tcp_input.c revision 1.118
1 /* $NetBSD: tcp_input.c,v 1.118 2000/10/17 02:57:02 thorpej Exp $ */ 2 3 /* 4 %%% portions-copyright-nrl-95 5 Portions of this software are Copyright 1995-1998 by Randall Atkinson, 6 Ronald Lee, Daniel McDonald, Bao Phan, and Chris Winters. All Rights 7 Reserved. All rights under this copyright have been assigned to the US 8 Naval Research Laboratory (NRL). The NRL Copyright Notice and License 9 Agreement Version 1.1 (January 17, 1995) applies to these portions of the 10 software. 11 You should have received a copy of the license with this software. If you 12 didn't get a copy, you may request one from <license (at) ipv6.nrl.navy.mil>. 13 14 */ 15 16 /* 17 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 18 * All rights reserved. 19 * 20 * Redistribution and use in source and binary forms, with or without 21 * modification, are permitted provided that the following conditions 22 * are met: 23 * 1. Redistributions of source code must retain the above copyright 24 * notice, this list of conditions and the following disclaimer. 25 * 2. Redistributions in binary form must reproduce the above copyright 26 * notice, this list of conditions and the following disclaimer in the 27 * documentation and/or other materials provided with the distribution. 28 * 3. Neither the name of the project nor the names of its contributors 29 * may be used to endorse or promote products derived from this software 30 * without specific prior written permission. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 33 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 34 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 35 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 36 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 37 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 38 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 39 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 40 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 41 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 42 * SUCH DAMAGE. 43 */ 44 45 /*- 46 * Copyright (c) 1997, 1998, 1999 The NetBSD Foundation, Inc. 47 * All rights reserved. 48 * 49 * This code is derived from software contributed to The NetBSD Foundation 50 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 51 * Facility, NASA Ames Research Center. 52 * 53 * Redistribution and use in source and binary forms, with or without 54 * modification, are permitted provided that the following conditions 55 * are met: 56 * 1. Redistributions of source code must retain the above copyright 57 * notice, this list of conditions and the following disclaimer. 58 * 2. Redistributions in binary form must reproduce the above copyright 59 * notice, this list of conditions and the following disclaimer in the 60 * documentation and/or other materials provided with the distribution. 61 * 3. All advertising materials mentioning features or use of this software 62 * must display the following acknowledgement: 63 * This product includes software developed by the NetBSD 64 * Foundation, Inc. and its contributors. 65 * 4. Neither the name of The NetBSD Foundation nor the names of its 66 * contributors may be used to endorse or promote products derived 67 * from this software without specific prior written permission. 68 * 69 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 70 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 71 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 72 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 73 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 74 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 75 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 76 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 77 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 78 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 79 * POSSIBILITY OF SUCH DAMAGE. 80 */ 81 82 /* 83 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 84 * The Regents of the University of California. All rights reserved. 85 * 86 * Redistribution and use in source and binary forms, with or without 87 * modification, are permitted provided that the following conditions 88 * are met: 89 * 1. Redistributions of source code must retain the above copyright 90 * notice, this list of conditions and the following disclaimer. 91 * 2. Redistributions in binary form must reproduce the above copyright 92 * notice, this list of conditions and the following disclaimer in the 93 * documentation and/or other materials provided with the distribution. 94 * 3. All advertising materials mentioning features or use of this software 95 * must display the following acknowledgement: 96 * This product includes software developed by the University of 97 * California, Berkeley and its contributors. 98 * 4. Neither the name of the University nor the names of its contributors 99 * may be used to endorse or promote products derived from this software 100 * without specific prior written permission. 101 * 102 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 103 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 104 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 105 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 106 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 107 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 108 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 109 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 110 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 111 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 112 * SUCH DAMAGE. 113 * 114 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 115 */ 116 117 /* 118 * TODO list for SYN cache stuff: 119 * 120 * Find room for a "state" field, which is needed to keep a 121 * compressed state for TIME_WAIT TCBs. It's been noted already 122 * that this is fairly important for very high-volume web and 123 * mail servers, which use a large number of short-lived 124 * connections. 125 */ 126 127 #include "opt_inet.h" 128 #include "opt_ipsec.h" 129 130 #include <sys/param.h> 131 #include <sys/systm.h> 132 #include <sys/malloc.h> 133 #include <sys/mbuf.h> 134 #include <sys/protosw.h> 135 #include <sys/socket.h> 136 #include <sys/socketvar.h> 137 #include <sys/errno.h> 138 #include <sys/syslog.h> 139 #include <sys/pool.h> 140 #include <sys/domain.h> 141 142 #include <net/if.h> 143 #include <net/route.h> 144 #include <net/if_types.h> 145 146 #include <netinet/in.h> 147 #include <netinet/in_systm.h> 148 #include <netinet/ip.h> 149 #include <netinet/in_pcb.h> 150 #include <netinet/ip_var.h> 151 152 #ifdef INET6 153 #ifndef INET 154 #include <netinet/in.h> 155 #endif 156 #include <netinet/ip6.h> 157 #include <netinet6/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 static int tcp_rst_ppslim_count = 0; 194 static struct timeval tcp_rst_ppslim_last; 195 196 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 197 198 /* for modulo comparisons of timestamps */ 199 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 200 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 201 202 /* 203 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 204 */ 205 #ifdef INET6 206 #define ND6_HINT(tp) \ 207 do { \ 208 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 209 && tp->t_in6pcb->in6p_route.ro_rt) { \ 210 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 211 } \ 212 } while (0) 213 #else 214 #define ND6_HINT(tp) 215 #endif 216 217 /* 218 * Macro to compute ACK transmission behavior. Delay the ACK unless 219 * we have already delayed an ACK (must send an ACK every two segments). 220 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 221 * option is enabled. 222 */ 223 #define TCP_SETUP_ACK(tp, th) \ 224 do { \ 225 if ((tp)->t_flags & TF_DELACK || \ 226 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 227 tp->t_flags |= TF_ACKNOW; \ 228 else \ 229 TCP_SET_DELACK(tp); \ 230 } while (0) 231 232 /* 233 * Convert TCP protocol fields to host order for easier processing. 234 */ 235 #define TCP_FIELDS_TO_HOST(th) \ 236 do { \ 237 NTOHL((th)->th_seq); \ 238 NTOHL((th)->th_ack); \ 239 NTOHS((th)->th_win); \ 240 NTOHS((th)->th_urp); \ 241 } while (0) 242 243 int 244 tcp_reass(tp, th, m, tlen) 245 struct tcpcb *tp; 246 struct tcphdr *th; 247 struct mbuf *m; 248 int *tlen; 249 { 250 struct ipqent *p, *q, *nq, *tiqe = NULL; 251 struct socket *so = NULL; 252 int pkt_flags; 253 tcp_seq pkt_seq; 254 unsigned pkt_len; 255 u_long rcvpartdupbyte = 0; 256 u_long rcvoobyte; 257 258 if (tp->t_inpcb) 259 so = tp->t_inpcb->inp_socket; 260 #ifdef INET6 261 else if (tp->t_in6pcb) 262 so = tp->t_in6pcb->in6p_socket; 263 #endif 264 265 TCP_REASS_LOCK_CHECK(tp); 266 267 /* 268 * Call with th==0 after become established to 269 * force pre-ESTABLISHED data up to user socket. 270 */ 271 if (th == 0) 272 goto present; 273 274 rcvoobyte = *tlen; 275 /* 276 * Copy these to local variables because the tcpiphdr 277 * gets munged while we are collapsing mbufs. 278 */ 279 pkt_seq = th->th_seq; 280 pkt_len = *tlen; 281 pkt_flags = th->th_flags; 282 /* 283 * Find a segment which begins after this one does. 284 */ 285 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 286 nq = q->ipqe_q.le_next; 287 /* 288 * If the received segment is just right after this 289 * fragment, merge the two together and then check 290 * for further overlaps. 291 */ 292 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 293 #ifdef TCPREASS_DEBUG 294 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 295 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 296 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 297 #endif 298 pkt_len += q->ipqe_len; 299 pkt_flags |= q->ipqe_flags; 300 pkt_seq = q->ipqe_seq; 301 m_cat(q->ipqe_m, m); 302 m = q->ipqe_m; 303 goto free_ipqe; 304 } 305 /* 306 * If the received segment is completely past this 307 * fragment, we need to go the next fragment. 308 */ 309 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 310 p = q; 311 continue; 312 } 313 /* 314 * If the fragment is past the received segment, 315 * it (or any following) can't be concatenated. 316 */ 317 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 318 break; 319 /* 320 * We've received all the data in this segment before. 321 * mark it as a duplicate and return. 322 */ 323 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 324 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 325 tcpstat.tcps_rcvduppack++; 326 tcpstat.tcps_rcvdupbyte += pkt_len; 327 m_freem(m); 328 if (tiqe != NULL) 329 pool_put(&ipqent_pool, tiqe); 330 return (0); 331 } 332 /* 333 * Received segment completely overlaps this fragment 334 * so we drop the fragment (this keeps the temporal 335 * ordering of segments correct). 336 */ 337 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 338 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 339 rcvpartdupbyte += q->ipqe_len; 340 m_freem(q->ipqe_m); 341 goto free_ipqe; 342 } 343 /* 344 * RX'ed segment extends past the end of the 345 * fragment. Drop the overlapping bytes. Then 346 * merge the fragment and segment then treat as 347 * a longer received packet. 348 */ 349 if (SEQ_LT(q->ipqe_seq, pkt_seq) 350 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 351 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 352 #ifdef TCPREASS_DEBUG 353 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 354 tp, overlap, 355 pkt_seq, pkt_seq + pkt_len, pkt_len); 356 #endif 357 m_adj(m, overlap); 358 rcvpartdupbyte += overlap; 359 m_cat(q->ipqe_m, m); 360 m = q->ipqe_m; 361 pkt_seq = q->ipqe_seq; 362 pkt_len += q->ipqe_len - overlap; 363 rcvoobyte -= overlap; 364 goto free_ipqe; 365 } 366 /* 367 * RX'ed segment extends past the front of the 368 * fragment. Drop the overlapping bytes on the 369 * received packet. The packet will then be 370 * contatentated with this fragment a bit later. 371 */ 372 if (SEQ_GT(q->ipqe_seq, pkt_seq) 373 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 374 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 375 #ifdef TCPREASS_DEBUG 376 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 377 tp, overlap, 378 pkt_seq, pkt_seq + pkt_len, pkt_len); 379 #endif 380 m_adj(m, -overlap); 381 pkt_len -= overlap; 382 rcvpartdupbyte += overlap; 383 rcvoobyte -= overlap; 384 } 385 /* 386 * If the received segment immediates precedes this 387 * fragment then tack the fragment onto this segment 388 * and reinsert the data. 389 */ 390 if (q->ipqe_seq == pkt_seq + pkt_len) { 391 #ifdef TCPREASS_DEBUG 392 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 393 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 394 pkt_seq, pkt_seq + pkt_len, pkt_len); 395 #endif 396 pkt_len += q->ipqe_len; 397 pkt_flags |= q->ipqe_flags; 398 m_cat(m, q->ipqe_m); 399 LIST_REMOVE(q, ipqe_q); 400 LIST_REMOVE(q, ipqe_timeq); 401 if (tiqe == NULL) { 402 tiqe = q; 403 } else { 404 pool_put(&ipqent_pool, q); 405 } 406 break; 407 } 408 /* 409 * If the fragment is before the segment, remember it. 410 * When this loop is terminated, p will contain the 411 * pointer to fragment that is right before the received 412 * segment. 413 */ 414 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 415 p = q; 416 417 continue; 418 419 /* 420 * This is a common operation. It also will allow 421 * to save doing a malloc/free in most instances. 422 */ 423 free_ipqe: 424 LIST_REMOVE(q, ipqe_q); 425 LIST_REMOVE(q, ipqe_timeq); 426 if (tiqe == NULL) { 427 tiqe = q; 428 } else { 429 pool_put(&ipqent_pool, q); 430 } 431 } 432 433 /* 434 * Allocate a new queue entry since the received segment did not 435 * collapse onto any other out-of-order block; thus we are allocating 436 * a new block. If it had collapsed, tiqe would not be NULL and 437 * we would be reusing it. 438 * XXX If we can't, just drop the packet. XXX 439 */ 440 if (tiqe == NULL) { 441 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 442 if (tiqe == NULL) { 443 tcpstat.tcps_rcvmemdrop++; 444 m_freem(m); 445 return (0); 446 } 447 } 448 449 /* 450 * Update the counters. 451 */ 452 tcpstat.tcps_rcvoopack++; 453 tcpstat.tcps_rcvoobyte += rcvoobyte; 454 if (rcvpartdupbyte) { 455 tcpstat.tcps_rcvpartduppack++; 456 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 457 } 458 459 /* 460 * Insert the new fragment queue entry into both queues. 461 */ 462 tiqe->ipqe_m = m; 463 tiqe->ipqe_seq = pkt_seq; 464 tiqe->ipqe_len = pkt_len; 465 tiqe->ipqe_flags = pkt_flags; 466 if (p == NULL) { 467 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 468 #ifdef TCPREASS_DEBUG 469 if (tiqe->ipqe_seq != tp->rcv_nxt) 470 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 471 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 472 #endif 473 } else { 474 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 475 #ifdef TCPREASS_DEBUG 476 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 477 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 478 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 479 #endif 480 } 481 482 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 483 484 present: 485 /* 486 * Present data to user, advancing rcv_nxt through 487 * completed sequence space. 488 */ 489 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 490 return (0); 491 q = tp->segq.lh_first; 492 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 493 return (0); 494 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 495 return (0); 496 497 tp->rcv_nxt += q->ipqe_len; 498 pkt_flags = q->ipqe_flags & TH_FIN; 499 ND6_HINT(tp); 500 501 LIST_REMOVE(q, ipqe_q); 502 LIST_REMOVE(q, ipqe_timeq); 503 if (so->so_state & SS_CANTRCVMORE) 504 m_freem(q->ipqe_m); 505 else 506 sbappend(&so->so_rcv, q->ipqe_m); 507 pool_put(&ipqent_pool, q); 508 sorwakeup(so); 509 return (pkt_flags); 510 } 511 512 #if defined(INET6) && !defined(TCP6) 513 int 514 tcp6_input(mp, offp, proto) 515 struct mbuf **mp; 516 int *offp, proto; 517 { 518 struct mbuf *m = *mp; 519 520 /* 521 * draft-itojun-ipv6-tcp-to-anycast 522 * better place to put this in? 523 */ 524 if (m->m_flags & M_ANYCAST6) { 525 struct ip6_hdr *ip6; 526 if (m->m_len < sizeof(struct ip6_hdr)) { 527 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 528 tcpstat.tcps_rcvshort++; 529 return IPPROTO_DONE; 530 } 531 } 532 ip6 = mtod(m, struct ip6_hdr *); 533 icmp6_error(m, ICMP6_DST_UNREACH, 534 ICMP6_DST_UNREACH_ADDR, 535 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 536 return IPPROTO_DONE; 537 } 538 539 tcp_input(m, *offp, proto); 540 return IPPROTO_DONE; 541 } 542 #endif 543 544 /* 545 * TCP input routine, follows pages 65-76 of the 546 * protocol specification dated September, 1981 very closely. 547 */ 548 void 549 #if __STDC__ 550 tcp_input(struct mbuf *m, ...) 551 #else 552 tcp_input(m, va_alist) 553 struct mbuf *m; 554 #endif 555 { 556 int proto; 557 struct tcphdr *th; 558 struct ip *ip; 559 struct inpcb *inp; 560 #ifdef INET6 561 struct ip6_hdr *ip6; 562 struct in6pcb *in6p; 563 #endif 564 caddr_t optp = NULL; 565 int optlen = 0; 566 int len, tlen, toff, hdroptlen = 0; 567 struct tcpcb *tp = 0; 568 int tiflags; 569 struct socket *so = NULL; 570 int todrop, acked, ourfinisacked, needoutput = 0; 571 short ostate = 0; 572 int iss = 0; 573 u_long tiwin; 574 struct tcp_opt_info opti; 575 int off, iphlen; 576 va_list ap; 577 int af; /* af on the wire */ 578 struct mbuf *tcp_saveti = NULL; 579 580 va_start(ap, m); 581 toff = va_arg(ap, int); 582 proto = va_arg(ap, int); 583 va_end(ap); 584 585 tcpstat.tcps_rcvtotal++; 586 587 bzero(&opti, sizeof(opti)); 588 opti.ts_present = 0; 589 opti.maxseg = 0; 590 591 /* 592 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 593 * 594 * TCP is, by definition, unicast, so we reject all 595 * multicast outright. 596 * 597 * Note, there are additional src/dst address checks in 598 * the AF-specific code below. 599 */ 600 if (m->m_flags & (M_BCAST|M_MCAST)) { 601 /* XXX stat */ 602 goto drop; 603 } 604 #ifdef INET6 605 if (m->m_flags & M_ANYCAST6) { 606 /* XXX stat */ 607 goto drop; 608 } 609 #endif 610 611 /* 612 * Get IP and TCP header together in first mbuf. 613 * Note: IP leaves IP header in first mbuf. 614 */ 615 ip = mtod(m, struct ip *); 616 #ifdef INET6 617 ip6 = NULL; 618 #endif 619 switch (ip->ip_v) { 620 case 4: 621 af = AF_INET; 622 iphlen = sizeof(struct ip); 623 #ifndef PULLDOWN_TEST 624 /* would like to get rid of this... */ 625 if (toff > sizeof (struct ip)) { 626 ip_stripoptions(m, (struct mbuf *)0); 627 toff = sizeof(struct ip); 628 } 629 if (m->m_len < toff + sizeof (struct tcphdr)) { 630 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 631 tcpstat.tcps_rcvshort++; 632 return; 633 } 634 } 635 ip = mtod(m, struct ip *); 636 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 637 #else 638 ip = mtod(m, struct ip *); 639 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 640 sizeof(struct tcphdr)); 641 if (th == NULL) { 642 tcpstat.tcps_rcvshort++; 643 return; 644 } 645 #endif 646 647 /* 648 * Make sure destination address is not multicast. 649 * Source address checked in ip_input(). 650 */ 651 if (IN_MULTICAST(ip->ip_dst.s_addr)) { 652 /* XXX stat */ 653 goto drop; 654 } 655 656 /* We do the checksum after PCB lookup... */ 657 len = ip->ip_len; 658 tlen = len - toff; 659 break; 660 #ifdef INET6 661 case 6: 662 ip = NULL; 663 iphlen = sizeof(struct ip6_hdr); 664 af = AF_INET6; 665 #ifndef PULLDOWN_TEST 666 if (m->m_len < toff + sizeof(struct tcphdr)) { 667 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 668 if (m == NULL) { 669 tcpstat.tcps_rcvshort++; 670 return; 671 } 672 } 673 ip6 = mtod(m, struct ip6_hdr *); 674 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 675 #else 676 ip6 = mtod(m, struct ip6_hdr *); 677 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 678 sizeof(struct tcphdr)); 679 if (th == NULL) { 680 tcpstat.tcps_rcvshort++; 681 return; 682 } 683 #endif 684 685 /* Be proactive about malicious use of IPv4 mapped address */ 686 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 687 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 688 /* XXX stat */ 689 goto drop; 690 } 691 692 /* 693 * Be proactive about unspecified IPv6 address in source. 694 * As we use all-zero to indicate unbounded/unconnected pcb, 695 * unspecified IPv6 address can be used to confuse us. 696 * 697 * Note that packets with unspecified IPv6 destination is 698 * already dropped in ip6_input. 699 */ 700 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 701 /* XXX stat */ 702 goto drop; 703 } 704 705 /* 706 * Make sure destination address is not multicast. 707 * Source address checked in ip6_input(). 708 */ 709 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 710 /* XXX stat */ 711 goto drop; 712 } 713 714 /* We do the checksum after PCB lookup... */ 715 len = m->m_pkthdr.len; 716 tlen = len - toff; 717 break; 718 #endif 719 default: 720 m_freem(m); 721 return; 722 } 723 724 /* 725 * Check that TCP offset makes sense, 726 * pull out TCP options and adjust length. XXX 727 */ 728 off = th->th_off << 2; 729 if (off < sizeof (struct tcphdr) || off > tlen) { 730 tcpstat.tcps_rcvbadoff++; 731 goto drop; 732 } 733 tlen -= off; 734 735 /* 736 * tcp_input() has been modified to use tlen to mean the TCP data 737 * length throughout the function. Other functions can use 738 * m->m_pkthdr.len as the basis for calculating the TCP data length. 739 * rja 740 */ 741 742 if (off > sizeof (struct tcphdr)) { 743 #ifndef PULLDOWN_TEST 744 if (m->m_len < toff + off) { 745 if ((m = m_pullup(m, toff + off)) == 0) { 746 tcpstat.tcps_rcvshort++; 747 return; 748 } 749 switch (af) { 750 case AF_INET: 751 ip = mtod(m, struct ip *); 752 break; 753 #ifdef INET6 754 case AF_INET6: 755 ip6 = mtod(m, struct ip6_hdr *); 756 break; 757 #endif 758 } 759 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 760 } 761 #else 762 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 763 if (th == NULL) { 764 tcpstat.tcps_rcvshort++; 765 return; 766 } 767 /* 768 * NOTE: ip/ip6 will not be affected by m_pulldown() 769 * (as they're before toff) and we don't need to update those. 770 */ 771 #endif 772 optlen = off - sizeof (struct tcphdr); 773 optp = ((caddr_t)th) + sizeof(struct tcphdr); 774 /* 775 * Do quick retrieval of timestamp options ("options 776 * prediction?"). If timestamp is the only option and it's 777 * formatted as recommended in RFC 1323 appendix A, we 778 * quickly get the values now and not bother calling 779 * tcp_dooptions(), etc. 780 */ 781 if ((optlen == TCPOLEN_TSTAMP_APPA || 782 (optlen > TCPOLEN_TSTAMP_APPA && 783 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 784 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 785 (th->th_flags & TH_SYN) == 0) { 786 opti.ts_present = 1; 787 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 788 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 789 optp = NULL; /* we've parsed the options */ 790 } 791 } 792 tiflags = th->th_flags; 793 794 /* 795 * Locate pcb for segment. 796 */ 797 findpcb: 798 inp = NULL; 799 #ifdef INET6 800 in6p = NULL; 801 #endif 802 switch (af) { 803 case AF_INET: 804 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 805 ip->ip_dst, th->th_dport); 806 if (inp == 0) { 807 ++tcpstat.tcps_pcbhashmiss; 808 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 809 } 810 #if defined(INET6) && !defined(TCP6) 811 if (inp == 0) { 812 struct in6_addr s, d; 813 814 /* mapped addr case */ 815 bzero(&s, sizeof(s)); 816 s.s6_addr16[5] = htons(0xffff); 817 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 818 bzero(&d, sizeof(d)); 819 d.s6_addr16[5] = htons(0xffff); 820 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 821 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 822 &d, th->th_dport, 0); 823 if (in6p == 0) { 824 ++tcpstat.tcps_pcbhashmiss; 825 in6p = in6_pcblookup_bind(&tcb6, &d, 826 th->th_dport, 0); 827 } 828 } 829 #endif 830 #ifndef INET6 831 if (inp == 0) 832 #else 833 if (inp == 0 && in6p == 0) 834 #endif 835 { 836 ++tcpstat.tcps_noport; 837 if (tcp_log_refused && (tiflags & TH_SYN)) { 838 #ifndef INET6 839 char src[4*sizeof "123"]; 840 char dst[4*sizeof "123"]; 841 #else 842 char src[INET6_ADDRSTRLEN]; 843 char dst[INET6_ADDRSTRLEN]; 844 #endif 845 if (ip) { 846 strcpy(src, inet_ntoa(ip->ip_src)); 847 strcpy(dst, inet_ntoa(ip->ip_dst)); 848 } 849 #ifdef INET6 850 else if (ip6) { 851 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 852 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 853 } 854 #endif 855 else { 856 strcpy(src, "(unknown)"); 857 strcpy(dst, "(unknown)"); 858 } 859 log(LOG_INFO, 860 "Connection attempt to TCP %s:%d from %s:%d\n", 861 dst, ntohs(th->th_dport), 862 src, ntohs(th->th_sport)); 863 } 864 TCP_FIELDS_TO_HOST(th); 865 goto dropwithreset_ratelim; 866 } 867 #ifdef IPSEC 868 if (inp && ipsec4_in_reject(m, inp)) { 869 ipsecstat.in_polvio++; 870 goto drop; 871 } 872 #ifdef INET6 873 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 874 ipsecstat.in_polvio++; 875 goto drop; 876 } 877 #endif 878 #endif /*IPSEC*/ 879 break; 880 #if defined(INET6) && !defined(TCP6) 881 case AF_INET6: 882 { 883 int faith; 884 885 #if defined(NFAITH) && NFAITH > 0 886 if (m->m_pkthdr.rcvif 887 && m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 888 faith = 1; 889 } else 890 faith = 0; 891 #else 892 faith = 0; 893 #endif 894 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 895 &ip6->ip6_dst, th->th_dport, faith); 896 if (in6p == NULL) { 897 ++tcpstat.tcps_pcbhashmiss; 898 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 899 th->th_dport, faith); 900 } 901 if (in6p == NULL) { 902 ++tcpstat.tcps_noport; 903 TCP_FIELDS_TO_HOST(th); 904 goto dropwithreset_ratelim; 905 } 906 #ifdef IPSEC 907 if (ipsec6_in_reject(m, in6p)) { 908 ipsec6stat.in_polvio++; 909 goto drop; 910 } 911 #endif /*IPSEC*/ 912 break; 913 } 914 #endif 915 } 916 917 /* 918 * If the state is CLOSED (i.e., TCB does not exist) then 919 * all data in the incoming segment is discarded. 920 * If the TCB exists but is in CLOSED state, it is embryonic, 921 * but should either do a listen or a connect soon. 922 */ 923 tp = NULL; 924 so = NULL; 925 if (inp) { 926 tp = intotcpcb(inp); 927 so = inp->inp_socket; 928 } 929 #ifdef INET6 930 else if (in6p) { 931 tp = in6totcpcb(in6p); 932 so = in6p->in6p_socket; 933 } 934 #endif 935 if (tp == 0) { 936 TCP_FIELDS_TO_HOST(th); 937 goto dropwithreset_ratelim; 938 } 939 if (tp->t_state == TCPS_CLOSED) 940 goto drop; 941 942 /* 943 * Checksum extended TCP header and data. 944 */ 945 switch (af) { 946 case AF_INET: 947 #ifndef PULLDOWN_TEST 948 { 949 struct ipovly *ipov; 950 ipov = (struct ipovly *)ip; 951 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 952 ipov->ih_len = htons(tlen + off); 953 954 if (in_cksum(m, len) != 0) { 955 tcpstat.tcps_rcvbadsum++; 956 goto drop; 957 } 958 } 959 #else 960 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) { 961 tcpstat.tcps_rcvbadsum++; 962 goto drop; 963 } 964 #endif 965 break; 966 967 #ifdef INET6 968 case AF_INET6: 969 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) { 970 tcpstat.tcps_rcvbadsum++; 971 goto drop; 972 } 973 break; 974 #endif 975 } 976 977 TCP_FIELDS_TO_HOST(th); 978 979 /* Unscale the window into a 32-bit value. */ 980 if ((tiflags & TH_SYN) == 0) 981 tiwin = th->th_win << tp->snd_scale; 982 else 983 tiwin = th->th_win; 984 985 #ifdef INET6 986 /* save packet options if user wanted */ 987 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 988 if (in6p->in6p_options) { 989 m_freem(in6p->in6p_options); 990 in6p->in6p_options = 0; 991 } 992 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 993 } 994 #endif 995 996 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 997 union syn_cache_sa src; 998 union syn_cache_sa dst; 999 1000 bzero(&src, sizeof(src)); 1001 bzero(&dst, sizeof(dst)); 1002 switch (af) { 1003 case AF_INET: 1004 src.sin.sin_len = sizeof(struct sockaddr_in); 1005 src.sin.sin_family = AF_INET; 1006 src.sin.sin_addr = ip->ip_src; 1007 src.sin.sin_port = th->th_sport; 1008 1009 dst.sin.sin_len = sizeof(struct sockaddr_in); 1010 dst.sin.sin_family = AF_INET; 1011 dst.sin.sin_addr = ip->ip_dst; 1012 dst.sin.sin_port = th->th_dport; 1013 break; 1014 #ifdef INET6 1015 case AF_INET6: 1016 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1017 src.sin6.sin6_family = AF_INET6; 1018 src.sin6.sin6_addr = ip6->ip6_src; 1019 src.sin6.sin6_port = th->th_sport; 1020 1021 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1022 dst.sin6.sin6_family = AF_INET6; 1023 dst.sin6.sin6_addr = ip6->ip6_dst; 1024 dst.sin6.sin6_port = th->th_dport; 1025 break; 1026 #endif /* INET6 */ 1027 default: 1028 goto badsyn; /*sanity*/ 1029 } 1030 1031 if (so->so_options & SO_DEBUG) { 1032 ostate = tp->t_state; 1033 1034 tcp_saveti = NULL; 1035 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1036 goto nosave; 1037 1038 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1039 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1040 if (!tcp_saveti) 1041 goto nosave; 1042 } else { 1043 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1044 if (!tcp_saveti) 1045 goto nosave; 1046 tcp_saveti->m_len = iphlen; 1047 m_copydata(m, 0, iphlen, 1048 mtod(tcp_saveti, caddr_t)); 1049 } 1050 1051 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1052 m_freem(tcp_saveti); 1053 tcp_saveti = NULL; 1054 } else { 1055 tcp_saveti->m_len += sizeof(struct tcphdr); 1056 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1057 sizeof(struct tcphdr)); 1058 } 1059 if (tcp_saveti) { 1060 /* 1061 * need to recover version # field, which was 1062 * overwritten on ip_cksum computation. 1063 */ 1064 struct ip *sip; 1065 sip = mtod(tcp_saveti, struct ip *); 1066 switch (af) { 1067 case AF_INET: 1068 sip->ip_v = 4; 1069 break; 1070 #ifdef INET6 1071 case AF_INET6: 1072 sip->ip_v = 6; 1073 break; 1074 #endif 1075 } 1076 } 1077 nosave:; 1078 } 1079 if (so->so_options & SO_ACCEPTCONN) { 1080 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1081 if (tiflags & TH_RST) { 1082 syn_cache_reset(&src.sa, &dst.sa, th); 1083 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1084 (TH_ACK|TH_SYN)) { 1085 /* 1086 * Received a SYN,ACK. This should 1087 * never happen while we are in 1088 * LISTEN. Send an RST. 1089 */ 1090 goto badsyn; 1091 } else if (tiflags & TH_ACK) { 1092 so = syn_cache_get(&src.sa, &dst.sa, 1093 th, toff, tlen, so, m); 1094 if (so == NULL) { 1095 /* 1096 * We don't have a SYN for 1097 * this ACK; send an RST. 1098 */ 1099 goto badsyn; 1100 } else if (so == 1101 (struct socket *)(-1)) { 1102 /* 1103 * We were unable to create 1104 * the connection. If the 1105 * 3-way handshake was 1106 * completed, and RST has 1107 * been sent to the peer. 1108 * Since the mbuf might be 1109 * in use for the reply, 1110 * do not free it. 1111 */ 1112 m = NULL; 1113 } else { 1114 /* 1115 * We have created a 1116 * full-blown connection. 1117 */ 1118 tp = NULL; 1119 inp = NULL; 1120 #ifdef INET6 1121 in6p = NULL; 1122 #endif 1123 switch (so->so_proto->pr_domain->dom_family) { 1124 case AF_INET: 1125 inp = sotoinpcb(so); 1126 tp = intotcpcb(inp); 1127 break; 1128 #ifdef INET6 1129 case AF_INET6: 1130 in6p = sotoin6pcb(so); 1131 tp = in6totcpcb(in6p); 1132 break; 1133 #endif 1134 } 1135 if (tp == NULL) 1136 goto badsyn; /*XXX*/ 1137 tiwin <<= tp->snd_scale; 1138 goto after_listen; 1139 } 1140 } else { 1141 /* 1142 * None of RST, SYN or ACK was set. 1143 * This is an invalid packet for a 1144 * TCB in LISTEN state. Send a RST. 1145 */ 1146 goto badsyn; 1147 } 1148 } else { 1149 /* 1150 * Received a SYN. 1151 */ 1152 1153 /* 1154 * LISTEN socket received a SYN 1155 * from itself? This can't possibly 1156 * be valid; drop the packet. 1157 */ 1158 if (th->th_sport == th->th_dport) { 1159 int i; 1160 1161 switch (af) { 1162 case AF_INET: 1163 i = in_hosteq(ip->ip_src, ip->ip_dst); 1164 break; 1165 #ifdef INET6 1166 case AF_INET6: 1167 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1168 break; 1169 #endif 1170 default: 1171 i = 1; 1172 } 1173 if (i) { 1174 tcpstat.tcps_badsyn++; 1175 goto drop; 1176 } 1177 } 1178 1179 /* 1180 * SYN looks ok; create compressed TCP 1181 * state for it. 1182 */ 1183 if (so->so_qlen <= so->so_qlimit && 1184 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1185 so, m, optp, optlen, &opti)) 1186 m = NULL; 1187 } 1188 goto drop; 1189 } 1190 } 1191 1192 after_listen: 1193 #ifdef DIAGNOSTIC 1194 /* 1195 * Should not happen now that all embryonic connections 1196 * are handled with compressed state. 1197 */ 1198 if (tp->t_state == TCPS_LISTEN) 1199 panic("tcp_input: TCPS_LISTEN"); 1200 #endif 1201 1202 /* 1203 * Segment received on connection. 1204 * Reset idle time and keep-alive timer. 1205 */ 1206 tp->t_idle = 0; 1207 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1208 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1209 1210 /* 1211 * Process options. 1212 */ 1213 if (optp) 1214 tcp_dooptions(tp, optp, optlen, th, &opti); 1215 1216 /* 1217 * Header prediction: check for the two common cases 1218 * of a uni-directional data xfer. If the packet has 1219 * no control flags, is in-sequence, the window didn't 1220 * change and we're not retransmitting, it's a 1221 * candidate. If the length is zero and the ack moved 1222 * forward, we're the sender side of the xfer. Just 1223 * free the data acked & wake any higher level process 1224 * that was blocked waiting for space. If the length 1225 * is non-zero and the ack didn't move, we're the 1226 * receiver side. If we're getting packets in-order 1227 * (the reassembly queue is empty), add the data to 1228 * the socket buffer and note that we need a delayed ack. 1229 */ 1230 if (tp->t_state == TCPS_ESTABLISHED && 1231 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1232 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1233 th->th_seq == tp->rcv_nxt && 1234 tiwin && tiwin == tp->snd_wnd && 1235 tp->snd_nxt == tp->snd_max) { 1236 1237 /* 1238 * If last ACK falls within this segment's sequence numbers, 1239 * record the timestamp. 1240 */ 1241 if (opti.ts_present && 1242 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1243 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1244 tp->ts_recent_age = tcp_now; 1245 tp->ts_recent = opti.ts_val; 1246 } 1247 1248 if (tlen == 0) { 1249 if (SEQ_GT(th->th_ack, tp->snd_una) && 1250 SEQ_LEQ(th->th_ack, tp->snd_max) && 1251 tp->snd_cwnd >= tp->snd_wnd && 1252 tp->t_dupacks < tcprexmtthresh) { 1253 /* 1254 * this is a pure ack for outstanding data. 1255 */ 1256 ++tcpstat.tcps_predack; 1257 if (opti.ts_present && opti.ts_ecr) 1258 tcp_xmit_timer(tp, 1259 tcp_now - opti.ts_ecr + 1); 1260 else if (tp->t_rtt && 1261 SEQ_GT(th->th_ack, tp->t_rtseq)) 1262 tcp_xmit_timer(tp, tp->t_rtt); 1263 acked = th->th_ack - tp->snd_una; 1264 tcpstat.tcps_rcvackpack++; 1265 tcpstat.tcps_rcvackbyte += acked; 1266 ND6_HINT(tp); 1267 sbdrop(&so->so_snd, acked); 1268 /* 1269 * We want snd_recover to track snd_una to 1270 * avoid sequence wraparound problems for 1271 * very large transfers. 1272 */ 1273 tp->snd_una = tp->snd_recover = th->th_ack; 1274 m_freem(m); 1275 1276 /* 1277 * If all outstanding data are acked, stop 1278 * retransmit timer, otherwise restart timer 1279 * using current (possibly backed-off) value. 1280 * If process is waiting for space, 1281 * wakeup/selwakeup/signal. If data 1282 * are ready to send, let tcp_output 1283 * decide between more output or persist. 1284 */ 1285 if (tp->snd_una == tp->snd_max) 1286 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1287 else if (TCP_TIMER_ISARMED(tp, 1288 TCPT_PERSIST) == 0) 1289 TCP_TIMER_ARM(tp, TCPT_REXMT, 1290 tp->t_rxtcur); 1291 1292 sowwakeup(so); 1293 if (so->so_snd.sb_cc) 1294 (void) tcp_output(tp); 1295 if (tcp_saveti) 1296 m_freem(tcp_saveti); 1297 return; 1298 } 1299 } else if (th->th_ack == tp->snd_una && 1300 tp->segq.lh_first == NULL && 1301 tlen <= sbspace(&so->so_rcv)) { 1302 /* 1303 * this is a pure, in-sequence data packet 1304 * with nothing on the reassembly queue and 1305 * we have enough buffer space to take it. 1306 */ 1307 ++tcpstat.tcps_preddat; 1308 tp->rcv_nxt += tlen; 1309 tcpstat.tcps_rcvpack++; 1310 tcpstat.tcps_rcvbyte += tlen; 1311 ND6_HINT(tp); 1312 /* 1313 * Drop TCP, IP headers and TCP options then add data 1314 * to socket buffer. 1315 */ 1316 m_adj(m, toff + off); 1317 sbappend(&so->so_rcv, m); 1318 sorwakeup(so); 1319 TCP_SETUP_ACK(tp, th); 1320 if (tp->t_flags & TF_ACKNOW) 1321 (void) tcp_output(tp); 1322 if (tcp_saveti) 1323 m_freem(tcp_saveti); 1324 return; 1325 } 1326 } 1327 1328 /* 1329 * Compute mbuf offset to TCP data segment. 1330 */ 1331 hdroptlen = toff + off; 1332 1333 /* 1334 * Calculate amount of space in receive window, 1335 * and then do TCP input processing. 1336 * Receive window is amount of space in rcv queue, 1337 * but not less than advertised window. 1338 */ 1339 { int win; 1340 1341 win = sbspace(&so->so_rcv); 1342 if (win < 0) 1343 win = 0; 1344 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1345 } 1346 1347 switch (tp->t_state) { 1348 1349 /* 1350 * If the state is SYN_SENT: 1351 * if seg contains an ACK, but not for our SYN, drop the input. 1352 * if seg contains a RST, then drop the connection. 1353 * if seg does not contain SYN, then drop it. 1354 * Otherwise this is an acceptable SYN segment 1355 * initialize tp->rcv_nxt and tp->irs 1356 * if seg contains ack then advance tp->snd_una 1357 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1358 * arrange for segment to be acked (eventually) 1359 * continue processing rest of data/controls, beginning with URG 1360 */ 1361 case TCPS_SYN_SENT: 1362 if ((tiflags & TH_ACK) && 1363 (SEQ_LEQ(th->th_ack, tp->iss) || 1364 SEQ_GT(th->th_ack, tp->snd_max))) 1365 goto dropwithreset; 1366 if (tiflags & TH_RST) { 1367 if (tiflags & TH_ACK) 1368 tp = tcp_drop(tp, ECONNREFUSED); 1369 goto drop; 1370 } 1371 if ((tiflags & TH_SYN) == 0) 1372 goto drop; 1373 if (tiflags & TH_ACK) { 1374 tp->snd_una = tp->snd_recover = th->th_ack; 1375 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1376 tp->snd_nxt = tp->snd_una; 1377 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1378 } 1379 tp->irs = th->th_seq; 1380 tcp_rcvseqinit(tp); 1381 tp->t_flags |= TF_ACKNOW; 1382 tcp_mss_from_peer(tp, opti.maxseg); 1383 1384 /* 1385 * Initialize the initial congestion window. If we 1386 * had to retransmit the SYN, we must initialize cwnd 1387 * to 1 segment (i.e. the Loss Window). 1388 */ 1389 if (tp->t_flags & TF_SYN_REXMT) 1390 tp->snd_cwnd = tp->t_peermss; 1391 else 1392 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1393 tp->t_peermss); 1394 1395 tcp_rmx_rtt(tp); 1396 if (tiflags & TH_ACK) { 1397 tcpstat.tcps_connects++; 1398 soisconnected(so); 1399 tcp_established(tp); 1400 /* Do window scaling on this connection? */ 1401 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1402 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1403 tp->snd_scale = tp->requested_s_scale; 1404 tp->rcv_scale = tp->request_r_scale; 1405 } 1406 TCP_REASS_LOCK(tp); 1407 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1408 TCP_REASS_UNLOCK(tp); 1409 /* 1410 * if we didn't have to retransmit the SYN, 1411 * use its rtt as our initial srtt & rtt var. 1412 */ 1413 if (tp->t_rtt) 1414 tcp_xmit_timer(tp, tp->t_rtt); 1415 } else 1416 tp->t_state = TCPS_SYN_RECEIVED; 1417 1418 /* 1419 * Advance th->th_seq to correspond to first data byte. 1420 * If data, trim to stay within window, 1421 * dropping FIN if necessary. 1422 */ 1423 th->th_seq++; 1424 if (tlen > tp->rcv_wnd) { 1425 todrop = tlen - tp->rcv_wnd; 1426 m_adj(m, -todrop); 1427 tlen = tp->rcv_wnd; 1428 tiflags &= ~TH_FIN; 1429 tcpstat.tcps_rcvpackafterwin++; 1430 tcpstat.tcps_rcvbyteafterwin += todrop; 1431 } 1432 tp->snd_wl1 = th->th_seq - 1; 1433 tp->rcv_up = th->th_seq; 1434 goto step6; 1435 1436 /* 1437 * If the state is SYN_RECEIVED: 1438 * If seg contains an ACK, but not for our SYN, drop the input 1439 * and generate an RST. See page 36, rfc793 1440 */ 1441 case TCPS_SYN_RECEIVED: 1442 if ((tiflags & TH_ACK) && 1443 (SEQ_LEQ(th->th_ack, tp->iss) || 1444 SEQ_GT(th->th_ack, tp->snd_max))) 1445 goto dropwithreset; 1446 break; 1447 } 1448 1449 /* 1450 * States other than LISTEN or SYN_SENT. 1451 * First check timestamp, if present. 1452 * Then check that at least some bytes of segment are within 1453 * receive window. If segment begins before rcv_nxt, 1454 * drop leading data (and SYN); if nothing left, just ack. 1455 * 1456 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1457 * and it's less than ts_recent, drop it. 1458 */ 1459 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1460 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1461 1462 /* Check to see if ts_recent is over 24 days old. */ 1463 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1464 /* 1465 * Invalidate ts_recent. If this segment updates 1466 * ts_recent, the age will be reset later and ts_recent 1467 * will get a valid value. If it does not, setting 1468 * ts_recent to zero will at least satisfy the 1469 * requirement that zero be placed in the timestamp 1470 * echo reply when ts_recent isn't valid. The 1471 * age isn't reset until we get a valid ts_recent 1472 * because we don't want out-of-order segments to be 1473 * dropped when ts_recent is old. 1474 */ 1475 tp->ts_recent = 0; 1476 } else { 1477 tcpstat.tcps_rcvduppack++; 1478 tcpstat.tcps_rcvdupbyte += tlen; 1479 tcpstat.tcps_pawsdrop++; 1480 goto dropafterack; 1481 } 1482 } 1483 1484 todrop = tp->rcv_nxt - th->th_seq; 1485 if (todrop > 0) { 1486 if (tiflags & TH_SYN) { 1487 tiflags &= ~TH_SYN; 1488 th->th_seq++; 1489 if (th->th_urp > 1) 1490 th->th_urp--; 1491 else { 1492 tiflags &= ~TH_URG; 1493 th->th_urp = 0; 1494 } 1495 todrop--; 1496 } 1497 if (todrop > tlen || 1498 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1499 /* 1500 * Any valid FIN must be to the left of the window. 1501 * At this point the FIN must be a duplicate or 1502 * out of sequence; drop it. 1503 */ 1504 tiflags &= ~TH_FIN; 1505 /* 1506 * Send an ACK to resynchronize and drop any data. 1507 * But keep on processing for RST or ACK. 1508 */ 1509 tp->t_flags |= TF_ACKNOW; 1510 todrop = tlen; 1511 tcpstat.tcps_rcvdupbyte += todrop; 1512 tcpstat.tcps_rcvduppack++; 1513 } else { 1514 tcpstat.tcps_rcvpartduppack++; 1515 tcpstat.tcps_rcvpartdupbyte += todrop; 1516 } 1517 hdroptlen += todrop; /*drop from head afterwards*/ 1518 th->th_seq += todrop; 1519 tlen -= todrop; 1520 if (th->th_urp > todrop) 1521 th->th_urp -= todrop; 1522 else { 1523 tiflags &= ~TH_URG; 1524 th->th_urp = 0; 1525 } 1526 } 1527 1528 /* 1529 * If new data are received on a connection after the 1530 * user processes are gone, then RST the other end. 1531 */ 1532 if ((so->so_state & SS_NOFDREF) && 1533 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1534 tp = tcp_close(tp); 1535 tcpstat.tcps_rcvafterclose++; 1536 goto dropwithreset; 1537 } 1538 1539 /* 1540 * If segment ends after window, drop trailing data 1541 * (and PUSH and FIN); if nothing left, just ACK. 1542 */ 1543 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1544 if (todrop > 0) { 1545 tcpstat.tcps_rcvpackafterwin++; 1546 if (todrop >= tlen) { 1547 tcpstat.tcps_rcvbyteafterwin += tlen; 1548 /* 1549 * If a new connection request is received 1550 * while in TIME_WAIT, drop the old connection 1551 * and start over if the sequence numbers 1552 * are above the previous ones. 1553 */ 1554 if (tiflags & TH_SYN && 1555 tp->t_state == TCPS_TIME_WAIT && 1556 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1557 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1558 tp->snd_nxt); 1559 tp = tcp_close(tp); 1560 goto findpcb; 1561 } 1562 /* 1563 * If window is closed can only take segments at 1564 * window edge, and have to drop data and PUSH from 1565 * incoming segments. Continue processing, but 1566 * remember to ack. Otherwise, drop segment 1567 * and ack. 1568 */ 1569 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1570 tp->t_flags |= TF_ACKNOW; 1571 tcpstat.tcps_rcvwinprobe++; 1572 } else 1573 goto dropafterack; 1574 } else 1575 tcpstat.tcps_rcvbyteafterwin += todrop; 1576 m_adj(m, -todrop); 1577 tlen -= todrop; 1578 tiflags &= ~(TH_PUSH|TH_FIN); 1579 } 1580 1581 /* 1582 * If last ACK falls within this segment's sequence numbers, 1583 * and the timestamp is newer, record it. 1584 */ 1585 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1586 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1587 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1588 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1589 tp->ts_recent_age = tcp_now; 1590 tp->ts_recent = opti.ts_val; 1591 } 1592 1593 /* 1594 * If the RST bit is set examine the state: 1595 * SYN_RECEIVED STATE: 1596 * If passive open, return to LISTEN state. 1597 * If active open, inform user that connection was refused. 1598 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1599 * Inform user that connection was reset, and close tcb. 1600 * CLOSING, LAST_ACK, TIME_WAIT STATES 1601 * Close the tcb. 1602 */ 1603 if (tiflags&TH_RST) switch (tp->t_state) { 1604 1605 case TCPS_SYN_RECEIVED: 1606 so->so_error = ECONNREFUSED; 1607 goto close; 1608 1609 case TCPS_ESTABLISHED: 1610 case TCPS_FIN_WAIT_1: 1611 case TCPS_FIN_WAIT_2: 1612 case TCPS_CLOSE_WAIT: 1613 so->so_error = ECONNRESET; 1614 close: 1615 tp->t_state = TCPS_CLOSED; 1616 tcpstat.tcps_drops++; 1617 tp = tcp_close(tp); 1618 goto drop; 1619 1620 case TCPS_CLOSING: 1621 case TCPS_LAST_ACK: 1622 case TCPS_TIME_WAIT: 1623 tp = tcp_close(tp); 1624 goto drop; 1625 } 1626 1627 /* 1628 * If a SYN is in the window, then this is an 1629 * error and we send an RST and drop the connection. 1630 */ 1631 if (tiflags & TH_SYN) { 1632 tp = tcp_drop(tp, ECONNRESET); 1633 goto dropwithreset; 1634 } 1635 1636 /* 1637 * If the ACK bit is off we drop the segment and return. 1638 */ 1639 if ((tiflags & TH_ACK) == 0) { 1640 if (tp->t_flags & TF_ACKNOW) 1641 goto dropafterack; 1642 else 1643 goto drop; 1644 } 1645 1646 /* 1647 * Ack processing. 1648 */ 1649 switch (tp->t_state) { 1650 1651 /* 1652 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1653 * ESTABLISHED state and continue processing, otherwise 1654 * send an RST. 1655 */ 1656 case TCPS_SYN_RECEIVED: 1657 if (SEQ_GT(tp->snd_una, th->th_ack) || 1658 SEQ_GT(th->th_ack, tp->snd_max)) 1659 goto dropwithreset; 1660 tcpstat.tcps_connects++; 1661 soisconnected(so); 1662 tcp_established(tp); 1663 /* Do window scaling? */ 1664 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1665 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1666 tp->snd_scale = tp->requested_s_scale; 1667 tp->rcv_scale = tp->request_r_scale; 1668 } 1669 TCP_REASS_LOCK(tp); 1670 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1671 TCP_REASS_UNLOCK(tp); 1672 tp->snd_wl1 = th->th_seq - 1; 1673 /* fall into ... */ 1674 1675 /* 1676 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1677 * ACKs. If the ack is in the range 1678 * tp->snd_una < th->th_ack <= tp->snd_max 1679 * then advance tp->snd_una to th->th_ack and drop 1680 * data from the retransmission queue. If this ACK reflects 1681 * more up to date window information we update our window information. 1682 */ 1683 case TCPS_ESTABLISHED: 1684 case TCPS_FIN_WAIT_1: 1685 case TCPS_FIN_WAIT_2: 1686 case TCPS_CLOSE_WAIT: 1687 case TCPS_CLOSING: 1688 case TCPS_LAST_ACK: 1689 case TCPS_TIME_WAIT: 1690 1691 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1692 if (tlen == 0 && tiwin == tp->snd_wnd) { 1693 tcpstat.tcps_rcvdupack++; 1694 /* 1695 * If we have outstanding data (other than 1696 * a window probe), this is a completely 1697 * duplicate ack (ie, window info didn't 1698 * change), the ack is the biggest we've 1699 * seen and we've seen exactly our rexmt 1700 * threshhold of them, assume a packet 1701 * has been dropped and retransmit it. 1702 * Kludge snd_nxt & the congestion 1703 * window so we send only this one 1704 * packet. 1705 * 1706 * We know we're losing at the current 1707 * window size so do congestion avoidance 1708 * (set ssthresh to half the current window 1709 * and pull our congestion window back to 1710 * the new ssthresh). 1711 * 1712 * Dup acks mean that packets have left the 1713 * network (they're now cached at the receiver) 1714 * so bump cwnd by the amount in the receiver 1715 * to keep a constant cwnd packets in the 1716 * network. 1717 */ 1718 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1719 th->th_ack != tp->snd_una) 1720 tp->t_dupacks = 0; 1721 else if (++tp->t_dupacks == tcprexmtthresh) { 1722 tcp_seq onxt = tp->snd_nxt; 1723 u_int win = 1724 min(tp->snd_wnd, tp->snd_cwnd) / 1725 2 / tp->t_segsz; 1726 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1727 tp->snd_recover)) { 1728 /* 1729 * False fast retransmit after 1730 * timeout. Do not cut window. 1731 */ 1732 tp->snd_cwnd += tp->t_segsz; 1733 tp->t_dupacks = 0; 1734 (void) tcp_output(tp); 1735 goto drop; 1736 } 1737 1738 if (win < 2) 1739 win = 2; 1740 tp->snd_ssthresh = win * tp->t_segsz; 1741 tp->snd_recover = tp->snd_max; 1742 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1743 tp->t_rtt = 0; 1744 tp->snd_nxt = th->th_ack; 1745 tp->snd_cwnd = tp->t_segsz; 1746 (void) tcp_output(tp); 1747 tp->snd_cwnd = tp->snd_ssthresh + 1748 tp->t_segsz * tp->t_dupacks; 1749 if (SEQ_GT(onxt, tp->snd_nxt)) 1750 tp->snd_nxt = onxt; 1751 goto drop; 1752 } else if (tp->t_dupacks > tcprexmtthresh) { 1753 tp->snd_cwnd += tp->t_segsz; 1754 (void) tcp_output(tp); 1755 goto drop; 1756 } 1757 } else 1758 tp->t_dupacks = 0; 1759 break; 1760 } 1761 /* 1762 * If the congestion window was inflated to account 1763 * for the other side's cached packets, retract it. 1764 */ 1765 if (tcp_do_newreno == 0) { 1766 if (tp->t_dupacks >= tcprexmtthresh && 1767 tp->snd_cwnd > tp->snd_ssthresh) 1768 tp->snd_cwnd = tp->snd_ssthresh; 1769 tp->t_dupacks = 0; 1770 } else if (tp->t_dupacks >= tcprexmtthresh && 1771 tcp_newreno(tp, th) == 0) { 1772 tp->snd_cwnd = tp->snd_ssthresh; 1773 /* 1774 * Window inflation should have left us with approx. 1775 * snd_ssthresh outstanding data. But in case we 1776 * would be inclined to send a burst, better to do 1777 * it via the slow start mechanism. 1778 */ 1779 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1780 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1781 + tp->t_segsz; 1782 tp->t_dupacks = 0; 1783 } 1784 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1785 tcpstat.tcps_rcvacktoomuch++; 1786 goto dropafterack; 1787 } 1788 acked = th->th_ack - tp->snd_una; 1789 tcpstat.tcps_rcvackpack++; 1790 tcpstat.tcps_rcvackbyte += acked; 1791 1792 /* 1793 * If we have a timestamp reply, update smoothed 1794 * round trip time. If no timestamp is present but 1795 * transmit timer is running and timed sequence 1796 * number was acked, update smoothed round trip time. 1797 * Since we now have an rtt measurement, cancel the 1798 * timer backoff (cf., Phil Karn's retransmit alg.). 1799 * Recompute the initial retransmit timer. 1800 */ 1801 if (opti.ts_present && opti.ts_ecr) 1802 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1803 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1804 tcp_xmit_timer(tp,tp->t_rtt); 1805 1806 /* 1807 * If all outstanding data is acked, stop retransmit 1808 * timer and remember to restart (more output or persist). 1809 * If there is more data to be acked, restart retransmit 1810 * timer, using current (possibly backed-off) value. 1811 */ 1812 if (th->th_ack == tp->snd_max) { 1813 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1814 needoutput = 1; 1815 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1816 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1817 /* 1818 * When new data is acked, open the congestion window. 1819 * If the window gives us less than ssthresh packets 1820 * in flight, open exponentially (segsz per packet). 1821 * Otherwise open linearly: segsz per window 1822 * (segsz^2 / cwnd per packet), plus a constant 1823 * fraction of a packet (segsz/8) to help larger windows 1824 * open quickly enough. 1825 */ 1826 { 1827 u_int cw = tp->snd_cwnd; 1828 u_int incr = tp->t_segsz; 1829 1830 if (cw > tp->snd_ssthresh) 1831 incr = incr * incr / cw; 1832 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1833 tp->snd_cwnd = min(cw + incr, 1834 TCP_MAXWIN << tp->snd_scale); 1835 } 1836 ND6_HINT(tp); 1837 if (acked > so->so_snd.sb_cc) { 1838 tp->snd_wnd -= so->so_snd.sb_cc; 1839 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1840 ourfinisacked = 1; 1841 } else { 1842 sbdrop(&so->so_snd, acked); 1843 tp->snd_wnd -= acked; 1844 ourfinisacked = 0; 1845 } 1846 sowwakeup(so); 1847 /* 1848 * We want snd_recover to track snd_una to 1849 * avoid sequence wraparound problems for 1850 * very large transfers. 1851 */ 1852 tp->snd_una = tp->snd_recover = th->th_ack; 1853 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1854 tp->snd_nxt = tp->snd_una; 1855 1856 switch (tp->t_state) { 1857 1858 /* 1859 * In FIN_WAIT_1 STATE in addition to the processing 1860 * for the ESTABLISHED state if our FIN is now acknowledged 1861 * then enter FIN_WAIT_2. 1862 */ 1863 case TCPS_FIN_WAIT_1: 1864 if (ourfinisacked) { 1865 /* 1866 * If we can't receive any more 1867 * data, then closing user can proceed. 1868 * Starting the timer is contrary to the 1869 * specification, but if we don't get a FIN 1870 * we'll hang forever. 1871 */ 1872 if (so->so_state & SS_CANTRCVMORE) { 1873 soisdisconnected(so); 1874 if (tcp_maxidle > 0) 1875 TCP_TIMER_ARM(tp, TCPT_2MSL, 1876 tcp_maxidle); 1877 } 1878 tp->t_state = TCPS_FIN_WAIT_2; 1879 } 1880 break; 1881 1882 /* 1883 * In CLOSING STATE in addition to the processing for 1884 * the ESTABLISHED state if the ACK acknowledges our FIN 1885 * then enter the TIME-WAIT state, otherwise ignore 1886 * the segment. 1887 */ 1888 case TCPS_CLOSING: 1889 if (ourfinisacked) { 1890 tp->t_state = TCPS_TIME_WAIT; 1891 tcp_canceltimers(tp); 1892 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1893 soisdisconnected(so); 1894 } 1895 break; 1896 1897 /* 1898 * In LAST_ACK, we may still be waiting for data to drain 1899 * and/or to be acked, as well as for the ack of our FIN. 1900 * If our FIN is now acknowledged, delete the TCB, 1901 * enter the closed state and return. 1902 */ 1903 case TCPS_LAST_ACK: 1904 if (ourfinisacked) { 1905 tp = tcp_close(tp); 1906 goto drop; 1907 } 1908 break; 1909 1910 /* 1911 * In TIME_WAIT state the only thing that should arrive 1912 * is a retransmission of the remote FIN. Acknowledge 1913 * it and restart the finack timer. 1914 */ 1915 case TCPS_TIME_WAIT: 1916 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1917 goto dropafterack; 1918 } 1919 } 1920 1921 step6: 1922 /* 1923 * Update window information. 1924 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1925 */ 1926 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1927 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1928 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1929 /* keep track of pure window updates */ 1930 if (tlen == 0 && 1931 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1932 tcpstat.tcps_rcvwinupd++; 1933 tp->snd_wnd = tiwin; 1934 tp->snd_wl1 = th->th_seq; 1935 tp->snd_wl2 = th->th_ack; 1936 if (tp->snd_wnd > tp->max_sndwnd) 1937 tp->max_sndwnd = tp->snd_wnd; 1938 needoutput = 1; 1939 } 1940 1941 /* 1942 * Process segments with URG. 1943 */ 1944 if ((tiflags & TH_URG) && th->th_urp && 1945 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1946 /* 1947 * This is a kludge, but if we receive and accept 1948 * random urgent pointers, we'll crash in 1949 * soreceive. It's hard to imagine someone 1950 * actually wanting to send this much urgent data. 1951 */ 1952 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1953 th->th_urp = 0; /* XXX */ 1954 tiflags &= ~TH_URG; /* XXX */ 1955 goto dodata; /* XXX */ 1956 } 1957 /* 1958 * If this segment advances the known urgent pointer, 1959 * then mark the data stream. This should not happen 1960 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1961 * a FIN has been received from the remote side. 1962 * In these states we ignore the URG. 1963 * 1964 * According to RFC961 (Assigned Protocols), 1965 * the urgent pointer points to the last octet 1966 * of urgent data. We continue, however, 1967 * to consider it to indicate the first octet 1968 * of data past the urgent section as the original 1969 * spec states (in one of two places). 1970 */ 1971 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1972 tp->rcv_up = th->th_seq + th->th_urp; 1973 so->so_oobmark = so->so_rcv.sb_cc + 1974 (tp->rcv_up - tp->rcv_nxt) - 1; 1975 if (so->so_oobmark == 0) 1976 so->so_state |= SS_RCVATMARK; 1977 sohasoutofband(so); 1978 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1979 } 1980 /* 1981 * Remove out of band data so doesn't get presented to user. 1982 * This can happen independent of advancing the URG pointer, 1983 * but if two URG's are pending at once, some out-of-band 1984 * data may creep in... ick. 1985 */ 1986 if (th->th_urp <= (u_int16_t) tlen 1987 #ifdef SO_OOBINLINE 1988 && (so->so_options & SO_OOBINLINE) == 0 1989 #endif 1990 ) 1991 tcp_pulloutofband(so, th, m, hdroptlen); 1992 } else 1993 /* 1994 * If no out of band data is expected, 1995 * pull receive urgent pointer along 1996 * with the receive window. 1997 */ 1998 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1999 tp->rcv_up = tp->rcv_nxt; 2000 dodata: /* XXX */ 2001 2002 /* 2003 * Process the segment text, merging it into the TCP sequencing queue, 2004 * and arranging for acknowledgement of receipt if necessary. 2005 * This process logically involves adjusting tp->rcv_wnd as data 2006 * is presented to the user (this happens in tcp_usrreq.c, 2007 * case PRU_RCVD). If a FIN has already been received on this 2008 * connection then we just ignore the text. 2009 */ 2010 if ((tlen || (tiflags & TH_FIN)) && 2011 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2012 /* 2013 * Insert segment ti into reassembly queue of tcp with 2014 * control block tp. Return TH_FIN if reassembly now includes 2015 * a segment with FIN. The macro form does the common case 2016 * inline (segment is the next to be received on an 2017 * established connection, and the queue is empty), 2018 * avoiding linkage into and removal from the queue and 2019 * repetition of various conversions. 2020 * Set DELACK for segments received in order, but ack 2021 * immediately when segments are out of order 2022 * (so fast retransmit can work). 2023 */ 2024 /* NOTE: this was TCP_REASS() macro, but used only once */ 2025 TCP_REASS_LOCK(tp); 2026 if (th->th_seq == tp->rcv_nxt && 2027 tp->segq.lh_first == NULL && 2028 tp->t_state == TCPS_ESTABLISHED) { 2029 TCP_SETUP_ACK(tp, th); 2030 tp->rcv_nxt += tlen; 2031 tiflags = th->th_flags & TH_FIN; 2032 tcpstat.tcps_rcvpack++; 2033 tcpstat.tcps_rcvbyte += tlen; 2034 ND6_HINT(tp); 2035 m_adj(m, hdroptlen); 2036 sbappend(&(so)->so_rcv, m); 2037 sorwakeup(so); 2038 } else { 2039 m_adj(m, hdroptlen); 2040 tiflags = tcp_reass(tp, th, m, &tlen); 2041 tp->t_flags |= TF_ACKNOW; 2042 } 2043 TCP_REASS_UNLOCK(tp); 2044 2045 /* 2046 * Note the amount of data that peer has sent into 2047 * our window, in order to estimate the sender's 2048 * buffer size. 2049 */ 2050 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2051 } else { 2052 m_freem(m); 2053 m = NULL; 2054 tiflags &= ~TH_FIN; 2055 } 2056 2057 /* 2058 * If FIN is received ACK the FIN and let the user know 2059 * that the connection is closing. Ignore a FIN received before 2060 * the connection is fully established. 2061 */ 2062 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2063 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2064 socantrcvmore(so); 2065 tp->t_flags |= TF_ACKNOW; 2066 tp->rcv_nxt++; 2067 } 2068 switch (tp->t_state) { 2069 2070 /* 2071 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2072 */ 2073 case TCPS_ESTABLISHED: 2074 tp->t_state = TCPS_CLOSE_WAIT; 2075 break; 2076 2077 /* 2078 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2079 * enter the CLOSING state. 2080 */ 2081 case TCPS_FIN_WAIT_1: 2082 tp->t_state = TCPS_CLOSING; 2083 break; 2084 2085 /* 2086 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2087 * starting the time-wait timer, turning off the other 2088 * standard timers. 2089 */ 2090 case TCPS_FIN_WAIT_2: 2091 tp->t_state = TCPS_TIME_WAIT; 2092 tcp_canceltimers(tp); 2093 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2094 soisdisconnected(so); 2095 break; 2096 2097 /* 2098 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2099 */ 2100 case TCPS_TIME_WAIT: 2101 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2102 break; 2103 } 2104 } 2105 if (so->so_options & SO_DEBUG) { 2106 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2107 } 2108 2109 /* 2110 * Return any desired output. 2111 */ 2112 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2113 (void) tcp_output(tp); 2114 if (tcp_saveti) 2115 m_freem(tcp_saveti); 2116 return; 2117 2118 badsyn: 2119 /* 2120 * Received a bad SYN. Increment counters and dropwithreset. 2121 */ 2122 tcpstat.tcps_badsyn++; 2123 tp = NULL; 2124 goto dropwithreset; 2125 2126 dropafterack: 2127 /* 2128 * Generate an ACK dropping incoming segment if it occupies 2129 * sequence space, where the ACK reflects our state. 2130 */ 2131 if (tiflags & TH_RST) 2132 goto drop; 2133 m_freem(m); 2134 tp->t_flags |= TF_ACKNOW; 2135 (void) tcp_output(tp); 2136 if (tcp_saveti) 2137 m_freem(tcp_saveti); 2138 return; 2139 2140 dropwithreset_ratelim: 2141 /* 2142 * We may want to rate-limit RSTs in certain situations, 2143 * particularly if we are sending an RST in response to 2144 * an attempt to connect to or otherwise communicate with 2145 * a port for which we have no socket. 2146 */ 2147 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2148 tcp_rst_ppslim) == 0) { 2149 /* XXX stat */ 2150 goto drop; 2151 } 2152 /* ...fall into dropwithreset... */ 2153 2154 dropwithreset: 2155 /* 2156 * Generate a RST, dropping incoming segment. 2157 * Make ACK acceptable to originator of segment. 2158 */ 2159 if (tiflags & TH_RST) 2160 goto drop; 2161 { 2162 /* 2163 * need to recover version # field, which was overwritten on 2164 * ip_cksum computation. 2165 */ 2166 struct ip *sip; 2167 sip = mtod(m, struct ip *); 2168 switch (af) { 2169 case AF_INET: 2170 sip->ip_v = 4; 2171 break; 2172 #ifdef INET6 2173 case AF_INET6: 2174 sip->ip_v = 6; 2175 break; 2176 #endif 2177 } 2178 } 2179 if (tiflags & TH_ACK) 2180 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2181 else { 2182 if (tiflags & TH_SYN) 2183 tlen++; 2184 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2185 TH_RST|TH_ACK); 2186 } 2187 if (tcp_saveti) 2188 m_freem(tcp_saveti); 2189 return; 2190 2191 drop: 2192 /* 2193 * Drop space held by incoming segment and return. 2194 */ 2195 if (tp) { 2196 if (tp->t_inpcb) 2197 so = tp->t_inpcb->inp_socket; 2198 #ifdef INET6 2199 else if (tp->t_in6pcb) 2200 so = tp->t_in6pcb->in6p_socket; 2201 #endif 2202 else 2203 so = NULL; 2204 if (so && (so->so_options & SO_DEBUG) != 0) 2205 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2206 } 2207 if (tcp_saveti) 2208 m_freem(tcp_saveti); 2209 m_freem(m); 2210 return; 2211 } 2212 2213 void 2214 tcp_dooptions(tp, cp, cnt, th, oi) 2215 struct tcpcb *tp; 2216 u_char *cp; 2217 int cnt; 2218 struct tcphdr *th; 2219 struct tcp_opt_info *oi; 2220 { 2221 u_int16_t mss; 2222 int opt, optlen; 2223 2224 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2225 opt = cp[0]; 2226 if (opt == TCPOPT_EOL) 2227 break; 2228 if (opt == TCPOPT_NOP) 2229 optlen = 1; 2230 else { 2231 if (cnt < 2) 2232 break; 2233 optlen = cp[1]; 2234 if (optlen < 2 || optlen > cnt) 2235 break; 2236 } 2237 switch (opt) { 2238 2239 default: 2240 continue; 2241 2242 case TCPOPT_MAXSEG: 2243 if (optlen != TCPOLEN_MAXSEG) 2244 continue; 2245 if (!(th->th_flags & TH_SYN)) 2246 continue; 2247 bcopy(cp + 2, &mss, sizeof(mss)); 2248 oi->maxseg = ntohs(mss); 2249 break; 2250 2251 case TCPOPT_WINDOW: 2252 if (optlen != TCPOLEN_WINDOW) 2253 continue; 2254 if (!(th->th_flags & TH_SYN)) 2255 continue; 2256 tp->t_flags |= TF_RCVD_SCALE; 2257 tp->requested_s_scale = cp[2]; 2258 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2259 #if 0 /*XXX*/ 2260 char *p; 2261 2262 if (ip) 2263 p = ntohl(ip->ip_src); 2264 #ifdef INET6 2265 else if (ip6) 2266 p = ip6_sprintf(&ip6->ip6_src); 2267 #endif 2268 else 2269 p = "(unknown)"; 2270 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2271 "assuming %d\n", 2272 tp->requested_s_scale, p, 2273 TCP_MAX_WINSHIFT); 2274 #else 2275 log(LOG_ERR, "TCP: invalid wscale %d, " 2276 "assuming %d\n", 2277 tp->requested_s_scale, 2278 TCP_MAX_WINSHIFT); 2279 #endif 2280 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2281 } 2282 break; 2283 2284 case TCPOPT_TIMESTAMP: 2285 if (optlen != TCPOLEN_TIMESTAMP) 2286 continue; 2287 oi->ts_present = 1; 2288 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2289 NTOHL(oi->ts_val); 2290 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2291 NTOHL(oi->ts_ecr); 2292 2293 /* 2294 * A timestamp received in a SYN makes 2295 * it ok to send timestamp requests and replies. 2296 */ 2297 if (th->th_flags & TH_SYN) { 2298 tp->t_flags |= TF_RCVD_TSTMP; 2299 tp->ts_recent = oi->ts_val; 2300 tp->ts_recent_age = tcp_now; 2301 } 2302 break; 2303 case TCPOPT_SACK_PERMITTED: 2304 if (optlen != TCPOLEN_SACK_PERMITTED) 2305 continue; 2306 if (!(th->th_flags & TH_SYN)) 2307 continue; 2308 tp->t_flags &= ~TF_CANT_TXSACK; 2309 break; 2310 2311 case TCPOPT_SACK: 2312 if (tp->t_flags & TF_IGNR_RXSACK) 2313 continue; 2314 if (optlen % 8 != 2 || optlen < 10) 2315 continue; 2316 cp += 2; 2317 optlen -= 2; 2318 for (; optlen > 0; cp -= 8, optlen -= 8) { 2319 tcp_seq lwe, rwe; 2320 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2321 NTOHL(lwe); 2322 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2323 NTOHL(rwe); 2324 /* tcp_mark_sacked(tp, lwe, rwe); */ 2325 } 2326 break; 2327 } 2328 } 2329 } 2330 2331 /* 2332 * Pull out of band byte out of a segment so 2333 * it doesn't appear in the user's data queue. 2334 * It is still reflected in the segment length for 2335 * sequencing purposes. 2336 */ 2337 void 2338 tcp_pulloutofband(so, th, m, off) 2339 struct socket *so; 2340 struct tcphdr *th; 2341 struct mbuf *m; 2342 int off; 2343 { 2344 int cnt = off + th->th_urp - 1; 2345 2346 while (cnt >= 0) { 2347 if (m->m_len > cnt) { 2348 char *cp = mtod(m, caddr_t) + cnt; 2349 struct tcpcb *tp = sototcpcb(so); 2350 2351 tp->t_iobc = *cp; 2352 tp->t_oobflags |= TCPOOB_HAVEDATA; 2353 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2354 m->m_len--; 2355 return; 2356 } 2357 cnt -= m->m_len; 2358 m = m->m_next; 2359 if (m == 0) 2360 break; 2361 } 2362 panic("tcp_pulloutofband"); 2363 } 2364 2365 /* 2366 * Collect new round-trip time estimate 2367 * and update averages and current timeout. 2368 */ 2369 void 2370 tcp_xmit_timer(tp, rtt) 2371 struct tcpcb *tp; 2372 short rtt; 2373 { 2374 short delta; 2375 short rttmin; 2376 2377 tcpstat.tcps_rttupdated++; 2378 --rtt; 2379 if (tp->t_srtt != 0) { 2380 /* 2381 * srtt is stored as fixed point with 3 bits after the 2382 * binary point (i.e., scaled by 8). The following magic 2383 * is equivalent to the smoothing algorithm in rfc793 with 2384 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2385 * point). Adjust rtt to origin 0. 2386 */ 2387 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2388 if ((tp->t_srtt += delta) <= 0) 2389 tp->t_srtt = 1 << 2; 2390 /* 2391 * We accumulate a smoothed rtt variance (actually, a 2392 * smoothed mean difference), then set the retransmit 2393 * timer to smoothed rtt + 4 times the smoothed variance. 2394 * rttvar is stored as fixed point with 2 bits after the 2395 * binary point (scaled by 4). The following is 2396 * equivalent to rfc793 smoothing with an alpha of .75 2397 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2398 * rfc793's wired-in beta. 2399 */ 2400 if (delta < 0) 2401 delta = -delta; 2402 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2403 if ((tp->t_rttvar += delta) <= 0) 2404 tp->t_rttvar = 1 << 2; 2405 } else { 2406 /* 2407 * No rtt measurement yet - use the unsmoothed rtt. 2408 * Set the variance to half the rtt (so our first 2409 * retransmit happens at 3*rtt). 2410 */ 2411 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2412 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2413 } 2414 tp->t_rtt = 0; 2415 tp->t_rxtshift = 0; 2416 2417 /* 2418 * the retransmit should happen at rtt + 4 * rttvar. 2419 * Because of the way we do the smoothing, srtt and rttvar 2420 * will each average +1/2 tick of bias. When we compute 2421 * the retransmit timer, we want 1/2 tick of rounding and 2422 * 1 extra tick because of +-1/2 tick uncertainty in the 2423 * firing of the timer. The bias will give us exactly the 2424 * 1.5 tick we need. But, because the bias is 2425 * statistical, we have to test that we don't drop below 2426 * the minimum feasible timer (which is 2 ticks). 2427 */ 2428 if (tp->t_rttmin > rtt + 2) 2429 rttmin = tp->t_rttmin; 2430 else 2431 rttmin = rtt + 2; 2432 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2433 2434 /* 2435 * We received an ack for a packet that wasn't retransmitted; 2436 * it is probably safe to discard any error indications we've 2437 * received recently. This isn't quite right, but close enough 2438 * for now (a route might have failed after we sent a segment, 2439 * and the return path might not be symmetrical). 2440 */ 2441 tp->t_softerror = 0; 2442 } 2443 2444 /* 2445 * Checks for partial ack. If partial ack arrives, force the retransmission 2446 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2447 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2448 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2449 */ 2450 int 2451 tcp_newreno(tp, th) 2452 struct tcpcb *tp; 2453 struct tcphdr *th; 2454 { 2455 tcp_seq onxt = tp->snd_nxt; 2456 u_long ocwnd = tp->snd_cwnd; 2457 2458 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2459 /* 2460 * snd_una has not yet been updated and the socket's send 2461 * buffer has not yet drained off the ACK'd data, so we 2462 * have to leave snd_una as it was to get the correct data 2463 * offset in tcp_output(). 2464 */ 2465 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2466 tp->t_rtt = 0; 2467 tp->snd_nxt = th->th_ack; 2468 /* 2469 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2470 * is not yet updated when we're called. 2471 */ 2472 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2473 (void) tcp_output(tp); 2474 tp->snd_cwnd = ocwnd; 2475 if (SEQ_GT(onxt, tp->snd_nxt)) 2476 tp->snd_nxt = onxt; 2477 /* 2478 * Partial window deflation. Relies on fact that tp->snd_una 2479 * not updated yet. 2480 */ 2481 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2482 return 1; 2483 } 2484 return 0; 2485 } 2486 2487 2488 /* 2489 * TCP compressed state engine. Currently used to hold compressed 2490 * state for SYN_RECEIVED. 2491 */ 2492 2493 u_long syn_cache_count; 2494 u_int32_t syn_hash1, syn_hash2; 2495 2496 #define SYN_HASH(sa, sp, dp) \ 2497 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2498 ((u_int32_t)(sp)))^syn_hash2))) 2499 #ifndef INET6 2500 #define SYN_HASHALL(hash, src, dst) \ 2501 do { \ 2502 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2503 ((struct sockaddr_in *)(src))->sin_port, \ 2504 ((struct sockaddr_in *)(dst))->sin_port); \ 2505 } while (0) 2506 #else 2507 #define SYN_HASH6(sa, sp, dp) \ 2508 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2509 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2510 & 0x7fffffff) 2511 2512 #define SYN_HASHALL(hash, src, dst) \ 2513 do { \ 2514 switch ((src)->sa_family) { \ 2515 case AF_INET: \ 2516 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2517 ((struct sockaddr_in *)(src))->sin_port, \ 2518 ((struct sockaddr_in *)(dst))->sin_port); \ 2519 break; \ 2520 case AF_INET6: \ 2521 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2522 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2523 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2524 break; \ 2525 default: \ 2526 hash = 0; \ 2527 } \ 2528 } while (0) 2529 #endif /* INET6 */ 2530 2531 #define SYN_CACHE_RM(sc) \ 2532 do { \ 2533 LIST_REMOVE((sc), sc_bucketq); \ 2534 (sc)->sc_tp = NULL; \ 2535 LIST_REMOVE((sc), sc_tpq); \ 2536 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2537 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2538 syn_cache_count--; \ 2539 } while (0) 2540 2541 #define SYN_CACHE_PUT(sc) \ 2542 do { \ 2543 if ((sc)->sc_ipopts) \ 2544 (void) m_free((sc)->sc_ipopts); \ 2545 if ((sc)->sc_route4.ro_rt != NULL) \ 2546 RTFREE((sc)->sc_route4.ro_rt); \ 2547 pool_put(&syn_cache_pool, (sc)); \ 2548 } while (0) 2549 2550 struct pool syn_cache_pool; 2551 2552 /* 2553 * We don't estimate RTT with SYNs, so each packet starts with the default 2554 * RTT and each timer queue has a fixed timeout value. This allows us to 2555 * optimize the timer queues somewhat. 2556 */ 2557 #define SYN_CACHE_TIMER_ARM(sc) \ 2558 do { \ 2559 TCPT_RANGESET((sc)->sc_rxtcur, \ 2560 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2561 TCPTV_REXMTMAX); \ 2562 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2563 } while (0) 2564 2565 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2566 2567 void 2568 syn_cache_init() 2569 { 2570 int i; 2571 2572 /* Initialize the hash buckets. */ 2573 for (i = 0; i < tcp_syn_cache_size; i++) 2574 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2575 2576 /* Initialize the timer queues. */ 2577 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2578 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2579 2580 /* Initialize the syn cache pool. */ 2581 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2582 "synpl", 0, NULL, NULL, M_PCB); 2583 } 2584 2585 void 2586 syn_cache_insert(sc, tp) 2587 struct syn_cache *sc; 2588 struct tcpcb *tp; 2589 { 2590 struct syn_cache_head *scp; 2591 struct syn_cache *sc2; 2592 int s, i; 2593 2594 /* 2595 * If there are no entries in the hash table, reinitialize 2596 * the hash secrets. 2597 */ 2598 if (syn_cache_count == 0) { 2599 struct timeval tv; 2600 microtime(&tv); 2601 syn_hash1 = random() ^ (u_long)≻ 2602 syn_hash2 = random() ^ tv.tv_usec; 2603 } 2604 2605 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2606 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2607 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2608 2609 /* 2610 * Make sure that we don't overflow the per-bucket 2611 * limit or the total cache size limit. 2612 */ 2613 s = splsoftnet(); 2614 if (scp->sch_length >= tcp_syn_bucket_limit) { 2615 tcpstat.tcps_sc_bucketoverflow++; 2616 /* 2617 * The bucket is full. Toss the oldest element in the 2618 * bucket. This will be the entry with our bucket 2619 * index closest to the front of the timer queue with 2620 * the largest timeout value. 2621 * 2622 * Note: This timer queue traversal may be expensive, so 2623 * we hope that this doesn't happen very often. It is 2624 * much more likely that we'll overflow the entire 2625 * cache, which is much easier to handle; see below. 2626 */ 2627 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2628 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2629 sc2 != NULL; 2630 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2631 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2632 SYN_CACHE_RM(sc2); 2633 SYN_CACHE_PUT(sc2); 2634 goto insert; /* 2 level break */ 2635 } 2636 } 2637 } 2638 #ifdef DIAGNOSTIC 2639 /* 2640 * This should never happen; we should always find an 2641 * entry in our bucket. 2642 */ 2643 panic("syn_cache_insert: bucketoverflow: impossible"); 2644 #endif 2645 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2646 tcpstat.tcps_sc_overflowed++; 2647 /* 2648 * The cache is full. Toss the oldest entry in the 2649 * entire cache. This is the front entry in the 2650 * first non-empty timer queue with the largest 2651 * timeout value. 2652 */ 2653 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2654 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2655 if (sc2 == NULL) 2656 continue; 2657 SYN_CACHE_RM(sc2); 2658 SYN_CACHE_PUT(sc2); 2659 goto insert; /* symmetry with above */ 2660 } 2661 #ifdef DIAGNOSTIC 2662 /* 2663 * This should never happen; we should always find an 2664 * entry in the cache. 2665 */ 2666 panic("syn_cache_insert: cache overflow: impossible"); 2667 #endif 2668 } 2669 2670 insert: 2671 /* 2672 * Initialize the entry's timer. 2673 */ 2674 sc->sc_rxttot = 0; 2675 sc->sc_rxtshift = 0; 2676 SYN_CACHE_TIMER_ARM(sc); 2677 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2678 2679 /* Link it from tcpcb entry */ 2680 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2681 2682 /* Put it into the bucket. */ 2683 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2684 scp->sch_length++; 2685 syn_cache_count++; 2686 2687 tcpstat.tcps_sc_added++; 2688 splx(s); 2689 } 2690 2691 /* 2692 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2693 * If we have retransmitted an entry the maximum number of times, expire 2694 * that entry. 2695 */ 2696 void 2697 syn_cache_timer() 2698 { 2699 struct syn_cache *sc, *nsc; 2700 int i, s; 2701 2702 s = splsoftnet(); 2703 2704 /* 2705 * First, get all the entries that need to be retransmitted, or 2706 * must be expired due to exceeding the initial keepalive time. 2707 */ 2708 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2709 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2710 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2711 sc = nsc) { 2712 nsc = TAILQ_NEXT(sc, sc_timeq); 2713 2714 /* 2715 * Compute the total amount of time this entry has 2716 * been on a queue. If this entry has been on longer 2717 * than the keep alive timer would allow, expire it. 2718 */ 2719 sc->sc_rxttot += sc->sc_rxtcur; 2720 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2721 tcpstat.tcps_sc_timed_out++; 2722 SYN_CACHE_RM(sc); 2723 SYN_CACHE_PUT(sc); 2724 continue; 2725 } 2726 2727 tcpstat.tcps_sc_retransmitted++; 2728 (void) syn_cache_respond(sc, NULL); 2729 2730 /* Advance this entry onto the next timer queue. */ 2731 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2732 sc->sc_rxtshift = i + 1; 2733 SYN_CACHE_TIMER_ARM(sc); 2734 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2735 sc, sc_timeq); 2736 } 2737 } 2738 2739 /* 2740 * Now get all the entries that are expired due to too many 2741 * retransmissions. 2742 */ 2743 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2744 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2745 sc = nsc) { 2746 nsc = TAILQ_NEXT(sc, sc_timeq); 2747 tcpstat.tcps_sc_timed_out++; 2748 SYN_CACHE_RM(sc); 2749 SYN_CACHE_PUT(sc); 2750 } 2751 splx(s); 2752 } 2753 2754 /* 2755 * Remove syn cache created by the specified tcb entry, 2756 * because this does not make sense to keep them 2757 * (if there's no tcb entry, syn cache entry will never be used) 2758 */ 2759 void 2760 syn_cache_cleanup(tp) 2761 struct tcpcb *tp; 2762 { 2763 struct syn_cache *sc, *nsc; 2764 int s; 2765 2766 s = splsoftnet(); 2767 2768 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2769 nsc = LIST_NEXT(sc, sc_tpq); 2770 2771 #ifdef DIAGNOSTIC 2772 if (sc->sc_tp != tp) 2773 panic("invalid sc_tp in syn_cache_cleanup"); 2774 #endif 2775 SYN_CACHE_RM(sc); 2776 SYN_CACHE_PUT(sc); 2777 } 2778 /* just for safety */ 2779 LIST_INIT(&tp->t_sc); 2780 2781 splx(s); 2782 } 2783 2784 /* 2785 * Find an entry in the syn cache. 2786 */ 2787 struct syn_cache * 2788 syn_cache_lookup(src, dst, headp) 2789 struct sockaddr *src; 2790 struct sockaddr *dst; 2791 struct syn_cache_head **headp; 2792 { 2793 struct syn_cache *sc; 2794 struct syn_cache_head *scp; 2795 u_int32_t hash; 2796 int s; 2797 2798 SYN_HASHALL(hash, src, dst); 2799 2800 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2801 *headp = scp; 2802 s = splsoftnet(); 2803 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2804 sc = LIST_NEXT(sc, sc_bucketq)) { 2805 if (sc->sc_hash != hash) 2806 continue; 2807 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2808 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2809 splx(s); 2810 return (sc); 2811 } 2812 } 2813 splx(s); 2814 return (NULL); 2815 } 2816 2817 /* 2818 * This function gets called when we receive an ACK for a 2819 * socket in the LISTEN state. We look up the connection 2820 * in the syn cache, and if its there, we pull it out of 2821 * the cache and turn it into a full-blown connection in 2822 * the SYN-RECEIVED state. 2823 * 2824 * The return values may not be immediately obvious, and their effects 2825 * can be subtle, so here they are: 2826 * 2827 * NULL SYN was not found in cache; caller should drop the 2828 * packet and send an RST. 2829 * 2830 * -1 We were unable to create the new connection, and are 2831 * aborting it. An ACK,RST is being sent to the peer 2832 * (unless we got screwey sequence numbners; see below), 2833 * because the 3-way handshake has been completed. Caller 2834 * should not free the mbuf, since we may be using it. If 2835 * we are not, we will free it. 2836 * 2837 * Otherwise, the return value is a pointer to the new socket 2838 * associated with the connection. 2839 */ 2840 struct socket * 2841 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2842 struct sockaddr *src; 2843 struct sockaddr *dst; 2844 struct tcphdr *th; 2845 unsigned int hlen, tlen; 2846 struct socket *so; 2847 struct mbuf *m; 2848 { 2849 struct syn_cache *sc; 2850 struct syn_cache_head *scp; 2851 struct inpcb *inp = NULL; 2852 #ifdef INET6 2853 struct in6pcb *in6p = NULL; 2854 #endif 2855 struct tcpcb *tp = 0; 2856 struct mbuf *am; 2857 int s; 2858 struct socket *oso; 2859 2860 s = splsoftnet(); 2861 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2862 splx(s); 2863 return (NULL); 2864 } 2865 2866 /* 2867 * Verify the sequence and ack numbers. Try getting the correct 2868 * response again. 2869 */ 2870 if ((th->th_ack != sc->sc_iss + 1) || 2871 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2872 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2873 (void) syn_cache_respond(sc, m); 2874 splx(s); 2875 return ((struct socket *)(-1)); 2876 } 2877 2878 /* Remove this cache entry */ 2879 SYN_CACHE_RM(sc); 2880 splx(s); 2881 2882 /* 2883 * Ok, create the full blown connection, and set things up 2884 * as they would have been set up if we had created the 2885 * connection when the SYN arrived. If we can't create 2886 * the connection, abort it. 2887 */ 2888 /* 2889 * inp still has the OLD in_pcb stuff, set the 2890 * v6-related flags on the new guy, too. This is 2891 * done particularly for the case where an AF_INET6 2892 * socket is bound only to a port, and a v4 connection 2893 * comes in on that port. 2894 * we also copy the flowinfo from the original pcb 2895 * to the new one. 2896 */ 2897 { 2898 struct inpcb *parentinpcb; 2899 2900 parentinpcb = (struct inpcb *)so->so_pcb; 2901 2902 oso = so; 2903 so = sonewconn(so, SS_ISCONNECTED); 2904 if (so == NULL) 2905 goto resetandabort; 2906 2907 switch (so->so_proto->pr_domain->dom_family) { 2908 case AF_INET: 2909 inp = sotoinpcb(so); 2910 break; 2911 #ifdef INET6 2912 case AF_INET6: 2913 in6p = sotoin6pcb(so); 2914 #if 0 /*def INET6*/ 2915 inp->inp_flags |= (parentinpcb->inp_flags & 2916 (INP_IPV6 | INP_IPV6_UNDEC | INP_IPV6_MAPPED)); 2917 if ((inp->inp_flags & INP_IPV6) && 2918 !(inp->inp_flags & INP_IPV6_MAPPED)) { 2919 inp->inp_ipv6.ip6_hlim = parentinpcb->inp_ipv6.ip6_hlim; 2920 inp->inp_ipv6.ip6_vfc = parentinpcb->inp_ipv6.ip6_vfc; 2921 } 2922 #endif 2923 break; 2924 #endif 2925 } 2926 } 2927 switch (src->sa_family) { 2928 case AF_INET: 2929 if (inp) { 2930 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2931 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2932 inp->inp_options = ip_srcroute(); 2933 in_pcbstate(inp, INP_BOUND); 2934 if (inp->inp_options == NULL) { 2935 inp->inp_options = sc->sc_ipopts; 2936 sc->sc_ipopts = NULL; 2937 } 2938 } 2939 #ifdef INET6 2940 else if (in6p) { 2941 /* IPv4 packet to AF_INET6 socket */ 2942 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2943 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2944 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2945 &in6p->in6p_laddr.s6_addr32[3], 2946 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2947 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2948 in6totcpcb(in6p)->t_family = AF_INET; 2949 } 2950 #endif 2951 break; 2952 #ifdef INET6 2953 case AF_INET6: 2954 if (in6p) { 2955 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 2956 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 2957 #if 0 2958 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 2959 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 2960 #endif 2961 } 2962 break; 2963 #endif 2964 } 2965 #ifdef INET6 2966 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 2967 struct in6pcb *oin6p = sotoin6pcb(oso); 2968 /* inherit socket options from the listening socket */ 2969 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 2970 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 2971 m_freem(in6p->in6p_options); 2972 in6p->in6p_options = 0; 2973 } 2974 ip6_savecontrol(in6p, &in6p->in6p_options, 2975 mtod(m, struct ip6_hdr *), m); 2976 } 2977 #endif 2978 2979 #ifdef IPSEC 2980 /* 2981 * we make a copy of policy, instead of sharing the policy, 2982 * for better behavior in terms of SA lookup and dead SA removal. 2983 */ 2984 if (inp) { 2985 /* copy old policy into new socket's */ 2986 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 2987 printf("tcp_input: could not copy policy\n"); 2988 } 2989 #ifdef INET6 2990 else if (in6p) { 2991 /* copy old policy into new socket's */ 2992 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 2993 printf("tcp_input: could not copy policy\n"); 2994 } 2995 #endif 2996 #endif 2997 2998 /* 2999 * Give the new socket our cached route reference. 3000 */ 3001 if (inp) 3002 inp->inp_route = sc->sc_route4; /* struct assignment */ 3003 #ifdef INET6 3004 else 3005 in6p->in6p_route = sc->sc_route6; 3006 #endif 3007 sc->sc_route4.ro_rt = NULL; 3008 3009 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3010 if (am == NULL) 3011 goto resetandabort; 3012 am->m_len = src->sa_len; 3013 bcopy(src, mtod(am, caddr_t), src->sa_len); 3014 if (inp) { 3015 if (in_pcbconnect(inp, am)) { 3016 (void) m_free(am); 3017 goto resetandabort; 3018 } 3019 } 3020 #ifdef INET6 3021 else if (in6p) { 3022 if (src->sa_family == AF_INET) { 3023 /* IPv4 packet to AF_INET6 socket */ 3024 struct sockaddr_in6 *sin6; 3025 sin6 = mtod(am, struct sockaddr_in6 *); 3026 am->m_len = sizeof(*sin6); 3027 bzero(sin6, sizeof(*sin6)); 3028 sin6->sin6_family = AF_INET6; 3029 sin6->sin6_len = sizeof(*sin6); 3030 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3031 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3032 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3033 &sin6->sin6_addr.s6_addr32[3], 3034 sizeof(sin6->sin6_addr.s6_addr32[3])); 3035 } 3036 if (in6_pcbconnect(in6p, am)) { 3037 (void) m_free(am); 3038 goto resetandabort; 3039 } 3040 } 3041 #endif 3042 else { 3043 (void) m_free(am); 3044 goto resetandabort; 3045 } 3046 (void) m_free(am); 3047 3048 if (inp) 3049 tp = intotcpcb(inp); 3050 #ifdef INET6 3051 else if (in6p) 3052 tp = in6totcpcb(in6p); 3053 #endif 3054 else 3055 tp = NULL; 3056 if (sc->sc_request_r_scale != 15) { 3057 tp->requested_s_scale = sc->sc_requested_s_scale; 3058 tp->request_r_scale = sc->sc_request_r_scale; 3059 tp->snd_scale = sc->sc_requested_s_scale; 3060 tp->rcv_scale = sc->sc_request_r_scale; 3061 tp->t_flags |= TF_RCVD_SCALE; 3062 } 3063 if (sc->sc_flags & SCF_TIMESTAMP) 3064 tp->t_flags |= TF_RCVD_TSTMP; 3065 3066 tp->t_template = tcp_template(tp); 3067 if (tp->t_template == 0) { 3068 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3069 so = NULL; 3070 m_freem(m); 3071 goto abort; 3072 } 3073 3074 tp->iss = sc->sc_iss; 3075 tp->irs = sc->sc_irs; 3076 tcp_sendseqinit(tp); 3077 tcp_rcvseqinit(tp); 3078 tp->t_state = TCPS_SYN_RECEIVED; 3079 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3080 tcpstat.tcps_accepts++; 3081 3082 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3083 tp->t_ourmss = sc->sc_ourmaxseg; 3084 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3085 3086 /* 3087 * Initialize the initial congestion window. If we 3088 * had to retransmit the SYN,ACK, we must initialize cwnd 3089 * to 1 segment (i.e. the Loss Window). 3090 */ 3091 if (sc->sc_rxtshift) 3092 tp->snd_cwnd = tp->t_peermss; 3093 else 3094 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3095 3096 tcp_rmx_rtt(tp); 3097 tp->snd_wl1 = sc->sc_irs; 3098 tp->rcv_up = sc->sc_irs + 1; 3099 3100 /* 3101 * This is what whould have happened in tcp_ouput() when 3102 * the SYN,ACK was sent. 3103 */ 3104 tp->snd_up = tp->snd_una; 3105 tp->snd_max = tp->snd_nxt = tp->iss+1; 3106 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3107 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3108 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3109 tp->last_ack_sent = tp->rcv_nxt; 3110 3111 tcpstat.tcps_sc_completed++; 3112 SYN_CACHE_PUT(sc); 3113 return (so); 3114 3115 resetandabort: 3116 (void) tcp_respond(NULL, m, m, th, 3117 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3118 abort: 3119 if (so != NULL) 3120 (void) soabort(so); 3121 SYN_CACHE_PUT(sc); 3122 tcpstat.tcps_sc_aborted++; 3123 return ((struct socket *)(-1)); 3124 } 3125 3126 /* 3127 * This function is called when we get a RST for a 3128 * non-existant connection, so that we can see if the 3129 * connection is in the syn cache. If it is, zap it. 3130 */ 3131 3132 void 3133 syn_cache_reset(src, dst, th) 3134 struct sockaddr *src; 3135 struct sockaddr *dst; 3136 struct tcphdr *th; 3137 { 3138 struct syn_cache *sc; 3139 struct syn_cache_head *scp; 3140 int s = splsoftnet(); 3141 3142 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3143 splx(s); 3144 return; 3145 } 3146 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3147 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3148 splx(s); 3149 return; 3150 } 3151 SYN_CACHE_RM(sc); 3152 splx(s); 3153 tcpstat.tcps_sc_reset++; 3154 SYN_CACHE_PUT(sc); 3155 } 3156 3157 void 3158 syn_cache_unreach(src, dst, th) 3159 struct sockaddr *src; 3160 struct sockaddr *dst; 3161 struct tcphdr *th; 3162 { 3163 struct syn_cache *sc; 3164 struct syn_cache_head *scp; 3165 int s; 3166 3167 s = splsoftnet(); 3168 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3169 splx(s); 3170 return; 3171 } 3172 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3173 if (ntohl (th->th_seq) != sc->sc_iss) { 3174 splx(s); 3175 return; 3176 } 3177 3178 /* 3179 * If we've rertransmitted 3 times and this is our second error, 3180 * we remove the entry. Otherwise, we allow it to continue on. 3181 * This prevents us from incorrectly nuking an entry during a 3182 * spurious network outage. 3183 * 3184 * See tcp_notify(). 3185 */ 3186 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3187 sc->sc_flags |= SCF_UNREACH; 3188 splx(s); 3189 return; 3190 } 3191 3192 SYN_CACHE_RM(sc); 3193 splx(s); 3194 tcpstat.tcps_sc_unreach++; 3195 SYN_CACHE_PUT(sc); 3196 } 3197 3198 /* 3199 * Given a LISTEN socket and an inbound SYN request, add 3200 * this to the syn cache, and send back a segment: 3201 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3202 * to the source. 3203 * 3204 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3205 * Doing so would require that we hold onto the data and deliver it 3206 * to the application. However, if we are the target of a SYN-flood 3207 * DoS attack, an attacker could send data which would eventually 3208 * consume all available buffer space if it were ACKed. By not ACKing 3209 * the data, we avoid this DoS scenario. 3210 */ 3211 3212 int 3213 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3214 struct sockaddr *src; 3215 struct sockaddr *dst; 3216 struct tcphdr *th; 3217 unsigned int hlen; 3218 struct socket *so; 3219 struct mbuf *m; 3220 u_char *optp; 3221 int optlen; 3222 struct tcp_opt_info *oi; 3223 { 3224 struct tcpcb tb, *tp; 3225 long win; 3226 struct syn_cache *sc; 3227 struct syn_cache_head *scp; 3228 struct mbuf *ipopts; 3229 3230 tp = sototcpcb(so); 3231 3232 /* 3233 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3234 * 3235 * Note this check is performed in tcp_input() very early on. 3236 */ 3237 3238 /* 3239 * Initialize some local state. 3240 */ 3241 win = sbspace(&so->so_rcv); 3242 if (win > TCP_MAXWIN) 3243 win = TCP_MAXWIN; 3244 3245 if (src->sa_family == AF_INET) { 3246 /* 3247 * Remember the IP options, if any. 3248 */ 3249 ipopts = ip_srcroute(); 3250 } else 3251 ipopts = NULL; 3252 3253 if (optp) { 3254 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3255 tcp_dooptions(&tb, optp, optlen, th, oi); 3256 } else 3257 tb.t_flags = 0; 3258 3259 /* 3260 * See if we already have an entry for this connection. 3261 * If we do, resend the SYN,ACK. We do not count this 3262 * as a retransmission (XXX though maybe we should). 3263 */ 3264 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3265 tcpstat.tcps_sc_dupesyn++; 3266 if (ipopts) { 3267 /* 3268 * If we were remembering a previous source route, 3269 * forget it and use the new one we've been given. 3270 */ 3271 if (sc->sc_ipopts) 3272 (void) m_free(sc->sc_ipopts); 3273 sc->sc_ipopts = ipopts; 3274 } 3275 sc->sc_timestamp = tb.ts_recent; 3276 if (syn_cache_respond(sc, m) == 0) { 3277 tcpstat.tcps_sndacks++; 3278 tcpstat.tcps_sndtotal++; 3279 } 3280 return (1); 3281 } 3282 3283 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3284 if (sc == NULL) { 3285 if (ipopts) 3286 (void) m_free(ipopts); 3287 return (0); 3288 } 3289 3290 /* 3291 * Fill in the cache, and put the necessary IP and TCP 3292 * options into the reply. 3293 */ 3294 bzero(sc, sizeof(struct syn_cache)); 3295 bcopy(src, &sc->sc_src, src->sa_len); 3296 bcopy(dst, &sc->sc_dst, dst->sa_len); 3297 sc->sc_flags = 0; 3298 sc->sc_ipopts = ipopts; 3299 sc->sc_irs = th->th_seq; 3300 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 3301 sc->sc_peermaxseg = oi->maxseg; 3302 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3303 m->m_pkthdr.rcvif : NULL, 3304 sc->sc_src.sa.sa_family); 3305 sc->sc_win = win; 3306 sc->sc_timestamp = tb.ts_recent; 3307 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3308 sc->sc_flags |= SCF_TIMESTAMP; 3309 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3310 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3311 sc->sc_requested_s_scale = tb.requested_s_scale; 3312 sc->sc_request_r_scale = 0; 3313 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3314 TCP_MAXWIN << sc->sc_request_r_scale < 3315 so->so_rcv.sb_hiwat) 3316 sc->sc_request_r_scale++; 3317 } else { 3318 sc->sc_requested_s_scale = 15; 3319 sc->sc_request_r_scale = 15; 3320 } 3321 sc->sc_tp = tp; 3322 if (syn_cache_respond(sc, m) == 0) { 3323 syn_cache_insert(sc, tp); 3324 tcpstat.tcps_sndacks++; 3325 tcpstat.tcps_sndtotal++; 3326 } else { 3327 SYN_CACHE_PUT(sc); 3328 tcpstat.tcps_sc_dropped++; 3329 } 3330 return (1); 3331 } 3332 3333 int 3334 syn_cache_respond(sc, m) 3335 struct syn_cache *sc; 3336 struct mbuf *m; 3337 { 3338 struct route *ro; 3339 u_int8_t *optp; 3340 int optlen, error; 3341 u_int16_t tlen; 3342 struct ip *ip = NULL; 3343 #ifdef INET6 3344 struct ip6_hdr *ip6 = NULL; 3345 #endif 3346 struct tcphdr *th; 3347 u_int hlen; 3348 3349 switch (sc->sc_src.sa.sa_family) { 3350 case AF_INET: 3351 hlen = sizeof(struct ip); 3352 ro = &sc->sc_route4; 3353 break; 3354 #ifdef INET6 3355 case AF_INET6: 3356 hlen = sizeof(struct ip6_hdr); 3357 ro = (struct route *)&sc->sc_route6; 3358 break; 3359 #endif 3360 default: 3361 if (m) 3362 m_freem(m); 3363 return EAFNOSUPPORT; 3364 } 3365 3366 /* Compute the size of the TCP options. */ 3367 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3368 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3369 3370 tlen = hlen + sizeof(struct tcphdr) + optlen; 3371 3372 /* 3373 * Create the IP+TCP header from scratch. 3374 */ 3375 if (m) 3376 m_freem(m); 3377 #ifdef DIAGNOSTIC 3378 if (max_linkhdr + tlen > MCLBYTES) 3379 return (ENOBUFS); 3380 #endif 3381 MGETHDR(m, M_DONTWAIT, MT_DATA); 3382 if (m && tlen > MHLEN) { 3383 MCLGET(m, M_DONTWAIT); 3384 if ((m->m_flags & M_EXT) == 0) { 3385 m_freem(m); 3386 m = NULL; 3387 } 3388 } 3389 if (m == NULL) 3390 return (ENOBUFS); 3391 3392 /* Fixup the mbuf. */ 3393 m->m_data += max_linkhdr; 3394 m->m_len = m->m_pkthdr.len = tlen; 3395 #ifdef IPSEC 3396 if (sc->sc_tp) { 3397 struct tcpcb *tp; 3398 struct socket *so; 3399 3400 tp = sc->sc_tp; 3401 if (tp->t_inpcb) 3402 so = tp->t_inpcb->inp_socket; 3403 #ifdef INET6 3404 else if (tp->t_in6pcb) 3405 so = tp->t_in6pcb->in6p_socket; 3406 #endif 3407 else 3408 so = NULL; 3409 /* use IPsec policy on listening socket, on SYN ACK */ 3410 ipsec_setsocket(m, so); 3411 } 3412 #endif 3413 m->m_pkthdr.rcvif = NULL; 3414 memset(mtod(m, u_char *), 0, tlen); 3415 3416 switch (sc->sc_src.sa.sa_family) { 3417 case AF_INET: 3418 ip = mtod(m, struct ip *); 3419 ip->ip_dst = sc->sc_src.sin.sin_addr; 3420 ip->ip_src = sc->sc_dst.sin.sin_addr; 3421 ip->ip_p = IPPROTO_TCP; 3422 th = (struct tcphdr *)(ip + 1); 3423 th->th_dport = sc->sc_src.sin.sin_port; 3424 th->th_sport = sc->sc_dst.sin.sin_port; 3425 break; 3426 #ifdef INET6 3427 case AF_INET6: 3428 ip6 = mtod(m, struct ip6_hdr *); 3429 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3430 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3431 ip6->ip6_nxt = IPPROTO_TCP; 3432 /* ip6_plen will be updated in ip6_output() */ 3433 th = (struct tcphdr *)(ip6 + 1); 3434 th->th_dport = sc->sc_src.sin6.sin6_port; 3435 th->th_sport = sc->sc_dst.sin6.sin6_port; 3436 break; 3437 #endif 3438 default: 3439 th = NULL; 3440 } 3441 3442 th->th_seq = htonl(sc->sc_iss); 3443 th->th_ack = htonl(sc->sc_irs + 1); 3444 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3445 th->th_flags = TH_SYN|TH_ACK; 3446 th->th_win = htons(sc->sc_win); 3447 /* th_sum already 0 */ 3448 /* th_urp already 0 */ 3449 3450 /* Tack on the TCP options. */ 3451 optp = (u_int8_t *)(th + 1); 3452 *optp++ = TCPOPT_MAXSEG; 3453 *optp++ = 4; 3454 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3455 *optp++ = sc->sc_ourmaxseg & 0xff; 3456 3457 if (sc->sc_request_r_scale != 15) { 3458 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3459 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3460 sc->sc_request_r_scale); 3461 optp += 4; 3462 } 3463 3464 if (sc->sc_flags & SCF_TIMESTAMP) { 3465 u_int32_t *lp = (u_int32_t *)(optp); 3466 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3467 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3468 *lp++ = htonl(tcp_now); 3469 *lp = htonl(sc->sc_timestamp); 3470 optp += TCPOLEN_TSTAMP_APPA; 3471 } 3472 3473 /* Compute the packet's checksum. */ 3474 switch (sc->sc_src.sa.sa_family) { 3475 case AF_INET: 3476 ip->ip_len = htons(tlen - hlen); 3477 th->th_sum = 0; 3478 th->th_sum = in_cksum(m, tlen); 3479 break; 3480 #ifdef INET6 3481 case AF_INET6: 3482 ip6->ip6_plen = htons(tlen - hlen); 3483 th->th_sum = 0; 3484 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3485 break; 3486 #endif 3487 } 3488 3489 /* 3490 * Fill in some straggling IP bits. Note the stack expects 3491 * ip_len to be in host order, for convenience. 3492 */ 3493 switch (sc->sc_src.sa.sa_family) { 3494 case AF_INET: 3495 ip->ip_len = tlen; 3496 ip->ip_ttl = ip_defttl; 3497 /* XXX tos? */ 3498 break; 3499 #ifdef INET6 3500 case AF_INET6: 3501 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3502 ip6->ip6_vfc |= IPV6_VERSION; 3503 ip6->ip6_plen = htons(tlen - hlen); 3504 /* ip6_hlim will be initialized afterwards */ 3505 /* XXX flowlabel? */ 3506 break; 3507 #endif 3508 } 3509 3510 switch (sc->sc_src.sa.sa_family) { 3511 case AF_INET: 3512 error = ip_output(m, sc->sc_ipopts, ro, 3513 (ip_mtudisc ? IP_MTUDISC : 0), 3514 NULL); 3515 break; 3516 #ifdef INET6 3517 case AF_INET6: 3518 ip6->ip6_hlim = in6_selecthlim(NULL, 3519 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3520 3521 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3522 0, NULL, NULL); 3523 break; 3524 #endif 3525 default: 3526 error = EAFNOSUPPORT; 3527 break; 3528 } 3529 return (error); 3530 } 3531
Indexes created Mon Sep 22 13:09:51 GMT 2025