1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_ipfw.h" /* for ipfw_fwd */ 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_mac.h" 41 #include "opt_tcpdebug.h" 42 #include "opt_tcp_input.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/mac.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/proc.h> /* for proc0 declaration */ 50 #include <sys/protosw.h> 51 #include <sys/signalvar.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/sysctl.h> 55 #include <sys/syslog.h> 56 #include <sys/systm.h> 57 58 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 59 60 #include <net/if.h> 61 #include <net/route.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip.h> 68 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70 #include <netinet/ip_var.h> 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/in6_pcb.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet6/tcp6_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif /* TCPDEBUG */ 86 87 #ifdef FAST_IPSEC 88 #include <netipsec/ipsec.h> 89 #include <netipsec/ipsec6.h> 90 #endif /*FAST_IPSEC*/ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #include <netinet6/ipsec6.h> 95 #include <netkey/key.h> 96 #endif /*IPSEC*/ 97 98 #include <machine/in_cksum.h> 99 100 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 101 102 static const int tcprexmtthresh = 3; 103 tcp_cc tcp_ccgen; 104 105 struct tcpstat tcpstat; 106 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 107 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 108 109 static int log_in_vain = 0; 110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 111 &log_in_vain, 0, "Log all incoming TCP connections"); 112 113 static int blackhole = 0; 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 115 &blackhole, 0, "Do not send RST when dropping refused connections"); 116 117 int tcp_delack_enabled = 1; 118 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 119 &tcp_delack_enabled, 0, 120 "Delay ACK to try and piggyback it onto a data packet"); 121 122 #ifdef TCP_DROP_SYNFIN 123 static int drop_synfin = 0; 124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 125 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 126 #endif 127 128 static int tcp_do_rfc3042 = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 130 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 131 132 static int tcp_do_rfc3390 = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 134 &tcp_do_rfc3390, 0, 135 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 136 137 struct inpcbhead tcb; 138 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 139 struct inpcbinfo tcbinfo; 140 struct mtx *tcbinfo_mtx; 141 142 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 143 static void tcp_pulloutofband(struct socket *, 144 struct tcphdr *, struct mbuf *, int); 145 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 146 struct mbuf *); 147 static void tcp_xmit_timer(struct tcpcb *, int); 148 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 149 static int tcp_timewait(struct tcptw *, struct tcpopt *, 150 struct tcphdr *, struct mbuf *, int); 151 152 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 153 #ifdef INET6 154 #define ND6_HINT(tp) \ 155 do { \ 156 if ((tp) && (tp)->t_inpcb && \ 157 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 158 nd6_nud_hint(NULL, NULL, 0); \ 159 } while (0) 160 #else 161 #define ND6_HINT(tp) 162 #endif 163 164 /* 165 * Indicate whether this ack should be delayed. We can delay the ack if 166 * - there is no delayed ack timer in progress and 167 * - our last ack wasn't a 0-sized window. We never want to delay 168 * the ack that opens up a 0-sized window and 169 * - delayed acks are enabled or 170 * - this is a half-synchronized T/TCP connection. 171 */ 172 #define DELAY_ACK(tp) \ 173 ((!callout_active(tp->tt_delack) && \ 174 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 175 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 176 177 static int 178 tcp_reass(tp, th, tlenp, m) 179 register struct tcpcb *tp; 180 register struct tcphdr *th; 181 int *tlenp; 182 struct mbuf *m; 183 { 184 struct tseg_qent *q; 185 struct tseg_qent *p = NULL; 186 struct tseg_qent *nq; 187 struct tseg_qent *te; 188 struct socket *so = tp->t_inpcb->inp_socket; 189 int flags; 190 191 /* 192 * Call with th==0 after become established to 193 * force pre-ESTABLISHED data up to user socket. 194 */ 195 if (th == 0) 196 goto present; 197 198 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 199 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 200 M_NOWAIT); 201 if (te == NULL) { 202 tcpstat.tcps_rcvmemdrop++; 203 m_freem(m); 204 return (0); 205 } 206 207 /* 208 * Find a segment which begins after this one does. 209 */ 210 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 211 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 212 break; 213 p = q; 214 } 215 216 /* 217 * If there is a preceding segment, it may provide some of 218 * our data already. If so, drop the data from the incoming 219 * segment. If it provides all of our data, drop us. 220 */ 221 if (p != NULL) { 222 register int i; 223 /* conversion to int (in i) handles seq wraparound */ 224 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 225 if (i > 0) { 226 if (i >= *tlenp) { 227 tcpstat.tcps_rcvduppack++; 228 tcpstat.tcps_rcvdupbyte += *tlenp; 229 m_freem(m); 230 FREE(te, M_TSEGQ); 231 /* 232 * Try to present any queued data 233 * at the left window edge to the user. 234 * This is needed after the 3-WHS 235 * completes. 236 */ 237 goto present; /* ??? */ 238 } 239 m_adj(m, i); 240 *tlenp -= i; 241 th->th_seq += i; 242 } 243 } 244 tcpstat.tcps_rcvoopack++; 245 tcpstat.tcps_rcvoobyte += *tlenp; 246 247 /* 248 * While we overlap succeeding segments trim them or, 249 * if they are completely covered, dequeue them. 250 */ 251 while (q) { 252 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 253 if (i <= 0) 254 break; 255 if (i < q->tqe_len) { 256 q->tqe_th->th_seq += i; 257 q->tqe_len -= i; 258 m_adj(q->tqe_m, i); 259 break; 260 } 261 262 nq = LIST_NEXT(q, tqe_q); 263 LIST_REMOVE(q, tqe_q); 264 m_freem(q->tqe_m); 265 FREE(q, M_TSEGQ); 266 q = nq; 267 } 268 269 /* Insert the new segment queue entry into place. */ 270 te->tqe_m = m; 271 te->tqe_th = th; 272 te->tqe_len = *tlenp; 273 274 if (p == NULL) { 275 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 276 } else { 277 LIST_INSERT_AFTER(p, te, tqe_q); 278 } 279 280 present: 281 /* 282 * Present data to user, advancing rcv_nxt through 283 * completed sequence space. 284 */ 285 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 286 return (0); 287 q = LIST_FIRST(&tp->t_segq); 288 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 289 return (0); 290 do { 291 tp->rcv_nxt += q->tqe_len; 292 flags = q->tqe_th->th_flags & TH_FIN; 293 nq = LIST_NEXT(q, tqe_q); 294 LIST_REMOVE(q, tqe_q); 295 if (so->so_state & SS_CANTRCVMORE) 296 m_freem(q->tqe_m); 297 else 298 sbappendstream(&so->so_rcv, q->tqe_m); 299 FREE(q, M_TSEGQ); 300 q = nq; 301 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 302 ND6_HINT(tp); 303 sorwakeup(so); 304 return (flags); 305 } 306 307 /* 308 * TCP input routine, follows pages 65-76 of the 309 * protocol specification dated September, 1981 very closely. 310 */ 311 #ifdef INET6 312 int 313 tcp6_input(mp, offp, proto) 314 struct mbuf **mp; 315 int *offp, proto; 316 { 317 register struct mbuf *m = *mp; 318 struct in6_ifaddr *ia6; 319 320 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 321 322 /* 323 * draft-itojun-ipv6-tcp-to-anycast 324 * better place to put this in? 325 */ 326 ia6 = ip6_getdstifaddr(m); 327 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 328 struct ip6_hdr *ip6; 329 330 ip6 = mtod(m, struct ip6_hdr *); 331 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 332 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 333 return IPPROTO_DONE; 334 } 335 336 tcp_input(m, *offp); 337 return IPPROTO_DONE; 338 } 339 #endif 340 341 void 342 tcp_input(m, off0) 343 register struct mbuf *m; 344 int off0; 345 { 346 register struct tcphdr *th; 347 register struct ip *ip = NULL; 348 register struct ipovly *ipov; 349 register struct inpcb *inp = NULL; 350 u_char *optp = NULL; 351 int optlen = 0; 352 int len, tlen, off; 353 int drop_hdrlen; 354 register struct tcpcb *tp = 0; 355 register int thflags; 356 struct socket *so = 0; 357 int todrop, acked, ourfinisacked, needoutput = 0; 358 u_long tiwin; 359 struct tcpopt to; /* options in this segment */ 360 struct rmxp_tao tao; /* our TAO cache entry */ 361 int headlocked = 0; 362 struct sockaddr_in *next_hop = NULL; 363 int rstreason; /* For badport_bandlim accounting purposes */ 364 365 struct ip6_hdr *ip6 = NULL; 366 #ifdef INET6 367 int isipv6; 368 #else 369 const int isipv6 = 0; 370 #endif 371 372 #ifdef TCPDEBUG 373 /* 374 * The size of tcp_saveipgen must be the size of the max ip header, 375 * now IPv6. 376 */ 377 u_char tcp_saveipgen[40]; 378 struct tcphdr tcp_savetcp; 379 short ostate = 0; 380 #endif 381 382 /* Grab info from MT_TAG mbufs prepended to the chain. */ 383 for (;m && m->m_type == MT_TAG; m = m->m_next) { 384 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 385 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 386 } 387 #ifdef INET6 388 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 389 #endif 390 bzero(&tao, sizeof(tao)); 391 bzero((char *)&to, sizeof(to)); 392 393 tcpstat.tcps_rcvtotal++; 394 395 if (isipv6) { 396 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 397 ip6 = mtod(m, struct ip6_hdr *); 398 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 399 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 400 tcpstat.tcps_rcvbadsum++; 401 goto drop; 402 } 403 th = (struct tcphdr *)((caddr_t)ip6 + off0); 404 405 /* 406 * Be proactive about unspecified IPv6 address in source. 407 * As we use all-zero to indicate unbounded/unconnected pcb, 408 * unspecified IPv6 address can be used to confuse us. 409 * 410 * Note that packets with unspecified IPv6 destination is 411 * already dropped in ip6_input. 412 */ 413 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 414 /* XXX stat */ 415 goto drop; 416 } 417 } else { 418 /* 419 * Get IP and TCP header together in first mbuf. 420 * Note: IP leaves IP header in first mbuf. 421 */ 422 if (off0 > sizeof (struct ip)) { 423 ip_stripoptions(m, (struct mbuf *)0); 424 off0 = sizeof(struct ip); 425 } 426 if (m->m_len < sizeof (struct tcpiphdr)) { 427 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 428 tcpstat.tcps_rcvshort++; 429 return; 430 } 431 } 432 ip = mtod(m, struct ip *); 433 ipov = (struct ipovly *)ip; 434 th = (struct tcphdr *)((caddr_t)ip + off0); 435 tlen = ip->ip_len; 436 437 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 438 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 439 th->th_sum = m->m_pkthdr.csum_data; 440 else 441 th->th_sum = in_pseudo(ip->ip_src.s_addr, 442 ip->ip_dst.s_addr, 443 htonl(m->m_pkthdr.csum_data + 444 ip->ip_len + 445 IPPROTO_TCP)); 446 th->th_sum ^= 0xffff; 447 #ifdef TCPDEBUG 448 ipov->ih_len = (u_short)tlen; 449 ipov->ih_len = htons(ipov->ih_len); 450 #endif 451 } else { 452 /* 453 * Checksum extended TCP header and data. 454 */ 455 len = sizeof (struct ip) + tlen; 456 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 457 ipov->ih_len = (u_short)tlen; 458 ipov->ih_len = htons(ipov->ih_len); 459 th->th_sum = in_cksum(m, len); 460 } 461 if (th->th_sum) { 462 tcpstat.tcps_rcvbadsum++; 463 goto drop; 464 } 465 #ifdef INET6 466 /* Re-initialization for later version check */ 467 ip->ip_v = IPVERSION; 468 #endif 469 } 470 471 /* 472 * Check that TCP offset makes sense, 473 * pull out TCP options and adjust length. XXX 474 */ 475 off = th->th_off << 2; 476 if (off < sizeof (struct tcphdr) || off > tlen) { 477 tcpstat.tcps_rcvbadoff++; 478 goto drop; 479 } 480 tlen -= off; /* tlen is used instead of ti->ti_len */ 481 if (off > sizeof (struct tcphdr)) { 482 if (isipv6) { 483 IP6_EXTHDR_CHECK(m, off0, off, ); 484 ip6 = mtod(m, struct ip6_hdr *); 485 th = (struct tcphdr *)((caddr_t)ip6 + off0); 486 } else { 487 if (m->m_len < sizeof(struct ip) + off) { 488 if ((m = m_pullup(m, sizeof (struct ip) + off)) 489 == 0) { 490 tcpstat.tcps_rcvshort++; 491 return; 492 } 493 ip = mtod(m, struct ip *); 494 ipov = (struct ipovly *)ip; 495 th = (struct tcphdr *)((caddr_t)ip + off0); 496 } 497 } 498 optlen = off - sizeof (struct tcphdr); 499 optp = (u_char *)(th + 1); 500 } 501 thflags = th->th_flags; 502 503 #ifdef TCP_DROP_SYNFIN 504 /* 505 * If the drop_synfin option is enabled, drop all packets with 506 * both the SYN and FIN bits set. This prevents e.g. nmap from 507 * identifying the TCP/IP stack. 508 * 509 * This is a violation of the TCP specification. 510 */ 511 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 512 goto drop; 513 #endif 514 515 /* 516 * Convert TCP protocol specific fields to host format. 517 */ 518 th->th_seq = ntohl(th->th_seq); 519 th->th_ack = ntohl(th->th_ack); 520 th->th_win = ntohs(th->th_win); 521 th->th_urp = ntohs(th->th_urp); 522 523 /* 524 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 525 * until after ip6_savecontrol() is called and before other functions 526 * which don't want those proto headers. 527 * Because ip6_savecontrol() is going to parse the mbuf to 528 * search for data to be passed up to user-land, it wants mbuf 529 * parameters to be unchanged. 530 * XXX: the call of ip6_savecontrol() has been obsoleted based on 531 * latest version of the advanced API (20020110). 532 */ 533 drop_hdrlen = off0 + off; 534 535 /* 536 * Locate pcb for segment. 537 */ 538 INP_INFO_WLOCK(&tcbinfo); 539 headlocked = 1; 540 findpcb: 541 /* IPFIREWALL_FORWARD section */ 542 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 543 /* 544 * Transparently forwarded. Pretend to be the destination. 545 * already got one like this? 546 */ 547 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 548 ip->ip_dst, th->th_dport, 549 0, m->m_pkthdr.rcvif); 550 if (!inp) { 551 /* It's new. Try find the ambushing socket. */ 552 inp = in_pcblookup_hash(&tcbinfo, 553 ip->ip_src, th->th_sport, 554 next_hop->sin_addr, 555 next_hop->sin_port ? 556 ntohs(next_hop->sin_port) : 557 th->th_dport, 558 1, m->m_pkthdr.rcvif); 559 } 560 } else { 561 if (isipv6) 562 inp = in6_pcblookup_hash(&tcbinfo, 563 &ip6->ip6_src, th->th_sport, 564 &ip6->ip6_dst, th->th_dport, 565 1, m->m_pkthdr.rcvif); 566 else 567 inp = in_pcblookup_hash(&tcbinfo, 568 ip->ip_src, th->th_sport, 569 ip->ip_dst, th->th_dport, 570 1, m->m_pkthdr.rcvif); 571 } 572 573 #ifdef IPSEC 574 if (isipv6) { 575 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 576 ipsec6stat.in_polvio++; 577 goto drop; 578 } 579 } else { 580 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 581 ipsecstat.in_polvio++; 582 goto drop; 583 } 584 } 585 #endif 586 #ifdef FAST_IPSEC 587 if (isipv6) { 588 if (inp != NULL && ipsec6_in_reject(m, inp)) { 589 goto drop; 590 } 591 } else 592 if (inp != NULL && ipsec4_in_reject(m, inp)) { 593 goto drop; 594 } 595 #endif /*FAST_IPSEC*/ 596 597 /* 598 * If the state is CLOSED (i.e., TCB does not exist) then 599 * all data in the incoming segment is discarded. 600 * If the TCB exists but is in CLOSED state, it is embryonic, 601 * but should either do a listen or a connect soon. 602 */ 603 if (inp == NULL) { 604 if (log_in_vain) { 605 #ifdef INET6 606 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 607 #else 608 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 609 #endif 610 611 if (isipv6) { 612 strcpy(dbuf, "["); 613 strcpy(sbuf, "["); 614 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 615 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 616 strcat(dbuf, "]"); 617 strcat(sbuf, "]"); 618 } else { 619 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 620 strcpy(sbuf, inet_ntoa(ip->ip_src)); 621 } 622 switch (log_in_vain) { 623 case 1: 624 if ((thflags & TH_SYN) == 0) 625 break; 626 /* FALLTHROUGH */ 627 case 2: 628 log(LOG_INFO, 629 "Connection attempt to TCP %s:%d " 630 "from %s:%d flags:0x%02x\n", 631 dbuf, ntohs(th->th_dport), sbuf, 632 ntohs(th->th_sport), thflags); 633 break; 634 default: 635 break; 636 } 637 } 638 if (blackhole) { 639 switch (blackhole) { 640 case 1: 641 if (thflags & TH_SYN) 642 goto drop; 643 break; 644 case 2: 645 goto drop; 646 default: 647 goto drop; 648 } 649 } 650 rstreason = BANDLIM_RST_CLOSEDPORT; 651 goto dropwithreset; 652 } 653 INP_LOCK(inp); 654 if (inp->inp_vflag & INP_TIMEWAIT) { 655 /* 656 * The only option of relevance is TOF_CC, and only if 657 * present in a SYN segment. See tcp_timewait(). 658 */ 659 if (thflags & TH_SYN) 660 tcp_dooptions(&to, optp, optlen, 1); 661 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 662 &to, th, m, tlen)) 663 goto findpcb; 664 /* 665 * tcp_timewait unlocks inp. 666 */ 667 INP_INFO_WUNLOCK(&tcbinfo); 668 return; 669 } 670 tp = intotcpcb(inp); 671 if (tp == 0) { 672 INP_UNLOCK(inp); 673 rstreason = BANDLIM_RST_CLOSEDPORT; 674 goto dropwithreset; 675 } 676 if (tp->t_state == TCPS_CLOSED) 677 goto drop; 678 679 /* Unscale the window into a 32-bit value. */ 680 if ((thflags & TH_SYN) == 0) 681 tiwin = th->th_win << tp->snd_scale; 682 else 683 tiwin = th->th_win; 684 685 #ifdef MAC 686 if (mac_check_inpcb_deliver(inp, m)) 687 goto drop; 688 #endif 689 so = inp->inp_socket; 690 #ifdef TCPDEBUG 691 if (so->so_options & SO_DEBUG) { 692 ostate = tp->t_state; 693 if (isipv6) 694 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 695 else 696 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 697 tcp_savetcp = *th; 698 } 699 #endif 700 if (so->so_options & SO_ACCEPTCONN) { 701 struct in_conninfo inc; 702 703 #ifdef INET6 704 inc.inc_isipv6 = isipv6; 705 #endif 706 if (isipv6) { 707 inc.inc6_faddr = ip6->ip6_src; 708 inc.inc6_laddr = ip6->ip6_dst; 709 } else { 710 inc.inc_faddr = ip->ip_src; 711 inc.inc_laddr = ip->ip_dst; 712 } 713 inc.inc_fport = th->th_sport; 714 inc.inc_lport = th->th_dport; 715 716 /* 717 * If the state is LISTEN then ignore segment if it contains 718 * a RST. If the segment contains an ACK then it is bad and 719 * send a RST. If it does not contain a SYN then it is not 720 * interesting; drop it. 721 * 722 * If the state is SYN_RECEIVED (syncache) and seg contains 723 * an ACK, but not for our SYN/ACK, send a RST. If the seg 724 * contains a RST, check the sequence number to see if it 725 * is a valid reset segment. 726 */ 727 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 728 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 729 if (!syncache_expand(&inc, th, &so, m)) { 730 /* 731 * No syncache entry, or ACK was not 732 * for our SYN/ACK. Send a RST. 733 */ 734 tcpstat.tcps_badsyn++; 735 rstreason = BANDLIM_RST_OPENPORT; 736 goto dropwithreset; 737 } 738 if (so == NULL) { 739 /* 740 * Could not complete 3-way handshake, 741 * connection is being closed down, and 742 * syncache will free mbuf. 743 */ 744 INP_UNLOCK(inp); 745 INP_INFO_WUNLOCK(&tcbinfo); 746 return; 747 } 748 /* 749 * Socket is created in state SYN_RECEIVED. 750 * Continue processing segment. 751 */ 752 INP_UNLOCK(inp); 753 inp = sotoinpcb(so); 754 INP_LOCK(inp); 755 tp = intotcpcb(inp); 756 /* 757 * This is what would have happened in 758 * tcp_output() when the SYN,ACK was sent. 759 */ 760 tp->snd_up = tp->snd_una; 761 tp->snd_max = tp->snd_nxt = tp->iss + 1; 762 tp->last_ack_sent = tp->rcv_nxt; 763 /* 764 * RFC1323: The window in SYN & SYN/ACK 765 * segments is never scaled. 766 */ 767 tp->snd_wnd = tiwin; /* unscaled */ 768 goto after_listen; 769 } 770 if (thflags & TH_RST) { 771 syncache_chkrst(&inc, th); 772 goto drop; 773 } 774 if (thflags & TH_ACK) { 775 syncache_badack(&inc); 776 tcpstat.tcps_badsyn++; 777 rstreason = BANDLIM_RST_OPENPORT; 778 goto dropwithreset; 779 } 780 goto drop; 781 } 782 783 /* 784 * Segment's flags are (SYN) or (SYN|FIN). 785 */ 786 #ifdef INET6 787 /* 788 * If deprecated address is forbidden, 789 * we do not accept SYN to deprecated interface 790 * address to prevent any new inbound connection from 791 * getting established. 792 * When we do not accept SYN, we send a TCP RST, 793 * with deprecated source address (instead of dropping 794 * it). We compromise it as it is much better for peer 795 * to send a RST, and RST will be the final packet 796 * for the exchange. 797 * 798 * If we do not forbid deprecated addresses, we accept 799 * the SYN packet. RFC2462 does not suggest dropping 800 * SYN in this case. 801 * If we decipher RFC2462 5.5.4, it says like this: 802 * 1. use of deprecated addr with existing 803 * communication is okay - "SHOULD continue to be 804 * used" 805 * 2. use of it with new communication: 806 * (2a) "SHOULD NOT be used if alternate address 807 * with sufficient scope is available" 808 * (2b) nothing mentioned otherwise. 809 * Here we fall into (2b) case as we have no choice in 810 * our source address selection - we must obey the peer. 811 * 812 * The wording in RFC2462 is confusing, and there are 813 * multiple description text for deprecated address 814 * handling - worse, they are not exactly the same. 815 * I believe 5.5.4 is the best one, so we follow 5.5.4. 816 */ 817 if (isipv6 && !ip6_use_deprecated) { 818 struct in6_ifaddr *ia6; 819 820 if ((ia6 = ip6_getdstifaddr(m)) && 821 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 822 INP_UNLOCK(inp); 823 tp = NULL; 824 rstreason = BANDLIM_RST_OPENPORT; 825 goto dropwithreset; 826 } 827 } 828 #endif 829 /* 830 * If it is from this socket, drop it, it must be forged. 831 * Don't bother responding if the destination was a broadcast. 832 */ 833 if (th->th_dport == th->th_sport) { 834 if (isipv6) { 835 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 836 &ip6->ip6_src)) 837 goto drop; 838 } else { 839 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 840 goto drop; 841 } 842 } 843 /* 844 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 845 * 846 * Note that it is quite possible to receive unicast 847 * link-layer packets with a broadcast IP address. Use 848 * in_broadcast() to find them. 849 */ 850 if (m->m_flags & (M_BCAST|M_MCAST)) 851 goto drop; 852 if (isipv6) { 853 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 854 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 855 goto drop; 856 } else { 857 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 858 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 859 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 860 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 861 goto drop; 862 } 863 /* 864 * SYN appears to be valid; create compressed TCP state 865 * for syncache, or perform t/tcp connection. 866 */ 867 if (so->so_qlen <= so->so_qlimit) { 868 #ifdef TCPDEBUG 869 if (so->so_options & SO_DEBUG) 870 tcp_trace(TA_INPUT, ostate, tp, 871 (void *)tcp_saveipgen, &tcp_savetcp, 0); 872 #endif 873 tcp_dooptions(&to, optp, optlen, 1); 874 if (!syncache_add(&inc, &to, th, &so, m)) 875 goto drop; 876 if (so == NULL) { 877 /* 878 * Entry added to syncache, mbuf used to 879 * send SYN,ACK packet. 880 */ 881 KASSERT(headlocked, ("headlocked")); 882 INP_UNLOCK(inp); 883 INP_INFO_WUNLOCK(&tcbinfo); 884 return; 885 } 886 /* 887 * Segment passed TAO tests. 888 */ 889 INP_UNLOCK(inp); 890 inp = sotoinpcb(so); 891 INP_LOCK(inp); 892 tp = intotcpcb(inp); 893 tp->snd_wnd = tiwin; 894 tp->t_starttime = ticks; 895 tp->t_state = TCPS_ESTABLISHED; 896 897 /* 898 * T/TCP logic: 899 * If there is a FIN or if there is data, then 900 * delay SYN,ACK(SYN) in the hope of piggy-backing 901 * it on a response segment. Otherwise must send 902 * ACK now in case the other side is slow starting. 903 */ 904 if (thflags & TH_FIN || tlen != 0) 905 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 906 else 907 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 908 tcpstat.tcps_connects++; 909 soisconnected(so); 910 goto trimthenstep6; 911 } 912 goto drop; 913 } 914 after_listen: 915 916 /* XXX temp debugging */ 917 /* should not happen - syncache should pick up these connections */ 918 if (tp->t_state == TCPS_LISTEN) 919 panic("tcp_input: TCPS_LISTEN"); 920 921 /* 922 * Segment received on connection. 923 * Reset idle time and keep-alive timer. 924 */ 925 tp->t_rcvtime = ticks; 926 if (TCPS_HAVEESTABLISHED(tp->t_state)) 927 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 928 929 /* 930 * Process options only when we get SYN/ACK back. The SYN case 931 * for incoming connections is handled in tcp_syncache. 932 * XXX this is traditional behavior, may need to be cleaned up. 933 */ 934 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 935 if (thflags & TH_SYN) { 936 if (to.to_flags & TOF_SCALE) { 937 tp->t_flags |= TF_RCVD_SCALE; 938 tp->requested_s_scale = to.to_requested_s_scale; 939 } 940 if (to.to_flags & TOF_TS) { 941 tp->t_flags |= TF_RCVD_TSTMP; 942 tp->ts_recent = to.to_tsval; 943 tp->ts_recent_age = ticks; 944 } 945 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 946 tp->t_flags |= TF_RCVD_CC; 947 if (to.to_flags & TOF_MSS) 948 tcp_mss(tp, to.to_mss); 949 } 950 951 /* 952 * Header prediction: check for the two common cases 953 * of a uni-directional data xfer. If the packet has 954 * no control flags, is in-sequence, the window didn't 955 * change and we're not retransmitting, it's a 956 * candidate. If the length is zero and the ack moved 957 * forward, we're the sender side of the xfer. Just 958 * free the data acked & wake any higher level process 959 * that was blocked waiting for space. If the length 960 * is non-zero and the ack didn't move, we're the 961 * receiver side. If we're getting packets in-order 962 * (the reassembly queue is empty), add the data to 963 * the socket buffer and note that we need a delayed ack. 964 * Make sure that the hidden state-flags are also off. 965 * Since we check for TCPS_ESTABLISHED above, it can only 966 * be TH_NEEDSYN. 967 */ 968 if (tp->t_state == TCPS_ESTABLISHED && 969 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 970 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 971 ((to.to_flags & TOF_TS) == 0 || 972 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 973 /* 974 * Using the CC option is compulsory if once started: 975 * the segment is OK if no T/TCP was negotiated or 976 * if the segment has a CC option equal to CCrecv 977 */ 978 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 979 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 980 th->th_seq == tp->rcv_nxt && 981 tiwin && tiwin == tp->snd_wnd && 982 tp->snd_nxt == tp->snd_max) { 983 984 /* 985 * If last ACK falls within this segment's sequence numbers, 986 * record the timestamp. 987 * NOTE that the test is modified according to the latest 988 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 989 */ 990 if ((to.to_flags & TOF_TS) != 0 && 991 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 992 tp->ts_recent_age = ticks; 993 tp->ts_recent = to.to_tsval; 994 } 995 996 if (tlen == 0) { 997 if (SEQ_GT(th->th_ack, tp->snd_una) && 998 SEQ_LEQ(th->th_ack, tp->snd_max) && 999 tp->snd_cwnd >= tp->snd_wnd && 1000 ((!tcp_do_newreno && 1001 tp->t_dupacks < tcprexmtthresh) || 1002 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1003 KASSERT(headlocked, ("headlocked")); 1004 INP_INFO_WUNLOCK(&tcbinfo); 1005 /* 1006 * this is a pure ack for outstanding data. 1007 */ 1008 ++tcpstat.tcps_predack; 1009 /* 1010 * "bad retransmit" recovery 1011 */ 1012 if (tp->t_rxtshift == 1 && 1013 ticks < tp->t_badrxtwin) { 1014 ++tcpstat.tcps_sndrexmitbad; 1015 tp->snd_cwnd = tp->snd_cwnd_prev; 1016 tp->snd_ssthresh = 1017 tp->snd_ssthresh_prev; 1018 tp->snd_recover = tp->snd_recover_prev; 1019 if (tp->t_flags & TF_WASFRECOVERY) 1020 ENTER_FASTRECOVERY(tp); 1021 tp->snd_nxt = tp->snd_max; 1022 tp->t_badrxtwin = 0; 1023 } 1024 1025 /* 1026 * Recalculate the transmit timer / rtt. 1027 * 1028 * Some boxes send broken timestamp replies 1029 * during the SYN+ACK phase, ignore 1030 * timestamps of 0 or we could calculate a 1031 * huge RTT and blow up the retransmit timer. 1032 */ 1033 if ((to.to_flags & TOF_TS) != 0 && 1034 to.to_tsecr) { 1035 tcp_xmit_timer(tp, 1036 ticks - to.to_tsecr + 1); 1037 } else if (tp->t_rtttime && 1038 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1039 tcp_xmit_timer(tp, 1040 ticks - tp->t_rtttime); 1041 } 1042 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1043 acked = th->th_ack - tp->snd_una; 1044 tcpstat.tcps_rcvackpack++; 1045 tcpstat.tcps_rcvackbyte += acked; 1046 sbdrop(&so->so_snd, acked); 1047 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1048 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1049 tp->snd_recover = th->th_ack - 1; 1050 tp->snd_una = th->th_ack; 1051 /* 1052 * pull snd_wl2 up to prevent seq wrap relative 1053 * to th_ack. 1054 */ 1055 tp->snd_wl2 = th->th_ack; 1056 tp->t_dupacks = 0; 1057 m_freem(m); 1058 ND6_HINT(tp); /* some progress has been done */ 1059 1060 /* 1061 * If all outstanding data are acked, stop 1062 * retransmit timer, otherwise restart timer 1063 * using current (possibly backed-off) value. 1064 * If process is waiting for space, 1065 * wakeup/selwakeup/signal. If data 1066 * are ready to send, let tcp_output 1067 * decide between more output or persist. 1068 1069 #ifdef TCPDEBUG 1070 if (so->so_options & SO_DEBUG) 1071 tcp_trace(TA_INPUT, ostate, tp, 1072 (void *)tcp_saveipgen, 1073 &tcp_savetcp, 0); 1074 #endif 1075 */ 1076 if (tp->snd_una == tp->snd_max) 1077 callout_stop(tp->tt_rexmt); 1078 else if (!callout_active(tp->tt_persist)) 1079 callout_reset(tp->tt_rexmt, 1080 tp->t_rxtcur, 1081 tcp_timer_rexmt, tp); 1082 1083 sowwakeup(so); 1084 if (so->so_snd.sb_cc) 1085 (void) tcp_output(tp); 1086 goto check_delack; 1087 } 1088 } else if (th->th_ack == tp->snd_una && 1089 LIST_EMPTY(&tp->t_segq) && 1090 tlen <= sbspace(&so->so_rcv)) { 1091 KASSERT(headlocked, ("headlocked")); 1092 INP_INFO_WUNLOCK(&tcbinfo); 1093 /* 1094 * this is a pure, in-sequence data packet 1095 * with nothing on the reassembly queue and 1096 * we have enough buffer space to take it. 1097 */ 1098 ++tcpstat.tcps_preddat; 1099 tp->rcv_nxt += tlen; 1100 /* 1101 * Pull snd_wl1 up to prevent seq wrap relative to 1102 * th_seq. 1103 */ 1104 tp->snd_wl1 = th->th_seq; 1105 /* 1106 * Pull rcv_up up to prevent seq wrap relative to 1107 * rcv_nxt. 1108 */ 1109 tp->rcv_up = tp->rcv_nxt; 1110 tcpstat.tcps_rcvpack++; 1111 tcpstat.tcps_rcvbyte += tlen; 1112 ND6_HINT(tp); /* some progress has been done */ 1113 /* 1114 #ifdef TCPDEBUG 1115 if (so->so_options & SO_DEBUG) 1116 tcp_trace(TA_INPUT, ostate, tp, 1117 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1118 #endif 1119 * Add data to socket buffer. 1120 */ 1121 if (so->so_state & SS_CANTRCVMORE) { 1122 m_freem(m); 1123 } else { 1124 m_adj(m, drop_hdrlen); /* delayed header drop */ 1125 sbappendstream(&so->so_rcv, m); 1126 } 1127 sorwakeup(so); 1128 if (DELAY_ACK(tp)) { 1129 tp->t_flags |= TF_DELACK; 1130 } else { 1131 tp->t_flags |= TF_ACKNOW; 1132 tcp_output(tp); 1133 } 1134 goto check_delack; 1135 } 1136 } 1137 1138 /* 1139 * Calculate amount of space in receive window, 1140 * and then do TCP input processing. 1141 * Receive window is amount of space in rcv queue, 1142 * but not less than advertised window. 1143 */ 1144 { int win; 1145 1146 win = sbspace(&so->so_rcv); 1147 if (win < 0) 1148 win = 0; 1149 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1150 } 1151 1152 switch (tp->t_state) { 1153 1154 /* 1155 * If the state is SYN_RECEIVED: 1156 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1157 */ 1158 case TCPS_SYN_RECEIVED: 1159 if ((thflags & TH_ACK) && 1160 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1161 SEQ_GT(th->th_ack, tp->snd_max))) { 1162 rstreason = BANDLIM_RST_OPENPORT; 1163 goto dropwithreset; 1164 } 1165 break; 1166 1167 /* 1168 * If the state is SYN_SENT: 1169 * if seg contains an ACK, but not for our SYN, drop the input. 1170 * if seg contains a RST, then drop the connection. 1171 * if seg does not contain SYN, then drop it. 1172 * Otherwise this is an acceptable SYN segment 1173 * initialize tp->rcv_nxt and tp->irs 1174 * if seg contains ack then advance tp->snd_una 1175 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1176 * arrange for segment to be acked (eventually) 1177 * continue processing rest of data/controls, beginning with URG 1178 */ 1179 case TCPS_SYN_SENT: 1180 if (tcp_do_rfc1644) 1181 tcp_hc_gettao(&inp->inp_inc, &tao); 1182 1183 if ((thflags & TH_ACK) && 1184 (SEQ_LEQ(th->th_ack, tp->iss) || 1185 SEQ_GT(th->th_ack, tp->snd_max))) { 1186 /* 1187 * If we have a cached CCsent for the remote host, 1188 * hence we haven't just crashed and restarted, 1189 * do not send a RST. This may be a retransmission 1190 * from the other side after our earlier ACK was lost. 1191 * Our new SYN, when it arrives, will serve as the 1192 * needed ACK. 1193 */ 1194 if (tao.tao_ccsent != 0) 1195 goto drop; 1196 else { 1197 rstreason = BANDLIM_UNLIMITED; 1198 goto dropwithreset; 1199 } 1200 } 1201 if (thflags & TH_RST) { 1202 if (thflags & TH_ACK) 1203 tp = tcp_drop(tp, ECONNREFUSED); 1204 goto drop; 1205 } 1206 if ((thflags & TH_SYN) == 0) 1207 goto drop; 1208 tp->snd_wnd = th->th_win; /* initial send window */ 1209 tp->cc_recv = to.to_cc; /* foreign CC */ 1210 1211 tp->irs = th->th_seq; 1212 tcp_rcvseqinit(tp); 1213 if (thflags & TH_ACK) { 1214 /* 1215 * Our SYN was acked. If segment contains CC.ECHO 1216 * option, check it to make sure this segment really 1217 * matches our SYN. If not, just drop it as old 1218 * duplicate, but send an RST if we're still playing 1219 * by the old rules. If no CC.ECHO option, make sure 1220 * we don't get fooled into using T/TCP. 1221 */ 1222 if (to.to_flags & TOF_CCECHO) { 1223 if (tp->cc_send != to.to_ccecho) { 1224 if (tao.tao_ccsent != 0) 1225 goto drop; 1226 else { 1227 rstreason = BANDLIM_UNLIMITED; 1228 goto dropwithreset; 1229 } 1230 } 1231 } else 1232 tp->t_flags &= ~TF_RCVD_CC; 1233 tcpstat.tcps_connects++; 1234 soisconnected(so); 1235 #ifdef MAC 1236 mac_set_socket_peer_from_mbuf(m, so); 1237 #endif 1238 /* Do window scaling on this connection? */ 1239 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1240 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1241 tp->snd_scale = tp->requested_s_scale; 1242 tp->rcv_scale = tp->request_r_scale; 1243 } 1244 /* Segment is acceptable, update cache if undefined. */ 1245 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1246 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1247 1248 tp->rcv_adv += tp->rcv_wnd; 1249 tp->snd_una++; /* SYN is acked */ 1250 /* 1251 * If there's data, delay ACK; if there's also a FIN 1252 * ACKNOW will be turned on later. 1253 */ 1254 if (DELAY_ACK(tp) && tlen != 0) 1255 callout_reset(tp->tt_delack, tcp_delacktime, 1256 tcp_timer_delack, tp); 1257 else 1258 tp->t_flags |= TF_ACKNOW; 1259 /* 1260 * Received <SYN,ACK> in SYN_SENT[*] state. 1261 * Transitions: 1262 * SYN_SENT --> ESTABLISHED 1263 * SYN_SENT* --> FIN_WAIT_1 1264 */ 1265 tp->t_starttime = ticks; 1266 if (tp->t_flags & TF_NEEDFIN) { 1267 tp->t_state = TCPS_FIN_WAIT_1; 1268 tp->t_flags &= ~TF_NEEDFIN; 1269 thflags &= ~TH_SYN; 1270 } else { 1271 tp->t_state = TCPS_ESTABLISHED; 1272 callout_reset(tp->tt_keep, tcp_keepidle, 1273 tcp_timer_keep, tp); 1274 } 1275 } else { 1276 /* 1277 * Received initial SYN in SYN-SENT[*] state => 1278 * simultaneous open. If segment contains CC option 1279 * and there is a cached CC, apply TAO test. 1280 * If it succeeds, connection is * half-synchronized. 1281 * Otherwise, do 3-way handshake: 1282 * SYN-SENT -> SYN-RECEIVED 1283 * SYN-SENT* -> SYN-RECEIVED* 1284 * If there was no CC option, clear cached CC value. 1285 */ 1286 tp->t_flags |= TF_ACKNOW; 1287 callout_stop(tp->tt_rexmt); 1288 if (to.to_flags & TOF_CC) { 1289 if (tao.tao_cc != 0 && 1290 CC_GT(to.to_cc, tao.tao_cc)) { 1291 /* 1292 * update cache and make transition: 1293 * SYN-SENT -> ESTABLISHED* 1294 * SYN-SENT* -> FIN-WAIT-1* 1295 */ 1296 tao.tao_cc = to.to_cc; 1297 tcp_hc_updatetao(&inp->inp_inc, 1298 TCP_HC_TAO_CC, to.to_cc, 0); 1299 tp->t_starttime = ticks; 1300 if (tp->t_flags & TF_NEEDFIN) { 1301 tp->t_state = TCPS_FIN_WAIT_1; 1302 tp->t_flags &= ~TF_NEEDFIN; 1303 } else { 1304 tp->t_state = TCPS_ESTABLISHED; 1305 callout_reset(tp->tt_keep, 1306 tcp_keepidle, 1307 tcp_timer_keep, 1308 tp); 1309 } 1310 tp->t_flags |= TF_NEEDSYN; 1311 } else 1312 tp->t_state = TCPS_SYN_RECEIVED; 1313 } else { 1314 if (tcp_do_rfc1644) { 1315 /* CC.NEW or no option => invalidate cache */ 1316 tao.tao_cc = 0; 1317 tcp_hc_updatetao(&inp->inp_inc, 1318 TCP_HC_TAO_CC, to.to_cc, 0); 1319 } 1320 tp->t_state = TCPS_SYN_RECEIVED; 1321 } 1322 } 1323 1324 trimthenstep6: 1325 /* 1326 * Advance th->th_seq to correspond to first data byte. 1327 * If data, trim to stay within window, 1328 * dropping FIN if necessary. 1329 */ 1330 th->th_seq++; 1331 if (tlen > tp->rcv_wnd) { 1332 todrop = tlen - tp->rcv_wnd; 1333 m_adj(m, -todrop); 1334 tlen = tp->rcv_wnd; 1335 thflags &= ~TH_FIN; 1336 tcpstat.tcps_rcvpackafterwin++; 1337 tcpstat.tcps_rcvbyteafterwin += todrop; 1338 } 1339 tp->snd_wl1 = th->th_seq - 1; 1340 tp->rcv_up = th->th_seq; 1341 /* 1342 * Client side of transaction: already sent SYN and data. 1343 * If the remote host used T/TCP to validate the SYN, 1344 * our data will be ACK'd; if so, enter normal data segment 1345 * processing in the middle of step 5, ack processing. 1346 * Otherwise, goto step 6. 1347 */ 1348 if (thflags & TH_ACK) 1349 goto process_ACK; 1350 1351 goto step6; 1352 1353 /* 1354 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1355 * if segment contains a SYN and CC [not CC.NEW] option: 1356 * if state == TIME_WAIT and connection duration > MSL, 1357 * drop packet and send RST; 1358 * 1359 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1360 * ack the FIN (and data) in retransmission queue. 1361 * Complete close and delete TCPCB. Then reprocess 1362 * segment, hoping to find new TCPCB in LISTEN state; 1363 * 1364 * else must be old SYN; drop it. 1365 * else do normal processing. 1366 */ 1367 case TCPS_LAST_ACK: 1368 case TCPS_CLOSING: 1369 case TCPS_TIME_WAIT: 1370 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1371 if ((thflags & TH_SYN) && 1372 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1373 if (tp->t_state == TCPS_TIME_WAIT && 1374 (ticks - tp->t_starttime) > tcp_msl) { 1375 rstreason = BANDLIM_UNLIMITED; 1376 goto dropwithreset; 1377 } 1378 if (CC_GT(to.to_cc, tp->cc_recv)) { 1379 tp = tcp_close(tp); 1380 goto findpcb; 1381 } 1382 else 1383 goto drop; 1384 } 1385 break; /* continue normal processing */ 1386 } 1387 1388 /* 1389 * States other than LISTEN or SYN_SENT. 1390 * First check the RST flag and sequence number since reset segments 1391 * are exempt from the timestamp and connection count tests. This 1392 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1393 * below which allowed reset segments in half the sequence space 1394 * to fall though and be processed (which gives forged reset 1395 * segments with a random sequence number a 50 percent chance of 1396 * killing a connection). 1397 * Then check timestamp, if present. 1398 * Then check the connection count, if present. 1399 * Then check that at least some bytes of segment are within 1400 * receive window. If segment begins before rcv_nxt, 1401 * drop leading data (and SYN); if nothing left, just ack. 1402 * 1403 * 1404 * If the RST bit is set, check the sequence number to see 1405 * if this is a valid reset segment. 1406 * RFC 793 page 37: 1407 * In all states except SYN-SENT, all reset (RST) segments 1408 * are validated by checking their SEQ-fields. A reset is 1409 * valid if its sequence number is in the window. 1410 * Note: this does not take into account delayed ACKs, so 1411 * we should test against last_ack_sent instead of rcv_nxt. 1412 * The sequence number in the reset segment is normally an 1413 * echo of our outgoing acknowlegement numbers, but some hosts 1414 * send a reset with the sequence number at the rightmost edge 1415 * of our receive window, and we have to handle this case. 1416 * If we have multiple segments in flight, the intial reset 1417 * segment sequence numbers will be to the left of last_ack_sent, 1418 * but they will eventually catch up. 1419 * In any case, it never made sense to trim reset segments to 1420 * fit the receive window since RFC 1122 says: 1421 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1422 * 1423 * A TCP SHOULD allow a received RST segment to include data. 1424 * 1425 * DISCUSSION 1426 * It has been suggested that a RST segment could contain 1427 * ASCII text that encoded and explained the cause of the 1428 * RST. No standard has yet been established for such 1429 * data. 1430 * 1431 * If the reset segment passes the sequence number test examine 1432 * the state: 1433 * SYN_RECEIVED STATE: 1434 * If passive open, return to LISTEN state. 1435 * If active open, inform user that connection was refused. 1436 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1437 * Inform user that connection was reset, and close tcb. 1438 * CLOSING, LAST_ACK STATES: 1439 * Close the tcb. 1440 * TIME_WAIT STATE: 1441 * Drop the segment - see Stevens, vol. 2, p. 964 and 1442 * RFC 1337. 1443 */ 1444 if (thflags & TH_RST) { 1445 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1446 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1447 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 break; 1463 1464 case TCPS_CLOSING: 1465 case TCPS_LAST_ACK: 1466 tp = tcp_close(tp); 1467 break; 1468 1469 case TCPS_TIME_WAIT: 1470 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1471 ("timewait")); 1472 break; 1473 } 1474 } 1475 goto drop; 1476 } 1477 1478 /* 1479 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1480 * and it's less than ts_recent, drop it. 1481 */ 1482 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1483 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1484 1485 /* Check to see if ts_recent is over 24 days old. */ 1486 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1487 /* 1488 * Invalidate ts_recent. If this segment updates 1489 * ts_recent, the age will be reset later and ts_recent 1490 * will get a valid value. If it does not, setting 1491 * ts_recent to zero will at least satisfy the 1492 * requirement that zero be placed in the timestamp 1493 * echo reply when ts_recent isn't valid. The 1494 * age isn't reset until we get a valid ts_recent 1495 * because we don't want out-of-order segments to be 1496 * dropped when ts_recent is old. 1497 */ 1498 tp->ts_recent = 0; 1499 } else { 1500 tcpstat.tcps_rcvduppack++; 1501 tcpstat.tcps_rcvdupbyte += tlen; 1502 tcpstat.tcps_pawsdrop++; 1503 if (tlen) 1504 goto dropafterack; 1505 goto drop; 1506 } 1507 } 1508 1509 /* 1510 * T/TCP mechanism 1511 * If T/TCP was negotiated and the segment doesn't have CC, 1512 * or if its CC is wrong then drop the segment. 1513 * RST segments do not have to comply with this. 1514 */ 1515 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1516 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1517 goto dropafterack; 1518 1519 /* 1520 * In the SYN-RECEIVED state, validate that the packet belongs to 1521 * this connection before trimming the data to fit the receive 1522 * window. Check the sequence number versus IRS since we know 1523 * the sequence numbers haven't wrapped. This is a partial fix 1524 * for the "LAND" DoS attack. 1525 */ 1526 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1527 rstreason = BANDLIM_RST_OPENPORT; 1528 goto dropwithreset; 1529 } 1530 1531 todrop = tp->rcv_nxt - th->th_seq; 1532 if (todrop > 0) { 1533 if (thflags & TH_SYN) { 1534 thflags &= ~TH_SYN; 1535 th->th_seq++; 1536 if (th->th_urp > 1) 1537 th->th_urp--; 1538 else 1539 thflags &= ~TH_URG; 1540 todrop--; 1541 } 1542 /* 1543 * Following if statement from Stevens, vol. 2, p. 960. 1544 */ 1545 if (todrop > tlen 1546 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1547 /* 1548 * Any valid FIN must be to the left of the window. 1549 * At this point the FIN must be a duplicate or out 1550 * of sequence; drop it. 1551 */ 1552 thflags &= ~TH_FIN; 1553 1554 /* 1555 * Send an ACK to resynchronize and drop any data. 1556 * But keep on processing for RST or ACK. 1557 */ 1558 tp->t_flags |= TF_ACKNOW; 1559 todrop = tlen; 1560 tcpstat.tcps_rcvduppack++; 1561 tcpstat.tcps_rcvdupbyte += todrop; 1562 } else { 1563 tcpstat.tcps_rcvpartduppack++; 1564 tcpstat.tcps_rcvpartdupbyte += todrop; 1565 } 1566 drop_hdrlen += todrop; /* drop from the top afterwards */ 1567 th->th_seq += todrop; 1568 tlen -= todrop; 1569 if (th->th_urp > todrop) 1570 th->th_urp -= todrop; 1571 else { 1572 thflags &= ~TH_URG; 1573 th->th_urp = 0; 1574 } 1575 } 1576 1577 /* 1578 * If new data are received on a connection after the 1579 * user processes are gone, then RST the other end. 1580 */ 1581 if ((so->so_state & SS_NOFDREF) && 1582 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1583 tp = tcp_close(tp); 1584 tcpstat.tcps_rcvafterclose++; 1585 rstreason = BANDLIM_UNLIMITED; 1586 goto dropwithreset; 1587 } 1588 1589 /* 1590 * If segment ends after window, drop trailing data 1591 * (and PUSH and FIN); if nothing left, just ACK. 1592 */ 1593 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1594 if (todrop > 0) { 1595 tcpstat.tcps_rcvpackafterwin++; 1596 if (todrop >= tlen) { 1597 tcpstat.tcps_rcvbyteafterwin += tlen; 1598 /* 1599 * If a new connection request is received 1600 * while in TIME_WAIT, drop the old connection 1601 * and start over if the sequence numbers 1602 * are above the previous ones. 1603 */ 1604 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1605 if (thflags & TH_SYN && 1606 tp->t_state == TCPS_TIME_WAIT && 1607 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1608 tp = tcp_close(tp); 1609 goto findpcb; 1610 } 1611 /* 1612 * If window is closed can only take segments at 1613 * window edge, and have to drop data and PUSH from 1614 * incoming segments. Continue processing, but 1615 * remember to ack. Otherwise, drop segment 1616 * and ack. 1617 */ 1618 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1619 tp->t_flags |= TF_ACKNOW; 1620 tcpstat.tcps_rcvwinprobe++; 1621 } else 1622 goto dropafterack; 1623 } else 1624 tcpstat.tcps_rcvbyteafterwin += todrop; 1625 m_adj(m, -todrop); 1626 tlen -= todrop; 1627 thflags &= ~(TH_PUSH|TH_FIN); 1628 } 1629 1630 /* 1631 * If last ACK falls within this segment's sequence numbers, 1632 * record its timestamp. 1633 * NOTE that the test is modified according to the latest 1634 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1635 */ 1636 if ((to.to_flags & TOF_TS) != 0 && 1637 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1638 tp->ts_recent_age = ticks; 1639 tp->ts_recent = to.to_tsval; 1640 } 1641 1642 /* 1643 * If a SYN is in the window, then this is an 1644 * error and we send an RST and drop the connection. 1645 */ 1646 if (thflags & TH_SYN) { 1647 tp = tcp_drop(tp, ECONNRESET); 1648 rstreason = BANDLIM_UNLIMITED; 1649 goto drop; 1650 } 1651 1652 /* 1653 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1654 * flag is on (half-synchronized state), then queue data for 1655 * later processing; else drop segment and return. 1656 */ 1657 if ((thflags & TH_ACK) == 0) { 1658 if (tp->t_state == TCPS_SYN_RECEIVED || 1659 (tp->t_flags & TF_NEEDSYN)) 1660 goto step6; 1661 else 1662 goto drop; 1663 } 1664 1665 /* 1666 * Ack processing. 1667 */ 1668 switch (tp->t_state) { 1669 1670 /* 1671 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1672 * ESTABLISHED state and continue processing. 1673 * The ACK was checked above. 1674 */ 1675 case TCPS_SYN_RECEIVED: 1676 1677 tcpstat.tcps_connects++; 1678 soisconnected(so); 1679 /* Do window scaling? */ 1680 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1681 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1682 tp->snd_scale = tp->requested_s_scale; 1683 tp->rcv_scale = tp->request_r_scale; 1684 } 1685 /* 1686 * Upon successful completion of 3-way handshake, 1687 * update cache.CC, pass any queued data to the user, 1688 * and advance state appropriately. 1689 */ 1690 if (tcp_do_rfc1644) { 1691 tao.tao_cc = tp->cc_recv; 1692 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1693 tp->cc_recv, 0); 1694 } 1695 /* 1696 * Make transitions: 1697 * SYN-RECEIVED -> ESTABLISHED 1698 * SYN-RECEIVED* -> FIN-WAIT-1 1699 */ 1700 tp->t_starttime = ticks; 1701 if (tp->t_flags & TF_NEEDFIN) { 1702 tp->t_state = TCPS_FIN_WAIT_1; 1703 tp->t_flags &= ~TF_NEEDFIN; 1704 } else { 1705 tp->t_state = TCPS_ESTABLISHED; 1706 callout_reset(tp->tt_keep, tcp_keepidle, 1707 tcp_timer_keep, tp); 1708 } 1709 /* 1710 * If segment contains data or ACK, will call tcp_reass() 1711 * later; if not, do so now to pass queued data to user. 1712 */ 1713 if (tlen == 0 && (thflags & TH_FIN) == 0) 1714 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1715 (struct mbuf *)0); 1716 tp->snd_wl1 = th->th_seq - 1; 1717 /* FALLTHROUGH */ 1718 1719 /* 1720 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1721 * ACKs. If the ack is in the range 1722 * tp->snd_una < th->th_ack <= tp->snd_max 1723 * then advance tp->snd_una to th->th_ack and drop 1724 * data from the retransmission queue. If this ACK reflects 1725 * more up to date window information we update our window information. 1726 */ 1727 case TCPS_ESTABLISHED: 1728 case TCPS_FIN_WAIT_1: 1729 case TCPS_FIN_WAIT_2: 1730 case TCPS_CLOSE_WAIT: 1731 case TCPS_CLOSING: 1732 case TCPS_LAST_ACK: 1733 case TCPS_TIME_WAIT: 1734 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1735 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1736 if (tlen == 0 && tiwin == tp->snd_wnd) { 1737 tcpstat.tcps_rcvdupack++; 1738 /* 1739 * If we have outstanding data (other than 1740 * a window probe), this is a completely 1741 * duplicate ack (ie, window info didn't 1742 * change), the ack is the biggest we've 1743 * seen and we've seen exactly our rexmt 1744 * threshhold of them, assume a packet 1745 * has been dropped and retransmit it. 1746 * Kludge snd_nxt & the congestion 1747 * window so we send only this one 1748 * packet. 1749 * 1750 * We know we're losing at the current 1751 * window size so do congestion avoidance 1752 * (set ssthresh to half the current window 1753 * and pull our congestion window back to 1754 * the new ssthresh). 1755 * 1756 * Dup acks mean that packets have left the 1757 * network (they're now cached at the receiver) 1758 * so bump cwnd by the amount in the receiver 1759 * to keep a constant cwnd packets in the 1760 * network. 1761 */ 1762 if (!callout_active(tp->tt_rexmt) || 1763 th->th_ack != tp->snd_una) 1764 tp->t_dupacks = 0; 1765 else if (++tp->t_dupacks > tcprexmtthresh || 1766 (tcp_do_newreno && 1767 IN_FASTRECOVERY(tp))) { 1768 tp->snd_cwnd += tp->t_maxseg; 1769 (void) tcp_output(tp); 1770 goto drop; 1771 } else if (tp->t_dupacks == tcprexmtthresh) { 1772 tcp_seq onxt = tp->snd_nxt; 1773 u_int win; 1774 if (tcp_do_newreno && 1775 SEQ_LEQ(th->th_ack, 1776 tp->snd_recover)) { 1777 tp->t_dupacks = 0; 1778 break; 1779 } 1780 win = min(tp->snd_wnd, tp->snd_cwnd) / 1781 2 / tp->t_maxseg; 1782 if (win < 2) 1783 win = 2; 1784 tp->snd_ssthresh = win * tp->t_maxseg; 1785 ENTER_FASTRECOVERY(tp); 1786 tp->snd_recover = tp->snd_max; 1787 callout_stop(tp->tt_rexmt); 1788 tp->t_rtttime = 0; 1789 tp->snd_nxt = th->th_ack; 1790 tp->snd_cwnd = tp->t_maxseg; 1791 (void) tcp_output(tp); 1792 KASSERT(tp->snd_limited <= 2, 1793 ("tp->snd_limited too big")); 1794 tp->snd_cwnd = tp->snd_ssthresh + 1795 tp->t_maxseg * 1796 (tp->t_dupacks - tp->snd_limited); 1797 if (SEQ_GT(onxt, tp->snd_nxt)) 1798 tp->snd_nxt = onxt; 1799 goto drop; 1800 } else if (tcp_do_rfc3042) { 1801 u_long oldcwnd = tp->snd_cwnd; 1802 tcp_seq oldsndmax = tp->snd_max; 1803 u_int sent; 1804 KASSERT(tp->t_dupacks == 1 || 1805 tp->t_dupacks == 2, 1806 ("dupacks not 1 or 2")); 1807 if (tp->t_dupacks == 1) { 1808 tp->snd_limited = 0; 1809 tp->snd_cwnd += tp->t_maxseg; 1810 } else { 1811 tp->snd_cwnd += 1812 tp->t_maxseg * 2; 1813 } 1814 (void) tcp_output(tp); 1815 sent = tp->snd_max - oldsndmax; 1816 if (sent > tp->t_maxseg) { 1817 KASSERT(tp->snd_limited == 0 && 1818 tp->t_dupacks == 2, 1819 ("sent too much")); 1820 tp->snd_limited = 2; 1821 } else if (sent > 0) 1822 ++tp->snd_limited; 1823 tp->snd_cwnd = oldcwnd; 1824 goto drop; 1825 } 1826 } else 1827 tp->t_dupacks = 0; 1828 break; 1829 } 1830 1831 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1832 1833 /* 1834 * If the congestion window was inflated to account 1835 * for the other side's cached packets, retract it. 1836 */ 1837 if (tcp_do_newreno) { 1838 if (IN_FASTRECOVERY(tp)) { 1839 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1840 tcp_newreno_partial_ack(tp, th); 1841 } else { 1842 /* 1843 * Window inflation should have left us 1844 * with approximately snd_ssthresh 1845 * outstanding data. 1846 * But in case we would be inclined to 1847 * send a burst, better to do it via 1848 * the slow start mechanism. 1849 */ 1850 if (SEQ_GT(th->th_ack + 1851 tp->snd_ssthresh, 1852 tp->snd_max)) 1853 tp->snd_cwnd = tp->snd_max - 1854 th->th_ack + 1855 tp->t_maxseg; 1856 else 1857 tp->snd_cwnd = tp->snd_ssthresh; 1858 } 1859 } 1860 } else { 1861 if (tp->t_dupacks >= tcprexmtthresh && 1862 tp->snd_cwnd > tp->snd_ssthresh) 1863 tp->snd_cwnd = tp->snd_ssthresh; 1864 } 1865 tp->t_dupacks = 0; 1866 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1867 tcpstat.tcps_rcvacktoomuch++; 1868 goto dropafterack; 1869 } 1870 /* 1871 * If we reach this point, ACK is not a duplicate, 1872 * i.e., it ACKs something we sent. 1873 */ 1874 if (tp->t_flags & TF_NEEDSYN) { 1875 /* 1876 * T/TCP: Connection was half-synchronized, and our 1877 * SYN has been ACK'd (so connection is now fully 1878 * synchronized). Go to non-starred state, 1879 * increment snd_una for ACK of SYN, and check if 1880 * we can do window scaling. 1881 */ 1882 tp->t_flags &= ~TF_NEEDSYN; 1883 tp->snd_una++; 1884 /* Do window scaling? */ 1885 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1886 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1887 tp->snd_scale = tp->requested_s_scale; 1888 tp->rcv_scale = tp->request_r_scale; 1889 } 1890 } 1891 1892 process_ACK: 1893 acked = th->th_ack - tp->snd_una; 1894 tcpstat.tcps_rcvackpack++; 1895 tcpstat.tcps_rcvackbyte += acked; 1896 1897 /* 1898 * If we just performed our first retransmit, and the ACK 1899 * arrives within our recovery window, then it was a mistake 1900 * to do the retransmit in the first place. Recover our 1901 * original cwnd and ssthresh, and proceed to transmit where 1902 * we left off. 1903 */ 1904 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1905 ++tcpstat.tcps_sndrexmitbad; 1906 tp->snd_cwnd = tp->snd_cwnd_prev; 1907 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1908 tp->snd_recover = tp->snd_recover_prev; 1909 if (tp->t_flags & TF_WASFRECOVERY) 1910 ENTER_FASTRECOVERY(tp); 1911 tp->snd_nxt = tp->snd_max; 1912 tp->t_badrxtwin = 0; /* XXX probably not required */ 1913 } 1914 1915 /* 1916 * If we have a timestamp reply, update smoothed 1917 * round trip time. If no timestamp is present but 1918 * transmit timer is running and timed sequence 1919 * number was acked, update smoothed round trip time. 1920 * Since we now have an rtt measurement, cancel the 1921 * timer backoff (cf., Phil Karn's retransmit alg.). 1922 * Recompute the initial retransmit timer. 1923 * 1924 * Some boxes send broken timestamp replies 1925 * during the SYN+ACK phase, ignore 1926 * timestamps of 0 or we could calculate a 1927 * huge RTT and blow up the retransmit timer. 1928 */ 1929 if ((to.to_flags & TOF_TS) != 0 && 1930 to.to_tsecr) { 1931 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1932 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1933 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1934 } 1935 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1936 1937 /* 1938 * If all outstanding data is acked, stop retransmit 1939 * timer and remember to restart (more output or persist). 1940 * If there is more data to be acked, restart retransmit 1941 * timer, using current (possibly backed-off) value. 1942 */ 1943 if (th->th_ack == tp->snd_max) { 1944 callout_stop(tp->tt_rexmt); 1945 needoutput = 1; 1946 } else if (!callout_active(tp->tt_persist)) 1947 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1948 tcp_timer_rexmt, tp); 1949 1950 /* 1951 * If no data (only SYN) was ACK'd, 1952 * skip rest of ACK processing. 1953 */ 1954 if (acked == 0) 1955 goto step6; 1956 1957 /* 1958 * When new data is acked, open the congestion window. 1959 * If the window gives us less than ssthresh packets 1960 * in flight, open exponentially (maxseg per packet). 1961 * Otherwise open linearly: maxseg per window 1962 * (maxseg^2 / cwnd per packet). 1963 */ 1964 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 1965 register u_int cw = tp->snd_cwnd; 1966 register u_int incr = tp->t_maxseg; 1967 if (cw > tp->snd_ssthresh) 1968 incr = incr * incr / cw; 1969 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1970 } 1971 if (acked > so->so_snd.sb_cc) { 1972 tp->snd_wnd -= so->so_snd.sb_cc; 1973 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1974 ourfinisacked = 1; 1975 } else { 1976 sbdrop(&so->so_snd, acked); 1977 tp->snd_wnd -= acked; 1978 ourfinisacked = 0; 1979 } 1980 sowwakeup(so); 1981 /* detect una wraparound */ 1982 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 1983 SEQ_GT(tp->snd_una, tp->snd_recover) && 1984 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1985 tp->snd_recover = th->th_ack - 1; 1986 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 1987 SEQ_GEQ(th->th_ack, tp->snd_recover)) 1988 EXIT_FASTRECOVERY(tp); 1989 tp->snd_una = th->th_ack; 1990 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1991 tp->snd_nxt = tp->snd_una; 1992 1993 switch (tp->t_state) { 1994 1995 /* 1996 * In FIN_WAIT_1 STATE in addition to the processing 1997 * for the ESTABLISHED state if our FIN is now acknowledged 1998 * then enter FIN_WAIT_2. 1999 */ 2000 case TCPS_FIN_WAIT_1: 2001 if (ourfinisacked) { 2002 /* 2003 * If we can't receive any more 2004 * data, then closing user can proceed. 2005 * Starting the timer is contrary to the 2006 * specification, but if we don't get a FIN 2007 * we'll hang forever. 2008 */ 2009 /* XXXjl 2010 * we should release the tp also, and use a 2011 * compressed state. 2012 */ 2013 if (so->so_state & SS_CANTRCVMORE) { 2014 soisdisconnected(so); 2015 callout_reset(tp->tt_2msl, tcp_maxidle, 2016 tcp_timer_2msl, tp); 2017 } 2018 tp->t_state = TCPS_FIN_WAIT_2; 2019 } 2020 break; 2021 2022 /* 2023 * In CLOSING STATE in addition to the processing for 2024 * the ESTABLISHED state if the ACK acknowledges our FIN 2025 * then enter the TIME-WAIT state, otherwise ignore 2026 * the segment. 2027 */ 2028 case TCPS_CLOSING: 2029 if (ourfinisacked) { 2030 KASSERT(headlocked, ("headlocked")); 2031 tcp_twstart(tp); 2032 INP_INFO_WUNLOCK(&tcbinfo); 2033 m_freem(m); 2034 return; 2035 } 2036 break; 2037 2038 /* 2039 * In LAST_ACK, we may still be waiting for data to drain 2040 * and/or to be acked, as well as for the ack of our FIN. 2041 * If our FIN is now acknowledged, delete the TCB, 2042 * enter the closed state and return. 2043 */ 2044 case TCPS_LAST_ACK: 2045 if (ourfinisacked) { 2046 tp = tcp_close(tp); 2047 goto drop; 2048 } 2049 break; 2050 2051 /* 2052 * In TIME_WAIT state the only thing that should arrive 2053 * is a retransmission of the remote FIN. Acknowledge 2054 * it and restart the finack timer. 2055 */ 2056 case TCPS_TIME_WAIT: 2057 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2058 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2059 tcp_timer_2msl, tp); 2060 goto dropafterack; 2061 } 2062 } 2063 2064 step6: 2065 /* 2066 * Update window information. 2067 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2068 */ 2069 if ((thflags & TH_ACK) && 2070 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2071 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2072 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2073 /* keep track of pure window updates */ 2074 if (tlen == 0 && 2075 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2076 tcpstat.tcps_rcvwinupd++; 2077 tp->snd_wnd = tiwin; 2078 tp->snd_wl1 = th->th_seq; 2079 tp->snd_wl2 = th->th_ack; 2080 if (tp->snd_wnd > tp->max_sndwnd) 2081 tp->max_sndwnd = tp->snd_wnd; 2082 needoutput = 1; 2083 } 2084 2085 /* 2086 * Process segments with URG. 2087 */ 2088 if ((thflags & TH_URG) && th->th_urp && 2089 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2090 /* 2091 * This is a kludge, but if we receive and accept 2092 * random urgent pointers, we'll crash in 2093 * soreceive. It's hard to imagine someone 2094 * actually wanting to send this much urgent data. 2095 */ 2096 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2097 th->th_urp = 0; /* XXX */ 2098 thflags &= ~TH_URG; /* XXX */ 2099 goto dodata; /* XXX */ 2100 } 2101 /* 2102 * If this segment advances the known urgent pointer, 2103 * then mark the data stream. This should not happen 2104 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2105 * a FIN has been received from the remote side. 2106 * In these states we ignore the URG. 2107 * 2108 * According to RFC961 (Assigned Protocols), 2109 * the urgent pointer points to the last octet 2110 * of urgent data. We continue, however, 2111 * to consider it to indicate the first octet 2112 * of data past the urgent section as the original 2113 * spec states (in one of two places). 2114 */ 2115 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2116 tp->rcv_up = th->th_seq + th->th_urp; 2117 so->so_oobmark = so->so_rcv.sb_cc + 2118 (tp->rcv_up - tp->rcv_nxt) - 1; 2119 if (so->so_oobmark == 0) 2120 so->so_state |= SS_RCVATMARK; 2121 sohasoutofband(so); 2122 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2123 } 2124 /* 2125 * Remove out of band data so doesn't get presented to user. 2126 * This can happen independent of advancing the URG pointer, 2127 * but if two URG's are pending at once, some out-of-band 2128 * data may creep in... ick. 2129 */ 2130 if (th->th_urp <= (u_long)tlen && 2131 !(so->so_options & SO_OOBINLINE)) { 2132 /* hdr drop is delayed */ 2133 tcp_pulloutofband(so, th, m, drop_hdrlen); 2134 } 2135 } else { 2136 /* 2137 * If no out of band data is expected, 2138 * pull receive urgent pointer along 2139 * with the receive window. 2140 */ 2141 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2142 tp->rcv_up = tp->rcv_nxt; 2143 } 2144 dodata: /* XXX */ 2145 KASSERT(headlocked, ("headlocked")); 2146 /* 2147 * Process the segment text, merging it into the TCP sequencing queue, 2148 * and arranging for acknowledgment of receipt if necessary. 2149 * This process logically involves adjusting tp->rcv_wnd as data 2150 * is presented to the user (this happens in tcp_usrreq.c, 2151 * case PRU_RCVD). If a FIN has already been received on this 2152 * connection then we just ignore the text. 2153 */ 2154 if ((tlen || (thflags & TH_FIN)) && 2155 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2156 m_adj(m, drop_hdrlen); /* delayed header drop */ 2157 /* 2158 * Insert segment which includes th into TCP reassembly queue 2159 * with control block tp. Set thflags to whether reassembly now 2160 * includes a segment with FIN. This handles the common case 2161 * inline (segment is the next to be received on an established 2162 * connection, and the queue is empty), avoiding linkage into 2163 * and removal from the queue and repetition of various 2164 * conversions. 2165 * Set DELACK for segments received in order, but ack 2166 * immediately when segments are out of order (so 2167 * fast retransmit can work). 2168 */ 2169 if (th->th_seq == tp->rcv_nxt && 2170 LIST_EMPTY(&tp->t_segq) && 2171 TCPS_HAVEESTABLISHED(tp->t_state)) { 2172 if (DELAY_ACK(tp)) 2173 tp->t_flags |= TF_DELACK; 2174 else 2175 tp->t_flags |= TF_ACKNOW; 2176 tp->rcv_nxt += tlen; 2177 thflags = th->th_flags & TH_FIN; 2178 tcpstat.tcps_rcvpack++; 2179 tcpstat.tcps_rcvbyte += tlen; 2180 ND6_HINT(tp); 2181 if (so->so_state & SS_CANTRCVMORE) 2182 m_freem(m); 2183 else 2184 sbappendstream(&so->so_rcv, m); 2185 sorwakeup(so); 2186 } else { 2187 thflags = tcp_reass(tp, th, &tlen, m); 2188 tp->t_flags |= TF_ACKNOW; 2189 } 2190 2191 /* 2192 * Note the amount of data that peer has sent into 2193 * our window, in order to estimate the sender's 2194 * buffer size. 2195 */ 2196 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2197 } else { 2198 m_freem(m); 2199 thflags &= ~TH_FIN; 2200 } 2201 2202 /* 2203 * If FIN is received ACK the FIN and let the user know 2204 * that the connection is closing. 2205 */ 2206 if (thflags & TH_FIN) { 2207 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2208 socantrcvmore(so); 2209 /* 2210 * If connection is half-synchronized 2211 * (ie NEEDSYN flag on) then delay ACK, 2212 * so it may be piggybacked when SYN is sent. 2213 * Otherwise, since we received a FIN then no 2214 * more input can be expected, send ACK now. 2215 */ 2216 if (tp->t_flags & TF_NEEDSYN) 2217 tp->t_flags |= TF_DELACK; 2218 else 2219 tp->t_flags |= TF_ACKNOW; 2220 tp->rcv_nxt++; 2221 } 2222 switch (tp->t_state) { 2223 2224 /* 2225 * In SYN_RECEIVED and ESTABLISHED STATES 2226 * enter the CLOSE_WAIT state. 2227 */ 2228 case TCPS_SYN_RECEIVED: 2229 tp->t_starttime = ticks; 2230 /*FALLTHROUGH*/ 2231 case TCPS_ESTABLISHED: 2232 tp->t_state = TCPS_CLOSE_WAIT; 2233 break; 2234 2235 /* 2236 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2237 * enter the CLOSING state. 2238 */ 2239 case TCPS_FIN_WAIT_1: 2240 tp->t_state = TCPS_CLOSING; 2241 break; 2242 2243 /* 2244 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2245 * starting the time-wait timer, turning off the other 2246 * standard timers. 2247 */ 2248 case TCPS_FIN_WAIT_2: 2249 KASSERT(headlocked == 1, ("headlocked should be 1")); 2250 tcp_twstart(tp); 2251 INP_INFO_WUNLOCK(&tcbinfo); 2252 return; 2253 2254 /* 2255 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2256 */ 2257 case TCPS_TIME_WAIT: 2258 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2259 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2260 tcp_timer_2msl, tp); 2261 break; 2262 } 2263 } 2264 INP_INFO_WUNLOCK(&tcbinfo); 2265 #ifdef TCPDEBUG 2266 if (so->so_options & SO_DEBUG) 2267 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2268 &tcp_savetcp, 0); 2269 #endif 2270 2271 /* 2272 * Return any desired output. 2273 */ 2274 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2275 (void) tcp_output(tp); 2276 2277 check_delack: 2278 if (tp->t_flags & TF_DELACK) { 2279 tp->t_flags &= ~TF_DELACK; 2280 callout_reset(tp->tt_delack, tcp_delacktime, 2281 tcp_timer_delack, tp); 2282 } 2283 INP_UNLOCK(inp); 2284 return; 2285 2286 dropafterack: 2287 /* 2288 * Generate an ACK dropping incoming segment if it occupies 2289 * sequence space, where the ACK reflects our state. 2290 * 2291 * We can now skip the test for the RST flag since all 2292 * paths to this code happen after packets containing 2293 * RST have been dropped. 2294 * 2295 * In the SYN-RECEIVED state, don't send an ACK unless the 2296 * segment we received passes the SYN-RECEIVED ACK test. 2297 * If it fails send a RST. This breaks the loop in the 2298 * "LAND" DoS attack, and also prevents an ACK storm 2299 * between two listening ports that have been sent forged 2300 * SYN segments, each with the source address of the other. 2301 */ 2302 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2303 (SEQ_GT(tp->snd_una, th->th_ack) || 2304 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2305 rstreason = BANDLIM_RST_OPENPORT; 2306 goto dropwithreset; 2307 } 2308 #ifdef TCPDEBUG 2309 if (so->so_options & SO_DEBUG) 2310 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2311 &tcp_savetcp, 0); 2312 #endif 2313 KASSERT(headlocked, ("headlocked should be 1")); 2314 INP_INFO_WUNLOCK(&tcbinfo); 2315 m_freem(m); 2316 tp->t_flags |= TF_ACKNOW; 2317 (void) tcp_output(tp); 2318 INP_UNLOCK(inp); 2319 return; 2320 2321 dropwithreset: 2322 /* 2323 * Generate a RST, dropping incoming segment. 2324 * Make ACK acceptable to originator of segment. 2325 * Don't bother to respond if destination was broadcast/multicast. 2326 */ 2327 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2328 goto drop; 2329 if (isipv6) { 2330 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2331 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2332 goto drop; 2333 } else { 2334 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2335 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2336 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2337 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2338 goto drop; 2339 } 2340 /* IPv6 anycast check is done at tcp6_input() */ 2341 2342 /* 2343 * Perform bandwidth limiting. 2344 */ 2345 if (badport_bandlim(rstreason) < 0) 2346 goto drop; 2347 2348 #ifdef TCPDEBUG 2349 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2350 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2351 &tcp_savetcp, 0); 2352 #endif 2353 2354 if (thflags & TH_ACK) 2355 /* mtod() below is safe as long as hdr dropping is delayed */ 2356 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2357 TH_RST); 2358 else { 2359 if (thflags & TH_SYN) 2360 tlen++; 2361 /* mtod() below is safe as long as hdr dropping is delayed */ 2362 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2363 (tcp_seq)0, TH_RST|TH_ACK); 2364 } 2365 2366 if (tp) 2367 INP_UNLOCK(inp); 2368 if (headlocked) 2369 INP_INFO_WUNLOCK(&tcbinfo); 2370 return; 2371 2372 drop: 2373 /* 2374 * Drop space held by incoming segment and return. 2375 */ 2376 #ifdef TCPDEBUG 2377 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2378 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2379 &tcp_savetcp, 0); 2380 #endif 2381 if (tp) 2382 INP_UNLOCK(inp); 2383 m_freem(m); 2384 if (headlocked) 2385 INP_INFO_WUNLOCK(&tcbinfo); 2386 return; 2387 } 2388 2389 /* 2390 * Parse TCP options and place in tcpopt. 2391 */ 2392 static void 2393 tcp_dooptions(to, cp, cnt, is_syn) 2394 struct tcpopt *to; 2395 u_char *cp; 2396 int cnt; 2397 int is_syn; 2398 { 2399 int opt, optlen; 2400 2401 to->to_flags = 0; 2402 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2403 opt = cp[0]; 2404 if (opt == TCPOPT_EOL) 2405 break; 2406 if (opt == TCPOPT_NOP) 2407 optlen = 1; 2408 else { 2409 if (cnt < 2) 2410 break; 2411 optlen = cp[1]; 2412 if (optlen < 2 || optlen > cnt) 2413 break; 2414 } 2415 switch (opt) { 2416 case TCPOPT_MAXSEG: 2417 if (optlen != TCPOLEN_MAXSEG) 2418 continue; 2419 if (!is_syn) 2420 continue; 2421 to->to_flags |= TOF_MSS; 2422 bcopy((char *)cp + 2, 2423 (char *)&to->to_mss, sizeof(to->to_mss)); 2424 to->to_mss = ntohs(to->to_mss); 2425 break; 2426 case TCPOPT_WINDOW: 2427 if (optlen != TCPOLEN_WINDOW) 2428 continue; 2429 if (! is_syn) 2430 continue; 2431 to->to_flags |= TOF_SCALE; 2432 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2433 break; 2434 case TCPOPT_TIMESTAMP: 2435 if (optlen != TCPOLEN_TIMESTAMP) 2436 continue; 2437 to->to_flags |= TOF_TS; 2438 bcopy((char *)cp + 2, 2439 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2440 to->to_tsval = ntohl(to->to_tsval); 2441 bcopy((char *)cp + 6, 2442 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2443 to->to_tsecr = ntohl(to->to_tsecr); 2444 break; 2445 case TCPOPT_CC: 2446 if (optlen != TCPOLEN_CC) 2447 continue; 2448 to->to_flags |= TOF_CC; 2449 bcopy((char *)cp + 2, 2450 (char *)&to->to_cc, sizeof(to->to_cc)); 2451 to->to_cc = ntohl(to->to_cc); 2452 break; 2453 case TCPOPT_CCNEW: 2454 if (optlen != TCPOLEN_CC) 2455 continue; 2456 if (!is_syn) 2457 continue; 2458 to->to_flags |= TOF_CCNEW; 2459 bcopy((char *)cp + 2, 2460 (char *)&to->to_cc, sizeof(to->to_cc)); 2461 to->to_cc = ntohl(to->to_cc); 2462 break; 2463 case TCPOPT_CCECHO: 2464 if (optlen != TCPOLEN_CC) 2465 continue; 2466 if (!is_syn) 2467 continue; 2468 to->to_flags |= TOF_CCECHO; 2469 bcopy((char *)cp + 2, 2470 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2471 to->to_ccecho = ntohl(to->to_ccecho); 2472 break; 2473 default: 2474 continue; 2475 } 2476 } 2477 } 2478 2479 /* 2480 * Pull out of band byte out of a segment so 2481 * it doesn't appear in the user's data queue. 2482 * It is still reflected in the segment length for 2483 * sequencing purposes. 2484 */ 2485 static void 2486 tcp_pulloutofband(so, th, m, off) 2487 struct socket *so; 2488 struct tcphdr *th; 2489 register struct mbuf *m; 2490 int off; /* delayed to be droped hdrlen */ 2491 { 2492 int cnt = off + th->th_urp - 1; 2493 2494 while (cnt >= 0) { 2495 if (m->m_len > cnt) { 2496 char *cp = mtod(m, caddr_t) + cnt; 2497 struct tcpcb *tp = sototcpcb(so); 2498 2499 tp->t_iobc = *cp; 2500 tp->t_oobflags |= TCPOOB_HAVEDATA; 2501 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2502 m->m_len--; 2503 if (m->m_flags & M_PKTHDR) 2504 m->m_pkthdr.len--; 2505 return; 2506 } 2507 cnt -= m->m_len; 2508 m = m->m_next; 2509 if (m == 0) 2510 break; 2511 } 2512 panic("tcp_pulloutofband"); 2513 } 2514 2515 /* 2516 * Collect new round-trip time estimate 2517 * and update averages and current timeout. 2518 */ 2519 static void 2520 tcp_xmit_timer(tp, rtt) 2521 register struct tcpcb *tp; 2522 int rtt; 2523 { 2524 register int delta; 2525 2526 tcpstat.tcps_rttupdated++; 2527 tp->t_rttupdated++; 2528 if (tp->t_srtt != 0) { 2529 /* 2530 * srtt is stored as fixed point with 5 bits after the 2531 * binary point (i.e., scaled by 8). The following magic 2532 * is equivalent to the smoothing algorithm in rfc793 with 2533 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2534 * point). Adjust rtt to origin 0. 2535 */ 2536 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2537 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2538 2539 if ((tp->t_srtt += delta) <= 0) 2540 tp->t_srtt = 1; 2541 2542 /* 2543 * We accumulate a smoothed rtt variance (actually, a 2544 * smoothed mean difference), then set the retransmit 2545 * timer to smoothed rtt + 4 times the smoothed variance. 2546 * rttvar is stored as fixed point with 4 bits after the 2547 * binary point (scaled by 16). The following is 2548 * equivalent to rfc793 smoothing with an alpha of .75 2549 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2550 * rfc793's wired-in beta. 2551 */ 2552 if (delta < 0) 2553 delta = -delta; 2554 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2555 if ((tp->t_rttvar += delta) <= 0) 2556 tp->t_rttvar = 1; 2557 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2558 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2559 } else { 2560 /* 2561 * No rtt measurement yet - use the unsmoothed rtt. 2562 * Set the variance to half the rtt (so our first 2563 * retransmit happens at 3*rtt). 2564 */ 2565 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2566 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2567 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2568 } 2569 tp->t_rtttime = 0; 2570 tp->t_rxtshift = 0; 2571 2572 /* 2573 * the retransmit should happen at rtt + 4 * rttvar. 2574 * Because of the way we do the smoothing, srtt and rttvar 2575 * will each average +1/2 tick of bias. When we compute 2576 * the retransmit timer, we want 1/2 tick of rounding and 2577 * 1 extra tick because of +-1/2 tick uncertainty in the 2578 * firing of the timer. The bias will give us exactly the 2579 * 1.5 tick we need. But, because the bias is 2580 * statistical, we have to test that we don't drop below 2581 * the minimum feasible timer (which is 2 ticks). 2582 */ 2583 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2584 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2585 2586 /* 2587 * We received an ack for a packet that wasn't retransmitted; 2588 * it is probably safe to discard any error indications we've 2589 * received recently. This isn't quite right, but close enough 2590 * for now (a route might have failed after we sent a segment, 2591 * and the return path might not be symmetrical). 2592 */ 2593 tp->t_softerror = 0; 2594 } 2595 2596 /* 2597 * Determine a reasonable value for maxseg size. 2598 * If the route is known, check route for mtu. 2599 * If none, use an mss that can be handled on the outgoing 2600 * interface without forcing IP to fragment; if bigger than 2601 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2602 * to utilize large mbufs. If no route is found, route has no mtu, 2603 * or the destination isn't local, use a default, hopefully conservative 2604 * size (usually 512 or the default IP max size, but no more than the mtu 2605 * of the interface), as we can't discover anything about intervening 2606 * gateways or networks. We also initialize the congestion/slow start 2607 * window to be a single segment if the destination isn't local. 2608 * While looking at the routing entry, we also initialize other path-dependent 2609 * parameters from pre-set or cached values in the routing entry. 2610 * 2611 * Also take into account the space needed for options that we 2612 * send regularly. Make maxseg shorter by that amount to assure 2613 * that we can send maxseg amount of data even when the options 2614 * are present. Store the upper limit of the length of options plus 2615 * data in maxopd. 2616 * 2617 * 2618 * In case of T/TCP, we call this routine during implicit connection 2619 * setup as well (offer = -1), to initialize maxseg from the cached 2620 * MSS of our peer. 2621 * 2622 * NOTE that this routine is only called when we process an incoming 2623 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2624 */ 2625 void 2626 tcp_mss(tp, offer) 2627 struct tcpcb *tp; 2628 int offer; 2629 { 2630 int rtt, mss; 2631 u_long bufsize; 2632 u_long maxmtu; 2633 struct inpcb *inp = tp->t_inpcb; 2634 struct socket *so; 2635 struct hc_metrics_lite metrics; 2636 struct rmxp_tao tao; 2637 int origoffer = offer; 2638 #ifdef INET6 2639 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2640 size_t min_protoh = isipv6 ? 2641 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2642 sizeof (struct tcpiphdr); 2643 #else 2644 const size_t min_protoh = sizeof(struct tcpiphdr); 2645 #endif 2646 bzero(&tao, sizeof(tao)); 2647 2648 /* initialize */ 2649 #ifdef INET6 2650 if (isipv6) { 2651 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2652 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2653 } else 2654 #endif 2655 { 2656 maxmtu = tcp_maxmtu(&inp->inp_inc); 2657 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2658 } 2659 so = inp->inp_socket; 2660 2661 /* 2662 * no route to sender, take default mss and return 2663 */ 2664 if (maxmtu == 0) 2665 return; 2666 2667 /* what have we got? */ 2668 switch (offer) { 2669 case 0: 2670 /* 2671 * Offer == 0 means that there was no MSS on the SYN 2672 * segment, in this case we use tcp_mssdflt. 2673 */ 2674 offer = 2675 #ifdef INET6 2676 isipv6 ? tcp_v6mssdflt : 2677 #endif 2678 tcp_mssdflt; 2679 break; 2680 2681 case -1: 2682 /* 2683 * Offer == -1 means that we didn't receive SYN yet, 2684 * use cached value in that case; 2685 */ 2686 if (tcp_do_rfc1644) 2687 tcp_hc_gettao(&inp->inp_inc, &tao); 2688 if (tao.tao_mssopt != 0) 2689 offer = tao.tao_mssopt; 2690 /* FALLTHROUGH */ 2691 2692 default: 2693 /* 2694 * Sanity check: make sure that maxopd will be large 2695 * enough to allow some data on segments even if the 2696 * all the option space is used (40bytes). Otherwise 2697 * funny things may happen in tcp_output. 2698 */ 2699 offer = max(offer, 64); 2700 if (tcp_do_rfc1644) 2701 tcp_hc_updatetao(&inp->inp_inc, 2702 TCP_HC_TAO_MSSOPT, 0, offer); 2703 } 2704 2705 /* 2706 * rmx information is now retrieved from tcp_hostcache 2707 */ 2708 tcp_hc_get(&inp->inp_inc, &metrics); 2709 2710 /* 2711 * if there's a discovered mtu int tcp hostcache, use it 2712 * else, use the link mtu. 2713 */ 2714 if (metrics.rmx_mtu) 2715 mss = metrics.rmx_mtu - min_protoh; 2716 else { 2717 #ifdef INET6 2718 if (isipv6) { 2719 mss = maxmtu - min_protoh; 2720 if (!path_mtu_discovery && 2721 !in6_localaddr(&inp->in6p_faddr)) 2722 mss = min(mss, tcp_v6mssdflt); 2723 } else 2724 #endif 2725 { 2726 mss = maxmtu - min_protoh; 2727 if (!path_mtu_discovery && 2728 !in_localaddr(inp->inp_faddr)) 2729 mss = min(mss, tcp_mssdflt); 2730 } 2731 } 2732 mss = min(mss, offer); 2733 2734 /* 2735 * maxopd stores the maximum length of data AND options 2736 * in a segment; maxseg is the amount of data in a normal 2737 * segment. We need to store this value (maxopd) apart 2738 * from maxseg, because now every segment carries options 2739 * and thus we normally have somewhat less data in segments. 2740 */ 2741 tp->t_maxopd = mss; 2742 2743 /* 2744 * In case of T/TCP, origoffer==-1 indicates, that no segments 2745 * were received yet. In this case we just guess, otherwise 2746 * we do the same as before T/TCP. 2747 */ 2748 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2749 (origoffer == -1 || 2750 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2751 mss -= TCPOLEN_TSTAMP_APPA; 2752 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2753 (origoffer == -1 || 2754 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2755 mss -= TCPOLEN_CC_APPA; 2756 tp->t_maxseg = mss; 2757 2758 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2759 if (mss > MCLBYTES) 2760 mss &= ~(MCLBYTES-1); 2761 #else 2762 if (mss > MCLBYTES) 2763 mss = mss / MCLBYTES * MCLBYTES; 2764 #endif 2765 tp->t_maxseg = mss; 2766 2767 /* 2768 * If there's a pipesize, change the socket buffer to that size, 2769 * don't change if sb_hiwat is different than default (then it 2770 * has been changed on purpose with setsockopt). 2771 * Make the socket buffers an integral number of mss units; 2772 * if the mss is larger than the socket buffer, decrease the mss. 2773 */ 2774 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2775 bufsize = metrics.rmx_sendpipe; 2776 else 2777 bufsize = so->so_snd.sb_hiwat; 2778 if (bufsize < mss) 2779 mss = bufsize; 2780 else { 2781 bufsize = roundup(bufsize, mss); 2782 if (bufsize > sb_max) 2783 bufsize = sb_max; 2784 if (bufsize > so->so_snd.sb_hiwat) 2785 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2786 } 2787 tp->t_maxseg = mss; 2788 2789 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2790 bufsize = metrics.rmx_recvpipe; 2791 else 2792 bufsize = so->so_rcv.sb_hiwat; 2793 if (bufsize > mss) { 2794 bufsize = roundup(bufsize, mss); 2795 if (bufsize > sb_max) 2796 bufsize = sb_max; 2797 if (bufsize > so->so_rcv.sb_hiwat) 2798 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2799 } 2800 /* 2801 * While we're here, check the others too 2802 */ 2803 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2804 tp->t_srtt = rtt; 2805 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2806 tcpstat.tcps_usedrtt++; 2807 if (metrics.rmx_rttvar) { 2808 tp->t_rttvar = metrics.rmx_rttvar; 2809 tcpstat.tcps_usedrttvar++; 2810 } else { 2811 /* default variation is +- 1 rtt */ 2812 tp->t_rttvar = 2813 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2814 } 2815 TCPT_RANGESET(tp->t_rxtcur, 2816 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2817 tp->t_rttmin, TCPTV_REXMTMAX); 2818 } 2819 if (metrics.rmx_ssthresh) { 2820 /* 2821 * There's some sort of gateway or interface 2822 * buffer limit on the path. Use this to set 2823 * the slow start threshhold, but set the 2824 * threshold to no less than 2*mss. 2825 */ 2826 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2827 tcpstat.tcps_usedssthresh++; 2828 } 2829 if (metrics.rmx_bandwidth) 2830 tp->snd_bandwidth = metrics.rmx_bandwidth; 2831 2832 /* 2833 * Set the slow-start flight size depending on whether this 2834 * is a local network or not. 2835 * 2836 * Extend this so we cache the cwnd too and retrieve it here. 2837 * Make cwnd even bigger than RFC3390 suggests but only if we 2838 * have previous experience with the remote host. Be careful 2839 * not make cwnd bigger than remote receive window or our own 2840 * send socket buffer. Maybe put some additional upper bound 2841 * on the retrieved cwnd. Should do incremental updates to 2842 * hostcache when cwnd collapses so next connection doesn't 2843 * overloads the path again. 2844 * 2845 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2846 * We currently check only in syncache_socket for that. 2847 */ 2848 #define TCP_METRICS_CWND 2849 #ifdef TCP_METRICS_CWND 2850 if (metrics.rmx_cwnd) 2851 tp->snd_cwnd = max(mss, 2852 min(metrics.rmx_cwnd / 2, 2853 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2854 else 2855 #endif 2856 if (tcp_do_rfc3390) 2857 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2858 #ifdef INET6 2859 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2860 (!isipv6 && in_localaddr(inp->inp_faddr))) 2861 #else 2862 else if (in_localaddr(inp->inp_faddr)) 2863 #endif 2864 tp->snd_cwnd = mss * ss_fltsz_local; 2865 else 2866 tp->snd_cwnd = mss * ss_fltsz; 2867 } 2868 2869 /* 2870 * Determine the MSS option to send on an outgoing SYN. 2871 */ 2872 int 2873 tcp_mssopt(inc) 2874 struct in_conninfo *inc; 2875 { 2876 int mss = 0; 2877 u_long maxmtu = 0; 2878 u_long thcmtu = 0; 2879 size_t min_protoh; 2880 #ifdef INET6 2881 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2882 #endif 2883 2884 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2885 2886 #ifdef INET6 2887 if (isipv6) { 2888 mss = tcp_v6mssdflt; 2889 maxmtu = tcp_maxmtu6(inc); 2890 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2891 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2892 } else 2893 #endif 2894 { 2895 mss = tcp_mssdflt; 2896 maxmtu = tcp_maxmtu(inc); 2897 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2898 min_protoh = sizeof(struct tcpiphdr); 2899 } 2900 if (maxmtu && thcmtu) 2901 mss = min(maxmtu, thcmtu) - min_protoh; 2902 else if (maxmtu || thcmtu) 2903 mss = max(maxmtu, thcmtu) - min_protoh; 2904 2905 return (mss); 2906 } 2907 2908 2909 /* 2910 * On a partial ack arrives, force the retransmission of the 2911 * next unacknowledged segment. Do not clear tp->t_dupacks. 2912 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2913 * be started again. 2914 */ 2915 static void 2916 tcp_newreno_partial_ack(tp, th) 2917 struct tcpcb *tp; 2918 struct tcphdr *th; 2919 { 2920 tcp_seq onxt = tp->snd_nxt; 2921 u_long ocwnd = tp->snd_cwnd; 2922 2923 callout_stop(tp->tt_rexmt); 2924 tp->t_rtttime = 0; 2925 tp->snd_nxt = th->th_ack; 2926 /* 2927 * Set snd_cwnd to one segment beyond acknowledged offset. 2928 * (tp->snd_una has not yet been updated when this function is called.) 2929 */ 2930 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2931 tp->t_flags |= TF_ACKNOW; 2932 (void) tcp_output(tp); 2933 tp->snd_cwnd = ocwnd; 2934 if (SEQ_GT(onxt, tp->snd_nxt)) 2935 tp->snd_nxt = onxt; 2936 /* 2937 * Partial window deflation. Relies on fact that tp->snd_una 2938 * not updated yet. 2939 */ 2940 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2941 } 2942 2943 /* 2944 * Returns 1 if the TIME_WAIT state was killed and we should start over, 2945 * looking for a pcb in the listen state. Returns 0 otherwise. 2946 */ 2947 static int 2948 tcp_timewait(tw, to, th, m, tlen) 2949 struct tcptw *tw; 2950 struct tcpopt *to; 2951 struct tcphdr *th; 2952 struct mbuf *m; 2953 int tlen; 2954 { 2955 int thflags; 2956 tcp_seq seq; 2957 #ifdef INET6 2958 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 2959 #else 2960 const int isipv6 = 0; 2961 #endif 2962 2963 thflags = th->th_flags; 2964 2965 /* 2966 * NOTE: for FIN_WAIT_2 (to be added later), 2967 * must validate sequence number before accepting RST 2968 */ 2969 2970 /* 2971 * If the segment contains RST: 2972 * Drop the segment - see Stevens, vol. 2, p. 964 and 2973 * RFC 1337. 2974 */ 2975 if (thflags & TH_RST) 2976 goto drop; 2977 2978 /* 2979 * If segment contains a SYN and CC [not CC.NEW] option: 2980 * if connection duration > MSL, drop packet and send RST; 2981 * 2982 * if SEG.CC > CCrecv then is new SYN. 2983 * Complete close and delete TCPCB. Then reprocess 2984 * segment, hoping to find new TCPCB in LISTEN state; 2985 * 2986 * else must be old SYN; drop it. 2987 * else do normal processing. 2988 */ 2989 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 2990 if ((ticks - tw->t_starttime) > tcp_msl) 2991 goto reset; 2992 if (CC_GT(to->to_cc, tw->cc_recv)) { 2993 (void) tcp_twclose(tw, 0); 2994 return (1); 2995 } 2996 goto drop; 2997 } 2998 2999 #if 0 3000 /* PAWS not needed at the moment */ 3001 /* 3002 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3003 * and it's less than ts_recent, drop it. 3004 */ 3005 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3006 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3007 if ((thflags & TH_ACK) == 0) 3008 goto drop; 3009 goto ack; 3010 } 3011 /* 3012 * ts_recent is never updated because we never accept new segments. 3013 */ 3014 #endif 3015 3016 /* 3017 * If a new connection request is received 3018 * while in TIME_WAIT, drop the old connection 3019 * and start over if the sequence numbers 3020 * are above the previous ones. 3021 */ 3022 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3023 (void) tcp_twclose(tw, 0); 3024 return (1); 3025 } 3026 3027 /* 3028 * Drop the the segment if it does not contain an ACK. 3029 */ 3030 if ((thflags & TH_ACK) == 0) 3031 goto drop; 3032 3033 /* 3034 * Reset the 2MSL timer if this is a duplicate FIN. 3035 */ 3036 if (thflags & TH_FIN) { 3037 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3038 if (seq + 1 == tw->rcv_nxt) 3039 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3040 } 3041 3042 /* 3043 * Acknowledge the segment if it has data or is not a duplicate ACK. 3044 */ 3045 if (thflags != TH_ACK || tlen != 0 || 3046 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3047 tcp_twrespond(tw, NULL, m, TH_ACK); 3048 goto drop; 3049 3050 reset: 3051 /* 3052 * Generate a RST, dropping incoming segment. 3053 * Make ACK acceptable to originator of segment. 3054 * Don't bother to respond if destination was broadcast/multicast. 3055 */ 3056 if (m->m_flags & (M_BCAST|M_MCAST)) 3057 goto drop; 3058 if (isipv6) { 3059 struct ip6_hdr *ip6; 3060 3061 /* IPv6 anycast check is done at tcp6_input() */ 3062 ip6 = mtod(m, struct ip6_hdr *); 3063 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3064 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3065 goto drop; 3066 } else { 3067 struct ip *ip; 3068 3069 ip = mtod(m, struct ip *); 3070 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3071 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3072 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3073 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3074 goto drop; 3075 } 3076 if (thflags & TH_ACK) { 3077 tcp_respond(NULL, 3078 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3079 } else { 3080 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3081 tcp_respond(NULL, 3082 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3083 } 3084 INP_UNLOCK(tw->tw_inpcb); 3085 return (0); 3086 3087 drop: 3088 INP_UNLOCK(tw->tw_inpcb); 3089 m_freem(m); 3090 return (0); 3091 } 3092