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 = 1; 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 = 1; 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 * This is the second part of the MSS DoS prevention code (after 923 * minmss on the sending side) and it deals with too many too small 924 * tcp packets in a too short timeframe (1 second). 925 * 926 * For every full second we count the number of received packets 927 * and bytes. If we get a lot of packets per second for this connection 928 * (tcp_minmssoverload) we take a closer look at it and compute the 929 * average packet size for the past second. If that is less than 930 * tcp_minmss we get too many packets with very small payload which 931 * is not good and burdens our system (and every packet generates 932 * a wakeup to the process connected to our socket). We can reasonable 933 * expect this to be small packet DoS attack to exhaust our CPU 934 * cycles. 935 * 936 * Care has to be taken for the minimum packet overload value. This 937 * value defines the minimum number of packets per second before we 938 * start to worry. This must not be too low to avoid killing for 939 * example interactive connections with many small packets like 940 * telnet or SSH. 941 * 942 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 943 * this check. 944 * 945 * Account for packet if payload packet, skip over ACK, etc. 946 */ 947 if (tcp_minmss && tcp_minmssoverload && 948 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 949 if (tp->rcv_second > ticks) { 950 tp->rcv_pps++; 951 tp->rcv_byps += tlen + off; 952 if (tp->rcv_pps > tcp_minmssoverload) { 953 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 954 printf("too many small tcp packets from " 955 "%s:%u, av. %lubyte/packet, " 956 "dropping connection\n", 957 #ifdef INET6 958 isipv6 ? 959 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 960 #endif 961 inet_ntoa(inp->inp_inc.inc_faddr), 962 inp->inp_inc.inc_fport, 963 tp->rcv_byps / tp->rcv_pps); 964 tp = tcp_drop(tp, ECONNRESET); 965 tcpstat.tcps_minmssdrops++; 966 goto drop; 967 } 968 } 969 } else { 970 tp->rcv_second = ticks + hz; 971 tp->rcv_pps = 1; 972 tp->rcv_byps = tlen + off; 973 } 974 } 975 976 /* 977 * Segment received on connection. 978 * Reset idle time and keep-alive timer. 979 */ 980 tp->t_rcvtime = ticks; 981 if (TCPS_HAVEESTABLISHED(tp->t_state)) 982 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 983 984 /* 985 * Process options only when we get SYN/ACK back. The SYN case 986 * for incoming connections is handled in tcp_syncache. 987 * XXX this is traditional behavior, may need to be cleaned up. 988 */ 989 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 990 if (thflags & TH_SYN) { 991 if (to.to_flags & TOF_SCALE) { 992 tp->t_flags |= TF_RCVD_SCALE; 993 tp->requested_s_scale = to.to_requested_s_scale; 994 } 995 if (to.to_flags & TOF_TS) { 996 tp->t_flags |= TF_RCVD_TSTMP; 997 tp->ts_recent = to.to_tsval; 998 tp->ts_recent_age = ticks; 999 } 1000 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1001 tp->t_flags |= TF_RCVD_CC; 1002 if (to.to_flags & TOF_MSS) 1003 tcp_mss(tp, to.to_mss); 1004 } 1005 1006 /* 1007 * Header prediction: check for the two common cases 1008 * of a uni-directional data xfer. If the packet has 1009 * no control flags, is in-sequence, the window didn't 1010 * change and we're not retransmitting, it's a 1011 * candidate. If the length is zero and the ack moved 1012 * forward, we're the sender side of the xfer. Just 1013 * free the data acked & wake any higher level process 1014 * that was blocked waiting for space. If the length 1015 * is non-zero and the ack didn't move, we're the 1016 * receiver side. If we're getting packets in-order 1017 * (the reassembly queue is empty), add the data to 1018 * the socket buffer and note that we need a delayed ack. 1019 * Make sure that the hidden state-flags are also off. 1020 * Since we check for TCPS_ESTABLISHED above, it can only 1021 * be TH_NEEDSYN. 1022 */ 1023 if (tp->t_state == TCPS_ESTABLISHED && 1024 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1025 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1026 ((to.to_flags & TOF_TS) == 0 || 1027 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1028 /* 1029 * Using the CC option is compulsory if once started: 1030 * the segment is OK if no T/TCP was negotiated or 1031 * if the segment has a CC option equal to CCrecv 1032 */ 1033 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1034 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1035 th->th_seq == tp->rcv_nxt && 1036 tiwin && tiwin == tp->snd_wnd && 1037 tp->snd_nxt == tp->snd_max) { 1038 1039 /* 1040 * If last ACK falls within this segment's sequence numbers, 1041 * record the timestamp. 1042 * NOTE that the test is modified according to the latest 1043 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1044 */ 1045 if ((to.to_flags & TOF_TS) != 0 && 1046 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1047 tp->ts_recent_age = ticks; 1048 tp->ts_recent = to.to_tsval; 1049 } 1050 1051 if (tlen == 0) { 1052 if (SEQ_GT(th->th_ack, tp->snd_una) && 1053 SEQ_LEQ(th->th_ack, tp->snd_max) && 1054 tp->snd_cwnd >= tp->snd_wnd && 1055 ((!tcp_do_newreno && 1056 tp->t_dupacks < tcprexmtthresh) || 1057 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1058 KASSERT(headlocked, ("headlocked")); 1059 INP_INFO_WUNLOCK(&tcbinfo); 1060 /* 1061 * this is a pure ack for outstanding data. 1062 */ 1063 ++tcpstat.tcps_predack; 1064 /* 1065 * "bad retransmit" recovery 1066 */ 1067 if (tp->t_rxtshift == 1 && 1068 ticks < tp->t_badrxtwin) { 1069 ++tcpstat.tcps_sndrexmitbad; 1070 tp->snd_cwnd = tp->snd_cwnd_prev; 1071 tp->snd_ssthresh = 1072 tp->snd_ssthresh_prev; 1073 tp->snd_recover = tp->snd_recover_prev; 1074 if (tp->t_flags & TF_WASFRECOVERY) 1075 ENTER_FASTRECOVERY(tp); 1076 tp->snd_nxt = tp->snd_max; 1077 tp->t_badrxtwin = 0; 1078 } 1079 1080 /* 1081 * Recalculate the transmit timer / rtt. 1082 * 1083 * Some boxes send broken timestamp replies 1084 * during the SYN+ACK phase, ignore 1085 * timestamps of 0 or we could calculate a 1086 * huge RTT and blow up the retransmit timer. 1087 */ 1088 if ((to.to_flags & TOF_TS) != 0 && 1089 to.to_tsecr) { 1090 tcp_xmit_timer(tp, 1091 ticks - to.to_tsecr + 1); 1092 } else if (tp->t_rtttime && 1093 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1094 tcp_xmit_timer(tp, 1095 ticks - tp->t_rtttime); 1096 } 1097 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1098 acked = th->th_ack - tp->snd_una; 1099 tcpstat.tcps_rcvackpack++; 1100 tcpstat.tcps_rcvackbyte += acked; 1101 sbdrop(&so->so_snd, acked); 1102 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1103 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1104 tp->snd_recover = th->th_ack - 1; 1105 tp->snd_una = th->th_ack; 1106 /* 1107 * pull snd_wl2 up to prevent seq wrap relative 1108 * to th_ack. 1109 */ 1110 tp->snd_wl2 = th->th_ack; 1111 tp->t_dupacks = 0; 1112 m_freem(m); 1113 ND6_HINT(tp); /* some progress has been done */ 1114 1115 /* 1116 * If all outstanding data are acked, stop 1117 * retransmit timer, otherwise restart timer 1118 * using current (possibly backed-off) value. 1119 * If process is waiting for space, 1120 * wakeup/selwakeup/signal. If data 1121 * are ready to send, let tcp_output 1122 * decide between more output or persist. 1123 1124 #ifdef TCPDEBUG 1125 if (so->so_options & SO_DEBUG) 1126 tcp_trace(TA_INPUT, ostate, tp, 1127 (void *)tcp_saveipgen, 1128 &tcp_savetcp, 0); 1129 #endif 1130 */ 1131 if (tp->snd_una == tp->snd_max) 1132 callout_stop(tp->tt_rexmt); 1133 else if (!callout_active(tp->tt_persist)) 1134 callout_reset(tp->tt_rexmt, 1135 tp->t_rxtcur, 1136 tcp_timer_rexmt, tp); 1137 1138 sowwakeup(so); 1139 if (so->so_snd.sb_cc) 1140 (void) tcp_output(tp); 1141 goto check_delack; 1142 } 1143 } else if (th->th_ack == tp->snd_una && 1144 LIST_EMPTY(&tp->t_segq) && 1145 tlen <= sbspace(&so->so_rcv)) { 1146 KASSERT(headlocked, ("headlocked")); 1147 INP_INFO_WUNLOCK(&tcbinfo); 1148 /* 1149 * this is a pure, in-sequence data packet 1150 * with nothing on the reassembly queue and 1151 * we have enough buffer space to take it. 1152 */ 1153 ++tcpstat.tcps_preddat; 1154 tp->rcv_nxt += tlen; 1155 /* 1156 * Pull snd_wl1 up to prevent seq wrap relative to 1157 * th_seq. 1158 */ 1159 tp->snd_wl1 = th->th_seq; 1160 /* 1161 * Pull rcv_up up to prevent seq wrap relative to 1162 * rcv_nxt. 1163 */ 1164 tp->rcv_up = tp->rcv_nxt; 1165 tcpstat.tcps_rcvpack++; 1166 tcpstat.tcps_rcvbyte += tlen; 1167 ND6_HINT(tp); /* some progress has been done */ 1168 /* 1169 #ifdef TCPDEBUG 1170 if (so->so_options & SO_DEBUG) 1171 tcp_trace(TA_INPUT, ostate, tp, 1172 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1173 #endif 1174 * Add data to socket buffer. 1175 */ 1176 if (so->so_state & SS_CANTRCVMORE) { 1177 m_freem(m); 1178 } else { 1179 m_adj(m, drop_hdrlen); /* delayed header drop */ 1180 sbappendstream(&so->so_rcv, m); 1181 } 1182 sorwakeup(so); 1183 if (DELAY_ACK(tp)) { 1184 tp->t_flags |= TF_DELACK; 1185 } else { 1186 tp->t_flags |= TF_ACKNOW; 1187 tcp_output(tp); 1188 } 1189 goto check_delack; 1190 } 1191 } 1192 1193 /* 1194 * Calculate amount of space in receive window, 1195 * and then do TCP input processing. 1196 * Receive window is amount of space in rcv queue, 1197 * but not less than advertised window. 1198 */ 1199 { int win; 1200 1201 win = sbspace(&so->so_rcv); 1202 if (win < 0) 1203 win = 0; 1204 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1205 } 1206 1207 switch (tp->t_state) { 1208 1209 /* 1210 * If the state is SYN_RECEIVED: 1211 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1212 */ 1213 case TCPS_SYN_RECEIVED: 1214 if ((thflags & TH_ACK) && 1215 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1216 SEQ_GT(th->th_ack, tp->snd_max))) { 1217 rstreason = BANDLIM_RST_OPENPORT; 1218 goto dropwithreset; 1219 } 1220 break; 1221 1222 /* 1223 * If the state is SYN_SENT: 1224 * if seg contains an ACK, but not for our SYN, drop the input. 1225 * if seg contains a RST, then drop the connection. 1226 * if seg does not contain SYN, then drop it. 1227 * Otherwise this is an acceptable SYN segment 1228 * initialize tp->rcv_nxt and tp->irs 1229 * if seg contains ack then advance tp->snd_una 1230 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1231 * arrange for segment to be acked (eventually) 1232 * continue processing rest of data/controls, beginning with URG 1233 */ 1234 case TCPS_SYN_SENT: 1235 if (tcp_do_rfc1644) 1236 tcp_hc_gettao(&inp->inp_inc, &tao); 1237 1238 if ((thflags & TH_ACK) && 1239 (SEQ_LEQ(th->th_ack, tp->iss) || 1240 SEQ_GT(th->th_ack, tp->snd_max))) { 1241 /* 1242 * If we have a cached CCsent for the remote host, 1243 * hence we haven't just crashed and restarted, 1244 * do not send a RST. This may be a retransmission 1245 * from the other side after our earlier ACK was lost. 1246 * Our new SYN, when it arrives, will serve as the 1247 * needed ACK. 1248 */ 1249 if (tao.tao_ccsent != 0) 1250 goto drop; 1251 else { 1252 rstreason = BANDLIM_UNLIMITED; 1253 goto dropwithreset; 1254 } 1255 } 1256 if (thflags & TH_RST) { 1257 if (thflags & TH_ACK) 1258 tp = tcp_drop(tp, ECONNREFUSED); 1259 goto drop; 1260 } 1261 if ((thflags & TH_SYN) == 0) 1262 goto drop; 1263 tp->snd_wnd = th->th_win; /* initial send window */ 1264 tp->cc_recv = to.to_cc; /* foreign CC */ 1265 1266 tp->irs = th->th_seq; 1267 tcp_rcvseqinit(tp); 1268 if (thflags & TH_ACK) { 1269 /* 1270 * Our SYN was acked. If segment contains CC.ECHO 1271 * option, check it to make sure this segment really 1272 * matches our SYN. If not, just drop it as old 1273 * duplicate, but send an RST if we're still playing 1274 * by the old rules. If no CC.ECHO option, make sure 1275 * we don't get fooled into using T/TCP. 1276 */ 1277 if (to.to_flags & TOF_CCECHO) { 1278 if (tp->cc_send != to.to_ccecho) { 1279 if (tao.tao_ccsent != 0) 1280 goto drop; 1281 else { 1282 rstreason = BANDLIM_UNLIMITED; 1283 goto dropwithreset; 1284 } 1285 } 1286 } else 1287 tp->t_flags &= ~TF_RCVD_CC; 1288 tcpstat.tcps_connects++; 1289 soisconnected(so); 1290 #ifdef MAC 1291 mac_set_socket_peer_from_mbuf(m, so); 1292 #endif 1293 /* Do window scaling on this connection? */ 1294 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1295 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1296 tp->snd_scale = tp->requested_s_scale; 1297 tp->rcv_scale = tp->request_r_scale; 1298 } 1299 /* Segment is acceptable, update cache if undefined. */ 1300 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1301 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1302 1303 tp->rcv_adv += tp->rcv_wnd; 1304 tp->snd_una++; /* SYN is acked */ 1305 /* 1306 * If there's data, delay ACK; if there's also a FIN 1307 * ACKNOW will be turned on later. 1308 */ 1309 if (DELAY_ACK(tp) && tlen != 0) 1310 callout_reset(tp->tt_delack, tcp_delacktime, 1311 tcp_timer_delack, tp); 1312 else 1313 tp->t_flags |= TF_ACKNOW; 1314 /* 1315 * Received <SYN,ACK> in SYN_SENT[*] state. 1316 * Transitions: 1317 * SYN_SENT --> ESTABLISHED 1318 * SYN_SENT* --> FIN_WAIT_1 1319 */ 1320 tp->t_starttime = ticks; 1321 if (tp->t_flags & TF_NEEDFIN) { 1322 tp->t_state = TCPS_FIN_WAIT_1; 1323 tp->t_flags &= ~TF_NEEDFIN; 1324 thflags &= ~TH_SYN; 1325 } else { 1326 tp->t_state = TCPS_ESTABLISHED; 1327 callout_reset(tp->tt_keep, tcp_keepidle, 1328 tcp_timer_keep, tp); 1329 } 1330 } else { 1331 /* 1332 * Received initial SYN in SYN-SENT[*] state => 1333 * simultaneous open. If segment contains CC option 1334 * and there is a cached CC, apply TAO test. 1335 * If it succeeds, connection is * half-synchronized. 1336 * Otherwise, do 3-way handshake: 1337 * SYN-SENT -> SYN-RECEIVED 1338 * SYN-SENT* -> SYN-RECEIVED* 1339 * If there was no CC option, clear cached CC value. 1340 */ 1341 tp->t_flags |= TF_ACKNOW; 1342 callout_stop(tp->tt_rexmt); 1343 if (to.to_flags & TOF_CC) { 1344 if (tao.tao_cc != 0 && 1345 CC_GT(to.to_cc, tao.tao_cc)) { 1346 /* 1347 * update cache and make transition: 1348 * SYN-SENT -> ESTABLISHED* 1349 * SYN-SENT* -> FIN-WAIT-1* 1350 */ 1351 tao.tao_cc = to.to_cc; 1352 tcp_hc_updatetao(&inp->inp_inc, 1353 TCP_HC_TAO_CC, to.to_cc, 0); 1354 tp->t_starttime = ticks; 1355 if (tp->t_flags & TF_NEEDFIN) { 1356 tp->t_state = TCPS_FIN_WAIT_1; 1357 tp->t_flags &= ~TF_NEEDFIN; 1358 } else { 1359 tp->t_state = TCPS_ESTABLISHED; 1360 callout_reset(tp->tt_keep, 1361 tcp_keepidle, 1362 tcp_timer_keep, 1363 tp); 1364 } 1365 tp->t_flags |= TF_NEEDSYN; 1366 } else 1367 tp->t_state = TCPS_SYN_RECEIVED; 1368 } else { 1369 if (tcp_do_rfc1644) { 1370 /* CC.NEW or no option => invalidate cache */ 1371 tao.tao_cc = 0; 1372 tcp_hc_updatetao(&inp->inp_inc, 1373 TCP_HC_TAO_CC, to.to_cc, 0); 1374 } 1375 tp->t_state = TCPS_SYN_RECEIVED; 1376 } 1377 } 1378 1379 trimthenstep6: 1380 /* 1381 * Advance th->th_seq to correspond to first data byte. 1382 * If data, trim to stay within window, 1383 * dropping FIN if necessary. 1384 */ 1385 th->th_seq++; 1386 if (tlen > tp->rcv_wnd) { 1387 todrop = tlen - tp->rcv_wnd; 1388 m_adj(m, -todrop); 1389 tlen = tp->rcv_wnd; 1390 thflags &= ~TH_FIN; 1391 tcpstat.tcps_rcvpackafterwin++; 1392 tcpstat.tcps_rcvbyteafterwin += todrop; 1393 } 1394 tp->snd_wl1 = th->th_seq - 1; 1395 tp->rcv_up = th->th_seq; 1396 /* 1397 * Client side of transaction: already sent SYN and data. 1398 * If the remote host used T/TCP to validate the SYN, 1399 * our data will be ACK'd; if so, enter normal data segment 1400 * processing in the middle of step 5, ack processing. 1401 * Otherwise, goto step 6. 1402 */ 1403 if (thflags & TH_ACK) 1404 goto process_ACK; 1405 1406 goto step6; 1407 1408 /* 1409 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1410 * if segment contains a SYN and CC [not CC.NEW] option: 1411 * if state == TIME_WAIT and connection duration > MSL, 1412 * drop packet and send RST; 1413 * 1414 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1415 * ack the FIN (and data) in retransmission queue. 1416 * Complete close and delete TCPCB. Then reprocess 1417 * segment, hoping to find new TCPCB in LISTEN state; 1418 * 1419 * else must be old SYN; drop it. 1420 * else do normal processing. 1421 */ 1422 case TCPS_LAST_ACK: 1423 case TCPS_CLOSING: 1424 case TCPS_TIME_WAIT: 1425 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1426 if ((thflags & TH_SYN) && 1427 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1428 if (tp->t_state == TCPS_TIME_WAIT && 1429 (ticks - tp->t_starttime) > tcp_msl) { 1430 rstreason = BANDLIM_UNLIMITED; 1431 goto dropwithreset; 1432 } 1433 if (CC_GT(to.to_cc, tp->cc_recv)) { 1434 tp = tcp_close(tp); 1435 goto findpcb; 1436 } 1437 else 1438 goto drop; 1439 } 1440 break; /* continue normal processing */ 1441 } 1442 1443 /* 1444 * States other than LISTEN or SYN_SENT. 1445 * First check the RST flag and sequence number since reset segments 1446 * are exempt from the timestamp and connection count tests. This 1447 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1448 * below which allowed reset segments in half the sequence space 1449 * to fall though and be processed (which gives forged reset 1450 * segments with a random sequence number a 50 percent chance of 1451 * killing a connection). 1452 * Then check timestamp, if present. 1453 * Then check the connection count, if present. 1454 * Then check that at least some bytes of segment are within 1455 * receive window. If segment begins before rcv_nxt, 1456 * drop leading data (and SYN); if nothing left, just ack. 1457 * 1458 * 1459 * If the RST bit is set, check the sequence number to see 1460 * if this is a valid reset segment. 1461 * RFC 793 page 37: 1462 * In all states except SYN-SENT, all reset (RST) segments 1463 * are validated by checking their SEQ-fields. A reset is 1464 * valid if its sequence number is in the window. 1465 * Note: this does not take into account delayed ACKs, so 1466 * we should test against last_ack_sent instead of rcv_nxt. 1467 * The sequence number in the reset segment is normally an 1468 * echo of our outgoing acknowlegement numbers, but some hosts 1469 * send a reset with the sequence number at the rightmost edge 1470 * of our receive window, and we have to handle this case. 1471 * If we have multiple segments in flight, the intial reset 1472 * segment sequence numbers will be to the left of last_ack_sent, 1473 * but they will eventually catch up. 1474 * In any case, it never made sense to trim reset segments to 1475 * fit the receive window since RFC 1122 says: 1476 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1477 * 1478 * A TCP SHOULD allow a received RST segment to include data. 1479 * 1480 * DISCUSSION 1481 * It has been suggested that a RST segment could contain 1482 * ASCII text that encoded and explained the cause of the 1483 * RST. No standard has yet been established for such 1484 * data. 1485 * 1486 * If the reset segment passes the sequence number test examine 1487 * the state: 1488 * SYN_RECEIVED STATE: 1489 * If passive open, return to LISTEN state. 1490 * If active open, inform user that connection was refused. 1491 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1492 * Inform user that connection was reset, and close tcb. 1493 * CLOSING, LAST_ACK STATES: 1494 * Close the tcb. 1495 * TIME_WAIT STATE: 1496 * Drop the segment - see Stevens, vol. 2, p. 964 and 1497 * RFC 1337. 1498 */ 1499 if (thflags & TH_RST) { 1500 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1501 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1502 switch (tp->t_state) { 1503 1504 case TCPS_SYN_RECEIVED: 1505 so->so_error = ECONNREFUSED; 1506 goto close; 1507 1508 case TCPS_ESTABLISHED: 1509 case TCPS_FIN_WAIT_1: 1510 case TCPS_FIN_WAIT_2: 1511 case TCPS_CLOSE_WAIT: 1512 so->so_error = ECONNRESET; 1513 close: 1514 tp->t_state = TCPS_CLOSED; 1515 tcpstat.tcps_drops++; 1516 tp = tcp_close(tp); 1517 break; 1518 1519 case TCPS_CLOSING: 1520 case TCPS_LAST_ACK: 1521 tp = tcp_close(tp); 1522 break; 1523 1524 case TCPS_TIME_WAIT: 1525 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1526 ("timewait")); 1527 break; 1528 } 1529 } 1530 goto drop; 1531 } 1532 1533 /* 1534 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1535 * and it's less than ts_recent, drop it. 1536 */ 1537 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1538 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1539 1540 /* Check to see if ts_recent is over 24 days old. */ 1541 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1542 /* 1543 * Invalidate ts_recent. If this segment updates 1544 * ts_recent, the age will be reset later and ts_recent 1545 * will get a valid value. If it does not, setting 1546 * ts_recent to zero will at least satisfy the 1547 * requirement that zero be placed in the timestamp 1548 * echo reply when ts_recent isn't valid. The 1549 * age isn't reset until we get a valid ts_recent 1550 * because we don't want out-of-order segments to be 1551 * dropped when ts_recent is old. 1552 */ 1553 tp->ts_recent = 0; 1554 } else { 1555 tcpstat.tcps_rcvduppack++; 1556 tcpstat.tcps_rcvdupbyte += tlen; 1557 tcpstat.tcps_pawsdrop++; 1558 if (tlen) 1559 goto dropafterack; 1560 goto drop; 1561 } 1562 } 1563 1564 /* 1565 * T/TCP mechanism 1566 * If T/TCP was negotiated and the segment doesn't have CC, 1567 * or if its CC is wrong then drop the segment. 1568 * RST segments do not have to comply with this. 1569 */ 1570 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1571 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1572 goto dropafterack; 1573 1574 /* 1575 * In the SYN-RECEIVED state, validate that the packet belongs to 1576 * this connection before trimming the data to fit the receive 1577 * window. Check the sequence number versus IRS since we know 1578 * the sequence numbers haven't wrapped. This is a partial fix 1579 * for the "LAND" DoS attack. 1580 */ 1581 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1582 rstreason = BANDLIM_RST_OPENPORT; 1583 goto dropwithreset; 1584 } 1585 1586 todrop = tp->rcv_nxt - th->th_seq; 1587 if (todrop > 0) { 1588 if (thflags & TH_SYN) { 1589 thflags &= ~TH_SYN; 1590 th->th_seq++; 1591 if (th->th_urp > 1) 1592 th->th_urp--; 1593 else 1594 thflags &= ~TH_URG; 1595 todrop--; 1596 } 1597 /* 1598 * Following if statement from Stevens, vol. 2, p. 960. 1599 */ 1600 if (todrop > tlen 1601 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1602 /* 1603 * Any valid FIN must be to the left of the window. 1604 * At this point the FIN must be a duplicate or out 1605 * of sequence; drop it. 1606 */ 1607 thflags &= ~TH_FIN; 1608 1609 /* 1610 * Send an ACK to resynchronize and drop any data. 1611 * But keep on processing for RST or ACK. 1612 */ 1613 tp->t_flags |= TF_ACKNOW; 1614 todrop = tlen; 1615 tcpstat.tcps_rcvduppack++; 1616 tcpstat.tcps_rcvdupbyte += todrop; 1617 } else { 1618 tcpstat.tcps_rcvpartduppack++; 1619 tcpstat.tcps_rcvpartdupbyte += todrop; 1620 } 1621 drop_hdrlen += todrop; /* drop from the top afterwards */ 1622 th->th_seq += todrop; 1623 tlen -= todrop; 1624 if (th->th_urp > todrop) 1625 th->th_urp -= todrop; 1626 else { 1627 thflags &= ~TH_URG; 1628 th->th_urp = 0; 1629 } 1630 } 1631 1632 /* 1633 * If new data are received on a connection after the 1634 * user processes are gone, then RST the other end. 1635 */ 1636 if ((so->so_state & SS_NOFDREF) && 1637 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1638 tp = tcp_close(tp); 1639 tcpstat.tcps_rcvafterclose++; 1640 rstreason = BANDLIM_UNLIMITED; 1641 goto dropwithreset; 1642 } 1643 1644 /* 1645 * If segment ends after window, drop trailing data 1646 * (and PUSH and FIN); if nothing left, just ACK. 1647 */ 1648 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1649 if (todrop > 0) { 1650 tcpstat.tcps_rcvpackafterwin++; 1651 if (todrop >= tlen) { 1652 tcpstat.tcps_rcvbyteafterwin += tlen; 1653 /* 1654 * If a new connection request is received 1655 * while in TIME_WAIT, drop the old connection 1656 * and start over if the sequence numbers 1657 * are above the previous ones. 1658 */ 1659 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1660 if (thflags & TH_SYN && 1661 tp->t_state == TCPS_TIME_WAIT && 1662 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1663 tp = tcp_close(tp); 1664 goto findpcb; 1665 } 1666 /* 1667 * If window is closed can only take segments at 1668 * window edge, and have to drop data and PUSH from 1669 * incoming segments. Continue processing, but 1670 * remember to ack. Otherwise, drop segment 1671 * and ack. 1672 */ 1673 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1674 tp->t_flags |= TF_ACKNOW; 1675 tcpstat.tcps_rcvwinprobe++; 1676 } else 1677 goto dropafterack; 1678 } else 1679 tcpstat.tcps_rcvbyteafterwin += todrop; 1680 m_adj(m, -todrop); 1681 tlen -= todrop; 1682 thflags &= ~(TH_PUSH|TH_FIN); 1683 } 1684 1685 /* 1686 * If last ACK falls within this segment's sequence numbers, 1687 * record its timestamp. 1688 * NOTE that the test is modified according to the latest 1689 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1690 */ 1691 if ((to.to_flags & TOF_TS) != 0 && 1692 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1693 tp->ts_recent_age = ticks; 1694 tp->ts_recent = to.to_tsval; 1695 } 1696 1697 /* 1698 * If a SYN is in the window, then this is an 1699 * error and we send an RST and drop the connection. 1700 */ 1701 if (thflags & TH_SYN) { 1702 tp = tcp_drop(tp, ECONNRESET); 1703 rstreason = BANDLIM_UNLIMITED; 1704 goto drop; 1705 } 1706 1707 /* 1708 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1709 * flag is on (half-synchronized state), then queue data for 1710 * later processing; else drop segment and return. 1711 */ 1712 if ((thflags & TH_ACK) == 0) { 1713 if (tp->t_state == TCPS_SYN_RECEIVED || 1714 (tp->t_flags & TF_NEEDSYN)) 1715 goto step6; 1716 else 1717 goto drop; 1718 } 1719 1720 /* 1721 * Ack processing. 1722 */ 1723 switch (tp->t_state) { 1724 1725 /* 1726 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1727 * ESTABLISHED state and continue processing. 1728 * The ACK was checked above. 1729 */ 1730 case TCPS_SYN_RECEIVED: 1731 1732 tcpstat.tcps_connects++; 1733 soisconnected(so); 1734 /* Do window scaling? */ 1735 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1736 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1737 tp->snd_scale = tp->requested_s_scale; 1738 tp->rcv_scale = tp->request_r_scale; 1739 } 1740 /* 1741 * Upon successful completion of 3-way handshake, 1742 * update cache.CC, pass any queued data to the user, 1743 * and advance state appropriately. 1744 */ 1745 if (tcp_do_rfc1644) { 1746 tao.tao_cc = tp->cc_recv; 1747 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1748 tp->cc_recv, 0); 1749 } 1750 /* 1751 * Make transitions: 1752 * SYN-RECEIVED -> ESTABLISHED 1753 * SYN-RECEIVED* -> FIN-WAIT-1 1754 */ 1755 tp->t_starttime = ticks; 1756 if (tp->t_flags & TF_NEEDFIN) { 1757 tp->t_state = TCPS_FIN_WAIT_1; 1758 tp->t_flags &= ~TF_NEEDFIN; 1759 } else { 1760 tp->t_state = TCPS_ESTABLISHED; 1761 callout_reset(tp->tt_keep, tcp_keepidle, 1762 tcp_timer_keep, tp); 1763 } 1764 /* 1765 * If segment contains data or ACK, will call tcp_reass() 1766 * later; if not, do so now to pass queued data to user. 1767 */ 1768 if (tlen == 0 && (thflags & TH_FIN) == 0) 1769 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1770 (struct mbuf *)0); 1771 tp->snd_wl1 = th->th_seq - 1; 1772 /* FALLTHROUGH */ 1773 1774 /* 1775 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1776 * ACKs. If the ack is in the range 1777 * tp->snd_una < th->th_ack <= tp->snd_max 1778 * then advance tp->snd_una to th->th_ack and drop 1779 * data from the retransmission queue. If this ACK reflects 1780 * more up to date window information we update our window information. 1781 */ 1782 case TCPS_ESTABLISHED: 1783 case TCPS_FIN_WAIT_1: 1784 case TCPS_FIN_WAIT_2: 1785 case TCPS_CLOSE_WAIT: 1786 case TCPS_CLOSING: 1787 case TCPS_LAST_ACK: 1788 case TCPS_TIME_WAIT: 1789 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1790 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1791 if (tlen == 0 && tiwin == tp->snd_wnd) { 1792 tcpstat.tcps_rcvdupack++; 1793 /* 1794 * If we have outstanding data (other than 1795 * a window probe), this is a completely 1796 * duplicate ack (ie, window info didn't 1797 * change), the ack is the biggest we've 1798 * seen and we've seen exactly our rexmt 1799 * threshhold of them, assume a packet 1800 * has been dropped and retransmit it. 1801 * Kludge snd_nxt & the congestion 1802 * window so we send only this one 1803 * packet. 1804 * 1805 * We know we're losing at the current 1806 * window size so do congestion avoidance 1807 * (set ssthresh to half the current window 1808 * and pull our congestion window back to 1809 * the new ssthresh). 1810 * 1811 * Dup acks mean that packets have left the 1812 * network (they're now cached at the receiver) 1813 * so bump cwnd by the amount in the receiver 1814 * to keep a constant cwnd packets in the 1815 * network. 1816 */ 1817 if (!callout_active(tp->tt_rexmt) || 1818 th->th_ack != tp->snd_una) 1819 tp->t_dupacks = 0; 1820 else if (++tp->t_dupacks > tcprexmtthresh || 1821 (tcp_do_newreno && 1822 IN_FASTRECOVERY(tp))) { 1823 tp->snd_cwnd += tp->t_maxseg; 1824 (void) tcp_output(tp); 1825 goto drop; 1826 } else if (tp->t_dupacks == tcprexmtthresh) { 1827 tcp_seq onxt = tp->snd_nxt; 1828 u_int win; 1829 if (tcp_do_newreno && 1830 SEQ_LEQ(th->th_ack, 1831 tp->snd_recover)) { 1832 tp->t_dupacks = 0; 1833 break; 1834 } 1835 win = min(tp->snd_wnd, tp->snd_cwnd) / 1836 2 / tp->t_maxseg; 1837 if (win < 2) 1838 win = 2; 1839 tp->snd_ssthresh = win * tp->t_maxseg; 1840 ENTER_FASTRECOVERY(tp); 1841 tp->snd_recover = tp->snd_max; 1842 callout_stop(tp->tt_rexmt); 1843 tp->t_rtttime = 0; 1844 tp->snd_nxt = th->th_ack; 1845 tp->snd_cwnd = tp->t_maxseg; 1846 (void) tcp_output(tp); 1847 KASSERT(tp->snd_limited <= 2, 1848 ("tp->snd_limited too big")); 1849 tp->snd_cwnd = tp->snd_ssthresh + 1850 tp->t_maxseg * 1851 (tp->t_dupacks - tp->snd_limited); 1852 if (SEQ_GT(onxt, tp->snd_nxt)) 1853 tp->snd_nxt = onxt; 1854 goto drop; 1855 } else if (tcp_do_rfc3042) { 1856 u_long oldcwnd = tp->snd_cwnd; 1857 tcp_seq oldsndmax = tp->snd_max; 1858 u_int sent; 1859 KASSERT(tp->t_dupacks == 1 || 1860 tp->t_dupacks == 2, 1861 ("dupacks not 1 or 2")); 1862 if (tp->t_dupacks == 1) 1863 tp->snd_limited = 0; 1864 tp->snd_cwnd = 1865 (tp->snd_nxt - tp->snd_una) + 1866 (tp->t_dupacks - tp->snd_limited) * 1867 tp->t_maxseg; 1868 (void) tcp_output(tp); 1869 sent = tp->snd_max - oldsndmax; 1870 if (sent > tp->t_maxseg) { 1871 KASSERT(tp->snd_limited == 0 && 1872 tp->t_dupacks == 2, 1873 ("sent too much")); 1874 tp->snd_limited = 2; 1875 } else if (sent > 0) 1876 ++tp->snd_limited; 1877 tp->snd_cwnd = oldcwnd; 1878 goto drop; 1879 } 1880 } else 1881 tp->t_dupacks = 0; 1882 break; 1883 } 1884 1885 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1886 1887 /* 1888 * If the congestion window was inflated to account 1889 * for the other side's cached packets, retract it. 1890 */ 1891 if (tcp_do_newreno) { 1892 if (IN_FASTRECOVERY(tp)) { 1893 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1894 tcp_newreno_partial_ack(tp, th); 1895 } else { 1896 /* 1897 * Window inflation should have left us 1898 * with approximately snd_ssthresh 1899 * outstanding data. 1900 * But in case we would be inclined to 1901 * send a burst, better to do it via 1902 * the slow start mechanism. 1903 */ 1904 if (SEQ_GT(th->th_ack + 1905 tp->snd_ssthresh, 1906 tp->snd_max)) 1907 tp->snd_cwnd = tp->snd_max - 1908 th->th_ack + 1909 tp->t_maxseg; 1910 else 1911 tp->snd_cwnd = tp->snd_ssthresh; 1912 } 1913 } 1914 } else { 1915 if (tp->t_dupacks >= tcprexmtthresh && 1916 tp->snd_cwnd > tp->snd_ssthresh) 1917 tp->snd_cwnd = tp->snd_ssthresh; 1918 } 1919 tp->t_dupacks = 0; 1920 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1921 tcpstat.tcps_rcvacktoomuch++; 1922 goto dropafterack; 1923 } 1924 /* 1925 * If we reach this point, ACK is not a duplicate, 1926 * i.e., it ACKs something we sent. 1927 */ 1928 if (tp->t_flags & TF_NEEDSYN) { 1929 /* 1930 * T/TCP: Connection was half-synchronized, and our 1931 * SYN has been ACK'd (so connection is now fully 1932 * synchronized). Go to non-starred state, 1933 * increment snd_una for ACK of SYN, and check if 1934 * we can do window scaling. 1935 */ 1936 tp->t_flags &= ~TF_NEEDSYN; 1937 tp->snd_una++; 1938 /* Do window scaling? */ 1939 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1940 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1941 tp->snd_scale = tp->requested_s_scale; 1942 tp->rcv_scale = tp->request_r_scale; 1943 } 1944 } 1945 1946 process_ACK: 1947 acked = th->th_ack - tp->snd_una; 1948 tcpstat.tcps_rcvackpack++; 1949 tcpstat.tcps_rcvackbyte += acked; 1950 1951 /* 1952 * If we just performed our first retransmit, and the ACK 1953 * arrives within our recovery window, then it was a mistake 1954 * to do the retransmit in the first place. Recover our 1955 * original cwnd and ssthresh, and proceed to transmit where 1956 * we left off. 1957 */ 1958 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1959 ++tcpstat.tcps_sndrexmitbad; 1960 tp->snd_cwnd = tp->snd_cwnd_prev; 1961 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1962 tp->snd_recover = tp->snd_recover_prev; 1963 if (tp->t_flags & TF_WASFRECOVERY) 1964 ENTER_FASTRECOVERY(tp); 1965 tp->snd_nxt = tp->snd_max; 1966 tp->t_badrxtwin = 0; /* XXX probably not required */ 1967 } 1968 1969 /* 1970 * If we have a timestamp reply, update smoothed 1971 * round trip time. If no timestamp is present but 1972 * transmit timer is running and timed sequence 1973 * number was acked, update smoothed round trip time. 1974 * Since we now have an rtt measurement, cancel the 1975 * timer backoff (cf., Phil Karn's retransmit alg.). 1976 * Recompute the initial retransmit timer. 1977 * 1978 * Some boxes send broken timestamp replies 1979 * during the SYN+ACK phase, ignore 1980 * timestamps of 0 or we could calculate a 1981 * huge RTT and blow up the retransmit timer. 1982 */ 1983 if ((to.to_flags & TOF_TS) != 0 && 1984 to.to_tsecr) { 1985 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1986 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1987 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1988 } 1989 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1990 1991 /* 1992 * If all outstanding data is acked, stop retransmit 1993 * timer and remember to restart (more output or persist). 1994 * If there is more data to be acked, restart retransmit 1995 * timer, using current (possibly backed-off) value. 1996 */ 1997 if (th->th_ack == tp->snd_max) { 1998 callout_stop(tp->tt_rexmt); 1999 needoutput = 1; 2000 } else if (!callout_active(tp->tt_persist)) 2001 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2002 tcp_timer_rexmt, tp); 2003 2004 /* 2005 * If no data (only SYN) was ACK'd, 2006 * skip rest of ACK processing. 2007 */ 2008 if (acked == 0) 2009 goto step6; 2010 2011 /* 2012 * When new data is acked, open the congestion window. 2013 * If the window gives us less than ssthresh packets 2014 * in flight, open exponentially (maxseg per packet). 2015 * Otherwise open linearly: maxseg per window 2016 * (maxseg^2 / cwnd per packet). 2017 */ 2018 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 2019 register u_int cw = tp->snd_cwnd; 2020 register u_int incr = tp->t_maxseg; 2021 if (cw > tp->snd_ssthresh) 2022 incr = incr * incr / cw; 2023 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2024 } 2025 if (acked > so->so_snd.sb_cc) { 2026 tp->snd_wnd -= so->so_snd.sb_cc; 2027 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2028 ourfinisacked = 1; 2029 } else { 2030 sbdrop(&so->so_snd, acked); 2031 tp->snd_wnd -= acked; 2032 ourfinisacked = 0; 2033 } 2034 sowwakeup(so); 2035 /* detect una wraparound */ 2036 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 2037 SEQ_GT(tp->snd_una, tp->snd_recover) && 2038 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2039 tp->snd_recover = th->th_ack - 1; 2040 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 2041 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2042 EXIT_FASTRECOVERY(tp); 2043 tp->snd_una = th->th_ack; 2044 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2045 tp->snd_nxt = tp->snd_una; 2046 2047 switch (tp->t_state) { 2048 2049 /* 2050 * In FIN_WAIT_1 STATE in addition to the processing 2051 * for the ESTABLISHED state if our FIN is now acknowledged 2052 * then enter FIN_WAIT_2. 2053 */ 2054 case TCPS_FIN_WAIT_1: 2055 if (ourfinisacked) { 2056 /* 2057 * If we can't receive any more 2058 * data, then closing user can proceed. 2059 * Starting the timer is contrary to the 2060 * specification, but if we don't get a FIN 2061 * we'll hang forever. 2062 */ 2063 /* XXXjl 2064 * we should release the tp also, and use a 2065 * compressed state. 2066 */ 2067 if (so->so_state & SS_CANTRCVMORE) { 2068 soisdisconnected(so); 2069 callout_reset(tp->tt_2msl, tcp_maxidle, 2070 tcp_timer_2msl, tp); 2071 } 2072 tp->t_state = TCPS_FIN_WAIT_2; 2073 } 2074 break; 2075 2076 /* 2077 * In CLOSING STATE in addition to the processing for 2078 * the ESTABLISHED state if the ACK acknowledges our FIN 2079 * then enter the TIME-WAIT state, otherwise ignore 2080 * the segment. 2081 */ 2082 case TCPS_CLOSING: 2083 if (ourfinisacked) { 2084 KASSERT(headlocked, ("headlocked")); 2085 tcp_twstart(tp); 2086 INP_INFO_WUNLOCK(&tcbinfo); 2087 m_freem(m); 2088 return; 2089 } 2090 break; 2091 2092 /* 2093 * In LAST_ACK, we may still be waiting for data to drain 2094 * and/or to be acked, as well as for the ack of our FIN. 2095 * If our FIN is now acknowledged, delete the TCB, 2096 * enter the closed state and return. 2097 */ 2098 case TCPS_LAST_ACK: 2099 if (ourfinisacked) { 2100 tp = tcp_close(tp); 2101 goto drop; 2102 } 2103 break; 2104 2105 /* 2106 * In TIME_WAIT state the only thing that should arrive 2107 * is a retransmission of the remote FIN. Acknowledge 2108 * it and restart the finack timer. 2109 */ 2110 case TCPS_TIME_WAIT: 2111 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2112 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2113 tcp_timer_2msl, tp); 2114 goto dropafterack; 2115 } 2116 } 2117 2118 step6: 2119 /* 2120 * Update window information. 2121 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2122 */ 2123 if ((thflags & TH_ACK) && 2124 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2125 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2126 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2127 /* keep track of pure window updates */ 2128 if (tlen == 0 && 2129 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2130 tcpstat.tcps_rcvwinupd++; 2131 tp->snd_wnd = tiwin; 2132 tp->snd_wl1 = th->th_seq; 2133 tp->snd_wl2 = th->th_ack; 2134 if (tp->snd_wnd > tp->max_sndwnd) 2135 tp->max_sndwnd = tp->snd_wnd; 2136 needoutput = 1; 2137 } 2138 2139 /* 2140 * Process segments with URG. 2141 */ 2142 if ((thflags & TH_URG) && th->th_urp && 2143 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2144 /* 2145 * This is a kludge, but if we receive and accept 2146 * random urgent pointers, we'll crash in 2147 * soreceive. It's hard to imagine someone 2148 * actually wanting to send this much urgent data. 2149 */ 2150 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2151 th->th_urp = 0; /* XXX */ 2152 thflags &= ~TH_URG; /* XXX */ 2153 goto dodata; /* XXX */ 2154 } 2155 /* 2156 * If this segment advances the known urgent pointer, 2157 * then mark the data stream. This should not happen 2158 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2159 * a FIN has been received from the remote side. 2160 * In these states we ignore the URG. 2161 * 2162 * According to RFC961 (Assigned Protocols), 2163 * the urgent pointer points to the last octet 2164 * of urgent data. We continue, however, 2165 * to consider it to indicate the first octet 2166 * of data past the urgent section as the original 2167 * spec states (in one of two places). 2168 */ 2169 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2170 tp->rcv_up = th->th_seq + th->th_urp; 2171 so->so_oobmark = so->so_rcv.sb_cc + 2172 (tp->rcv_up - tp->rcv_nxt) - 1; 2173 if (so->so_oobmark == 0) 2174 so->so_state |= SS_RCVATMARK; 2175 sohasoutofband(so); 2176 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2177 } 2178 /* 2179 * Remove out of band data so doesn't get presented to user. 2180 * This can happen independent of advancing the URG pointer, 2181 * but if two URG's are pending at once, some out-of-band 2182 * data may creep in... ick. 2183 */ 2184 if (th->th_urp <= (u_long)tlen && 2185 !(so->so_options & SO_OOBINLINE)) { 2186 /* hdr drop is delayed */ 2187 tcp_pulloutofband(so, th, m, drop_hdrlen); 2188 } 2189 } else { 2190 /* 2191 * If no out of band data is expected, 2192 * pull receive urgent pointer along 2193 * with the receive window. 2194 */ 2195 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2196 tp->rcv_up = tp->rcv_nxt; 2197 } 2198 dodata: /* XXX */ 2199 KASSERT(headlocked, ("headlocked")); 2200 /* 2201 * Process the segment text, merging it into the TCP sequencing queue, 2202 * and arranging for acknowledgment of receipt if necessary. 2203 * This process logically involves adjusting tp->rcv_wnd as data 2204 * is presented to the user (this happens in tcp_usrreq.c, 2205 * case PRU_RCVD). If a FIN has already been received on this 2206 * connection then we just ignore the text. 2207 */ 2208 if ((tlen || (thflags & TH_FIN)) && 2209 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2210 m_adj(m, drop_hdrlen); /* delayed header drop */ 2211 /* 2212 * Insert segment which includes th into TCP reassembly queue 2213 * with control block tp. Set thflags to whether reassembly now 2214 * includes a segment with FIN. This handles the common case 2215 * inline (segment is the next to be received on an established 2216 * connection, and the queue is empty), avoiding linkage into 2217 * and removal from the queue and repetition of various 2218 * conversions. 2219 * Set DELACK for segments received in order, but ack 2220 * immediately when segments are out of order (so 2221 * fast retransmit can work). 2222 */ 2223 if (th->th_seq == tp->rcv_nxt && 2224 LIST_EMPTY(&tp->t_segq) && 2225 TCPS_HAVEESTABLISHED(tp->t_state)) { 2226 if (DELAY_ACK(tp)) 2227 tp->t_flags |= TF_DELACK; 2228 else 2229 tp->t_flags |= TF_ACKNOW; 2230 tp->rcv_nxt += tlen; 2231 thflags = th->th_flags & TH_FIN; 2232 tcpstat.tcps_rcvpack++; 2233 tcpstat.tcps_rcvbyte += tlen; 2234 ND6_HINT(tp); 2235 if (so->so_state & SS_CANTRCVMORE) 2236 m_freem(m); 2237 else 2238 sbappendstream(&so->so_rcv, m); 2239 sorwakeup(so); 2240 } else { 2241 thflags = tcp_reass(tp, th, &tlen, m); 2242 tp->t_flags |= TF_ACKNOW; 2243 } 2244 2245 /* 2246 * Note the amount of data that peer has sent into 2247 * our window, in order to estimate the sender's 2248 * buffer size. 2249 */ 2250 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2251 } else { 2252 m_freem(m); 2253 thflags &= ~TH_FIN; 2254 } 2255 2256 /* 2257 * If FIN is received ACK the FIN and let the user know 2258 * that the connection is closing. 2259 */ 2260 if (thflags & TH_FIN) { 2261 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2262 socantrcvmore(so); 2263 /* 2264 * If connection is half-synchronized 2265 * (ie NEEDSYN flag on) then delay ACK, 2266 * so it may be piggybacked when SYN is sent. 2267 * Otherwise, since we received a FIN then no 2268 * more input can be expected, send ACK now. 2269 */ 2270 if (tp->t_flags & TF_NEEDSYN) 2271 tp->t_flags |= TF_DELACK; 2272 else 2273 tp->t_flags |= TF_ACKNOW; 2274 tp->rcv_nxt++; 2275 } 2276 switch (tp->t_state) { 2277 2278 /* 2279 * In SYN_RECEIVED and ESTABLISHED STATES 2280 * enter the CLOSE_WAIT state. 2281 */ 2282 case TCPS_SYN_RECEIVED: 2283 tp->t_starttime = ticks; 2284 /*FALLTHROUGH*/ 2285 case TCPS_ESTABLISHED: 2286 tp->t_state = TCPS_CLOSE_WAIT; 2287 break; 2288 2289 /* 2290 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2291 * enter the CLOSING state. 2292 */ 2293 case TCPS_FIN_WAIT_1: 2294 tp->t_state = TCPS_CLOSING; 2295 break; 2296 2297 /* 2298 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2299 * starting the time-wait timer, turning off the other 2300 * standard timers. 2301 */ 2302 case TCPS_FIN_WAIT_2: 2303 KASSERT(headlocked == 1, ("headlocked should be 1")); 2304 tcp_twstart(tp); 2305 INP_INFO_WUNLOCK(&tcbinfo); 2306 return; 2307 2308 /* 2309 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2310 */ 2311 case TCPS_TIME_WAIT: 2312 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2313 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2314 tcp_timer_2msl, tp); 2315 break; 2316 } 2317 } 2318 INP_INFO_WUNLOCK(&tcbinfo); 2319 #ifdef TCPDEBUG 2320 if (so->so_options & SO_DEBUG) 2321 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2322 &tcp_savetcp, 0); 2323 #endif 2324 2325 /* 2326 * Return any desired output. 2327 */ 2328 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2329 (void) tcp_output(tp); 2330 2331 check_delack: 2332 if (tp->t_flags & TF_DELACK) { 2333 tp->t_flags &= ~TF_DELACK; 2334 callout_reset(tp->tt_delack, tcp_delacktime, 2335 tcp_timer_delack, tp); 2336 } 2337 INP_UNLOCK(inp); 2338 return; 2339 2340 dropafterack: 2341 /* 2342 * Generate an ACK dropping incoming segment if it occupies 2343 * sequence space, where the ACK reflects our state. 2344 * 2345 * We can now skip the test for the RST flag since all 2346 * paths to this code happen after packets containing 2347 * RST have been dropped. 2348 * 2349 * In the SYN-RECEIVED state, don't send an ACK unless the 2350 * segment we received passes the SYN-RECEIVED ACK test. 2351 * If it fails send a RST. This breaks the loop in the 2352 * "LAND" DoS attack, and also prevents an ACK storm 2353 * between two listening ports that have been sent forged 2354 * SYN segments, each with the source address of the other. 2355 */ 2356 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2357 (SEQ_GT(tp->snd_una, th->th_ack) || 2358 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2359 rstreason = BANDLIM_RST_OPENPORT; 2360 goto dropwithreset; 2361 } 2362 #ifdef TCPDEBUG 2363 if (so->so_options & SO_DEBUG) 2364 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2365 &tcp_savetcp, 0); 2366 #endif 2367 KASSERT(headlocked, ("headlocked should be 1")); 2368 INP_INFO_WUNLOCK(&tcbinfo); 2369 m_freem(m); 2370 tp->t_flags |= TF_ACKNOW; 2371 (void) tcp_output(tp); 2372 INP_UNLOCK(inp); 2373 return; 2374 2375 dropwithreset: 2376 /* 2377 * Generate a RST, dropping incoming segment. 2378 * Make ACK acceptable to originator of segment. 2379 * Don't bother to respond if destination was broadcast/multicast. 2380 */ 2381 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2382 goto drop; 2383 if (isipv6) { 2384 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2385 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2386 goto drop; 2387 } else { 2388 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2389 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2390 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2391 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2392 goto drop; 2393 } 2394 /* IPv6 anycast check is done at tcp6_input() */ 2395 2396 /* 2397 * Perform bandwidth limiting. 2398 */ 2399 if (badport_bandlim(rstreason) < 0) 2400 goto drop; 2401 2402 #ifdef TCPDEBUG 2403 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2404 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2405 &tcp_savetcp, 0); 2406 #endif 2407 2408 if (thflags & TH_ACK) 2409 /* mtod() below is safe as long as hdr dropping is delayed */ 2410 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2411 TH_RST); 2412 else { 2413 if (thflags & TH_SYN) 2414 tlen++; 2415 /* mtod() below is safe as long as hdr dropping is delayed */ 2416 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2417 (tcp_seq)0, TH_RST|TH_ACK); 2418 } 2419 2420 if (tp) 2421 INP_UNLOCK(inp); 2422 if (headlocked) 2423 INP_INFO_WUNLOCK(&tcbinfo); 2424 return; 2425 2426 drop: 2427 /* 2428 * Drop space held by incoming segment and return. 2429 */ 2430 #ifdef TCPDEBUG 2431 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2432 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2433 &tcp_savetcp, 0); 2434 #endif 2435 if (tp) 2436 INP_UNLOCK(inp); 2437 m_freem(m); 2438 if (headlocked) 2439 INP_INFO_WUNLOCK(&tcbinfo); 2440 return; 2441 } 2442 2443 /* 2444 * Parse TCP options and place in tcpopt. 2445 */ 2446 static void 2447 tcp_dooptions(to, cp, cnt, is_syn) 2448 struct tcpopt *to; 2449 u_char *cp; 2450 int cnt; 2451 int is_syn; 2452 { 2453 int opt, optlen; 2454 2455 to->to_flags = 0; 2456 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2457 opt = cp[0]; 2458 if (opt == TCPOPT_EOL) 2459 break; 2460 if (opt == TCPOPT_NOP) 2461 optlen = 1; 2462 else { 2463 if (cnt < 2) 2464 break; 2465 optlen = cp[1]; 2466 if (optlen < 2 || optlen > cnt) 2467 break; 2468 } 2469 switch (opt) { 2470 case TCPOPT_MAXSEG: 2471 if (optlen != TCPOLEN_MAXSEG) 2472 continue; 2473 if (!is_syn) 2474 continue; 2475 to->to_flags |= TOF_MSS; 2476 bcopy((char *)cp + 2, 2477 (char *)&to->to_mss, sizeof(to->to_mss)); 2478 to->to_mss = ntohs(to->to_mss); 2479 break; 2480 case TCPOPT_WINDOW: 2481 if (optlen != TCPOLEN_WINDOW) 2482 continue; 2483 if (! is_syn) 2484 continue; 2485 to->to_flags |= TOF_SCALE; 2486 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2487 break; 2488 case TCPOPT_TIMESTAMP: 2489 if (optlen != TCPOLEN_TIMESTAMP) 2490 continue; 2491 to->to_flags |= TOF_TS; 2492 bcopy((char *)cp + 2, 2493 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2494 to->to_tsval = ntohl(to->to_tsval); 2495 bcopy((char *)cp + 6, 2496 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2497 to->to_tsecr = ntohl(to->to_tsecr); 2498 break; 2499 case TCPOPT_CC: 2500 if (optlen != TCPOLEN_CC) 2501 continue; 2502 to->to_flags |= TOF_CC; 2503 bcopy((char *)cp + 2, 2504 (char *)&to->to_cc, sizeof(to->to_cc)); 2505 to->to_cc = ntohl(to->to_cc); 2506 break; 2507 case TCPOPT_CCNEW: 2508 if (optlen != TCPOLEN_CC) 2509 continue; 2510 if (!is_syn) 2511 continue; 2512 to->to_flags |= TOF_CCNEW; 2513 bcopy((char *)cp + 2, 2514 (char *)&to->to_cc, sizeof(to->to_cc)); 2515 to->to_cc = ntohl(to->to_cc); 2516 break; 2517 case TCPOPT_CCECHO: 2518 if (optlen != TCPOLEN_CC) 2519 continue; 2520 if (!is_syn) 2521 continue; 2522 to->to_flags |= TOF_CCECHO; 2523 bcopy((char *)cp + 2, 2524 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2525 to->to_ccecho = ntohl(to->to_ccecho); 2526 break; 2527 default: 2528 continue; 2529 } 2530 } 2531 } 2532 2533 /* 2534 * Pull out of band byte out of a segment so 2535 * it doesn't appear in the user's data queue. 2536 * It is still reflected in the segment length for 2537 * sequencing purposes. 2538 */ 2539 static void 2540 tcp_pulloutofband(so, th, m, off) 2541 struct socket *so; 2542 struct tcphdr *th; 2543 register struct mbuf *m; 2544 int off; /* delayed to be droped hdrlen */ 2545 { 2546 int cnt = off + th->th_urp - 1; 2547 2548 while (cnt >= 0) { 2549 if (m->m_len > cnt) { 2550 char *cp = mtod(m, caddr_t) + cnt; 2551 struct tcpcb *tp = sototcpcb(so); 2552 2553 tp->t_iobc = *cp; 2554 tp->t_oobflags |= TCPOOB_HAVEDATA; 2555 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2556 m->m_len--; 2557 if (m->m_flags & M_PKTHDR) 2558 m->m_pkthdr.len--; 2559 return; 2560 } 2561 cnt -= m->m_len; 2562 m = m->m_next; 2563 if (m == 0) 2564 break; 2565 } 2566 panic("tcp_pulloutofband"); 2567 } 2568 2569 /* 2570 * Collect new round-trip time estimate 2571 * and update averages and current timeout. 2572 */ 2573 static void 2574 tcp_xmit_timer(tp, rtt) 2575 register struct tcpcb *tp; 2576 int rtt; 2577 { 2578 register int delta; 2579 2580 tcpstat.tcps_rttupdated++; 2581 tp->t_rttupdated++; 2582 if (tp->t_srtt != 0) { 2583 /* 2584 * srtt is stored as fixed point with 5 bits after the 2585 * binary point (i.e., scaled by 8). The following magic 2586 * is equivalent to the smoothing algorithm in rfc793 with 2587 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2588 * point). Adjust rtt to origin 0. 2589 */ 2590 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2591 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2592 2593 if ((tp->t_srtt += delta) <= 0) 2594 tp->t_srtt = 1; 2595 2596 /* 2597 * We accumulate a smoothed rtt variance (actually, a 2598 * smoothed mean difference), then set the retransmit 2599 * timer to smoothed rtt + 4 times the smoothed variance. 2600 * rttvar is stored as fixed point with 4 bits after the 2601 * binary point (scaled by 16). The following is 2602 * equivalent to rfc793 smoothing with an alpha of .75 2603 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2604 * rfc793's wired-in beta. 2605 */ 2606 if (delta < 0) 2607 delta = -delta; 2608 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2609 if ((tp->t_rttvar += delta) <= 0) 2610 tp->t_rttvar = 1; 2611 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2612 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2613 } else { 2614 /* 2615 * No rtt measurement yet - use the unsmoothed rtt. 2616 * Set the variance to half the rtt (so our first 2617 * retransmit happens at 3*rtt). 2618 */ 2619 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2620 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2621 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2622 } 2623 tp->t_rtttime = 0; 2624 tp->t_rxtshift = 0; 2625 2626 /* 2627 * the retransmit should happen at rtt + 4 * rttvar. 2628 * Because of the way we do the smoothing, srtt and rttvar 2629 * will each average +1/2 tick of bias. When we compute 2630 * the retransmit timer, we want 1/2 tick of rounding and 2631 * 1 extra tick because of +-1/2 tick uncertainty in the 2632 * firing of the timer. The bias will give us exactly the 2633 * 1.5 tick we need. But, because the bias is 2634 * statistical, we have to test that we don't drop below 2635 * the minimum feasible timer (which is 2 ticks). 2636 */ 2637 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2638 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2639 2640 /* 2641 * We received an ack for a packet that wasn't retransmitted; 2642 * it is probably safe to discard any error indications we've 2643 * received recently. This isn't quite right, but close enough 2644 * for now (a route might have failed after we sent a segment, 2645 * and the return path might not be symmetrical). 2646 */ 2647 tp->t_softerror = 0; 2648 } 2649 2650 /* 2651 * Determine a reasonable value for maxseg size. 2652 * If the route is known, check route for mtu. 2653 * If none, use an mss that can be handled on the outgoing 2654 * interface without forcing IP to fragment; if bigger than 2655 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2656 * to utilize large mbufs. If no route is found, route has no mtu, 2657 * or the destination isn't local, use a default, hopefully conservative 2658 * size (usually 512 or the default IP max size, but no more than the mtu 2659 * of the interface), as we can't discover anything about intervening 2660 * gateways or networks. We also initialize the congestion/slow start 2661 * window to be a single segment if the destination isn't local. 2662 * While looking at the routing entry, we also initialize other path-dependent 2663 * parameters from pre-set or cached values in the routing entry. 2664 * 2665 * Also take into account the space needed for options that we 2666 * send regularly. Make maxseg shorter by that amount to assure 2667 * that we can send maxseg amount of data even when the options 2668 * are present. Store the upper limit of the length of options plus 2669 * data in maxopd. 2670 * 2671 * 2672 * In case of T/TCP, we call this routine during implicit connection 2673 * setup as well (offer = -1), to initialize maxseg from the cached 2674 * MSS of our peer. 2675 * 2676 * NOTE that this routine is only called when we process an incoming 2677 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2678 */ 2679 void 2680 tcp_mss(tp, offer) 2681 struct tcpcb *tp; 2682 int offer; 2683 { 2684 int rtt, mss; 2685 u_long bufsize; 2686 u_long maxmtu; 2687 struct inpcb *inp = tp->t_inpcb; 2688 struct socket *so; 2689 struct hc_metrics_lite metrics; 2690 struct rmxp_tao tao; 2691 int origoffer = offer; 2692 #ifdef INET6 2693 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2694 size_t min_protoh = isipv6 ? 2695 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2696 sizeof (struct tcpiphdr); 2697 #else 2698 const size_t min_protoh = sizeof(struct tcpiphdr); 2699 #endif 2700 bzero(&tao, sizeof(tao)); 2701 2702 /* initialize */ 2703 #ifdef INET6 2704 if (isipv6) { 2705 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2706 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2707 } else 2708 #endif 2709 { 2710 maxmtu = tcp_maxmtu(&inp->inp_inc); 2711 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2712 } 2713 so = inp->inp_socket; 2714 2715 /* 2716 * no route to sender, take default mss and return 2717 */ 2718 if (maxmtu == 0) 2719 return; 2720 2721 /* what have we got? */ 2722 switch (offer) { 2723 case 0: 2724 /* 2725 * Offer == 0 means that there was no MSS on the SYN 2726 * segment, in this case we use tcp_mssdflt. 2727 */ 2728 offer = 2729 #ifdef INET6 2730 isipv6 ? tcp_v6mssdflt : 2731 #endif 2732 tcp_mssdflt; 2733 break; 2734 2735 case -1: 2736 /* 2737 * Offer == -1 means that we didn't receive SYN yet, 2738 * use cached value in that case; 2739 */ 2740 if (tcp_do_rfc1644) 2741 tcp_hc_gettao(&inp->inp_inc, &tao); 2742 if (tao.tao_mssopt != 0) 2743 offer = tao.tao_mssopt; 2744 /* FALLTHROUGH */ 2745 2746 default: 2747 /* 2748 * Prevent DoS attack with too small MSS. Round up 2749 * to at least minmss. 2750 */ 2751 offer = max(offer, tcp_minmss); 2752 /* 2753 * Sanity check: make sure that maxopd will be large 2754 * enough to allow some data on segments even if the 2755 * all the option space is used (40bytes). Otherwise 2756 * funny things may happen in tcp_output. 2757 */ 2758 offer = max(offer, 64); 2759 if (tcp_do_rfc1644) 2760 tcp_hc_updatetao(&inp->inp_inc, 2761 TCP_HC_TAO_MSSOPT, 0, offer); 2762 } 2763 2764 /* 2765 * rmx information is now retrieved from tcp_hostcache 2766 */ 2767 tcp_hc_get(&inp->inp_inc, &metrics); 2768 2769 /* 2770 * if there's a discovered mtu int tcp hostcache, use it 2771 * else, use the link mtu. 2772 */ 2773 if (metrics.rmx_mtu) 2774 mss = metrics.rmx_mtu - min_protoh; 2775 else { 2776 #ifdef INET6 2777 if (isipv6) { 2778 mss = maxmtu - min_protoh; 2779 if (!path_mtu_discovery && 2780 !in6_localaddr(&inp->in6p_faddr)) 2781 mss = min(mss, tcp_v6mssdflt); 2782 } else 2783 #endif 2784 { 2785 mss = maxmtu - min_protoh; 2786 if (!path_mtu_discovery && 2787 !in_localaddr(inp->inp_faddr)) 2788 mss = min(mss, tcp_mssdflt); 2789 } 2790 } 2791 mss = min(mss, offer); 2792 2793 /* 2794 * maxopd stores the maximum length of data AND options 2795 * in a segment; maxseg is the amount of data in a normal 2796 * segment. We need to store this value (maxopd) apart 2797 * from maxseg, because now every segment carries options 2798 * and thus we normally have somewhat less data in segments. 2799 */ 2800 tp->t_maxopd = mss; 2801 2802 /* 2803 * In case of T/TCP, origoffer==-1 indicates, that no segments 2804 * were received yet. In this case we just guess, otherwise 2805 * we do the same as before T/TCP. 2806 */ 2807 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2808 (origoffer == -1 || 2809 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2810 mss -= TCPOLEN_TSTAMP_APPA; 2811 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2812 (origoffer == -1 || 2813 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2814 mss -= TCPOLEN_CC_APPA; 2815 tp->t_maxseg = mss; 2816 2817 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2818 if (mss > MCLBYTES) 2819 mss &= ~(MCLBYTES-1); 2820 #else 2821 if (mss > MCLBYTES) 2822 mss = mss / MCLBYTES * MCLBYTES; 2823 #endif 2824 tp->t_maxseg = mss; 2825 2826 /* 2827 * If there's a pipesize, change the socket buffer to that size, 2828 * don't change if sb_hiwat is different than default (then it 2829 * has been changed on purpose with setsockopt). 2830 * Make the socket buffers an integral number of mss units; 2831 * if the mss is larger than the socket buffer, decrease the mss. 2832 */ 2833 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2834 bufsize = metrics.rmx_sendpipe; 2835 else 2836 bufsize = so->so_snd.sb_hiwat; 2837 if (bufsize < mss) 2838 mss = bufsize; 2839 else { 2840 bufsize = roundup(bufsize, mss); 2841 if (bufsize > sb_max) 2842 bufsize = sb_max; 2843 if (bufsize > so->so_snd.sb_hiwat) 2844 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2845 } 2846 tp->t_maxseg = mss; 2847 2848 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2849 bufsize = metrics.rmx_recvpipe; 2850 else 2851 bufsize = so->so_rcv.sb_hiwat; 2852 if (bufsize > mss) { 2853 bufsize = roundup(bufsize, mss); 2854 if (bufsize > sb_max) 2855 bufsize = sb_max; 2856 if (bufsize > so->so_rcv.sb_hiwat) 2857 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2858 } 2859 /* 2860 * While we're here, check the others too 2861 */ 2862 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2863 tp->t_srtt = rtt; 2864 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2865 tcpstat.tcps_usedrtt++; 2866 if (metrics.rmx_rttvar) { 2867 tp->t_rttvar = metrics.rmx_rttvar; 2868 tcpstat.tcps_usedrttvar++; 2869 } else { 2870 /* default variation is +- 1 rtt */ 2871 tp->t_rttvar = 2872 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2873 } 2874 TCPT_RANGESET(tp->t_rxtcur, 2875 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2876 tp->t_rttmin, TCPTV_REXMTMAX); 2877 } 2878 if (metrics.rmx_ssthresh) { 2879 /* 2880 * There's some sort of gateway or interface 2881 * buffer limit on the path. Use this to set 2882 * the slow start threshhold, but set the 2883 * threshold to no less than 2*mss. 2884 */ 2885 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2886 tcpstat.tcps_usedssthresh++; 2887 } 2888 if (metrics.rmx_bandwidth) 2889 tp->snd_bandwidth = metrics.rmx_bandwidth; 2890 2891 /* 2892 * Set the slow-start flight size depending on whether this 2893 * is a local network or not. 2894 * 2895 * Extend this so we cache the cwnd too and retrieve it here. 2896 * Make cwnd even bigger than RFC3390 suggests but only if we 2897 * have previous experience with the remote host. Be careful 2898 * not make cwnd bigger than remote receive window or our own 2899 * send socket buffer. Maybe put some additional upper bound 2900 * on the retrieved cwnd. Should do incremental updates to 2901 * hostcache when cwnd collapses so next connection doesn't 2902 * overloads the path again. 2903 * 2904 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2905 * We currently check only in syncache_socket for that. 2906 */ 2907 #define TCP_METRICS_CWND 2908 #ifdef TCP_METRICS_CWND 2909 if (metrics.rmx_cwnd) 2910 tp->snd_cwnd = max(mss, 2911 min(metrics.rmx_cwnd / 2, 2912 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2913 else 2914 #endif 2915 if (tcp_do_rfc3390) 2916 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2917 #ifdef INET6 2918 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2919 (!isipv6 && in_localaddr(inp->inp_faddr))) 2920 #else 2921 else if (in_localaddr(inp->inp_faddr)) 2922 #endif 2923 tp->snd_cwnd = mss * ss_fltsz_local; 2924 else 2925 tp->snd_cwnd = mss * ss_fltsz; 2926 } 2927 2928 /* 2929 * Determine the MSS option to send on an outgoing SYN. 2930 */ 2931 int 2932 tcp_mssopt(inc) 2933 struct in_conninfo *inc; 2934 { 2935 int mss = 0; 2936 u_long maxmtu = 0; 2937 u_long thcmtu = 0; 2938 size_t min_protoh; 2939 #ifdef INET6 2940 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2941 #endif 2942 2943 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2944 2945 #ifdef INET6 2946 if (isipv6) { 2947 mss = tcp_v6mssdflt; 2948 maxmtu = tcp_maxmtu6(inc); 2949 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2950 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2951 } else 2952 #endif 2953 { 2954 mss = tcp_mssdflt; 2955 maxmtu = tcp_maxmtu(inc); 2956 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2957 min_protoh = sizeof(struct tcpiphdr); 2958 } 2959 if (maxmtu && thcmtu) 2960 mss = min(maxmtu, thcmtu) - min_protoh; 2961 else if (maxmtu || thcmtu) 2962 mss = max(maxmtu, thcmtu) - min_protoh; 2963 2964 return (mss); 2965 } 2966 2967 2968 /* 2969 * On a partial ack arrives, force the retransmission of the 2970 * next unacknowledged segment. Do not clear tp->t_dupacks. 2971 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2972 * be started again. 2973 */ 2974 static void 2975 tcp_newreno_partial_ack(tp, th) 2976 struct tcpcb *tp; 2977 struct tcphdr *th; 2978 { 2979 tcp_seq onxt = tp->snd_nxt; 2980 u_long ocwnd = tp->snd_cwnd; 2981 2982 callout_stop(tp->tt_rexmt); 2983 tp->t_rtttime = 0; 2984 tp->snd_nxt = th->th_ack; 2985 /* 2986 * Set snd_cwnd to one segment beyond acknowledged offset. 2987 * (tp->snd_una has not yet been updated when this function is called.) 2988 */ 2989 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2990 tp->t_flags |= TF_ACKNOW; 2991 (void) tcp_output(tp); 2992 tp->snd_cwnd = ocwnd; 2993 if (SEQ_GT(onxt, tp->snd_nxt)) 2994 tp->snd_nxt = onxt; 2995 /* 2996 * Partial window deflation. Relies on fact that tp->snd_una 2997 * not updated yet. 2998 */ 2999 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3000 } 3001 3002 /* 3003 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3004 * looking for a pcb in the listen state. Returns 0 otherwise. 3005 */ 3006 static int 3007 tcp_timewait(tw, to, th, m, tlen) 3008 struct tcptw *tw; 3009 struct tcpopt *to; 3010 struct tcphdr *th; 3011 struct mbuf *m; 3012 int tlen; 3013 { 3014 int thflags; 3015 tcp_seq seq; 3016 #ifdef INET6 3017 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3018 #else 3019 const int isipv6 = 0; 3020 #endif 3021 3022 thflags = th->th_flags; 3023 3024 /* 3025 * NOTE: for FIN_WAIT_2 (to be added later), 3026 * must validate sequence number before accepting RST 3027 */ 3028 3029 /* 3030 * If the segment contains RST: 3031 * Drop the segment - see Stevens, vol. 2, p. 964 and 3032 * RFC 1337. 3033 */ 3034 if (thflags & TH_RST) 3035 goto drop; 3036 3037 /* 3038 * If segment contains a SYN and CC [not CC.NEW] option: 3039 * if connection duration > MSL, drop packet and send RST; 3040 * 3041 * if SEG.CC > CCrecv then is new SYN. 3042 * Complete close and delete TCPCB. Then reprocess 3043 * segment, hoping to find new TCPCB in LISTEN state; 3044 * 3045 * else must be old SYN; drop it. 3046 * else do normal processing. 3047 */ 3048 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3049 if ((ticks - tw->t_starttime) > tcp_msl) 3050 goto reset; 3051 if (CC_GT(to->to_cc, tw->cc_recv)) { 3052 (void) tcp_twclose(tw, 0); 3053 return (1); 3054 } 3055 goto drop; 3056 } 3057 3058 #if 0 3059 /* PAWS not needed at the moment */ 3060 /* 3061 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3062 * and it's less than ts_recent, drop it. 3063 */ 3064 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3065 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3066 if ((thflags & TH_ACK) == 0) 3067 goto drop; 3068 goto ack; 3069 } 3070 /* 3071 * ts_recent is never updated because we never accept new segments. 3072 */ 3073 #endif 3074 3075 /* 3076 * If a new connection request is received 3077 * while in TIME_WAIT, drop the old connection 3078 * and start over if the sequence numbers 3079 * are above the previous ones. 3080 */ 3081 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3082 (void) tcp_twclose(tw, 0); 3083 return (1); 3084 } 3085 3086 /* 3087 * Drop the the segment if it does not contain an ACK. 3088 */ 3089 if ((thflags & TH_ACK) == 0) 3090 goto drop; 3091 3092 /* 3093 * Reset the 2MSL timer if this is a duplicate FIN. 3094 */ 3095 if (thflags & TH_FIN) { 3096 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3097 if (seq + 1 == tw->rcv_nxt) 3098 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3099 } 3100 3101 /* 3102 * Acknowledge the segment if it has data or is not a duplicate ACK. 3103 */ 3104 if (thflags != TH_ACK || tlen != 0 || 3105 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3106 tcp_twrespond(tw, NULL, m, TH_ACK); 3107 goto drop; 3108 3109 reset: 3110 /* 3111 * Generate a RST, dropping incoming segment. 3112 * Make ACK acceptable to originator of segment. 3113 * Don't bother to respond if destination was broadcast/multicast. 3114 */ 3115 if (m->m_flags & (M_BCAST|M_MCAST)) 3116 goto drop; 3117 if (isipv6) { 3118 struct ip6_hdr *ip6; 3119 3120 /* IPv6 anycast check is done at tcp6_input() */ 3121 ip6 = mtod(m, struct ip6_hdr *); 3122 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3123 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3124 goto drop; 3125 } else { 3126 struct ip *ip; 3127 3128 ip = mtod(m, struct ip *); 3129 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3130 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3131 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3132 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3133 goto drop; 3134 } 3135 if (thflags & TH_ACK) { 3136 tcp_respond(NULL, 3137 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3138 } else { 3139 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3140 tcp_respond(NULL, 3141 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3142 } 3143 INP_UNLOCK(tw->tw_inpcb); 3144 return (0); 3145 3146 drop: 3147 INP_UNLOCK(tw->tw_inpcb); 3148 m_freem(m); 3149 return (0); 3150 } 3151