tcp_input.c revision 1.129
1 /* $NetBSD: tcp_input.c,v 1.129 2001/09/10 22:14:26 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, 2001 The NetBSD Foundation, Inc. 47 * All rights reserved. 48 * 49 * This code is derived from software contributed to The NetBSD Foundation 50 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 51 * Facility, NASA Ames Research Center. 52 * 53 * Redistribution and use in source and binary forms, with or without 54 * modification, are permitted provided that the following conditions 55 * are met: 56 * 1. Redistributions of source code must retain the above copyright 57 * notice, this list of conditions and the following disclaimer. 58 * 2. Redistributions in binary form must reproduce the above copyright 59 * notice, this list of conditions and the following disclaimer in the 60 * documentation and/or other materials provided with the distribution. 61 * 3. All advertising materials mentioning features or use of this software 62 * must display the following acknowledgement: 63 * This product includes software developed by the NetBSD 64 * Foundation, Inc. and its contributors. 65 * 4. Neither the name of The NetBSD Foundation nor the names of its 66 * contributors may be used to endorse or promote products derived 67 * from this software without specific prior written permission. 68 * 69 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 70 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 71 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 72 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 73 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 74 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 75 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 76 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 77 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 78 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 79 * POSSIBILITY OF SUCH DAMAGE. 80 */ 81 82 /* 83 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 84 * The Regents of the University of California. All rights reserved. 85 * 86 * Redistribution and use in source and binary forms, with or without 87 * modification, are permitted provided that the following conditions 88 * are met: 89 * 1. Redistributions of source code must retain the above copyright 90 * notice, this list of conditions and the following disclaimer. 91 * 2. Redistributions in binary form must reproduce the above copyright 92 * notice, this list of conditions and the following disclaimer in the 93 * documentation and/or other materials provided with the distribution. 94 * 3. All advertising materials mentioning features or use of this software 95 * must display the following acknowledgement: 96 * This product includes software developed by the University of 97 * California, Berkeley and its contributors. 98 * 4. Neither the name of the University nor the names of its contributors 99 * may be used to endorse or promote products derived from this software 100 * without specific prior written permission. 101 * 102 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 103 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 104 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 105 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 106 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 107 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 108 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 109 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 110 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 111 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 112 * SUCH DAMAGE. 113 * 114 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 115 */ 116 117 /* 118 * TODO list for SYN cache stuff: 119 * 120 * Find room for a "state" field, which is needed to keep a 121 * compressed state for TIME_WAIT TCBs. It's been noted already 122 * that this is fairly important for very high-volume web and 123 * mail servers, which use a large number of short-lived 124 * connections. 125 */ 126 127 #include "opt_inet.h" 128 #include "opt_ipsec.h" 129 #include "opt_inet_csum.h" 130 #include "opt_tcp_debug.h" 131 132 #include <sys/param.h> 133 #include <sys/systm.h> 134 #include <sys/malloc.h> 135 #include <sys/mbuf.h> 136 #include <sys/protosw.h> 137 #include <sys/socket.h> 138 #include <sys/socketvar.h> 139 #include <sys/errno.h> 140 #include <sys/syslog.h> 141 #include <sys/pool.h> 142 #include <sys/domain.h> 143 #include <sys/kernel.h> 144 145 #include <net/if.h> 146 #include <net/route.h> 147 #include <net/if_types.h> 148 149 #include <netinet/in.h> 150 #include <netinet/in_systm.h> 151 #include <netinet/ip.h> 152 #include <netinet/in_pcb.h> 153 #include <netinet/ip_var.h> 154 155 #ifdef INET6 156 #ifndef INET 157 #include <netinet/in.h> 158 #endif 159 #include <netinet/ip6.h> 160 #include <netinet6/ip6_var.h> 161 #include <netinet6/in6_pcb.h> 162 #include <netinet6/ip6_var.h> 163 #include <netinet6/in6_var.h> 164 #include <netinet/icmp6.h> 165 #include <netinet6/nd6.h> 166 #endif 167 168 #ifdef PULLDOWN_TEST 169 #ifndef INET6 170 /* always need ip6.h for IP6_EXTHDR_GET */ 171 #include <netinet/ip6.h> 172 #endif 173 #endif 174 175 #include <netinet/tcp.h> 176 #include <netinet/tcp_fsm.h> 177 #include <netinet/tcp_seq.h> 178 #include <netinet/tcp_timer.h> 179 #include <netinet/tcp_var.h> 180 #include <netinet/tcpip.h> 181 #include <netinet/tcp_debug.h> 182 183 #include <machine/stdarg.h> 184 185 #ifdef IPSEC 186 #include <netinet6/ipsec.h> 187 #include <netkey/key.h> 188 #endif /*IPSEC*/ 189 #ifdef INET6 190 #include "faith.h" 191 #if defined(NFAITH) && NFAITH > 0 192 #include <net/if_faith.h> 193 #endif 194 #endif 195 196 int tcprexmtthresh = 3; 197 int tcp_log_refused; 198 199 static int tcp_rst_ppslim_count = 0; 200 static struct timeval tcp_rst_ppslim_last; 201 202 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 203 204 /* for modulo comparisons of timestamps */ 205 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 206 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 207 208 /* 209 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 210 */ 211 #ifdef INET6 212 #define ND6_HINT(tp) \ 213 do { \ 214 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \ 215 && tp->t_in6pcb->in6p_route.ro_rt) { \ 216 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 217 } \ 218 } while (0) 219 #else 220 #define ND6_HINT(tp) 221 #endif 222 223 /* 224 * Macro to compute ACK transmission behavior. Delay the ACK unless 225 * we have already delayed an ACK (must send an ACK every two segments). 226 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 227 * option is enabled. 228 */ 229 #define TCP_SETUP_ACK(tp, th) \ 230 do { \ 231 if ((tp)->t_flags & TF_DELACK || \ 232 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 233 tp->t_flags |= TF_ACKNOW; \ 234 else \ 235 TCP_SET_DELACK(tp); \ 236 } while (0) 237 238 /* 239 * Convert TCP protocol fields to host order for easier processing. 240 */ 241 #define TCP_FIELDS_TO_HOST(th) \ 242 do { \ 243 NTOHL((th)->th_seq); \ 244 NTOHL((th)->th_ack); \ 245 NTOHS((th)->th_win); \ 246 NTOHS((th)->th_urp); \ 247 } while (0) 248 249 #ifdef TCP_CSUM_COUNTERS 250 #include <sys/device.h> 251 252 extern struct evcnt tcp_hwcsum_ok; 253 extern struct evcnt tcp_hwcsum_bad; 254 extern struct evcnt tcp_hwcsum_data; 255 extern struct evcnt tcp_swcsum; 256 257 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 258 259 #else 260 261 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 262 263 #endif /* TCP_CSUM_COUNTERS */ 264 265 int 266 tcp_reass(tp, th, m, tlen) 267 struct tcpcb *tp; 268 struct tcphdr *th; 269 struct mbuf *m; 270 int *tlen; 271 { 272 struct ipqent *p, *q, *nq, *tiqe = NULL; 273 struct socket *so = NULL; 274 int pkt_flags; 275 tcp_seq pkt_seq; 276 unsigned pkt_len; 277 u_long rcvpartdupbyte = 0; 278 u_long rcvoobyte; 279 280 if (tp->t_inpcb) 281 so = tp->t_inpcb->inp_socket; 282 #ifdef INET6 283 else if (tp->t_in6pcb) 284 so = tp->t_in6pcb->in6p_socket; 285 #endif 286 287 TCP_REASS_LOCK_CHECK(tp); 288 289 /* 290 * Call with th==0 after become established to 291 * force pre-ESTABLISHED data up to user socket. 292 */ 293 if (th == 0) 294 goto present; 295 296 rcvoobyte = *tlen; 297 /* 298 * Copy these to local variables because the tcpiphdr 299 * gets munged while we are collapsing mbufs. 300 */ 301 pkt_seq = th->th_seq; 302 pkt_len = *tlen; 303 pkt_flags = th->th_flags; 304 /* 305 * Find a segment which begins after this one does. 306 */ 307 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 308 nq = q->ipqe_q.le_next; 309 /* 310 * If the received segment is just right after this 311 * fragment, merge the two together and then check 312 * for further overlaps. 313 */ 314 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 315 #ifdef TCPREASS_DEBUG 316 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 317 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 318 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 319 #endif 320 pkt_len += q->ipqe_len; 321 pkt_flags |= q->ipqe_flags; 322 pkt_seq = q->ipqe_seq; 323 m_cat(q->ipqe_m, m); 324 m = q->ipqe_m; 325 goto free_ipqe; 326 } 327 /* 328 * If the received segment is completely past this 329 * fragment, we need to go the next fragment. 330 */ 331 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 332 p = q; 333 continue; 334 } 335 /* 336 * If the fragment is past the received segment, 337 * it (or any following) can't be concatenated. 338 */ 339 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 340 break; 341 /* 342 * We've received all the data in this segment before. 343 * mark it as a duplicate and return. 344 */ 345 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 346 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 347 tcpstat.tcps_rcvduppack++; 348 tcpstat.tcps_rcvdupbyte += pkt_len; 349 m_freem(m); 350 if (tiqe != NULL) 351 pool_put(&ipqent_pool, tiqe); 352 return (0); 353 } 354 /* 355 * Received segment completely overlaps this fragment 356 * so we drop the fragment (this keeps the temporal 357 * ordering of segments correct). 358 */ 359 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 360 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 361 rcvpartdupbyte += q->ipqe_len; 362 m_freem(q->ipqe_m); 363 goto free_ipqe; 364 } 365 /* 366 * RX'ed segment extends past the end of the 367 * fragment. Drop the overlapping bytes. Then 368 * merge the fragment and segment then treat as 369 * a longer received packet. 370 */ 371 if (SEQ_LT(q->ipqe_seq, pkt_seq) 372 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 373 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 374 #ifdef TCPREASS_DEBUG 375 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 376 tp, overlap, 377 pkt_seq, pkt_seq + pkt_len, pkt_len); 378 #endif 379 m_adj(m, overlap); 380 rcvpartdupbyte += overlap; 381 m_cat(q->ipqe_m, m); 382 m = q->ipqe_m; 383 pkt_seq = q->ipqe_seq; 384 pkt_len += q->ipqe_len - overlap; 385 rcvoobyte -= overlap; 386 goto free_ipqe; 387 } 388 /* 389 * RX'ed segment extends past the front of the 390 * fragment. Drop the overlapping bytes on the 391 * received packet. The packet will then be 392 * contatentated with this fragment a bit later. 393 */ 394 if (SEQ_GT(q->ipqe_seq, pkt_seq) 395 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 396 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 397 #ifdef TCPREASS_DEBUG 398 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 399 tp, overlap, 400 pkt_seq, pkt_seq + pkt_len, pkt_len); 401 #endif 402 m_adj(m, -overlap); 403 pkt_len -= overlap; 404 rcvpartdupbyte += overlap; 405 rcvoobyte -= overlap; 406 } 407 /* 408 * If the received segment immediates precedes this 409 * fragment then tack the fragment onto this segment 410 * and reinsert the data. 411 */ 412 if (q->ipqe_seq == pkt_seq + pkt_len) { 413 #ifdef TCPREASS_DEBUG 414 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 415 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 416 pkt_seq, pkt_seq + pkt_len, pkt_len); 417 #endif 418 pkt_len += q->ipqe_len; 419 pkt_flags |= q->ipqe_flags; 420 m_cat(m, q->ipqe_m); 421 LIST_REMOVE(q, ipqe_q); 422 LIST_REMOVE(q, ipqe_timeq); 423 if (tiqe == NULL) { 424 tiqe = q; 425 } else { 426 pool_put(&ipqent_pool, q); 427 } 428 break; 429 } 430 /* 431 * If the fragment is before the segment, remember it. 432 * When this loop is terminated, p will contain the 433 * pointer to fragment that is right before the received 434 * segment. 435 */ 436 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 437 p = q; 438 439 continue; 440 441 /* 442 * This is a common operation. It also will allow 443 * to save doing a malloc/free in most instances. 444 */ 445 free_ipqe: 446 LIST_REMOVE(q, ipqe_q); 447 LIST_REMOVE(q, ipqe_timeq); 448 if (tiqe == NULL) { 449 tiqe = q; 450 } else { 451 pool_put(&ipqent_pool, q); 452 } 453 } 454 455 /* 456 * Allocate a new queue entry since the received segment did not 457 * collapse onto any other out-of-order block; thus we are allocating 458 * a new block. If it had collapsed, tiqe would not be NULL and 459 * we would be reusing it. 460 * XXX If we can't, just drop the packet. XXX 461 */ 462 if (tiqe == NULL) { 463 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 464 if (tiqe == NULL) { 465 tcpstat.tcps_rcvmemdrop++; 466 m_freem(m); 467 return (0); 468 } 469 } 470 471 /* 472 * Update the counters. 473 */ 474 tcpstat.tcps_rcvoopack++; 475 tcpstat.tcps_rcvoobyte += rcvoobyte; 476 if (rcvpartdupbyte) { 477 tcpstat.tcps_rcvpartduppack++; 478 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 479 } 480 481 /* 482 * Insert the new fragment queue entry into both queues. 483 */ 484 tiqe->ipqe_m = m; 485 tiqe->ipqe_seq = pkt_seq; 486 tiqe->ipqe_len = pkt_len; 487 tiqe->ipqe_flags = pkt_flags; 488 if (p == NULL) { 489 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 490 #ifdef TCPREASS_DEBUG 491 if (tiqe->ipqe_seq != tp->rcv_nxt) 492 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 493 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 494 #endif 495 } else { 496 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 497 #ifdef TCPREASS_DEBUG 498 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 499 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 500 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 501 #endif 502 } 503 504 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 505 506 present: 507 /* 508 * Present data to user, advancing rcv_nxt through 509 * completed sequence space. 510 */ 511 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 512 return (0); 513 q = tp->segq.lh_first; 514 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 515 return (0); 516 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 517 return (0); 518 519 tp->rcv_nxt += q->ipqe_len; 520 pkt_flags = q->ipqe_flags & TH_FIN; 521 ND6_HINT(tp); 522 523 LIST_REMOVE(q, ipqe_q); 524 LIST_REMOVE(q, ipqe_timeq); 525 if (so->so_state & SS_CANTRCVMORE) 526 m_freem(q->ipqe_m); 527 else 528 sbappend(&so->so_rcv, q->ipqe_m); 529 pool_put(&ipqent_pool, q); 530 sorwakeup(so); 531 return (pkt_flags); 532 } 533 534 #ifdef INET6 535 int 536 tcp6_input(mp, offp, proto) 537 struct mbuf **mp; 538 int *offp, proto; 539 { 540 struct mbuf *m = *mp; 541 542 /* 543 * draft-itojun-ipv6-tcp-to-anycast 544 * better place to put this in? 545 */ 546 if (m->m_flags & M_ANYCAST6) { 547 struct ip6_hdr *ip6; 548 if (m->m_len < sizeof(struct ip6_hdr)) { 549 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 550 tcpstat.tcps_rcvshort++; 551 return IPPROTO_DONE; 552 } 553 } 554 ip6 = mtod(m, struct ip6_hdr *); 555 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 556 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 557 return IPPROTO_DONE; 558 } 559 560 tcp_input(m, *offp, proto); 561 return IPPROTO_DONE; 562 } 563 #endif 564 565 /* 566 * TCP input routine, follows pages 65-76 of the 567 * protocol specification dated September, 1981 very closely. 568 */ 569 void 570 #if __STDC__ 571 tcp_input(struct mbuf *m, ...) 572 #else 573 tcp_input(m, va_alist) 574 struct mbuf *m; 575 #endif 576 { 577 int proto; 578 struct tcphdr *th; 579 struct ip *ip; 580 struct inpcb *inp; 581 #ifdef INET6 582 struct ip6_hdr *ip6; 583 struct in6pcb *in6p; 584 #endif 585 caddr_t optp = NULL; 586 int optlen = 0; 587 int len, tlen, toff, hdroptlen = 0; 588 struct tcpcb *tp = 0; 589 int tiflags; 590 struct socket *so = NULL; 591 int todrop, acked, ourfinisacked, needoutput = 0; 592 short ostate = 0; 593 int iss = 0; 594 u_long tiwin; 595 struct tcp_opt_info opti; 596 int off, iphlen; 597 va_list ap; 598 int af; /* af on the wire */ 599 struct mbuf *tcp_saveti = NULL; 600 601 va_start(ap, m); 602 toff = va_arg(ap, int); 603 proto = va_arg(ap, int); 604 va_end(ap); 605 606 tcpstat.tcps_rcvtotal++; 607 608 bzero(&opti, sizeof(opti)); 609 opti.ts_present = 0; 610 opti.maxseg = 0; 611 612 /* 613 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 614 * 615 * TCP is, by definition, unicast, so we reject all 616 * multicast outright. 617 * 618 * Note, there are additional src/dst address checks in 619 * the AF-specific code below. 620 */ 621 if (m->m_flags & (M_BCAST|M_MCAST)) { 622 /* XXX stat */ 623 goto drop; 624 } 625 #ifdef INET6 626 if (m->m_flags & M_ANYCAST6) { 627 /* XXX stat */ 628 goto drop; 629 } 630 #endif 631 632 /* 633 * Get IP and TCP header together in first mbuf. 634 * Note: IP leaves IP header in first mbuf. 635 */ 636 ip = mtod(m, struct ip *); 637 #ifdef INET6 638 ip6 = NULL; 639 #endif 640 switch (ip->ip_v) { 641 #ifdef INET 642 case 4: 643 af = AF_INET; 644 iphlen = sizeof(struct ip); 645 #ifndef PULLDOWN_TEST 646 /* would like to get rid of this... */ 647 if (toff > sizeof (struct ip)) { 648 ip_stripoptions(m, (struct mbuf *)0); 649 toff = sizeof(struct ip); 650 } 651 if (m->m_len < toff + sizeof (struct tcphdr)) { 652 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 653 tcpstat.tcps_rcvshort++; 654 return; 655 } 656 } 657 ip = mtod(m, struct ip *); 658 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 659 #else 660 ip = mtod(m, struct ip *); 661 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 662 sizeof(struct tcphdr)); 663 if (th == NULL) { 664 tcpstat.tcps_rcvshort++; 665 return; 666 } 667 #endif 668 669 /* 670 * Make sure destination address is not multicast. 671 * Source address checked in ip_input(). 672 */ 673 if (IN_MULTICAST(ip->ip_dst.s_addr)) { 674 /* XXX stat */ 675 goto drop; 676 } 677 678 /* We do the checksum after PCB lookup... */ 679 len = ip->ip_len; 680 tlen = len - toff; 681 break; 682 #endif 683 #ifdef INET6 684 case 6: 685 ip = NULL; 686 iphlen = sizeof(struct ip6_hdr); 687 af = AF_INET6; 688 #ifndef PULLDOWN_TEST 689 if (m->m_len < toff + sizeof(struct tcphdr)) { 690 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 691 if (m == NULL) { 692 tcpstat.tcps_rcvshort++; 693 return; 694 } 695 } 696 ip6 = mtod(m, struct ip6_hdr *); 697 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 698 #else 699 ip6 = mtod(m, struct ip6_hdr *); 700 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 701 sizeof(struct tcphdr)); 702 if (th == NULL) { 703 tcpstat.tcps_rcvshort++; 704 return; 705 } 706 #endif 707 708 /* Be proactive about malicious use of IPv4 mapped address */ 709 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 710 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 711 /* XXX stat */ 712 goto drop; 713 } 714 715 /* 716 * Be proactive about unspecified IPv6 address in source. 717 * As we use all-zero to indicate unbounded/unconnected pcb, 718 * unspecified IPv6 address can be used to confuse us. 719 * 720 * Note that packets with unspecified IPv6 destination is 721 * already dropped in ip6_input. 722 */ 723 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 724 /* XXX stat */ 725 goto drop; 726 } 727 728 /* 729 * Make sure destination address is not multicast. 730 * Source address checked in ip6_input(). 731 */ 732 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 733 /* XXX stat */ 734 goto drop; 735 } 736 737 /* We do the checksum after PCB lookup... */ 738 len = m->m_pkthdr.len; 739 tlen = len - toff; 740 break; 741 #endif 742 default: 743 m_freem(m); 744 return; 745 } 746 747 /* 748 * Check that TCP offset makes sense, 749 * pull out TCP options and adjust length. XXX 750 */ 751 off = th->th_off << 2; 752 if (off < sizeof (struct tcphdr) || off > tlen) { 753 tcpstat.tcps_rcvbadoff++; 754 goto drop; 755 } 756 tlen -= off; 757 758 /* 759 * tcp_input() has been modified to use tlen to mean the TCP data 760 * length throughout the function. Other functions can use 761 * m->m_pkthdr.len as the basis for calculating the TCP data length. 762 * rja 763 */ 764 765 if (off > sizeof (struct tcphdr)) { 766 #ifndef PULLDOWN_TEST 767 if (m->m_len < toff + off) { 768 if ((m = m_pullup(m, toff + off)) == 0) { 769 tcpstat.tcps_rcvshort++; 770 return; 771 } 772 switch (af) { 773 #ifdef INET 774 case AF_INET: 775 ip = mtod(m, struct ip *); 776 break; 777 #endif 778 #ifdef INET6 779 case AF_INET6: 780 ip6 = mtod(m, struct ip6_hdr *); 781 break; 782 #endif 783 } 784 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 785 } 786 #else 787 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 788 if (th == NULL) { 789 tcpstat.tcps_rcvshort++; 790 return; 791 } 792 /* 793 * NOTE: ip/ip6 will not be affected by m_pulldown() 794 * (as they're before toff) and we don't need to update those. 795 */ 796 #endif 797 optlen = off - sizeof (struct tcphdr); 798 optp = ((caddr_t)th) + sizeof(struct tcphdr); 799 /* 800 * Do quick retrieval of timestamp options ("options 801 * prediction?"). If timestamp is the only option and it's 802 * formatted as recommended in RFC 1323 appendix A, we 803 * quickly get the values now and not bother calling 804 * tcp_dooptions(), etc. 805 */ 806 if ((optlen == TCPOLEN_TSTAMP_APPA || 807 (optlen > TCPOLEN_TSTAMP_APPA && 808 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 809 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 810 (th->th_flags & TH_SYN) == 0) { 811 opti.ts_present = 1; 812 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 813 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 814 optp = NULL; /* we've parsed the options */ 815 } 816 } 817 tiflags = th->th_flags; 818 819 /* 820 * Locate pcb for segment. 821 */ 822 findpcb: 823 inp = NULL; 824 #ifdef INET6 825 in6p = NULL; 826 #endif 827 switch (af) { 828 #ifdef INET 829 case AF_INET: 830 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 831 ip->ip_dst, th->th_dport); 832 if (inp == 0) { 833 ++tcpstat.tcps_pcbhashmiss; 834 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 835 } 836 #ifdef INET6 837 if (inp == 0) { 838 struct in6_addr s, d; 839 840 /* mapped addr case */ 841 bzero(&s, sizeof(s)); 842 s.s6_addr16[5] = htons(0xffff); 843 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 844 bzero(&d, sizeof(d)); 845 d.s6_addr16[5] = htons(0xffff); 846 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 847 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 848 &d, th->th_dport, 0); 849 if (in6p == 0) { 850 ++tcpstat.tcps_pcbhashmiss; 851 in6p = in6_pcblookup_bind(&tcb6, &d, 852 th->th_dport, 0); 853 } 854 } 855 #endif 856 #ifndef INET6 857 if (inp == 0) 858 #else 859 if (inp == 0 && in6p == 0) 860 #endif 861 { 862 ++tcpstat.tcps_noport; 863 if (tcp_log_refused && (tiflags & TH_SYN)) { 864 #ifndef INET6 865 char src[4*sizeof "123"]; 866 char dst[4*sizeof "123"]; 867 #else 868 char src[INET6_ADDRSTRLEN]; 869 char dst[INET6_ADDRSTRLEN]; 870 #endif 871 if (ip) { 872 strcpy(src, inet_ntoa(ip->ip_src)); 873 strcpy(dst, inet_ntoa(ip->ip_dst)); 874 } 875 #ifdef INET6 876 else if (ip6) { 877 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 878 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 879 } 880 #endif 881 else { 882 strcpy(src, "(unknown)"); 883 strcpy(dst, "(unknown)"); 884 } 885 log(LOG_INFO, 886 "Connection attempt to TCP %s:%d from %s:%d\n", 887 dst, ntohs(th->th_dport), 888 src, ntohs(th->th_sport)); 889 } 890 TCP_FIELDS_TO_HOST(th); 891 goto dropwithreset_ratelim; 892 } 893 #ifdef IPSEC 894 if (inp && ipsec4_in_reject(m, inp)) { 895 ipsecstat.in_polvio++; 896 goto drop; 897 } 898 #ifdef INET6 899 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 900 ipsecstat.in_polvio++; 901 goto drop; 902 } 903 #endif 904 #endif /*IPSEC*/ 905 break; 906 #endif /*INET*/ 907 #ifdef INET6 908 case AF_INET6: 909 { 910 int faith; 911 912 #if defined(NFAITH) && NFAITH > 0 913 faith = faithprefix(&ip6->ip6_dst); 914 #else 915 faith = 0; 916 #endif 917 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 918 &ip6->ip6_dst, th->th_dport, faith); 919 if (in6p == NULL) { 920 ++tcpstat.tcps_pcbhashmiss; 921 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 922 th->th_dport, faith); 923 } 924 if (in6p == NULL) { 925 ++tcpstat.tcps_noport; 926 TCP_FIELDS_TO_HOST(th); 927 goto dropwithreset_ratelim; 928 } 929 #ifdef IPSEC 930 if (ipsec6_in_reject(m, in6p)) { 931 ipsec6stat.in_polvio++; 932 goto drop; 933 } 934 #endif /*IPSEC*/ 935 break; 936 } 937 #endif 938 } 939 940 /* 941 * If the state is CLOSED (i.e., TCB does not exist) then 942 * all data in the incoming segment is discarded. 943 * If the TCB exists but is in CLOSED state, it is embryonic, 944 * but should either do a listen or a connect soon. 945 */ 946 tp = NULL; 947 so = NULL; 948 if (inp) { 949 tp = intotcpcb(inp); 950 so = inp->inp_socket; 951 } 952 #ifdef INET6 953 else if (in6p) { 954 tp = in6totcpcb(in6p); 955 so = in6p->in6p_socket; 956 } 957 #endif 958 if (tp == 0) { 959 TCP_FIELDS_TO_HOST(th); 960 goto dropwithreset_ratelim; 961 } 962 if (tp->t_state == TCPS_CLOSED) 963 goto drop; 964 965 /* 966 * Checksum extended TCP header and data. 967 */ 968 switch (af) { 969 #ifdef INET 970 case AF_INET: 971 switch (m->m_pkthdr.csum_flags & 972 ((m->m_pkthdr.rcvif->if_csum_flags & M_CSUM_TCPv4) | 973 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 974 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 975 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 976 goto badcsum; 977 978 case M_CSUM_TCPv4|M_CSUM_DATA: 979 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 980 if ((m->m_pkthdr.csum_data ^ 0xffff) != 0) 981 goto badcsum; 982 break; 983 984 case M_CSUM_TCPv4: 985 /* Checksum was okay. */ 986 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 987 break; 988 989 default: 990 /* Must compute it ourselves. */ 991 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 992 #ifndef PULLDOWN_TEST 993 { 994 struct ipovly *ipov; 995 ipov = (struct ipovly *)ip; 996 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 997 ipov->ih_len = htons(tlen + off); 998 999 if (in_cksum(m, len) != 0) 1000 goto badcsum; 1001 } 1002 #else 1003 if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1004 goto badcsum; 1005 #endif /* ! PULLDOWN_TEST */ 1006 break; 1007 } 1008 break; 1009 #endif /* INET4 */ 1010 1011 #ifdef INET6 1012 case AF_INET6: 1013 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 1014 goto badcsum; 1015 break; 1016 #endif /* INET6 */ 1017 } 1018 1019 TCP_FIELDS_TO_HOST(th); 1020 1021 /* Unscale the window into a 32-bit value. */ 1022 if ((tiflags & TH_SYN) == 0) 1023 tiwin = th->th_win << tp->snd_scale; 1024 else 1025 tiwin = th->th_win; 1026 1027 #ifdef INET6 1028 /* save packet options if user wanted */ 1029 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1030 if (in6p->in6p_options) { 1031 m_freem(in6p->in6p_options); 1032 in6p->in6p_options = 0; 1033 } 1034 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1035 } 1036 #endif 1037 1038 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1039 union syn_cache_sa src; 1040 union syn_cache_sa dst; 1041 1042 bzero(&src, sizeof(src)); 1043 bzero(&dst, sizeof(dst)); 1044 switch (af) { 1045 #ifdef INET 1046 case AF_INET: 1047 src.sin.sin_len = sizeof(struct sockaddr_in); 1048 src.sin.sin_family = AF_INET; 1049 src.sin.sin_addr = ip->ip_src; 1050 src.sin.sin_port = th->th_sport; 1051 1052 dst.sin.sin_len = sizeof(struct sockaddr_in); 1053 dst.sin.sin_family = AF_INET; 1054 dst.sin.sin_addr = ip->ip_dst; 1055 dst.sin.sin_port = th->th_dport; 1056 break; 1057 #endif 1058 #ifdef INET6 1059 case AF_INET6: 1060 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1061 src.sin6.sin6_family = AF_INET6; 1062 src.sin6.sin6_addr = ip6->ip6_src; 1063 src.sin6.sin6_port = th->th_sport; 1064 1065 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1066 dst.sin6.sin6_family = AF_INET6; 1067 dst.sin6.sin6_addr = ip6->ip6_dst; 1068 dst.sin6.sin6_port = th->th_dport; 1069 break; 1070 #endif /* INET6 */ 1071 default: 1072 goto badsyn; /*sanity*/ 1073 } 1074 1075 if (so->so_options & SO_DEBUG) { 1076 ostate = tp->t_state; 1077 1078 tcp_saveti = NULL; 1079 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1080 goto nosave; 1081 1082 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1083 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1084 if (!tcp_saveti) 1085 goto nosave; 1086 } else { 1087 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1088 if (!tcp_saveti) 1089 goto nosave; 1090 tcp_saveti->m_len = iphlen; 1091 m_copydata(m, 0, iphlen, 1092 mtod(tcp_saveti, caddr_t)); 1093 } 1094 1095 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1096 m_freem(tcp_saveti); 1097 tcp_saveti = NULL; 1098 } else { 1099 tcp_saveti->m_len += sizeof(struct tcphdr); 1100 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1101 sizeof(struct tcphdr)); 1102 } 1103 if (tcp_saveti) { 1104 /* 1105 * need to recover version # field, which was 1106 * overwritten on ip_cksum computation. 1107 */ 1108 struct ip *sip; 1109 sip = mtod(tcp_saveti, struct ip *); 1110 switch (af) { 1111 #ifdef INET 1112 case AF_INET: 1113 sip->ip_v = 4; 1114 break; 1115 #endif 1116 #ifdef INET6 1117 case AF_INET6: 1118 sip->ip_v = 6; 1119 break; 1120 #endif 1121 } 1122 } 1123 nosave:; 1124 } 1125 if (so->so_options & SO_ACCEPTCONN) { 1126 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1127 if (tiflags & TH_RST) { 1128 syn_cache_reset(&src.sa, &dst.sa, th); 1129 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1130 (TH_ACK|TH_SYN)) { 1131 /* 1132 * Received a SYN,ACK. This should 1133 * never happen while we are in 1134 * LISTEN. Send an RST. 1135 */ 1136 goto badsyn; 1137 } else if (tiflags & TH_ACK) { 1138 so = syn_cache_get(&src.sa, &dst.sa, 1139 th, toff, tlen, so, m); 1140 if (so == NULL) { 1141 /* 1142 * We don't have a SYN for 1143 * this ACK; send an RST. 1144 */ 1145 goto badsyn; 1146 } else if (so == 1147 (struct socket *)(-1)) { 1148 /* 1149 * We were unable to create 1150 * the connection. If the 1151 * 3-way handshake was 1152 * completed, and RST has 1153 * been sent to the peer. 1154 * Since the mbuf might be 1155 * in use for the reply, 1156 * do not free it. 1157 */ 1158 m = NULL; 1159 } else { 1160 /* 1161 * We have created a 1162 * full-blown connection. 1163 */ 1164 tp = NULL; 1165 inp = NULL; 1166 #ifdef INET6 1167 in6p = NULL; 1168 #endif 1169 switch (so->so_proto->pr_domain->dom_family) { 1170 #ifdef INET 1171 case AF_INET: 1172 inp = sotoinpcb(so); 1173 tp = intotcpcb(inp); 1174 break; 1175 #endif 1176 #ifdef INET6 1177 case AF_INET6: 1178 in6p = sotoin6pcb(so); 1179 tp = in6totcpcb(in6p); 1180 break; 1181 #endif 1182 } 1183 if (tp == NULL) 1184 goto badsyn; /*XXX*/ 1185 tiwin <<= tp->snd_scale; 1186 goto after_listen; 1187 } 1188 } else { 1189 /* 1190 * None of RST, SYN or ACK was set. 1191 * This is an invalid packet for a 1192 * TCB in LISTEN state. Send a RST. 1193 */ 1194 goto badsyn; 1195 } 1196 } else { 1197 /* 1198 * Received a SYN. 1199 */ 1200 1201 /* 1202 * LISTEN socket received a SYN 1203 * from itself? This can't possibly 1204 * be valid; drop the packet. 1205 */ 1206 if (th->th_sport == th->th_dport) { 1207 int i; 1208 1209 switch (af) { 1210 #ifdef INET 1211 case AF_INET: 1212 i = in_hosteq(ip->ip_src, ip->ip_dst); 1213 break; 1214 #endif 1215 #ifdef INET6 1216 case AF_INET6: 1217 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1218 break; 1219 #endif 1220 default: 1221 i = 1; 1222 } 1223 if (i) { 1224 tcpstat.tcps_badsyn++; 1225 goto drop; 1226 } 1227 } 1228 1229 /* 1230 * SYN looks ok; create compressed TCP 1231 * state for it. 1232 */ 1233 if (so->so_qlen <= so->so_qlimit && 1234 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1235 so, m, optp, optlen, &opti)) 1236 m = NULL; 1237 } 1238 goto drop; 1239 } 1240 } 1241 1242 after_listen: 1243 #ifdef DIAGNOSTIC 1244 /* 1245 * Should not happen now that all embryonic connections 1246 * are handled with compressed state. 1247 */ 1248 if (tp->t_state == TCPS_LISTEN) 1249 panic("tcp_input: TCPS_LISTEN"); 1250 #endif 1251 1252 /* 1253 * Segment received on connection. 1254 * Reset idle time and keep-alive timer. 1255 */ 1256 tp->t_rcvtime = tcp_now; 1257 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1258 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1259 1260 /* 1261 * Process options. 1262 */ 1263 if (optp) 1264 tcp_dooptions(tp, optp, optlen, th, &opti); 1265 1266 /* 1267 * Header prediction: check for the two common cases 1268 * of a uni-directional data xfer. If the packet has 1269 * no control flags, is in-sequence, the window didn't 1270 * change and we're not retransmitting, it's a 1271 * candidate. If the length is zero and the ack moved 1272 * forward, we're the sender side of the xfer. Just 1273 * free the data acked & wake any higher level process 1274 * that was blocked waiting for space. If the length 1275 * is non-zero and the ack didn't move, we're the 1276 * receiver side. If we're getting packets in-order 1277 * (the reassembly queue is empty), add the data to 1278 * the socket buffer and note that we need a delayed ack. 1279 */ 1280 if (tp->t_state == TCPS_ESTABLISHED && 1281 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1282 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1283 th->th_seq == tp->rcv_nxt && 1284 tiwin && tiwin == tp->snd_wnd && 1285 tp->snd_nxt == tp->snd_max) { 1286 1287 /* 1288 * If last ACK falls within this segment's sequence numbers, 1289 * record the timestamp. 1290 */ 1291 if (opti.ts_present && 1292 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1293 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1294 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1295 tp->ts_recent = opti.ts_val; 1296 } 1297 1298 if (tlen == 0) { 1299 if (SEQ_GT(th->th_ack, tp->snd_una) && 1300 SEQ_LEQ(th->th_ack, tp->snd_max) && 1301 tp->snd_cwnd >= tp->snd_wnd && 1302 tp->t_dupacks < tcprexmtthresh) { 1303 /* 1304 * this is a pure ack for outstanding data. 1305 */ 1306 ++tcpstat.tcps_predack; 1307 if (opti.ts_present && opti.ts_ecr) 1308 tcp_xmit_timer(tp, 1309 TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1310 else if (tp->t_rtttime && 1311 SEQ_GT(th->th_ack, tp->t_rtseq)) 1312 tcp_xmit_timer(tp, 1313 tcp_now - tp->t_rtttime); 1314 acked = th->th_ack - tp->snd_una; 1315 tcpstat.tcps_rcvackpack++; 1316 tcpstat.tcps_rcvackbyte += acked; 1317 ND6_HINT(tp); 1318 sbdrop(&so->so_snd, acked); 1319 /* 1320 * We want snd_recover to track snd_una to 1321 * avoid sequence wraparound problems for 1322 * very large transfers. 1323 */ 1324 tp->snd_una = tp->snd_recover = th->th_ack; 1325 m_freem(m); 1326 1327 /* 1328 * If all outstanding data are acked, stop 1329 * retransmit timer, otherwise restart timer 1330 * using current (possibly backed-off) value. 1331 * If process is waiting for space, 1332 * wakeup/selwakeup/signal. If data 1333 * are ready to send, let tcp_output 1334 * decide between more output or persist. 1335 */ 1336 if (tp->snd_una == tp->snd_max) 1337 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1338 else if (TCP_TIMER_ISARMED(tp, 1339 TCPT_PERSIST) == 0) 1340 TCP_TIMER_ARM(tp, TCPT_REXMT, 1341 tp->t_rxtcur); 1342 1343 sowwakeup(so); 1344 if (so->so_snd.sb_cc) 1345 (void) tcp_output(tp); 1346 if (tcp_saveti) 1347 m_freem(tcp_saveti); 1348 return; 1349 } 1350 } else if (th->th_ack == tp->snd_una && 1351 tp->segq.lh_first == NULL && 1352 tlen <= sbspace(&so->so_rcv)) { 1353 /* 1354 * this is a pure, in-sequence data packet 1355 * with nothing on the reassembly queue and 1356 * we have enough buffer space to take it. 1357 */ 1358 ++tcpstat.tcps_preddat; 1359 tp->rcv_nxt += tlen; 1360 tcpstat.tcps_rcvpack++; 1361 tcpstat.tcps_rcvbyte += tlen; 1362 ND6_HINT(tp); 1363 /* 1364 * Drop TCP, IP headers and TCP options then add data 1365 * to socket buffer. 1366 */ 1367 m_adj(m, toff + off); 1368 sbappend(&so->so_rcv, m); 1369 sorwakeup(so); 1370 TCP_SETUP_ACK(tp, th); 1371 if (tp->t_flags & TF_ACKNOW) 1372 (void) tcp_output(tp); 1373 if (tcp_saveti) 1374 m_freem(tcp_saveti); 1375 return; 1376 } 1377 } 1378 1379 /* 1380 * Compute mbuf offset to TCP data segment. 1381 */ 1382 hdroptlen = toff + off; 1383 1384 /* 1385 * Calculate amount of space in receive window, 1386 * and then do TCP input processing. 1387 * Receive window is amount of space in rcv queue, 1388 * but not less than advertised window. 1389 */ 1390 { int win; 1391 1392 win = sbspace(&so->so_rcv); 1393 if (win < 0) 1394 win = 0; 1395 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1396 } 1397 1398 switch (tp->t_state) { 1399 1400 /* 1401 * If the state is SYN_SENT: 1402 * if seg contains an ACK, but not for our SYN, drop the input. 1403 * if seg contains a RST, then drop the connection. 1404 * if seg does not contain SYN, then drop it. 1405 * Otherwise this is an acceptable SYN segment 1406 * initialize tp->rcv_nxt and tp->irs 1407 * if seg contains ack then advance tp->snd_una 1408 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1409 * arrange for segment to be acked (eventually) 1410 * continue processing rest of data/controls, beginning with URG 1411 */ 1412 case TCPS_SYN_SENT: 1413 if ((tiflags & TH_ACK) && 1414 (SEQ_LEQ(th->th_ack, tp->iss) || 1415 SEQ_GT(th->th_ack, tp->snd_max))) 1416 goto dropwithreset; 1417 if (tiflags & TH_RST) { 1418 if (tiflags & TH_ACK) 1419 tp = tcp_drop(tp, ECONNREFUSED); 1420 goto drop; 1421 } 1422 if ((tiflags & TH_SYN) == 0) 1423 goto drop; 1424 if (tiflags & TH_ACK) { 1425 tp->snd_una = tp->snd_recover = th->th_ack; 1426 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1427 tp->snd_nxt = tp->snd_una; 1428 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1429 } 1430 tp->irs = th->th_seq; 1431 tcp_rcvseqinit(tp); 1432 tp->t_flags |= TF_ACKNOW; 1433 tcp_mss_from_peer(tp, opti.maxseg); 1434 1435 /* 1436 * Initialize the initial congestion window. If we 1437 * had to retransmit the SYN, we must initialize cwnd 1438 * to 1 segment (i.e. the Loss Window). 1439 */ 1440 if (tp->t_flags & TF_SYN_REXMT) 1441 tp->snd_cwnd = tp->t_peermss; 1442 else 1443 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1444 tp->t_peermss); 1445 1446 tcp_rmx_rtt(tp); 1447 if (tiflags & TH_ACK) { 1448 tcpstat.tcps_connects++; 1449 soisconnected(so); 1450 tcp_established(tp); 1451 /* Do window scaling on this connection? */ 1452 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1453 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1454 tp->snd_scale = tp->requested_s_scale; 1455 tp->rcv_scale = tp->request_r_scale; 1456 } 1457 TCP_REASS_LOCK(tp); 1458 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1459 TCP_REASS_UNLOCK(tp); 1460 /* 1461 * if we didn't have to retransmit the SYN, 1462 * use its rtt as our initial srtt & rtt var. 1463 */ 1464 if (tp->t_rtttime) 1465 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1466 } else 1467 tp->t_state = TCPS_SYN_RECEIVED; 1468 1469 /* 1470 * Advance th->th_seq to correspond to first data byte. 1471 * If data, trim to stay within window, 1472 * dropping FIN if necessary. 1473 */ 1474 th->th_seq++; 1475 if (tlen > tp->rcv_wnd) { 1476 todrop = tlen - tp->rcv_wnd; 1477 m_adj(m, -todrop); 1478 tlen = tp->rcv_wnd; 1479 tiflags &= ~TH_FIN; 1480 tcpstat.tcps_rcvpackafterwin++; 1481 tcpstat.tcps_rcvbyteafterwin += todrop; 1482 } 1483 tp->snd_wl1 = th->th_seq - 1; 1484 tp->rcv_up = th->th_seq; 1485 goto step6; 1486 1487 /* 1488 * If the state is SYN_RECEIVED: 1489 * If seg contains an ACK, but not for our SYN, drop the input 1490 * and generate an RST. See page 36, rfc793 1491 */ 1492 case TCPS_SYN_RECEIVED: 1493 if ((tiflags & TH_ACK) && 1494 (SEQ_LEQ(th->th_ack, tp->iss) || 1495 SEQ_GT(th->th_ack, tp->snd_max))) 1496 goto dropwithreset; 1497 break; 1498 } 1499 1500 /* 1501 * States other than LISTEN or SYN_SENT. 1502 * First check timestamp, if present. 1503 * Then check that at least some bytes of segment are within 1504 * receive window. If segment begins before rcv_nxt, 1505 * drop leading data (and SYN); if nothing left, just ack. 1506 * 1507 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1508 * and it's less than ts_recent, drop it. 1509 */ 1510 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1511 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1512 1513 /* Check to see if ts_recent is over 24 days old. */ 1514 if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) > 1515 TCP_PAWS_IDLE) { 1516 /* 1517 * Invalidate ts_recent. If this segment updates 1518 * ts_recent, the age will be reset later and ts_recent 1519 * will get a valid value. If it does not, setting 1520 * ts_recent to zero will at least satisfy the 1521 * requirement that zero be placed in the timestamp 1522 * echo reply when ts_recent isn't valid. The 1523 * age isn't reset until we get a valid ts_recent 1524 * because we don't want out-of-order segments to be 1525 * dropped when ts_recent is old. 1526 */ 1527 tp->ts_recent = 0; 1528 } else { 1529 tcpstat.tcps_rcvduppack++; 1530 tcpstat.tcps_rcvdupbyte += tlen; 1531 tcpstat.tcps_pawsdrop++; 1532 goto dropafterack; 1533 } 1534 } 1535 1536 todrop = tp->rcv_nxt - th->th_seq; 1537 if (todrop > 0) { 1538 if (tiflags & TH_SYN) { 1539 tiflags &= ~TH_SYN; 1540 th->th_seq++; 1541 if (th->th_urp > 1) 1542 th->th_urp--; 1543 else { 1544 tiflags &= ~TH_URG; 1545 th->th_urp = 0; 1546 } 1547 todrop--; 1548 } 1549 if (todrop > tlen || 1550 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1551 /* 1552 * Any valid FIN must be to the left of the window. 1553 * At this point the FIN must be a duplicate or 1554 * out of sequence; drop it. 1555 */ 1556 tiflags &= ~TH_FIN; 1557 /* 1558 * Send an ACK to resynchronize and drop any data. 1559 * But keep on processing for RST or ACK. 1560 */ 1561 tp->t_flags |= TF_ACKNOW; 1562 todrop = tlen; 1563 tcpstat.tcps_rcvdupbyte += todrop; 1564 tcpstat.tcps_rcvduppack++; 1565 } else { 1566 tcpstat.tcps_rcvpartduppack++; 1567 tcpstat.tcps_rcvpartdupbyte += todrop; 1568 } 1569 hdroptlen += todrop; /*drop from head afterwards*/ 1570 th->th_seq += todrop; 1571 tlen -= todrop; 1572 if (th->th_urp > todrop) 1573 th->th_urp -= todrop; 1574 else { 1575 tiflags &= ~TH_URG; 1576 th->th_urp = 0; 1577 } 1578 } 1579 1580 /* 1581 * If new data are received on a connection after the 1582 * user processes are gone, then RST the other end. 1583 */ 1584 if ((so->so_state & SS_NOFDREF) && 1585 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1586 tp = tcp_close(tp); 1587 tcpstat.tcps_rcvafterclose++; 1588 goto dropwithreset; 1589 } 1590 1591 /* 1592 * If segment ends after window, drop trailing data 1593 * (and PUSH and FIN); if nothing left, just ACK. 1594 */ 1595 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1596 if (todrop > 0) { 1597 tcpstat.tcps_rcvpackafterwin++; 1598 if (todrop >= tlen) { 1599 tcpstat.tcps_rcvbyteafterwin += tlen; 1600 /* 1601 * If a new connection request is received 1602 * while in TIME_WAIT, drop the old connection 1603 * and start over if the sequence numbers 1604 * are above the previous ones. 1605 */ 1606 if (tiflags & TH_SYN && 1607 tp->t_state == TCPS_TIME_WAIT && 1608 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1609 iss = tcp_new_iss(tp, tp->snd_nxt); 1610 tp = tcp_close(tp); 1611 goto findpcb; 1612 } 1613 /* 1614 * If window is closed can only take segments at 1615 * window edge, and have to drop data and PUSH from 1616 * incoming segments. Continue processing, but 1617 * remember to ack. Otherwise, drop segment 1618 * and ack. 1619 */ 1620 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1621 tp->t_flags |= TF_ACKNOW; 1622 tcpstat.tcps_rcvwinprobe++; 1623 } else 1624 goto dropafterack; 1625 } else 1626 tcpstat.tcps_rcvbyteafterwin += todrop; 1627 m_adj(m, -todrop); 1628 tlen -= todrop; 1629 tiflags &= ~(TH_PUSH|TH_FIN); 1630 } 1631 1632 /* 1633 * If last ACK falls within this segment's sequence numbers, 1634 * and the timestamp is newer, record it. 1635 */ 1636 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1637 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1638 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1639 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1640 tp->ts_recent_age = TCP_TIMESTAMP(tp); 1641 tp->ts_recent = opti.ts_val; 1642 } 1643 1644 /* 1645 * If the RST bit is set examine the state: 1646 * SYN_RECEIVED STATE: 1647 * If passive open, return to LISTEN state. 1648 * If active open, inform user that connection was refused. 1649 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1650 * Inform user that connection was reset, and close tcb. 1651 * CLOSING, LAST_ACK, TIME_WAIT STATES 1652 * Close the tcb. 1653 */ 1654 if (tiflags&TH_RST) switch (tp->t_state) { 1655 1656 case TCPS_SYN_RECEIVED: 1657 so->so_error = ECONNREFUSED; 1658 goto close; 1659 1660 case TCPS_ESTABLISHED: 1661 case TCPS_FIN_WAIT_1: 1662 case TCPS_FIN_WAIT_2: 1663 case TCPS_CLOSE_WAIT: 1664 so->so_error = ECONNRESET; 1665 close: 1666 tp->t_state = TCPS_CLOSED; 1667 tcpstat.tcps_drops++; 1668 tp = tcp_close(tp); 1669 goto drop; 1670 1671 case TCPS_CLOSING: 1672 case TCPS_LAST_ACK: 1673 case TCPS_TIME_WAIT: 1674 tp = tcp_close(tp); 1675 goto drop; 1676 } 1677 1678 /* 1679 * If a SYN is in the window, then this is an 1680 * error and we send an RST and drop the connection. 1681 */ 1682 if (tiflags & TH_SYN) { 1683 tp = tcp_drop(tp, ECONNRESET); 1684 goto dropwithreset; 1685 } 1686 1687 /* 1688 * If the ACK bit is off we drop the segment and return. 1689 */ 1690 if ((tiflags & TH_ACK) == 0) { 1691 if (tp->t_flags & TF_ACKNOW) 1692 goto dropafterack; 1693 else 1694 goto drop; 1695 } 1696 1697 /* 1698 * Ack processing. 1699 */ 1700 switch (tp->t_state) { 1701 1702 /* 1703 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1704 * ESTABLISHED state and continue processing, otherwise 1705 * send an RST. 1706 */ 1707 case TCPS_SYN_RECEIVED: 1708 if (SEQ_GT(tp->snd_una, th->th_ack) || 1709 SEQ_GT(th->th_ack, tp->snd_max)) 1710 goto dropwithreset; 1711 tcpstat.tcps_connects++; 1712 soisconnected(so); 1713 tcp_established(tp); 1714 /* Do window scaling? */ 1715 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1716 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1717 tp->snd_scale = tp->requested_s_scale; 1718 tp->rcv_scale = tp->request_r_scale; 1719 } 1720 TCP_REASS_LOCK(tp); 1721 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1722 TCP_REASS_UNLOCK(tp); 1723 tp->snd_wl1 = th->th_seq - 1; 1724 /* fall into ... */ 1725 1726 /* 1727 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1728 * ACKs. If the ack is in the range 1729 * tp->snd_una < th->th_ack <= tp->snd_max 1730 * then advance tp->snd_una to th->th_ack and drop 1731 * data from the retransmission queue. If this ACK reflects 1732 * more up to date window information we update our window information. 1733 */ 1734 case TCPS_ESTABLISHED: 1735 case TCPS_FIN_WAIT_1: 1736 case TCPS_FIN_WAIT_2: 1737 case TCPS_CLOSE_WAIT: 1738 case TCPS_CLOSING: 1739 case TCPS_LAST_ACK: 1740 case TCPS_TIME_WAIT: 1741 1742 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1743 if (tlen == 0 && tiwin == tp->snd_wnd) { 1744 tcpstat.tcps_rcvdupack++; 1745 /* 1746 * If we have outstanding data (other than 1747 * a window probe), this is a completely 1748 * duplicate ack (ie, window info didn't 1749 * change), the ack is the biggest we've 1750 * seen and we've seen exactly our rexmt 1751 * threshhold of them, assume a packet 1752 * has been dropped and retransmit it. 1753 * Kludge snd_nxt & the congestion 1754 * window so we send only this one 1755 * packet. 1756 * 1757 * We know we're losing at the current 1758 * window size so do congestion avoidance 1759 * (set ssthresh to half the current window 1760 * and pull our congestion window back to 1761 * the new ssthresh). 1762 * 1763 * Dup acks mean that packets have left the 1764 * network (they're now cached at the receiver) 1765 * so bump cwnd by the amount in the receiver 1766 * to keep a constant cwnd packets in the 1767 * network. 1768 */ 1769 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1770 th->th_ack != tp->snd_una) 1771 tp->t_dupacks = 0; 1772 else if (++tp->t_dupacks == tcprexmtthresh) { 1773 tcp_seq onxt = tp->snd_nxt; 1774 u_int win = 1775 min(tp->snd_wnd, tp->snd_cwnd) / 1776 2 / tp->t_segsz; 1777 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1778 tp->snd_recover)) { 1779 /* 1780 * False fast retransmit after 1781 * timeout. Do not cut window. 1782 */ 1783 tp->snd_cwnd += tp->t_segsz; 1784 tp->t_dupacks = 0; 1785 (void) tcp_output(tp); 1786 goto drop; 1787 } 1788 1789 if (win < 2) 1790 win = 2; 1791 tp->snd_ssthresh = win * tp->t_segsz; 1792 tp->snd_recover = tp->snd_max; 1793 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1794 tp->t_rtttime = 0; 1795 tp->snd_nxt = th->th_ack; 1796 tp->snd_cwnd = tp->t_segsz; 1797 (void) tcp_output(tp); 1798 tp->snd_cwnd = tp->snd_ssthresh + 1799 tp->t_segsz * tp->t_dupacks; 1800 if (SEQ_GT(onxt, tp->snd_nxt)) 1801 tp->snd_nxt = onxt; 1802 goto drop; 1803 } else if (tp->t_dupacks > tcprexmtthresh) { 1804 tp->snd_cwnd += tp->t_segsz; 1805 (void) tcp_output(tp); 1806 goto drop; 1807 } 1808 } else 1809 tp->t_dupacks = 0; 1810 break; 1811 } 1812 /* 1813 * If the congestion window was inflated to account 1814 * for the other side's cached packets, retract it. 1815 */ 1816 if (tcp_do_newreno == 0) { 1817 if (tp->t_dupacks >= tcprexmtthresh && 1818 tp->snd_cwnd > tp->snd_ssthresh) 1819 tp->snd_cwnd = tp->snd_ssthresh; 1820 tp->t_dupacks = 0; 1821 } else if (tp->t_dupacks >= tcprexmtthresh && 1822 tcp_newreno(tp, th) == 0) { 1823 tp->snd_cwnd = tp->snd_ssthresh; 1824 /* 1825 * Window inflation should have left us with approx. 1826 * snd_ssthresh outstanding data. But in case we 1827 * would be inclined to send a burst, better to do 1828 * it via the slow start mechanism. 1829 */ 1830 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1831 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1832 + tp->t_segsz; 1833 tp->t_dupacks = 0; 1834 } 1835 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1836 tcpstat.tcps_rcvacktoomuch++; 1837 goto dropafterack; 1838 } 1839 acked = th->th_ack - tp->snd_una; 1840 tcpstat.tcps_rcvackpack++; 1841 tcpstat.tcps_rcvackbyte += acked; 1842 1843 /* 1844 * If we have a timestamp reply, update smoothed 1845 * round trip time. If no timestamp is present but 1846 * transmit timer is running and timed sequence 1847 * number was acked, update smoothed round trip time. 1848 * Since we now have an rtt measurement, cancel the 1849 * timer backoff (cf., Phil Karn's retransmit alg.). 1850 * Recompute the initial retransmit timer. 1851 */ 1852 if (opti.ts_present && opti.ts_ecr) 1853 tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1); 1854 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1855 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1856 1857 /* 1858 * If all outstanding data is acked, stop retransmit 1859 * timer and remember to restart (more output or persist). 1860 * If there is more data to be acked, restart retransmit 1861 * timer, using current (possibly backed-off) value. 1862 */ 1863 if (th->th_ack == tp->snd_max) { 1864 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1865 needoutput = 1; 1866 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1867 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1868 /* 1869 * When new data is acked, open the congestion window. 1870 * If the window gives us less than ssthresh packets 1871 * in flight, open exponentially (segsz per packet). 1872 * Otherwise open linearly: segsz per window 1873 * (segsz^2 / cwnd per packet), plus a constant 1874 * fraction of a packet (segsz/8) to help larger windows 1875 * open quickly enough. 1876 */ 1877 { 1878 u_int cw = tp->snd_cwnd; 1879 u_int incr = tp->t_segsz; 1880 1881 if (cw > tp->snd_ssthresh) 1882 incr = incr * incr / cw; 1883 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1884 tp->snd_cwnd = min(cw + incr, 1885 TCP_MAXWIN << tp->snd_scale); 1886 } 1887 ND6_HINT(tp); 1888 if (acked > so->so_snd.sb_cc) { 1889 tp->snd_wnd -= so->so_snd.sb_cc; 1890 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1891 ourfinisacked = 1; 1892 } else { 1893 sbdrop(&so->so_snd, acked); 1894 tp->snd_wnd -= acked; 1895 ourfinisacked = 0; 1896 } 1897 sowwakeup(so); 1898 /* 1899 * We want snd_recover to track snd_una to 1900 * avoid sequence wraparound problems for 1901 * very large transfers. 1902 */ 1903 tp->snd_una = tp->snd_recover = th->th_ack; 1904 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1905 tp->snd_nxt = tp->snd_una; 1906 1907 switch (tp->t_state) { 1908 1909 /* 1910 * In FIN_WAIT_1 STATE in addition to the processing 1911 * for the ESTABLISHED state if our FIN is now acknowledged 1912 * then enter FIN_WAIT_2. 1913 */ 1914 case TCPS_FIN_WAIT_1: 1915 if (ourfinisacked) { 1916 /* 1917 * If we can't receive any more 1918 * data, then closing user can proceed. 1919 * Starting the timer is contrary to the 1920 * specification, but if we don't get a FIN 1921 * we'll hang forever. 1922 */ 1923 if (so->so_state & SS_CANTRCVMORE) { 1924 soisdisconnected(so); 1925 if (tcp_maxidle > 0) 1926 TCP_TIMER_ARM(tp, TCPT_2MSL, 1927 tcp_maxidle); 1928 } 1929 tp->t_state = TCPS_FIN_WAIT_2; 1930 } 1931 break; 1932 1933 /* 1934 * In CLOSING STATE in addition to the processing for 1935 * the ESTABLISHED state if the ACK acknowledges our FIN 1936 * then enter the TIME-WAIT state, otherwise ignore 1937 * the segment. 1938 */ 1939 case TCPS_CLOSING: 1940 if (ourfinisacked) { 1941 tp->t_state = TCPS_TIME_WAIT; 1942 tcp_canceltimers(tp); 1943 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1944 soisdisconnected(so); 1945 } 1946 break; 1947 1948 /* 1949 * In LAST_ACK, we may still be waiting for data to drain 1950 * and/or to be acked, as well as for the ack of our FIN. 1951 * If our FIN is now acknowledged, delete the TCB, 1952 * enter the closed state and return. 1953 */ 1954 case TCPS_LAST_ACK: 1955 if (ourfinisacked) { 1956 tp = tcp_close(tp); 1957 goto drop; 1958 } 1959 break; 1960 1961 /* 1962 * In TIME_WAIT state the only thing that should arrive 1963 * is a retransmission of the remote FIN. Acknowledge 1964 * it and restart the finack timer. 1965 */ 1966 case TCPS_TIME_WAIT: 1967 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1968 goto dropafterack; 1969 } 1970 } 1971 1972 step6: 1973 /* 1974 * Update window information. 1975 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1976 */ 1977 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1978 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1979 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1980 /* keep track of pure window updates */ 1981 if (tlen == 0 && 1982 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1983 tcpstat.tcps_rcvwinupd++; 1984 tp->snd_wnd = tiwin; 1985 tp->snd_wl1 = th->th_seq; 1986 tp->snd_wl2 = th->th_ack; 1987 if (tp->snd_wnd > tp->max_sndwnd) 1988 tp->max_sndwnd = tp->snd_wnd; 1989 needoutput = 1; 1990 } 1991 1992 /* 1993 * Process segments with URG. 1994 */ 1995 if ((tiflags & TH_URG) && th->th_urp && 1996 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1997 /* 1998 * This is a kludge, but if we receive and accept 1999 * random urgent pointers, we'll crash in 2000 * soreceive. It's hard to imagine someone 2001 * actually wanting to send this much urgent data. 2002 */ 2003 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2004 th->th_urp = 0; /* XXX */ 2005 tiflags &= ~TH_URG; /* XXX */ 2006 goto dodata; /* XXX */ 2007 } 2008 /* 2009 * If this segment advances the known urgent pointer, 2010 * then mark the data stream. This should not happen 2011 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2012 * a FIN has been received from the remote side. 2013 * In these states we ignore the URG. 2014 * 2015 * According to RFC961 (Assigned Protocols), 2016 * the urgent pointer points to the last octet 2017 * of urgent data. We continue, however, 2018 * to consider it to indicate the first octet 2019 * of data past the urgent section as the original 2020 * spec states (in one of two places). 2021 */ 2022 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2023 tp->rcv_up = th->th_seq + th->th_urp; 2024 so->so_oobmark = so->so_rcv.sb_cc + 2025 (tp->rcv_up - tp->rcv_nxt) - 1; 2026 if (so->so_oobmark == 0) 2027 so->so_state |= SS_RCVATMARK; 2028 sohasoutofband(so); 2029 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2030 } 2031 /* 2032 * Remove out of band data so doesn't get presented to user. 2033 * This can happen independent of advancing the URG pointer, 2034 * but if two URG's are pending at once, some out-of-band 2035 * data may creep in... ick. 2036 */ 2037 if (th->th_urp <= (u_int16_t) tlen 2038 #ifdef SO_OOBINLINE 2039 && (so->so_options & SO_OOBINLINE) == 0 2040 #endif 2041 ) 2042 tcp_pulloutofband(so, th, m, hdroptlen); 2043 } else 2044 /* 2045 * If no out of band data is expected, 2046 * pull receive urgent pointer along 2047 * with the receive window. 2048 */ 2049 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2050 tp->rcv_up = tp->rcv_nxt; 2051 dodata: /* XXX */ 2052 2053 /* 2054 * Process the segment text, merging it into the TCP sequencing queue, 2055 * and arranging for acknowledgement of receipt if necessary. 2056 * This process logically involves adjusting tp->rcv_wnd as data 2057 * is presented to the user (this happens in tcp_usrreq.c, 2058 * case PRU_RCVD). If a FIN has already been received on this 2059 * connection then we just ignore the text. 2060 */ 2061 if ((tlen || (tiflags & TH_FIN)) && 2062 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2063 /* 2064 * Insert segment ti into reassembly queue of tcp with 2065 * control block tp. Return TH_FIN if reassembly now includes 2066 * a segment with FIN. The macro form does the common case 2067 * inline (segment is the next to be received on an 2068 * established connection, and the queue is empty), 2069 * avoiding linkage into and removal from the queue and 2070 * repetition of various conversions. 2071 * Set DELACK for segments received in order, but ack 2072 * immediately when segments are out of order 2073 * (so fast retransmit can work). 2074 */ 2075 /* NOTE: this was TCP_REASS() macro, but used only once */ 2076 TCP_REASS_LOCK(tp); 2077 if (th->th_seq == tp->rcv_nxt && 2078 tp->segq.lh_first == NULL && 2079 tp->t_state == TCPS_ESTABLISHED) { 2080 TCP_SETUP_ACK(tp, th); 2081 tp->rcv_nxt += tlen; 2082 tiflags = th->th_flags & TH_FIN; 2083 tcpstat.tcps_rcvpack++; 2084 tcpstat.tcps_rcvbyte += tlen; 2085 ND6_HINT(tp); 2086 m_adj(m, hdroptlen); 2087 sbappend(&(so)->so_rcv, m); 2088 sorwakeup(so); 2089 } else { 2090 m_adj(m, hdroptlen); 2091 tiflags = tcp_reass(tp, th, m, &tlen); 2092 tp->t_flags |= TF_ACKNOW; 2093 } 2094 TCP_REASS_UNLOCK(tp); 2095 2096 /* 2097 * Note the amount of data that peer has sent into 2098 * our window, in order to estimate the sender's 2099 * buffer size. 2100 */ 2101 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2102 } else { 2103 m_freem(m); 2104 m = NULL; 2105 tiflags &= ~TH_FIN; 2106 } 2107 2108 /* 2109 * If FIN is received ACK the FIN and let the user know 2110 * that the connection is closing. Ignore a FIN received before 2111 * the connection is fully established. 2112 */ 2113 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2114 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2115 socantrcvmore(so); 2116 tp->t_flags |= TF_ACKNOW; 2117 tp->rcv_nxt++; 2118 } 2119 switch (tp->t_state) { 2120 2121 /* 2122 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2123 */ 2124 case TCPS_ESTABLISHED: 2125 tp->t_state = TCPS_CLOSE_WAIT; 2126 break; 2127 2128 /* 2129 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2130 * enter the CLOSING state. 2131 */ 2132 case TCPS_FIN_WAIT_1: 2133 tp->t_state = TCPS_CLOSING; 2134 break; 2135 2136 /* 2137 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2138 * starting the time-wait timer, turning off the other 2139 * standard timers. 2140 */ 2141 case TCPS_FIN_WAIT_2: 2142 tp->t_state = TCPS_TIME_WAIT; 2143 tcp_canceltimers(tp); 2144 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2145 soisdisconnected(so); 2146 break; 2147 2148 /* 2149 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2150 */ 2151 case TCPS_TIME_WAIT: 2152 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2153 break; 2154 } 2155 } 2156 #ifdef TCP_DEBUG 2157 if (so->so_options & SO_DEBUG) 2158 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2159 #endif 2160 2161 /* 2162 * Return any desired output. 2163 */ 2164 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2165 (void) tcp_output(tp); 2166 if (tcp_saveti) 2167 m_freem(tcp_saveti); 2168 return; 2169 2170 badsyn: 2171 /* 2172 * Received a bad SYN. Increment counters and dropwithreset. 2173 */ 2174 tcpstat.tcps_badsyn++; 2175 tp = NULL; 2176 goto dropwithreset; 2177 2178 dropafterack: 2179 /* 2180 * Generate an ACK dropping incoming segment if it occupies 2181 * sequence space, where the ACK reflects our state. 2182 */ 2183 if (tiflags & TH_RST) 2184 goto drop; 2185 m_freem(m); 2186 tp->t_flags |= TF_ACKNOW; 2187 (void) tcp_output(tp); 2188 if (tcp_saveti) 2189 m_freem(tcp_saveti); 2190 return; 2191 2192 dropwithreset_ratelim: 2193 /* 2194 * We may want to rate-limit RSTs in certain situations, 2195 * particularly if we are sending an RST in response to 2196 * an attempt to connect to or otherwise communicate with 2197 * a port for which we have no socket. 2198 */ 2199 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2200 tcp_rst_ppslim) == 0) { 2201 /* XXX stat */ 2202 goto drop; 2203 } 2204 /* ...fall into dropwithreset... */ 2205 2206 dropwithreset: 2207 /* 2208 * Generate a RST, dropping incoming segment. 2209 * Make ACK acceptable to originator of segment. 2210 */ 2211 if (tiflags & TH_RST) 2212 goto drop; 2213 { 2214 /* 2215 * need to recover version # field, which was overwritten on 2216 * ip_cksum computation. 2217 */ 2218 struct ip *sip; 2219 sip = mtod(m, struct ip *); 2220 switch (af) { 2221 #ifdef INET 2222 case AF_INET: 2223 sip->ip_v = 4; 2224 break; 2225 #endif 2226 #ifdef INET6 2227 case AF_INET6: 2228 sip->ip_v = 6; 2229 break; 2230 #endif 2231 } 2232 } 2233 if (tiflags & TH_ACK) 2234 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2235 else { 2236 if (tiflags & TH_SYN) 2237 tlen++; 2238 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2239 TH_RST|TH_ACK); 2240 } 2241 if (tcp_saveti) 2242 m_freem(tcp_saveti); 2243 return; 2244 2245 badcsum: 2246 tcpstat.tcps_rcvbadsum++; 2247 drop: 2248 /* 2249 * Drop space held by incoming segment and return. 2250 */ 2251 if (tp) { 2252 if (tp->t_inpcb) 2253 so = tp->t_inpcb->inp_socket; 2254 #ifdef INET6 2255 else if (tp->t_in6pcb) 2256 so = tp->t_in6pcb->in6p_socket; 2257 #endif 2258 else 2259 so = NULL; 2260 #ifdef TCP_DEBUG 2261 if (so && (so->so_options & SO_DEBUG) != 0) 2262 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2263 #endif 2264 } 2265 if (tcp_saveti) 2266 m_freem(tcp_saveti); 2267 m_freem(m); 2268 return; 2269 } 2270 2271 void 2272 tcp_dooptions(tp, cp, cnt, th, oi) 2273 struct tcpcb *tp; 2274 u_char *cp; 2275 int cnt; 2276 struct tcphdr *th; 2277 struct tcp_opt_info *oi; 2278 { 2279 u_int16_t mss; 2280 int opt, optlen; 2281 2282 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2283 opt = cp[0]; 2284 if (opt == TCPOPT_EOL) 2285 break; 2286 if (opt == TCPOPT_NOP) 2287 optlen = 1; 2288 else { 2289 if (cnt < 2) 2290 break; 2291 optlen = cp[1]; 2292 if (optlen < 2 || optlen > cnt) 2293 break; 2294 } 2295 switch (opt) { 2296 2297 default: 2298 continue; 2299 2300 case TCPOPT_MAXSEG: 2301 if (optlen != TCPOLEN_MAXSEG) 2302 continue; 2303 if (!(th->th_flags & TH_SYN)) 2304 continue; 2305 bcopy(cp + 2, &mss, sizeof(mss)); 2306 oi->maxseg = ntohs(mss); 2307 break; 2308 2309 case TCPOPT_WINDOW: 2310 if (optlen != TCPOLEN_WINDOW) 2311 continue; 2312 if (!(th->th_flags & TH_SYN)) 2313 continue; 2314 tp->t_flags |= TF_RCVD_SCALE; 2315 tp->requested_s_scale = cp[2]; 2316 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2317 #if 0 /*XXX*/ 2318 char *p; 2319 2320 if (ip) 2321 p = ntohl(ip->ip_src); 2322 #ifdef INET6 2323 else if (ip6) 2324 p = ip6_sprintf(&ip6->ip6_src); 2325 #endif 2326 else 2327 p = "(unknown)"; 2328 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2329 "assuming %d\n", 2330 tp->requested_s_scale, p, 2331 TCP_MAX_WINSHIFT); 2332 #else 2333 log(LOG_ERR, "TCP: invalid wscale %d, " 2334 "assuming %d\n", 2335 tp->requested_s_scale, 2336 TCP_MAX_WINSHIFT); 2337 #endif 2338 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2339 } 2340 break; 2341 2342 case TCPOPT_TIMESTAMP: 2343 if (optlen != TCPOLEN_TIMESTAMP) 2344 continue; 2345 oi->ts_present = 1; 2346 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2347 NTOHL(oi->ts_val); 2348 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2349 NTOHL(oi->ts_ecr); 2350 2351 /* 2352 * A timestamp received in a SYN makes 2353 * it ok to send timestamp requests and replies. 2354 */ 2355 if (th->th_flags & TH_SYN) { 2356 tp->t_flags |= TF_RCVD_TSTMP; 2357 tp->ts_recent = oi->ts_val; 2358 tp->ts_recent_age = TCP_TIMESTAMP(tp); 2359 } 2360 break; 2361 case TCPOPT_SACK_PERMITTED: 2362 if (optlen != TCPOLEN_SACK_PERMITTED) 2363 continue; 2364 if (!(th->th_flags & TH_SYN)) 2365 continue; 2366 tp->t_flags &= ~TF_CANT_TXSACK; 2367 break; 2368 2369 case TCPOPT_SACK: 2370 if (tp->t_flags & TF_IGNR_RXSACK) 2371 continue; 2372 if (optlen % 8 != 2 || optlen < 10) 2373 continue; 2374 cp += 2; 2375 optlen -= 2; 2376 for (; optlen > 0; cp -= 8, optlen -= 8) { 2377 tcp_seq lwe, rwe; 2378 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2379 NTOHL(lwe); 2380 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2381 NTOHL(rwe); 2382 /* tcp_mark_sacked(tp, lwe, rwe); */ 2383 } 2384 break; 2385 } 2386 } 2387 } 2388 2389 /* 2390 * Pull out of band byte out of a segment so 2391 * it doesn't appear in the user's data queue. 2392 * It is still reflected in the segment length for 2393 * sequencing purposes. 2394 */ 2395 void 2396 tcp_pulloutofband(so, th, m, off) 2397 struct socket *so; 2398 struct tcphdr *th; 2399 struct mbuf *m; 2400 int off; 2401 { 2402 int cnt = off + th->th_urp - 1; 2403 2404 while (cnt >= 0) { 2405 if (m->m_len > cnt) { 2406 char *cp = mtod(m, caddr_t) + cnt; 2407 struct tcpcb *tp = sototcpcb(so); 2408 2409 tp->t_iobc = *cp; 2410 tp->t_oobflags |= TCPOOB_HAVEDATA; 2411 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2412 m->m_len--; 2413 return; 2414 } 2415 cnt -= m->m_len; 2416 m = m->m_next; 2417 if (m == 0) 2418 break; 2419 } 2420 panic("tcp_pulloutofband"); 2421 } 2422 2423 /* 2424 * Collect new round-trip time estimate 2425 * and update averages and current timeout. 2426 */ 2427 void 2428 tcp_xmit_timer(tp, rtt) 2429 struct tcpcb *tp; 2430 uint32_t rtt; 2431 { 2432 int32_t delta; 2433 2434 tcpstat.tcps_rttupdated++; 2435 if (tp->t_srtt != 0) { 2436 /* 2437 * srtt is stored as fixed point with 3 bits after the 2438 * binary point (i.e., scaled by 8). The following magic 2439 * is equivalent to the smoothing algorithm in rfc793 with 2440 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2441 * point). Adjust rtt to origin 0. 2442 */ 2443 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2444 if ((tp->t_srtt += delta) <= 0) 2445 tp->t_srtt = 1 << 2; 2446 /* 2447 * We accumulate a smoothed rtt variance (actually, a 2448 * smoothed mean difference), then set the retransmit 2449 * timer to smoothed rtt + 4 times the smoothed variance. 2450 * rttvar is stored as fixed point with 2 bits after the 2451 * binary point (scaled by 4). The following is 2452 * equivalent to rfc793 smoothing with an alpha of .75 2453 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2454 * rfc793's wired-in beta. 2455 */ 2456 if (delta < 0) 2457 delta = -delta; 2458 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2459 if ((tp->t_rttvar += delta) <= 0) 2460 tp->t_rttvar = 1 << 2; 2461 } else { 2462 /* 2463 * No rtt measurement yet - use the unsmoothed rtt. 2464 * Set the variance to half the rtt (so our first 2465 * retransmit happens at 3*rtt). 2466 */ 2467 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2468 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2469 } 2470 tp->t_rtttime = 0; 2471 tp->t_rxtshift = 0; 2472 2473 /* 2474 * the retransmit should happen at rtt + 4 * rttvar. 2475 * Because of the way we do the smoothing, srtt and rttvar 2476 * will each average +1/2 tick of bias. When we compute 2477 * the retransmit timer, we want 1/2 tick of rounding and 2478 * 1 extra tick because of +-1/2 tick uncertainty in the 2479 * firing of the timer. The bias will give us exactly the 2480 * 1.5 tick we need. But, because the bias is 2481 * statistical, we have to test that we don't drop below 2482 * the minimum feasible timer (which is 2 ticks). 2483 */ 2484 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2485 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2486 2487 /* 2488 * We received an ack for a packet that wasn't retransmitted; 2489 * it is probably safe to discard any error indications we've 2490 * received recently. This isn't quite right, but close enough 2491 * for now (a route might have failed after we sent a segment, 2492 * and the return path might not be symmetrical). 2493 */ 2494 tp->t_softerror = 0; 2495 } 2496 2497 /* 2498 * Checks for partial ack. If partial ack arrives, force the retransmission 2499 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2500 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2501 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2502 */ 2503 int 2504 tcp_newreno(tp, th) 2505 struct tcpcb *tp; 2506 struct tcphdr *th; 2507 { 2508 tcp_seq onxt = tp->snd_nxt; 2509 u_long ocwnd = tp->snd_cwnd; 2510 2511 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2512 /* 2513 * snd_una has not yet been updated and the socket's send 2514 * buffer has not yet drained off the ACK'd data, so we 2515 * have to leave snd_una as it was to get the correct data 2516 * offset in tcp_output(). 2517 */ 2518 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2519 tp->t_rtttime = 0; 2520 tp->snd_nxt = th->th_ack; 2521 /* 2522 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2523 * is not yet updated when we're called. 2524 */ 2525 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2526 (void) tcp_output(tp); 2527 tp->snd_cwnd = ocwnd; 2528 if (SEQ_GT(onxt, tp->snd_nxt)) 2529 tp->snd_nxt = onxt; 2530 /* 2531 * Partial window deflation. Relies on fact that tp->snd_una 2532 * not updated yet. 2533 */ 2534 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2535 return 1; 2536 } 2537 return 0; 2538 } 2539 2540 2541 /* 2542 * TCP compressed state engine. Currently used to hold compressed 2543 * state for SYN_RECEIVED. 2544 */ 2545 2546 u_long syn_cache_count; 2547 u_int32_t syn_hash1, syn_hash2; 2548 2549 #define SYN_HASH(sa, sp, dp) \ 2550 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2551 ((u_int32_t)(sp)))^syn_hash2))) 2552 #ifndef INET6 2553 #define SYN_HASHALL(hash, src, dst) \ 2554 do { \ 2555 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2556 ((struct sockaddr_in *)(src))->sin_port, \ 2557 ((struct sockaddr_in *)(dst))->sin_port); \ 2558 } while (0) 2559 #else 2560 #define SYN_HASH6(sa, sp, dp) \ 2561 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2562 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2563 & 0x7fffffff) 2564 2565 #define SYN_HASHALL(hash, src, dst) \ 2566 do { \ 2567 switch ((src)->sa_family) { \ 2568 case AF_INET: \ 2569 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2570 ((struct sockaddr_in *)(src))->sin_port, \ 2571 ((struct sockaddr_in *)(dst))->sin_port); \ 2572 break; \ 2573 case AF_INET6: \ 2574 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2575 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2576 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2577 break; \ 2578 default: \ 2579 hash = 0; \ 2580 } \ 2581 } while (0) 2582 #endif /* INET6 */ 2583 2584 #define SYN_CACHE_RM(sc) \ 2585 do { \ 2586 LIST_REMOVE((sc), sc_bucketq); \ 2587 (sc)->sc_tp = NULL; \ 2588 LIST_REMOVE((sc), sc_tpq); \ 2589 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2590 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2591 syn_cache_count--; \ 2592 } while (0) 2593 2594 #define SYN_CACHE_PUT(sc) \ 2595 do { \ 2596 if ((sc)->sc_ipopts) \ 2597 (void) m_free((sc)->sc_ipopts); \ 2598 if ((sc)->sc_route4.ro_rt != NULL) \ 2599 RTFREE((sc)->sc_route4.ro_rt); \ 2600 pool_put(&syn_cache_pool, (sc)); \ 2601 } while (0) 2602 2603 struct pool syn_cache_pool; 2604 2605 /* 2606 * We don't estimate RTT with SYNs, so each packet starts with the default 2607 * RTT and each timer queue has a fixed timeout value. This allows us to 2608 * optimize the timer queues somewhat. 2609 */ 2610 #define SYN_CACHE_TIMER_ARM(sc) \ 2611 do { \ 2612 TCPT_RANGESET((sc)->sc_rxtcur, \ 2613 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2614 TCPTV_REXMTMAX); \ 2615 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2616 } while (0) 2617 2618 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2619 2620 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 2621 2622 void 2623 syn_cache_init() 2624 { 2625 int i; 2626 2627 /* Initialize the hash buckets. */ 2628 for (i = 0; i < tcp_syn_cache_size; i++) 2629 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2630 2631 /* Initialize the timer queues. */ 2632 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2633 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2634 2635 /* Initialize the syn cache pool. */ 2636 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2637 "synpl", 0, NULL, NULL, M_PCB); 2638 } 2639 2640 void 2641 syn_cache_insert(sc, tp) 2642 struct syn_cache *sc; 2643 struct tcpcb *tp; 2644 { 2645 struct syn_cache_head *scp; 2646 struct syn_cache *sc2; 2647 int s, i; 2648 2649 /* 2650 * If there are no entries in the hash table, reinitialize 2651 * the hash secrets. 2652 */ 2653 if (syn_cache_count == 0) { 2654 struct timeval tv; 2655 microtime(&tv); 2656 syn_hash1 = random() ^ (u_long)≻ 2657 syn_hash2 = random() ^ tv.tv_usec; 2658 } 2659 2660 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2661 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2662 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2663 2664 /* 2665 * Make sure that we don't overflow the per-bucket 2666 * limit or the total cache size limit. 2667 */ 2668 s = splsoftnet(); 2669 if (scp->sch_length >= tcp_syn_bucket_limit) { 2670 tcpstat.tcps_sc_bucketoverflow++; 2671 /* 2672 * The bucket is full. Toss the oldest element in the 2673 * bucket. This will be the entry with our bucket 2674 * index closest to the front of the timer queue with 2675 * the largest timeout value. 2676 * 2677 * Note: This timer queue traversal may be expensive, so 2678 * we hope that this doesn't happen very often. It is 2679 * much more likely that we'll overflow the entire 2680 * cache, which is much easier to handle; see below. 2681 */ 2682 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2683 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2684 sc2 != NULL; 2685 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2686 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2687 SYN_CACHE_RM(sc2); 2688 SYN_CACHE_PUT(sc2); 2689 goto insert; /* 2 level break */ 2690 } 2691 } 2692 } 2693 #ifdef DIAGNOSTIC 2694 /* 2695 * This should never happen; we should always find an 2696 * entry in our bucket. 2697 */ 2698 panic("syn_cache_insert: bucketoverflow: impossible"); 2699 #endif 2700 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2701 tcpstat.tcps_sc_overflowed++; 2702 /* 2703 * The cache is full. Toss the oldest entry in the 2704 * entire cache. This is the front entry in the 2705 * first non-empty timer queue with the largest 2706 * timeout value. 2707 */ 2708 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2709 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2710 if (sc2 == NULL) 2711 continue; 2712 SYN_CACHE_RM(sc2); 2713 SYN_CACHE_PUT(sc2); 2714 goto insert; /* symmetry with above */ 2715 } 2716 #ifdef DIAGNOSTIC 2717 /* 2718 * This should never happen; we should always find an 2719 * entry in the cache. 2720 */ 2721 panic("syn_cache_insert: cache overflow: impossible"); 2722 #endif 2723 } 2724 2725 insert: 2726 /* 2727 * Initialize the entry's timer. 2728 */ 2729 sc->sc_rxttot = 0; 2730 sc->sc_rxtshift = 0; 2731 SYN_CACHE_TIMER_ARM(sc); 2732 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2733 2734 /* Link it from tcpcb entry */ 2735 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2736 2737 /* Put it into the bucket. */ 2738 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2739 scp->sch_length++; 2740 syn_cache_count++; 2741 2742 tcpstat.tcps_sc_added++; 2743 splx(s); 2744 } 2745 2746 /* 2747 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2748 * If we have retransmitted an entry the maximum number of times, expire 2749 * that entry. 2750 */ 2751 void 2752 syn_cache_timer() 2753 { 2754 struct syn_cache *sc, *nsc; 2755 int i, s; 2756 2757 s = splsoftnet(); 2758 2759 /* 2760 * First, get all the entries that need to be retransmitted, or 2761 * must be expired due to exceeding the initial keepalive time. 2762 */ 2763 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2764 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2765 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2766 sc = nsc) { 2767 nsc = TAILQ_NEXT(sc, sc_timeq); 2768 2769 /* 2770 * Compute the total amount of time this entry has 2771 * been on a queue. If this entry has been on longer 2772 * than the keep alive timer would allow, expire it. 2773 */ 2774 sc->sc_rxttot += sc->sc_rxtcur; 2775 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2776 tcpstat.tcps_sc_timed_out++; 2777 SYN_CACHE_RM(sc); 2778 SYN_CACHE_PUT(sc); 2779 continue; 2780 } 2781 2782 tcpstat.tcps_sc_retransmitted++; 2783 (void) syn_cache_respond(sc, NULL); 2784 2785 /* Advance this entry onto the next timer queue. */ 2786 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2787 sc->sc_rxtshift = i + 1; 2788 SYN_CACHE_TIMER_ARM(sc); 2789 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2790 sc, sc_timeq); 2791 } 2792 } 2793 2794 /* 2795 * Now get all the entries that are expired due to too many 2796 * retransmissions. 2797 */ 2798 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2799 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2800 sc = nsc) { 2801 nsc = TAILQ_NEXT(sc, sc_timeq); 2802 tcpstat.tcps_sc_timed_out++; 2803 SYN_CACHE_RM(sc); 2804 SYN_CACHE_PUT(sc); 2805 } 2806 splx(s); 2807 } 2808 2809 /* 2810 * Remove syn cache created by the specified tcb entry, 2811 * because this does not make sense to keep them 2812 * (if there's no tcb entry, syn cache entry will never be used) 2813 */ 2814 void 2815 syn_cache_cleanup(tp) 2816 struct tcpcb *tp; 2817 { 2818 struct syn_cache *sc, *nsc; 2819 int s; 2820 2821 s = splsoftnet(); 2822 2823 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2824 nsc = LIST_NEXT(sc, sc_tpq); 2825 2826 #ifdef DIAGNOSTIC 2827 if (sc->sc_tp != tp) 2828 panic("invalid sc_tp in syn_cache_cleanup"); 2829 #endif 2830 SYN_CACHE_RM(sc); 2831 SYN_CACHE_PUT(sc); 2832 } 2833 /* just for safety */ 2834 LIST_INIT(&tp->t_sc); 2835 2836 splx(s); 2837 } 2838 2839 /* 2840 * Find an entry in the syn cache. 2841 */ 2842 struct syn_cache * 2843 syn_cache_lookup(src, dst, headp) 2844 struct sockaddr *src; 2845 struct sockaddr *dst; 2846 struct syn_cache_head **headp; 2847 { 2848 struct syn_cache *sc; 2849 struct syn_cache_head *scp; 2850 u_int32_t hash; 2851 int s; 2852 2853 SYN_HASHALL(hash, src, dst); 2854 2855 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2856 *headp = scp; 2857 s = splsoftnet(); 2858 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2859 sc = LIST_NEXT(sc, sc_bucketq)) { 2860 if (sc->sc_hash != hash) 2861 continue; 2862 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2863 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2864 splx(s); 2865 return (sc); 2866 } 2867 } 2868 splx(s); 2869 return (NULL); 2870 } 2871 2872 /* 2873 * This function gets called when we receive an ACK for a 2874 * socket in the LISTEN state. We look up the connection 2875 * in the syn cache, and if its there, we pull it out of 2876 * the cache and turn it into a full-blown connection in 2877 * the SYN-RECEIVED state. 2878 * 2879 * The return values may not be immediately obvious, and their effects 2880 * can be subtle, so here they are: 2881 * 2882 * NULL SYN was not found in cache; caller should drop the 2883 * packet and send an RST. 2884 * 2885 * -1 We were unable to create the new connection, and are 2886 * aborting it. An ACK,RST is being sent to the peer 2887 * (unless we got screwey sequence numbners; see below), 2888 * because the 3-way handshake has been completed. Caller 2889 * should not free the mbuf, since we may be using it. If 2890 * we are not, we will free it. 2891 * 2892 * Otherwise, the return value is a pointer to the new socket 2893 * associated with the connection. 2894 */ 2895 struct socket * 2896 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2897 struct sockaddr *src; 2898 struct sockaddr *dst; 2899 struct tcphdr *th; 2900 unsigned int hlen, tlen; 2901 struct socket *so; 2902 struct mbuf *m; 2903 { 2904 struct syn_cache *sc; 2905 struct syn_cache_head *scp; 2906 struct inpcb *inp = NULL; 2907 #ifdef INET6 2908 struct in6pcb *in6p = NULL; 2909 #endif 2910 struct tcpcb *tp = 0; 2911 struct mbuf *am; 2912 int s; 2913 struct socket *oso; 2914 2915 s = splsoftnet(); 2916 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2917 splx(s); 2918 return (NULL); 2919 } 2920 2921 /* 2922 * Verify the sequence and ack numbers. Try getting the correct 2923 * response again. 2924 */ 2925 if ((th->th_ack != sc->sc_iss + 1) || 2926 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2927 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2928 (void) syn_cache_respond(sc, m); 2929 splx(s); 2930 return ((struct socket *)(-1)); 2931 } 2932 2933 /* Remove this cache entry */ 2934 SYN_CACHE_RM(sc); 2935 splx(s); 2936 2937 /* 2938 * Ok, create the full blown connection, and set things up 2939 * as they would have been set up if we had created the 2940 * connection when the SYN arrived. If we can't create 2941 * the connection, abort it. 2942 */ 2943 /* 2944 * inp still has the OLD in_pcb stuff, set the 2945 * v6-related flags on the new guy, too. This is 2946 * done particularly for the case where an AF_INET6 2947 * socket is bound only to a port, and a v4 connection 2948 * comes in on that port. 2949 * we also copy the flowinfo from the original pcb 2950 * to the new one. 2951 */ 2952 { 2953 struct inpcb *parentinpcb; 2954 2955 parentinpcb = (struct inpcb *)so->so_pcb; 2956 2957 oso = so; 2958 so = sonewconn(so, SS_ISCONNECTED); 2959 if (so == NULL) 2960 goto resetandabort; 2961 2962 switch (so->so_proto->pr_domain->dom_family) { 2963 #ifdef INET 2964 case AF_INET: 2965 inp = sotoinpcb(so); 2966 break; 2967 #endif 2968 #ifdef INET6 2969 case AF_INET6: 2970 in6p = sotoin6pcb(so); 2971 break; 2972 #endif 2973 } 2974 } 2975 switch (src->sa_family) { 2976 #ifdef INET 2977 case AF_INET: 2978 if (inp) { 2979 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2980 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2981 inp->inp_options = ip_srcroute(); 2982 in_pcbstate(inp, INP_BOUND); 2983 if (inp->inp_options == NULL) { 2984 inp->inp_options = sc->sc_ipopts; 2985 sc->sc_ipopts = NULL; 2986 } 2987 } 2988 #ifdef INET6 2989 else if (in6p) { 2990 /* IPv4 packet to AF_INET6 socket */ 2991 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2992 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2993 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2994 &in6p->in6p_laddr.s6_addr32[3], 2995 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2996 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2997 in6totcpcb(in6p)->t_family = AF_INET; 2998 } 2999 #endif 3000 break; 3001 #endif 3002 #ifdef INET6 3003 case AF_INET6: 3004 if (in6p) { 3005 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3006 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3007 #if 0 3008 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 3009 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 3010 #endif 3011 } 3012 break; 3013 #endif 3014 } 3015 #ifdef INET6 3016 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3017 struct in6pcb *oin6p = sotoin6pcb(oso); 3018 /* inherit socket options from the listening socket */ 3019 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3020 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3021 m_freem(in6p->in6p_options); 3022 in6p->in6p_options = 0; 3023 } 3024 ip6_savecontrol(in6p, &in6p->in6p_options, 3025 mtod(m, struct ip6_hdr *), m); 3026 } 3027 #endif 3028 3029 #ifdef IPSEC 3030 /* 3031 * we make a copy of policy, instead of sharing the policy, 3032 * for better behavior in terms of SA lookup and dead SA removal. 3033 */ 3034 if (inp) { 3035 /* copy old policy into new socket's */ 3036 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3037 printf("tcp_input: could not copy policy\n"); 3038 } 3039 #ifdef INET6 3040 else if (in6p) { 3041 /* copy old policy into new socket's */ 3042 if (ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp, in6p->in6p_sp)) 3043 printf("tcp_input: could not copy policy\n"); 3044 } 3045 #endif 3046 #endif 3047 3048 /* 3049 * Give the new socket our cached route reference. 3050 */ 3051 if (inp) 3052 inp->inp_route = sc->sc_route4; /* struct assignment */ 3053 #ifdef INET6 3054 else 3055 in6p->in6p_route = sc->sc_route6; 3056 #endif 3057 sc->sc_route4.ro_rt = NULL; 3058 3059 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3060 if (am == NULL) 3061 goto resetandabort; 3062 am->m_len = src->sa_len; 3063 bcopy(src, mtod(am, caddr_t), src->sa_len); 3064 if (inp) { 3065 if (in_pcbconnect(inp, am)) { 3066 (void) m_free(am); 3067 goto resetandabort; 3068 } 3069 } 3070 #ifdef INET6 3071 else if (in6p) { 3072 if (src->sa_family == AF_INET) { 3073 /* IPv4 packet to AF_INET6 socket */ 3074 struct sockaddr_in6 *sin6; 3075 sin6 = mtod(am, struct sockaddr_in6 *); 3076 am->m_len = sizeof(*sin6); 3077 bzero(sin6, sizeof(*sin6)); 3078 sin6->sin6_family = AF_INET6; 3079 sin6->sin6_len = sizeof(*sin6); 3080 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3081 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3082 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3083 &sin6->sin6_addr.s6_addr32[3], 3084 sizeof(sin6->sin6_addr.s6_addr32[3])); 3085 } 3086 if (in6_pcbconnect(in6p, am)) { 3087 (void) m_free(am); 3088 goto resetandabort; 3089 } 3090 } 3091 #endif 3092 else { 3093 (void) m_free(am); 3094 goto resetandabort; 3095 } 3096 (void) m_free(am); 3097 3098 if (inp) 3099 tp = intotcpcb(inp); 3100 #ifdef INET6 3101 else if (in6p) 3102 tp = in6totcpcb(in6p); 3103 #endif 3104 else 3105 tp = NULL; 3106 if (sc->sc_request_r_scale != 15) { 3107 tp->requested_s_scale = sc->sc_requested_s_scale; 3108 tp->request_r_scale = sc->sc_request_r_scale; 3109 tp->snd_scale = sc->sc_requested_s_scale; 3110 tp->rcv_scale = sc->sc_request_r_scale; 3111 tp->t_flags |= TF_RCVD_SCALE; 3112 } 3113 if (sc->sc_flags & SCF_TIMESTAMP) 3114 tp->t_flags |= TF_RCVD_TSTMP; 3115 tp->ts_timebase = sc->sc_timebase; 3116 3117 tp->t_template = tcp_template(tp); 3118 if (tp->t_template == 0) { 3119 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3120 so = NULL; 3121 m_freem(m); 3122 goto abort; 3123 } 3124 3125 tp->iss = sc->sc_iss; 3126 tp->irs = sc->sc_irs; 3127 tcp_sendseqinit(tp); 3128 tcp_rcvseqinit(tp); 3129 tp->t_state = TCPS_SYN_RECEIVED; 3130 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3131 tcpstat.tcps_accepts++; 3132 3133 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3134 tp->t_ourmss = sc->sc_ourmaxseg; 3135 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3136 3137 /* 3138 * Initialize the initial congestion window. If we 3139 * had to retransmit the SYN,ACK, we must initialize cwnd 3140 * to 1 segment (i.e. the Loss Window). 3141 */ 3142 if (sc->sc_rxtshift) 3143 tp->snd_cwnd = tp->t_peermss; 3144 else 3145 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 3146 3147 tcp_rmx_rtt(tp); 3148 tp->snd_wl1 = sc->sc_irs; 3149 tp->rcv_up = sc->sc_irs + 1; 3150 3151 /* 3152 * This is what whould have happened in tcp_ouput() when 3153 * the SYN,ACK was sent. 3154 */ 3155 tp->snd_up = tp->snd_una; 3156 tp->snd_max = tp->snd_nxt = tp->iss+1; 3157 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3158 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3159 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3160 tp->last_ack_sent = tp->rcv_nxt; 3161 3162 tcpstat.tcps_sc_completed++; 3163 SYN_CACHE_PUT(sc); 3164 return (so); 3165 3166 resetandabort: 3167 (void) tcp_respond(NULL, m, m, th, 3168 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 3169 abort: 3170 if (so != NULL) 3171 (void) soabort(so); 3172 SYN_CACHE_PUT(sc); 3173 tcpstat.tcps_sc_aborted++; 3174 return ((struct socket *)(-1)); 3175 } 3176 3177 /* 3178 * This function is called when we get a RST for a 3179 * non-existent connection, so that we can see if the 3180 * connection is in the syn cache. If it is, zap it. 3181 */ 3182 3183 void 3184 syn_cache_reset(src, dst, th) 3185 struct sockaddr *src; 3186 struct sockaddr *dst; 3187 struct tcphdr *th; 3188 { 3189 struct syn_cache *sc; 3190 struct syn_cache_head *scp; 3191 int s = splsoftnet(); 3192 3193 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3194 splx(s); 3195 return; 3196 } 3197 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3198 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3199 splx(s); 3200 return; 3201 } 3202 SYN_CACHE_RM(sc); 3203 splx(s); 3204 tcpstat.tcps_sc_reset++; 3205 SYN_CACHE_PUT(sc); 3206 } 3207 3208 void 3209 syn_cache_unreach(src, dst, th) 3210 struct sockaddr *src; 3211 struct sockaddr *dst; 3212 struct tcphdr *th; 3213 { 3214 struct syn_cache *sc; 3215 struct syn_cache_head *scp; 3216 int s; 3217 3218 s = splsoftnet(); 3219 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3220 splx(s); 3221 return; 3222 } 3223 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3224 if (ntohl (th->th_seq) != sc->sc_iss) { 3225 splx(s); 3226 return; 3227 } 3228 3229 /* 3230 * If we've rertransmitted 3 times and this is our second error, 3231 * we remove the entry. Otherwise, we allow it to continue on. 3232 * This prevents us from incorrectly nuking an entry during a 3233 * spurious network outage. 3234 * 3235 * See tcp_notify(). 3236 */ 3237 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3238 sc->sc_flags |= SCF_UNREACH; 3239 splx(s); 3240 return; 3241 } 3242 3243 SYN_CACHE_RM(sc); 3244 splx(s); 3245 tcpstat.tcps_sc_unreach++; 3246 SYN_CACHE_PUT(sc); 3247 } 3248 3249 /* 3250 * Given a LISTEN socket and an inbound SYN request, add 3251 * this to the syn cache, and send back a segment: 3252 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3253 * to the source. 3254 * 3255 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3256 * Doing so would require that we hold onto the data and deliver it 3257 * to the application. However, if we are the target of a SYN-flood 3258 * DoS attack, an attacker could send data which would eventually 3259 * consume all available buffer space if it were ACKed. By not ACKing 3260 * the data, we avoid this DoS scenario. 3261 */ 3262 3263 int 3264 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3265 struct sockaddr *src; 3266 struct sockaddr *dst; 3267 struct tcphdr *th; 3268 unsigned int hlen; 3269 struct socket *so; 3270 struct mbuf *m; 3271 u_char *optp; 3272 int optlen; 3273 struct tcp_opt_info *oi; 3274 { 3275 struct tcpcb tb, *tp; 3276 long win; 3277 struct syn_cache *sc; 3278 struct syn_cache_head *scp; 3279 struct mbuf *ipopts; 3280 3281 tp = sototcpcb(so); 3282 3283 /* 3284 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3285 * 3286 * Note this check is performed in tcp_input() very early on. 3287 */ 3288 3289 /* 3290 * Initialize some local state. 3291 */ 3292 win = sbspace(&so->so_rcv); 3293 if (win > TCP_MAXWIN) 3294 win = TCP_MAXWIN; 3295 3296 switch (src->sa_family) { 3297 #ifdef INET 3298 case AF_INET: 3299 /* 3300 * Remember the IP options, if any. 3301 */ 3302 ipopts = ip_srcroute(); 3303 break; 3304 #endif 3305 default: 3306 ipopts = NULL; 3307 } 3308 3309 if (optp) { 3310 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3311 tcp_dooptions(&tb, optp, optlen, th, oi); 3312 } else 3313 tb.t_flags = 0; 3314 3315 /* 3316 * See if we already have an entry for this connection. 3317 * If we do, resend the SYN,ACK. We do not count this 3318 * as a retransmission (XXX though maybe we should). 3319 */ 3320 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3321 tcpstat.tcps_sc_dupesyn++; 3322 if (ipopts) { 3323 /* 3324 * If we were remembering a previous source route, 3325 * forget it and use the new one we've been given. 3326 */ 3327 if (sc->sc_ipopts) 3328 (void) m_free(sc->sc_ipopts); 3329 sc->sc_ipopts = ipopts; 3330 } 3331 sc->sc_timestamp = tb.ts_recent; 3332 if (syn_cache_respond(sc, m) == 0) { 3333 tcpstat.tcps_sndacks++; 3334 tcpstat.tcps_sndtotal++; 3335 } 3336 return (1); 3337 } 3338 3339 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3340 if (sc == NULL) { 3341 if (ipopts) 3342 (void) m_free(ipopts); 3343 return (0); 3344 } 3345 3346 /* 3347 * Fill in the cache, and put the necessary IP and TCP 3348 * options into the reply. 3349 */ 3350 bzero(sc, sizeof(struct syn_cache)); 3351 bcopy(src, &sc->sc_src, src->sa_len); 3352 bcopy(dst, &sc->sc_dst, dst->sa_len); 3353 sc->sc_flags = 0; 3354 sc->sc_ipopts = ipopts; 3355 sc->sc_irs = th->th_seq; 3356 switch (src->sa_family) { 3357 #ifdef INET 3358 case AF_INET: 3359 { 3360 struct sockaddr_in *srcin = (void *) src; 3361 struct sockaddr_in *dstin = (void *) dst; 3362 3363 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3364 &srcin->sin_addr, dstin->sin_port, 3365 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3366 break; 3367 } 3368 #endif /* INET */ 3369 #ifdef INET6 3370 case AF_INET6: 3371 { 3372 struct sockaddr_in6 *srcin6 = (void *) src; 3373 struct sockaddr_in6 *dstin6 = (void *) dst; 3374 3375 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3376 &srcin6->sin6_addr, dstin6->sin6_port, 3377 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3378 break; 3379 } 3380 #endif /* INET6 */ 3381 } 3382 sc->sc_peermaxseg = oi->maxseg; 3383 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3384 m->m_pkthdr.rcvif : NULL, 3385 sc->sc_src.sa.sa_family); 3386 sc->sc_win = win; 3387 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 3388 sc->sc_timestamp = tb.ts_recent; 3389 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3390 sc->sc_flags |= SCF_TIMESTAMP; 3391 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3392 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3393 sc->sc_requested_s_scale = tb.requested_s_scale; 3394 sc->sc_request_r_scale = 0; 3395 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3396 TCP_MAXWIN << sc->sc_request_r_scale < 3397 so->so_rcv.sb_hiwat) 3398 sc->sc_request_r_scale++; 3399 } else { 3400 sc->sc_requested_s_scale = 15; 3401 sc->sc_request_r_scale = 15; 3402 } 3403 sc->sc_tp = tp; 3404 if (syn_cache_respond(sc, m) == 0) { 3405 syn_cache_insert(sc, tp); 3406 tcpstat.tcps_sndacks++; 3407 tcpstat.tcps_sndtotal++; 3408 } else { 3409 SYN_CACHE_PUT(sc); 3410 tcpstat.tcps_sc_dropped++; 3411 } 3412 return (1); 3413 } 3414 3415 int 3416 syn_cache_respond(sc, m) 3417 struct syn_cache *sc; 3418 struct mbuf *m; 3419 { 3420 struct route *ro; 3421 u_int8_t *optp; 3422 int optlen, error; 3423 u_int16_t tlen; 3424 struct ip *ip = NULL; 3425 #ifdef INET6 3426 struct ip6_hdr *ip6 = NULL; 3427 #endif 3428 struct tcphdr *th; 3429 u_int hlen; 3430 3431 switch (sc->sc_src.sa.sa_family) { 3432 case AF_INET: 3433 hlen = sizeof(struct ip); 3434 ro = &sc->sc_route4; 3435 break; 3436 #ifdef INET6 3437 case AF_INET6: 3438 hlen = sizeof(struct ip6_hdr); 3439 ro = (struct route *)&sc->sc_route6; 3440 break; 3441 #endif 3442 default: 3443 if (m) 3444 m_freem(m); 3445 return EAFNOSUPPORT; 3446 } 3447 3448 /* Compute the size of the TCP options. */ 3449 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3450 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3451 3452 tlen = hlen + sizeof(struct tcphdr) + optlen; 3453 3454 /* 3455 * Create the IP+TCP header from scratch. 3456 */ 3457 if (m) 3458 m_freem(m); 3459 #ifdef DIAGNOSTIC 3460 if (max_linkhdr + tlen > MCLBYTES) 3461 return (ENOBUFS); 3462 #endif 3463 MGETHDR(m, M_DONTWAIT, MT_DATA); 3464 if (m && tlen > MHLEN) { 3465 MCLGET(m, M_DONTWAIT); 3466 if ((m->m_flags & M_EXT) == 0) { 3467 m_freem(m); 3468 m = NULL; 3469 } 3470 } 3471 if (m == NULL) 3472 return (ENOBUFS); 3473 3474 /* Fixup the mbuf. */ 3475 m->m_data += max_linkhdr; 3476 m->m_len = m->m_pkthdr.len = tlen; 3477 #ifdef IPSEC 3478 if (sc->sc_tp) { 3479 struct tcpcb *tp; 3480 struct socket *so; 3481 3482 tp = sc->sc_tp; 3483 if (tp->t_inpcb) 3484 so = tp->t_inpcb->inp_socket; 3485 #ifdef INET6 3486 else if (tp->t_in6pcb) 3487 so = tp->t_in6pcb->in6p_socket; 3488 #endif 3489 else 3490 so = NULL; 3491 /* use IPsec policy on listening socket, on SYN ACK */ 3492 if (ipsec_setsocket(m, so) != 0) { 3493 m_freem(m); 3494 return ENOBUFS; 3495 } 3496 } 3497 #endif 3498 m->m_pkthdr.rcvif = NULL; 3499 memset(mtod(m, u_char *), 0, tlen); 3500 3501 switch (sc->sc_src.sa.sa_family) { 3502 case AF_INET: 3503 ip = mtod(m, struct ip *); 3504 ip->ip_dst = sc->sc_src.sin.sin_addr; 3505 ip->ip_src = sc->sc_dst.sin.sin_addr; 3506 ip->ip_p = IPPROTO_TCP; 3507 th = (struct tcphdr *)(ip + 1); 3508 th->th_dport = sc->sc_src.sin.sin_port; 3509 th->th_sport = sc->sc_dst.sin.sin_port; 3510 break; 3511 #ifdef INET6 3512 case AF_INET6: 3513 ip6 = mtod(m, struct ip6_hdr *); 3514 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3515 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3516 ip6->ip6_nxt = IPPROTO_TCP; 3517 /* ip6_plen will be updated in ip6_output() */ 3518 th = (struct tcphdr *)(ip6 + 1); 3519 th->th_dport = sc->sc_src.sin6.sin6_port; 3520 th->th_sport = sc->sc_dst.sin6.sin6_port; 3521 break; 3522 #endif 3523 default: 3524 th = NULL; 3525 } 3526 3527 th->th_seq = htonl(sc->sc_iss); 3528 th->th_ack = htonl(sc->sc_irs + 1); 3529 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3530 th->th_flags = TH_SYN|TH_ACK; 3531 th->th_win = htons(sc->sc_win); 3532 /* th_sum already 0 */ 3533 /* th_urp already 0 */ 3534 3535 /* Tack on the TCP options. */ 3536 optp = (u_int8_t *)(th + 1); 3537 *optp++ = TCPOPT_MAXSEG; 3538 *optp++ = 4; 3539 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3540 *optp++ = sc->sc_ourmaxseg & 0xff; 3541 3542 if (sc->sc_request_r_scale != 15) { 3543 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3544 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3545 sc->sc_request_r_scale); 3546 optp += 4; 3547 } 3548 3549 if (sc->sc_flags & SCF_TIMESTAMP) { 3550 u_int32_t *lp = (u_int32_t *)(optp); 3551 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3552 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3553 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 3554 *lp = htonl(sc->sc_timestamp); 3555 optp += TCPOLEN_TSTAMP_APPA; 3556 } 3557 3558 /* Compute the packet's checksum. */ 3559 switch (sc->sc_src.sa.sa_family) { 3560 case AF_INET: 3561 ip->ip_len = htons(tlen - hlen); 3562 th->th_sum = 0; 3563 th->th_sum = in_cksum(m, tlen); 3564 break; 3565 #ifdef INET6 3566 case AF_INET6: 3567 ip6->ip6_plen = htons(tlen - hlen); 3568 th->th_sum = 0; 3569 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3570 break; 3571 #endif 3572 } 3573 3574 /* 3575 * Fill in some straggling IP bits. Note the stack expects 3576 * ip_len to be in host order, for convenience. 3577 */ 3578 switch (sc->sc_src.sa.sa_family) { 3579 #ifdef INET 3580 case AF_INET: 3581 ip->ip_len = tlen; 3582 ip->ip_ttl = ip_defttl; 3583 /* XXX tos? */ 3584 break; 3585 #endif 3586 #ifdef INET6 3587 case AF_INET6: 3588 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3589 ip6->ip6_vfc |= IPV6_VERSION; 3590 ip6->ip6_plen = htons(tlen - hlen); 3591 /* ip6_hlim will be initialized afterwards */ 3592 /* XXX flowlabel? */ 3593 break; 3594 #endif 3595 } 3596 3597 switch (sc->sc_src.sa.sa_family) { 3598 #ifdef INET 3599 case AF_INET: 3600 error = ip_output(m, sc->sc_ipopts, ro, 3601 (ip_mtudisc ? IP_MTUDISC : 0), 3602 NULL); 3603 break; 3604 #endif 3605 #ifdef INET6 3606 case AF_INET6: 3607 ip6->ip6_hlim = in6_selecthlim(NULL, 3608 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 3609 3610 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3611 0, NULL, NULL); 3612 break; 3613 #endif 3614 default: 3615 error = EAFNOSUPPORT; 3616 break; 3617 } 3618 return (error); 3619 } 3620
Indexes created Mon Sep 22 05:09:51 GMT 2025