ(th->th_ack, tp->iss) || 1412 SEQ_GT(th->th_ack, tp->snd_max))) 1413 goto dropwithreset; 1414 if (tiflags & TH_RST) { 1415 if (tiflags & TH_ACK) 1416 tp = tcp_drop(tp, ECONNREFUSED); 1417 goto drop; 1418 } 1419 if ((tiflags & TH_SYN) == 0) 1420 goto drop; 1421 if (tiflags & TH_ACK) { 1422 tp->snd_una = tp->snd_recover = th->th_ack; 1423 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1424 tp->snd_nxt = tp->snd_una; 1425 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1426 } 1427 tp->irs = th->th_seq; 1428 tcp_rcvseqinit(tp); 1429 tp->t_flags |= TF_ACKNOW; 1430 tcp_mss_from_peer(tp, opti.maxseg); 1431 1432 /* 1433 * Initialize the initial congestion window. If we 1434 * had to retransmit the SYN, we must initialize cwnd 1435 * to 1 segment (i.e. the Loss Window). 1436 */ 1437 if (tp->t_flags & TF_SYN_REXMT) 1438 tp->snd_cwnd = tp->t_peermss; 1439 else 1440 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1441 tp->t_peermss); 1442 1443 tcp_rmx_rtt(tp); 1444 if (tiflags & TH_ACK) { 1445 tcpstat.tcps_connects++; 1446 soisconnected(so); 1447 tcp_established(tp); 1448 /* Do window scaling on this connection? */ 1449 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1450 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1451 tp->snd_scale = tp->requested_s_scale; 1452 tp->rcv_scale = tp->request_r_scale; 1453 } 1454 TCP_REASS_LOCK(tp); 1455 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1456 TCP_REASS_UNLOCK(tp); 1457 /* 1458 * if we didn't have to retransmit the SYN, 1459 * use its rtt as our initial srtt & rtt var. 1460 */ 1461 if (tp->t_rtt) 1462 tcp_xmit_timer(tp, tp->t_rtt); 1463 } else 1464 tp->t_state = TCPS_SYN_RECEIVED; 1465 1466 /* 1467 * Advance th->th_seq to correspond to first data byte. 1468 * If data, trim to stay within window, 1469 * dropping FIN if necessary. 1470 */ 1471 th->th_seq++; 1472 if (tlen > tp->rcv_wnd) { 1473 todrop = tlen - tp->rcv_wnd; 1474 m_adj(m, -todrop); 1475 tlen = tp->rcv_wnd; 1476 tiflags &= ~TH_FIN; 1477 tcpstat.tcps_rcvpackafterwin++; 1478 tcpstat.tcps_rcvbyteafterwin += todrop; 1479 } 1480 tp->snd_wl1 = th->th_seq - 1; 1481 tp->rcv_up = th->th_seq; 1482 goto step6; 1483 1484 /* 1485 * If the state is SYN_RECEIVED: 1486 * If seg contains an ACK, but not for our SYN, drop the input 1487 * and generate an RST. See page 36, rfc793 1488 */ 1489 case TCPS_SYN_RECEIVED: 1490 if ((tiflags & TH_ACK) && 1491 (SEQ_LEQ(th->th_ack, tp->iss) || 1492 SEQ_GT(th->th_ack, tp->snd_max))) 1493 goto dropwithreset; 1494 break; 1495 } 1496 1497 /* 1498 * States other than LISTEN or SYN_SENT. 1499 * First check timestamp, if present. 1500 * Then check that at least some bytes of segment are within 1501 * receive window. If segment begins before rcv_nxt, 1502 * drop leading data (and SYN); if nothing left, just ack. 1503 * 1504 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1505 * and it's less than ts_recent, drop it. 1506 */ 1507 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1508 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1509 1510 /* Check to see if ts_recent is over 24 days old. */ 1511 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1512 TCP_PAWS_IDLE) { 1513 /* 1514 * Invalidate ts_recent. If this segment updates 1515 * ts_recent, the age will be reset later and ts_recent 1516 * will get a valid value. If it does not, setting 1517 * ts_recent to zero will at least satisfy the 1518 * requirement that zero be placed in the timestamp 1519 * echo reply when ts_recent isn't valid. The 1520 * age isn't reset until we get a valid ts_recent 1521 * because we don't want out-of-order segments to be 1522 * dropped when ts_recent is old. 1523 */ 1524 tp->ts_recent = 0; 1525 } else { 1526 tcpstat.tcps_rcvduppack++; 1527 tcpstat.tcps_rcvdupbyte += tlen; 1528 tcpstat.tcps_pawsdrop++; 1529 goto dropafterack; 1530 } 1531 } 1532 1533 todrop = tp->rcv_nxt - th->th_seq; 1534 if (todrop > 0) { 1535 if (tiflags & TH_SYN) { 1536 tiflags &= ~TH_SYN; 1537 th->th_seq++; 1538 if (th->th_urp > 1) 1539 th->th_urp--; 1540 else { 1541 tiflags &= ~TH_URG; 1542 th->th_urp = 0; 1543 } 1544 todrop--; 1545 } 1546 if (todrop > tlen || 1547 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1548 /* 1549 * Any valid FIN must be to the left of the window. 1550 * At this point the FIN must be a duplicate or 1551 * out of sequence; drop it. 1552 */ 1553 tiflags &= ~TH_FIN; 1554 /* 1555 * Send an ACK to resynchronize and drop any data. 1556 * But keep on processing for RST or ACK. 1557 */ 1558 tp->t_flags |= TF_ACKNOW; 1559 todrop = tlen; 1560 tcpstat.tcps_rcvdupbyte += todrop; 1561 tcpstat.tcps_rcvduppack++; 1562 } else { 1563 tcpstat.tcps_rcvpartduppack++; 1564 tcpstat.tcps_rcvpartdupbyte += todrop; 1565 } 1566 hdroptlen += todrop; /*drop from head afterwards*/ 1567 th->th_seq += todrop; 1568 tlen -= todrop; 1569 if (th->th_urp > todrop) 1570 th->th_urp -= todrop; 1571 else { 1572 tiflags &= ~TH_URG; 1573 th->th_urp = 0; 1574 } 1575 } 1576 1577 /* 1578 * If new data are received on a connection after the 1579 * user processes are gone, then RST the other end. 1580 */ 1581 if ((so->so_state & SS_NOFDREF) && 1582 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1583 tp = tcp_close(tp); 1584 tcpstat.tcps_rcvafterclose++; 1585 goto dropwithreset; 1586 } 1587 1588 /* 1589 * If segment ends after window, drop trailing data 1590 * (and PUSH and FIN); if nothing left, just ACK. 1591 */ 1592 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1593 if (todrop > 0) { 1594 tcpstat.tcps_rcvpackafterwin++; 1595 if (todrop >= tlen) { 1596 tcpstat.tcps_rcvbyteafterwin += tlen; 1597 /* 1598 * If a new connection request is received 1599 * while in TIME_WAIT, drop the old connection 1600 * and start over if the sequence numbers 1601 * are above the previous ones. 1602 */ 1603 if (tiflags & TH_SYN && 1604 tp->t_state == TCPS_TIME_WAIT && 1605 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1606 iss = tcp_new_iss(tp, tp->snd_nxt); 1607 tp = tcp_close(tp); 1608 goto findpcb; 1609 } 1610 /* 1611 * If window is closed can only take segments at 1612 * window edge, and have to drop data and PUSH from 1613 * incoming segments. Continue processing, but 1614 * remember to ack. Otherwise, drop segment 1615 * and ack. 1616 */ 1617 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1618 tp->t_flags |= TF_ACKNOW; 1619 tcpstat.tcps_rcvwinprobe++; 1620 } else 1621 goto dropafterack; 1622 } else 1623 tcpstat.tcps_rcvbyteafterwin += todrop; 1624 m_adj(m, -todrop); 1625 tlen -= todrop; 1626 tiflags &= ~(TH_PUSH|TH_FIN); 1627 } 1628 1629 /* 1630 * If last ACK falls within this segment's sequence numbers, 1631 * and the timestamp is newer, record it. 1632 */ 1633 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1634 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1635 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1636 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1637 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1638 tp->ts_recent = opti.ts_val; 1639 } 1640 1641 /* 1642 * If the RST bit is set examine the state: 1643 * SYN_RECEIVED STATE: 1644 * If passive open, return to LISTEN state. 1645 * If active open, inform user that connection was refused. 1646 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1647 * Inform user that connection was reset, and close tcb. 1648 * CLOSING, LAST_ACK, TIME_WAIT STATES 1649 * Close the tcb. 1650 */ 1651 if (tiflags&TH_RST) switch (tp->t_state) { 1652 1653 case TCPS_SYN_RECEIVED: 1654 so->so_error = ECONNREFUSED; 1655 goto close; 1656 1657 case TCPS_ESTABLISHED: 1658 case TCPS_FIN_WAIT_1: 1659 case TCPS_FIN_WAIT_2: 1660 case TCPS_CLOSE_WAIT: 1661 so->so_error = ECONNRESET; 1662 close: 1663 tp->t_state = TCPS_CLOSED; 1664 tcpstat.tcps_drops++; 1665 tp = tcp_close(tp); 1666 goto drop; 1667 1668 case TCPS_CLOSING: 1669 case TCPS_LAST_ACK: 1670 case TCPS_TIME_WAIT: 1671 tp = tcp_close(tp); 1672 goto drop; 1673 } 1674 1675 /* 1676 * If a SYN is in the window, then this is an 1677 * error and we send an RST and drop the connection. 1678 */ 1679 if (tiflags & TH_SYN) { 1680 tp = tcp_drop(tp, ECONNRESET); 1681 goto dropwithreset; 1682 } 1683 1684 /* 1685 * If the ACK bit is off we drop the segment and return. 1686 */ 1687 if ((tiflags & TH_ACK) == 0) { 1688 if (tp->t_flags & TF_ACKNOW) 1689 goto dropafterack; 1690 else 1691 goto drop; 1692 } 1693 1694 /* 1695 * Ack processing. 1696 */ 1697 switch (tp->t_state) { 1698 1699 /* 1700 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1701 * ESTABLISHED state and continue processing, otherwise 1702 * send an RST. 1703 */ 1704 case TCPS_SYN_RECEIVED: 1705 if (SEQ_GT(tp->snd_una, th->th_ack) || 1706 SEQ_GT(th->th_ack, tp->snd_max)) 1707 goto dropwithreset; 1708 tcpstat.tcps_connects++; 1709 soisconnected(so); 1710 tcp_established(tp); 1711 /* Do window scaling? */ 1712 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1713 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1714 tp->snd_scale = tp->requested_s_scale; 1715 tp->rcv_scale = tp->request_r_scale; 1716 } 1717 TCP_REASS_LOCK(tp); 1718 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1719 TCP_REASS_UNLOCK(tp); 1720 tp->snd_wl1 = th->th_seq - 1; 1721 /* fall into ... */ 1722 1723 /* 1724 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1725 * ACKs. If the ack is in the range 1726 * tp->snd_una < th->th_ack <= tp->snd_max 1727 * then advance tp->snd_una to th->th_ack and drop 1728 * data from the retransmission queue. If this ACK reflects 1729 * more up to date window information we update our window information. 1730 */ 1731 case TCPS_ESTABLISHED: 1732 case TCPS_FIN_WAIT_1: 1733 case TCPS_FIN_WAIT_2: 1734 case TCPS_CLOSE_WAIT: 1735 case TCPS_CLOSING: 1736 case TCPS_LAST_ACK: 1737 case TCPS_TIME_WAIT: 1738 1739 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1740 if (tlen == 0 && tiwin == tp->snd_wnd) { 1741 tcpstat.tcps_rcvdupack++; 1742 /* 1743 * If we have outstanding data (other than 1744 * a window probe), this is a completely 1745 * duplicate ack (ie, window info didn't 1746 * change), the ack is the biggest we've 1747 * seen and we've seen exactly our rexmt 1748 * threshhold of them, assume a packet 1749 * has been dropped and retransmit it. 1750 * Kludge snd_nxt & the congestion 1751 * window so we send only this one 1752 * packet. 1753 * 1754 * We know we're losing at the current 1755 * window size so do congestion avoidance 1756 * (set ssthresh to half the current window 1757 * and pull our congestion window back to 1758 * the new ssthresh). 1759 * 1760 * Dup acks mean that packets have left the 1761 * network (they're now cached at the receiver) 1762 * so bump cwnd by the amount in the receiver 1763 * to keep a constant cwnd packets in the 1764 * network. 1765 */ 1766 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1767 th->th_ack != tp->snd_una) 1768 tp->t_dupacks = 0; 1769 else if (++tp->t_dupacks == tcprexmtthresh) { 1770 tcp_seq onxt = tp->snd_nxt; 1771 u_int win = 1772 min(tp->snd_wnd, tp->snd_cwnd) / 1773 2 / tp->t_segsz; 1774 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1775 tp->snd_recover)) { 1776 /* 1777 * False fast retransmit after 1778 * timeout. Do not cut window. 1779 */ 1780 tp->snd_cwnd += tp->t_segsz; 1781 tp->t_dupacks = 0; 1782 (void) tcp_output(tp); 1783 goto drop; 1784 } 1785 1786 if (win < 2) 1787 win = 2; 1788 tp->snd_ssthresh = win * tp->t_segsz; 1789 tp->snd_recover = tp->snd_max; 1790 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1791 tp->t_rtt = 0; 1792 tp->snd_nxt = th->th_ack; 1793 tp->snd_cwnd = tp->t_segsz; 1794 (void) tcp_output(tp); 1795 tp->snd_cwnd = tp->snd_ssthresh + 1796 tp->t_segsz * tp->t_dupacks; 1797 if (SEQ_GT(onxt, tp->snd_nxt)) 1798 tp->snd_nxt = onxt; 1799 goto drop; 1800 } else if (tp->t_dupacks > tcprexmtthresh) { 1801 tp->snd_cwnd += tp->t_segsz; 1802 (void) tcp_output(tp); 1803 goto drop; 1804 } 1805 } else 1806 tp->t_dupacks = 0; 1807 break; 1808 } 1809 /* 1810 * If the congestion window was inflated to account 1811 * for the other side's cached packets, retract it. 1812 */ 1813 if (tcp_do_newreno == 0) { 1814 if (tp->t_dupacks >= tcprexmtthresh && 1815 tp->snd_cwnd > tp->snd_ssthresh) 1816 tp->snd_cwnd = tp->snd_ssthresh; 1817 tp->t_dupacks = 0; 1818 } else if (tp->t_dupacks >= tcprexmtthresh && 1819 tcp_newreno(tp, th) == 0) { 1820 tp->snd_cwnd = tp->snd_ssthresh; 1821 /* 1822 * Window inflation should have left us with approx. 1823 * snd_ssthresh outstanding data. But in case we 1824 * would be inclined to send a burst, better to do 1825 * it via the slow start mechanism. 1826 */ 1827 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1828 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1829 + tp->t_segsz; 1830 tp->t_dupacks = 0; 1831 } 1832 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1833 tcpstat.tcps_rcvacktoomuch++; 1834 goto dropafterack; 1835 } 1836 acked = th->th_ack - tp->snd_una; 1837 tcpstat.tcps_rcvackpack++; 1838 tcpstat.tcps_rcvackbyte += acked; 1839 1840 /* 1841 * If we have a timestamp reply, update smoothed 1842 * round trip time. If no timestamp is present but 1843 * transmit timer is running and timed sequence 1844 * number was acked, update smoothed round trip time. 1845 * Since we now have an rtt measurement, cancel the 1846 * timer backoff (cf., Phil Karn's retransmit alg.). 1847 * Recompute the initial retransmit timer. 1848 */ 1849 if (opti.ts_present && opti.ts_ecr) 1850 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1851 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1852 tcp_xmit_timer(tp,tp->t_rtt); 1853 1854 /* 1855 * If all outstanding data is acked, stop retransmit 1856 * timer and remember to restart (more output or persist). 1857 * If there is more data to be acked, restart retransmit 1858 * timer, using current (possibly backed-off) value. 1859 */ 1860 if (th->th_ack == tp->snd_max) { 1861 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1862 needoutput = 1; 1863 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1864 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1865 /* 1866 * When new data is acked, open the congestion window. 1867 * If the window gives us less than ssthresh packets 1868 * in flight, open exponentially (segsz per packet). 1869 * Otherwise open linearly: segsz per window 1870 * (segsz^2 / cwnd per packet), plus a constant 1871 * fraction of a packet (segsz/8) to help larger windows 1872 * open quickly enough. 1873 */ 1874 { 1875 u_int cw = tp->snd_cwnd; 1876 u_int incr = tp->t_segsz; 1877 1878 if (cw > tp->snd_ssthresh) 1879 incr = incr * incr / cw; 1880 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1881 tp->snd_cwnd = min(cw + incr, 1882 TCP_MAXWIN << tp->snd_scale); 1883 } 1884 ND6_HINT(tp); 1885 if (acked > so->so_snd.sb_cc) { 1886 tp->snd_wnd -= so->so_snd.sb_cc; 1887 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1888 ourfinisacked = 1; 1889 } else { 1890 sbdrop(&so->so_snd, acked); 1891 tp->snd_wnd -= acked; 1892 ourfinisacked = 0; 1893 } 1894 sowwakeup(so); 1895 /* 1896 * We want snd_recover to track snd_una to 1897 * avoid sequence wraparound problems for 1898 * very large transfers. 1899 */ 1900 tp->snd_una = tp->snd_recover = th->th_ack; 1901 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1902 tp->snd_nxt = tp->snd_una; 1903 1904 switch (tp->t_state) { 1905 1906 /* 1907 * In FIN_WAIT_1 STATE in addition to the processing 1908 * for the ESTABLISHED state if our FIN is now acknowledged 1909 * then enter FIN_WAIT_2. 1910 */ 1911 case TCPS_FIN_WAIT_1: 1912 if (ourfinisacked) { 1913 /* 1914 * If we can't receive any more 1915 * data, then closing user can proceed. 1916 * Starting the timer is contrary to the 1917 * specification, but if we don't get a FIN 1918 * we'll hang forever. 1919 */ 1920 if (so->so_state & SS_CANTRCVMORE) { 1921 soisdisconnected(so); 1922 if (tcp_maxidle > 0) 1923 TCP_TIMER_ARM(tp, TCPT_2MSL, 1924 tcp_maxidle); 1925 } 1926 tp->t_state = TCPS_FIN_WAIT_2; 1927 } 1928 break; 1929 1930 /* 1931 * In CLOSING STATE in addition to the processing for 1932 * the ESTABLISHED state if the ACK acknowledges our FIN 1933 * then enter the TIME-WAIT state, otherwise ignore 1934 * the segment. 1935 */ 1936 case TCPS_CLOSING: 1937 if (ourfinisacked) { 1938 tp->t_state = TCPS_TIME_WAIT; 1939 tcp_canceltimers(tp); 1940 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1941 soisdisconnected(so); 1942 } 1943 break; 1944 1945 /* 1946 * In LAST_ACK, we may still be waiting for data to drain 1947 * and/or to be acked, as well as for the ack of our FIN. 1948 * If our FIN is now acknowledged, delete the TCB, 1949 * enter the closed state and return. 1950 */ 1951 case TCPS_LAST_ACK: 1952 if (ourfinisacked) { 1953 tp = tcp_close(tp); 1954 goto drop; 1955 } 1956 break; 1957 1958 /* 1959 * In TIME_WAIT state the only thing that should arrive 1960 * is a retransmission of the remote FIN. Acknowledge 1961 * it and restart the finack timer. 1962 */ 1963 case TCPS_TIME_WAIT: 1964 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1965 goto dropafterack; 1966 } 1967 } 1968 1969 step6: 1970 /* 1971 * Update window information. 1972 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1973 */ 1974 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1975 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1976 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1977 /* keep track of pure window updates */ 1978 if (tlen == 0 && 1979 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1980 tcpstat.tcps_rcvwinupd++; 1981 tp->snd_wnd = tiwin; 1982 tp->snd_wl1 = th->th_seq; 1983 tp->snd_wl2 = th->th_ack; 1984 if (tp->snd_wnd > tp->max_sndwnd) 1985 tp->max_sndwnd = tp->snd_wnd; 1986 needoutput = 1; 1987 } 1988 1989 /* 1990 * Process segments with URG. 1991 */ 1992 if ((tiflags & TH_URG) && th->th_urp && 1993 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1994 /* 1995 * This is a kludge, but if we receive and accept 1996 * random urgent pointers, we'll crash in 1997 * soreceive. It's hard to imagine someone 1998 * actually wanting to send this much urgent data. 1999 */ 2000 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2001 th->th_urp = 0; /* XXX */ 2002 tiflags &= ~TH_URG; /* XXX */ 2003 goto dodata; /* XXX */ 2004 } 2005 /* 2006 * If this segment advances the known urgent pointer, 2007 * then mark the data stream. This should not happen 2008 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2009 * a FIN has been received from the remote side. 2010 * In these states we ignore the URG. 2011 * 2012 * According to RFC961 (Assigned Protocols), 2013 * the urgent pointer points to the last octet 2014 * of urgent data. We continue, however, 2015 * to consider it to indicate the first octet 2016 * of data past the urgent section as the original 2017 * spec states (in one of two places). 2018 */ 2019 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2020 tp->rcv_up = th->th_seq + th->th_urp; 2021 so->so_oobmark = so->so_rcv.sb_cc + 2022 (tp->rcv_up - tp->rcv_nxt) - 1; 2023 if (so->so_oobmark == 0) 2024 so->so_state |= SS_RCVATMARK; 2025 sohasoutofband(so); 2026 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2027 } 2028 /* 2029 * Remove out of band data so doesn't get presented to user. 2030 * This can happen independent of advancing the URG pointer, 2031 * but if two URG's are pending at once, some out-of-band 2032 * data may creep in... ick. 2033 */ 2034 if (th->th_urp <= (u_int16_t) tlen 2035 #ifdef SO_OOBINLINE 2036 && (so->so_options & SO_OOBINLINE) == 0 2037 #endif 2038 ) 2039 tcp_pulloutofband(so, th, m, hdroptlen); 2040 } else 2041 /* 2042 * If no out of band data is expected, 2043 * pull receive urgent pointer along 2044 * with the receive window. 2045 */ 2046 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2047 tp->rcv_up = tp->rcv_nxt; 2048 dodata: /* XXX */ 2049 2050 /* 2051 * Process the segment text, merging it into the TCP sequencing queue, 2052 * and arranging for acknowledgement of receipt if necessary. 2053 * This process logically involves adjusting tp->rcv_wnd as data 2054 * is presented to the user (this happens in tcp_usrreq.c, 2055 * case PRU_RCVD). If a FIN has already been received on this 2056 * connection then we just ignore the text. 2057 */ 2058 if ((tlen || (tiflags & TH_FIN)) && 2059 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2060 /* 2061 * Insert segment ti into reassembly queue of tcp with 2062 * control block tp. Return TH_FIN if reassembly now includes 2063 * a segment with FIN. The macro form does the common case 2064 * inline (segment is the next to be received on an 2065 * established connection, and the queue is empty), 2066 * avoiding linkage into and removal from the queue and 2067 * repetition of various conversions. 2068 * Set DELACK for segments received in order, but ack 2069 * immediately when segments are out of order 2070 * (so fast retransmit can work). 2071 */ 2072 /* NOTE: this was TCP_REASS() macro, but used only once */ 2073 TCP_REASS_LOCK(tp); 2074 if (th->th_seq == tp->rcv_nxt && 2075 tp->segq.lh_first == NULL && 2076 tp->t_state == TCPS_ESTABLISHED) { 2077 TCP_SETUP_ACK(tp, th); 2078 tp->rcv_nxt += tlen; 2079 tiflags = th->th_flags & TH_FIN; 2080 tcpstat.tcps_rcvpack++; 2081 tcpstat.tcps_rcvbyte += tlen; 2082 ND6_HINT(tp); 2083 m_adj(m, hdroptlen); 2084 sbappend(&(so)->so_rcv, m); 2085 sorwakeup(so); 2086 } else { 2087 m_adj(m, hdroptlen); 2088 tiflags = tcp_reass(tp, th, m, &tlen); 2089 tp->t_flags |= TF_ACKNOW; 2090 } 2091 TCP_REASS_UNLOCK(tp); 2092 2093 /* 2094 * Note the amount of data that peer has sent into 2095 * our window, in order to estimate the sender's 2096 * buffer size. 2097 */ 2098 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2099 } else { 2100 m_freem(m); 2101 m = NULL; 2102 tiflags &= ~TH_FIN; 2103 } 2104 2105 /* 2106 * If FIN is received ACK the FIN and let the user know 2107 * that the connection is closing. Ignore a FIN received before 2108 * the connection is fully established. 2109 */ 2110 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2111 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2112 socantrcvmore(so); 2113 tp->t_flags |= TF_ACKNOW; 2114 tp->rcv_nxt++; 2115 } 2116 switch (tp->t_state) { 2117 2118 /* 2119 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2120 */ 2121 case TCPS_ESTABLISHED: 2122 tp->t_state = TCPS_CLOSE_WAIT; 2123 break; 2124 2125 /* 2126 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2127 * enter the CLOSING state. 2128 */ 2129 case TCPS_FIN_WAIT_1: 2130 tp->t_state = TCPS_CLOSING; 2131 break; 2132 2133 /* 2134 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2135 * starting the time-wait timer, turning off the other 2136 * standard timers. 2137 */ 2138 case TCPS_FIN_WAIT_2: 2139 tp->t_state = TCPS_TIME_WAIT; 2140 tcp_canceltimers(tp); 2141 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2142 soisdisconnected(so); 2143 break; 2144 2145 /* 2146 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2147 */ 2148 case TCPS_TIME_WAIT: 2149 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2150 break; 2151 } 2152 } 2153 if (so->so_options & SO_DEBUG) { 2154 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2155 } 2156 2157 /* 2158 * Return any desired output. 2159 */ 2160 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2161 (void) tcp_output(tp); 2162 if (tcp_saveti) 2163 m_freem(tcp_saveti); 2164 return; 2165 2166 badsyn: 2167 /* 2168 * Received a bad SYN. Increment counters and dropwithreset. 2169 */ 2170 tcpstat.tcps_badsyn++; 2171 tp = NULL; 2172 goto dropwithreset; 2173 2174 dropafterack: 2175 /* 2176 * Generate an ACK dropping incoming segment if it occupies 2177 * sequence space, where the ACK reflects our state. 2178 */ 2179 if (tiflags & TH_RST) 2180 goto drop; 2181 m_freem(m); 2182 tp->t_flags |= TF_ACKNOW; 2183 (void) tcp_output(tp); 2184 if (tcp_saveti) 2185 m_freem(tcp_saveti); 2186 return; 2187 2188 dropwithreset_ratelim: 2189 /* 2190 * We may want to rate-limit RSTs in certain situations, 2191 * particularly if we are sending an RST in response to 2192 * an attempt to connect to or otherwise communicate with 2193 * a port for which we have no socket. 2194 */ 2195 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2196 tcp_rst_ppslim) == 0) { 2197 /* XXX stat */ 2198 goto drop; 2199 } 2200 /* ...fall into dropwithreset... */ 2201 2202 dropwithreset: 2203 /* 2204 * Generate a RST, dropping incoming segment. 2205 * Make ACK acceptable to originator of segment. 2206 */ 2207 if (tiflags & TH_RST) 2208 goto drop; 2209 { 2210 /* 2211 * need to recover version # field, which was overwritten on 2212 * ip_cksum computation. 2213 */ 2214 struct ip *sip; 2215 sip = mtod(m, struct ip *); 2216 switch (af) { 2217 #ifdef INET 2218 case AF_INET: 2219 sip->ip_v = 4; 2220 break; 2221 #endif 2222 #ifdef INET6 2223 case AF_INET6: 2224 sip->ip_v = 6; 2225 break; 2226 #endif 2227 } 2228 } 2229 if (tiflags & TH_ACK) 2230 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2231 else { 2232 if (tiflags & TH_SYN) 2233 tlen++; 2234 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2235 TH_RST|TH_ACK); 2236 } 2237 if (tcp_saveti) 2238 m_freem(tcp_saveti); 2239 return; 2240 2241 badcsum: 2242 tcpstat.tcps_rcvbadsum++; 2243 drop: 2244 /* 2245 * Drop space held by incoming segment and return. 2246 */ 2247 if (tp) { 2248 if (tp->t_inpcb) 2249 so = tp->t_inpcb->inp_socket; 2250 #ifdef INET6 2251 else if (tp->t_in6pcb) 2252 so = tp->t_in6pcb->in6p_socket; 2253 #endif 2254 else 2255 so = NULL; 2256 if (so && (so->so_options & SO_DEBUG) != 0) 2257 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2258 } 2259 if (tcp_saveti) 2260 m_freem(tcp_saveti); 2261 m_freem(m); 2262 return; 2263 } 2264 2265 void 2266 tcp_dooptions(tp, cp, cnt, th, oi) 2267 struct tcpcb *tp; 2268 u_char *cp; 2269 int cnt; 2270 struct tcphdr *th; 2271 struct tcp_opt_info *oi; 2272 { 2273 u_int16_t mss; 2274 int opt, optlen; 2275 2276 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2277 opt = cp[0]; 2278 if (opt == TCPOPT_EOL) 2279 break; 2280 if (opt == TCPOPT_NOP) 2281 optlen = 1; 2282 else { 2283 if (cnt < 2) 2284 break; 2285 optlen = cp[1]; 2286 if (optlen < 2 || optlen > cnt) 2287 break; 2288 } 2289 switch (opt) { 2290 2291 default: 2292 continue; 2293 2294 case TCPOPT_MAXSEG: 2295 if (optlen != TCPOLEN_MAXSEG) 2296 continue; 2297 if (!(th->th_flags & TH_SYN)) 2298 continue; 2299 bcopy(cp + 2, &mss, sizeof(mss)); 2300 oi->maxseg = ntohs(mss); 2301 break; 2302 2303 case TCPOPT_WINDOW: 2304 if (optlen != TCPOLEN_WINDOW) 2305 continue; 2306 if (!(th->th_flags & TH_SYN)) 2307 continue; 2308 tp->t_flags |= TF_RCVD_SCALE; 2309 tp->requested_s_scale = cp[2]; 2310 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2311 #if 0 /*XXX*/ 2312 char *p; 2313 2314 if (ip) 2315 p = ntohl(ip->ip_src); 2316 #ifdef INET6 2317 else if (ip6) 2318 p = ip6_sprintf(&ip6->ip6_src); 2319 #endif 2320 else 2321 p = "(unknown)"; 2322 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2323 "assuming %d\n", 2324 tp->requested_s_scale, p, 2325 TCP_MAX_WINSHIFT); 2326 #else 2327 log(LOG_ERR, "TCP: invalid wscale %d, " 2328 "assuming %d\n", 2329 tp->requested_s_scale, 2330 TCP_MAX_WINSHIFT); 2331 #endif 2332 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2333 } 2334 break; 2335 2336 case TCPOPT_TIMESTAMP: 2337 if (optlen != TCPOLEN_TIMESTAMP) 2338 continue; 2339 oi->ts_present = 1; 2340 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2341 NTOHL(oi->ts_val); 2342 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2343 NTOHL(oi->ts_ecr); 2344 2345 /* 2346 * A timestamp received in a SYN makes 2347 * it ok to send timestamp requests and replies. 2348 */ 2349 if (th->th_flags & TH_SYN) { 2350 tp->t_flags |= TF_RCVD_TSTMP; 2351 tp->ts_recent = oi->ts_val; 2352 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2353 } 2354 break; 2355 case TCPOPT_SACK_PERMITTED: 2356 if (optlen != TCPOLEN_SACK_PERMITTED) 2357 continue; 2358 if (!(th->th_flags & TH_SYN)) 2359 continue; 2360 tp->t_flags &= ~TF_CANT_TXSACK; 2361 break; 2362 2363 case TCPOPT_SACK: 2364 if (tp->t_flags & TF_IGNR_RXSACK) 2365 continue; 2366 if (optlen % 8 != 2 || optlen < 10) 2367 continue; 2368 cp += 2; 2369 optlen -= 2; 2370 for (; optlen > 0; cp -= 8, optlen -= 8) { 2371 tcp_seq lwe, rwe; 2372 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2373 NTOHL(lwe); 2374 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2375 NTOHL(rwe); 2376 /* tcp_mark_sacked(tp, lwe, rwe); */ 2377 } 2378 break; 2379 } 2380 } 2381 } 2382 2383 /* 2384 * Pull out of band byte out of a segment so 2385 * it doesn't appear in the user's data queue. 2386 * It is still reflected in the segment length for 2387 * sequencing purposes. 2388 */ 2389 void 2390 tcp_pulloutofband(so, th, m, off) 2391 struct socket *so; 2392 struct tcphdr *th; 2393 struct mbuf *m; 2394 int off; 2395 { 2396 int cnt = off + th->th_urp - 1; 2397 2398 while (cnt >= 0) { 2399 if (m->m_len > cnt) { 2400 char *cp = mtod(m, caddr_t) + cnt; 2401 struct tcpcb *tp = sototcpcb(so); 2402 2403 tp->t_iobc = *cp; 2404 tp->t_oobflags |= TCPOOB_HAVEDATA; 2405 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2406 m->m_len--; 2407 return; 2408 } 2409 cnt -= m->m_len; 2410 m = m->m_next; 2411 if (m == 0) 2412 break; 2413 } 2414 panic("tcp_pulloutofband"); 2415 } 2416 2417 /* 2418 * Collect new round-trip time estimate 2419 * and update averages and current timeout. 2420 */ 2421 void 2422 tcp_xmit_timer(tp, rtt) 2423 struct tcpcb *tp; 2424 short rtt; 2425 { 2426 short delta; 2427 short rttmin; 2428 2429 tcpstat.tcps_rttupdated++; 2430 --rtt; 2431 if (tp->t_srtt != 0) { 2432 /* 2433 * srtt is stored as fixed point with 3 bits after the 2434 * binary point (i.e., scaled by 8). The following magic 2435 * is equivalent to the smoothing algorithm in rfc793 with 2436 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2437 * point). Adjust rtt to origin 0. 2438 */ 2439 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2440 if ((tp->t_srtt += delta) <= 0) 2441 tp->t_srtt = 1 << 2; 2442 /* 2443 * We accumulate a smoothed rtt variance (actually, a 2444 * smoothed mean difference), then set the retransmit 2445 * timer to smoothed rtt + 4 times the smoothed variance. 2446 * rttvar is stored as fixed point with 2 bits after the 2447 * binary point (scaled by 4). The following is 2448 * equivalent to rfc793 smoothing with an alpha of .75 2449 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2450 * rfc793's wired-in beta. 2451 */ 2452 if (delta < 0) 2453 delta = -delta; 2454 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2455 if ((tp->t_rttvar += delta) <= 0) 2456 tp->t_rttvar = 1 << 2; 2457 } else { 2458 /* 2459 * No rtt measurement yet - use the unsmoothed rtt. 2460 * Set the variance to half the rtt (so our first 2461 * retransmit happens at 3*rtt). 2462 */ 2463 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2464 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2465 } 2466 tp->t_rtt = 0; 2467 tp->t_rxtshift = 0; 2468 2469 /* 2470 * the retransmit should happen at rtt + 4 * rttvar. 2471 * Because of the way we do the smoothing, srtt and rttvar 2472 * will each average +1/2 tick of bias. When we compute 2473 * the retransmit timer, we want 1/2 tick of rounding and 2474 * 1 extra tick because of +-1/2 tick uncertainty in the 2475 * firing of the timer. The bias will give us exactly the 2476 * 1.5 tick we need. But, because the bias is 2477 * statistical, we have to test that we don't drop below 2478 * the minimum feasible timer (which is 2 ticks). 2479 */ 2480 if (tp->t_rttmin > rtt + 2) 2481 rttmin = tp->t_rttmin; 2482 else 2483 rttmin = rtt + 2; 2484 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2485 2486 /* 2487 * We received an ack for a packet that wasn't retransmitted; 2488 * it is probably safe to discard any error indications we've 2489 * received recently. This isn't quite right, but close enough 2490 * for now (a route might have failed after we sent a segment, 2491 * and the return path might not be symmetrical). 2492 */ 2493 tp->t_softerror = 0; 2494 } 2495 2496 /* 2497 * Checks for partial ack. If partial ack arrives, force the retransmission 2498 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2499 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2500 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2501 */ 2502 int 2503 tcp_newreno(tp, th) 2504 struct tcpcb *tp; 2505 struct tcphdr *th; 2506 { 2507 tcp_seq onxt = tp->snd_nxt; 2508 u_long ocwnd = tp->snd_cwnd; 2509 2510 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2511 /* 2512 * snd_una has not yet been updated and the socket's send 2513 * buffer has not yet drained off the ACK'd data, so we 2514 * have to leave snd_una as it was to get the correct data 2515 * offset in tcp_output(). 2516 */ 2517 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2518 tp->t_rtt = 0; 2519 tp->snd_nxt = th->th_ack; 2520 /* 2521 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2522 * is not yet updated when we're called. 2523 */ 2524 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2525 (void) tcp_output(tp); 2526 tp->snd_cwnd = ocwnd; 2527 if (SEQ_GT(onxt, tp->snd_nxt)) 2528 tp->snd_nxt = onxt; 2529 /* 2530 * Partial window deflation. Relies on fact that tp->snd_una 2531 * not updated yet. 2532 */ 2533 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2534 return 1; 2535 } 2536 return 0; 2537 } 2538 2539 2540 /* 2541 * TCP compressed state engine. Currently used to hold compressed 2542 * state for SYN_RECEIVED. 2543 */ 2544 2545 u_long syn_cache_count; 2546 u_int32_t syn_hash1, syn_hash2; 2547 2548 #define SYN_HASH(sa, sp, dp) \ 2549 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2550 ((u_int32_t)(sp)))^syn_hash2))) 2551 #ifndef INET6 2552 #define SYN_HASHALL(hash, src, dst) \ 2553 do { \ 2554 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2555 ((struct sockaddr_in *)(src))->sin_port, \ 2556 ((struct sockaddr_in *)(dst))->sin_port); \ 2557 } while (0) 2558 #else 2559 #define SYN_HASH6(sa, sp, dp) \ 2560 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2561 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2562 & 0x7fffffff) 2563 2564 #define SYN_HASHALL(hash, src, dst) \ 2565 do { \ 2566 switch ((src)->sa_family) { \ 2567 case AF_INET: \ 2568 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2569 ((struct sockaddr_in *)(src))->sin_port, \ 2570 ((struct sockaddr_in *)(dst))->sin_port); \ 2571 break; \ 2572 case AF_INET6: \ 2573 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2574 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2575 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2576 break; \ 2577 default: \ 2578 hash = 0; \ 2579 } \ 2580 } while (0) 2581 #endif /* INET6 */ 2582 2583 #define SYN_CACHE_RM(sc) \ 2584 do { \ 2585 LIST_REMOVE((sc), sc_bucketq); \ 2586 (sc)->sc_tp = NULL; \ 2587 LIST_REMOVE((sc), sc_tpq); \ 2588 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2589 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2590 syn_cache_count--; \ 2591 } while (0) 2592 2593 #define SYN_CACHE_PUT(sc) \ 2594 do { \ 2595 if ((sc)->sc_ipopts) \ 2596 (void) m_free((sc)->sc_ipopts); \ 2597 if ((sc)->sc_route4.ro_rt != NULL) \ 2598 RTFREE((sc)->sc_route4.ro_rt); \ 2599 pool_put(&syn_cache_pool, (sc)); \ 2600 } while (0) 2601 2602 struct pool syn_cache_pool; 2603 2604 /* 2605 * We don't estimate RTT with SYNs, so each packet starts with the default 2606 * RTT and each timer queue has a fixed timeout value. This allows us to 2607 * optimize the timer queues somewhat. 2608 */ 2609 #define SYN_CACHE_TIMER_ARM(sc) \ 2610 do { \ 2611 TCPT_RANGESET((sc)->sc_rxtcur, \ 2612 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2613 TCPTV_REXMTMAX); \ 2614 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2615 } while (0) 2616 2617 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2618 2619 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2620 2621 void 2622 syn_cache_init() 2623 { 2624 int i; 2625 2626 /* Initialize the hash buckets. */ 2627 for (i = 0; i < tcp_syn_cache_size; i++) 2628 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2629 2630 /* Initialize the timer queues. */ 2631 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2632 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2633 2634 /* Initialize the syn cache pool. */ 2635 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2636 "synpl", 0, NULL, NULL, M_PCB); 2637 } 2638 2639 void 2640 syn_cache_insert(sc, tp) 2641 struct syn_cache *sc; 2642 struct tcpcb *tp; 2643 { 2644 struct syn_cache_head *scp; 2645 struct syn_cache *sc2; 2646 int s, i; 2647 2648 /* 2649 * If there are no entries in the hash table, reinitialize 2650 * the hash secrets. 2651 */ 2652 if (syn_cache_count == 0) { 2653 struct timeval tv; 2654 microtime(&tv); 2655 syn_hash1 = random() ^ (u_long)≻ 2656 syn_hash2 = random() ^ tv.tv_usec; 2657 } 2658 2659 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2660 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2661 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2662 2663 /* 2664 * Make sure that we don't overflow the per-bucket 2665 * limit or the total cache size limit. 2666 */ 2667 s = splsoftnet(); 2668 if (scp->sch_length >= tcp_syn_bucket_limit) { 2669 tcpstat.tcps_sc_bucketoverflow++; 2670 /* 2671 * The bucket is full. Toss the oldest element in the 2672 * bucket. This will be the entry with our bucket 2673 * index closest to the front of the timer queue with 2674 * the largest timeout value. 2675 * 2676 * Note: This timer queue traversal may be expensive, so 2677 * we hope that this doesn't happen very often. It is 2678 * much more likely that we'll overflow the entire 2679 * cache, which is much easier to handle; see below. 2680 */ 2681 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2682 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2683 sc2 != NULL; 2684 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2685 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2686 SYN_CACHE_RM(sc2); 2687 SYN_CACHE_PUT(sc2); 2688 goto insert; /* 2 level break */ 2689 } 2690 } 2691 } 2692 #ifdef DIAGNOSTIC 2693 /* 2694 * This should never happen; we should always find an 2695 * entry in our bucket. 2696 */ 2697 panic("syn_cache_insert: bucketoverflow: impossible"); 2698 #endif 2699 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2700 tcpstat.tcps_sc_overflowed++; 2701 /* 2702 * The cache is full. Toss the oldest entry in the 2703 * entire cache. This is the front entry in the 2704 * first non-empty timer queue with the largest 2705 * timeout value. 2706 */ 2707 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2708 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2709 if (sc2 == NULL) 2710 continue; 2711 SYN_CACHE_RM(sc2); 2712 SYN_CACHE_PUT(sc2); 2713 goto insert; /* symmetry with above */ 2714 } 2715 #ifdef DIAGNOSTIC 2716 /* 2717 * This should never happen; we should always find an 2718 * entry in the cache. 2719 */ 2720 panic("syn_cache_insert: cache overflow: impossible"); 2721 #endif 2722 } 2723 2724 insert: 2725 /* 2726 * Initialize the entry's timer. 2727 */ 2728 sc->sc_rxttot = 0; 2729 sc->sc_rxtshift = 0; 2730 SYN_CACHE_TIMER_ARM(sc); 2731 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2732 2733 /* Link it from tcpcb entry */ 2734 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2735 2736 /* Put it into the bucket. */ 2737 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2738 scp->sch_length++; 2739 syn_cache_count++; 2740 2741 tcpstat.tcps_sc_added++; 2742 splx(s); 2743 } 2744 2745 /* 2746 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2747 * If we have retransmitted an entry the maximum number of times, expire 2748 * that entry. 2749 */ 2750 void 2751 syn_cache_timer() 2752 { 2753 struct syn_cache *sc, *nsc; 2754 int i, s; 2755 2756 s = splsoftnet(); 2757 2758 /* 2759 * First, get all the entries that need to be retransmitted, or 2760 * must be expired due to exceeding the initial keepalive time. 2761 */ 2762 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2763 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2764 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2765 sc = nsc) { 2766 nsc = TAILQ_NEXT(sc, sc_timeq); 2767 2768 /* 2769 * Compute the total amount of time this entry has 2770 * been on a queue. If this entry has been on longer 2771 * than the keep alive timer would allow, expire it. 2772 */ 2773 sc->sc_rxttot += sc->sc_rxtcur; 2774 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2775 tcpstat.tcps_sc_timed_out++; 2776 SYN_CACHE_RM(sc); 2777 SYN_CACHE_PUT(sc); 2778 continue; 2779 } 2780 2781 tcpstat.tcps_sc_retransmitted++; 2782 (void) syn_cache_respond(sc, NULL); 2783 2784 /* Advance this entry onto the next timer queue. */ 2785 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2786 sc->sc_rxtshift = i + 1; 2787 SYN_CACHE_TIMER_ARM(sc); 2788 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2789 sc, sc_timeq); 2790 } 2791 } 2792 2793 /* 2794 * Now get all the entries that are expired due to too many 2795 * retransmissions. 2796 */ 2797 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2798 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2799 sc = nsc) { 2800 nsc = TAILQ_NEXT(sc, sc_timeq); 2801 tcpstat.tcps_sc_timed_out++; 2802 SYN_CACHE_RM(sc); 2803 SYN_CACHE_PUT(sc); 2804 } 2805 splx(s); 2806 } 2807 2808 /* 2809 * Remove syn cache created by the specified tcb entry, 2810 * because this does not make sense to keep them 2811 * (if there's no tcb entry, syn cache entry will never be used) 2812 */ 2813 void 2814 syn_cache_cleanup(tp) 2815 struct tcpcb *tp; 2816 { 2817 struct syn_cache *sc, *nsc; 2818 int s; 2819 2820 s = splsoftnet(); 2821 2822 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2823 nsc = LIST_NEXT(sc, sc_tpq); 2824 2825 #ifdef DIAGNOSTIC 2826 if (sc->sc_tp != tp) 2827 panic("invalid sc_tp in syn_cache_cleanup"); 2828 #endif 2829 SYN_CACHE_RM(sc); 2830 SYN_CACHE_PUT(sc); 2831 } 2832 /* just for safety */ 2833 LIST_INIT(&tp->t_sc); 2834 2835 splx(s); 2836 } 2837 2838 /* 2839 * Find an entry in the syn cache. 2840 */ 2841 struct syn_cache * 2842 syn_cache_lookup(src, dst, headp) 2843 struct sockaddr *src; 2844 struct sockaddr *dst; 2845 struct syn_cache_head **headp; 2846 { 2847 struct syn_cache *sc; 2848 struct syn_cache_head *scp; 2849 u_int32_t hash; 2850 int s; 2851 2852 SYN_HASHALL(hash, src, dst); 2853 2854 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2855 *headp = scp; 2856 s = splsoftnet(); 2857 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2858 sc = LIST_NEXT(sc, sc_bucketq)) { 2859 if (sc->sc_hash != hash) 2860 continue; 2861 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2862 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2863 splx(s); 2864 return (sc); 2865 } 2866 } 2867 splx(s); 2868 return (NULL); 2869 } 2870 2871 /* 2872 * This function gets called when we receive an ACK for a 2873 * socket in the LISTEN state. We look up the connection 2874 * in the syn cache, and if its there, we pull it out of 2875 * the cache and turn it into a full-blown connection in 2876 * the SYN-RECEIVED state. 2877 * 2878 * The return values may not be immediately obvious, and their effects 2879 * can be subtle, so here they are: 2880 * 2881 * NULL SYN was not found in cache; caller should drop the 2882 * packet and send an RST. 2883 * 2884 * -1 We were unable to create the new connection, and are 2885 * aborting it. An ACK,RST is being sent to the peer 2886 * (unless we got screwey sequence numbners; see below), 2887 * because the 3-way handshake has been completed. Caller 2888 * should not free the mbuf, since we may be using it. If 2889 * we are not, we will free it. 2890 * 2891 * Otherwise, the return value is a pointer to the new socket 2892 * associated with the connection. 2893 */ 2894 struct socket * 2895 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2896 struct sockaddr *src; 2897 struct sockaddr *dst; 2898 struct tcphdr *th; 2899 unsigned int hlen, tlen; 2900 struct socket *so; 2901 struct mbuf *m; 2902 { 2903 struct syn_cache *sc; 2904 struct syn_cache_head *scp; 2905 struct inpcb *inp = NULL; 2906 #ifdef INET6 2907 struct in6pcb *in6p = NULL; 2908 #endif 2909 struct tcpcb *tp = 0; 2910 struct mbuf *am; 2911 int s; 2912 struct socket *oso; 2913 2914 s = splsoftnet(); 2915 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2916 splx(s); 2917 return (NULL); 2918 } 2919 2920 /* 2921 * Verify the sequence and ack numbers. Try getting the correct 2922 * response again. 2923 */ 2924 if ((th->th_ack != sc->sc_iss + 1) || 2925 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2926 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2927 (void) syn_cache_respond(sc, m); 2928 splx(s); 2929 return ((struct socket *)(-1)); 2930 } 2931 2932 /* Remove this cache entry */ 2933 SYN_CACHE_RM(sc); 2934 splx(s); 2935 2936 /* 2937 * Ok, create the full blown connection, and set things up 2938 * as they would have been set up if we had created the 2939 * connection when the SYN arrived. If we can't create 2940 * the connection, abort it. 2941 */ 2942 /* 2943 * inp still has the OLD in_pcb stuff, set the 2944 * v6-related flags on the new guy, too. This is 2945 * done particularly for the case where an AF_INET6 2946 * socket is bound only to a port, and a v4 connection 2947 * comes in on that port. 2948 * we also copy the flowinfo from the original pcb 2949 * to the new one. 2950 */ 2951 { 2952 struct inpcb *parentinpcb; 2953 2954 parentinpcb = (struct inpcb *)so->so_pcb; 2955 2956 oso = so; 2957 so = sonewconn(so, SS_ISCONNECTED); 2958 if (so == NULL) 2959 goto resetandabort; 2960 2961 switch (so->so_proto->pr_domain->dom_family) { 2962 #ifdef INET 2963 case AF_INET: 2964 inp = sotoinpcb(so); 2965 break; 2966 #endif 2967 #ifdef INET6 2968 case AF_INET6: 2969 in6p = sotoin6pcb(so); 2970 break; 2971 #endif 2972 } 2973 } 2974 switch (src->sa_family) { 2975 #ifdef INET 2976 case AF_INET: 2977 if (inp) { 2978 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2979 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2980 inp->inp_options = ip_srcroute(); 2981 in_pcbstate(inp, INP_BOUND); 2982 if (inp->inp_options == NULL) { 2983 inp->inp_options = sc->sc_ipopts; 2984 sc->sc_ipopts = NULL; 2985 } 2986 } 2987 #ifdef INET6 2988 else if (in6p) { 2989 /* IPv4 packet to AF_INET6 socket */ 2990 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2991 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2992 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2993 &in6p->in6p_laddr.s6_addr32[3], 2994 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2995 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2996 in6totcpcb(in6p)->t_family = AF_INET; 2997 } 2998 #endif 2999 break; 3000 #endif 3001 #ifdef INET6 3002 case AF_INET6: 3003 if (in6p) { 3004 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3005 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3006 #if 0 3007 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3008 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3009 #endif 3010 } 3011 break; 3012 #endif 3013 } 3014 #ifdef INET6 3015 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3016 struct in6pcb *oin6p = sotoin6pcb(oso); 3017 /* inherit socket options from the listening socket */ 3018 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3019 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3020 m_freem(in6p->in6p_options); 3021 in6p->in6p_options = 0; 3022 } 3023 ip6_savecontrol(in6p, &in6p->in6p_options, 3024 mtod(m, struct ip6_hdr *), m); 3025 } 3026 #endif 3027 3028 #ifdef IPSEC 3029 /* 3030 * we make a copy of policy, instead of sharing the policy, 3031 * for better behavior in terms of SA lookup and dead SA removal. 3032 */ 3033 if (inp) { 3034 /* copy old policy into new socket's */ 3035 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3036 printf("tcp_input: could not copy policy\n"); 3037 } 3038 #ifdef INET6 3039 else if (in6p) { 3040 /* copy old policy into new socket's */ 3041 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 3042 printf("tcp_input: could not copy policy\n"); 3043 } 3044 #endif 3045 #endif 3046 3047 /* 3048 * Give the new socket our cached route reference. 3049 */ 3050 if (inp) 3051 inp->inp_route = sc->sc_route4; /* struct assignment */ 3052 #ifdef INET6 3053 else 3054 in6p->in6p_route = sc->sc_route6; 3055 #endif 3056 sc->sc_route4.ro_rt = NULL; 3057 3058 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3059 if (am == NULL) 3060 goto resetandabort; 3061 am->m_len = src->sa_len; 3062 bcopy(src, mtod(am, caddr_t), src->sa_len); 3063 if (inp) { 3064 if (in_pcbconnect(inp, am)) { 3065 (void) m_free(am); 3066 goto resetandabort; 3067 } 3068 } 3069 #ifdef INET6 3070 else if (in6p) { 3071 if (src->sa_family == AF_INET) { 3072 /* IPv4 packet to AF_INET6 socket */ 3073 struct sockaddr_in6 *sin6; 3074 sin6 = mtod(am, struct sockaddr_in6 *); 3075 am->m_len = sizeof(*sin6); 3076 bzero(sin6, sizeof(*sin6)); 3077 sin6->sin6_family = AF_INET6; 3078 sin6->sin6_len = sizeof(*sin6); 3079 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3080 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3081 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3082 &sin6->sin6_addr.s6_addr32[3], 3083 sizeof(sin6->sin6_addr.s6_addr32[3])); 3084 } 3085 if (in6_pcbconnect(in6p, am)) { 3086 (void) m_free(am); 3087 goto resetandabort; 3088 } 3089 } 3090 #endif 3091 else { 3092 (void) m_free(am); 3093 goto resetandabort; 3094 } 3095 (void) m_free(am); 3096 3097 if (inp) 3098 tp = intotcpcb(inp); 3099 #ifdef INET6 3100 else if (in6p) 3101 tp = in6totcpcb(in6p); 3102 #endif 3103 else 3104 tp = NULL; 3105 if (sc->sc_request_r_scale != 15) { 3106 tp->requested_s_scale = sc->sc_requested_s_scale; 3107 tp->request_r_scale = sc->sc_request_r_scale; 3108 tp->snd_scale = sc->sc_requested_s_scale; 3109 tp->rcv_scale = sc->sc_request_r_scale; 3110 tp->t_flags |= TF_RCVD_SCALE; 3111 } 3112 if (sc->sc_flags & SCF_TIMESTAMP) 3113 tp->t_flags |= TF_RCVD_TSTMP; 3114 tp->ts_timebase = sc->sc_timebase; 3115 3116 tp->t_template = tcp_template(tp); 3117 if (tp->t_template == 0) { 3118 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3119 so = NULL; 3120 m_freem(m); 3121 goto abort; 3122 } 3123 3124 tp->iss = sc->sc_iss; 3125 tp->irs = sc->sc_irs; 3126 tcp_sendseqinit(tp); 3127 tcp_rcvseqinit(tp); 3128 tp->t_state = TCPS_SYN_RECEIVED; 3129 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3130 tcpstat.tcps_accepts++; 3131 3132 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3133 tp->t_ourmss = sc->sc_ourmaxseg; 3134 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3135 3136 /* 3137 * Initialize the initial congestion window. If we 3138 * had to retransmit the SYN,ACK, we must initialize cwnd 3139 * to 1 segment (i.e. the Loss Window). 3140 */ 3141 if (sc->sc_rxtshift) 3142 tp->snd_cwnd = tp->t_peermss; 3143 else 3144 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3145 3146 tcp_rmx_rtt(tp); 3147 tp->snd_wl1 = sc->sc_irs; 3148 tp->rcv_up = sc->sc_irs + 1; 3149 3150 /* 3151 * This is what whould have happened in tcp_ouput() when 3152 * the SYN,ACK was sent. 3153 */ 3154 tp->snd_up = tp->snd_una; 3155 tp->snd_max = tp->snd_nxt = tp->iss+1; 3156 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3157 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3158 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3159 tp->last_ack_sent = tp->rcv_nxt; 3160 3161 tcpstat.tcps_sc_completed++; 3162 SYN_CACHE_PUT(sc); 3163 return (so); 3164 3165 resetandabort: 3166 (void) tcp_respond(NULL, m, m, th, 3167 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3168 abort: 3169 if (so != NULL) 3170 (void) soabort(so); 3171 SYN_CACHE_PUT(sc); 3172 tcpstat.tcps_sc_aborted++; 3173 return ((struct socket *)(-1)); 3174 } 3175 3176 /* 3177 * This function is called when we get a RST for a 3178 * non-existent connection, so that we can see if the 3179 * connection is in the syn cache. If it is, zap it. 3180 */ 3181 3182 void 3183 syn_cache_reset(src, dst, th) 3184 struct sockaddr *src; 3185 struct sockaddr *dst; 3186 struct tcphdr *th; 3187 { 3188 struct syn_cache *sc; 3189 struct syn_cache_head *scp; 3190 int s = splsoftnet(); 3191 3192 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3193 splx(s); 3194 return; 3195 } 3196 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3197 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3198 splx(s); 3199 return; 3200 } 3201 SYN_CACHE_RM(sc); 3202 splx(s); 3203 tcpstat.tcps_sc_reset++; 3204 SYN_CACHE_PUT(sc); 3205 } 3206 3207 void 3208 syn_cache_unreach(src, dst, th) 3209 struct sockaddr *src; 3210 struct sockaddr *dst; 3211 struct tcphdr *th; 3212 { 3213 struct syn_cache *sc; 3214 struct syn_cache_head *scp; 3215 int s; 3216 3217 s = splsoftnet(); 3218 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3219 splx(s); 3220 return; 3221 } 3222 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3223 if (ntohl (th->th_seq) != sc->sc_iss) { 3224 splx(s); 3225 return; 3226 } 3227 3228 /* 3229 * If we've rertransmitted 3 times and this is our second error, 3230 * we remove the entry. Otherwise, we allow it to continue on. 3231 * This prevents us from incorrectly nuking an entry during a 3232 * spurious network outage. 3233 * 3234 * See tcp_notify(). 3235 */ 3236 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3237 sc->sc_flags |= SCF_UNREACH; 3238 splx(s); 3239 return; 3240 } 3241 3242 SYN_CACHE_RM(sc); 3243 splx(s); 3244 tcpstat.tcps_sc_unreach++; 3245 SYN_CACHE_PUT(sc); 3246 } 3247 3248 /* 3249 * Given a LISTEN socket and an inbound SYN request, add 3250 * this to the syn cache, and send back a segment: 3251 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3252 * to the source. 3253 * 3254 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3255 * Doing so would require that we hold onto the data and deliver it 3256 * to the application. However, if we are the target of a SYN-flood 3257 * DoS attack, an attacker could send data which would eventually 3258 * consume all available buffer space if it were ACKed. By not ACKing 3259 * the data, we avoid this DoS scenario. 3260 */ 3261 3262 int 3263 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3264 struct sockaddr *src; 3265 struct sockaddr *dst; 3266 struct tcphdr *th; 3267 unsigned int hlen; 3268 struct socket *so; 3269 struct mbuf *m; 3270 u_char *optp; 3271 int optlen; 3272 struct tcp_opt_info *oi; 3273 { 3274 struct tcpcb tb, *tp; 3275 long win; 3276 struct syn_cache *sc; 3277 struct syn_cache_head *scp; 3278 struct mbuf *ipopts; 3279 3280 tp = sototcpcb(so); 3281 3282 /* 3283 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3284 * 3285 * Note this check is performed in tcp_input() very early on. 3286 */ 3287 3288 /* 3289 * Initialize some local state. 3290 */ 3291 win = sbspace(&so->so_rcv); 3292 if (win > TCP_MAXWIN) 3293 win = TCP_MAXWIN; 3294 3295 switch (src->sa_family) { 3296 #ifdef INET 3297 case AF_INET: 3298 /* 3299 * Remember the IP options, if any. 3300 */ 3301 ipopts = ip_srcroute(); 3302 break; 3303 #endif 3304 default: 3305 ipopts = NULL; 3306 } 3307 3308 if (optp) { 3309 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3310 tcp_dooptions(&tb, optp, optlen, th, oi); 3311 } else 3312 tb.t_flags = 0; 3313 3314 /* 3315 * See if we already have an entry for this connection. 3316 * If we do, resend the SYN,ACK. We do not count this 3317 * as a retransmission (XXX though maybe we should). 3318 */ 3319 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3320 tcpstat.tcps_sc_dupesyn++; 3321 if (ipopts) { 3322 /* 3323 * If we were remembering a previous source route, 3324 * forget it and use the new one we've been given. 3325 */ 3326 if (sc->sc_ipopts) 3327 (void) m_free(sc->sc_ipopts); 3328 sc->sc_ipopts = ipopts; 3329 } 3330 sc->sc_timestamp = tb.ts_recent; 3331 if (syn_cache_respond(sc, m) == 0) { 3332 tcpstat.tcps_sndacks++; 3333 tcpstat.tcps_sndtotal++; 3334 } 3335 return (1); 3336 } 3337 3338 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3339 if (sc == NULL) { 3340 if (ipopts) 3341 (void) m_free(ipopts); 3342 return (0); 3343 } 3344 3345 /* 3346 * Fill in the cache, and put the necessary IP and TCP 3347 * options into the reply. 3348 */ 3349 bzero(sc, sizeof(struct syn_cache)); 3350 bcopy(src, &sc->sc_src, src->sa_len); 3351 bcopy(dst, &sc->sc_dst, dst->sa_len); 3352 sc->sc_flags = 0; 3353 sc->sc_ipopts = ipopts; 3354 sc->sc_irs = th->th_seq; 3355 switch (src->sa_family) { 3356 #ifdef INET 3357 case AF_INET: 3358 { 3359 struct sockaddr_in *srcin = (void *) src; 3360 struct sockaddr_in *dstin = (void *) dst; 3361 3362 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3363 &srcin->sin_addr, dstin->sin_port, 3364 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3365 break; 3366 } 3367 #endif /* INET */ 3368 #ifdef INET6 3369 case AF_INET6: 3370 { 3371 struct sockaddr_in6 *srcin6 = (void *) src; 3372 struct sockaddr_in6 *dstin6 = (void *) dst; 3373 3374 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3375 &srcin6->sin6_addr, dstin6->sin6_port, 3376 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3377 break; 3378 } 3379 #endif /* INET6 */ 3380 } 3381 sc->sc_peermaxseg = oi->maxseg; 3382 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3383 m->m_pkthdr.rcvif : NULL, 3384 sc->sc_src.sa.sa_family); 3385 sc->sc_win = win; 3386 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3387 sc->sc_timestamp = tb.ts_recent; 3388 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3389 sc->sc_flags |= SCF_TIMESTAMP; 3390 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3391 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3392 sc->sc_requested_s_scale = tb.requested_s_scale; 3393 sc->sc_request_r_scale = 0; 3394 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3395 TCP_MAXWIN << sc->sc_request_r_scale < 3396 so->so_rcv.sb_hiwat) 3397 sc->sc_request_r_scale++; 3398 } else { 3399 sc->sc_requested_s_scale = 15; 3400 sc->sc_request_r_scale = 15; 3401 } 3402 sc->sc_tp = tp; 3403 if (syn_cache_respond(sc, m) == 0) { 3404 syn_cache_insert(sc, tp); 3405 tcpstat.tcps_sndacks++; 3406 tcpstat.tcps_sndtotal++; 3407 } else { 3408 SYN_CACHE_PUT(sc); 3409 tcpstat.tcps_sc_dropped++; 3410 } 3411 return (1); 3412 } 3413 3414 int 3415 syn_cache_respond(sc, m) 3416 struct syn_cache *sc; 3417 struct mbuf *m; 3418 { 3419 struct route *ro; 3420 u_int8_t *optp; 3421 int optlen, error; 3422 u_int16_t tlen; 3423 struct ip *ip = NULL; 3424 #ifdef INET6 3425 struct ip6_hdr *ip6 = NULL; 3426 #endif 3427 struct tcphdr *th; 3428 u_int hlen; 3429 3430 switch (sc->sc_src.sa.sa_family) { 3431 case AF_INET: 3432 hlen = sizeof(struct ip); 3433 ro = &sc->sc_route4; 3434 break; 3435 #ifdef INET6 3436 case AF_INET6: 3437 hlen = sizeof(struct ip6_hdr); 3438 ro = (struct route *)&sc->sc_route6; 3439 break; 3440 #endif 3441 default: 3442 if (m) 3443 m_freem(m); 3444 return EAFNOSUPPORT; 3445 } 3446 3447 /* Compute the size of the TCP options. */ 3448 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3449 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3450 3451 tlen = hlen + sizeof(struct tcphdr) + optlen; 3452 3453 /* 3454 * Create the IP+TCP header from scratch. 3455 */ 3456 if (m) 3457 m_freem(m); 3458 #ifdef DIAGNOSTIC 3459 if (max_linkhdr + tlen > MCLBYTES) 3460 return (ENOBUFS); 3461 #endif 3462 MGETHDR(m, M_DONTWAIT, MT_DATA); 3463 if (m && tlen > MHLEN) { 3464 MCLGET(m, M_DONTWAIT); 3465 if ((m->m_flags & M_EXT) == 0) { 3466 m_freem(m); 3467 m = NULL; 3468 } 3469 } 3470 if (m == NULL) 3471 return (ENOBUFS); 3472 3473 /* Fixup the mbuf. */ 3474 m->m_data += max_linkhdr; 3475 m->m_len = m->m_pkthdr.len = tlen; 3476 #ifdef IPSEC 3477 if (sc->sc_tp) { 3478 struct tcpcb *tp; 3479 struct socket *so; 3480 3481 tp = sc->sc_tp; 3482 if (tp->t_inpcb) 3483 so = tp->t_inpcb->inp_socket; 3484 #ifdef INET6 3485 else if (tp->t_in6pcb) 3486 so = tp->t_in6pcb->in6p_socket; 3487 #endif 3488 else 3489 so = NULL; 3490 /* use IPsec policy on listening socket, on SYN ACK */ 3491 if (ipsec_setsocket(m, so) != 0) { 3492 m_freem(m); 3493 return ENOBUFS; 3494 } 3495 } 3496 #endif 3497 m->m_pkthdr.rcvif = NULL; 3498 memset(mtod(m, u_char *), 0, tlen); 3499 3500 switch (sc->sc_src.sa.sa_family) { 3501 case AF_INET: 3502 ip = mtod(m, struct ip *); 3503 ip->ip_dst = sc->sc_src.sin.sin_addr; 3504 ip->ip_src = sc->sc_dst.sin.sin_addr; 3505 ip->ip_p = IPPROTO_TCP; 3506 th = (struct tcphdr *)(ip + 1); 3507 th->th_dport = sc->sc_src.sin.sin_port; 3508 th->th_sport = sc->sc_dst.sin.sin_port; 3509 break; 3510 #ifdef INET6 3511 case AF_INET6: 3512 ip6 = mtod(m, struct ip6_hdr *); 3513 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3514 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3515 ip6->ip6_nxt = IPPROTO_TCP; 3516 /* ip6_plen will be updated in ip6_output() */ 3517 th = (struct tcphdr *)(ip6 + 1); 3518 th->th_dport = sc->sc_src.sin6.sin6_port; 3519 th->th_sport = sc->sc_dst.sin6.sin6_port; 3520 break; 3521 #endif 3522 default: 3523 th = NULL; 3524 } 3525 3526 th->th_seq = htonl(sc->sc_iss); 3527 th->th_ack = htonl(sc->sc_irs + 1); 3528 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3529 th->th_flags = TH_SYN|TH_ACK; 3530 th->th_win = htons(sc->sc_win); 3531 /* th_sum already 0 */ 3532 /* th_urp already 0 */ 3533 3534 /* Tack on the TCP options. */ 3535 optp = (u_int8_t *)(th + 1); 3536 *optp++ = TCPOPT_MAXSEG; 3537 *optp++ = 4; 3538 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3539 *optp++ = sc->sc_ourmaxseg & 0xff; 3540 3541 if (sc->sc_request_r_scale != 15) { 3542 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3543 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3544 sc->sc_request_r_scale); 3545 optp += 4; 3546 } 3547 3548 if (sc->sc_flags & SCF_TIMESTAMP) { 3549 u_int32_t *lp = (u_int32_t *)(optp); 3550 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3551 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3552 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3553 *lp = htonl(sc->sc_timestamp); 3554 optp += TCPOLEN_TSTAMP_APPA; 3555 } 3556 3557 /* Compute the packet's checksum. */ 3558 switch (sc->sc_src.sa.sa_family) { 3559 case AF_INET: 3560 ip->ip_len = htons(tlen - hlen); 3561 th->th_sum = 0; 3562 th->th_sum = in_cksum(m, tlen); 3563 break; 3564 #ifdef INET6 3565 case AF_INET6: 3566 ip6->ip6_plen = htons(tlen - hlen); 3567 th->th_sum = 0; 3568 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3569 break; 3570 #endif 3571 } 3572 3573 /* 3574 * Fill in some straggling IP bits. Note the stack expects 3575 * ip_len to be in host order, for convenience. 3576 */ 3577 switch (sc->sc_src.sa.sa_family) { 3578 #ifdef INET 3579 case AF_INET: 3580 ip->ip_len = tlen; 3581 ip->ip_ttl = ip_defttl; 3582 /* XXX tos? */ 3583 break; 3584 #endif 3585 #ifdef INET6 3586 case AF_INET6: 3587 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3588 ip6->ip6_vfc |= IPV6_VERSION; 3589 ip6->ip6_plen = htons(tlen - hlen); 3590 /* ip6_hlim will be initialized afterwards */ 3591 /* XXX flowlabel? */ 3592 break; 3593 #endif 3594 } 3595 3596 switch (sc->sc_src.sa.sa_family) { 3597 #ifdef INET 3598 case AF_INET: 3599 error = ip_output(m, sc->sc_ipopts, ro, 3600 (ip_mtudisc ? IP_MTUDISC : 0), 3601 NULL); 3602 break; 3603 #endif 3604 #ifdef INET6 3605 case AF_INET6: 3606 ip6->ip6_hlim = in6_selecthlim(NULL, 3607 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3608 3609 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3610 0, NULL, NULL); 3611 break; 3612 #endif 3613 default: 3614 error = EAFNOSUPPORT; 3615 break; 3616 } 3617 return (error); 3618 } 3619