tcp_input.c revision 1.115
1 /* $NetBSD: tcp_input.c,v 1.115 2000/07/27 06:18:13 itojun Exp $ */ 2 3 /* 4 %%% portions-copyright-nrl-95 5 Portions of this software are Copyright 1995-1998 by Randall Atkinson, 6 Ronald Lee, Daniel McDonald, Bao Phan, and Chris Winters. All Rights 7 Reserved. All rights under this copyright have been assigned to the US 8 Naval Research Laboratory (NRL). The NRL Copyright Notice and License 9 Agreement Version 1.1 (January 17, 1995) applies to these portions of the 10 software. 11 You should have received a copy of the license with this software. If you 12 didn't get a copy, you may request one from <license (at) ipv6.nrl.navy.mil>. 13 14 */ 15 16 /* 17 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 18 * All rights reserved. 19 * 20 * Redistribution and use in source and binary forms, with or without 21 * modification, are permitted provided that the following conditions 22 * are met: 23 * 1. Redistributions of source code must retain the above copyright 24 * notice, this list of conditions and the following disclaimer. 25 * 2. Redistributions in binary form must reproduce the above copyright 26 * notice, this list of conditions and the following disclaimer in the 27 * documentation and/or other materials provided with the distribution. 28 * 3. Neither the name of the project nor the names of its contributors 29 * may be used to endorse or promote products derived from this software 30 * without specific prior written permission. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 33 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 34 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 35 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 36 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 37 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 38 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 39 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 40 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 41 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 42 * SUCH DAMAGE. 43 */ 44 45 /*- 46 * Copyright (c) 1997, 1998, 1999 The NetBSD Foundation, Inc. 47 * All rights reserved. 48 * 49 * This code is derived from software contributed to The NetBSD Foundation 50 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 51 * Facility, NASA Ames Research Center. 52 * 53 * Redistribution and use in source and binary forms, with or without 54 * modification, are permitted provided that the following conditions 55 * are met: 56 * 1. Redistributions of source code must retain the above copyright 57 * notice, this list of conditions and the following disclaimer. 58 * 2. Redistributions in binary form must reproduce the above copyright 59 * notice, this list of conditions and the following disclaimer in the 60 * documentation and/or other materials provided with the distribution. 61 * 3. All advertising materials mentioning features or use of this software 62 * must display the following acknowledgement: 63 * This product includes software developed by the NetBSD 64 * Foundation, Inc. and its contributors. 65 * 4. Neither the name of The NetBSD Foundation nor the names of its 66 * contributors may be used to endorse or promote products derived 67 * from this software without specific prior written permission. 68 * 69 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 70 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 71 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 72 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 73 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 74 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 75 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 76 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 77 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 78 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 79 * POSSIBILITY OF SUCH DAMAGE. 80 */ 81 82 /* 83 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 84 * The Regents of the University of California. All rights reserved. 85 * 86 * Redistribution and use in source and binary forms, with or without 87 * modification, are permitted provided that the following conditions 88 * are met: 89 * 1. Redistributions of source code must retain the above copyright 90 * notice, this list of conditions and the following disclaimer. 91 * 2. Redistributions in binary form must reproduce the above copyright 92 * notice, this list of conditions and the following disclaimer in the 93 * documentation and/or other materials provided with the distribution. 94 * 3. All advertising materials mentioning features or use of this software 95 * must display the following acknowledgement: 96 * This product includes software developed by the University of 97 * California, Berkeley and its contributors. 98 * 4. Neither the name of the University nor the names of its contributors 99 * may be used to endorse or promote products derived from this software 100 * without specific prior written permission. 101 * 102 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 103 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 104 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 105 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 106 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 107 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 108 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 109 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 110 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 111 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 112 * SUCH DAMAGE. 113 * 114 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 115 */ 116 117 /* 118 * TODO list for SYN cache stuff: 119 * 120 * Find room for a "state" field, which is needed to keep a 121 * compressed state for TIME_WAIT TCBs. It's been noted already 122 * that this is fairly important for very high-volume web and 123 * mail servers, which use a large number of short-lived 124 * connections. 125 */ 126 127 #include "opt_inet.h" 128 #include "opt_ipsec.h" 129 130 #include <sys/param.h> 131 #include <sys/systm.h> 132 #include <sys/malloc.h> 133 #include <sys/mbuf.h> 134 #include <sys/protosw.h> 135 #include <sys/socket.h> 136 #include <sys/socketvar.h> 137 #include <sys/errno.h> 138 #include <sys/syslog.h> 139 #include <sys/pool.h> 140 #include <sys/domain.h> 141 142 #include <net/if.h> 143 #include <net/route.h> 144 #include <net/if_types.h> 145 146 #include <netinet/in.h> 147 #include <netinet/in_systm.h> 148 #include <netinet/ip.h> 149 #include <netinet/in_pcb.h> 150 #include <netinet/ip_var.h> 151 152 #ifdef INET6 153 #ifndef INET 154 #include <netinet/in.h> 155 #endif 156 #include <netinet/ip6.h> 157 #include <netinet6/ip6_var.h> 158 #include <netinet6/in6_pcb.h> 159 #include <netinet6/ip6_var.h> 160 #include <netinet6/in6_var.h> 161 #include <netinet/icmp6.h> 162 #include <netinet6/nd6.h> 163 #endif 164 165 #ifdef PULLDOWN_TEST 166 #ifndef INET6 167 /* always need ip6.h for IP6_EXTHDR_GET */ 168 #include <netinet/ip6.h> 169 #endif 170 #endif 171 172 #include <netinet/tcp.h> 173 #include <netinet/tcp_fsm.h> 174 #include <netinet/tcp_seq.h> 175 #include <netinet/tcp_timer.h> 176 #include <netinet/tcp_var.h> 177 #include <netinet/tcpip.h> 178 #include <netinet/tcp_debug.h> 179 180 #include <machine/stdarg.h> 181 182 #ifdef IPSEC 183 #include <netinet6/ipsec.h> 184 #include <netkey/key.h> 185 #endif /*IPSEC*/ 186 #ifdef INET6 187 #include "faith.h" 188 #endif 189 190 int tcprexmtthresh = 3; 191 int tcp_log_refused; 192 193 struct timeval tcp_rst_ratelim_last; 194 195 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 196 197 /* for modulo comparisons of timestamps */ 198 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 199 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 200 201 /* 202 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 203 */ 204 #ifdef INET6 205 #define ND6_HINT(tp) \ 206 do { \ 207 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 208 && tp->t_in6pcb->in6p_route.ro_rt) { \ 209 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 210 } \ 211 } while (0) 212 #else 213 #define ND6_HINT(tp) 214 #endif 215 216 /* 217 * Macro to compute ACK transmission behavior. Delay the ACK unless 218 * we have already delayed an ACK (must send an ACK every two segments). 219 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 220 * option is enabled. 221 */ 222 #define TCP_SETUP_ACK(tp, th) \ 223 do { \ 224 if ((tp)->t_flags & TF_DELACK || \ 225 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 226 tp->t_flags |= TF_ACKNOW; \ 227 else \ 228 TCP_SET_DELACK(tp); \ 229 } while (0) 230 231 /* 232 * Convert TCP protocol fields to host order for easier processing. 233 */ 234 #define TCP_FIELDS_TO_HOST(th) \ 235 do { \ 236 NTOHL((th)->th_seq); \ 237 NTOHL((th)->th_ack); \ 238 NTOHS((th)->th_win); \ 239 NTOHS((th)->th_urp); \ 240 } while (0) 241 242 int 243 tcp_reass(tp, th, m, tlen) 244 struct tcpcb *tp; 245 struct tcphdr *th; 246 struct mbuf *m; 247 int *tlen; 248 { 249 struct ipqent *p, *q, *nq, *tiqe = NULL; 250 struct socket *so = NULL; 251 int pkt_flags; 252 tcp_seq pkt_seq; 253 unsigned pkt_len; 254 u_long rcvpartdupbyte = 0; 255 u_long rcvoobyte; 256 257 if (tp->t_inpcb) 258 so = tp->t_inpcb->inp_socket; 259 #ifdef INET6 260 else if (tp->t_in6pcb) 261 so = tp->t_in6pcb->in6p_socket; 262 #endif 263 264 TCP_REASS_LOCK_CHECK(tp); 265 266 /* 267 * Call with th==0 after become established to 268 * force pre-ESTABLISHED data up to user socket. 269 */ 270 if (th == 0) 271 goto present; 272 273 rcvoobyte = *tlen; 274 /* 275 * Copy these to local variables because the tcpiphdr 276 * gets munged while we are collapsing mbufs. 277 */ 278 pkt_seq = th->th_seq; 279 pkt_len = *tlen; 280 pkt_flags = th->th_flags; 281 /* 282 * Find a segment which begins after this one does. 283 */ 284 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 285 nq = q->ipqe_q.le_next; 286 /* 287 * If the received segment is just right after this 288 * fragment, merge the two together and then check 289 * for further overlaps. 290 */ 291 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 292 #ifdef TCPREASS_DEBUG 293 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 294 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 295 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 296 #endif 297 pkt_len += q->ipqe_len; 298 pkt_flags |= q->ipqe_flags; 299 pkt_seq = q->ipqe_seq; 300 m_cat(q->ipqe_m, m); 301 m = q->ipqe_m; 302 goto free_ipqe; 303 } 304 /* 305 * If the received segment is completely past this 306 * fragment, we need to go the next fragment. 307 */ 308 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 309 p = q; 310 continue; 311 } 312 /* 313 * If the fragment is past the received segment, 314 * it (or any following) can't be concatenated. 315 */ 316 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 317 break; 318 /* 319 * We've received all the data in this segment before. 320 * mark it as a duplicate and return. 321 */ 322 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 323 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 324 tcpstat.tcps_rcvduppack++; 325 tcpstat.tcps_rcvdupbyte += pkt_len; 326 m_freem(m); 327 if (tiqe != NULL) 328 pool_put(&ipqent_pool, tiqe); 329 return (0); 330 } 331 /* 332 * Received segment completely overlaps this fragment 333 * so we drop the fragment (this keeps the temporal 334 * ordering of segments correct). 335 */ 336 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 337 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 338 rcvpartdupbyte += q->ipqe_len; 339 m_freem(q->ipqe_m); 340 goto free_ipqe; 341 } 342 /* 343 * RX'ed segment extends past the end of the 344 * fragment. Drop the overlapping bytes. Then 345 * merge the fragment and segment then treat as 346 * a longer received packet. 347 */ 348 if (SEQ_LT(q->ipqe_seq, pkt_seq) 349 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 350 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 351 #ifdef TCPREASS_DEBUG 352 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 353 tp, overlap, 354 pkt_seq, pkt_seq + pkt_len, pkt_len); 355 #endif 356 m_adj(m, overlap); 357 rcvpartdupbyte += overlap; 358 m_cat(q->ipqe_m, m); 359 m = q->ipqe_m; 360 pkt_seq = q->ipqe_seq; 361 pkt_len += q->ipqe_len - overlap; 362 rcvoobyte -= overlap; 363 goto free_ipqe; 364 } 365 /* 366 * RX'ed segment extends past the front of the 367 * fragment. Drop the overlapping bytes on the 368 * received packet. The packet will then be 369 * contatentated with this fragment a bit later. 370 */ 371 if (SEQ_GT(q->ipqe_seq, pkt_seq) 372 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 373 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 374 #ifdef TCPREASS_DEBUG 375 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 376 tp, overlap, 377 pkt_seq, pkt_seq + pkt_len, pkt_len); 378 #endif 379 m_adj(m, -overlap); 380 pkt_len -= overlap; 381 rcvpartdupbyte += overlap; 382 rcvoobyte -= overlap; 383 } 384 /* 385 * If the received segment immediates precedes this 386 * fragment then tack the fragment onto this segment 387 * and reinsert the data. 388 */ 389 if (q->ipqe_seq == pkt_seq + pkt_len) { 390 #ifdef TCPREASS_DEBUG 391 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 392 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 393 pkt_seq, pkt_seq + pkt_len, pkt_len); 394 #endif 395 pkt_len += q->ipqe_len; 396 pkt_flags |= q->ipqe_flags; 397 m_cat(m, q->ipqe_m); 398 LIST_REMOVE(q, ipqe_q); 399 LIST_REMOVE(q, ipqe_timeq); 400 if (tiqe == NULL) { 401 tiqe = q; 402 } else { 403 pool_put(&ipqent_pool, q); 404 } 405 break; 406 } 407 /* 408 * If the fragment is before the segment, remember it. 409 * When this loop is terminated, p will contain the 410 * pointer to fragment that is right before the received 411 * segment. 412 */ 413 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 414 p = q; 415 416 continue; 417 418 /* 419 * This is a common operation. It also will allow 420 * to save doing a malloc/free in most instances. 421 */ 422 free_ipqe: 423 LIST_REMOVE(q, ipqe_q); 424 LIST_REMOVE(q, ipqe_timeq); 425 if (tiqe == NULL) { 426 tiqe = q; 427 } else { 428 pool_put(&ipqent_pool, q); 429 } 430 } 431 432 /* 433 * Allocate a new queue entry since the received segment did not 434 * collapse onto any other out-of-order block; thus we are allocating 435 * a new block. If it had collapsed, tiqe would not be NULL and 436 * we would be reusing it. 437 * XXX If we can't, just drop the packet. XXX 438 */ 439 if (tiqe == NULL) { 440 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 441 if (tiqe == NULL) { 442 tcpstat.tcps_rcvmemdrop++; 443 m_freem(m); 444 return (0); 445 } 446 } 447 448 /* 449 * Update the counters. 450 */ 451 tcpstat.tcps_rcvoopack++; 452 tcpstat.tcps_rcvoobyte += rcvoobyte; 453 if (rcvpartdupbyte) { 454 tcpstat.tcps_rcvpartduppack++; 455 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 456 } 457 458 /* 459 * Insert the new fragment queue entry into both queues. 460 */ 461 tiqe->ipqe_m = m; 462 tiqe->ipqe_seq = pkt_seq; 463 tiqe->ipqe_len = pkt_len; 464 tiqe->ipqe_flags = pkt_flags; 465 if (p == NULL) { 466 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 467 #ifdef TCPREASS_DEBUG 468 if (tiqe->ipqe_seq != tp->rcv_nxt) 469 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 470 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 471 #endif 472 } else { 473 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 474 #ifdef TCPREASS_DEBUG 475 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 476 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 477 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 478 #endif 479 } 480 481 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 482 483 present: 484 /* 485 * Present data to user, advancing rcv_nxt through 486 * completed sequence space. 487 */ 488 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 489 return (0); 490 q = tp->segq.lh_first; 491 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 492 return (0); 493 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 494 return (0); 495 496 tp->rcv_nxt += q->ipqe_len; 497 pkt_flags = q->ipqe_flags & TH_FIN; 498 ND6_HINT(tp); 499 500 LIST_REMOVE(q, ipqe_q); 501 LIST_REMOVE(q, ipqe_timeq); 502 if (so->so_state & SS_CANTRCVMORE) 503 m_freem(q->ipqe_m); 504 else 505 sbappend(&so->so_rcv, q->ipqe_m); 506 pool_put(&ipqent_pool, q); 507 sorwakeup(so); 508 return (pkt_flags); 509 } 510 511 #if defined(INET6) && !defined(TCP6) 512 int 513 tcp6_input(mp, offp, proto) 514 struct mbuf **mp; 515 int *offp, proto; 516 { 517 struct mbuf *m = *mp; 518 519 /* 520 * draft-itojun-ipv6-tcp-to-anycast 521 * better place to put this in? 522 */ 523 if (m->m_flags & M_ANYCAST6) { 524 struct ip6_hdr *ip6; 525 if (m->m_len < sizeof(struct ip6_hdr)) { 526 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 527 tcpstat.tcps_rcvshort++; 528 return IPPROTO_DONE; 529 } 530 } 531 ip6 = mtod(m, struct ip6_hdr *); 532 icmp6_error(m, ICMP6_DST_UNREACH, 533 ICMP6_DST_UNREACH_ADDR, 534 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 535 return IPPROTO_DONE; 536 } 537 538 tcp_input(m, *offp, proto); 539 return IPPROTO_DONE; 540 } 541 #endif 542 543 /* 544 * TCP input routine, follows pages 65-76 of the 545 * protocol specification dated September, 1981 very closely. 546 */ 547 void 548 #if __STDC__ 549 tcp_input(struct mbuf *m, ...) 550 #else 551 tcp_input(m, va_alist) 552 struct mbuf *m; 553 #endif 554 { 555 int proto; 556 struct tcphdr *th; 557 struct ip *ip; 558 struct inpcb *inp; 559 #ifdef INET6 560 struct ip6_hdr *ip6; 561 struct in6pcb *in6p; 562 #endif 563 caddr_t optp = NULL; 564 int optlen = 0; 565 int len, tlen, toff, hdroptlen = 0; 566 struct tcpcb *tp = 0; 567 int tiflags; 568 struct socket *so = NULL; 569 int todrop, acked, ourfinisacked, needoutput = 0; 570 short ostate = 0; 571 int iss = 0; 572 u_long tiwin; 573 struct tcp_opt_info opti; 574 int off, iphlen; 575 va_list ap; 576 int af; /* af on the wire */ 577 struct mbuf *tcp_saveti = NULL; 578 579 va_start(ap, m); 580 toff = va_arg(ap, int); 581 proto = va_arg(ap, int); 582 va_end(ap); 583 584 tcpstat.tcps_rcvtotal++; 585 586 bzero(&opti, sizeof(opti)); 587 opti.ts_present = 0; 588 opti.maxseg = 0; 589 590 /* 591 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 592 * 593 * TCP is, by definition, unicast, so we reject all 594 * multicast outright. 595 * 596 * Note, there are additional src/dst address checks in 597 * the AF-specific code below. 598 */ 599 if (m->m_flags & (M_BCAST|M_MCAST)) { 600 /* XXX stat */ 601 goto drop; 602 } 603 #ifdef INET6 604 if (m->m_flags & M_ANYCAST6) { 605 /* XXX stat */ 606 goto drop; 607 } 608 #endif 609 610 /* 611 * Get IP and TCP header together in first mbuf. 612 * Note: IP leaves IP header in first mbuf. 613 */ 614 ip = mtod(m, struct ip *); 615 #ifdef INET6 616 ip6 = NULL; 617 #endif 618 switch (ip->ip_v) { 619 case 4: 620 af = AF_INET; 621 iphlen = sizeof(struct ip); 622 #ifndef PULLDOWN_TEST 623 /* would like to get rid of this... */ 624 if (toff > sizeof (struct ip)) { 625 ip_stripoptions(m, (struct mbuf *)0); 626 toff = sizeof(struct ip); 627 } 628 if (m->m_len < toff + sizeof (struct tcphdr)) { 629 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 630 tcpstat.tcps_rcvshort++; 631 return; 632 } 633 } 634 ip = mtod(m, struct ip *); 635 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 636 #else 637 ip = mtod(m, struct ip *); 638 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 639 sizeof(struct tcphdr)); 640 if (th == NULL) { 641 tcpstat.tcps_rcvshort++; 642 return; 643 } 644 #endif 645 646 /* 647 * Make sure destination address is not multicast. 648 * Source address checked in ip_input(). 649 */ 650 if (IN_MULTICAST(ip->ip_dst.s_addr)) { 651 /* XXX stat */ 652 goto drop; 653 } 654 655 /* We do the checksum after PCB lookup... */ 656 len = ip->ip_len; 657 tlen = len - toff; 658 break; 659 #ifdef INET6 660 case 6: 661 ip = NULL; 662 iphlen = sizeof(struct ip6_hdr); 663 af = AF_INET6; 664 #ifndef PULLDOWN_TEST 665 if (m->m_len < toff + sizeof(struct tcphdr)) { 666 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 667 if (m == NULL) { 668 tcpstat.tcps_rcvshort++; 669 return; 670 } 671 } 672 ip6 = mtod(m, struct ip6_hdr *); 673 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 674 #else 675 ip6 = mtod(m, struct ip6_hdr *); 676 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 677 sizeof(struct tcphdr)); 678 if (th == NULL) { 679 tcpstat.tcps_rcvshort++; 680 return; 681 } 682 #endif 683 684 /* Be proactive about malicious use of IPv4 mapped address */ 685 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 686 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 687 /* XXX stat */ 688 goto drop; 689 } 690 691 /* 692 * Be proactive about unspecified IPv6 address in source. 693 * As we use all-zero to indicate unbounded/unconnected pcb, 694 * unspecified IPv6 address can be used to confuse us. 695 * 696 * Note that packets with unspecified IPv6 destination is 697 * already dropped in ip6_input. 698 */ 699 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 700 /* XXX stat */ 701 goto drop; 702 } 703 704 /* 705 * Make sure destination address is not multicast. 706 * Source address checked in ip6_input(). 707 */ 708 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 709 /* XXX stat */ 710 goto drop; 711 } 712 713 /* We do the checksum after PCB lookup... */ 714 len = m->m_pkthdr.len; 715 tlen = len - toff; 716 break; 717 #endif 718 default: 719 m_freem(m); 720 return; 721 } 722 723 /* 724 * Check that TCP offset makes sense, 725 * pull out TCP options and adjust length. XXX 726 */ 727 off = th->th_off << 2; 728 if (off < sizeof (struct tcphdr) || off > tlen) { 729 tcpstat.tcps_rcvbadoff++; 730 goto drop; 731 } 732 tlen -= off; 733 734 /* 735 * tcp_input() has been modified to use tlen to mean the TCP data 736 * length throughout the function. Other functions can use 737 * m->m_pkthdr.len as the basis for calculating the TCP data length. 738 * rja 739 */ 740 741 if (off > sizeof (struct tcphdr)) { 742 #ifndef PULLDOWN_TEST 743 if (m->m_len < toff + off) { 744 if ((m = m_pullup(m, toff + off)) == 0) { 745 tcpstat.tcps_rcvshort++; 746 return; 747 } 748 switch (af) { 749 case AF_INET: 750 ip = mtod(m, struct ip *); 751 break; 752 #ifdef INET6 753 case AF_INET6: 754 ip6 = mtod(m, struct ip6_hdr *); 755 break; 756 #endif 757 } 758 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 759 } 760 #else 761 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 762 if (th == NULL) { 763 tcpstat.tcps_rcvshort++; 764 return; 765 } 766 /* 767 * NOTE: ip/ip6 will not be affected by m_pulldown() 768 * (as they're before toff) and we don't need to update those. 769 */ 770 #endif 771 optlen = off - sizeof (struct tcphdr); 772 optp = ((caddr_t)th) + sizeof(struct tcphdr); 773 /* 774 * Do quick retrieval of timestamp options ("options 775 * prediction?"). If timestamp is the only option and it's 776 * formatted as recommended in RFC 1323 appendix A, we 777 * quickly get the values now and not bother calling 778 * tcp_dooptions(), etc. 779 */ 780 if ((optlen == TCPOLEN_TSTAMP_APPA || 781 (optlen > TCPOLEN_TSTAMP_APPA && 782 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 783 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 784 (th->th_flags & TH_SYN) == 0) { 785 opti.ts_present = 1; 786 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 787 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 788 optp = NULL; /* we've parsed the options */ 789 } 790 } 791 tiflags = th->th_flags; 792 793 /* 794 * Locate pcb for segment. 795 */ 796 findpcb: 797 inp = NULL; 798 #ifdef INET6 799 in6p = NULL; 800 #endif 801 switch (af) { 802 case AF_INET: 803 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 804 ip->ip_dst, th->th_dport); 805 if (inp == 0) { 806 ++tcpstat.tcps_pcbhashmiss; 807 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 808 } 809 #if defined(INET6) && !defined(TCP6) 810 if (inp == 0) { 811 struct in6_addr s, d; 812 813 /* mapped addr case */ 814 bzero(&s, sizeof(s)); 815 s.s6_addr16[5] = htons(0xffff); 816 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 817 bzero(&d, sizeof(d)); 818 d.s6_addr16[5] = htons(0xffff); 819 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 820 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 821 &d, th->th_dport, 0); 822 if (in6p == 0) { 823 ++tcpstat.tcps_pcbhashmiss; 824 in6p = in6_pcblookup_bind(&tcb6, &d, 825 th->th_dport, 0); 826 } 827 } 828 #endif 829 #ifndef INET6 830 if (inp == 0) 831 #else 832 if (inp == 0 && in6p == 0) 833 #endif 834 { 835 ++tcpstat.tcps_noport; 836 if (tcp_log_refused && (tiflags & TH_SYN)) { 837 #ifndef INET6 838 char src[4*sizeof "123"]; 839 char dst[4*sizeof "123"]; 840 #else 841 char src[INET6_ADDRSTRLEN]; 842 char dst[INET6_ADDRSTRLEN]; 843 #endif 844 if (ip) { 845 strcpy(src, inet_ntoa(ip->ip_src)); 846 strcpy(dst, inet_ntoa(ip->ip_dst)); 847 } 848 #ifdef INET6 849 else if (ip6) { 850 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 851 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 852 } 853 #endif 854 else { 855 strcpy(src, "(unknown)"); 856 strcpy(dst, "(unknown)"); 857 } 858 log(LOG_INFO, 859 "Connection attempt to TCP %s:%d from %s:%d\n", 860 dst, ntohs(th->th_dport), 861 src, ntohs(th->th_sport)); 862 } 863 TCP_FIELDS_TO_HOST(th); 864 goto dropwithreset_ratelim; 865 } 866 #ifdef IPSEC 867 if (inp && ipsec4_in_reject(m, inp)) { 868 ipsecstat.in_polvio++; 869 goto drop; 870 } 871 #ifdef INET6 872 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 873 ipsecstat.in_polvio++; 874 goto drop; 875 } 876 #endif 877 #endif /*IPSEC*/ 878 break; 879 #if defined(INET6) && !defined(TCP6) 880 case AF_INET6: 881 { 882 int faith; 883 884 #if defined(NFAITH) && NFAITH > 0 885 if (m->m_pkthdr.rcvif 886 && m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 887 faith = 1; 888 } else 889 faith = 0; 890 #else 891 faith = 0; 892 #endif 893 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 894 &ip6->ip6_dst, th->th_dport, faith); 895 if (in6p == NULL) { 896 ++tcpstat.tcps_pcbhashmiss; 897 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 898 th->th_dport, faith); 899 } 900 if (in6p == NULL) { 901 ++tcpstat.tcps_noport; 902 TCP_FIELDS_TO_HOST(th); 903 goto dropwithreset_ratelim; 904 } 905 #ifdef IPSEC 906 if (ipsec6_in_reject(m, in6p)) { 907 ipsec6stat.in_polvio++; 908 goto drop; 909 } 910 #endif /*IPSEC*/ 911 break; 912 } 913 #endif 914 } 915 916 /* 917 * If the state is CLOSED (i.e., TCB does not exist) then 918 * all data in the incoming segment is discarded. 919 * If the TCB exists but is in CLOSED state, it is embryonic, 920 * but should either do a listen or a connect soon. 921 */ 922 tp = NULL; 923 so = NULL; 924 if (inp) { 925 tp = intotcpcb(inp); 926 so = inp->inp_socket; 927 } 928 #ifdef INET6 929 else if (in6p) { 930 tp = in6totcpcb(in6p); 931 so = in6p->in6p_socket; 932 } 933 #endif 934 if (tp == 0) { 935 TCP_FIELDS_TO_HOST(th); 936 goto dropwithreset_ratelim; 937 } 938 if (tp->t_state == TCPS_CLOSED) 939 goto drop; 940 941 /* 942 * Checksum extended TCP header and data. 943 */ 944 switch (af) { 945 case AF_INET: 946 #ifndef PULLDOWN_TEST 947 { 948 struct ipovly *ipov; 949 ipov = (struct ipovly *)ip; 950 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 951 ipov->ih_len = htons(tlen + off); 952 953 if (in_cksum(m, len) != 0) { 954 tcpstat.tcps_rcvbadsum++; 955 goto drop; 956 } 957 } 958 #else 959 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) { 960 tcpstat.tcps_rcvbadsum++; 961 goto drop; 962 } 963 #endif 964 break; 965 966 #ifdef INET6 967 case AF_INET6: 968 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) { 969 tcpstat.tcps_rcvbadsum++; 970 goto drop; 971 } 972 break; 973 #endif 974 } 975 976 TCP_FIELDS_TO_HOST(th); 977 978 /* Unscale the window into a 32-bit value. */ 979 if ((tiflags & TH_SYN) == 0) 980 tiwin = th->th_win << tp->snd_scale; 981 else 982 tiwin = th->th_win; 983 984 #ifdef INET6 985 /* save packet options if user wanted */ 986 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 987 if (in6p->in6p_options) { 988 m_freem(in6p->in6p_options); 989 in6p->in6p_options = 0; 990 } 991 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 992 } 993 #endif 994 995 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 996 union syn_cache_sa src; 997 union syn_cache_sa dst; 998 999 bzero(&src, sizeof(src)); 1000 bzero(&dst, sizeof(dst)); 1001 switch (af) { 1002 case AF_INET: 1003 src.sin.sin_len = sizeof(struct sockaddr_in); 1004 src.sin.sin_family = AF_INET; 1005 src.sin.sin_addr = ip->ip_src; 1006 src.sin.sin_port = th->th_sport; 1007 1008 dst.sin.sin_len = sizeof(struct sockaddr_in); 1009 dst.sin.sin_family = AF_INET; 1010 dst.sin.sin_addr = ip->ip_dst; 1011 dst.sin.sin_port = th->th_dport; 1012 break; 1013 #ifdef INET6 1014 case AF_INET6: 1015 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1016 src.sin6.sin6_family = AF_INET6; 1017 src.sin6.sin6_addr = ip6->ip6_src; 1018 src.sin6.sin6_port = th->th_sport; 1019 1020 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1021 dst.sin6.sin6_family = AF_INET6; 1022 dst.sin6.sin6_addr = ip6->ip6_dst; 1023 dst.sin6.sin6_port = th->th_dport; 1024 break; 1025 #endif /* INET6 */ 1026 default: 1027 goto badsyn; /*sanity*/ 1028 } 1029 1030 if (so->so_options & SO_DEBUG) { 1031 ostate = tp->t_state; 1032 1033 tcp_saveti = NULL; 1034 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1035 goto nosave; 1036 1037 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1038 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1039 if (!tcp_saveti) 1040 goto nosave; 1041 } else { 1042 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1043 if (!tcp_saveti) 1044 goto nosave; 1045 tcp_saveti->m_len = iphlen; 1046 m_copydata(m, 0, iphlen, 1047 mtod(tcp_saveti, caddr_t)); 1048 } 1049 1050 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1051 m_freem(tcp_saveti); 1052 tcp_saveti = NULL; 1053 } else { 1054 tcp_saveti->m_len += sizeof(struct tcphdr); 1055 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1056 sizeof(struct tcphdr)); 1057 } 1058 if (tcp_saveti) { 1059 /* 1060 * need to recover version # field, which was 1061 * overwritten on ip_cksum computation. 1062 */ 1063 struct ip *sip; 1064 sip = mtod(tcp_saveti, struct ip *); 1065 switch (af) { 1066 case AF_INET: 1067 sip->ip_v = 4; 1068 break; 1069 #ifdef INET6 1070 case AF_INET6: 1071 sip->ip_v = 6; 1072 break; 1073 #endif 1074 } 1075 } 1076 nosave:; 1077 } 1078 if (so->so_options & SO_ACCEPTCONN) { 1079 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1080 if (tiflags & TH_RST) { 1081 syn_cache_reset(&src.sa, &dst.sa, th); 1082 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1083 (TH_ACK|TH_SYN)) { 1084 /* 1085 * Received a SYN,ACK. This should 1086 * never happen while we are in 1087 * LISTEN. Send an RST. 1088 */ 1089 goto badsyn; 1090 } else if (tiflags & TH_ACK) { 1091 so = syn_cache_get(&src.sa, &dst.sa, 1092 th, toff, tlen, so, m); 1093 if (so == NULL) { 1094 /* 1095 * We don't have a SYN for 1096 * this ACK; send an RST. 1097 */ 1098 goto badsyn; 1099 } else if (so == 1100 (struct socket *)(-1)) { 1101 /* 1102 * We were unable to create 1103 * the connection. If the 1104 * 3-way handshake was 1105 * completed, and RST has 1106 * been sent to the peer. 1107 * Since the mbuf might be 1108 * in use for the reply, 1109 * do not free it. 1110 */ 1111 m = NULL; 1112 } else { 1113 /* 1114 * We have created a 1115 * full-blown connection. 1116 */ 1117 tp = NULL; 1118 inp = NULL; 1119 #ifdef INET6 1120 in6p = NULL; 1121 #endif 1122 switch (so->so_proto->pr_domain->dom_family) { 1123 case AF_INET: 1124 inp = sotoinpcb(so); 1125 tp = intotcpcb(inp); 1126 break; 1127 #ifdef INET6 1128 case AF_INET6: 1129 in6p = sotoin6pcb(so); 1130 tp = in6totcpcb(in6p); 1131 break; 1132 #endif 1133 } 1134 if (tp == NULL) 1135 goto badsyn; /*XXX*/ 1136 tiwin <<= tp->snd_scale; 1137 goto after_listen; 1138 } 1139 } else { 1140 /* 1141 * None of RST, SYN or ACK was set. 1142 * This is an invalid packet for a 1143 * TCB in LISTEN state. Send a RST. 1144 */ 1145 goto badsyn; 1146 } 1147 } else { 1148 /* 1149 * Received a SYN. 1150 */ 1151 1152 /* 1153 * LISTEN socket received a SYN 1154 * from itself? This can't possibly 1155 * be valid; drop the packet. 1156 */ 1157 if (th->th_sport == th->th_dport) { 1158 int i; 1159 1160 switch (af) { 1161 case AF_INET: 1162 i = in_hosteq(ip->ip_src, ip->ip_dst); 1163 break; 1164 #ifdef INET6 1165 case AF_INET6: 1166 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1167 break; 1168 #endif 1169 default: 1170 i = 1; 1171 } 1172 if (i) { 1173 tcpstat.tcps_badsyn++; 1174 goto drop; 1175 } 1176 } 1177 1178 /* 1179 * SYN looks ok; create compressed TCP 1180 * state for it. 1181 */ 1182 if (so->so_qlen <= so->so_qlimit && 1183 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1184 so, m, optp, optlen, &opti)) 1185 m = NULL; 1186 } 1187 goto drop; 1188 } 1189 } 1190 1191 after_listen: 1192 #ifdef DIAGNOSTIC 1193 /* 1194 * Should not happen now that all embryonic connections 1195 * are handled with compressed state. 1196 */ 1197 if (tp->t_state == TCPS_LISTEN) 1198 panic("tcp_input: TCPS_LISTEN"); 1199 #endif 1200 1201 /* 1202 * Segment received on connection. 1203 * Reset idle time and keep-alive timer. 1204 */ 1205 tp->t_idle = 0; 1206 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1207 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1208 1209 /* 1210 * Process options. 1211 */ 1212 if (optp) 1213 tcp_dooptions(tp, optp, optlen, th, &opti); 1214 1215 /* 1216 * Header prediction: check for the two common cases 1217 * of a uni-directional data xfer. If the packet has 1218 * no control flags, is in-sequence, the window didn't 1219 * change and we're not retransmitting, it's a 1220 * candidate. If the length is zero and the ack moved 1221 * forward, we're the sender side of the xfer. Just 1222 * free the data acked & wake any higher level process 1223 * that was blocked waiting for space. If the length 1224 * is non-zero and the ack didn't move, we're the 1225 * receiver side. If we're getting packets in-order 1226 * (the reassembly queue is empty), add the data to 1227 * the socket buffer and note that we need a delayed ack. 1228 */ 1229 if (tp->t_state == TCPS_ESTABLISHED && 1230 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1231 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1232 th->th_seq == tp->rcv_nxt && 1233 tiwin && tiwin == tp->snd_wnd && 1234 tp->snd_nxt == tp->snd_max) { 1235 1236 /* 1237 * If last ACK falls within this segment's sequence numbers, 1238 * record the timestamp. 1239 */ 1240 if (opti.ts_present && 1241 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1242 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1243 tp->ts_recent_age = tcp_now; 1244 tp->ts_recent = opti.ts_val; 1245 } 1246 1247 if (tlen == 0) { 1248 if (SEQ_GT(th->th_ack, tp->snd_una) && 1249 SEQ_LEQ(th->th_ack, tp->snd_max) && 1250 tp->snd_cwnd >= tp->snd_wnd && 1251 tp->t_dupacks < tcprexmtthresh) { 1252 /* 1253 * this is a pure ack for outstanding data. 1254 */ 1255 ++tcpstat.tcps_predack; 1256 if (opti.ts_present && opti.ts_ecr) 1257 tcp_xmit_timer(tp, 1258 tcp_now - opti.ts_ecr + 1); 1259 else if (tp->t_rtt && 1260 SEQ_GT(th->th_ack, tp->t_rtseq)) 1261 tcp_xmit_timer(tp, tp->t_rtt); 1262 acked = th->th_ack - tp->snd_una; 1263 tcpstat.tcps_rcvackpack++; 1264 tcpstat.tcps_rcvackbyte += acked; 1265 ND6_HINT(tp); 1266 sbdrop(&so->so_snd, acked); 1267 /* 1268 * We want snd_recover to track snd_una to 1269 * avoid sequence wraparound problems for 1270 * very large transfers. 1271 */ 1272 tp->snd_una = tp->snd_recover = th->th_ack; 1273 m_freem(m); 1274 1275 /* 1276 * If all outstanding data are acked, stop 1277 * retransmit timer, otherwise restart timer 1278 * using current (possibly backed-off) value. 1279 * If process is waiting for space, 1280 * wakeup/selwakeup/signal. If data 1281 * are ready to send, let tcp_output 1282 * decide between more output or persist. 1283 */ 1284 if (tp->snd_una == tp->snd_max) 1285 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1286 else if (TCP_TIMER_ISARMED(tp, 1287 TCPT_PERSIST) == 0) 1288 TCP_TIMER_ARM(tp, TCPT_REXMT, 1289 tp->t_rxtcur); 1290 1291 sowwakeup(so); 1292 if (so->so_snd.sb_cc) 1293 (void) tcp_output(tp); 1294 if (tcp_saveti) 1295 m_freem(tcp_saveti); 1296 return; 1297 } 1298 } else if (th->th_ack == tp->snd_una && 1299 tp->segq.lh_first == NULL && 1300 tlen <= sbspace(&so->so_rcv)) { 1301 /* 1302 * this is a pure, in-sequence data packet 1303 * with nothing on the reassembly queue and 1304 * we have enough buffer space to take it. 1305 */ 1306 ++tcpstat.tcps_preddat; 1307 tp->rcv_nxt += tlen; 1308 tcpstat.tcps_rcvpack++; 1309 tcpstat.tcps_rcvbyte += tlen; 1310 ND6_HINT(tp); 1311 /* 1312 * Drop TCP, IP headers and TCP options then add data 1313 * to socket buffer. 1314 */ 1315 m_adj(m, toff + off); 1316 sbappend(&so->so_rcv, m); 1317 sorwakeup(so); 1318 TCP_SETUP_ACK(tp, th); 1319 if (tp->t_flags & TF_ACKNOW) 1320 (void) tcp_output(tp); 1321 if (tcp_saveti) 1322 m_freem(tcp_saveti); 1323 return; 1324 } 1325 } 1326 1327 /* 1328 * Compute mbuf offset to TCP data segment. 1329 */ 1330 hdroptlen = toff + off; 1331 1332 /* 1333 * Calculate amount of space in receive window, 1334 * and then do TCP input processing. 1335 * Receive window is amount of space in rcv queue, 1336 * but not less than advertised window. 1337 */ 1338 { int win; 1339 1340 win = sbspace(&so->so_rcv); 1341 if (win < 0) 1342 win = 0; 1343 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1344 } 1345 1346 switch (tp->t_state) { 1347 1348 /* 1349 * If the state is SYN_SENT: 1350 * if seg contains an ACK, but not for our SYN, drop the input. 1351 * if seg contains a RST, then drop the connection. 1352 * if seg does not contain SYN, then drop it. 1353 * Otherwise this is an acceptable SYN segment 1354 * initialize tp->rcv_nxt and tp->irs 1355 * if seg contains ack then advance tp->snd_una 1356 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1357 * arrange for segment to be acked (eventually) 1358 * continue processing rest of data/controls, beginning with URG 1359 */ 1360 case TCPS_SYN_SENT: 1361 if ((tiflags & TH_ACK) && 1362 (SEQ_LEQ(th->th_ack, tp->iss) || 1363 SEQ_GT(th->th_ack, tp->snd_max))) 1364 goto dropwithreset; 1365 if (tiflags & TH_RST) { 1366 if (tiflags & TH_ACK) 1367 tp = tcp_drop(tp, ECONNREFUSED); 1368 goto drop; 1369 } 1370 if ((tiflags & TH_SYN) == 0) 1371 goto drop; 1372 if (tiflags & TH_ACK) { 1373 tp->snd_una = tp->snd_recover = th->th_ack; 1374 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1375 tp->snd_nxt = tp->snd_una; 1376 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1377 } 1378 tp->irs = th->th_seq; 1379 tcp_rcvseqinit(tp); 1380 tp->t_flags |= TF_ACKNOW; 1381 tcp_mss_from_peer(tp, opti.maxseg); 1382 1383 /* 1384 * Initialize the initial congestion window. If we 1385 * had to retransmit the SYN, we must initialize cwnd 1386 * to 1 segment (i.e. the Loss Window). 1387 */ 1388 if (tp->t_flags & TF_SYN_REXMT) 1389 tp->snd_cwnd = tp->t_peermss; 1390 else 1391 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1392 tp->t_peermss); 1393 1394 tcp_rmx_rtt(tp); 1395 if (tiflags & TH_ACK) { 1396 tcpstat.tcps_connects++; 1397 soisconnected(so); 1398 tcp_established(tp); 1399 /* Do window scaling on this connection? */ 1400 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1401 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1402 tp->snd_scale = tp->requested_s_scale; 1403 tp->rcv_scale = tp->request_r_scale; 1404 } 1405 TCP_REASS_LOCK(tp); 1406 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1407 TCP_REASS_UNLOCK(tp); 1408 /* 1409 * if we didn't have to retransmit the SYN, 1410 * use its rtt as our initial srtt & rtt var. 1411 */ 1412 if (tp->t_rtt) 1413 tcp_xmit_timer(tp, tp->t_rtt); 1414 } else 1415 tp->t_state = TCPS_SYN_RECEIVED; 1416 1417 /* 1418 * Advance th->th_seq to correspond to first data byte. 1419 * If data, trim to stay within window, 1420 * dropping FIN if necessary. 1421 */ 1422 th->th_seq++; 1423 if (tlen > tp->rcv_wnd) { 1424 todrop = tlen - tp->rcv_wnd; 1425 m_adj(m, -todrop); 1426 tlen = tp->rcv_wnd; 1427 tiflags &= ~TH_FIN; 1428 tcpstat.tcps_rcvpackafterwin++; 1429 tcpstat.tcps_rcvbyteafterwin += todrop; 1430 } 1431 tp->snd_wl1 = th->th_seq - 1; 1432 tp->rcv_up = th->th_seq; 1433 goto step6; 1434 1435 /* 1436 * If the state is SYN_RECEIVED: 1437 * If seg contains an ACK, but not for our SYN, drop the input 1438 * and generate an RST. See page 36, rfc793 1439 */ 1440 case TCPS_SYN_RECEIVED: 1441 if ((tiflags & TH_ACK) && 1442 (SEQ_LEQ(th->th_ack, tp->iss) || 1443 SEQ_GT(th->th_ack, tp->snd_max))) 1444 goto dropwithreset; 1445 break; 1446 } 1447 1448 /* 1449 * States other than LISTEN or SYN_SENT. 1450 * First check timestamp, if present. 1451 * Then check that at least some bytes of segment are within 1452 * receive window. If segment begins before rcv_nxt, 1453 * drop leading data (and SYN); if nothing left, just ack. 1454 * 1455 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1456 * and it's less than ts_recent, drop it. 1457 */ 1458 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1459 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1460 1461 /* Check to see if ts_recent is over 24 days old. */ 1462 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1463 /* 1464 * Invalidate ts_recent. If this segment updates 1465 * ts_recent, the age will be reset later and ts_recent 1466 * will get a valid value. If it does not, setting 1467 * ts_recent to zero will at least satisfy the 1468 * requirement that zero be placed in the timestamp 1469 * echo reply when ts_recent isn't valid. The 1470 * age isn't reset until we get a valid ts_recent 1471 * because we don't want out-of-order segments to be 1472 * dropped when ts_recent is old. 1473 */ 1474 tp->ts_recent = 0; 1475 } else { 1476 tcpstat.tcps_rcvduppack++; 1477 tcpstat.tcps_rcvdupbyte += tlen; 1478 tcpstat.tcps_pawsdrop++; 1479 goto dropafterack; 1480 } 1481 } 1482 1483 todrop = tp->rcv_nxt - th->th_seq; 1484 if (todrop > 0) { 1485 if (tiflags & TH_SYN) { 1486 tiflags &= ~TH_SYN; 1487 th->th_seq++; 1488 if (th->th_urp > 1) 1489 th->th_urp--; 1490 else { 1491 tiflags &= ~TH_URG; 1492 th->th_urp = 0; 1493 } 1494 todrop--; 1495 } 1496 if (todrop > tlen || 1497 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1498 /* 1499 * Any valid FIN must be to the left of the window. 1500 * At this point the FIN must be a duplicate or 1501 * out of sequence; drop it. 1502 */ 1503 tiflags &= ~TH_FIN; 1504 /* 1505 * Send an ACK to resynchronize and drop any data. 1506 * But keep on processing for RST or ACK. 1507 */ 1508 tp->t_flags |= TF_ACKNOW; 1509 todrop = tlen; 1510 tcpstat.tcps_rcvdupbyte += todrop; 1511 tcpstat.tcps_rcvduppack++; 1512 } else { 1513 tcpstat.tcps_rcvpartduppack++; 1514 tcpstat.tcps_rcvpartdupbyte += todrop; 1515 } 1516 hdroptlen += todrop; /*drop from head afterwards*/ 1517 th->th_seq += todrop; 1518 tlen -= todrop; 1519 if (th->th_urp > todrop) 1520 th->th_urp -= todrop; 1521 else { 1522 tiflags &= ~TH_URG; 1523 th->th_urp = 0; 1524 } 1525 } 1526 1527 /* 1528 * If new data are received on a connection after the 1529 * user processes are gone, then RST the other end. 1530 */ 1531 if ((so->so_state & SS_NOFDREF) && 1532 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1533 tp = tcp_close(tp); 1534 tcpstat.tcps_rcvafterclose++; 1535 goto dropwithreset; 1536 } 1537 1538 /* 1539 * If segment ends after window, drop trailing data 1540 * (and PUSH and FIN); if nothing left, just ACK. 1541 */ 1542 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1543 if (todrop > 0) { 1544 tcpstat.tcps_rcvpackafterwin++; 1545 if (todrop >= tlen) { 1546 tcpstat.tcps_rcvbyteafterwin += tlen; 1547 /* 1548 * If a new connection request is received 1549 * while in TIME_WAIT, drop the old connection 1550 * and start over if the sequence numbers 1551 * are above the previous ones. 1552 */ 1553 if (tiflags & TH_SYN && 1554 tp->t_state == TCPS_TIME_WAIT && 1555 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1556 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1557 tp->snd_nxt); 1558 tp = tcp_close(tp); 1559 goto findpcb; 1560 } 1561 /* 1562 * If window is closed can only take segments at 1563 * window edge, and have to drop data and PUSH from 1564 * incoming segments. Continue processing, but 1565 * remember to ack. Otherwise, drop segment 1566 * and ack. 1567 */ 1568 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1569 tp->t_flags |= TF_ACKNOW; 1570 tcpstat.tcps_rcvwinprobe++; 1571 } else 1572 goto dropafterack; 1573 } else 1574 tcpstat.tcps_rcvbyteafterwin += todrop; 1575 m_adj(m, -todrop); 1576 tlen -= todrop; 1577 tiflags &= ~(TH_PUSH|TH_FIN); 1578 } 1579 1580 /* 1581 * If last ACK falls within this segment's sequence numbers, 1582 * and the timestamp is newer, record it. 1583 */ 1584 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1585 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1586 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1587 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1588 tp->ts_recent_age = tcp_now; 1589 tp->ts_recent = opti.ts_val; 1590 } 1591 1592 /* 1593 * If the RST bit is set examine the state: 1594 * SYN_RECEIVED STATE: 1595 * If passive open, return to LISTEN state. 1596 * If active open, inform user that connection was refused. 1597 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1598 * Inform user that connection was reset, and close tcb. 1599 * CLOSING, LAST_ACK, TIME_WAIT STATES 1600 * Close the tcb. 1601 */ 1602 if (tiflags&TH_RST) switch (tp->t_state) { 1603 1604 case TCPS_SYN_RECEIVED: 1605 so->so_error = ECONNREFUSED; 1606 goto close; 1607 1608 case TCPS_ESTABLISHED: 1609 case TCPS_FIN_WAIT_1: 1610 case TCPS_FIN_WAIT_2: 1611 case TCPS_CLOSE_WAIT: 1612 so->so_error = ECONNRESET; 1613 close: 1614 tp->t_state = TCPS_CLOSED; 1615 tcpstat.tcps_drops++; 1616 tp = tcp_close(tp); 1617 goto drop; 1618 1619 case TCPS_CLOSING: 1620 case TCPS_LAST_ACK: 1621 case TCPS_TIME_WAIT: 1622 tp = tcp_close(tp); 1623 goto drop; 1624 } 1625 1626 /* 1627 * If a SYN is in the window, then this is an 1628 * error and we send an RST and drop the connection. 1629 */ 1630 if (tiflags & TH_SYN) { 1631 tp = tcp_drop(tp, ECONNRESET); 1632 goto dropwithreset; 1633 } 1634 1635 /* 1636 * If the ACK bit is off we drop the segment and return. 1637 */ 1638 if ((tiflags & TH_ACK) == 0) { 1639 if (tp->t_flags & TF_ACKNOW) 1640 goto dropafterack; 1641 else 1642 goto drop; 1643 } 1644 1645 /* 1646 * Ack processing. 1647 */ 1648 switch (tp->t_state) { 1649 1650 /* 1651 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1652 * ESTABLISHED state and continue processing, otherwise 1653 * send an RST. 1654 */ 1655 case TCPS_SYN_RECEIVED: 1656 if (SEQ_GT(tp->snd_una, th->th_ack) || 1657 SEQ_GT(th->th_ack, tp->snd_max)) 1658 goto dropwithreset; 1659 tcpstat.tcps_connects++; 1660 soisconnected(so); 1661 tcp_established(tp); 1662 /* Do window scaling? */ 1663 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1664 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1665 tp->snd_scale = tp->requested_s_scale; 1666 tp->rcv_scale = tp->request_r_scale; 1667 } 1668 TCP_REASS_LOCK(tp); 1669 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1670 TCP_REASS_UNLOCK(tp); 1671 tp->snd_wl1 = th->th_seq - 1; 1672 /* fall into ... */ 1673 1674 /* 1675 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1676 * ACKs. If the ack is in the range 1677 * tp->snd_una < th->th_ack <= tp->snd_max 1678 * then advance tp->snd_una to th->th_ack and drop 1679 * data from the retransmission queue. If this ACK reflects 1680 * more up to date window information we update our window information. 1681 */ 1682 case TCPS_ESTABLISHED: 1683 case TCPS_FIN_WAIT_1: 1684 case TCPS_FIN_WAIT_2: 1685 case TCPS_CLOSE_WAIT: 1686 case TCPS_CLOSING: 1687 case TCPS_LAST_ACK: 1688 case TCPS_TIME_WAIT: 1689 1690 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1691 if (tlen == 0 && tiwin == tp->snd_wnd) { 1692 tcpstat.tcps_rcvdupack++; 1693 /* 1694 * If we have outstanding data (other than 1695 * a window probe), this is a completely 1696 * duplicate ack (ie, window info didn't 1697 * change), the ack is the biggest we've 1698 * seen and we've seen exactly our rexmt 1699 * threshhold of them, assume a packet 1700 * has been dropped and retransmit it. 1701 * Kludge snd_nxt & the congestion 1702 * window so we send only this one 1703 * packet. 1704 * 1705 * We know we're losing at the current 1706 * window size so do congestion avoidance 1707 * (set ssthresh to half the current window 1708 * and pull our congestion window back to 1709 * the new ssthresh). 1710 * 1711 * Dup acks mean that packets have left the 1712 * network (they're now cached at the receiver) 1713 * so bump cwnd by the amount in the receiver 1714 * to keep a constant cwnd packets in the 1715 * network. 1716 */ 1717 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1718 th->th_ack != tp->snd_una) 1719 tp->t_dupacks = 0; 1720 else if (++tp->t_dupacks == tcprexmtthresh) { 1721 tcp_seq onxt = tp->snd_nxt; 1722 u_int win = 1723 min(tp->snd_wnd, tp->snd_cwnd) / 1724 2 / tp->t_segsz; 1725 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1726 tp->snd_recover)) { 1727 /* 1728 * False fast retransmit after 1729 * timeout. Do not cut window. 1730 */ 1731 tp->snd_cwnd += tp->t_segsz; 1732 tp->t_dupacks = 0; 1733 (void) tcp_output(tp); 1734 goto drop; 1735 } 1736 1737 if (win < 2) 1738 win = 2; 1739 tp->snd_ssthresh = win * tp->t_segsz; 1740 tp->snd_recover = tp->snd_max; 1741 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1742 tp->t_rtt = 0; 1743 tp->snd_nxt = th->th_ack; 1744 tp->snd_cwnd = tp->t_segsz; 1745 (void) tcp_output(tp); 1746 tp->snd_cwnd = tp->snd_ssthresh + 1747 tp->t_segsz * tp->t_dupacks; 1748 if (SEQ_GT(onxt, tp->snd_nxt)) 1749 tp->snd_nxt = onxt; 1750 goto drop; 1751 } else if (tp->t_dupacks > tcprexmtthresh) { 1752 tp->snd_cwnd += tp->t_segsz; 1753 (void) tcp_output(tp); 1754 goto drop; 1755 } 1756 } else 1757 tp->t_dupacks = 0; 1758 break; 1759 } 1760 /* 1761 * If the congestion window was inflated to account 1762 * for the other side's cached packets, retract it. 1763 */ 1764 if (tcp_do_newreno == 0) { 1765 if (tp->t_dupacks >= tcprexmtthresh && 1766 tp->snd_cwnd > tp->snd_ssthresh) 1767 tp->snd_cwnd = tp->snd_ssthresh; 1768 tp->t_dupacks = 0; 1769 } else if (tp->t_dupacks >= tcprexmtthresh && 1770 tcp_newreno(tp, th) == 0) { 1771 tp->snd_cwnd = tp->snd_ssthresh; 1772 /* 1773 * Window inflation should have left us with approx. 1774 * snd_ssthresh outstanding data. But in case we 1775 * would be inclined to send a burst, better to do 1776 * it via the slow start mechanism. 1777 */ 1778 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1779 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1780 + tp->t_segsz; 1781 tp->t_dupacks = 0; 1782 } 1783 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1784 tcpstat.tcps_rcvacktoomuch++; 1785 goto dropafterack; 1786 } 1787 acked = th->th_ack - tp->snd_una; 1788 tcpstat.tcps_rcvackpack++; 1789 tcpstat.tcps_rcvackbyte += acked; 1790 1791 /* 1792 * If we have a timestamp reply, update smoothed 1793 * round trip time. If no timestamp is present but 1794 * transmit timer is running and timed sequence 1795 * number was acked, update smoothed round trip time. 1796 * Since we now have an rtt measurement, cancel the 1797 * timer backoff (cf., Phil Karn's retransmit alg.). 1798 * Recompute the initial retransmit timer. 1799 */ 1800 if (opti.ts_present && opti.ts_ecr) 1801 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1802 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1803 tcp_xmit_timer(tp,tp->t_rtt); 1804 1805 /* 1806 * If all outstanding data is acked, stop retransmit 1807 * timer and remember to restart (more output or persist). 1808 * If there is more data to be acked, restart retransmit 1809 * timer, using current (possibly backed-off) value. 1810 */ 1811 if (th->th_ack == tp->snd_max) { 1812 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1813 needoutput = 1; 1814 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1815 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1816 /* 1817 * When new data is acked, open the congestion window. 1818 * If the window gives us less than ssthresh packets 1819 * in flight, open exponentially (segsz per packet). 1820 * Otherwise open linearly: segsz per window 1821 * (segsz^2 / cwnd per packet), plus a constant 1822 * fraction of a packet (segsz/8) to help larger windows 1823 * open quickly enough. 1824 */ 1825 { 1826 u_int cw = tp->snd_cwnd; 1827 u_int incr = tp->t_segsz; 1828 1829 if (cw > tp->snd_ssthresh) 1830 incr = incr * incr / cw; 1831 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1832 tp->snd_cwnd = min(cw + incr, 1833 TCP_MAXWIN << tp->snd_scale); 1834 } 1835 ND6_HINT(tp); 1836 if (acked > so->so_snd.sb_cc) { 1837 tp->snd_wnd -= so->so_snd.sb_cc; 1838 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1839 ourfinisacked = 1; 1840 } else { 1841 sbdrop(&so->so_snd, acked); 1842 tp->snd_wnd -= acked; 1843 ourfinisacked = 0; 1844 } 1845 sowwakeup(so); 1846 /* 1847 * We want snd_recover to track snd_una to 1848 * avoid sequence wraparound problems for 1849 * very large transfers. 1850 */ 1851 tp->snd_una = tp->snd_recover = th->th_ack; 1852 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1853 tp->snd_nxt = tp->snd_una; 1854 1855 switch (tp->t_state) { 1856 1857 /* 1858 * In FIN_WAIT_1 STATE in addition to the processing 1859 * for the ESTABLISHED state if our FIN is now acknowledged 1860 * then enter FIN_WAIT_2. 1861 */ 1862 case TCPS_FIN_WAIT_1: 1863 if (ourfinisacked) { 1864 /* 1865 * If we can't receive any more 1866 * data, then closing user can proceed. 1867 * Starting the timer is contrary to the 1868 * specification, but if we don't get a FIN 1869 * we'll hang forever. 1870 */ 1871 if (so->so_state & SS_CANTRCVMORE) { 1872 soisdisconnected(so); 1873 if (tcp_maxidle > 0) 1874 TCP_TIMER_ARM(tp, TCPT_2MSL, 1875 tcp_maxidle); 1876 } 1877 tp->t_state = TCPS_FIN_WAIT_2; 1878 } 1879 break; 1880 1881 /* 1882 * In CLOSING STATE in addition to the processing for 1883 * the ESTABLISHED state if the ACK acknowledges our FIN 1884 * then enter the TIME-WAIT state, otherwise ignore 1885 * the segment. 1886 */ 1887 case TCPS_CLOSING: 1888 if (ourfinisacked) { 1889 tp->t_state = TCPS_TIME_WAIT; 1890 tcp_canceltimers(tp); 1891 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1892 soisdisconnected(so); 1893 } 1894 break; 1895 1896 /* 1897 * In LAST_ACK, we may still be waiting for data to drain 1898 * and/or to be acked, as well as for the ack of our FIN. 1899 * If our FIN is now acknowledged, delete the TCB, 1900 * enter the closed state and return. 1901 */ 1902 case TCPS_LAST_ACK: 1903 if (ourfinisacked) { 1904 tp = tcp_close(tp); 1905 goto drop; 1906 } 1907 break; 1908 1909 /* 1910 * In TIME_WAIT state the only thing that should arrive 1911 * is a retransmission of the remote FIN. Acknowledge 1912 * it and restart the finack timer. 1913 */ 1914 case TCPS_TIME_WAIT: 1915 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1916 goto dropafterack; 1917 } 1918 } 1919 1920 step6: 1921 /* 1922 * Update window information. 1923 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1924 */ 1925 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1926 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1927 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1928 /* keep track of pure window updates */ 1929 if (tlen == 0 && 1930 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1931 tcpstat.tcps_rcvwinupd++; 1932 tp->snd_wnd = tiwin; 1933 tp->snd_wl1 = th->th_seq; 1934 tp->snd_wl2 = th->th_ack; 1935 if (tp->snd_wnd > tp->max_sndwnd) 1936 tp->max_sndwnd = tp->snd_wnd; 1937 needoutput = 1; 1938 } 1939 1940 /* 1941 * Process segments with URG. 1942 */ 1943 if ((tiflags & TH_URG) && th->th_urp && 1944 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1945 /* 1946 * This is a kludge, but if we receive and accept 1947 * random urgent pointers, we'll crash in 1948 * soreceive. It's hard to imagine someone 1949 * actually wanting to send this much urgent data. 1950 */ 1951 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1952 th->th_urp = 0; /* XXX */ 1953 tiflags &= ~TH_URG; /* XXX */ 1954 goto dodata; /* XXX */ 1955 } 1956 /* 1957 * If this segment advances the known urgent pointer, 1958 * then mark the data stream. This should not happen 1959 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1960 * a FIN has been received from the remote side. 1961 * In these states we ignore the URG. 1962 * 1963 * According to RFC961 (Assigned Protocols), 1964 * the urgent pointer points to the last octet 1965 * of urgent data. We continue, however, 1966 * to consider it to indicate the first octet 1967 * of data past the urgent section as the original 1968 * spec states (in one of two places). 1969 */ 1970 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1971 tp->rcv_up = th->th_seq + th->th_urp; 1972 so->so_oobmark = so->so_rcv.sb_cc + 1973 (tp->rcv_up - tp->rcv_nxt) - 1; 1974 if (so->so_oobmark == 0) 1975 so->so_state |= SS_RCVATMARK; 1976 sohasoutofband(so); 1977 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1978 } 1979 /* 1980 * Remove out of band data so doesn't get presented to user. 1981 * This can happen independent of advancing the URG pointer, 1982 * but if two URG's are pending at once, some out-of-band 1983 * data may creep in... ick. 1984 */ 1985 if (th->th_urp <= (u_int16_t) tlen 1986 #ifdef SO_OOBINLINE 1987 && (so->so_options & SO_OOBINLINE) == 0 1988 #endif 1989 ) 1990 tcp_pulloutofband(so, th, m, hdroptlen); 1991 } else 1992 /* 1993 * If no out of band data is expected, 1994 * pull receive urgent pointer along 1995 * with the receive window. 1996 */ 1997 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1998 tp->rcv_up = tp->rcv_nxt; 1999 dodata: /* XXX */ 2000 2001 /* 2002 * Process the segment text, merging it into the TCP sequencing queue, 2003 * and arranging for acknowledgement of receipt if necessary. 2004 * This process logically involves adjusting tp->rcv_wnd as data 2005 * is presented to the user (this happens in tcp_usrreq.c, 2006 * case PRU_RCVD). If a FIN has already been received on this 2007 * connection then we just ignore the text. 2008 */ 2009 if ((tlen || (tiflags & TH_FIN)) && 2010 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2011 /* 2012 * Insert segment ti into reassembly queue of tcp with 2013 * control block tp. Return TH_FIN if reassembly now includes 2014 * a segment with FIN. The macro form does the common case 2015 * inline (segment is the next to be received on an 2016 * established connection, and the queue is empty), 2017 * avoiding linkage into and removal from the queue and 2018 * repetition of various conversions. 2019 * Set DELACK for segments received in order, but ack 2020 * immediately when segments are out of order 2021 * (so fast retransmit can work). 2022 */ 2023 /* NOTE: this was TCP_REASS() macro, but used only once */ 2024 TCP_REASS_LOCK(tp); 2025 if (th->th_seq == tp->rcv_nxt && 2026 tp->segq.lh_first == NULL && 2027 tp->t_state == TCPS_ESTABLISHED) { 2028 TCP_SETUP_ACK(tp, th); 2029 tp->rcv_nxt += tlen; 2030 tiflags = th->th_flags & TH_FIN; 2031 tcpstat.tcps_rcvpack++; 2032 tcpstat.tcps_rcvbyte += tlen; 2033 ND6_HINT(tp); 2034 m_adj(m, hdroptlen); 2035 sbappend(&(so)->so_rcv, m); 2036 sorwakeup(so); 2037 } else { 2038 m_adj(m, hdroptlen); 2039 tiflags = tcp_reass(tp, th, m, &tlen); 2040 tp->t_flags |= TF_ACKNOW; 2041 } 2042 TCP_REASS_UNLOCK(tp); 2043 2044 /* 2045 * Note the amount of data that peer has sent into 2046 * our window, in order to estimate the sender's 2047 * buffer size. 2048 */ 2049 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2050 } else { 2051 m_freem(m); 2052 m = NULL; 2053 tiflags &= ~TH_FIN; 2054 } 2055 2056 /* 2057 * If FIN is received ACK the FIN and let the user know 2058 * that the connection is closing. Ignore a FIN received before 2059 * the connection is fully established. 2060 */ 2061 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2062 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2063 socantrcvmore(so); 2064 tp->t_flags |= TF_ACKNOW; 2065 tp->rcv_nxt++; 2066 } 2067 switch (tp->t_state) { 2068 2069 /* 2070 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2071 */ 2072 case TCPS_ESTABLISHED: 2073 tp->t_state = TCPS_CLOSE_WAIT; 2074 break; 2075 2076 /* 2077 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2078 * enter the CLOSING state. 2079 */ 2080 case TCPS_FIN_WAIT_1: 2081 tp->t_state = TCPS_CLOSING; 2082 break; 2083 2084 /* 2085 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2086 * starting the time-wait timer, turning off the other 2087 * standard timers. 2088 */ 2089 case TCPS_FIN_WAIT_2: 2090 tp->t_state = TCPS_TIME_WAIT; 2091 tcp_canceltimers(tp); 2092 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2093 soisdisconnected(so); 2094 break; 2095 2096 /* 2097 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2098 */ 2099 case TCPS_TIME_WAIT: 2100 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2101 break; 2102 } 2103 } 2104 if (so->so_options & SO_DEBUG) { 2105 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2106 } 2107 2108 /* 2109 * Return any desired output. 2110 */ 2111 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2112 (void) tcp_output(tp); 2113 if (tcp_saveti) 2114 m_freem(tcp_saveti); 2115 return; 2116 2117 badsyn: 2118 /* 2119 * Received a bad SYN. Increment counters and dropwithreset. 2120 */ 2121 tcpstat.tcps_badsyn++; 2122 tp = NULL; 2123 goto dropwithreset; 2124 2125 dropafterack: 2126 /* 2127 * Generate an ACK dropping incoming segment if it occupies 2128 * sequence space, where the ACK reflects our state. 2129 */ 2130 if (tiflags & TH_RST) 2131 goto drop; 2132 m_freem(m); 2133 tp->t_flags |= TF_ACKNOW; 2134 (void) tcp_output(tp); 2135 if (tcp_saveti) 2136 m_freem(tcp_saveti); 2137 return; 2138 2139 dropwithreset_ratelim: 2140 /* 2141 * We may want to rate-limit RSTs in certain situations, 2142 * particularly if we are sending an RST in response to 2143 * an attempt to connect to or otherwise communicate with 2144 * a port for which we have no socket. 2145 */ 2146 if (ratecheck(&tcp_rst_ratelim_last, &tcp_rst_ratelim) == 0) { 2147 /* XXX stat */ 2148 goto drop; 2149 } 2150 /* ...fall into dropwithreset... */ 2151 2152 dropwithreset: 2153 /* 2154 * Generate a RST, dropping incoming segment. 2155 * Make ACK acceptable to originator of segment. 2156 */ 2157 if (tiflags & TH_RST) 2158 goto drop; 2159 { 2160 /* 2161 * need to recover version # field, which was overwritten on 2162 * ip_cksum computation. 2163 */ 2164 struct ip *sip; 2165 sip = mtod(m, struct ip *); 2166 switch (af) { 2167 case AF_INET: 2168 sip->ip_v = 4; 2169 break; 2170 #ifdef INET6 2171 case AF_INET6: 2172 sip->ip_v = 6; 2173 break; 2174 #endif 2175 } 2176 } 2177 if (tiflags & TH_ACK) 2178 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2179 else { 2180 if (tiflags & TH_SYN) 2181 tlen++; 2182 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2183 TH_RST|TH_ACK); 2184 } 2185 if (tcp_saveti) 2186 m_freem(tcp_saveti); 2187 return; 2188 2189 drop: 2190 /* 2191 * Drop space held by incoming segment and return. 2192 */ 2193 if (tp) { 2194 if (tp->t_inpcb) 2195 so = tp->t_inpcb->inp_socket; 2196 #ifdef INET6 2197 else if (tp->t_in6pcb) 2198 so = tp->t_in6pcb->in6p_socket; 2199 #endif 2200 else 2201 so = NULL; 2202 if (so && (so->so_options & SO_DEBUG) != 0) 2203 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2204 } 2205 if (tcp_saveti) 2206 m_freem(tcp_saveti); 2207 m_freem(m); 2208 return; 2209 } 2210 2211 void 2212 tcp_dooptions(tp, cp, cnt, th, oi) 2213 struct tcpcb *tp; 2214 u_char *cp; 2215 int cnt; 2216 struct tcphdr *th; 2217 struct tcp_opt_info *oi; 2218 { 2219 u_int16_t mss; 2220 int opt, optlen; 2221 2222 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2223 opt = cp[0]; 2224 if (opt == TCPOPT_EOL) 2225 break; 2226 if (opt == TCPOPT_NOP) 2227 optlen = 1; 2228 else { 2229 if (cnt < 2) 2230 break; 2231 optlen = cp[1]; 2232 if (optlen < 2 || optlen > cnt) 2233 break; 2234 } 2235 switch (opt) { 2236 2237 default: 2238 continue; 2239 2240 case TCPOPT_MAXSEG: 2241 if (optlen != TCPOLEN_MAXSEG) 2242 continue; 2243 if (!(th->th_flags & TH_SYN)) 2244 continue; 2245 bcopy(cp + 2, &mss, sizeof(mss)); 2246 oi->maxseg = ntohs(mss); 2247 break; 2248 2249 case TCPOPT_WINDOW: 2250 if (optlen != TCPOLEN_WINDOW) 2251 continue; 2252 if (!(th->th_flags & TH_SYN)) 2253 continue; 2254 tp->t_flags |= TF_RCVD_SCALE; 2255 tp->requested_s_scale = cp[2]; 2256 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2257 #if 0 /*XXX*/ 2258 char *p; 2259 2260 if (ip) 2261 p = ntohl(ip->ip_src); 2262 #ifdef INET6 2263 else if (ip6) 2264 p = ip6_sprintf(&ip6->ip6_src); 2265 #endif 2266 else 2267 p = "(unknown)"; 2268 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2269 "assuming %d\n", 2270 tp->requested_s_scale, p, 2271 TCP_MAX_WINSHIFT); 2272 #else 2273 log(LOG_ERR, "TCP: invalid wscale %d, " 2274 "assuming %d\n", 2275 tp->requested_s_scale, 2276 TCP_MAX_WINSHIFT); 2277 #endif 2278 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2279 } 2280 break; 2281 2282 case TCPOPT_TIMESTAMP: 2283 if (optlen != TCPOLEN_TIMESTAMP) 2284 continue; 2285 oi->ts_present = 1; 2286 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2287 NTOHL(oi->ts_val); 2288 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2289 NTOHL(oi->ts_ecr); 2290 2291 /* 2292 * A timestamp received in a SYN makes 2293 * it ok to send timestamp requests and replies. 2294 */ 2295 if (th->th_flags & TH_SYN) { 2296 tp->t_flags |= TF_RCVD_TSTMP; 2297 tp->ts_recent = oi->ts_val; 2298 tp->ts_recent_age = tcp_now; 2299 } 2300 break; 2301 case TCPOPT_SACK_PERMITTED: 2302 if (optlen != TCPOLEN_SACK_PERMITTED) 2303 continue; 2304 if (!(th->th_flags & TH_SYN)) 2305 continue; 2306 tp->t_flags &= ~TF_CANT_TXSACK; 2307 break; 2308 2309 case TCPOPT_SACK: 2310 if (tp->t_flags & TF_IGNR_RXSACK) 2311 continue; 2312 if (optlen % 8 != 2 || optlen < 10) 2313 continue; 2314 cp += 2; 2315 optlen -= 2; 2316 for (; optlen > 0; cp -= 8, optlen -= 8) { 2317 tcp_seq lwe, rwe; 2318 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2319 NTOHL(lwe); 2320 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2321 NTOHL(rwe); 2322 /* tcp_mark_sacked(tp, lwe, rwe); */ 2323 } 2324 break; 2325 } 2326 } 2327 } 2328 2329 /* 2330 * Pull out of band byte out of a segment so 2331 * it doesn't appear in the user's data queue. 2332 * It is still reflected in the segment length for 2333 * sequencing purposes. 2334 */ 2335 void 2336 tcp_pulloutofband(so, th, m, off) 2337 struct socket *so; 2338 struct tcphdr *th; 2339 struct mbuf *m; 2340 int off; 2341 { 2342 int cnt = off + th->th_urp - 1; 2343 2344 while (cnt >= 0) { 2345 if (m->m_len > cnt) { 2346 char *cp = mtod(m, caddr_t) + cnt; 2347 struct tcpcb *tp = sototcpcb(so); 2348 2349 tp->t_iobc = *cp; 2350 tp->t_oobflags |= TCPOOB_HAVEDATA; 2351 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2352 m->m_len--; 2353 return; 2354 } 2355 cnt -= m->m_len; 2356 m = m->m_next; 2357 if (m == 0) 2358 break; 2359 } 2360 panic("tcp_pulloutofband"); 2361 } 2362 2363 /* 2364 * Collect new round-trip time estimate 2365 * and update averages and current timeout. 2366 */ 2367 void 2368 tcp_xmit_timer(tp, rtt) 2369 struct tcpcb *tp; 2370 short rtt; 2371 { 2372 short delta; 2373 short rttmin; 2374 2375 tcpstat.tcps_rttupdated++; 2376 --rtt; 2377 if (tp->t_srtt != 0) { 2378 /* 2379 * srtt is stored as fixed point with 3 bits after the 2380 * binary point (i.e., scaled by 8). The following magic 2381 * is equivalent to the smoothing algorithm in rfc793 with 2382 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2383 * point). Adjust rtt to origin 0. 2384 */ 2385 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2386 if ((tp->t_srtt += delta) <= 0) 2387 tp->t_srtt = 1 << 2; 2388 /* 2389 * We accumulate a smoothed rtt variance (actually, a 2390 * smoothed mean difference), then set the retransmit 2391 * timer to smoothed rtt + 4 times the smoothed variance. 2392 * rttvar is stored as fixed point with 2 bits after the 2393 * binary point (scaled by 4). The following is 2394 * equivalent to rfc793 smoothing with an alpha of .75 2395 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2396 * rfc793's wired-in beta. 2397 */ 2398 if (delta < 0) 2399 delta = -delta; 2400 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2401 if ((tp->t_rttvar += delta) <= 0) 2402 tp->t_rttvar = 1 << 2; 2403 } else { 2404 /* 2405 * No rtt measurement yet - use the unsmoothed rtt. 2406 * Set the variance to half the rtt (so our first 2407 * retransmit happens at 3*rtt). 2408 */ 2409 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2410 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2411 } 2412 tp->t_rtt = 0; 2413 tp->t_rxtshift = 0; 2414 2415 /* 2416 * the retransmit should happen at rtt + 4 * rttvar. 2417 * Because of the way we do the smoothing, srtt and rttvar 2418 * will each average +1/2 tick of bias. When we compute 2419 * the retransmit timer, we want 1/2 tick of rounding and 2420 * 1 extra tick because of +-1/2 tick uncertainty in the 2421 * firing of the timer. The bias will give us exactly the 2422 * 1.5 tick we need. But, because the bias is 2423 * statistical, we have to test that we don't drop below 2424 * the minimum feasible timer (which is 2 ticks). 2425 */ 2426 if (tp->t_rttmin > rtt + 2) 2427 rttmin = tp->t_rttmin; 2428 else 2429 rttmin = rtt + 2; 2430 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2431 2432 /* 2433 * We received an ack for a packet that wasn't retransmitted; 2434 * it is probably safe to discard any error indications we've 2435 * received recently. This isn't quite right, but close enough 2436 * for now (a route might have failed after we sent a segment, 2437 * and the return path might not be symmetrical). 2438 */ 2439 tp->t_softerror = 0; 2440 } 2441 2442 /* 2443 * Checks for partial ack. If partial ack arrives, force the retransmission 2444 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2445 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2446 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2447 */ 2448 int 2449 tcp_newreno(tp, th) 2450 struct tcpcb *tp; 2451 struct tcphdr *th; 2452 { 2453 tcp_seq onxt = tp->snd_nxt; 2454 u_long ocwnd = tp->snd_cwnd; 2455 2456 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2457 /* 2458 * snd_una has not yet been updated and the socket's send 2459 * buffer has not yet drained off the ACK'd data, so we 2460 * have to leave snd_una as it was to get the correct data 2461 * offset in tcp_output(). 2462 */ 2463 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2464 tp->t_rtt = 0; 2465 tp->snd_nxt = th->th_ack; 2466 /* 2467 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2468 * is not yet updated when we're called. 2469 */ 2470 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2471 (void) tcp_output(tp); 2472 tp->snd_cwnd = ocwnd; 2473 if (SEQ_GT(onxt, tp->snd_nxt)) 2474 tp->snd_nxt = onxt; 2475 /* 2476 * Partial window deflation. Relies on fact that tp->snd_una 2477 * not updated yet. 2478 */ 2479 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2480 return 1; 2481 } 2482 return 0; 2483 } 2484 2485 2486 /* 2487 * TCP compressed state engine. Currently used to hold compressed 2488 * state for SYN_RECEIVED. 2489 */ 2490 2491 u_long syn_cache_count; 2492 u_int32_t syn_hash1, syn_hash2; 2493 2494 #define SYN_HASH(sa, sp, dp) \ 2495 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2496 ((u_int32_t)(sp)))^syn_hash2))) 2497 #ifndef INET6 2498 #define SYN_HASHALL(hash, src, dst) \ 2499 do { \ 2500 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2501 ((struct sockaddr_in *)(src))->sin_port, \ 2502 ((struct sockaddr_in *)(dst))->sin_port); \ 2503 } while (0) 2504 #else 2505 #define SYN_HASH6(sa, sp, dp) \ 2506 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2507 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2508 & 0x7fffffff) 2509 2510 #define SYN_HASHALL(hash, src, dst) \ 2511 do { \ 2512 switch ((src)->sa_family) { \ 2513 case AF_INET: \ 2514 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2515 ((struct sockaddr_in *)(src))->sin_port, \ 2516 ((struct sockaddr_in *)(dst))->sin_port); \ 2517 break; \ 2518 case AF_INET6: \ 2519 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2520 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2521 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2522 break; \ 2523 default: \ 2524 hash = 0; \ 2525 } \ 2526 } while (0) 2527 #endif /* INET6 */ 2528 2529 #define SYN_CACHE_RM(sc) \ 2530 do { \ 2531 LIST_REMOVE((sc), sc_bucketq); \ 2532 (sc)->sc_tp = NULL; \ 2533 LIST_REMOVE((sc), sc_tpq); \ 2534 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2535 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2536 syn_cache_count--; \ 2537 } while (0) 2538 2539 #define SYN_CACHE_PUT(sc) \ 2540 do { \ 2541 if ((sc)->sc_ipopts) \ 2542 (void) m_free((sc)->sc_ipopts); \ 2543 if ((sc)->sc_route4.ro_rt != NULL) \ 2544 RTFREE((sc)->sc_route4.ro_rt); \ 2545 pool_put(&syn_cache_pool, (sc)); \ 2546 } while (0) 2547 2548 struct pool syn_cache_pool; 2549 2550 /* 2551 * We don't estimate RTT with SYNs, so each packet starts with the default 2552 * RTT and each timer queue has a fixed timeout value. This allows us to 2553 * optimize the timer queues somewhat. 2554 */ 2555 #define SYN_CACHE_TIMER_ARM(sc) \ 2556 do { \ 2557 TCPT_RANGESET((sc)->sc_rxtcur, \ 2558 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2559 TCPTV_REXMTMAX); \ 2560 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2561 } while (0) 2562 2563 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2564 2565 void 2566 syn_cache_init() 2567 { 2568 int i; 2569 2570 /* Initialize the hash buckets. */ 2571 for (i = 0; i < tcp_syn_cache_size; i++) 2572 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2573 2574 /* Initialize the timer queues. */ 2575 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2576 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2577 2578 /* Initialize the syn cache pool. */ 2579 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2580 "synpl", 0, NULL, NULL, M_PCB); 2581 } 2582 2583 void 2584 syn_cache_insert(sc, tp) 2585 struct syn_cache *sc; 2586 struct tcpcb *tp; 2587 { 2588 struct syn_cache_head *scp; 2589 struct syn_cache *sc2; 2590 int s, i; 2591 2592 /* 2593 * If there are no entries in the hash table, reinitialize 2594 * the hash secrets. 2595 */ 2596 if (syn_cache_count == 0) { 2597 struct timeval tv; 2598 microtime(&tv); 2599 syn_hash1 = random() ^ (u_long)≻ 2600 syn_hash2 = random() ^ tv.tv_usec; 2601 } 2602 2603 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2604 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2605 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2606 2607 /* 2608 * Make sure that we don't overflow the per-bucket 2609 * limit or the total cache size limit. 2610 */ 2611 s = splsoftnet(); 2612 if (scp->sch_length >= tcp_syn_bucket_limit) { 2613 tcpstat.tcps_sc_bucketoverflow++; 2614 /* 2615 * The bucket is full. Toss the oldest element in the 2616 * bucket. This will be the entry with our bucket 2617 * index closest to the front of the timer queue with 2618 * the largest timeout value. 2619 * 2620 * Note: This timer queue traversal may be expensive, so 2621 * we hope that this doesn't happen very often. It is 2622 * much more likely that we'll overflow the entire 2623 * cache, which is much easier to handle; see below. 2624 */ 2625 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2626 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2627 sc2 != NULL; 2628 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2629 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2630 SYN_CACHE_RM(sc2); 2631 SYN_CACHE_PUT(sc2); 2632 goto insert; /* 2 level break */ 2633 } 2634 } 2635 } 2636 #ifdef DIAGNOSTIC 2637 /* 2638 * This should never happen; we should always find an 2639 * entry in our bucket. 2640 */ 2641 panic("syn_cache_insert: bucketoverflow: impossible"); 2642 #endif 2643 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2644 tcpstat.tcps_sc_overflowed++; 2645 /* 2646 * The cache is full. Toss the oldest entry in the 2647 * entire cache. This is the front entry in the 2648 * first non-empty timer queue with the largest 2649 * timeout value. 2650 */ 2651 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2652 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2653 if (sc2 == NULL) 2654 continue; 2655 SYN_CACHE_RM(sc2); 2656 SYN_CACHE_PUT(sc2); 2657 goto insert; /* symmetry with above */ 2658 } 2659 #ifdef DIAGNOSTIC 2660 /* 2661 * This should never happen; we should always find an 2662 * entry in the cache. 2663 */ 2664 panic("syn_cache_insert: cache overflow: impossible"); 2665 #endif 2666 } 2667 2668 insert: 2669 /* 2670 * Initialize the entry's timer. 2671 */ 2672 sc->sc_rxttot = 0; 2673 sc->sc_rxtshift = 0; 2674 SYN_CACHE_TIMER_ARM(sc); 2675 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2676 2677 /* Link it from tcpcb entry */ 2678 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2679 2680 /* Put it into the bucket. */ 2681 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2682 scp->sch_length++; 2683 syn_cache_count++; 2684 2685 tcpstat.tcps_sc_added++; 2686 splx(s); 2687 } 2688 2689 /* 2690 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2691 * If we have retransmitted an entry the maximum number of times, expire 2692 * that entry. 2693 */ 2694 void 2695 syn_cache_timer() 2696 { 2697 struct syn_cache *sc, *nsc; 2698 int i, s; 2699 2700 s = splsoftnet(); 2701 2702 /* 2703 * First, get all the entries that need to be retransmitted, or 2704 * must be expired due to exceeding the initial keepalive time. 2705 */ 2706 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2707 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2708 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2709 sc = nsc) { 2710 nsc = TAILQ_NEXT(sc, sc_timeq); 2711 2712 /* 2713 * Compute the total amount of time this entry has 2714 * been on a queue. If this entry has been on longer 2715 * than the keep alive timer would allow, expire it. 2716 */ 2717 sc->sc_rxttot += sc->sc_rxtcur; 2718 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2719 tcpstat.tcps_sc_timed_out++; 2720 SYN_CACHE_RM(sc); 2721 SYN_CACHE_PUT(sc); 2722 continue; 2723 } 2724 2725 tcpstat.tcps_sc_retransmitted++; 2726 (void) syn_cache_respond(sc, NULL); 2727 2728 /* Advance this entry onto the next timer queue. */ 2729 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2730 sc->sc_rxtshift = i + 1; 2731 SYN_CACHE_TIMER_ARM(sc); 2732 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2733 sc, sc_timeq); 2734 } 2735 } 2736 2737 /* 2738 * Now get all the entries that are expired due to too many 2739 * retransmissions. 2740 */ 2741 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2742 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2743 sc = nsc) { 2744 nsc = TAILQ_NEXT(sc, sc_timeq); 2745 tcpstat.tcps_sc_timed_out++; 2746 SYN_CACHE_RM(sc); 2747 SYN_CACHE_PUT(sc); 2748 } 2749 splx(s); 2750 } 2751 2752 /* 2753 * Remove syn cache created by the specified tcb entry, 2754 * because this does not make sense to keep them 2755 * (if there's no tcb entry, syn cache entry will never be used) 2756 */ 2757 void 2758 syn_cache_cleanup(tp) 2759 struct tcpcb *tp; 2760 { 2761 struct syn_cache *sc, *nsc; 2762 int s; 2763 2764 s = splsoftnet(); 2765 2766 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2767 nsc = LIST_NEXT(sc, sc_tpq); 2768 2769 #ifdef DIAGNOSTIC 2770 if (sc->sc_tp != tp) 2771 panic("invalid sc_tp in syn_cache_cleanup"); 2772 #endif 2773 SYN_CACHE_RM(sc); 2774 SYN_CACHE_PUT(sc); 2775 } 2776 /* just for safety */ 2777 LIST_INIT(&tp->t_sc); 2778 2779 splx(s); 2780 } 2781 2782 /* 2783 * Find an entry in the syn cache. 2784 */ 2785 struct syn_cache * 2786 syn_cache_lookup(src, dst, headp) 2787 struct sockaddr *src; 2788 struct sockaddr *dst; 2789 struct syn_cache_head **headp; 2790 { 2791 struct syn_cache *sc; 2792 struct syn_cache_head *scp; 2793 u_int32_t hash; 2794 int s; 2795 2796 SYN_HASHALL(hash, src, dst); 2797 2798 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2799 *headp = scp; 2800 s = splsoftnet(); 2801 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2802 sc = LIST_NEXT(sc, sc_bucketq)) { 2803 if (sc->sc_hash != hash) 2804 continue; 2805 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2806 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2807 splx(s); 2808 return (sc); 2809 } 2810 } 2811 splx(s); 2812 return (NULL); 2813 } 2814 2815 /* 2816 * This function gets called when we receive an ACK for a 2817 * socket in the LISTEN state. We look up the connection 2818 * in the syn cache, and if its there, we pull it out of 2819 * the cache and turn it into a full-blown connection in 2820 * the SYN-RECEIVED state. 2821 * 2822 * The return values may not be immediately obvious, and their effects 2823 * can be subtle, so here they are: 2824 * 2825 * NULL SYN was not found in cache; caller should drop the 2826 * packet and send an RST. 2827 * 2828 * -1 We were unable to create the new connection, and are 2829 * aborting it. An ACK,RST is being sent to the peer 2830 * (unless we got screwey sequence numbners; see below), 2831 * because the 3-way handshake has been completed. Caller 2832 * should not free the mbuf, since we may be using it. If 2833 * we are not, we will free it. 2834 * 2835 * Otherwise, the return value is a pointer to the new socket 2836 * associated with the connection. 2837 */ 2838 struct socket * 2839 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2840 struct sockaddr *src; 2841 struct sockaddr *dst; 2842 struct tcphdr *th; 2843 unsigned int hlen, tlen; 2844 struct socket *so; 2845 struct mbuf *m; 2846 { 2847 struct syn_cache *sc; 2848 struct syn_cache_head *scp; 2849 struct inpcb *inp = NULL; 2850 #ifdef INET6 2851 struct in6pcb *in6p = NULL; 2852 #endif 2853 struct tcpcb *tp = 0; 2854 struct mbuf *am; 2855 int s; 2856 struct socket *oso; 2857 2858 s = splsoftnet(); 2859 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2860 splx(s); 2861 return (NULL); 2862 } 2863 2864 /* 2865 * Verify the sequence and ack numbers. Try getting the correct 2866 * response again. 2867 */ 2868 if ((th->th_ack != sc->sc_iss + 1) || 2869 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2870 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2871 (void) syn_cache_respond(sc, m); 2872 splx(s); 2873 return ((struct socket *)(-1)); 2874 } 2875 2876 /* Remove this cache entry */ 2877 SYN_CACHE_RM(sc); 2878 splx(s); 2879 2880 /* 2881 * Ok, create the full blown connection, and set things up 2882 * as they would have been set up if we had created the 2883 * connection when the SYN arrived. If we can't create 2884 * the connection, abort it. 2885 */ 2886 /* 2887 * inp still has the OLD in_pcb stuff, set the 2888 * v6-related flags on the new guy, too. This is 2889 * done particularly for the case where an AF_INET6 2890 * socket is bound only to a port, and a v4 connection 2891 * comes in on that port. 2892 * we also copy the flowinfo from the original pcb 2893 * to the new one. 2894 */ 2895 { 2896 struct inpcb *parentinpcb; 2897 2898 parentinpcb = (struct inpcb *)so->so_pcb; 2899 2900 oso = so; 2901 so = sonewconn(so, SS_ISCONNECTED); 2902 if (so == NULL) 2903 goto resetandabort; 2904 2905 switch (so->so_proto->pr_domain->dom_family) { 2906 case AF_INET: 2907 inp = sotoinpcb(so); 2908 break; 2909 #ifdef INET6 2910 case AF_INET6: 2911 in6p = sotoin6pcb(so); 2912 #if 0 /*def INET6*/ 2913 inp->inp_flags |= (parentinpcb->inp_flags & 2914 (INP_IPV6 | INP_IPV6_UNDEC | INP_IPV6_MAPPED)); 2915 if ((inp->inp_flags & INP_IPV6) && 2916 !(inp->inp_flags & INP_IPV6_MAPPED)) { 2917 inp->inp_ipv6.ip6_hlim = parentinpcb->inp_ipv6.ip6_hlim; 2918 inp->inp_ipv6.ip6_vfc = parentinpcb->inp_ipv6.ip6_vfc; 2919 } 2920 #endif 2921 break; 2922 #endif 2923 } 2924 } 2925 switch (src->sa_family) { 2926 case AF_INET: 2927 if (inp) { 2928 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2929 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2930 inp->inp_options = ip_srcroute(); 2931 in_pcbstate(inp, INP_BOUND); 2932 if (inp->inp_options == NULL) { 2933 inp->inp_options = sc->sc_ipopts; 2934 sc->sc_ipopts = NULL; 2935 } 2936 } 2937 #ifdef INET6 2938 else if (in6p) { 2939 /* IPv4 packet to AF_INET6 socket */ 2940 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2941 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2942 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2943 &in6p->in6p_laddr.s6_addr32[3], 2944 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2945 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2946 in6totcpcb(in6p)->t_family = AF_INET; 2947 } 2948 #endif 2949 break; 2950 #ifdef INET6 2951 case AF_INET6: 2952 if (in6p) { 2953 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 2954 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 2955 #if 0 2956 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 2957 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 2958 #endif 2959 } 2960 break; 2961 #endif 2962 } 2963 #ifdef INET6 2964 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 2965 struct in6pcb *oin6p = sotoin6pcb(oso); 2966 /* inherit socket options from the listening socket */ 2967 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 2968 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 2969 m_freem(in6p->in6p_options); 2970 in6p->in6p_options = 0; 2971 } 2972 ip6_savecontrol(in6p, &in6p->in6p_options, 2973 mtod(m, struct ip6_hdr *), m); 2974 } 2975 #endif 2976 2977 #ifdef IPSEC 2978 /* 2979 * we make a copy of policy, instead of sharing the policy, 2980 * for better behavior in terms of SA lookup and dead SA removal. 2981 */ 2982 if (inp) { 2983 /* copy old policy into new socket's */ 2984 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 2985 printf("tcp_input: could not copy policy\n"); 2986 } 2987 #ifdef INET6 2988 else if (in6p) { 2989 /* copy old policy into new socket's */ 2990 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 2991 printf("tcp_input: could not copy policy\n"); 2992 } 2993 #endif 2994 #endif 2995 2996 /* 2997 * Give the new socket our cached route reference. 2998 */ 2999 if (inp) 3000 inp->inp_route = sc->sc_route4; /* struct assignment */ 3001 #ifdef INET6 3002 else 3003 in6p->in6p_route = sc->sc_route6; 3004 #endif 3005 sc->sc_route4.ro_rt = NULL; 3006 3007 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3008 if (am == NULL) 3009 goto resetandabort; 3010 am->m_len = src->sa_len; 3011 bcopy(src, mtod(am, caddr_t), src->sa_len); 3012 if (inp) { 3013 if (in_pcbconnect(inp, am)) { 3014 (void) m_free(am); 3015 goto resetandabort; 3016 } 3017 } 3018 #ifdef INET6 3019 else if (in6p) { 3020 if (src->sa_family == AF_INET) { 3021 /* IPv4 packet to AF_INET6 socket */ 3022 struct sockaddr_in6 *sin6; 3023 sin6 = mtod(am, struct sockaddr_in6 *); 3024 am->m_len = sizeof(*sin6); 3025 bzero(sin6, sizeof(*sin6)); 3026 sin6->sin6_family = AF_INET6; 3027 sin6->sin6_len = sizeof(*sin6); 3028 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3029 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3030 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3031 &sin6->sin6_addr.s6_addr32[3], 3032 sizeof(sin6->sin6_addr.s6_addr32[3])); 3033 } 3034 if (in6_pcbconnect(in6p, am)) { 3035 (void) m_free(am); 3036 goto resetandabort; 3037 } 3038 } 3039 #endif 3040 else { 3041 (void) m_free(am); 3042 goto resetandabort; 3043 } 3044 (void) m_free(am); 3045 3046 if (inp) 3047 tp = intotcpcb(inp); 3048 #ifdef INET6 3049 else if (in6p) 3050 tp = in6totcpcb(in6p); 3051 #endif 3052 else 3053 tp = NULL; 3054 if (sc->sc_request_r_scale != 15) { 3055 tp->requested_s_scale = sc->sc_requested_s_scale; 3056 tp->request_r_scale = sc->sc_request_r_scale; 3057 tp->snd_scale = sc->sc_requested_s_scale; 3058 tp->rcv_scale = sc->sc_request_r_scale; 3059 tp->t_flags |= TF_RCVD_SCALE; 3060 } 3061 if (sc->sc_flags & SCF_TIMESTAMP) 3062 tp->t_flags |= TF_RCVD_TSTMP; 3063 3064 tp->t_template = tcp_template(tp); 3065 if (tp->t_template == 0) { 3066 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3067 so = NULL; 3068 m_freem(m); 3069 goto abort; 3070 } 3071 3072 tp->iss = sc->sc_iss; 3073 tp->irs = sc->sc_irs; 3074 tcp_sendseqinit(tp); 3075 tcp_rcvseqinit(tp); 3076 tp->t_state = TCPS_SYN_RECEIVED; 3077 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3078 tcpstat.tcps_accepts++; 3079 3080 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3081 tp->t_ourmss = sc->sc_ourmaxseg; 3082 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3083 3084 /* 3085 * Initialize the initial congestion window. If we 3086 * had to retransmit the SYN,ACK, we must initialize cwnd 3087 * to 1 segment (i.e. the Loss Window). 3088 */ 3089 if (sc->sc_rxtshift) 3090 tp->snd_cwnd = tp->t_peermss; 3091 else 3092 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3093 3094 tcp_rmx_rtt(tp); 3095 tp->snd_wl1 = sc->sc_irs; 3096 tp->rcv_up = sc->sc_irs + 1; 3097 3098 /* 3099 * This is what whould have happened in tcp_ouput() when 3100 * the SYN,ACK was sent. 3101 */ 3102 tp->snd_up = tp->snd_una; 3103 tp->snd_max = tp->snd_nxt = tp->iss+1; 3104 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3105 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3106 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3107 tp->last_ack_sent = tp->rcv_nxt; 3108 3109 tcpstat.tcps_sc_completed++; 3110 SYN_CACHE_PUT(sc); 3111 return (so); 3112 3113 resetandabort: 3114 (void) tcp_respond(NULL, m, m, th, 3115 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3116 abort: 3117 if (so != NULL) 3118 (void) soabort(so); 3119 SYN_CACHE_PUT(sc); 3120 tcpstat.tcps_sc_aborted++; 3121 return ((struct socket *)(-1)); 3122 } 3123 3124 /* 3125 * This function is called when we get a RST for a 3126 * non-existant connection, so that we can see if the 3127 * connection is in the syn cache. If it is, zap it. 3128 */ 3129 3130 void 3131 syn_cache_reset(src, dst, th) 3132 struct sockaddr *src; 3133 struct sockaddr *dst; 3134 struct tcphdr *th; 3135 { 3136 struct syn_cache *sc; 3137 struct syn_cache_head *scp; 3138 int s = splsoftnet(); 3139 3140 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3141 splx(s); 3142 return; 3143 } 3144 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3145 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3146 splx(s); 3147 return; 3148 } 3149 SYN_CACHE_RM(sc); 3150 splx(s); 3151 tcpstat.tcps_sc_reset++; 3152 SYN_CACHE_PUT(sc); 3153 } 3154 3155 void 3156 syn_cache_unreach(src, dst, th) 3157 struct sockaddr *src; 3158 struct sockaddr *dst; 3159 struct tcphdr *th; 3160 { 3161 struct syn_cache *sc; 3162 struct syn_cache_head *scp; 3163 int s; 3164 3165 s = splsoftnet(); 3166 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3167 splx(s); 3168 return; 3169 } 3170 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3171 if (ntohl (th->th_seq) != sc->sc_iss) { 3172 splx(s); 3173 return; 3174 } 3175 3176 /* 3177 * If we've rertransmitted 3 times and this is our second error, 3178 * we remove the entry. Otherwise, we allow it to continue on. 3179 * This prevents us from incorrectly nuking an entry during a 3180 * spurious network outage. 3181 * 3182 * See tcp_notify(). 3183 */ 3184 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3185 sc->sc_flags |= SCF_UNREACH; 3186 splx(s); 3187 return; 3188 } 3189 3190 SYN_CACHE_RM(sc); 3191 splx(s); 3192 tcpstat.tcps_sc_unreach++; 3193 SYN_CACHE_PUT(sc); 3194 } 3195 3196 /* 3197 * Given a LISTEN socket and an inbound SYN request, add 3198 * this to the syn cache, and send back a segment: 3199 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3200 * to the source. 3201 * 3202 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3203 * Doing so would require that we hold onto the data and deliver it 3204 * to the application. However, if we are the target of a SYN-flood 3205 * DoS attack, an attacker could send data which would eventually 3206 * consume all available buffer space if it were ACKed. By not ACKing 3207 * the data, we avoid this DoS scenario. 3208 */ 3209 3210 int 3211 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3212 struct sockaddr *src; 3213 struct sockaddr *dst; 3214 struct tcphdr *th; 3215 unsigned int hlen; 3216 struct socket *so; 3217 struct mbuf *m; 3218 u_char *optp; 3219 int optlen; 3220 struct tcp_opt_info *oi; 3221 { 3222 struct tcpcb tb, *tp; 3223 long win; 3224 struct syn_cache *sc; 3225 struct syn_cache_head *scp; 3226 struct mbuf *ipopts; 3227 3228 tp = sototcpcb(so); 3229 3230 /* 3231 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3232 * 3233 * Note this check is performed in tcp_input() very early on. 3234 */ 3235 3236 /* 3237 * Initialize some local state. 3238 */ 3239 win = sbspace(&so->so_rcv); 3240 if (win > TCP_MAXWIN) 3241 win = TCP_MAXWIN; 3242 3243 if (src->sa_family == AF_INET) { 3244 /* 3245 * Remember the IP options, if any. 3246 */ 3247 ipopts = ip_srcroute(); 3248 } else 3249 ipopts = NULL; 3250 3251 if (optp) { 3252 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3253 tcp_dooptions(&tb, optp, optlen, th, oi); 3254 } else 3255 tb.t_flags = 0; 3256 3257 /* 3258 * See if we already have an entry for this connection. 3259 * If we do, resend the SYN,ACK. We do not count this 3260 * as a retransmission (XXX though maybe we should). 3261 */ 3262 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3263 tcpstat.tcps_sc_dupesyn++; 3264 if (ipopts) { 3265 /* 3266 * If we were remembering a previous source route, 3267 * forget it and use the new one we've been given. 3268 */ 3269 if (sc->sc_ipopts) 3270 (void) m_free(sc->sc_ipopts); 3271 sc->sc_ipopts = ipopts; 3272 } 3273 sc->sc_timestamp = tb.ts_recent; 3274 if (syn_cache_respond(sc, m) == 0) { 3275 tcpstat.tcps_sndacks++; 3276 tcpstat.tcps_sndtotal++; 3277 } 3278 return (1); 3279 } 3280 3281 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3282 if (sc == NULL) { 3283 if (ipopts) 3284 (void) m_free(ipopts); 3285 return (0); 3286 } 3287 3288 /* 3289 * Fill in the cache, and put the necessary IP and TCP 3290 * options into the reply. 3291 */ 3292 bzero(sc, sizeof(struct syn_cache)); 3293 bcopy(src, &sc->sc_src, src->sa_len); 3294 bcopy(dst, &sc->sc_dst, dst->sa_len); 3295 sc->sc_flags = 0; 3296 sc->sc_ipopts = ipopts; 3297 sc->sc_irs = th->th_seq; 3298 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 3299 sc->sc_peermaxseg = oi->maxseg; 3300 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3301 m->m_pkthdr.rcvif : NULL, 3302 sc->sc_src.sa.sa_family); 3303 sc->sc_win = win; 3304 sc->sc_timestamp = tb.ts_recent; 3305 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3306 sc->sc_flags |= SCF_TIMESTAMP; 3307 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3308 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3309 sc->sc_requested_s_scale = tb.requested_s_scale; 3310 sc->sc_request_r_scale = 0; 3311 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3312 TCP_MAXWIN << sc->sc_request_r_scale < 3313 so->so_rcv.sb_hiwat) 3314 sc->sc_request_r_scale++; 3315 } else { 3316 sc->sc_requested_s_scale = 15; 3317 sc->sc_request_r_scale = 15; 3318 } 3319 sc->sc_tp = tp; 3320 if (syn_cache_respond(sc, m) == 0) { 3321 syn_cache_insert(sc, tp); 3322 tcpstat.tcps_sndacks++; 3323 tcpstat.tcps_sndtotal++; 3324 } else { 3325 SYN_CACHE_PUT(sc); 3326 tcpstat.tcps_sc_dropped++; 3327 } 3328 return (1); 3329 } 3330 3331 int 3332 syn_cache_respond(sc, m) 3333 struct syn_cache *sc; 3334 struct mbuf *m; 3335 { 3336 struct route *ro; 3337 struct rtentry *rt; 3338 u_int8_t *optp; 3339 int optlen, error; 3340 u_int16_t tlen; 3341 struct ip *ip = NULL; 3342 #ifdef INET6 3343 struct ip6_hdr *ip6 = NULL; 3344 #endif 3345 struct tcphdr *th; 3346 u_int hlen; 3347 3348 switch (sc->sc_src.sa.sa_family) { 3349 case AF_INET: 3350 hlen = sizeof(struct ip); 3351 ro = &sc->sc_route4; 3352 break; 3353 #ifdef INET6 3354 case AF_INET6: 3355 hlen = sizeof(struct ip6_hdr); 3356 ro = (struct route *)&sc->sc_route6; 3357 break; 3358 #endif 3359 default: 3360 if (m) 3361 m_freem(m); 3362 return EAFNOSUPPORT; 3363 } 3364 3365 /* Compute the size of the TCP options. */ 3366 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3367 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3368 3369 tlen = hlen + sizeof(struct tcphdr) + optlen; 3370 3371 /* 3372 * Create the IP+TCP header from scratch. 3373 */ 3374 if (m) 3375 m_freem(m); 3376 #ifdef DIAGNOSTIC 3377 if (max_linkhdr + tlen > MCLBYTES) 3378 return (ENOBUFS); 3379 #endif 3380 MGETHDR(m, M_DONTWAIT, MT_DATA); 3381 if (m && tlen > MHLEN) { 3382 MCLGET(m, M_DONTWAIT); 3383 if ((m->m_flags & M_EXT) == 0) { 3384 m_freem(m); 3385 m = NULL; 3386 } 3387 } 3388 if (m == NULL) 3389 return (ENOBUFS); 3390 3391 /* Fixup the mbuf. */ 3392 m->m_data += max_linkhdr; 3393 m->m_len = m->m_pkthdr.len = tlen; 3394 #ifdef IPSEC 3395 if (sc->sc_tp) { 3396 struct tcpcb *tp; 3397 struct socket *so; 3398 3399 tp = sc->sc_tp; 3400 if (tp->t_inpcb) 3401 so = tp->t_inpcb->inp_socket; 3402 #ifdef INET6 3403 else if (tp->t_in6pcb) 3404 so = tp->t_in6pcb->in6p_socket; 3405 #endif 3406 else 3407 so = NULL; 3408 /* use IPsec policy on listening socket, on SYN ACK */ 3409 ipsec_setsocket(m, so); 3410 } 3411 #endif 3412 m->m_pkthdr.rcvif = NULL; 3413 memset(mtod(m, u_char *), 0, tlen); 3414 3415 switch (sc->sc_src.sa.sa_family) { 3416 case AF_INET: 3417 ip = mtod(m, struct ip *); 3418 ip->ip_dst = sc->sc_src.sin.sin_addr; 3419 ip->ip_src = sc->sc_dst.sin.sin_addr; 3420 ip->ip_p = IPPROTO_TCP; 3421 th = (struct tcphdr *)(ip + 1); 3422 th->th_dport = sc->sc_src.sin.sin_port; 3423 th->th_sport = sc->sc_dst.sin.sin_port; 3424 break; 3425 #ifdef INET6 3426 case AF_INET6: 3427 ip6 = mtod(m, struct ip6_hdr *); 3428 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3429 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3430 ip6->ip6_nxt = IPPROTO_TCP; 3431 /* ip6_plen will be updated in ip6_output() */ 3432 th = (struct tcphdr *)(ip6 + 1); 3433 th->th_dport = sc->sc_src.sin6.sin6_port; 3434 th->th_sport = sc->sc_dst.sin6.sin6_port; 3435 break; 3436 #endif 3437 default: 3438 th = NULL; 3439 } 3440 3441 th->th_seq = htonl(sc->sc_iss); 3442 th->th_ack = htonl(sc->sc_irs + 1); 3443 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3444 th->th_flags = TH_SYN|TH_ACK; 3445 th->th_win = htons(sc->sc_win); 3446 /* th_sum already 0 */ 3447 /* th_urp already 0 */ 3448 3449 /* Tack on the TCP options. */ 3450 optp = (u_int8_t *)(th + 1); 3451 *optp++ = TCPOPT_MAXSEG; 3452 *optp++ = 4; 3453 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3454 *optp++ = sc->sc_ourmaxseg & 0xff; 3455 3456 if (sc->sc_request_r_scale != 15) { 3457 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3458 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3459 sc->sc_request_r_scale); 3460 optp += 4; 3461 } 3462 3463 if (sc->sc_flags & SCF_TIMESTAMP) { 3464 u_int32_t *lp = (u_int32_t *)(optp); 3465 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3466 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3467 *lp++ = htonl(tcp_now); 3468 *lp = htonl(sc->sc_timestamp); 3469 optp += TCPOLEN_TSTAMP_APPA; 3470 } 3471 3472 /* Compute the packet's checksum. */ 3473 switch (sc->sc_src.sa.sa_family) { 3474 case AF_INET: 3475 ip->ip_len = htons(tlen - hlen); 3476 th->th_sum = 0; 3477 th->th_sum = in_cksum(m, tlen); 3478 break; 3479 #ifdef INET6 3480 case AF_INET6: 3481 ip6->ip6_plen = htons(tlen - hlen); 3482 th->th_sum = 0; 3483 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3484 break; 3485 #endif 3486 } 3487 3488 /* 3489 * Fill in some straggling IP bits. Note the stack expects 3490 * ip_len to be in host order, for convenience. 3491 */ 3492 switch (sc->sc_src.sa.sa_family) { 3493 case AF_INET: 3494 ip->ip_len = tlen; 3495 ip->ip_ttl = ip_defttl; 3496 /* XXX tos? */ 3497 break; 3498 #ifdef INET6 3499 case AF_INET6: 3500 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3501 ip6->ip6_vfc |= IPV6_VERSION; 3502 ip6->ip6_plen = htons(tlen - hlen); 3503 /* ip6_hlim will be initialized afterwards */ 3504 /* XXX flowlabel? */ 3505 break; 3506 #endif 3507 } 3508 3509 /* 3510 * If we're doing Path MTU discovery, we need to set DF unless 3511 * the route's MTU is locked. If we don't yet know the route, 3512 * look it up now. We will copy this reference to the inpcb 3513 * when we finish creating the connection. 3514 */ 3515 if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) { 3516 if (ro->ro_rt != NULL) { 3517 RTFREE(ro->ro_rt); 3518 ro->ro_rt = NULL; 3519 } 3520 bcopy(&sc->sc_src, &ro->ro_dst, sc->sc_src.sa.sa_len); 3521 rtalloc(ro); 3522 if ((rt = ro->ro_rt) == NULL) { 3523 m_freem(m); 3524 switch (sc->sc_src.sa.sa_family) { 3525 case AF_INET: 3526 ipstat.ips_noroute++; 3527 break; 3528 #ifdef INET6 3529 case AF_INET6: 3530 ip6stat.ip6s_noroute++; 3531 break; 3532 #endif 3533 } 3534 return (EHOSTUNREACH); 3535 } 3536 } 3537 3538 switch (sc->sc_src.sa.sa_family) { 3539 case AF_INET: 3540 if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0) 3541 ip->ip_off |= IP_DF; 3542 3543 /* ...and send it off! */ 3544 error = ip_output(m, sc->sc_ipopts, ro, 0, NULL); 3545 break; 3546 #ifdef INET6 3547 case AF_INET6: 3548 ip6->ip6_hlim = in6_selecthlim(NULL, 3549 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3550 3551 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3552 0, NULL, NULL); 3553 break; 3554 #endif 3555 default: 3556 error = EAFNOSUPPORT; 3557 break; 3558 } 3559 return (error); 3560 } 3561
Indexes created Mon Sep 22 10:09:38 GMT 2025