tcp_input.c revision 1.97
1 /* $NetBSD: tcp_input.c,v 1.97 1999/12/08 16:22:20 itojun Exp $ */ 2 3 /* 4 %%% portions-copyright-nrl-95 5 Portions of this software are Copyright 1995-1998 by Randall Atkinson, 6 Ronald Lee, Daniel McDonald, Bao Phan, and Chris Winters. All Rights 7 Reserved. All rights under this copyright have been assigned to the US 8 Naval Research Laboratory (NRL). The NRL Copyright Notice and License 9 Agreement Version 1.1 (January 17, 1995) applies to these portions of the 10 software. 11 You should have received a copy of the license with this software. If you 12 didn't get a copy, you may request one from <license (at) ipv6.nrl.navy.mil>. 13 14 */ 15 16 /* 17 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 18 * All rights reserved. 19 * 20 * Redistribution and use in source and binary forms, with or without 21 * modification, are permitted provided that the following conditions 22 * are met: 23 * 1. Redistributions of source code must retain the above copyright 24 * notice, this list of conditions and the following disclaimer. 25 * 2. Redistributions in binary form must reproduce the above copyright 26 * notice, this list of conditions and the following disclaimer in the 27 * documentation and/or other materials provided with the distribution. 28 * 3. Neither the name of the project nor the names of its contributors 29 * may be used to endorse or promote products derived from this software 30 * without specific prior written permission. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 33 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 34 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 35 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 36 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 37 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 38 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 39 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 40 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 41 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 42 * SUCH DAMAGE. 43 */ 44 45 /*- 46 * Copyright (c) 1997, 1998, 1999 The NetBSD Foundation, Inc. 47 * All rights reserved. 48 * 49 * This code is derived from software contributed to The NetBSD Foundation 50 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 51 * Facility, NASA Ames Research Center. 52 * 53 * Redistribution and use in source and binary forms, with or without 54 * modification, are permitted provided that the following conditions 55 * are met: 56 * 1. Redistributions of source code must retain the above copyright 57 * notice, this list of conditions and the following disclaimer. 58 * 2. Redistributions in binary form must reproduce the above copyright 59 * notice, this list of conditions and the following disclaimer in the 60 * documentation and/or other materials provided with the distribution. 61 * 3. All advertising materials mentioning features or use of this software 62 * must display the following acknowledgement: 63 * This product includes software developed by the NetBSD 64 * Foundation, Inc. and its contributors. 65 * 4. Neither the name of The NetBSD Foundation nor the names of its 66 * contributors may be used to endorse or promote products derived 67 * from this software without specific prior written permission. 68 * 69 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 70 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 71 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 72 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 73 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 74 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 75 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 76 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 77 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 78 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 79 * POSSIBILITY OF SUCH DAMAGE. 80 */ 81 82 /* 83 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 84 * The Regents of the University of California. All rights reserved. 85 * 86 * Redistribution and use in source and binary forms, with or without 87 * modification, are permitted provided that the following conditions 88 * are met: 89 * 1. Redistributions of source code must retain the above copyright 90 * notice, this list of conditions and the following disclaimer. 91 * 2. Redistributions in binary form must reproduce the above copyright 92 * notice, this list of conditions and the following disclaimer in the 93 * documentation and/or other materials provided with the distribution. 94 * 3. All advertising materials mentioning features or use of this software 95 * must display the following acknowledgement: 96 * This product includes software developed by the University of 97 * California, Berkeley and its contributors. 98 * 4. Neither the name of the University nor the names of its contributors 99 * may be used to endorse or promote products derived from this software 100 * without specific prior written permission. 101 * 102 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 103 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 104 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 105 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 106 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 107 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 108 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 109 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 110 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 111 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 112 * SUCH DAMAGE. 113 * 114 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 115 */ 116 117 /* 118 * TODO list for SYN cache stuff: 119 * 120 * Find room for a "state" field, which is needed to keep a 121 * compressed state for TIME_WAIT TCBs. It's been noted already 122 * that this is fairly important for very high-volume web and 123 * mail servers, which use a large number of short-lived 124 * connections. 125 */ 126 127 #include "opt_inet.h" 128 #include "opt_ipsec.h" 129 130 #include <sys/param.h> 131 #include <sys/systm.h> 132 #include <sys/malloc.h> 133 #include <sys/mbuf.h> 134 #include <sys/protosw.h> 135 #include <sys/socket.h> 136 #include <sys/socketvar.h> 137 #include <sys/errno.h> 138 #include <sys/syslog.h> 139 #include <sys/pool.h> 140 #include <sys/domain.h> 141 142 #include <net/if.h> 143 #include <net/route.h> 144 #include <net/if_types.h> 145 146 #include <netinet/in.h> 147 #include <netinet/in_systm.h> 148 #include <netinet/ip.h> 149 #include <netinet/in_pcb.h> 150 #include <netinet/ip_var.h> 151 152 #ifdef INET6 153 #ifndef INET 154 #include <netinet/in.h> 155 #endif 156 #include <netinet/ip6.h> 157 #include <netinet6/in6_pcb.h> 158 #include <netinet6/ip6_var.h> 159 #include <netinet6/in6_var.h> 160 #include <netinet/icmp6.h> 161 #endif 162 163 #include <netinet/tcp.h> 164 #include <netinet/tcp_fsm.h> 165 #include <netinet/tcp_seq.h> 166 #include <netinet/tcp_timer.h> 167 #include <netinet/tcp_var.h> 168 #include <netinet/tcpip.h> 169 #include <netinet/tcp_debug.h> 170 171 #include <machine/stdarg.h> 172 173 #ifdef IPSEC 174 #include <netinet6/ipsec.h> 175 #include <netkey/key.h> 176 #include <netkey/key_debug.h> 177 #endif /*IPSEC*/ 178 #ifdef INET6 179 #include "faith.h" 180 #endif 181 182 int tcprexmtthresh = 3; 183 int tcp_log_refused; 184 185 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 186 187 /* for modulo comparisons of timestamps */ 188 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 189 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 190 191 /* 192 * Macro to compute ACK transmission behavior. Delay the ACK unless 193 * we have already delayed an ACK (must send an ACK every two segments). 194 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 195 * option is enabled. 196 */ 197 #define TCP_SETUP_ACK(tp, th) \ 198 do { \ 199 if ((tp)->t_flags & TF_DELACK || \ 200 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 201 tp->t_flags |= TF_ACKNOW; \ 202 else \ 203 TCP_SET_DELACK(tp); \ 204 } while (0) 205 206 int 207 tcp_reass(tp, th, m, tlen) 208 register struct tcpcb *tp; 209 register struct tcphdr *th; 210 struct mbuf *m; 211 int *tlen; 212 { 213 register struct ipqent *p, *q, *nq, *tiqe = NULL; 214 struct socket *so = NULL; 215 int pkt_flags; 216 tcp_seq pkt_seq; 217 unsigned pkt_len; 218 u_long rcvpartdupbyte = 0; 219 u_long rcvoobyte; 220 221 if (tp->t_inpcb) 222 so = tp->t_inpcb->inp_socket; 223 #ifdef INET6 224 else if (tp->t_in6pcb) 225 so = tp->t_in6pcb->in6p_socket; 226 #endif 227 228 TCP_REASS_LOCK_CHECK(tp); 229 230 /* 231 * Call with th==0 after become established to 232 * force pre-ESTABLISHED data up to user socket. 233 */ 234 if (th == 0) 235 goto present; 236 237 rcvoobyte = *tlen; 238 /* 239 * Copy these to local variables because the tcpiphdr 240 * gets munged while we are collapsing mbufs. 241 */ 242 pkt_seq = th->th_seq; 243 pkt_len = *tlen; 244 pkt_flags = th->th_flags; 245 /* 246 * Find a segment which begins after this one does. 247 */ 248 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 249 nq = q->ipqe_q.le_next; 250 /* 251 * If the received segment is just right after this 252 * fragment, merge the two together and then check 253 * for further overlaps. 254 */ 255 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 256 #ifdef TCPREASS_DEBUG 257 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 258 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 259 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 260 #endif 261 pkt_len += q->ipqe_len; 262 pkt_flags |= q->ipqe_flags; 263 pkt_seq = q->ipqe_seq; 264 m_cat(q->ipqe_m, m); 265 m = q->ipqe_m; 266 goto free_ipqe; 267 } 268 /* 269 * If the received segment is completely past this 270 * fragment, we need to go the next fragment. 271 */ 272 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 273 p = q; 274 continue; 275 } 276 /* 277 * If the fragment is past the received segment, 278 * it (or any following) can't be concatenated. 279 */ 280 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 281 break; 282 /* 283 * We've received all the data in this segment before. 284 * mark it as a duplicate and return. 285 */ 286 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 287 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 288 tcpstat.tcps_rcvduppack++; 289 tcpstat.tcps_rcvdupbyte += pkt_len; 290 m_freem(m); 291 if (tiqe != NULL) 292 pool_put(&ipqent_pool, tiqe); 293 return (0); 294 } 295 /* 296 * Received segment completely overlaps this fragment 297 * so we drop the fragment (this keeps the temporal 298 * ordering of segments correct). 299 */ 300 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 301 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 302 rcvpartdupbyte += q->ipqe_len; 303 m_freem(q->ipqe_m); 304 goto free_ipqe; 305 } 306 /* 307 * RX'ed segment extends past the end of the 308 * fragment. Drop the overlapping bytes. Then 309 * merge the fragment and segment then treat as 310 * a longer received packet. 311 */ 312 if (SEQ_LT(q->ipqe_seq, pkt_seq) 313 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 314 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 315 #ifdef TCPREASS_DEBUG 316 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 317 tp, overlap, 318 pkt_seq, pkt_seq + pkt_len, pkt_len); 319 #endif 320 m_adj(m, overlap); 321 rcvpartdupbyte += overlap; 322 m_cat(q->ipqe_m, m); 323 m = q->ipqe_m; 324 pkt_seq = q->ipqe_seq; 325 pkt_len += q->ipqe_len - overlap; 326 rcvoobyte -= overlap; 327 goto free_ipqe; 328 } 329 /* 330 * RX'ed segment extends past the front of the 331 * fragment. Drop the overlapping bytes on the 332 * received packet. The packet will then be 333 * contatentated with this fragment a bit later. 334 */ 335 if (SEQ_GT(q->ipqe_seq, pkt_seq) 336 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 337 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 338 #ifdef TCPREASS_DEBUG 339 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 340 tp, overlap, 341 pkt_seq, pkt_seq + pkt_len, pkt_len); 342 #endif 343 m_adj(m, -overlap); 344 pkt_len -= overlap; 345 rcvpartdupbyte += overlap; 346 rcvoobyte -= overlap; 347 } 348 /* 349 * If the received segment immediates precedes this 350 * fragment then tack the fragment onto this segment 351 * and reinsert the data. 352 */ 353 if (q->ipqe_seq == pkt_seq + pkt_len) { 354 #ifdef TCPREASS_DEBUG 355 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 356 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 357 pkt_seq, pkt_seq + pkt_len, pkt_len); 358 #endif 359 pkt_len += q->ipqe_len; 360 pkt_flags |= q->ipqe_flags; 361 m_cat(m, q->ipqe_m); 362 LIST_REMOVE(q, ipqe_q); 363 LIST_REMOVE(q, ipqe_timeq); 364 if (tiqe == NULL) { 365 tiqe = q; 366 } else { 367 pool_put(&ipqent_pool, q); 368 } 369 break; 370 } 371 /* 372 * If the fragment is before the segment, remember it. 373 * When this loop is terminated, p will contain the 374 * pointer to fragment that is right before the received 375 * segment. 376 */ 377 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 378 p = q; 379 380 continue; 381 382 /* 383 * This is a common operation. It also will allow 384 * to save doing a malloc/free in most instances. 385 */ 386 free_ipqe: 387 LIST_REMOVE(q, ipqe_q); 388 LIST_REMOVE(q, ipqe_timeq); 389 if (tiqe == NULL) { 390 tiqe = q; 391 } else { 392 pool_put(&ipqent_pool, q); 393 } 394 } 395 396 /* 397 * Allocate a new queue entry since the received segment did not 398 * collapse onto any other out-of-order block; thus we are allocating 399 * a new block. If it had collapsed, tiqe would not be NULL and 400 * we would be reusing it. 401 * XXX If we can't, just drop the packet. XXX 402 */ 403 if (tiqe == NULL) { 404 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 405 if (tiqe == NULL) { 406 tcpstat.tcps_rcvmemdrop++; 407 m_freem(m); 408 return (0); 409 } 410 } 411 412 /* 413 * Update the counters. 414 */ 415 tcpstat.tcps_rcvoopack++; 416 tcpstat.tcps_rcvoobyte += rcvoobyte; 417 if (rcvpartdupbyte) { 418 tcpstat.tcps_rcvpartduppack++; 419 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 420 } 421 422 /* 423 * Insert the new fragment queue entry into both queues. 424 */ 425 tiqe->ipqe_m = m; 426 tiqe->ipqe_seq = pkt_seq; 427 tiqe->ipqe_len = pkt_len; 428 tiqe->ipqe_flags = pkt_flags; 429 if (p == NULL) { 430 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 431 #ifdef TCPREASS_DEBUG 432 if (tiqe->ipqe_seq != tp->rcv_nxt) 433 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 434 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 435 #endif 436 } else { 437 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 438 #ifdef TCPREASS_DEBUG 439 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 440 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 441 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 442 #endif 443 } 444 445 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 446 447 present: 448 /* 449 * Present data to user, advancing rcv_nxt through 450 * completed sequence space. 451 */ 452 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 453 return (0); 454 q = tp->segq.lh_first; 455 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 456 return (0); 457 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 458 return (0); 459 460 tp->rcv_nxt += q->ipqe_len; 461 pkt_flags = q->ipqe_flags & TH_FIN; 462 463 LIST_REMOVE(q, ipqe_q); 464 LIST_REMOVE(q, ipqe_timeq); 465 if (so->so_state & SS_CANTRCVMORE) 466 m_freem(q->ipqe_m); 467 else 468 sbappend(&so->so_rcv, q->ipqe_m); 469 pool_put(&ipqent_pool, q); 470 sorwakeup(so); 471 return (pkt_flags); 472 } 473 474 #if defined(INET6) && !defined(TCP6) 475 int 476 tcp6_input(mp, offp, proto) 477 struct mbuf **mp; 478 int *offp, proto; 479 { 480 struct mbuf *m = *mp; 481 482 #if defined(NFAITH) && 0 < NFAITH 483 if (m->m_pkthdr.rcvif) { 484 if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 485 /* XXX send icmp6 host/port unreach? */ 486 m_freem(m); 487 return IPPROTO_DONE; 488 } 489 } 490 #endif 491 492 /* 493 * draft-itojun-ipv6-tcp-to-anycast 494 * better place to put this in? 495 */ 496 if (m->m_flags & M_ANYCAST6) { 497 if (m->m_len >= sizeof(struct ip6_hdr)) { 498 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 499 icmp6_error(m, ICMP6_DST_UNREACH, 500 ICMP6_DST_UNREACH_ADDR, 501 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 502 } else 503 m_freem(m); 504 return IPPROTO_DONE; 505 } 506 507 tcp_input(m, *offp, proto); 508 return IPPROTO_DONE; 509 } 510 #endif 511 512 /* 513 * TCP input routine, follows pages 65-76 of the 514 * protocol specification dated September, 1981 very closely. 515 */ 516 void 517 #if __STDC__ 518 tcp_input(struct mbuf *m, ...) 519 #else 520 tcp_input(m, va_alist) 521 register struct mbuf *m; 522 #endif 523 { 524 int proto; 525 register struct tcphdr *th; 526 struct ip *ip; 527 register struct inpcb *inp; 528 #ifdef INET6 529 struct ip6_hdr *ip6; 530 register struct in6pcb *in6p; 531 #endif 532 caddr_t optp = NULL; 533 int optlen = 0; 534 int len, tlen, toff, hdroptlen = 0; 535 register struct tcpcb *tp = 0; 536 register int tiflags; 537 struct socket *so = NULL; 538 int todrop, acked, ourfinisacked, needoutput = 0; 539 short ostate = 0; 540 int iss = 0; 541 u_long tiwin; 542 struct tcp_opt_info opti; 543 int off, iphlen; 544 va_list ap; 545 int af; /* af on the wire */ 546 struct mbuf *tcp_saveti = NULL; 547 548 va_start(ap, m); 549 toff = va_arg(ap, int); 550 proto = va_arg(ap, int); 551 va_end(ap); 552 553 tcpstat.tcps_rcvtotal++; 554 555 bzero(&opti, sizeof(opti)); 556 opti.ts_present = 0; 557 opti.maxseg = 0; 558 559 /* 560 * Get IP and TCP header together in first mbuf. 561 * Note: IP leaves IP header in first mbuf. 562 */ 563 ip = mtod(m, struct ip *); 564 #ifdef INET6 565 ip6 = NULL; 566 #endif 567 switch (ip->ip_v) { 568 case 4: 569 af = AF_INET; 570 iphlen = sizeof(struct ip); 571 /* would like to get rid of this... */ 572 if (toff > sizeof (struct ip)) { 573 ip_stripoptions(m, (struct mbuf *)0); 574 toff = sizeof(struct ip); 575 } 576 if (m->m_len < toff + sizeof (struct tcphdr)) { 577 if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) { 578 tcpstat.tcps_rcvshort++; 579 return; 580 } 581 } 582 ip = mtod(m, struct ip *); 583 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 584 585 /* 586 * Checksum extended TCP header and data. 587 */ 588 len = ip->ip_len; 589 tlen = len - toff; 590 { 591 struct ipovly *ipov; 592 ipov = (struct ipovly *)ip; 593 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 594 ipov->ih_len = htons(tlen); 595 } 596 if (in_cksum(m, len) != 0) { 597 tcpstat.tcps_rcvbadsum++; 598 goto drop; 599 } 600 break; 601 #ifdef INET6 602 case 6: 603 ip = NULL; 604 iphlen = sizeof(struct ip6_hdr); 605 af = AF_INET6; 606 if (m->m_len < toff + sizeof(struct tcphdr)) { 607 m = m_pullup(m, toff + sizeof(struct tcphdr)); /*XXX*/ 608 if (m == NULL) { 609 tcpstat.tcps_rcvshort++; 610 return; 611 } 612 } 613 ip6 = mtod(m, struct ip6_hdr *); 614 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 615 616 /* 617 * Checksum extended TCP header and data. 618 */ 619 len = m->m_pkthdr.len; 620 tlen = len - toff; 621 if (in6_cksum(m, IPPROTO_TCP, toff, tlen)) { 622 tcpstat.tcps_rcvbadsum++; 623 goto drop; 624 } 625 break; 626 #endif 627 default: 628 m_freem(m); 629 return; 630 } 631 632 /* 633 * Check that TCP offset makes sense, 634 * pull out TCP options and adjust length. XXX 635 */ 636 off = th->th_off << 2; 637 if (off < sizeof (struct tcphdr) || off > tlen) { 638 tcpstat.tcps_rcvbadoff++; 639 goto drop; 640 } 641 tlen -= off; 642 643 /* 644 * tcp_input() has been modified to use tlen to mean the TCP data 645 * length throughout the function. Other functions can use 646 * m->m_pkthdr.len as the basis for calculating the TCP data length. 647 * rja 648 */ 649 650 if (off > sizeof (struct tcphdr)) { 651 if (m->m_len < toff + off) { 652 if ((m = m_pullup(m, toff + off)) == 0) { 653 tcpstat.tcps_rcvshort++; 654 return; 655 } 656 switch (af) { 657 case AF_INET: 658 ip = mtod(m, struct ip *); 659 break; 660 #ifdef INET6 661 case AF_INET6: 662 ip6 = mtod(m, struct ip6_hdr *); 663 break; 664 #endif 665 } 666 th = (struct tcphdr *)(mtod(m, caddr_t) + toff); 667 } 668 optlen = off - sizeof (struct tcphdr); 669 optp = mtod(m, caddr_t) + toff + sizeof (struct tcphdr); 670 /* 671 * Do quick retrieval of timestamp options ("options 672 * prediction?"). If timestamp is the only option and it's 673 * formatted as recommended in RFC 1323 appendix A, we 674 * quickly get the values now and not bother calling 675 * tcp_dooptions(), etc. 676 */ 677 if ((optlen == TCPOLEN_TSTAMP_APPA || 678 (optlen > TCPOLEN_TSTAMP_APPA && 679 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 680 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 681 (th->th_flags & TH_SYN) == 0) { 682 opti.ts_present = 1; 683 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 684 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 685 optp = NULL; /* we've parsed the options */ 686 } 687 } 688 tiflags = th->th_flags; 689 690 /* 691 * Convert TCP protocol specific fields to host format. 692 */ 693 NTOHL(th->th_seq); 694 NTOHL(th->th_ack); 695 NTOHS(th->th_win); 696 NTOHS(th->th_urp); 697 698 /* 699 * Locate pcb for segment. 700 */ 701 findpcb: 702 inp = NULL; 703 #ifdef INET6 704 in6p = NULL; 705 #endif 706 switch (af) { 707 case AF_INET: 708 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 709 ip->ip_dst, th->th_dport); 710 if (inp == 0) { 711 ++tcpstat.tcps_pcbhashmiss; 712 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 713 } 714 #if defined(INET6) && !defined(TCP6) 715 if (inp == 0) { 716 struct in6_addr s, d; 717 718 /* mapped addr case */ 719 bzero(&s, sizeof(s)); 720 s.s6_addr16[5] = htons(0xffff); 721 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 722 bzero(&d, sizeof(d)); 723 d.s6_addr16[5] = htons(0xffff); 724 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 725 in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport, 726 &d, th->th_dport, 0); 727 if (in6p == 0) { 728 ++tcpstat.tcps_pcbhashmiss; 729 in6p = in6_pcblookup_bind(&tcb6, &d, 730 th->th_dport, 0); 731 } 732 } 733 #endif 734 #ifndef INET6 735 if (inp == 0) 736 #else 737 if (inp == 0 && in6p == 0) 738 #endif 739 { 740 ++tcpstat.tcps_noport; 741 if (tcp_log_refused && (tiflags & TH_SYN)) { 742 #ifndef INET6 743 char src[4*sizeof "123"]; 744 char dst[4*sizeof "123"]; 745 #else 746 char src[INET6_ADDRSTRLEN]; 747 char dst[INET6_ADDRSTRLEN]; 748 #endif 749 if (ip) { 750 strcpy(src, inet_ntoa(ip->ip_src)); 751 strcpy(dst, inet_ntoa(ip->ip_dst)); 752 } 753 #ifdef INET6 754 else if (ip6) { 755 strcpy(src, ip6_sprintf(&ip6->ip6_src)); 756 strcpy(dst, ip6_sprintf(&ip6->ip6_dst)); 757 } 758 #endif 759 else { 760 strcpy(src, "(unknown)"); 761 strcpy(dst, "(unknown)"); 762 } 763 log(LOG_INFO, 764 "Connection attempt to TCP %s:%d from %s:%d\n", 765 dst, ntohs(th->th_dport), 766 src, ntohs(th->th_sport)); 767 } 768 goto dropwithreset; 769 } 770 #ifdef IPSEC 771 if (inp && ipsec4_in_reject(m, inp)) { 772 ipsecstat.in_polvio++; 773 goto drop; 774 } 775 #ifdef INET6 776 else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) { 777 ipsecstat.in_polvio++; 778 goto drop; 779 } 780 #endif 781 #endif /*IPSEC*/ 782 break; 783 #if defined(INET6) && !defined(TCP6) 784 case AF_INET6: 785 { 786 int faith; 787 788 #if defined(NFAITH) && NFAITH > 0 789 if (m->m_pkthdr.rcvif 790 && m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 791 faith = 1; 792 } else 793 faith = 0; 794 #else 795 faith = 0; 796 #endif 797 in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport, 798 &ip6->ip6_dst, th->th_dport, faith); 799 if (in6p == NULL) { 800 ++tcpstat.tcps_pcbhashmiss; 801 in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst, 802 th->th_dport, faith); 803 } 804 if (in6p == NULL) { 805 ++tcpstat.tcps_noport; 806 goto dropwithreset; 807 } 808 #ifdef IPSEC 809 if (ipsec6_in_reject(m, in6p)) { 810 ipsec6stat.in_polvio++; 811 goto drop; 812 } 813 #endif /*IPSEC*/ 814 break; 815 } 816 #endif 817 } 818 819 /* 820 * If the state is CLOSED (i.e., TCB does not exist) then 821 * all data in the incoming segment is discarded. 822 * If the TCB exists but is in CLOSED state, it is embryonic, 823 * but should either do a listen or a connect soon. 824 */ 825 tp = NULL; 826 so = NULL; 827 if (inp) { 828 tp = intotcpcb(inp); 829 so = inp->inp_socket; 830 } 831 #ifdef INET6 832 else if (in6p) { 833 tp = in6totcpcb(in6p); 834 so = in6p->in6p_socket; 835 } 836 #endif 837 if (tp == 0) { 838 goto dropwithreset; 839 } 840 if (tp->t_state == TCPS_CLOSED) 841 goto drop; 842 843 /* Unscale the window into a 32-bit value. */ 844 if ((tiflags & TH_SYN) == 0) 845 tiwin = th->th_win << tp->snd_scale; 846 else 847 tiwin = th->th_win; 848 849 #ifdef INET6 850 /* save packet options if user wanted */ 851 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 852 if (in6p->in6p_options) { 853 m_freem(in6p->in6p_options); 854 in6p->in6p_options = 0; 855 } 856 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 857 } 858 #endif 859 860 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 861 union syn_cache_sa src; 862 union syn_cache_sa dst; 863 864 bzero(&src, sizeof(src)); 865 bzero(&dst, sizeof(dst)); 866 switch (af) { 867 case AF_INET: 868 src.sin.sin_len = sizeof(struct sockaddr_in); 869 src.sin.sin_family = AF_INET; 870 src.sin.sin_addr = ip->ip_src; 871 src.sin.sin_port = th->th_sport; 872 873 dst.sin.sin_len = sizeof(struct sockaddr_in); 874 dst.sin.sin_family = AF_INET; 875 dst.sin.sin_addr = ip->ip_dst; 876 dst.sin.sin_port = th->th_dport; 877 break; 878 #ifdef INET6 879 case AF_INET6: 880 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 881 src.sin6.sin6_family = AF_INET6; 882 src.sin6.sin6_addr = ip6->ip6_src; 883 src.sin6.sin6_port = th->th_sport; 884 885 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 886 dst.sin6.sin6_family = AF_INET6; 887 dst.sin6.sin6_addr = ip6->ip6_dst; 888 dst.sin6.sin6_port = th->th_dport; 889 break; 890 #endif /* INET6 */ 891 default: 892 goto badsyn; /*sanity*/ 893 } 894 895 if (so->so_options & SO_DEBUG) { 896 ostate = tp->t_state; 897 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 898 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 899 m_freem(tcp_saveti); 900 tcp_saveti = NULL; 901 } else { 902 tcp_saveti->m_len += sizeof(struct tcphdr); 903 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 904 sizeof(struct tcphdr)); 905 } 906 if (tcp_saveti) { 907 /* 908 * need to recover version # field, which was 909 * overwritten on ip_cksum computation. 910 */ 911 struct ip *sip; 912 sip = mtod(tcp_saveti, struct ip *); 913 switch (af) { 914 case AF_INET: 915 sip->ip_v = 4; 916 break; 917 #ifdef INET6 918 case AF_INET6: 919 sip->ip_v = 6; 920 break; 921 #endif 922 } 923 } 924 } 925 if (so->so_options & SO_ACCEPTCONN) { 926 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 927 if (tiflags & TH_RST) { 928 syn_cache_reset(&src.sa, &dst.sa, th); 929 } else if ((tiflags & (TH_ACK|TH_SYN)) == 930 (TH_ACK|TH_SYN)) { 931 /* 932 * Received a SYN,ACK. This should 933 * never happen while we are in 934 * LISTEN. Send an RST. 935 */ 936 goto badsyn; 937 } else if (tiflags & TH_ACK) { 938 so = syn_cache_get(&src.sa, &dst.sa, 939 th, toff, tlen, so, m); 940 if (so == NULL) { 941 /* 942 * We don't have a SYN for 943 * this ACK; send an RST. 944 */ 945 goto badsyn; 946 } else if (so == 947 (struct socket *)(-1)) { 948 /* 949 * We were unable to create 950 * the connection. If the 951 * 3-way handshake was 952 * completed, and RST has 953 * been sent to the peer. 954 * Since the mbuf might be 955 * in use for the reply, 956 * do not free it. 957 */ 958 m = NULL; 959 } else { 960 /* 961 * We have created a 962 * full-blown connection. 963 */ 964 tp = NULL; 965 inp = NULL; 966 #ifdef INET6 967 in6p = NULL; 968 #endif 969 switch (so->so_proto->pr_domain->dom_family) { 970 case AF_INET: 971 inp = sotoinpcb(so); 972 tp = intotcpcb(inp); 973 break; 974 #ifdef INET6 975 case AF_INET6: 976 in6p = sotoin6pcb(so); 977 tp = in6totcpcb(in6p); 978 break; 979 #endif 980 } 981 if (tp == NULL) 982 goto badsyn; /*XXX*/ 983 tiwin <<= tp->snd_scale; 984 goto after_listen; 985 } 986 } else { 987 /* 988 * None of RST, SYN or ACK was set. 989 * This is an invalid packet for a 990 * TCB in LISTEN state. Send a RST. 991 */ 992 goto badsyn; 993 } 994 } else { 995 /* 996 * Received a SYN. 997 */ 998 999 /* 1000 * LISTEN socket received a SYN 1001 * from itself? This can't possibly 1002 * be valid; drop the packet. 1003 */ 1004 if (th->th_sport == th->th_dport) { 1005 int i; 1006 1007 switch (af) { 1008 case AF_INET: 1009 i = in_hosteq(ip->ip_src, ip->ip_dst); 1010 break; 1011 #ifdef INET6 1012 case AF_INET6: 1013 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1014 break; 1015 #endif 1016 default: 1017 i = 1; 1018 } 1019 if (i) { 1020 tcpstat.tcps_badsyn++; 1021 goto drop; 1022 } 1023 } 1024 1025 /* 1026 * SYN looks ok; create compressed TCP 1027 * state for it. 1028 */ 1029 if (so->so_qlen <= so->so_qlimit && 1030 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1031 so, m, optp, optlen, &opti)) 1032 m = NULL; 1033 } 1034 goto drop; 1035 } 1036 } 1037 1038 after_listen: 1039 #ifdef DIAGNOSTIC 1040 /* 1041 * Should not happen now that all embryonic connections 1042 * are handled with compressed state. 1043 */ 1044 if (tp->t_state == TCPS_LISTEN) 1045 panic("tcp_input: TCPS_LISTEN"); 1046 #endif 1047 1048 /* 1049 * Segment received on connection. 1050 * Reset idle time and keep-alive timer. 1051 */ 1052 tp->t_idle = 0; 1053 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1054 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1055 1056 /* 1057 * Process options. 1058 */ 1059 if (optp) 1060 tcp_dooptions(tp, optp, optlen, th, &opti); 1061 1062 /* 1063 * Header prediction: check for the two common cases 1064 * of a uni-directional data xfer. If the packet has 1065 * no control flags, is in-sequence, the window didn't 1066 * change and we're not retransmitting, it's a 1067 * candidate. If the length is zero and the ack moved 1068 * forward, we're the sender side of the xfer. Just 1069 * free the data acked & wake any higher level process 1070 * that was blocked waiting for space. If the length 1071 * is non-zero and the ack didn't move, we're the 1072 * receiver side. If we're getting packets in-order 1073 * (the reassembly queue is empty), add the data to 1074 * the socket buffer and note that we need a delayed ack. 1075 */ 1076 if (tp->t_state == TCPS_ESTABLISHED && 1077 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1078 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1079 th->th_seq == tp->rcv_nxt && 1080 tiwin && tiwin == tp->snd_wnd && 1081 tp->snd_nxt == tp->snd_max) { 1082 1083 /* 1084 * If last ACK falls within this segment's sequence numbers, 1085 * record the timestamp. 1086 */ 1087 if (opti.ts_present && 1088 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1089 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1090 tp->ts_recent_age = tcp_now; 1091 tp->ts_recent = opti.ts_val; 1092 } 1093 1094 if (tlen == 0) { 1095 if (SEQ_GT(th->th_ack, tp->snd_una) && 1096 SEQ_LEQ(th->th_ack, tp->snd_max) && 1097 tp->snd_cwnd >= tp->snd_wnd && 1098 tp->t_dupacks < tcprexmtthresh) { 1099 /* 1100 * this is a pure ack for outstanding data. 1101 */ 1102 ++tcpstat.tcps_predack; 1103 if (opti.ts_present && opti.ts_ecr) 1104 tcp_xmit_timer(tp, 1105 tcp_now - opti.ts_ecr + 1); 1106 else if (tp->t_rtt && 1107 SEQ_GT(th->th_ack, tp->t_rtseq)) 1108 tcp_xmit_timer(tp, tp->t_rtt); 1109 acked = th->th_ack - tp->snd_una; 1110 tcpstat.tcps_rcvackpack++; 1111 tcpstat.tcps_rcvackbyte += acked; 1112 sbdrop(&so->so_snd, acked); 1113 /* 1114 * We want snd_recover to track snd_una to 1115 * avoid sequence wraparound problems for 1116 * very large transfers. 1117 */ 1118 tp->snd_una = tp->snd_recover = th->th_ack; 1119 m_freem(m); 1120 1121 /* 1122 * If all outstanding data are acked, stop 1123 * retransmit timer, otherwise restart timer 1124 * using current (possibly backed-off) value. 1125 * If process is waiting for space, 1126 * wakeup/selwakeup/signal. If data 1127 * are ready to send, let tcp_output 1128 * decide between more output or persist. 1129 */ 1130 if (tp->snd_una == tp->snd_max) 1131 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1132 else if (TCP_TIMER_ISARMED(tp, 1133 TCPT_PERSIST) == 0) 1134 TCP_TIMER_ARM(tp, TCPT_REXMT, 1135 tp->t_rxtcur); 1136 1137 sowwakeup(so); 1138 if (so->so_snd.sb_cc) 1139 (void) tcp_output(tp); 1140 if (tcp_saveti) 1141 m_freem(tcp_saveti); 1142 return; 1143 } 1144 } else if (th->th_ack == tp->snd_una && 1145 tp->segq.lh_first == NULL && 1146 tlen <= sbspace(&so->so_rcv)) { 1147 /* 1148 * this is a pure, in-sequence data packet 1149 * with nothing on the reassembly queue and 1150 * we have enough buffer space to take it. 1151 */ 1152 ++tcpstat.tcps_preddat; 1153 tp->rcv_nxt += tlen; 1154 tcpstat.tcps_rcvpack++; 1155 tcpstat.tcps_rcvbyte += tlen; 1156 /* 1157 * Drop TCP, IP headers and TCP options then add data 1158 * to socket buffer. 1159 */ 1160 m_adj(m, toff + off); 1161 sbappend(&so->so_rcv, m); 1162 sorwakeup(so); 1163 TCP_SETUP_ACK(tp, th); 1164 if (tp->t_flags & TF_ACKNOW) 1165 (void) tcp_output(tp); 1166 if (tcp_saveti) 1167 m_freem(tcp_saveti); 1168 return; 1169 } 1170 } 1171 1172 /* 1173 * Compute mbuf offset to TCP data segment. 1174 */ 1175 hdroptlen = toff + off; 1176 1177 /* 1178 * Calculate amount of space in receive window, 1179 * and then do TCP input processing. 1180 * Receive window is amount of space in rcv queue, 1181 * but not less than advertised window. 1182 */ 1183 { int win; 1184 1185 win = sbspace(&so->so_rcv); 1186 if (win < 0) 1187 win = 0; 1188 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1189 } 1190 1191 switch (tp->t_state) { 1192 1193 /* 1194 * If the state is SYN_SENT: 1195 * if seg contains an ACK, but not for our SYN, drop the input. 1196 * if seg contains a RST, then drop the connection. 1197 * if seg does not contain SYN, then drop it. 1198 * Otherwise this is an acceptable SYN segment 1199 * initialize tp->rcv_nxt and tp->irs 1200 * if seg contains ack then advance tp->snd_una 1201 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1202 * arrange for segment to be acked (eventually) 1203 * continue processing rest of data/controls, beginning with URG 1204 */ 1205 case TCPS_SYN_SENT: 1206 if ((tiflags & TH_ACK) && 1207 (SEQ_LEQ(th->th_ack, tp->iss) || 1208 SEQ_GT(th->th_ack, tp->snd_max))) 1209 goto dropwithreset; 1210 if (tiflags & TH_RST) { 1211 if (tiflags & TH_ACK) 1212 tp = tcp_drop(tp, ECONNREFUSED); 1213 goto drop; 1214 } 1215 if ((tiflags & TH_SYN) == 0) 1216 goto drop; 1217 if (tiflags & TH_ACK) { 1218 tp->snd_una = tp->snd_recover = th->th_ack; 1219 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1220 tp->snd_nxt = tp->snd_una; 1221 } 1222 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1223 tp->irs = th->th_seq; 1224 tcp_rcvseqinit(tp); 1225 tp->t_flags |= TF_ACKNOW; 1226 tcp_mss_from_peer(tp, opti.maxseg); 1227 1228 /* 1229 * Initialize the initial congestion window. If we 1230 * had to retransmit the SYN, we must initialize cwnd 1231 * to 1 segment (i.e. the Loss Window). 1232 */ 1233 if (tp->t_flags & TF_SYN_REXMT) 1234 tp->snd_cwnd = tp->t_peermss; 1235 else 1236 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 1237 tp->t_peermss); 1238 1239 tcp_rmx_rtt(tp); 1240 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1241 tcpstat.tcps_connects++; 1242 soisconnected(so); 1243 tcp_established(tp); 1244 /* Do window scaling on this connection? */ 1245 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1246 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1247 tp->snd_scale = tp->requested_s_scale; 1248 tp->rcv_scale = tp->request_r_scale; 1249 } 1250 TCP_REASS_LOCK(tp); 1251 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1252 TCP_REASS_UNLOCK(tp); 1253 /* 1254 * if we didn't have to retransmit the SYN, 1255 * use its rtt as our initial srtt & rtt var. 1256 */ 1257 if (tp->t_rtt) 1258 tcp_xmit_timer(tp, tp->t_rtt); 1259 } else 1260 tp->t_state = TCPS_SYN_RECEIVED; 1261 1262 /* 1263 * Advance th->th_seq to correspond to first data byte. 1264 * If data, trim to stay within window, 1265 * dropping FIN if necessary. 1266 */ 1267 th->th_seq++; 1268 if (tlen > tp->rcv_wnd) { 1269 todrop = tlen - tp->rcv_wnd; 1270 m_adj(m, -todrop); 1271 tlen = tp->rcv_wnd; 1272 tiflags &= ~TH_FIN; 1273 tcpstat.tcps_rcvpackafterwin++; 1274 tcpstat.tcps_rcvbyteafterwin += todrop; 1275 } 1276 tp->snd_wl1 = th->th_seq - 1; 1277 tp->rcv_up = th->th_seq; 1278 goto step6; 1279 1280 /* 1281 * If the state is SYN_RECEIVED: 1282 * If seg contains an ACK, but not for our SYN, drop the input 1283 * and generate an RST. See page 36, rfc793 1284 */ 1285 case TCPS_SYN_RECEIVED: 1286 if ((tiflags & TH_ACK) && 1287 (SEQ_LEQ(th->th_ack, tp->iss) || 1288 SEQ_GT(th->th_ack, tp->snd_max))) 1289 goto dropwithreset; 1290 break; 1291 } 1292 1293 /* 1294 * States other than LISTEN or SYN_SENT. 1295 * First check timestamp, if present. 1296 * Then check that at least some bytes of segment are within 1297 * receive window. If segment begins before rcv_nxt, 1298 * drop leading data (and SYN); if nothing left, just ack. 1299 * 1300 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1301 * and it's less than ts_recent, drop it. 1302 */ 1303 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1304 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1305 1306 /* Check to see if ts_recent is over 24 days old. */ 1307 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1308 /* 1309 * Invalidate ts_recent. If this segment updates 1310 * ts_recent, the age will be reset later and ts_recent 1311 * will get a valid value. If it does not, setting 1312 * ts_recent to zero will at least satisfy the 1313 * requirement that zero be placed in the timestamp 1314 * echo reply when ts_recent isn't valid. The 1315 * age isn't reset until we get a valid ts_recent 1316 * because we don't want out-of-order segments to be 1317 * dropped when ts_recent is old. 1318 */ 1319 tp->ts_recent = 0; 1320 } else { 1321 tcpstat.tcps_rcvduppack++; 1322 tcpstat.tcps_rcvdupbyte += tlen; 1323 tcpstat.tcps_pawsdrop++; 1324 goto dropafterack; 1325 } 1326 } 1327 1328 todrop = tp->rcv_nxt - th->th_seq; 1329 if (todrop > 0) { 1330 if (tiflags & TH_SYN) { 1331 tiflags &= ~TH_SYN; 1332 th->th_seq++; 1333 if (th->th_urp > 1) 1334 th->th_urp--; 1335 else { 1336 tiflags &= ~TH_URG; 1337 th->th_urp = 0; 1338 } 1339 todrop--; 1340 } 1341 if (todrop > tlen || 1342 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1343 /* 1344 * Any valid FIN must be to the left of the window. 1345 * At this point the FIN must be a duplicate or 1346 * out of sequence; drop it. 1347 */ 1348 tiflags &= ~TH_FIN; 1349 /* 1350 * Send an ACK to resynchronize and drop any data. 1351 * But keep on processing for RST or ACK. 1352 */ 1353 tp->t_flags |= TF_ACKNOW; 1354 todrop = tlen; 1355 tcpstat.tcps_rcvdupbyte += todrop; 1356 tcpstat.tcps_rcvduppack++; 1357 } else { 1358 tcpstat.tcps_rcvpartduppack++; 1359 tcpstat.tcps_rcvpartdupbyte += todrop; 1360 } 1361 hdroptlen += todrop; /*drop from head afterwards*/ 1362 th->th_seq += todrop; 1363 tlen -= todrop; 1364 if (th->th_urp > todrop) 1365 th->th_urp -= todrop; 1366 else { 1367 tiflags &= ~TH_URG; 1368 th->th_urp = 0; 1369 } 1370 } 1371 1372 /* 1373 * If new data are received on a connection after the 1374 * user processes are gone, then RST the other end. 1375 */ 1376 if ((so->so_state & SS_NOFDREF) && 1377 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1378 tp = tcp_close(tp); 1379 tcpstat.tcps_rcvafterclose++; 1380 goto dropwithreset; 1381 } 1382 1383 /* 1384 * If segment ends after window, drop trailing data 1385 * (and PUSH and FIN); if nothing left, just ACK. 1386 */ 1387 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1388 if (todrop > 0) { 1389 tcpstat.tcps_rcvpackafterwin++; 1390 if (todrop >= tlen) { 1391 tcpstat.tcps_rcvbyteafterwin += tlen; 1392 /* 1393 * If a new connection request is received 1394 * while in TIME_WAIT, drop the old connection 1395 * and start over if the sequence numbers 1396 * are above the previous ones. 1397 */ 1398 if (tiflags & TH_SYN && 1399 tp->t_state == TCPS_TIME_WAIT && 1400 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1401 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1402 tp->snd_nxt); 1403 tp = tcp_close(tp); 1404 goto findpcb; 1405 } 1406 /* 1407 * If window is closed can only take segments at 1408 * window edge, and have to drop data and PUSH from 1409 * incoming segments. Continue processing, but 1410 * remember to ack. Otherwise, drop segment 1411 * and ack. 1412 */ 1413 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1414 tp->t_flags |= TF_ACKNOW; 1415 tcpstat.tcps_rcvwinprobe++; 1416 } else 1417 goto dropafterack; 1418 } else 1419 tcpstat.tcps_rcvbyteafterwin += todrop; 1420 m_adj(m, -todrop); 1421 tlen -= todrop; 1422 tiflags &= ~(TH_PUSH|TH_FIN); 1423 } 1424 1425 /* 1426 * If last ACK falls within this segment's sequence numbers, 1427 * and the timestamp is newer, record it. 1428 */ 1429 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1430 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1431 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1432 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1433 tp->ts_recent_age = tcp_now; 1434 tp->ts_recent = opti.ts_val; 1435 } 1436 1437 /* 1438 * If the RST bit is set examine the state: 1439 * SYN_RECEIVED STATE: 1440 * If passive open, return to LISTEN state. 1441 * If active open, inform user that connection was refused. 1442 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1443 * Inform user that connection was reset, and close tcb. 1444 * CLOSING, LAST_ACK, TIME_WAIT STATES 1445 * Close the tcb. 1446 */ 1447 if (tiflags&TH_RST) switch (tp->t_state) { 1448 1449 case TCPS_SYN_RECEIVED: 1450 so->so_error = ECONNREFUSED; 1451 goto close; 1452 1453 case TCPS_ESTABLISHED: 1454 case TCPS_FIN_WAIT_1: 1455 case TCPS_FIN_WAIT_2: 1456 case TCPS_CLOSE_WAIT: 1457 so->so_error = ECONNRESET; 1458 close: 1459 tp->t_state = TCPS_CLOSED; 1460 tcpstat.tcps_drops++; 1461 tp = tcp_close(tp); 1462 goto drop; 1463 1464 case TCPS_CLOSING: 1465 case TCPS_LAST_ACK: 1466 case TCPS_TIME_WAIT: 1467 tp = tcp_close(tp); 1468 goto drop; 1469 } 1470 1471 /* 1472 * If a SYN is in the window, then this is an 1473 * error and we send an RST and drop the connection. 1474 */ 1475 if (tiflags & TH_SYN) { 1476 tp = tcp_drop(tp, ECONNRESET); 1477 goto dropwithreset; 1478 } 1479 1480 /* 1481 * If the ACK bit is off we drop the segment and return. 1482 */ 1483 if ((tiflags & TH_ACK) == 0) { 1484 if (tp->t_flags & TF_ACKNOW) 1485 goto dropafterack; 1486 else 1487 goto drop; 1488 } 1489 1490 /* 1491 * Ack processing. 1492 */ 1493 switch (tp->t_state) { 1494 1495 /* 1496 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1497 * ESTABLISHED state and continue processing, otherwise 1498 * send an RST. 1499 */ 1500 case TCPS_SYN_RECEIVED: 1501 if (SEQ_GT(tp->snd_una, th->th_ack) || 1502 SEQ_GT(th->th_ack, tp->snd_max)) 1503 goto dropwithreset; 1504 tcpstat.tcps_connects++; 1505 soisconnected(so); 1506 tcp_established(tp); 1507 /* Do window scaling? */ 1508 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1509 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1510 tp->snd_scale = tp->requested_s_scale; 1511 tp->rcv_scale = tp->request_r_scale; 1512 } 1513 TCP_REASS_LOCK(tp); 1514 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1515 TCP_REASS_UNLOCK(tp); 1516 tp->snd_wl1 = th->th_seq - 1; 1517 /* fall into ... */ 1518 1519 /* 1520 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1521 * ACKs. If the ack is in the range 1522 * tp->snd_una < th->th_ack <= tp->snd_max 1523 * then advance tp->snd_una to th->th_ack and drop 1524 * data from the retransmission queue. If this ACK reflects 1525 * more up to date window information we update our window information. 1526 */ 1527 case TCPS_ESTABLISHED: 1528 case TCPS_FIN_WAIT_1: 1529 case TCPS_FIN_WAIT_2: 1530 case TCPS_CLOSE_WAIT: 1531 case TCPS_CLOSING: 1532 case TCPS_LAST_ACK: 1533 case TCPS_TIME_WAIT: 1534 1535 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1536 if (tlen == 0 && tiwin == tp->snd_wnd) { 1537 tcpstat.tcps_rcvdupack++; 1538 /* 1539 * If we have outstanding data (other than 1540 * a window probe), this is a completely 1541 * duplicate ack (ie, window info didn't 1542 * change), the ack is the biggest we've 1543 * seen and we've seen exactly our rexmt 1544 * threshhold of them, assume a packet 1545 * has been dropped and retransmit it. 1546 * Kludge snd_nxt & the congestion 1547 * window so we send only this one 1548 * packet. 1549 * 1550 * We know we're losing at the current 1551 * window size so do congestion avoidance 1552 * (set ssthresh to half the current window 1553 * and pull our congestion window back to 1554 * the new ssthresh). 1555 * 1556 * Dup acks mean that packets have left the 1557 * network (they're now cached at the receiver) 1558 * so bump cwnd by the amount in the receiver 1559 * to keep a constant cwnd packets in the 1560 * network. 1561 */ 1562 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1563 th->th_ack != tp->snd_una) 1564 tp->t_dupacks = 0; 1565 else if (++tp->t_dupacks == tcprexmtthresh) { 1566 tcp_seq onxt = tp->snd_nxt; 1567 u_int win = 1568 min(tp->snd_wnd, tp->snd_cwnd) / 1569 2 / tp->t_segsz; 1570 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1571 tp->snd_recover)) { 1572 /* 1573 * False fast retransmit after 1574 * timeout. Do not cut window. 1575 */ 1576 tp->snd_cwnd += tp->t_segsz; 1577 tp->t_dupacks = 0; 1578 (void) tcp_output(tp); 1579 goto drop; 1580 } 1581 1582 if (win < 2) 1583 win = 2; 1584 tp->snd_ssthresh = win * tp->t_segsz; 1585 tp->snd_recover = tp->snd_max; 1586 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1587 tp->t_rtt = 0; 1588 tp->snd_nxt = th->th_ack; 1589 tp->snd_cwnd = tp->t_segsz; 1590 (void) tcp_output(tp); 1591 tp->snd_cwnd = tp->snd_ssthresh + 1592 tp->t_segsz * tp->t_dupacks; 1593 if (SEQ_GT(onxt, tp->snd_nxt)) 1594 tp->snd_nxt = onxt; 1595 goto drop; 1596 } else if (tp->t_dupacks > tcprexmtthresh) { 1597 tp->snd_cwnd += tp->t_segsz; 1598 (void) tcp_output(tp); 1599 goto drop; 1600 } 1601 } else 1602 tp->t_dupacks = 0; 1603 break; 1604 } 1605 /* 1606 * If the congestion window was inflated to account 1607 * for the other side's cached packets, retract it. 1608 */ 1609 if (tcp_do_newreno == 0) { 1610 if (tp->t_dupacks >= tcprexmtthresh && 1611 tp->snd_cwnd > tp->snd_ssthresh) 1612 tp->snd_cwnd = tp->snd_ssthresh; 1613 tp->t_dupacks = 0; 1614 } else if (tp->t_dupacks >= tcprexmtthresh && 1615 tcp_newreno(tp, th) == 0) { 1616 tp->snd_cwnd = tp->snd_ssthresh; 1617 /* 1618 * Window inflation should have left us with approx. 1619 * snd_ssthresh outstanding data. But in case we 1620 * would be inclined to send a burst, better to do 1621 * it via the slow start mechanism. 1622 */ 1623 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 1624 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 1625 + tp->t_segsz; 1626 tp->t_dupacks = 0; 1627 } 1628 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1629 tcpstat.tcps_rcvacktoomuch++; 1630 goto dropafterack; 1631 } 1632 acked = th->th_ack - tp->snd_una; 1633 tcpstat.tcps_rcvackpack++; 1634 tcpstat.tcps_rcvackbyte += acked; 1635 1636 /* 1637 * If we have a timestamp reply, update smoothed 1638 * round trip time. If no timestamp is present but 1639 * transmit timer is running and timed sequence 1640 * number was acked, update smoothed round trip time. 1641 * Since we now have an rtt measurement, cancel the 1642 * timer backoff (cf., Phil Karn's retransmit alg.). 1643 * Recompute the initial retransmit timer. 1644 */ 1645 if (opti.ts_present && opti.ts_ecr) 1646 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1647 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1648 tcp_xmit_timer(tp,tp->t_rtt); 1649 1650 /* 1651 * If all outstanding data is acked, stop retransmit 1652 * timer and remember to restart (more output or persist). 1653 * If there is more data to be acked, restart retransmit 1654 * timer, using current (possibly backed-off) value. 1655 */ 1656 if (th->th_ack == tp->snd_max) { 1657 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1658 needoutput = 1; 1659 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1660 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1661 /* 1662 * When new data is acked, open the congestion window. 1663 * If the window gives us less than ssthresh packets 1664 * in flight, open exponentially (segsz per packet). 1665 * Otherwise open linearly: segsz per window 1666 * (segsz^2 / cwnd per packet), plus a constant 1667 * fraction of a packet (segsz/8) to help larger windows 1668 * open quickly enough. 1669 */ 1670 { 1671 register u_int cw = tp->snd_cwnd; 1672 register u_int incr = tp->t_segsz; 1673 1674 if (cw > tp->snd_ssthresh) 1675 incr = incr * incr / cw; 1676 if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover)) 1677 tp->snd_cwnd = min(cw + incr, 1678 TCP_MAXWIN << tp->snd_scale); 1679 } 1680 if (acked > so->so_snd.sb_cc) { 1681 tp->snd_wnd -= so->so_snd.sb_cc; 1682 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1683 ourfinisacked = 1; 1684 } else { 1685 sbdrop(&so->so_snd, acked); 1686 tp->snd_wnd -= acked; 1687 ourfinisacked = 0; 1688 } 1689 sowwakeup(so); 1690 /* 1691 * We want snd_recover to track snd_una to 1692 * avoid sequence wraparound problems for 1693 * very large transfers. 1694 */ 1695 tp->snd_una = tp->snd_recover = th->th_ack; 1696 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1697 tp->snd_nxt = tp->snd_una; 1698 1699 switch (tp->t_state) { 1700 1701 /* 1702 * In FIN_WAIT_1 STATE in addition to the processing 1703 * for the ESTABLISHED state if our FIN is now acknowledged 1704 * then enter FIN_WAIT_2. 1705 */ 1706 case TCPS_FIN_WAIT_1: 1707 if (ourfinisacked) { 1708 /* 1709 * If we can't receive any more 1710 * data, then closing user can proceed. 1711 * Starting the timer is contrary to the 1712 * specification, but if we don't get a FIN 1713 * we'll hang forever. 1714 */ 1715 if (so->so_state & SS_CANTRCVMORE) { 1716 soisdisconnected(so); 1717 if (tcp_maxidle > 0) 1718 TCP_TIMER_ARM(tp, TCPT_2MSL, 1719 tcp_maxidle); 1720 } 1721 tp->t_state = TCPS_FIN_WAIT_2; 1722 } 1723 break; 1724 1725 /* 1726 * In CLOSING STATE in addition to the processing for 1727 * the ESTABLISHED state if the ACK acknowledges our FIN 1728 * then enter the TIME-WAIT state, otherwise ignore 1729 * the segment. 1730 */ 1731 case TCPS_CLOSING: 1732 if (ourfinisacked) { 1733 tp->t_state = TCPS_TIME_WAIT; 1734 tcp_canceltimers(tp); 1735 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1736 soisdisconnected(so); 1737 } 1738 break; 1739 1740 /* 1741 * In LAST_ACK, we may still be waiting for data to drain 1742 * and/or to be acked, as well as for the ack of our FIN. 1743 * If our FIN is now acknowledged, delete the TCB, 1744 * enter the closed state and return. 1745 */ 1746 case TCPS_LAST_ACK: 1747 if (ourfinisacked) { 1748 tp = tcp_close(tp); 1749 goto drop; 1750 } 1751 break; 1752 1753 /* 1754 * In TIME_WAIT state the only thing that should arrive 1755 * is a retransmission of the remote FIN. Acknowledge 1756 * it and restart the finack timer. 1757 */ 1758 case TCPS_TIME_WAIT: 1759 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1760 goto dropafterack; 1761 } 1762 } 1763 1764 step6: 1765 /* 1766 * Update window information. 1767 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1768 */ 1769 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1770 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1771 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1772 /* keep track of pure window updates */ 1773 if (tlen == 0 && 1774 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1775 tcpstat.tcps_rcvwinupd++; 1776 tp->snd_wnd = tiwin; 1777 tp->snd_wl1 = th->th_seq; 1778 tp->snd_wl2 = th->th_ack; 1779 if (tp->snd_wnd > tp->max_sndwnd) 1780 tp->max_sndwnd = tp->snd_wnd; 1781 needoutput = 1; 1782 } 1783 1784 /* 1785 * Process segments with URG. 1786 */ 1787 if ((tiflags & TH_URG) && th->th_urp && 1788 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1789 /* 1790 * This is a kludge, but if we receive and accept 1791 * random urgent pointers, we'll crash in 1792 * soreceive. It's hard to imagine someone 1793 * actually wanting to send this much urgent data. 1794 */ 1795 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1796 th->th_urp = 0; /* XXX */ 1797 tiflags &= ~TH_URG; /* XXX */ 1798 goto dodata; /* XXX */ 1799 } 1800 /* 1801 * If this segment advances the known urgent pointer, 1802 * then mark the data stream. This should not happen 1803 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1804 * a FIN has been received from the remote side. 1805 * In these states we ignore the URG. 1806 * 1807 * According to RFC961 (Assigned Protocols), 1808 * the urgent pointer points to the last octet 1809 * of urgent data. We continue, however, 1810 * to consider it to indicate the first octet 1811 * of data past the urgent section as the original 1812 * spec states (in one of two places). 1813 */ 1814 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1815 tp->rcv_up = th->th_seq + th->th_urp; 1816 so->so_oobmark = so->so_rcv.sb_cc + 1817 (tp->rcv_up - tp->rcv_nxt) - 1; 1818 if (so->so_oobmark == 0) 1819 so->so_state |= SS_RCVATMARK; 1820 sohasoutofband(so); 1821 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1822 } 1823 /* 1824 * Remove out of band data so doesn't get presented to user. 1825 * This can happen independent of advancing the URG pointer, 1826 * but if two URG's are pending at once, some out-of-band 1827 * data may creep in... ick. 1828 */ 1829 if (th->th_urp <= (u_int16_t) tlen 1830 #ifdef SO_OOBINLINE 1831 && (so->so_options & SO_OOBINLINE) == 0 1832 #endif 1833 ) 1834 tcp_pulloutofband(so, th, m, hdroptlen); 1835 } else 1836 /* 1837 * If no out of band data is expected, 1838 * pull receive urgent pointer along 1839 * with the receive window. 1840 */ 1841 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1842 tp->rcv_up = tp->rcv_nxt; 1843 dodata: /* XXX */ 1844 1845 /* 1846 * Process the segment text, merging it into the TCP sequencing queue, 1847 * and arranging for acknowledgement of receipt if necessary. 1848 * This process logically involves adjusting tp->rcv_wnd as data 1849 * is presented to the user (this happens in tcp_usrreq.c, 1850 * case PRU_RCVD). If a FIN has already been received on this 1851 * connection then we just ignore the text. 1852 */ 1853 if ((tlen || (tiflags & TH_FIN)) && 1854 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1855 /* 1856 * Insert segment ti into reassembly queue of tcp with 1857 * control block tp. Return TH_FIN if reassembly now includes 1858 * a segment with FIN. The macro form does the common case 1859 * inline (segment is the next to be received on an 1860 * established connection, and the queue is empty), 1861 * avoiding linkage into and removal from the queue and 1862 * repetition of various conversions. 1863 * Set DELACK for segments received in order, but ack 1864 * immediately when segments are out of order 1865 * (so fast retransmit can work). 1866 */ 1867 /* NOTE: this was TCP_REASS() macro, but used only once */ 1868 TCP_REASS_LOCK(tp); 1869 if (th->th_seq == tp->rcv_nxt && 1870 tp->segq.lh_first == NULL && 1871 tp->t_state == TCPS_ESTABLISHED) { 1872 TCP_SETUP_ACK(tp, th); 1873 tp->rcv_nxt += tlen; 1874 tiflags = th->th_flags & TH_FIN; 1875 tcpstat.tcps_rcvpack++; 1876 tcpstat.tcps_rcvbyte += tlen; 1877 m_adj(m, hdroptlen); 1878 sbappend(&(so)->so_rcv, m); 1879 sorwakeup(so); 1880 } else { 1881 m_adj(m, hdroptlen); 1882 tiflags = tcp_reass(tp, th, m, &tlen); 1883 tp->t_flags |= TF_ACKNOW; 1884 } 1885 TCP_REASS_UNLOCK(tp); 1886 1887 /* 1888 * Note the amount of data that peer has sent into 1889 * our window, in order to estimate the sender's 1890 * buffer size. 1891 */ 1892 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1893 } else { 1894 m_freem(m); 1895 m = NULL; 1896 tiflags &= ~TH_FIN; 1897 } 1898 1899 /* 1900 * If FIN is received ACK the FIN and let the user know 1901 * that the connection is closing. Ignore a FIN received before 1902 * the connection is fully established. 1903 */ 1904 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 1905 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1906 socantrcvmore(so); 1907 tp->t_flags |= TF_ACKNOW; 1908 tp->rcv_nxt++; 1909 } 1910 switch (tp->t_state) { 1911 1912 /* 1913 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 1914 */ 1915 case TCPS_ESTABLISHED: 1916 tp->t_state = TCPS_CLOSE_WAIT; 1917 break; 1918 1919 /* 1920 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1921 * enter the CLOSING state. 1922 */ 1923 case TCPS_FIN_WAIT_1: 1924 tp->t_state = TCPS_CLOSING; 1925 break; 1926 1927 /* 1928 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1929 * starting the time-wait timer, turning off the other 1930 * standard timers. 1931 */ 1932 case TCPS_FIN_WAIT_2: 1933 tp->t_state = TCPS_TIME_WAIT; 1934 tcp_canceltimers(tp); 1935 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1936 soisdisconnected(so); 1937 break; 1938 1939 /* 1940 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1941 */ 1942 case TCPS_TIME_WAIT: 1943 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1944 break; 1945 } 1946 } 1947 if (so->so_options & SO_DEBUG) { 1948 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 1949 } 1950 1951 /* 1952 * Return any desired output. 1953 */ 1954 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1955 (void) tcp_output(tp); 1956 if (tcp_saveti) 1957 m_freem(tcp_saveti); 1958 return; 1959 1960 badsyn: 1961 /* 1962 * Received a bad SYN. Increment counters and dropwithreset. 1963 */ 1964 tcpstat.tcps_badsyn++; 1965 tp = NULL; 1966 goto dropwithreset; 1967 1968 dropafterack: 1969 /* 1970 * Generate an ACK dropping incoming segment if it occupies 1971 * sequence space, where the ACK reflects our state. 1972 */ 1973 if (tiflags & TH_RST) 1974 goto drop; 1975 m_freem(m); 1976 tp->t_flags |= TF_ACKNOW; 1977 (void) tcp_output(tp); 1978 if (tcp_saveti) 1979 m_freem(tcp_saveti); 1980 return; 1981 1982 dropwithreset: 1983 /* 1984 * Generate a RST, dropping incoming segment. 1985 * Make ACK acceptable to originator of segment. 1986 * Don't bother to respond if destination was broadcast/multicast. 1987 */ 1988 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 1989 goto drop; 1990 if (ip && IN_MULTICAST(ip->ip_dst.s_addr)) 1991 goto drop; 1992 #ifdef INET6 1993 if (m->m_flags & M_ANYCAST6) 1994 goto drop; 1995 else if (ip6 && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1996 goto drop; 1997 #endif 1998 { 1999 /* 2000 * need to recover version # field, which was overwritten on 2001 * ip_cksum computation. 2002 */ 2003 struct ip *sip; 2004 sip = mtod(m, struct ip *); 2005 switch (af) { 2006 case AF_INET: 2007 sip->ip_v = 4; 2008 break; 2009 #ifdef INET6 2010 case AF_INET6: 2011 sip->ip_v = 6; 2012 break; 2013 #endif 2014 } 2015 } 2016 if (tiflags & TH_ACK) 2017 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2018 else { 2019 if (tiflags & TH_SYN) 2020 tlen++; 2021 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2022 TH_RST|TH_ACK); 2023 } 2024 if (tcp_saveti) 2025 m_freem(tcp_saveti); 2026 return; 2027 2028 drop: 2029 /* 2030 * Drop space held by incoming segment and return. 2031 */ 2032 if (tp) { 2033 if (tp->t_inpcb) 2034 so = tp->t_inpcb->inp_socket; 2035 #ifdef INET6 2036 else if (tp->t_in6pcb) 2037 so = tp->t_in6pcb->in6p_socket; 2038 #endif 2039 else 2040 so = NULL; 2041 if (so && (so->so_options & SO_DEBUG) != 0) 2042 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2043 } 2044 if (tcp_saveti) 2045 m_freem(tcp_saveti); 2046 m_freem(m); 2047 return; 2048 } 2049 2050 void 2051 tcp_dooptions(tp, cp, cnt, th, oi) 2052 struct tcpcb *tp; 2053 u_char *cp; 2054 int cnt; 2055 struct tcphdr *th; 2056 struct tcp_opt_info *oi; 2057 { 2058 u_int16_t mss; 2059 int opt, optlen; 2060 2061 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2062 opt = cp[0]; 2063 if (opt == TCPOPT_EOL) 2064 break; 2065 if (opt == TCPOPT_NOP) 2066 optlen = 1; 2067 else { 2068 optlen = cp[1]; 2069 if (optlen <= 0) 2070 break; 2071 } 2072 switch (opt) { 2073 2074 default: 2075 continue; 2076 2077 case TCPOPT_MAXSEG: 2078 if (optlen != TCPOLEN_MAXSEG) 2079 continue; 2080 if (!(th->th_flags & TH_SYN)) 2081 continue; 2082 bcopy(cp + 2, &mss, sizeof(mss)); 2083 oi->maxseg = ntohs(mss); 2084 break; 2085 2086 case TCPOPT_WINDOW: 2087 if (optlen != TCPOLEN_WINDOW) 2088 continue; 2089 if (!(th->th_flags & TH_SYN)) 2090 continue; 2091 tp->t_flags |= TF_RCVD_SCALE; 2092 tp->requested_s_scale = cp[2]; 2093 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2094 #if 0 /*XXX*/ 2095 char *p; 2096 2097 if (ip) 2098 p = ntohl(ip->ip_src); 2099 #ifdef INET6 2100 else if (ip6) 2101 p = ip6_sprintf(&ip6->ip6_src); 2102 #endif 2103 else 2104 p = "(unknown)"; 2105 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2106 "assuming %d\n", 2107 tp->requested_s_scale, p, 2108 TCP_MAX_WINSHIFT); 2109 #else 2110 log(LOG_ERR, "TCP: invalid wscale %d, " 2111 "assuming %d\n", 2112 tp->requested_s_scale, 2113 TCP_MAX_WINSHIFT); 2114 #endif 2115 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2116 } 2117 break; 2118 2119 case TCPOPT_TIMESTAMP: 2120 if (optlen != TCPOLEN_TIMESTAMP) 2121 continue; 2122 oi->ts_present = 1; 2123 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2124 NTOHL(oi->ts_val); 2125 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2126 NTOHL(oi->ts_ecr); 2127 2128 /* 2129 * A timestamp received in a SYN makes 2130 * it ok to send timestamp requests and replies. 2131 */ 2132 if (th->th_flags & TH_SYN) { 2133 tp->t_flags |= TF_RCVD_TSTMP; 2134 tp->ts_recent = oi->ts_val; 2135 tp->ts_recent_age = tcp_now; 2136 } 2137 break; 2138 case TCPOPT_SACK_PERMITTED: 2139 if (optlen != TCPOLEN_SACK_PERMITTED) 2140 continue; 2141 if (!(th->th_flags & TH_SYN)) 2142 continue; 2143 tp->t_flags &= ~TF_CANT_TXSACK; 2144 break; 2145 2146 case TCPOPT_SACK: 2147 if (tp->t_flags & TF_IGNR_RXSACK) 2148 continue; 2149 if (optlen % 8 != 2 || optlen < 10) 2150 continue; 2151 cp += 2; 2152 optlen -= 2; 2153 for (; optlen > 0; cp -= 8, optlen -= 8) { 2154 tcp_seq lwe, rwe; 2155 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 2156 NTOHL(lwe); 2157 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 2158 NTOHL(rwe); 2159 /* tcp_mark_sacked(tp, lwe, rwe); */ 2160 } 2161 break; 2162 } 2163 } 2164 } 2165 2166 /* 2167 * Pull out of band byte out of a segment so 2168 * it doesn't appear in the user's data queue. 2169 * It is still reflected in the segment length for 2170 * sequencing purposes. 2171 */ 2172 void 2173 tcp_pulloutofband(so, th, m, off) 2174 struct socket *so; 2175 struct tcphdr *th; 2176 register struct mbuf *m; 2177 int off; 2178 { 2179 int cnt = off + th->th_urp - 1; 2180 2181 while (cnt >= 0) { 2182 if (m->m_len > cnt) { 2183 char *cp = mtod(m, caddr_t) + cnt; 2184 struct tcpcb *tp = sototcpcb(so); 2185 2186 tp->t_iobc = *cp; 2187 tp->t_oobflags |= TCPOOB_HAVEDATA; 2188 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2189 m->m_len--; 2190 return; 2191 } 2192 cnt -= m->m_len; 2193 m = m->m_next; 2194 if (m == 0) 2195 break; 2196 } 2197 panic("tcp_pulloutofband"); 2198 } 2199 2200 /* 2201 * Collect new round-trip time estimate 2202 * and update averages and current timeout. 2203 */ 2204 void 2205 tcp_xmit_timer(tp, rtt) 2206 register struct tcpcb *tp; 2207 short rtt; 2208 { 2209 register short delta; 2210 short rttmin; 2211 2212 tcpstat.tcps_rttupdated++; 2213 --rtt; 2214 if (tp->t_srtt != 0) { 2215 /* 2216 * srtt is stored as fixed point with 3 bits after the 2217 * binary point (i.e., scaled by 8). The following magic 2218 * is equivalent to the smoothing algorithm in rfc793 with 2219 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2220 * point). Adjust rtt to origin 0. 2221 */ 2222 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2223 if ((tp->t_srtt += delta) <= 0) 2224 tp->t_srtt = 1 << 2; 2225 /* 2226 * We accumulate a smoothed rtt variance (actually, a 2227 * smoothed mean difference), then set the retransmit 2228 * timer to smoothed rtt + 4 times the smoothed variance. 2229 * rttvar is stored as fixed point with 2 bits after the 2230 * binary point (scaled by 4). The following is 2231 * equivalent to rfc793 smoothing with an alpha of .75 2232 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2233 * rfc793's wired-in beta. 2234 */ 2235 if (delta < 0) 2236 delta = -delta; 2237 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2238 if ((tp->t_rttvar += delta) <= 0) 2239 tp->t_rttvar = 1 << 2; 2240 } else { 2241 /* 2242 * No rtt measurement yet - use the unsmoothed rtt. 2243 * Set the variance to half the rtt (so our first 2244 * retransmit happens at 3*rtt). 2245 */ 2246 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2247 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2248 } 2249 tp->t_rtt = 0; 2250 tp->t_rxtshift = 0; 2251 2252 /* 2253 * the retransmit should happen at rtt + 4 * rttvar. 2254 * Because of the way we do the smoothing, srtt and rttvar 2255 * will each average +1/2 tick of bias. When we compute 2256 * the retransmit timer, we want 1/2 tick of rounding and 2257 * 1 extra tick because of +-1/2 tick uncertainty in the 2258 * firing of the timer. The bias will give us exactly the 2259 * 1.5 tick we need. But, because the bias is 2260 * statistical, we have to test that we don't drop below 2261 * the minimum feasible timer (which is 2 ticks). 2262 */ 2263 if (tp->t_rttmin > rtt + 2) 2264 rttmin = tp->t_rttmin; 2265 else 2266 rttmin = rtt + 2; 2267 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2268 2269 /* 2270 * We received an ack for a packet that wasn't retransmitted; 2271 * it is probably safe to discard any error indications we've 2272 * received recently. This isn't quite right, but close enough 2273 * for now (a route might have failed after we sent a segment, 2274 * and the return path might not be symmetrical). 2275 */ 2276 tp->t_softerror = 0; 2277 } 2278 2279 /* 2280 * Checks for partial ack. If partial ack arrives, force the retransmission 2281 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2282 * 1. By setting snd_nxt to th_ack, this forces retransmission timer to 2283 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2284 */ 2285 int 2286 tcp_newreno(tp, th) 2287 struct tcpcb *tp; 2288 struct tcphdr *th; 2289 { 2290 tcp_seq onxt = tp->snd_nxt; 2291 u_long ocwnd = tp->snd_cwnd; 2292 2293 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2294 /* 2295 * snd_una has not yet been updated and the socket's send 2296 * buffer has not yet drained off the ACK'd data, so we 2297 * have to leave snd_una as it was to get the correct data 2298 * offset in tcp_output(). 2299 */ 2300 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2301 tp->t_rtt = 0; 2302 tp->snd_nxt = th->th_ack; 2303 /* 2304 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 2305 * is not yet updated when we're called. 2306 */ 2307 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 2308 (void) tcp_output(tp); 2309 tp->snd_cwnd = ocwnd; 2310 if (SEQ_GT(onxt, tp->snd_nxt)) 2311 tp->snd_nxt = onxt; 2312 /* 2313 * Partial window deflation. Relies on fact that tp->snd_una 2314 * not updated yet. 2315 */ 2316 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 2317 return 1; 2318 } 2319 return 0; 2320 } 2321 2322 2323 /* 2324 * TCP compressed state engine. Currently used to hold compressed 2325 * state for SYN_RECEIVED. 2326 */ 2327 2328 u_long syn_cache_count; 2329 u_int32_t syn_hash1, syn_hash2; 2330 2331 #define SYN_HASH(sa, sp, dp) \ 2332 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 2333 ((u_int32_t)(sp)))^syn_hash2))) 2334 #ifndef INET6 2335 #define SYN_HASHALL(hash, src, dst) \ 2336 do { \ 2337 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2338 ((struct sockaddr_in *)(src))->sin_port, \ 2339 ((struct sockaddr_in *)(dst))->sin_port); \ 2340 } while (0) 2341 #else 2342 #define SYN_HASH6(sa, sp, dp) \ 2343 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 2344 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 2345 & 0x7fffffff) 2346 2347 #define SYN_HASHALL(hash, src, dst) \ 2348 do { \ 2349 switch ((src)->sa_family) { \ 2350 case AF_INET: \ 2351 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 2352 ((struct sockaddr_in *)(src))->sin_port, \ 2353 ((struct sockaddr_in *)(dst))->sin_port); \ 2354 break; \ 2355 case AF_INET6: \ 2356 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 2357 ((struct sockaddr_in6 *)(src))->sin6_port, \ 2358 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 2359 break; \ 2360 default: \ 2361 hash = 0; \ 2362 } \ 2363 } while (0) 2364 #endif /* INET6 */ 2365 2366 #define SYN_CACHE_RM(sc) \ 2367 do { \ 2368 LIST_REMOVE((sc), sc_bucketq); \ 2369 (sc)->sc_tp = NULL; \ 2370 LIST_REMOVE((sc), sc_tpq); \ 2371 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 2372 TAILQ_REMOVE(&tcp_syn_cache_timeq[(sc)->sc_rxtshift], (sc), sc_timeq); \ 2373 syn_cache_count--; \ 2374 } while (0) 2375 2376 #define SYN_CACHE_PUT(sc) \ 2377 do { \ 2378 if ((sc)->sc_ipopts) \ 2379 (void) m_free((sc)->sc_ipopts); \ 2380 if ((sc)->sc_route4.ro_rt != NULL) \ 2381 RTFREE((sc)->sc_route4.ro_rt); \ 2382 pool_put(&syn_cache_pool, (sc)); \ 2383 } while (0) 2384 2385 struct pool syn_cache_pool; 2386 2387 /* 2388 * We don't estimate RTT with SYNs, so each packet starts with the default 2389 * RTT and each timer queue has a fixed timeout value. This allows us to 2390 * optimize the timer queues somewhat. 2391 */ 2392 #define SYN_CACHE_TIMER_ARM(sc) \ 2393 do { \ 2394 TCPT_RANGESET((sc)->sc_rxtcur, \ 2395 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 2396 TCPTV_REXMTMAX); \ 2397 PRT_SLOW_ARM((sc)->sc_rexmt, (sc)->sc_rxtcur); \ 2398 } while (0) 2399 2400 TAILQ_HEAD(, syn_cache) tcp_syn_cache_timeq[TCP_MAXRXTSHIFT + 1]; 2401 2402 void 2403 syn_cache_init() 2404 { 2405 int i; 2406 2407 /* Initialize the hash buckets. */ 2408 for (i = 0; i < tcp_syn_cache_size; i++) 2409 LIST_INIT(&tcp_syn_cache[i].sch_bucket); 2410 2411 /* Initialize the timer queues. */ 2412 for (i = 0; i <= TCP_MAXRXTSHIFT; i++) 2413 TAILQ_INIT(&tcp_syn_cache_timeq[i]); 2414 2415 /* Initialize the syn cache pool. */ 2416 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 2417 "synpl", 0, NULL, NULL, M_PCB); 2418 } 2419 2420 void 2421 syn_cache_insert(sc, tp) 2422 struct syn_cache *sc; 2423 struct tcpcb *tp; 2424 { 2425 struct syn_cache_head *scp; 2426 struct syn_cache *sc2; 2427 int s, i; 2428 2429 /* 2430 * If there are no entries in the hash table, reinitialize 2431 * the hash secrets. 2432 */ 2433 if (syn_cache_count == 0) { 2434 struct timeval tv; 2435 microtime(&tv); 2436 syn_hash1 = random() ^ (u_long)≻ 2437 syn_hash2 = random() ^ tv.tv_usec; 2438 } 2439 2440 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 2441 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 2442 scp = &tcp_syn_cache[sc->sc_bucketidx]; 2443 2444 /* 2445 * Make sure that we don't overflow the per-bucket 2446 * limit or the total cache size limit. 2447 */ 2448 s = splsoftnet(); 2449 if (scp->sch_length >= tcp_syn_bucket_limit) { 2450 tcpstat.tcps_sc_bucketoverflow++; 2451 /* 2452 * The bucket is full. Toss the oldest element in the 2453 * bucket. This will be the entry with our bucket 2454 * index closest to the front of the timer queue with 2455 * the largest timeout value. 2456 * 2457 * Note: This timer queue traversal may be expensive, so 2458 * we hope that this doesn't happen very often. It is 2459 * much more likely that we'll overflow the entire 2460 * cache, which is much easier to handle; see below. 2461 */ 2462 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2463 for (sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2464 sc2 != NULL; 2465 sc2 = TAILQ_NEXT(sc2, sc_timeq)) { 2466 if (sc2->sc_bucketidx == sc->sc_bucketidx) { 2467 SYN_CACHE_RM(sc2); 2468 SYN_CACHE_PUT(sc2); 2469 goto insert; /* 2 level break */ 2470 } 2471 } 2472 } 2473 #ifdef DIAGNOSTIC 2474 /* 2475 * This should never happen; we should always find an 2476 * entry in our bucket. 2477 */ 2478 panic("syn_cache_insert: bucketoverflow: impossible"); 2479 #endif 2480 } else if (syn_cache_count >= tcp_syn_cache_limit) { 2481 tcpstat.tcps_sc_overflowed++; 2482 /* 2483 * The cache is full. Toss the oldest entry in the 2484 * entire cache. This is the front entry in the 2485 * first non-empty timer queue with the largest 2486 * timeout value. 2487 */ 2488 for (i = TCP_MAXRXTSHIFT; i >= 0; i--) { 2489 sc2 = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2490 if (sc2 == NULL) 2491 continue; 2492 SYN_CACHE_RM(sc2); 2493 SYN_CACHE_PUT(sc2); 2494 goto insert; /* symmetry with above */ 2495 } 2496 #ifdef DIAGNOSTIC 2497 /* 2498 * This should never happen; we should always find an 2499 * entry in the cache. 2500 */ 2501 panic("syn_cache_insert: cache overflow: impossible"); 2502 #endif 2503 } 2504 2505 insert: 2506 /* 2507 * Initialize the entry's timer. 2508 */ 2509 sc->sc_rxttot = 0; 2510 sc->sc_rxtshift = 0; 2511 SYN_CACHE_TIMER_ARM(sc); 2512 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], sc, sc_timeq); 2513 2514 /* Link it from tcpcb entry */ 2515 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 2516 2517 /* Put it into the bucket. */ 2518 LIST_INSERT_HEAD(&scp->sch_bucket, sc, sc_bucketq); 2519 scp->sch_length++; 2520 syn_cache_count++; 2521 2522 tcpstat.tcps_sc_added++; 2523 splx(s); 2524 } 2525 2526 /* 2527 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 2528 * If we have retransmitted an entry the maximum number of times, expire 2529 * that entry. 2530 */ 2531 void 2532 syn_cache_timer() 2533 { 2534 struct syn_cache *sc, *nsc; 2535 int i, s; 2536 2537 s = splsoftnet(); 2538 2539 /* 2540 * First, get all the entries that need to be retransmitted, or 2541 * must be expired due to exceeding the initial keepalive time. 2542 */ 2543 for (i = 0; i < TCP_MAXRXTSHIFT; i++) { 2544 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[i]); 2545 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2546 sc = nsc) { 2547 nsc = TAILQ_NEXT(sc, sc_timeq); 2548 2549 /* 2550 * Compute the total amount of time this entry has 2551 * been on a queue. If this entry has been on longer 2552 * than the keep alive timer would allow, expire it. 2553 */ 2554 sc->sc_rxttot += sc->sc_rxtcur; 2555 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) { 2556 tcpstat.tcps_sc_timed_out++; 2557 SYN_CACHE_RM(sc); 2558 SYN_CACHE_PUT(sc); 2559 continue; 2560 } 2561 2562 tcpstat.tcps_sc_retransmitted++; 2563 (void) syn_cache_respond(sc, NULL); 2564 2565 /* Advance this entry onto the next timer queue. */ 2566 TAILQ_REMOVE(&tcp_syn_cache_timeq[i], sc, sc_timeq); 2567 sc->sc_rxtshift = i + 1; 2568 SYN_CACHE_TIMER_ARM(sc); 2569 TAILQ_INSERT_TAIL(&tcp_syn_cache_timeq[sc->sc_rxtshift], 2570 sc, sc_timeq); 2571 } 2572 } 2573 2574 /* 2575 * Now get all the entries that are expired due to too many 2576 * retransmissions. 2577 */ 2578 for (sc = TAILQ_FIRST(&tcp_syn_cache_timeq[TCP_MAXRXTSHIFT]); 2579 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_rexmt); 2580 sc = nsc) { 2581 nsc = TAILQ_NEXT(sc, sc_timeq); 2582 tcpstat.tcps_sc_timed_out++; 2583 SYN_CACHE_RM(sc); 2584 SYN_CACHE_PUT(sc); 2585 } 2586 splx(s); 2587 } 2588 2589 /* 2590 * Remove syn cache created by the specified tcb entry, 2591 * because this does not make sense to keep them 2592 * (if there's no tcb entry, syn cache entry will never be used) 2593 */ 2594 void 2595 syn_cache_cleanup(tp) 2596 struct tcpcb *tp; 2597 { 2598 struct syn_cache *sc, *nsc; 2599 int s; 2600 2601 s = splsoftnet(); 2602 2603 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 2604 nsc = LIST_NEXT(sc, sc_tpq); 2605 2606 #ifdef DIAGNOSTIC 2607 if (sc->sc_tp != tp) 2608 panic("invalid sc_tp in syn_cache_cleanup"); 2609 #endif 2610 SYN_CACHE_RM(sc); 2611 SYN_CACHE_PUT(sc); 2612 } 2613 /* just for safety */ 2614 LIST_INIT(&tp->t_sc); 2615 2616 splx(s); 2617 } 2618 2619 /* 2620 * Find an entry in the syn cache. 2621 */ 2622 struct syn_cache * 2623 syn_cache_lookup(src, dst, headp) 2624 struct sockaddr *src; 2625 struct sockaddr *dst; 2626 struct syn_cache_head **headp; 2627 { 2628 struct syn_cache *sc; 2629 struct syn_cache_head *scp; 2630 u_int32_t hash; 2631 int s; 2632 2633 SYN_HASHALL(hash, src, dst); 2634 2635 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 2636 *headp = scp; 2637 s = splsoftnet(); 2638 for (sc = LIST_FIRST(&scp->sch_bucket); sc != NULL; 2639 sc = LIST_NEXT(sc, sc_bucketq)) { 2640 if (sc->sc_hash != hash) 2641 continue; 2642 if (!bcmp(&sc->sc_src, src, src->sa_len) && 2643 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 2644 splx(s); 2645 return (sc); 2646 } 2647 } 2648 splx(s); 2649 return (NULL); 2650 } 2651 2652 /* 2653 * This function gets called when we receive an ACK for a 2654 * socket in the LISTEN state. We look up the connection 2655 * in the syn cache, and if its there, we pull it out of 2656 * the cache and turn it into a full-blown connection in 2657 * the SYN-RECEIVED state. 2658 * 2659 * The return values may not be immediately obvious, and their effects 2660 * can be subtle, so here they are: 2661 * 2662 * NULL SYN was not found in cache; caller should drop the 2663 * packet and send an RST. 2664 * 2665 * -1 We were unable to create the new connection, and are 2666 * aborting it. An ACK,RST is being sent to the peer 2667 * (unless we got screwey sequence numbners; see below), 2668 * because the 3-way handshake has been completed. Caller 2669 * should not free the mbuf, since we may be using it. If 2670 * we are not, we will free it. 2671 * 2672 * Otherwise, the return value is a pointer to the new socket 2673 * associated with the connection. 2674 */ 2675 struct socket * 2676 syn_cache_get(src, dst, th, hlen, tlen, so, m) 2677 struct sockaddr *src; 2678 struct sockaddr *dst; 2679 struct tcphdr *th; 2680 unsigned int hlen, tlen; 2681 struct socket *so; 2682 struct mbuf *m; 2683 { 2684 struct syn_cache *sc; 2685 struct syn_cache_head *scp; 2686 register struct inpcb *inp = NULL; 2687 #ifdef INET6 2688 register struct in6pcb *in6p = NULL; 2689 #endif 2690 register struct tcpcb *tp = 0; 2691 struct mbuf *am; 2692 int s; 2693 struct socket *oso; 2694 2695 s = splsoftnet(); 2696 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2697 splx(s); 2698 return (NULL); 2699 } 2700 2701 /* 2702 * Verify the sequence and ack numbers. Try getting the correct 2703 * response again. 2704 */ 2705 if ((th->th_ack != sc->sc_iss + 1) || 2706 SEQ_LEQ(th->th_seq, sc->sc_irs) || 2707 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 2708 (void) syn_cache_respond(sc, m); 2709 splx(s); 2710 return ((struct socket *)(-1)); 2711 } 2712 2713 /* Remove this cache entry */ 2714 SYN_CACHE_RM(sc); 2715 splx(s); 2716 2717 /* 2718 * Ok, create the full blown connection, and set things up 2719 * as they would have been set up if we had created the 2720 * connection when the SYN arrived. If we can't create 2721 * the connection, abort it. 2722 */ 2723 /* 2724 * inp still has the OLD in_pcb stuff, set the 2725 * v6-related flags on the new guy, too. This is 2726 * done particularly for the case where an AF_INET6 2727 * socket is bound only to a port, and a v4 connection 2728 * comes in on that port. 2729 * we also copy the flowinfo from the original pcb 2730 * to the new one. 2731 */ 2732 { 2733 struct inpcb *parentinpcb; 2734 2735 parentinpcb = (struct inpcb *)so->so_pcb; 2736 2737 oso = so; 2738 so = sonewconn(so, SS_ISCONNECTED); 2739 if (so == NULL) 2740 goto resetandabort; 2741 2742 switch (so->so_proto->pr_domain->dom_family) { 2743 case AF_INET: 2744 inp = sotoinpcb(so); 2745 break; 2746 #ifdef INET6 2747 case AF_INET6: 2748 in6p = sotoin6pcb(so); 2749 #if 0 /*def INET6*/ 2750 inp->inp_flags |= (parentinpcb->inp_flags & 2751 (INP_IPV6 | INP_IPV6_UNDEC | INP_IPV6_MAPPED)); 2752 if ((inp->inp_flags & INP_IPV6) && 2753 !(inp->inp_flags & INP_IPV6_MAPPED)) { 2754 inp->inp_ipv6.ip6_hlim = parentinpcb->inp_ipv6.ip6_hlim; 2755 inp->inp_ipv6.ip6_vfc = parentinpcb->inp_ipv6.ip6_vfc; 2756 } 2757 #endif 2758 break; 2759 #endif 2760 } 2761 } 2762 switch (src->sa_family) { 2763 case AF_INET: 2764 if (inp) { 2765 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 2766 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 2767 inp->inp_options = ip_srcroute(); 2768 in_pcbstate(inp, INP_BOUND); 2769 if (inp->inp_options == NULL) { 2770 inp->inp_options = sc->sc_ipopts; 2771 sc->sc_ipopts = NULL; 2772 } 2773 } 2774 #ifdef INET6 2775 else if (in6p) { 2776 /* IPv4 packet to AF_INET6 socket */ 2777 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 2778 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 2779 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 2780 &in6p->in6p_laddr.s6_addr32[3], 2781 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 2782 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 2783 in6totcpcb(in6p)->t_family = AF_INET; 2784 } 2785 #endif 2786 break; 2787 #ifdef INET6 2788 case AF_INET6: 2789 if (in6p) { 2790 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 2791 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 2792 #if 0 2793 in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK; 2794 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 2795 #endif 2796 } 2797 break; 2798 #endif 2799 } 2800 #ifdef INET6 2801 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 2802 struct in6pcb *oin6p = sotoin6pcb(oso); 2803 /* inherit socket options from the listening socket */ 2804 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 2805 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 2806 m_freem(in6p->in6p_options); 2807 in6p->in6p_options = 0; 2808 } 2809 ip6_savecontrol(in6p, &in6p->in6p_options, 2810 mtod(m, struct ip6_hdr *), m); 2811 } 2812 #endif 2813 2814 #ifdef IPSEC 2815 { 2816 struct secpolicy *sp; 2817 if (inp) { 2818 sp = ipsec_copy_policy(sotoinpcb(oso)->inp_sp); 2819 if (sp) { 2820 key_freesp(inp->inp_sp); 2821 inp->inp_sp = sp; 2822 } else 2823 printf("tcp_input: could not copy policy\n"); 2824 } 2825 #ifdef INET6 2826 else if (in6p) { 2827 sp = ipsec_copy_policy(sotoin6pcb(oso)->in6p_sp); 2828 if (sp) { 2829 key_freesp(in6p->in6p_sp); 2830 in6p->in6p_sp = sp; 2831 } else 2832 printf("tcp_input: could not copy policy\n"); 2833 } 2834 #endif 2835 } 2836 #endif 2837 2838 /* 2839 * Give the new socket our cached route reference. 2840 */ 2841 if (inp) 2842 inp->inp_route = sc->sc_route4; /* struct assignment */ 2843 #ifdef INET6 2844 else 2845 in6p->in6p_route = sc->sc_route6; 2846 #endif 2847 sc->sc_route4.ro_rt = NULL; 2848 2849 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 2850 if (am == NULL) 2851 goto resetandabort; 2852 am->m_len = src->sa_len; 2853 bcopy(src, mtod(am, caddr_t), src->sa_len); 2854 if (inp) { 2855 if (in_pcbconnect(inp, am)) { 2856 (void) m_free(am); 2857 goto resetandabort; 2858 } 2859 } 2860 #ifdef INET6 2861 else if (in6p) { 2862 if (src->sa_family == AF_INET) { 2863 /* IPv4 packet to AF_INET6 socket */ 2864 struct sockaddr_in6 *sin6; 2865 sin6 = mtod(am, struct sockaddr_in6 *); 2866 am->m_len = sizeof(*sin6); 2867 bzero(sin6, sizeof(*sin6)); 2868 sin6->sin6_family = AF_INET6; 2869 sin6->sin6_len = sizeof(*sin6); 2870 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 2871 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 2872 bcopy(&((struct sockaddr_in *)src)->sin_addr, 2873 &sin6->sin6_addr.s6_addr32[3], 2874 sizeof(sin6->sin6_addr.s6_addr32[3])); 2875 } 2876 if (in6_pcbconnect(in6p, am)) { 2877 (void) m_free(am); 2878 goto resetandabort; 2879 } 2880 } 2881 #endif 2882 else { 2883 (void) m_free(am); 2884 goto resetandabort; 2885 } 2886 (void) m_free(am); 2887 2888 if (inp) 2889 tp = intotcpcb(inp); 2890 #ifdef INET6 2891 else if (in6p) 2892 tp = in6totcpcb(in6p); 2893 #endif 2894 else 2895 tp = NULL; 2896 if (sc->sc_request_r_scale != 15) { 2897 tp->requested_s_scale = sc->sc_requested_s_scale; 2898 tp->request_r_scale = sc->sc_request_r_scale; 2899 tp->snd_scale = sc->sc_requested_s_scale; 2900 tp->rcv_scale = sc->sc_request_r_scale; 2901 tp->t_flags |= TF_RCVD_SCALE; 2902 } 2903 if (sc->sc_flags & SCF_TIMESTAMP) 2904 tp->t_flags |= TF_RCVD_TSTMP; 2905 2906 tp->t_template = tcp_template(tp); 2907 if (tp->t_template == 0) { 2908 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 2909 so = NULL; 2910 m_freem(m); 2911 goto abort; 2912 } 2913 2914 tp->iss = sc->sc_iss; 2915 tp->irs = sc->sc_irs; 2916 tcp_sendseqinit(tp); 2917 tcp_rcvseqinit(tp); 2918 tp->t_state = TCPS_SYN_RECEIVED; 2919 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 2920 tcpstat.tcps_accepts++; 2921 2922 /* Initialize tp->t_ourmss before we deal with the peer's! */ 2923 tp->t_ourmss = sc->sc_ourmaxseg; 2924 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 2925 2926 /* 2927 * Initialize the initial congestion window. If we 2928 * had to retransmit the SYN,ACK, we must initialize cwnd 2929 * to 1 segment (i.e. the Loss Window). 2930 */ 2931 if (sc->sc_rxtshift) 2932 tp->snd_cwnd = tp->t_peermss; 2933 else 2934 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 2935 2936 tcp_rmx_rtt(tp); 2937 tp->snd_wl1 = sc->sc_irs; 2938 tp->rcv_up = sc->sc_irs + 1; 2939 2940 /* 2941 * This is what whould have happened in tcp_ouput() when 2942 * the SYN,ACK was sent. 2943 */ 2944 tp->snd_up = tp->snd_una; 2945 tp->snd_max = tp->snd_nxt = tp->iss+1; 2946 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2947 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 2948 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 2949 tp->last_ack_sent = tp->rcv_nxt; 2950 2951 tcpstat.tcps_sc_completed++; 2952 SYN_CACHE_PUT(sc); 2953 return (so); 2954 2955 resetandabort: 2956 (void) tcp_respond(NULL, m, m, th, 2957 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 2958 abort: 2959 if (so != NULL) 2960 (void) soabort(so); 2961 SYN_CACHE_PUT(sc); 2962 tcpstat.tcps_sc_aborted++; 2963 return ((struct socket *)(-1)); 2964 } 2965 2966 /* 2967 * This function is called when we get a RST for a 2968 * non-existant connection, so that we can see if the 2969 * connection is in the syn cache. If it is, zap it. 2970 */ 2971 2972 void 2973 syn_cache_reset(src, dst, th) 2974 struct sockaddr *src; 2975 struct sockaddr *dst; 2976 struct tcphdr *th; 2977 { 2978 struct syn_cache *sc; 2979 struct syn_cache_head *scp; 2980 int s = splsoftnet(); 2981 2982 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 2983 splx(s); 2984 return; 2985 } 2986 if (SEQ_LT(th->th_seq, sc->sc_irs) || 2987 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 2988 splx(s); 2989 return; 2990 } 2991 SYN_CACHE_RM(sc); 2992 splx(s); 2993 tcpstat.tcps_sc_reset++; 2994 SYN_CACHE_PUT(sc); 2995 } 2996 2997 void 2998 syn_cache_unreach(src, dst, th) 2999 struct sockaddr *src; 3000 struct sockaddr *dst; 3001 struct tcphdr *th; 3002 { 3003 struct syn_cache *sc; 3004 struct syn_cache_head *scp; 3005 int s; 3006 3007 s = splsoftnet(); 3008 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3009 splx(s); 3010 return; 3011 } 3012 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3013 if (ntohl (th->th_seq) != sc->sc_iss) { 3014 splx(s); 3015 return; 3016 } 3017 3018 /* 3019 * If we've rertransmitted 3 times and this is our second error, 3020 * we remove the entry. Otherwise, we allow it to continue on. 3021 * This prevents us from incorrectly nuking an entry during a 3022 * spurious network outage. 3023 * 3024 * See tcp_notify(). 3025 */ 3026 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3027 sc->sc_flags |= SCF_UNREACH; 3028 splx(s); 3029 return; 3030 } 3031 3032 SYN_CACHE_RM(sc); 3033 splx(s); 3034 tcpstat.tcps_sc_unreach++; 3035 SYN_CACHE_PUT(sc); 3036 } 3037 3038 /* 3039 * Given a LISTEN socket and an inbound SYN request, add 3040 * this to the syn cache, and send back a segment: 3041 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3042 * to the source. 3043 * 3044 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3045 * Doing so would require that we hold onto the data and deliver it 3046 * to the application. However, if we are the target of a SYN-flood 3047 * DoS attack, an attacker could send data which would eventually 3048 * consume all available buffer space if it were ACKed. By not ACKing 3049 * the data, we avoid this DoS scenario. 3050 */ 3051 3052 int 3053 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi) 3054 struct sockaddr *src; 3055 struct sockaddr *dst; 3056 struct tcphdr *th; 3057 unsigned int hlen; 3058 struct socket *so; 3059 struct mbuf *m; 3060 u_char *optp; 3061 int optlen; 3062 struct tcp_opt_info *oi; 3063 { 3064 struct tcpcb tb, *tp; 3065 long win; 3066 struct syn_cache *sc; 3067 struct syn_cache_head *scp; 3068 struct mbuf *ipopts; 3069 3070 tp = sototcpcb(so); 3071 3072 /* 3073 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3074 * in_broadcast() should never return true on a received 3075 * packet with M_BCAST not set. 3076 */ 3077 if (m->m_flags & (M_BCAST|M_MCAST)) 3078 return 0; 3079 #ifdef INET6 3080 if (m->m_flags & M_ANYCAST6) 3081 return 0; 3082 #endif 3083 3084 switch (src->sa_family) { 3085 case AF_INET: 3086 if (IN_MULTICAST(((struct sockaddr_in *)src)->sin_addr.s_addr) 3087 || IN_MULTICAST(((struct sockaddr_in *)dst)->sin_addr.s_addr)) 3088 return 0; 3089 break; 3090 #ifdef INET6 3091 case AF_INET6: 3092 if (IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)src)->sin6_addr) 3093 || IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)dst)->sin6_addr)) 3094 return 0; 3095 break; 3096 #endif 3097 default: 3098 return 0; 3099 } 3100 3101 /* 3102 * Initialize some local state. 3103 */ 3104 win = sbspace(&so->so_rcv); 3105 if (win > TCP_MAXWIN) 3106 win = TCP_MAXWIN; 3107 3108 if (src->sa_family == AF_INET) { 3109 /* 3110 * Remember the IP options, if any. 3111 */ 3112 ipopts = ip_srcroute(); 3113 } else 3114 ipopts = NULL; 3115 3116 if (optp) { 3117 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3118 tcp_dooptions(&tb, optp, optlen, th, oi); 3119 } else 3120 tb.t_flags = 0; 3121 3122 /* 3123 * See if we already have an entry for this connection. 3124 * If we do, resend the SYN,ACK. We do not count this 3125 * as a retransmission (XXX though maybe we should). 3126 */ 3127 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3128 tcpstat.tcps_sc_dupesyn++; 3129 if (ipopts) { 3130 /* 3131 * If we were remembering a previous source route, 3132 * forget it and use the new one we've been given. 3133 */ 3134 if (sc->sc_ipopts) 3135 (void) m_free(sc->sc_ipopts); 3136 sc->sc_ipopts = ipopts; 3137 } 3138 sc->sc_timestamp = tb.ts_recent; 3139 if (syn_cache_respond(sc, m) == 0) { 3140 tcpstat.tcps_sndacks++; 3141 tcpstat.tcps_sndtotal++; 3142 } 3143 return (1); 3144 } 3145 3146 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3147 if (sc == NULL) { 3148 if (ipopts) 3149 (void) m_free(ipopts); 3150 return (0); 3151 } 3152 3153 /* 3154 * Fill in the cache, and put the necessary IP and TCP 3155 * options into the reply. 3156 */ 3157 bzero(sc, sizeof(struct syn_cache)); 3158 bcopy(src, &sc->sc_src, src->sa_len); 3159 bcopy(dst, &sc->sc_dst, dst->sa_len); 3160 sc->sc_flags = 0; 3161 sc->sc_ipopts = ipopts; 3162 sc->sc_irs = th->th_seq; 3163 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 3164 sc->sc_peermaxseg = oi->maxseg; 3165 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3166 m->m_pkthdr.rcvif : NULL, 3167 sc->sc_src.sa.sa_family); 3168 sc->sc_win = win; 3169 sc->sc_timestamp = tb.ts_recent; 3170 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 3171 sc->sc_flags |= SCF_TIMESTAMP; 3172 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3173 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3174 sc->sc_requested_s_scale = tb.requested_s_scale; 3175 sc->sc_request_r_scale = 0; 3176 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3177 TCP_MAXWIN << sc->sc_request_r_scale < 3178 so->so_rcv.sb_hiwat) 3179 sc->sc_request_r_scale++; 3180 } else { 3181 sc->sc_requested_s_scale = 15; 3182 sc->sc_request_r_scale = 15; 3183 } 3184 sc->sc_tp = tp; 3185 if (syn_cache_respond(sc, m) == 0) { 3186 syn_cache_insert(sc, tp); 3187 tcpstat.tcps_sndacks++; 3188 tcpstat.tcps_sndtotal++; 3189 } else { 3190 SYN_CACHE_PUT(sc); 3191 tcpstat.tcps_sc_dropped++; 3192 } 3193 return (1); 3194 } 3195 3196 int 3197 syn_cache_respond(sc, m) 3198 struct syn_cache *sc; 3199 struct mbuf *m; 3200 { 3201 struct route *ro; 3202 struct rtentry *rt; 3203 u_int8_t *optp; 3204 int optlen, error; 3205 u_int16_t tlen; 3206 struct ip *ip = NULL; 3207 #ifdef INET6 3208 struct ip6_hdr *ip6 = NULL; 3209 #endif 3210 struct tcphdr *th; 3211 u_int hlen; 3212 3213 switch (sc->sc_src.sa.sa_family) { 3214 case AF_INET: 3215 hlen = sizeof(struct ip); 3216 ro = &sc->sc_route4; 3217 break; 3218 #ifdef INET6 3219 case AF_INET6: 3220 hlen = sizeof(struct ip6_hdr); 3221 ro = (struct route *)&sc->sc_route6; 3222 break; 3223 #endif 3224 default: 3225 if (m) 3226 m_freem(m); 3227 return EAFNOSUPPORT; 3228 } 3229 3230 /* Compute the size of the TCP options. */ 3231 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3232 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3233 3234 tlen = hlen + sizeof(struct tcphdr) + optlen; 3235 3236 /* 3237 * Create the IP+TCP header from scratch. Reuse the received mbuf 3238 * if possible. 3239 */ 3240 if (m != NULL) { 3241 m_freem(m->m_next); 3242 m->m_next = NULL; 3243 MRESETDATA(m); 3244 } else { 3245 MGETHDR(m, M_DONTWAIT, MT_DATA); 3246 if (m == NULL) 3247 return (ENOBUFS); 3248 } 3249 3250 /* Fixup the mbuf. */ 3251 m->m_data += max_linkhdr; 3252 m->m_len = m->m_pkthdr.len = tlen; 3253 #ifdef IPSEC 3254 if (sc->sc_tp) { 3255 struct tcpcb *tp; 3256 struct socket *so; 3257 3258 tp = sc->sc_tp; 3259 if (tp->t_inpcb) 3260 so = tp->t_inpcb->inp_socket; 3261 #ifdef INET6 3262 else if (tp->t_in6pcb) 3263 so = tp->t_in6pcb->in6p_socket; 3264 #endif 3265 else 3266 so = NULL; 3267 /* use IPsec policy on listening socket, on SYN ACK */ 3268 m->m_pkthdr.rcvif = (struct ifnet *)so; 3269 } 3270 #else 3271 m->m_pkthdr.rcvif = NULL; 3272 #endif 3273 memset(mtod(m, u_char *), 0, tlen); 3274 3275 switch (sc->sc_src.sa.sa_family) { 3276 case AF_INET: 3277 ip = mtod(m, struct ip *); 3278 ip->ip_dst = sc->sc_src.sin.sin_addr; 3279 ip->ip_src = sc->sc_dst.sin.sin_addr; 3280 ip->ip_p = IPPROTO_TCP; 3281 th = (struct tcphdr *)(ip + 1); 3282 th->th_dport = sc->sc_src.sin.sin_port; 3283 th->th_sport = sc->sc_dst.sin.sin_port; 3284 break; 3285 #ifdef INET6 3286 case AF_INET6: 3287 ip6 = mtod(m, struct ip6_hdr *); 3288 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3289 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3290 ip6->ip6_nxt = IPPROTO_TCP; 3291 /* ip6_plen will be updated in ip6_output() */ 3292 th = (struct tcphdr *)(ip6 + 1); 3293 th->th_dport = sc->sc_src.sin6.sin6_port; 3294 th->th_sport = sc->sc_dst.sin6.sin6_port; 3295 break; 3296 #endif 3297 default: 3298 th = NULL; 3299 } 3300 3301 th->th_seq = htonl(sc->sc_iss); 3302 th->th_ack = htonl(sc->sc_irs + 1); 3303 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3304 th->th_flags = TH_SYN|TH_ACK; 3305 th->th_win = htons(sc->sc_win); 3306 /* th_sum already 0 */ 3307 /* th_urp already 0 */ 3308 3309 /* Tack on the TCP options. */ 3310 optp = (u_int8_t *)(th + 1); 3311 *optp++ = TCPOPT_MAXSEG; 3312 *optp++ = 4; 3313 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3314 *optp++ = sc->sc_ourmaxseg & 0xff; 3315 3316 if (sc->sc_request_r_scale != 15) { 3317 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3318 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 3319 sc->sc_request_r_scale); 3320 optp += 4; 3321 } 3322 3323 if (sc->sc_flags & SCF_TIMESTAMP) { 3324 u_int32_t *lp = (u_int32_t *)(optp); 3325 /* Form timestamp option as shown in appendix A of RFC 1323. */ 3326 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 3327 *lp++ = htonl(tcp_now); 3328 *lp = htonl(sc->sc_timestamp); 3329 optp += TCPOLEN_TSTAMP_APPA; 3330 } 3331 3332 /* Compute the packet's checksum. */ 3333 switch (sc->sc_src.sa.sa_family) { 3334 case AF_INET: 3335 ip->ip_len = htons(tlen - hlen); 3336 th->th_sum = 0; 3337 th->th_sum = in_cksum(m, tlen); 3338 break; 3339 #ifdef INET6 3340 case AF_INET6: 3341 ip6->ip6_plen = htons(tlen - hlen); 3342 th->th_sum = 0; 3343 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 3344 break; 3345 #endif 3346 } 3347 3348 /* 3349 * Fill in some straggling IP bits. Note the stack expects 3350 * ip_len to be in host order, for convenience. 3351 */ 3352 switch (sc->sc_src.sa.sa_family) { 3353 case AF_INET: 3354 ip->ip_len = tlen; 3355 ip->ip_ttl = ip_defttl; 3356 /* XXX tos? */ 3357 break; 3358 #ifdef INET6 3359 case AF_INET6: 3360 ip6->ip6_vfc = IPV6_VERSION; 3361 ip6->ip6_plen = htons(tlen - hlen); 3362 ip6->ip6_hlim = ip6_defhlim; 3363 /* XXX flowlabel? */ 3364 break; 3365 #endif 3366 } 3367 3368 /* 3369 * If we're doing Path MTU discovery, we need to set DF unless 3370 * the route's MTU is locked. If we don't yet know the route, 3371 * look it up now. We will copy this reference to the inpcb 3372 * when we finish creating the connection. 3373 */ 3374 if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) { 3375 if (ro->ro_rt != NULL) { 3376 RTFREE(ro->ro_rt); 3377 ro->ro_rt = NULL; 3378 } 3379 bcopy(&sc->sc_src, &ro->ro_dst, sc->sc_src.sa.sa_len); 3380 rtalloc(ro); 3381 if ((rt = ro->ro_rt) == NULL) { 3382 m_freem(m); 3383 switch (sc->sc_src.sa.sa_family) { 3384 case AF_INET: 3385 ipstat.ips_noroute++; 3386 break; 3387 #ifdef INET6 3388 case AF_INET6: 3389 ip6stat.ip6s_noroute++; 3390 break; 3391 #endif 3392 } 3393 return (EHOSTUNREACH); 3394 } 3395 } 3396 3397 switch (sc->sc_src.sa.sa_family) { 3398 case AF_INET: 3399 if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0) 3400 ip->ip_off |= IP_DF; 3401 3402 /* ...and send it off! */ 3403 error = ip_output(m, sc->sc_ipopts, ro, 0, NULL); 3404 break; 3405 #ifdef INET6 3406 case AF_INET6: 3407 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 3408 0, NULL); 3409 break; 3410 #endif 3411 default: 3412 error = EAFNOSUPPORT; 3413 break; 3414 } 3415 return (error); 3416 } 3417
Indexes created Mon Sep 22 05:09:51 GMT 2025