tcp_input.c revision 125784
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: head/sys/netinet/tcp_input.c 125784 2004-02-13 19:14:16Z mlaier $ 35 */ 36 37#include "opt_ipfw.h" /* for ipfw_fwd */ 38#include "opt_inet.h" 39#include "opt_inet6.h" 40#include "opt_ipsec.h" 41#include "opt_mac.h" 42#include "opt_tcpdebug.h" 43#include "opt_tcp_input.h" 44 45#include <sys/param.h> 46#include <sys/kernel.h> 47#include <sys/mac.h> 48#include <sys/malloc.h> 49#include <sys/mbuf.h> 50#include <sys/proc.h> /* for proc0 declaration */ 51#include <sys/protosw.h> 52#include <sys/signalvar.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55#include <sys/sysctl.h> 56#include <sys/syslog.h> 57#include <sys/systm.h> 58 59#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 60 61#include <net/if.h> 62#include <net/route.h> 63 64#include <netinet/in.h> 65#include <netinet/in_pcb.h> 66#include <netinet/in_systm.h> 67#include <netinet/in_var.h> 68#include <netinet/ip.h> 69#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 70#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 71#include <netinet/ip_var.h> 72#include <netinet/ip6.h> 73#include <netinet/icmp6.h> 74#include <netinet6/in6_pcb.h> 75#include <netinet6/ip6_var.h> 76#include <netinet6/nd6.h> 77#include <netinet/tcp.h> 78#include <netinet/tcp_fsm.h> 79#include <netinet/tcp_seq.h> 80#include <netinet/tcp_timer.h> 81#include <netinet/tcp_var.h> 82#include <netinet6/tcp6_var.h> 83#include <netinet/tcpip.h> 84#ifdef TCPDEBUG 85#include <netinet/tcp_debug.h> 86#endif /* TCPDEBUG */ 87 88#ifdef FAST_IPSEC 89#include <netipsec/ipsec.h> 90#include <netipsec/ipsec6.h> 91#endif /*FAST_IPSEC*/ 92 93#ifdef IPSEC 94#include <netinet6/ipsec.h> 95#include <netinet6/ipsec6.h> 96#include <netkey/key.h> 97#endif /*IPSEC*/ 98 99#include <machine/in_cksum.h> 100 101MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 102 103static const int tcprexmtthresh = 3; 104tcp_cc tcp_ccgen; 105 106struct tcpstat tcpstat; 107SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 109 110static int log_in_vain = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 112 &log_in_vain, 0, "Log all incoming TCP connections"); 113 114static int blackhole = 0; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 116 &blackhole, 0, "Do not send RST when dropping refused connections"); 117 118int tcp_delack_enabled = 1; 119SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 120 &tcp_delack_enabled, 0, 121 "Delay ACK to try and piggyback it onto a data packet"); 122 123#ifdef TCP_DROP_SYNFIN 124static int drop_synfin = 0; 125SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 127#endif 128 129static int tcp_do_rfc3042 = 1; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 131 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 132 133static int tcp_do_rfc3390 = 1; 134SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 135 &tcp_do_rfc3390, 0, 136 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 137 138struct inpcbhead tcb; 139#define tcb6 tcb /* for KAME src sync over BSD*'s */ 140struct inpcbinfo tcbinfo; 141struct mtx *tcbinfo_mtx; 142 143static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 144static void tcp_pulloutofband(struct socket *, 145 struct tcphdr *, struct mbuf *, int); 146static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 147 struct mbuf *); 148static void tcp_xmit_timer(struct tcpcb *, int); 149static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 150static int tcp_timewait(struct tcptw *, struct tcpopt *, 151 struct tcphdr *, struct mbuf *, int); 152 153/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 154#ifdef INET6 155#define ND6_HINT(tp) \ 156do { \ 157 if ((tp) && (tp)->t_inpcb && \ 158 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 159 nd6_nud_hint(NULL, NULL, 0); \ 160} while (0) 161#else 162#define ND6_HINT(tp) 163#endif 164 165/* 166 * Indicate whether this ack should be delayed. We can delay the ack if 167 * - there is no delayed ack timer in progress and 168 * - our last ack wasn't a 0-sized window. We never want to delay 169 * the ack that opens up a 0-sized window and 170 * - delayed acks are enabled or 171 * - this is a half-synchronized T/TCP connection. 172 */ 173#define DELAY_ACK(tp) \ 174 ((!callout_active(tp->tt_delack) && \ 175 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 176 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 177 178static int 179tcp_reass(tp, th, tlenp, m) 180 register struct tcpcb *tp; 181 register struct tcphdr *th; 182 int *tlenp; 183 struct mbuf *m; 184{ 185 struct tseg_qent *q; 186 struct tseg_qent *p = NULL; 187 struct tseg_qent *nq; 188 struct tseg_qent *te; 189 struct socket *so = tp->t_inpcb->inp_socket; 190 int flags; 191 192 /* 193 * Call with th==0 after become established to 194 * force pre-ESTABLISHED data up to user socket. 195 */ 196 if (th == 0) 197 goto present; 198 199 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 200 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 201 M_NOWAIT); 202 if (te == NULL) { 203 tcpstat.tcps_rcvmemdrop++; 204 m_freem(m); 205 return (0); 206 } 207 208 /* 209 * Find a segment which begins after this one does. 210 */ 211 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 212 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 213 break; 214 p = q; 215 } 216 217 /* 218 * If there is a preceding segment, it may provide some of 219 * our data already. If so, drop the data from the incoming 220 * segment. If it provides all of our data, drop us. 221 */ 222 if (p != NULL) { 223 register int i; 224 /* conversion to int (in i) handles seq wraparound */ 225 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 226 if (i > 0) { 227 if (i >= *tlenp) { 228 tcpstat.tcps_rcvduppack++; 229 tcpstat.tcps_rcvdupbyte += *tlenp; 230 m_freem(m); 231 FREE(te, M_TSEGQ); 232 /* 233 * Try to present any queued data 234 * at the left window edge to the user. 235 * This is needed after the 3-WHS 236 * completes. 237 */ 238 goto present; /* ??? */ 239 } 240 m_adj(m, i); 241 *tlenp -= i; 242 th->th_seq += i; 243 } 244 } 245 tcpstat.tcps_rcvoopack++; 246 tcpstat.tcps_rcvoobyte += *tlenp; 247 248 /* 249 * While we overlap succeeding segments trim them or, 250 * if they are completely covered, dequeue them. 251 */ 252 while (q) { 253 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 254 if (i <= 0) 255 break; 256 if (i < q->tqe_len) { 257 q->tqe_th->th_seq += i; 258 q->tqe_len -= i; 259 m_adj(q->tqe_m, i); 260 break; 261 } 262 263 nq = LIST_NEXT(q, tqe_q); 264 LIST_REMOVE(q, tqe_q); 265 m_freem(q->tqe_m); 266 FREE(q, M_TSEGQ); 267 q = nq; 268 } 269 270 /* Insert the new segment queue entry into place. */ 271 te->tqe_m = m; 272 te->tqe_th = th; 273 te->tqe_len = *tlenp; 274 275 if (p == NULL) { 276 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 277 } else { 278 LIST_INSERT_AFTER(p, te, tqe_q); 279 } 280 281present: 282 /* 283 * Present data to user, advancing rcv_nxt through 284 * completed sequence space. 285 */ 286 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 287 return (0); 288 q = LIST_FIRST(&tp->t_segq); 289 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 290 return (0); 291 do { 292 tp->rcv_nxt += q->tqe_len; 293 flags = q->tqe_th->th_flags & TH_FIN; 294 nq = LIST_NEXT(q, tqe_q); 295 LIST_REMOVE(q, tqe_q); 296 if (so->so_state & SS_CANTRCVMORE) 297 m_freem(q->tqe_m); 298 else 299 sbappendstream(&so->so_rcv, q->tqe_m); 300 FREE(q, M_TSEGQ); 301 q = nq; 302 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 303 ND6_HINT(tp); 304 sorwakeup(so); 305 return (flags); 306} 307 308/* 309 * TCP input routine, follows pages 65-76 of the 310 * protocol specification dated September, 1981 very closely. 311 */ 312#ifdef INET6 313int 314tcp6_input(mp, offp, proto) 315 struct mbuf **mp; 316 int *offp, proto; 317{ 318 register struct mbuf *m = *mp; 319 struct in6_ifaddr *ia6; 320 321 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 322 323 /* 324 * draft-itojun-ipv6-tcp-to-anycast 325 * better place to put this in? 326 */ 327 ia6 = ip6_getdstifaddr(m); 328 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 329 struct ip6_hdr *ip6; 330 331 ip6 = mtod(m, struct ip6_hdr *); 332 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 333 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 334 return IPPROTO_DONE; 335 } 336 337 tcp_input(m, *offp); 338 return IPPROTO_DONE; 339} 340#endif 341 342void 343tcp_input(m, off0) 344 register struct mbuf *m; 345 int off0; 346{ 347 register struct tcphdr *th; 348 register struct ip *ip = NULL; 349 register struct ipovly *ipov; 350 register struct inpcb *inp = NULL; 351 u_char *optp = NULL; 352 int optlen = 0; 353 int len, tlen, off; 354 int drop_hdrlen; 355 register struct tcpcb *tp = 0; 356 register int thflags; 357 struct socket *so = 0; 358 int todrop, acked, ourfinisacked, needoutput = 0; 359 u_long tiwin; 360 struct tcpopt to; /* options in this segment */ 361 struct rmxp_tao tao; /* our TAO cache entry */ 362 int headlocked = 0; 363 struct sockaddr_in *next_hop; 364 int rstreason; /* For badport_bandlim accounting purposes */ 365 366 struct ip6_hdr *ip6 = NULL; 367#ifdef INET6 368 int isipv6; 369#else 370 const int isipv6 = 0; 371#endif 372 373#ifdef TCPDEBUG 374 /* 375 * The size of tcp_saveipgen must be the size of the max ip header, 376 * now IPv6. 377 */ 378 u_char tcp_saveipgen[40]; 379 struct tcphdr tcp_savetcp; 380 short ostate = 0; 381#endif 382 383 next_hop = ip_claim_next_hop(m); 384#ifdef INET6 385 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 386#endif 387 bzero(&tao, sizeof(tao)); 388 bzero((char *)&to, sizeof(to)); 389 390 tcpstat.tcps_rcvtotal++; 391 392 if (isipv6) { 393 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 394 ip6 = mtod(m, struct ip6_hdr *); 395 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 396 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 397 tcpstat.tcps_rcvbadsum++; 398 goto drop; 399 } 400 th = (struct tcphdr *)((caddr_t)ip6 + off0); 401 402 /* 403 * Be proactive about unspecified IPv6 address in source. 404 * As we use all-zero to indicate unbounded/unconnected pcb, 405 * unspecified IPv6 address can be used to confuse us. 406 * 407 * Note that packets with unspecified IPv6 destination is 408 * already dropped in ip6_input. 409 */ 410 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 411 /* XXX stat */ 412 goto drop; 413 } 414 } else { 415 /* 416 * Get IP and TCP header together in first mbuf. 417 * Note: IP leaves IP header in first mbuf. 418 */ 419 if (off0 > sizeof (struct ip)) { 420 ip_stripoptions(m, (struct mbuf *)0); 421 off0 = sizeof(struct ip); 422 } 423 if (m->m_len < sizeof (struct tcpiphdr)) { 424 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 425 tcpstat.tcps_rcvshort++; 426 return; 427 } 428 } 429 ip = mtod(m, struct ip *); 430 ipov = (struct ipovly *)ip; 431 th = (struct tcphdr *)((caddr_t)ip + off0); 432 tlen = ip->ip_len; 433 434 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 435 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 436 th->th_sum = m->m_pkthdr.csum_data; 437 else 438 th->th_sum = in_pseudo(ip->ip_src.s_addr, 439 ip->ip_dst.s_addr, 440 htonl(m->m_pkthdr.csum_data + 441 ip->ip_len + 442 IPPROTO_TCP)); 443 th->th_sum ^= 0xffff; 444#ifdef TCPDEBUG 445 ipov->ih_len = (u_short)tlen; 446 ipov->ih_len = htons(ipov->ih_len); 447#endif 448 } else { 449 /* 450 * Checksum extended TCP header and data. 451 */ 452 len = sizeof (struct ip) + tlen; 453 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 454 ipov->ih_len = (u_short)tlen; 455 ipov->ih_len = htons(ipov->ih_len); 456 th->th_sum = in_cksum(m, len); 457 } 458 if (th->th_sum) { 459 tcpstat.tcps_rcvbadsum++; 460 goto drop; 461 } 462#ifdef INET6 463 /* Re-initialization for later version check */ 464 ip->ip_v = IPVERSION; 465#endif 466 } 467 468 /* 469 * Check that TCP offset makes sense, 470 * pull out TCP options and adjust length. XXX 471 */ 472 off = th->th_off << 2; 473 if (off < sizeof (struct tcphdr) || off > tlen) { 474 tcpstat.tcps_rcvbadoff++; 475 goto drop; 476 } 477 tlen -= off; /* tlen is used instead of ti->ti_len */ 478 if (off > sizeof (struct tcphdr)) { 479 if (isipv6) { 480 IP6_EXTHDR_CHECK(m, off0, off, ); 481 ip6 = mtod(m, struct ip6_hdr *); 482 th = (struct tcphdr *)((caddr_t)ip6 + off0); 483 } else { 484 if (m->m_len < sizeof(struct ip) + off) { 485 if ((m = m_pullup(m, sizeof (struct ip) + off)) 486 == 0) { 487 tcpstat.tcps_rcvshort++; 488 return; 489 } 490 ip = mtod(m, struct ip *); 491 ipov = (struct ipovly *)ip; 492 th = (struct tcphdr *)((caddr_t)ip + off0); 493 } 494 } 495 optlen = off - sizeof (struct tcphdr); 496 optp = (u_char *)(th + 1); 497 } 498 thflags = th->th_flags; 499 500#ifdef TCP_DROP_SYNFIN 501 /* 502 * If the drop_synfin option is enabled, drop all packets with 503 * both the SYN and FIN bits set. This prevents e.g. nmap from 504 * identifying the TCP/IP stack. 505 * 506 * This is a violation of the TCP specification. 507 */ 508 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 509 goto drop; 510#endif 511 512 /* 513 * Convert TCP protocol specific fields to host format. 514 */ 515 th->th_seq = ntohl(th->th_seq); 516 th->th_ack = ntohl(th->th_ack); 517 th->th_win = ntohs(th->th_win); 518 th->th_urp = ntohs(th->th_urp); 519 520 /* 521 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 522 * until after ip6_savecontrol() is called and before other functions 523 * which don't want those proto headers. 524 * Because ip6_savecontrol() is going to parse the mbuf to 525 * search for data to be passed up to user-land, it wants mbuf 526 * parameters to be unchanged. 527 * XXX: the call of ip6_savecontrol() has been obsoleted based on 528 * latest version of the advanced API (20020110). 529 */ 530 drop_hdrlen = off0 + off; 531 532 /* 533 * Locate pcb for segment. 534 */ 535 INP_INFO_WLOCK(&tcbinfo); 536 headlocked = 1; 537findpcb: 538 /* IPFIREWALL_FORWARD section */ 539 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 540 /* 541 * Transparently forwarded. Pretend to be the destination. 542 * already got one like this? 543 */ 544 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 545 ip->ip_dst, th->th_dport, 546 0, m->m_pkthdr.rcvif); 547 if (!inp) { 548 /* It's new. Try find the ambushing socket. */ 549 inp = in_pcblookup_hash(&tcbinfo, 550 ip->ip_src, th->th_sport, 551 next_hop->sin_addr, 552 next_hop->sin_port ? 553 ntohs(next_hop->sin_port) : 554 th->th_dport, 555 1, m->m_pkthdr.rcvif); 556 } 557 } else { 558 if (isipv6) 559 inp = in6_pcblookup_hash(&tcbinfo, 560 &ip6->ip6_src, th->th_sport, 561 &ip6->ip6_dst, th->th_dport, 562 1, m->m_pkthdr.rcvif); 563 else 564 inp = in_pcblookup_hash(&tcbinfo, 565 ip->ip_src, th->th_sport, 566 ip->ip_dst, th->th_dport, 567 1, m->m_pkthdr.rcvif); 568 } 569 570#ifdef IPSEC 571 if (isipv6) { 572 if (inp != NULL && ipsec6_in_reject(m, inp)) { 573 ipsec6stat.in_polvio++; 574 goto drop; 575 } 576 } else { 577 if (inp != NULL && ipsec4_in_reject(m, inp)) { 578 ipsecstat.in_polvio++; 579 goto drop; 580 } 581 } 582#endif 583#ifdef FAST_IPSEC 584 if (isipv6) { 585 if (inp != NULL && ipsec6_in_reject(m, inp)) { 586 goto drop; 587 } 588 } else 589 if (inp != NULL && ipsec4_in_reject(m, inp)) { 590 goto drop; 591 } 592#endif /*FAST_IPSEC*/ 593 594 /* 595 * If the state is CLOSED (i.e., TCB does not exist) then 596 * all data in the incoming segment is discarded. 597 * If the TCB exists but is in CLOSED state, it is embryonic, 598 * but should either do a listen or a connect soon. 599 */ 600 if (inp == NULL) { 601 if (log_in_vain) { 602#ifdef INET6 603 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 604#else 605 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 606#endif 607 608 if (isipv6) { 609 strcpy(dbuf, "["); 610 strcpy(sbuf, "["); 611 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 612 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 613 strcat(dbuf, "]"); 614 strcat(sbuf, "]"); 615 } else { 616 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 617 strcpy(sbuf, inet_ntoa(ip->ip_src)); 618 } 619 switch (log_in_vain) { 620 case 1: 621 if ((thflags & TH_SYN) == 0) 622 break; 623 /* FALLTHROUGH */ 624 case 2: 625 log(LOG_INFO, 626 "Connection attempt to TCP %s:%d " 627 "from %s:%d flags:0x%02x\n", 628 dbuf, ntohs(th->th_dport), sbuf, 629 ntohs(th->th_sport), thflags); 630 break; 631 default: 632 break; 633 } 634 } 635 if (blackhole) { 636 switch (blackhole) { 637 case 1: 638 if (thflags & TH_SYN) 639 goto drop; 640 break; 641 case 2: 642 goto drop; 643 default: 644 goto drop; 645 } 646 } 647 rstreason = BANDLIM_RST_CLOSEDPORT; 648 goto dropwithreset; 649 } 650 INP_LOCK(inp); 651 if (inp->inp_vflag & INP_TIMEWAIT) { 652 /* 653 * The only option of relevance is TOF_CC, and only if 654 * present in a SYN segment. See tcp_timewait(). 655 */ 656 if (thflags & TH_SYN) 657 tcp_dooptions(&to, optp, optlen, 1); 658 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 659 &to, th, m, tlen)) 660 goto findpcb; 661 /* 662 * tcp_timewait unlocks inp. 663 */ 664 INP_INFO_WUNLOCK(&tcbinfo); 665 return; 666 } 667 tp = intotcpcb(inp); 668 if (tp == 0) { 669 INP_UNLOCK(inp); 670 rstreason = BANDLIM_RST_CLOSEDPORT; 671 goto dropwithreset; 672 } 673 if (tp->t_state == TCPS_CLOSED) 674 goto drop; 675 676 /* Unscale the window into a 32-bit value. */ 677 if ((thflags & TH_SYN) == 0) 678 tiwin = th->th_win << tp->snd_scale; 679 else 680 tiwin = th->th_win; 681 682#ifdef MAC 683 if (mac_check_inpcb_deliver(inp, m)) 684 goto drop; 685#endif 686 so = inp->inp_socket; 687#ifdef TCPDEBUG 688 if (so->so_options & SO_DEBUG) { 689 ostate = tp->t_state; 690 if (isipv6) 691 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 692 else 693 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 694 tcp_savetcp = *th; 695 } 696#endif 697 if (so->so_options & SO_ACCEPTCONN) { 698 struct in_conninfo inc; 699 700#ifdef INET6 701 inc.inc_isipv6 = isipv6; 702#endif 703 if (isipv6) { 704 inc.inc6_faddr = ip6->ip6_src; 705 inc.inc6_laddr = ip6->ip6_dst; 706 } else { 707 inc.inc_faddr = ip->ip_src; 708 inc.inc_laddr = ip->ip_dst; 709 } 710 inc.inc_fport = th->th_sport; 711 inc.inc_lport = th->th_dport; 712 713 /* 714 * If the state is LISTEN then ignore segment if it contains 715 * a RST. If the segment contains an ACK then it is bad and 716 * send a RST. If it does not contain a SYN then it is not 717 * interesting; drop it. 718 * 719 * If the state is SYN_RECEIVED (syncache) and seg contains 720 * an ACK, but not for our SYN/ACK, send a RST. If the seg 721 * contains a RST, check the sequence number to see if it 722 * is a valid reset segment. 723 */ 724 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 725 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 726 if (!syncache_expand(&inc, th, &so, m)) { 727 /* 728 * No syncache entry, or ACK was not 729 * for our SYN/ACK. Send a RST. 730 */ 731 tcpstat.tcps_badsyn++; 732 rstreason = BANDLIM_RST_OPENPORT; 733 goto dropwithreset; 734 } 735 if (so == NULL) { 736 /* 737 * Could not complete 3-way handshake, 738 * connection is being closed down, and 739 * syncache will free mbuf. 740 */ 741 INP_UNLOCK(inp); 742 INP_INFO_WUNLOCK(&tcbinfo); 743 return; 744 } 745 /* 746 * Socket is created in state SYN_RECEIVED. 747 * Continue processing segment. 748 */ 749 INP_UNLOCK(inp); 750 inp = sotoinpcb(so); 751 INP_LOCK(inp); 752 tp = intotcpcb(inp); 753 /* 754 * This is what would have happened in 755 * tcp_output() when the SYN,ACK was sent. 756 */ 757 tp->snd_up = tp->snd_una; 758 tp->snd_max = tp->snd_nxt = tp->iss + 1; 759 tp->last_ack_sent = tp->rcv_nxt; 760 /* 761 * RFC1323: The window in SYN & SYN/ACK 762 * segments is never scaled. 763 */ 764 tp->snd_wnd = tiwin; /* unscaled */ 765 goto after_listen; 766 } 767 if (thflags & TH_RST) { 768 syncache_chkrst(&inc, th); 769 goto drop; 770 } 771 if (thflags & TH_ACK) { 772 syncache_badack(&inc); 773 tcpstat.tcps_badsyn++; 774 rstreason = BANDLIM_RST_OPENPORT; 775 goto dropwithreset; 776 } 777 goto drop; 778 } 779 780 /* 781 * Segment's flags are (SYN) or (SYN|FIN). 782 */ 783#ifdef INET6 784 /* 785 * If deprecated address is forbidden, 786 * we do not accept SYN to deprecated interface 787 * address to prevent any new inbound connection from 788 * getting established. 789 * When we do not accept SYN, we send a TCP RST, 790 * with deprecated source address (instead of dropping 791 * it). We compromise it as it is much better for peer 792 * to send a RST, and RST will be the final packet 793 * for the exchange. 794 * 795 * If we do not forbid deprecated addresses, we accept 796 * the SYN packet. RFC2462 does not suggest dropping 797 * SYN in this case. 798 * If we decipher RFC2462 5.5.4, it says like this: 799 * 1. use of deprecated addr with existing 800 * communication is okay - "SHOULD continue to be 801 * used" 802 * 2. use of it with new communication: 803 * (2a) "SHOULD NOT be used if alternate address 804 * with sufficient scope is available" 805 * (2b) nothing mentioned otherwise. 806 * Here we fall into (2b) case as we have no choice in 807 * our source address selection - we must obey the peer. 808 * 809 * The wording in RFC2462 is confusing, and there are 810 * multiple description text for deprecated address 811 * handling - worse, they are not exactly the same. 812 * I believe 5.5.4 is the best one, so we follow 5.5.4. 813 */ 814 if (isipv6 && !ip6_use_deprecated) { 815 struct in6_ifaddr *ia6; 816 817 if ((ia6 = ip6_getdstifaddr(m)) && 818 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 819 INP_UNLOCK(inp); 820 tp = NULL; 821 rstreason = BANDLIM_RST_OPENPORT; 822 goto dropwithreset; 823 } 824 } 825#endif 826 /* 827 * If it is from this socket, drop it, it must be forged. 828 * Don't bother responding if the destination was a broadcast. 829 */ 830 if (th->th_dport == th->th_sport) { 831 if (isipv6) { 832 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 833 &ip6->ip6_src)) 834 goto drop; 835 } else { 836 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 837 goto drop; 838 } 839 } 840 /* 841 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 842 * 843 * Note that it is quite possible to receive unicast 844 * link-layer packets with a broadcast IP address. Use 845 * in_broadcast() to find them. 846 */ 847 if (m->m_flags & (M_BCAST|M_MCAST)) 848 goto drop; 849 if (isipv6) { 850 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 851 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 852 goto drop; 853 } else { 854 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 855 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 856 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 857 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 858 goto drop; 859 } 860 /* 861 * SYN appears to be valid; create compressed TCP state 862 * for syncache, or perform t/tcp connection. 863 */ 864 if (so->so_qlen <= so->so_qlimit) { 865#ifdef TCPDEBUG 866 if (so->so_options & SO_DEBUG) 867 tcp_trace(TA_INPUT, ostate, tp, 868 (void *)tcp_saveipgen, &tcp_savetcp, 0); 869#endif 870 tcp_dooptions(&to, optp, optlen, 1); 871 if (!syncache_add(&inc, &to, th, &so, m)) 872 goto drop; 873 if (so == NULL) { 874 /* 875 * Entry added to syncache, mbuf used to 876 * send SYN,ACK packet. 877 */ 878 KASSERT(headlocked, ("headlocked")); 879 INP_UNLOCK(inp); 880 INP_INFO_WUNLOCK(&tcbinfo); 881 return; 882 } 883 /* 884 * Segment passed TAO tests. 885 */ 886 INP_UNLOCK(inp); 887 inp = sotoinpcb(so); 888 INP_LOCK(inp); 889 tp = intotcpcb(inp); 890 tp->snd_wnd = tiwin; 891 tp->t_starttime = ticks; 892 tp->t_state = TCPS_ESTABLISHED; 893 894 /* 895 * T/TCP logic: 896 * If there is a FIN or if there is data, then 897 * delay SYN,ACK(SYN) in the hope of piggy-backing 898 * it on a response segment. Otherwise must send 899 * ACK now in case the other side is slow starting. 900 */ 901 if (thflags & TH_FIN || tlen != 0) 902 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 903 else 904 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 905 tcpstat.tcps_connects++; 906 soisconnected(so); 907 goto trimthenstep6; 908 } 909 goto drop; 910 } 911after_listen: 912 913 /* XXX temp debugging */ 914 /* should not happen - syncache should pick up these connections */ 915 if (tp->t_state == TCPS_LISTEN) 916 panic("tcp_input: TCPS_LISTEN"); 917 918 /* 919 * This is the second part of the MSS DoS prevention code (after 920 * minmss on the sending side) and it deals with too many too small 921 * tcp packets in a too short timeframe (1 second). 922 * 923 * For every full second we count the number of received packets 924 * and bytes. If we get a lot of packets per second for this connection 925 * (tcp_minmssoverload) we take a closer look at it and compute the 926 * average packet size for the past second. If that is less than 927 * tcp_minmss we get too many packets with very small payload which 928 * is not good and burdens our system (and every packet generates 929 * a wakeup to the process connected to our socket). We can reasonable 930 * expect this to be small packet DoS attack to exhaust our CPU 931 * cycles. 932 * 933 * Care has to be taken for the minimum packet overload value. This 934 * value defines the minimum number of packets per second before we 935 * start to worry. This must not be too low to avoid killing for 936 * example interactive connections with many small packets like 937 * telnet or SSH. 938 * 939 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 940 * this check. 941 * 942 * Account for packet if payload packet, skip over ACK, etc. 943 */ 944 if (tcp_minmss && tcp_minmssoverload && 945 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 946 if (tp->rcv_second > ticks) { 947 tp->rcv_pps++; 948 tp->rcv_byps += tlen + off; 949 if (tp->rcv_pps > tcp_minmssoverload) { 950 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 951 printf("too many small tcp packets from " 952 "%s:%u, av. %lubyte/packet, " 953 "dropping connection\n", 954#ifdef INET6 955 isipv6 ? 956 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 957#endif 958 inet_ntoa(inp->inp_inc.inc_faddr), 959 inp->inp_inc.inc_fport, 960 tp->rcv_byps / tp->rcv_pps); 961 tp = tcp_drop(tp, ECONNRESET); 962 tcpstat.tcps_minmssdrops++; 963 goto drop; 964 } 965 } 966 } else { 967 tp->rcv_second = ticks + hz; 968 tp->rcv_pps = 1; 969 tp->rcv_byps = tlen + off; 970 } 971 } 972 973 /* 974 * Segment received on connection. 975 * Reset idle time and keep-alive timer. 976 */ 977 tp->t_rcvtime = ticks; 978 if (TCPS_HAVEESTABLISHED(tp->t_state)) 979 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 980 981 /* 982 * Process options only when we get SYN/ACK back. The SYN case 983 * for incoming connections is handled in tcp_syncache. 984 * XXX this is traditional behavior, may need to be cleaned up. 985 */ 986 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 987 if (thflags & TH_SYN) { 988 if (to.to_flags & TOF_SCALE) { 989 tp->t_flags |= TF_RCVD_SCALE; 990 tp->requested_s_scale = to.to_requested_s_scale; 991 } 992 if (to.to_flags & TOF_TS) { 993 tp->t_flags |= TF_RCVD_TSTMP; 994 tp->ts_recent = to.to_tsval; 995 tp->ts_recent_age = ticks; 996 } 997 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 998 tp->t_flags |= TF_RCVD_CC; 999 if (to.to_flags & TOF_MSS) 1000 tcp_mss(tp, to.to_mss); 1001 } 1002 1003 /* 1004 * Header prediction: check for the two common cases 1005 * of a uni-directional data xfer. If the packet has 1006 * no control flags, is in-sequence, the window didn't 1007 * change and we're not retransmitting, it's a 1008 * candidate. If the length is zero and the ack moved 1009 * forward, we're the sender side of the xfer. Just 1010 * free the data acked & wake any higher level process 1011 * that was blocked waiting for space. If the length 1012 * is non-zero and the ack didn't move, we're the 1013 * receiver side. If we're getting packets in-order 1014 * (the reassembly queue is empty), add the data to 1015 * the socket buffer and note that we need a delayed ack. 1016 * Make sure that the hidden state-flags are also off. 1017 * Since we check for TCPS_ESTABLISHED above, it can only 1018 * be TH_NEEDSYN. 1019 */ 1020 if (tp->t_state == TCPS_ESTABLISHED && 1021 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1022 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1023 ((to.to_flags & TOF_TS) == 0 || 1024 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1025 /* 1026 * Using the CC option is compulsory if once started: 1027 * the segment is OK if no T/TCP was negotiated or 1028 * if the segment has a CC option equal to CCrecv 1029 */ 1030 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1031 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1032 th->th_seq == tp->rcv_nxt && 1033 tiwin && tiwin == tp->snd_wnd && 1034 tp->snd_nxt == tp->snd_max) { 1035 1036 /* 1037 * If last ACK falls within this segment's sequence numbers, 1038 * record the timestamp. 1039 * NOTE that the test is modified according to the latest 1040 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1041 */ 1042 if ((to.to_flags & TOF_TS) != 0 && 1043 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1044 tp->ts_recent_age = ticks; 1045 tp->ts_recent = to.to_tsval; 1046 } 1047 1048 if (tlen == 0) { 1049 if (SEQ_GT(th->th_ack, tp->snd_una) && 1050 SEQ_LEQ(th->th_ack, tp->snd_max) && 1051 tp->snd_cwnd >= tp->snd_wnd && 1052 ((!tcp_do_newreno && 1053 tp->t_dupacks < tcprexmtthresh) || 1054 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1055 KASSERT(headlocked, ("headlocked")); 1056 INP_INFO_WUNLOCK(&tcbinfo); 1057 /* 1058 * this is a pure ack for outstanding data. 1059 */ 1060 ++tcpstat.tcps_predack; 1061 /* 1062 * "bad retransmit" recovery 1063 */ 1064 if (tp->t_rxtshift == 1 && 1065 ticks < tp->t_badrxtwin) { 1066 ++tcpstat.tcps_sndrexmitbad; 1067 tp->snd_cwnd = tp->snd_cwnd_prev; 1068 tp->snd_ssthresh = 1069 tp->snd_ssthresh_prev; 1070 tp->snd_recover = tp->snd_recover_prev; 1071 if (tp->t_flags & TF_WASFRECOVERY) 1072 ENTER_FASTRECOVERY(tp); 1073 tp->snd_nxt = tp->snd_max; 1074 tp->t_badrxtwin = 0; 1075 } 1076 1077 /* 1078 * Recalculate the transmit timer / rtt. 1079 * 1080 * Some boxes send broken timestamp replies 1081 * during the SYN+ACK phase, ignore 1082 * timestamps of 0 or we could calculate a 1083 * huge RTT and blow up the retransmit timer. 1084 */ 1085 if ((to.to_flags & TOF_TS) != 0 && 1086 to.to_tsecr) { 1087 tcp_xmit_timer(tp, 1088 ticks - to.to_tsecr + 1); 1089 } else if (tp->t_rtttime && 1090 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1091 tcp_xmit_timer(tp, 1092 ticks - tp->t_rtttime); 1093 } 1094 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1095 acked = th->th_ack - tp->snd_una; 1096 tcpstat.tcps_rcvackpack++; 1097 tcpstat.tcps_rcvackbyte += acked; 1098 sbdrop(&so->so_snd, acked); 1099 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1100 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1101 tp->snd_recover = th->th_ack - 1; 1102 tp->snd_una = th->th_ack; 1103 /* 1104 * pull snd_wl2 up to prevent seq wrap relative 1105 * to th_ack. 1106 */ 1107 tp->snd_wl2 = th->th_ack; 1108 tp->t_dupacks = 0; 1109 m_freem(m); 1110 ND6_HINT(tp); /* some progress has been done */ 1111 1112 /* 1113 * If all outstanding data are acked, stop 1114 * retransmit timer, otherwise restart timer 1115 * using current (possibly backed-off) value. 1116 * If process is waiting for space, 1117 * wakeup/selwakeup/signal. If data 1118 * are ready to send, let tcp_output 1119 * decide between more output or persist. 1120 1121#ifdef TCPDEBUG 1122 if (so->so_options & SO_DEBUG) 1123 tcp_trace(TA_INPUT, ostate, tp, 1124 (void *)tcp_saveipgen, 1125 &tcp_savetcp, 0); 1126#endif 1127 */ 1128 if (tp->snd_una == tp->snd_max) 1129 callout_stop(tp->tt_rexmt); 1130 else if (!callout_active(tp->tt_persist)) 1131 callout_reset(tp->tt_rexmt, 1132 tp->t_rxtcur, 1133 tcp_timer_rexmt, tp); 1134 1135 sowwakeup(so); 1136 if (so->so_snd.sb_cc) 1137 (void) tcp_output(tp); 1138 goto check_delack; 1139 } 1140 } else if (th->th_ack == tp->snd_una && 1141 LIST_EMPTY(&tp->t_segq) && 1142 tlen <= sbspace(&so->so_rcv)) { 1143 KASSERT(headlocked, ("headlocked")); 1144 INP_INFO_WUNLOCK(&tcbinfo); 1145 /* 1146 * this is a pure, in-sequence data packet 1147 * with nothing on the reassembly queue and 1148 * we have enough buffer space to take it. 1149 */ 1150 ++tcpstat.tcps_preddat; 1151 tp->rcv_nxt += tlen; 1152 /* 1153 * Pull snd_wl1 up to prevent seq wrap relative to 1154 * th_seq. 1155 */ 1156 tp->snd_wl1 = th->th_seq; 1157 /* 1158 * Pull rcv_up up to prevent seq wrap relative to 1159 * rcv_nxt. 1160 */ 1161 tp->rcv_up = tp->rcv_nxt; 1162 tcpstat.tcps_rcvpack++; 1163 tcpstat.tcps_rcvbyte += tlen; 1164 ND6_HINT(tp); /* some progress has been done */ 1165 /* 1166#ifdef TCPDEBUG 1167 if (so->so_options & SO_DEBUG) 1168 tcp_trace(TA_INPUT, ostate, tp, 1169 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1170#endif 1171 * Add data to socket buffer. 1172 */ 1173 if (so->so_state & SS_CANTRCVMORE) { 1174 m_freem(m); 1175 } else { 1176 m_adj(m, drop_hdrlen); /* delayed header drop */ 1177 sbappendstream(&so->so_rcv, m); 1178 } 1179 sorwakeup(so); 1180 if (DELAY_ACK(tp)) { 1181 tp->t_flags |= TF_DELACK; 1182 } else { 1183 tp->t_flags |= TF_ACKNOW; 1184 tcp_output(tp); 1185 } 1186 goto check_delack; 1187 } 1188 } 1189 1190 /* 1191 * Calculate amount of space in receive window, 1192 * and then do TCP input processing. 1193 * Receive window is amount of space in rcv queue, 1194 * but not less than advertised window. 1195 */ 1196 { int win; 1197 1198 win = sbspace(&so->so_rcv); 1199 if (win < 0) 1200 win = 0; 1201 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1202 } 1203 1204 switch (tp->t_state) { 1205 1206 /* 1207 * If the state is SYN_RECEIVED: 1208 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1209 */ 1210 case TCPS_SYN_RECEIVED: 1211 if ((thflags & TH_ACK) && 1212 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1213 SEQ_GT(th->th_ack, tp->snd_max))) { 1214 rstreason = BANDLIM_RST_OPENPORT; 1215 goto dropwithreset; 1216 } 1217 break; 1218 1219 /* 1220 * If the state is SYN_SENT: 1221 * if seg contains an ACK, but not for our SYN, drop the input. 1222 * if seg contains a RST, then drop the connection. 1223 * if seg does not contain SYN, then drop it. 1224 * Otherwise this is an acceptable SYN segment 1225 * initialize tp->rcv_nxt and tp->irs 1226 * if seg contains ack then advance tp->snd_una 1227 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1228 * arrange for segment to be acked (eventually) 1229 * continue processing rest of data/controls, beginning with URG 1230 */ 1231 case TCPS_SYN_SENT: 1232 if (tcp_do_rfc1644) 1233 tcp_hc_gettao(&inp->inp_inc, &tao); 1234 1235 if ((thflags & TH_ACK) && 1236 (SEQ_LEQ(th->th_ack, tp->iss) || 1237 SEQ_GT(th->th_ack, tp->snd_max))) { 1238 /* 1239 * If we have a cached CCsent for the remote host, 1240 * hence we haven't just crashed and restarted, 1241 * do not send a RST. This may be a retransmission 1242 * from the other side after our earlier ACK was lost. 1243 * Our new SYN, when it arrives, will serve as the 1244 * needed ACK. 1245 */ 1246 if (tao.tao_ccsent != 0) 1247 goto drop; 1248 else { 1249 rstreason = BANDLIM_UNLIMITED; 1250 goto dropwithreset; 1251 } 1252 } 1253 if (thflags & TH_RST) { 1254 if (thflags & TH_ACK) 1255 tp = tcp_drop(tp, ECONNREFUSED); 1256 goto drop; 1257 } 1258 if ((thflags & TH_SYN) == 0) 1259 goto drop; 1260 tp->snd_wnd = th->th_win; /* initial send window */ 1261 tp->cc_recv = to.to_cc; /* foreign CC */ 1262 1263 tp->irs = th->th_seq; 1264 tcp_rcvseqinit(tp); 1265 if (thflags & TH_ACK) { 1266 /* 1267 * Our SYN was acked. If segment contains CC.ECHO 1268 * option, check it to make sure this segment really 1269 * matches our SYN. If not, just drop it as old 1270 * duplicate, but send an RST if we're still playing 1271 * by the old rules. If no CC.ECHO option, make sure 1272 * we don't get fooled into using T/TCP. 1273 */ 1274 if (to.to_flags & TOF_CCECHO) { 1275 if (tp->cc_send != to.to_ccecho) { 1276 if (tao.tao_ccsent != 0) 1277 goto drop; 1278 else { 1279 rstreason = BANDLIM_UNLIMITED; 1280 goto dropwithreset; 1281 } 1282 } 1283 } else 1284 tp->t_flags &= ~TF_RCVD_CC; 1285 tcpstat.tcps_connects++; 1286 soisconnected(so); 1287#ifdef MAC 1288 mac_set_socket_peer_from_mbuf(m, so); 1289#endif 1290 /* Do window scaling on this connection? */ 1291 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1292 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1293 tp->snd_scale = tp->requested_s_scale; 1294 tp->rcv_scale = tp->request_r_scale; 1295 } 1296 /* Segment is acceptable, update cache if undefined. */ 1297 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1298 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1299 1300 tp->rcv_adv += tp->rcv_wnd; 1301 tp->snd_una++; /* SYN is acked */ 1302 /* 1303 * If there's data, delay ACK; if there's also a FIN 1304 * ACKNOW will be turned on later. 1305 */ 1306 if (DELAY_ACK(tp) && tlen != 0) 1307 callout_reset(tp->tt_delack, tcp_delacktime, 1308 tcp_timer_delack, tp); 1309 else 1310 tp->t_flags |= TF_ACKNOW; 1311 /* 1312 * Received <SYN,ACK> in SYN_SENT[*] state. 1313 * Transitions: 1314 * SYN_SENT --> ESTABLISHED 1315 * SYN_SENT* --> FIN_WAIT_1 1316 */ 1317 tp->t_starttime = ticks; 1318 if (tp->t_flags & TF_NEEDFIN) { 1319 tp->t_state = TCPS_FIN_WAIT_1; 1320 tp->t_flags &= ~TF_NEEDFIN; 1321 thflags &= ~TH_SYN; 1322 } else { 1323 tp->t_state = TCPS_ESTABLISHED; 1324 callout_reset(tp->tt_keep, tcp_keepidle, 1325 tcp_timer_keep, tp); 1326 } 1327 } else { 1328 /* 1329 * Received initial SYN in SYN-SENT[*] state => 1330 * simultaneous open. If segment contains CC option 1331 * and there is a cached CC, apply TAO test. 1332 * If it succeeds, connection is * half-synchronized. 1333 * Otherwise, do 3-way handshake: 1334 * SYN-SENT -> SYN-RECEIVED 1335 * SYN-SENT* -> SYN-RECEIVED* 1336 * If there was no CC option, clear cached CC value. 1337 */ 1338 tp->t_flags |= TF_ACKNOW; 1339 callout_stop(tp->tt_rexmt); 1340 if (to.to_flags & TOF_CC) { 1341 if (tao.tao_cc != 0 && 1342 CC_GT(to.to_cc, tao.tao_cc)) { 1343 /* 1344 * update cache and make transition: 1345 * SYN-SENT -> ESTABLISHED* 1346 * SYN-SENT* -> FIN-WAIT-1* 1347 */ 1348 tao.tao_cc = to.to_cc; 1349 tcp_hc_updatetao(&inp->inp_inc, 1350 TCP_HC_TAO_CC, to.to_cc, 0); 1351 tp->t_starttime = ticks; 1352 if (tp->t_flags & TF_NEEDFIN) { 1353 tp->t_state = TCPS_FIN_WAIT_1; 1354 tp->t_flags &= ~TF_NEEDFIN; 1355 } else { 1356 tp->t_state = TCPS_ESTABLISHED; 1357 callout_reset(tp->tt_keep, 1358 tcp_keepidle, 1359 tcp_timer_keep, 1360 tp); 1361 } 1362 tp->t_flags |= TF_NEEDSYN; 1363 } else 1364 tp->t_state = TCPS_SYN_RECEIVED; 1365 } else { 1366 if (tcp_do_rfc1644) { 1367 /* CC.NEW or no option => invalidate cache */ 1368 tao.tao_cc = 0; 1369 tcp_hc_updatetao(&inp->inp_inc, 1370 TCP_HC_TAO_CC, to.to_cc, 0); 1371 } 1372 tp->t_state = TCPS_SYN_RECEIVED; 1373 } 1374 } 1375 1376trimthenstep6: 1377 /* 1378 * Advance th->th_seq to correspond to first data byte. 1379 * If data, trim to stay within window, 1380 * dropping FIN if necessary. 1381 */ 1382 th->th_seq++; 1383 if (tlen > tp->rcv_wnd) { 1384 todrop = tlen - tp->rcv_wnd; 1385 m_adj(m, -todrop); 1386 tlen = tp->rcv_wnd; 1387 thflags &= ~TH_FIN; 1388 tcpstat.tcps_rcvpackafterwin++; 1389 tcpstat.tcps_rcvbyteafterwin += todrop; 1390 } 1391 tp->snd_wl1 = th->th_seq - 1; 1392 tp->rcv_up = th->th_seq; 1393 /* 1394 * Client side of transaction: already sent SYN and data. 1395 * If the remote host used T/TCP to validate the SYN, 1396 * our data will be ACK'd; if so, enter normal data segment 1397 * processing in the middle of step 5, ack processing. 1398 * Otherwise, goto step 6. 1399 */ 1400 if (thflags & TH_ACK) 1401 goto process_ACK; 1402 1403 goto step6; 1404 1405 /* 1406 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1407 * if segment contains a SYN and CC [not CC.NEW] option: 1408 * if state == TIME_WAIT and connection duration > MSL, 1409 * drop packet and send RST; 1410 * 1411 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1412 * ack the FIN (and data) in retransmission queue. 1413 * Complete close and delete TCPCB. Then reprocess 1414 * segment, hoping to find new TCPCB in LISTEN state; 1415 * 1416 * else must be old SYN; drop it. 1417 * else do normal processing. 1418 */ 1419 case TCPS_LAST_ACK: 1420 case TCPS_CLOSING: 1421 case TCPS_TIME_WAIT: 1422 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1423 if ((thflags & TH_SYN) && 1424 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1425 if (tp->t_state == TCPS_TIME_WAIT && 1426 (ticks - tp->t_starttime) > tcp_msl) { 1427 rstreason = BANDLIM_UNLIMITED; 1428 goto dropwithreset; 1429 } 1430 if (CC_GT(to.to_cc, tp->cc_recv)) { 1431 tp = tcp_close(tp); 1432 goto findpcb; 1433 } 1434 else 1435 goto drop; 1436 } 1437 break; /* continue normal processing */ 1438 } 1439 1440 /* 1441 * States other than LISTEN or SYN_SENT. 1442 * First check the RST flag and sequence number since reset segments 1443 * are exempt from the timestamp and connection count tests. This 1444 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1445 * below which allowed reset segments in half the sequence space 1446 * to fall though and be processed (which gives forged reset 1447 * segments with a random sequence number a 50 percent chance of 1448 * killing a connection). 1449 * Then check timestamp, if present. 1450 * Then check the connection count, if present. 1451 * Then check that at least some bytes of segment are within 1452 * receive window. If segment begins before rcv_nxt, 1453 * drop leading data (and SYN); if nothing left, just ack. 1454 * 1455 * 1456 * If the RST bit is set, check the sequence number to see 1457 * if this is a valid reset segment. 1458 * RFC 793 page 37: 1459 * In all states except SYN-SENT, all reset (RST) segments 1460 * are validated by checking their SEQ-fields. A reset is 1461 * valid if its sequence number is in the window. 1462 * Note: this does not take into account delayed ACKs, so 1463 * we should test against last_ack_sent instead of rcv_nxt. 1464 * The sequence number in the reset segment is normally an 1465 * echo of our outgoing acknowlegement numbers, but some hosts 1466 * send a reset with the sequence number at the rightmost edge 1467 * of our receive window, and we have to handle this case. 1468 * If we have multiple segments in flight, the intial reset 1469 * segment sequence numbers will be to the left of last_ack_sent, 1470 * but they will eventually catch up. 1471 * In any case, it never made sense to trim reset segments to 1472 * fit the receive window since RFC 1122 says: 1473 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1474 * 1475 * A TCP SHOULD allow a received RST segment to include data. 1476 * 1477 * DISCUSSION 1478 * It has been suggested that a RST segment could contain 1479 * ASCII text that encoded and explained the cause of the 1480 * RST. No standard has yet been established for such 1481 * data. 1482 * 1483 * If the reset segment passes the sequence number test examine 1484 * the state: 1485 * SYN_RECEIVED STATE: 1486 * If passive open, return to LISTEN state. 1487 * If active open, inform user that connection was refused. 1488 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1489 * Inform user that connection was reset, and close tcb. 1490 * CLOSING, LAST_ACK STATES: 1491 * Close the tcb. 1492 * TIME_WAIT STATE: 1493 * Drop the segment - see Stevens, vol. 2, p. 964 and 1494 * RFC 1337. 1495 */ 1496 if (thflags & TH_RST) { 1497 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1498 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1499 switch (tp->t_state) { 1500 1501 case TCPS_SYN_RECEIVED: 1502 so->so_error = ECONNREFUSED; 1503 goto close; 1504 1505 case TCPS_ESTABLISHED: 1506 case TCPS_FIN_WAIT_1: 1507 case TCPS_FIN_WAIT_2: 1508 case TCPS_CLOSE_WAIT: 1509 so->so_error = ECONNRESET; 1510 close: 1511 tp->t_state = TCPS_CLOSED; 1512 tcpstat.tcps_drops++; 1513 tp = tcp_close(tp); 1514 break; 1515 1516 case TCPS_CLOSING: 1517 case TCPS_LAST_ACK: 1518 tp = tcp_close(tp); 1519 break; 1520 1521 case TCPS_TIME_WAIT: 1522 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1523 ("timewait")); 1524 break; 1525 } 1526 } 1527 goto drop; 1528 } 1529 1530 /* 1531 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1532 * and it's less than ts_recent, drop it. 1533 */ 1534 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1535 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1536 1537 /* Check to see if ts_recent is over 24 days old. */ 1538 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1539 /* 1540 * Invalidate ts_recent. If this segment updates 1541 * ts_recent, the age will be reset later and ts_recent 1542 * will get a valid value. If it does not, setting 1543 * ts_recent to zero will at least satisfy the 1544 * requirement that zero be placed in the timestamp 1545 * echo reply when ts_recent isn't valid. The 1546 * age isn't reset until we get a valid ts_recent 1547 * because we don't want out-of-order segments to be 1548 * dropped when ts_recent is old. 1549 */ 1550 tp->ts_recent = 0; 1551 } else { 1552 tcpstat.tcps_rcvduppack++; 1553 tcpstat.tcps_rcvdupbyte += tlen; 1554 tcpstat.tcps_pawsdrop++; 1555 if (tlen) 1556 goto dropafterack; 1557 goto drop; 1558 } 1559 } 1560 1561 /* 1562 * T/TCP mechanism 1563 * If T/TCP was negotiated and the segment doesn't have CC, 1564 * or if its CC is wrong then drop the segment. 1565 * RST segments do not have to comply with this. 1566 */ 1567 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1568 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1569 goto dropafterack; 1570 1571 /* 1572 * In the SYN-RECEIVED state, validate that the packet belongs to 1573 * this connection before trimming the data to fit the receive 1574 * window. Check the sequence number versus IRS since we know 1575 * the sequence numbers haven't wrapped. This is a partial fix 1576 * for the "LAND" DoS attack. 1577 */ 1578 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1579 rstreason = BANDLIM_RST_OPENPORT; 1580 goto dropwithreset; 1581 } 1582 1583 todrop = tp->rcv_nxt - th->th_seq; 1584 if (todrop > 0) { 1585 if (thflags & TH_SYN) { 1586 thflags &= ~TH_SYN; 1587 th->th_seq++; 1588 if (th->th_urp > 1) 1589 th->th_urp--; 1590 else 1591 thflags &= ~TH_URG; 1592 todrop--; 1593 } 1594 /* 1595 * Following if statement from Stevens, vol. 2, p. 960. 1596 */ 1597 if (todrop > tlen 1598 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1599 /* 1600 * Any valid FIN must be to the left of the window. 1601 * At this point the FIN must be a duplicate or out 1602 * of sequence; drop it. 1603 */ 1604 thflags &= ~TH_FIN; 1605 1606 /* 1607 * Send an ACK to resynchronize and drop any data. 1608 * But keep on processing for RST or ACK. 1609 */ 1610 tp->t_flags |= TF_ACKNOW; 1611 todrop = tlen; 1612 tcpstat.tcps_rcvduppack++; 1613 tcpstat.tcps_rcvdupbyte += todrop; 1614 } else { 1615 tcpstat.tcps_rcvpartduppack++; 1616 tcpstat.tcps_rcvpartdupbyte += todrop; 1617 } 1618 drop_hdrlen += todrop; /* drop from the top afterwards */ 1619 th->th_seq += todrop; 1620 tlen -= todrop; 1621 if (th->th_urp > todrop) 1622 th->th_urp -= todrop; 1623 else { 1624 thflags &= ~TH_URG; 1625 th->th_urp = 0; 1626 } 1627 } 1628 1629 /* 1630 * If new data are received on a connection after the 1631 * user processes are gone, then RST the other end. 1632 */ 1633 if ((so->so_state & SS_NOFDREF) && 1634 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1635 tp = tcp_close(tp); 1636 tcpstat.tcps_rcvafterclose++; 1637 rstreason = BANDLIM_UNLIMITED; 1638 goto dropwithreset; 1639 } 1640 1641 /* 1642 * If segment ends after window, drop trailing data 1643 * (and PUSH and FIN); if nothing left, just ACK. 1644 */ 1645 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1646 if (todrop > 0) { 1647 tcpstat.tcps_rcvpackafterwin++; 1648 if (todrop >= tlen) { 1649 tcpstat.tcps_rcvbyteafterwin += tlen; 1650 /* 1651 * If a new connection request is received 1652 * while in TIME_WAIT, drop the old connection 1653 * and start over if the sequence numbers 1654 * are above the previous ones. 1655 */ 1656 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1657 if (thflags & TH_SYN && 1658 tp->t_state == TCPS_TIME_WAIT && 1659 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1660 tp = tcp_close(tp); 1661 goto findpcb; 1662 } 1663 /* 1664 * If window is closed can only take segments at 1665 * window edge, and have to drop data and PUSH from 1666 * incoming segments. Continue processing, but 1667 * remember to ack. Otherwise, drop segment 1668 * and ack. 1669 */ 1670 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1671 tp->t_flags |= TF_ACKNOW; 1672 tcpstat.tcps_rcvwinprobe++; 1673 } else 1674 goto dropafterack; 1675 } else 1676 tcpstat.tcps_rcvbyteafterwin += todrop; 1677 m_adj(m, -todrop); 1678 tlen -= todrop; 1679 thflags &= ~(TH_PUSH|TH_FIN); 1680 } 1681 1682 /* 1683 * If last ACK falls within this segment's sequence numbers, 1684 * record its timestamp. 1685 * NOTE that the test is modified according to the latest 1686 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1687 */ 1688 if ((to.to_flags & TOF_TS) != 0 && 1689 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1690 tp->ts_recent_age = ticks; 1691 tp->ts_recent = to.to_tsval; 1692 } 1693 1694 /* 1695 * If a SYN is in the window, then this is an 1696 * error and we send an RST and drop the connection. 1697 */ 1698 if (thflags & TH_SYN) { 1699 tp = tcp_drop(tp, ECONNRESET); 1700 rstreason = BANDLIM_UNLIMITED; 1701 goto drop; 1702 } 1703 1704 /* 1705 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1706 * flag is on (half-synchronized state), then queue data for 1707 * later processing; else drop segment and return. 1708 */ 1709 if ((thflags & TH_ACK) == 0) { 1710 if (tp->t_state == TCPS_SYN_RECEIVED || 1711 (tp->t_flags & TF_NEEDSYN)) 1712 goto step6; 1713 else 1714 goto drop; 1715 } 1716 1717 /* 1718 * Ack processing. 1719 */ 1720 switch (tp->t_state) { 1721 1722 /* 1723 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1724 * ESTABLISHED state and continue processing. 1725 * The ACK was checked above. 1726 */ 1727 case TCPS_SYN_RECEIVED: 1728 1729 tcpstat.tcps_connects++; 1730 soisconnected(so); 1731 /* Do window scaling? */ 1732 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1733 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1734 tp->snd_scale = tp->requested_s_scale; 1735 tp->rcv_scale = tp->request_r_scale; 1736 } 1737 /* 1738 * Upon successful completion of 3-way handshake, 1739 * update cache.CC, pass any queued data to the user, 1740 * and advance state appropriately. 1741 */ 1742 if (tcp_do_rfc1644) { 1743 tao.tao_cc = tp->cc_recv; 1744 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1745 tp->cc_recv, 0); 1746 } 1747 /* 1748 * Make transitions: 1749 * SYN-RECEIVED -> ESTABLISHED 1750 * SYN-RECEIVED* -> FIN-WAIT-1 1751 */ 1752 tp->t_starttime = ticks; 1753 if (tp->t_flags & TF_NEEDFIN) { 1754 tp->t_state = TCPS_FIN_WAIT_1; 1755 tp->t_flags &= ~TF_NEEDFIN; 1756 } else { 1757 tp->t_state = TCPS_ESTABLISHED; 1758 callout_reset(tp->tt_keep, tcp_keepidle, 1759 tcp_timer_keep, tp); 1760 } 1761 /* 1762 * If segment contains data or ACK, will call tcp_reass() 1763 * later; if not, do so now to pass queued data to user. 1764 */ 1765 if (tlen == 0 && (thflags & TH_FIN) == 0) 1766 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1767 (struct mbuf *)0); 1768 tp->snd_wl1 = th->th_seq - 1; 1769 /* FALLTHROUGH */ 1770 1771 /* 1772 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1773 * ACKs. If the ack is in the range 1774 * tp->snd_una < th->th_ack <= tp->snd_max 1775 * then advance tp->snd_una to th->th_ack and drop 1776 * data from the retransmission queue. If this ACK reflects 1777 * more up to date window information we update our window information. 1778 */ 1779 case TCPS_ESTABLISHED: 1780 case TCPS_FIN_WAIT_1: 1781 case TCPS_FIN_WAIT_2: 1782 case TCPS_CLOSE_WAIT: 1783 case TCPS_CLOSING: 1784 case TCPS_LAST_ACK: 1785 case TCPS_TIME_WAIT: 1786 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1787 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1788 if (tlen == 0 && tiwin == tp->snd_wnd) { 1789 tcpstat.tcps_rcvdupack++; 1790 /* 1791 * If we have outstanding data (other than 1792 * a window probe), this is a completely 1793 * duplicate ack (ie, window info didn't 1794 * change), the ack is the biggest we've 1795 * seen and we've seen exactly our rexmt 1796 * threshhold of them, assume a packet 1797 * has been dropped and retransmit it. 1798 * Kludge snd_nxt & the congestion 1799 * window so we send only this one 1800 * packet. 1801 * 1802 * We know we're losing at the current 1803 * window size so do congestion avoidance 1804 * (set ssthresh to half the current window 1805 * and pull our congestion window back to 1806 * the new ssthresh). 1807 * 1808 * Dup acks mean that packets have left the 1809 * network (they're now cached at the receiver) 1810 * so bump cwnd by the amount in the receiver 1811 * to keep a constant cwnd packets in the 1812 * network. 1813 */ 1814 if (!callout_active(tp->tt_rexmt) || 1815 th->th_ack != tp->snd_una) 1816 tp->t_dupacks = 0; 1817 else if (++tp->t_dupacks > tcprexmtthresh || 1818 (tcp_do_newreno && 1819 IN_FASTRECOVERY(tp))) { 1820 tp->snd_cwnd += tp->t_maxseg; 1821 (void) tcp_output(tp); 1822 goto drop; 1823 } else if (tp->t_dupacks == tcprexmtthresh) { 1824 tcp_seq onxt = tp->snd_nxt; 1825 u_int win; 1826 if (tcp_do_newreno && 1827 SEQ_LEQ(th->th_ack, 1828 tp->snd_recover)) { 1829 tp->t_dupacks = 0; 1830 break; 1831 } 1832 win = min(tp->snd_wnd, tp->snd_cwnd) / 1833 2 / tp->t_maxseg; 1834 if (win < 2) 1835 win = 2; 1836 tp->snd_ssthresh = win * tp->t_maxseg; 1837 ENTER_FASTRECOVERY(tp); 1838 tp->snd_recover = tp->snd_max; 1839 callout_stop(tp->tt_rexmt); 1840 tp->t_rtttime = 0; 1841 tp->snd_nxt = th->th_ack; 1842 tp->snd_cwnd = tp->t_maxseg; 1843 (void) tcp_output(tp); 1844 KASSERT(tp->snd_limited <= 2, 1845 ("tp->snd_limited too big")); 1846 tp->snd_cwnd = tp->snd_ssthresh + 1847 tp->t_maxseg * 1848 (tp->t_dupacks - tp->snd_limited); 1849 if (SEQ_GT(onxt, tp->snd_nxt)) 1850 tp->snd_nxt = onxt; 1851 goto drop; 1852 } else if (tcp_do_rfc3042) { 1853 u_long oldcwnd = tp->snd_cwnd; 1854 tcp_seq oldsndmax = tp->snd_max; 1855 u_int sent; 1856 KASSERT(tp->t_dupacks == 1 || 1857 tp->t_dupacks == 2, 1858 ("dupacks not 1 or 2")); 1859 if (tp->t_dupacks == 1) 1860 tp->snd_limited = 0; 1861 tp->snd_cwnd = 1862 (tp->snd_nxt - tp->snd_una) + 1863 (tp->t_dupacks - tp->snd_limited) * 1864 tp->t_maxseg; 1865 (void) tcp_output(tp); 1866 sent = tp->snd_max - oldsndmax; 1867 if (sent > tp->t_maxseg) { 1868 KASSERT(tp->snd_limited == 0 && 1869 tp->t_dupacks == 2, 1870 ("sent too much")); 1871 tp->snd_limited = 2; 1872 } else if (sent > 0) 1873 ++tp->snd_limited; 1874 tp->snd_cwnd = oldcwnd; 1875 goto drop; 1876 } 1877 } else 1878 tp->t_dupacks = 0; 1879 break; 1880 } 1881 1882 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1883 1884 /* 1885 * If the congestion window was inflated to account 1886 * for the other side's cached packets, retract it. 1887 */ 1888 if (tcp_do_newreno) { 1889 if (IN_FASTRECOVERY(tp)) { 1890 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1891 tcp_newreno_partial_ack(tp, th); 1892 } else { 1893 /* 1894 * Window inflation should have left us 1895 * with approximately snd_ssthresh 1896 * outstanding data. 1897 * But in case we would be inclined to 1898 * send a burst, better to do it via 1899 * the slow start mechanism. 1900 */ 1901 if (SEQ_GT(th->th_ack + 1902 tp->snd_ssthresh, 1903 tp->snd_max)) 1904 tp->snd_cwnd = tp->snd_max - 1905 th->th_ack + 1906 tp->t_maxseg; 1907 else 1908 tp->snd_cwnd = tp->snd_ssthresh; 1909 } 1910 } 1911 } else { 1912 if (tp->t_dupacks >= tcprexmtthresh && 1913 tp->snd_cwnd > tp->snd_ssthresh) 1914 tp->snd_cwnd = tp->snd_ssthresh; 1915 } 1916 tp->t_dupacks = 0; 1917 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1918 tcpstat.tcps_rcvacktoomuch++; 1919 goto dropafterack; 1920 } 1921 /* 1922 * If we reach this point, ACK is not a duplicate, 1923 * i.e., it ACKs something we sent. 1924 */ 1925 if (tp->t_flags & TF_NEEDSYN) { 1926 /* 1927 * T/TCP: Connection was half-synchronized, and our 1928 * SYN has been ACK'd (so connection is now fully 1929 * synchronized). Go to non-starred state, 1930 * increment snd_una for ACK of SYN, and check if 1931 * we can do window scaling. 1932 */ 1933 tp->t_flags &= ~TF_NEEDSYN; 1934 tp->snd_una++; 1935 /* Do window scaling? */ 1936 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1937 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1938 tp->snd_scale = tp->requested_s_scale; 1939 tp->rcv_scale = tp->request_r_scale; 1940 } 1941 } 1942 1943process_ACK: 1944 acked = th->th_ack - tp->snd_una; 1945 tcpstat.tcps_rcvackpack++; 1946 tcpstat.tcps_rcvackbyte += acked; 1947 1948 /* 1949 * If we just performed our first retransmit, and the ACK 1950 * arrives within our recovery window, then it was a mistake 1951 * to do the retransmit in the first place. Recover our 1952 * original cwnd and ssthresh, and proceed to transmit where 1953 * we left off. 1954 */ 1955 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1956 ++tcpstat.tcps_sndrexmitbad; 1957 tp->snd_cwnd = tp->snd_cwnd_prev; 1958 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1959 tp->snd_recover = tp->snd_recover_prev; 1960 if (tp->t_flags & TF_WASFRECOVERY) 1961 ENTER_FASTRECOVERY(tp); 1962 tp->snd_nxt = tp->snd_max; 1963 tp->t_badrxtwin = 0; /* XXX probably not required */ 1964 } 1965 1966 /* 1967 * If we have a timestamp reply, update smoothed 1968 * round trip time. If no timestamp is present but 1969 * transmit timer is running and timed sequence 1970 * number was acked, update smoothed round trip time. 1971 * Since we now have an rtt measurement, cancel the 1972 * timer backoff (cf., Phil Karn's retransmit alg.). 1973 * Recompute the initial retransmit timer. 1974 * 1975 * Some boxes send broken timestamp replies 1976 * during the SYN+ACK phase, ignore 1977 * timestamps of 0 or we could calculate a 1978 * huge RTT and blow up the retransmit timer. 1979 */ 1980 if ((to.to_flags & TOF_TS) != 0 && 1981 to.to_tsecr) { 1982 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1983 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1984 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1985 } 1986 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1987 1988 /* 1989 * If all outstanding data is acked, stop retransmit 1990 * timer and remember to restart (more output or persist). 1991 * If there is more data to be acked, restart retransmit 1992 * timer, using current (possibly backed-off) value. 1993 */ 1994 if (th->th_ack == tp->snd_max) { 1995 callout_stop(tp->tt_rexmt); 1996 needoutput = 1; 1997 } else if (!callout_active(tp->tt_persist)) 1998 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1999 tcp_timer_rexmt, tp); 2000 2001 /* 2002 * If no data (only SYN) was ACK'd, 2003 * skip rest of ACK processing. 2004 */ 2005 if (acked == 0) 2006 goto step6; 2007 2008 /* 2009 * When new data is acked, open the congestion window. 2010 * If the window gives us less than ssthresh packets 2011 * in flight, open exponentially (maxseg per packet). 2012 * Otherwise open linearly: maxseg per window 2013 * (maxseg^2 / cwnd per packet). 2014 */ 2015 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 2016 register u_int cw = tp->snd_cwnd; 2017 register u_int incr = tp->t_maxseg; 2018 if (cw > tp->snd_ssthresh) 2019 incr = incr * incr / cw; 2020 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2021 } 2022 if (acked > so->so_snd.sb_cc) { 2023 tp->snd_wnd -= so->so_snd.sb_cc; 2024 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2025 ourfinisacked = 1; 2026 } else { 2027 sbdrop(&so->so_snd, acked); 2028 tp->snd_wnd -= acked; 2029 ourfinisacked = 0; 2030 } 2031 sowwakeup(so); 2032 /* detect una wraparound */ 2033 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 2034 SEQ_GT(tp->snd_una, tp->snd_recover) && 2035 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2036 tp->snd_recover = th->th_ack - 1; 2037 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 2038 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2039 EXIT_FASTRECOVERY(tp); 2040 tp->snd_una = th->th_ack; 2041 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2042 tp->snd_nxt = tp->snd_una; 2043 2044 switch (tp->t_state) { 2045 2046 /* 2047 * In FIN_WAIT_1 STATE in addition to the processing 2048 * for the ESTABLISHED state if our FIN is now acknowledged 2049 * then enter FIN_WAIT_2. 2050 */ 2051 case TCPS_FIN_WAIT_1: 2052 if (ourfinisacked) { 2053 /* 2054 * If we can't receive any more 2055 * data, then closing user can proceed. 2056 * Starting the timer is contrary to the 2057 * specification, but if we don't get a FIN 2058 * we'll hang forever. 2059 */ 2060 /* XXXjl 2061 * we should release the tp also, and use a 2062 * compressed state. 2063 */ 2064 if (so->so_state & SS_CANTRCVMORE) { 2065 soisdisconnected(so); 2066 callout_reset(tp->tt_2msl, tcp_maxidle, 2067 tcp_timer_2msl, tp); 2068 } 2069 tp->t_state = TCPS_FIN_WAIT_2; 2070 } 2071 break; 2072 2073 /* 2074 * In CLOSING STATE in addition to the processing for 2075 * the ESTABLISHED state if the ACK acknowledges our FIN 2076 * then enter the TIME-WAIT state, otherwise ignore 2077 * the segment. 2078 */ 2079 case TCPS_CLOSING: 2080 if (ourfinisacked) { 2081 KASSERT(headlocked, ("headlocked")); 2082 tcp_twstart(tp); 2083 INP_INFO_WUNLOCK(&tcbinfo); 2084 m_freem(m); 2085 return; 2086 } 2087 break; 2088 2089 /* 2090 * In LAST_ACK, we may still be waiting for data to drain 2091 * and/or to be acked, as well as for the ack of our FIN. 2092 * If our FIN is now acknowledged, delete the TCB, 2093 * enter the closed state and return. 2094 */ 2095 case TCPS_LAST_ACK: 2096 if (ourfinisacked) { 2097 tp = tcp_close(tp); 2098 goto drop; 2099 } 2100 break; 2101 2102 /* 2103 * In TIME_WAIT state the only thing that should arrive 2104 * is a retransmission of the remote FIN. Acknowledge 2105 * it and restart the finack timer. 2106 */ 2107 case TCPS_TIME_WAIT: 2108 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2109 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2110 tcp_timer_2msl, tp); 2111 goto dropafterack; 2112 } 2113 } 2114 2115step6: 2116 /* 2117 * Update window information. 2118 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2119 */ 2120 if ((thflags & TH_ACK) && 2121 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2122 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2123 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2124 /* keep track of pure window updates */ 2125 if (tlen == 0 && 2126 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2127 tcpstat.tcps_rcvwinupd++; 2128 tp->snd_wnd = tiwin; 2129 tp->snd_wl1 = th->th_seq; 2130 tp->snd_wl2 = th->th_ack; 2131 if (tp->snd_wnd > tp->max_sndwnd) 2132 tp->max_sndwnd = tp->snd_wnd; 2133 needoutput = 1; 2134 } 2135 2136 /* 2137 * Process segments with URG. 2138 */ 2139 if ((thflags & TH_URG) && th->th_urp && 2140 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2141 /* 2142 * This is a kludge, but if we receive and accept 2143 * random urgent pointers, we'll crash in 2144 * soreceive. It's hard to imagine someone 2145 * actually wanting to send this much urgent data. 2146 */ 2147 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2148 th->th_urp = 0; /* XXX */ 2149 thflags &= ~TH_URG; /* XXX */ 2150 goto dodata; /* XXX */ 2151 } 2152 /* 2153 * If this segment advances the known urgent pointer, 2154 * then mark the data stream. This should not happen 2155 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2156 * a FIN has been received from the remote side. 2157 * In these states we ignore the URG. 2158 * 2159 * According to RFC961 (Assigned Protocols), 2160 * the urgent pointer points to the last octet 2161 * of urgent data. We continue, however, 2162 * to consider it to indicate the first octet 2163 * of data past the urgent section as the original 2164 * spec states (in one of two places). 2165 */ 2166 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2167 tp->rcv_up = th->th_seq + th->th_urp; 2168 so->so_oobmark = so->so_rcv.sb_cc + 2169 (tp->rcv_up - tp->rcv_nxt) - 1; 2170 if (so->so_oobmark == 0) 2171 so->so_state |= SS_RCVATMARK; 2172 sohasoutofband(so); 2173 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2174 } 2175 /* 2176 * Remove out of band data so doesn't get presented to user. 2177 * This can happen independent of advancing the URG pointer, 2178 * but if two URG's are pending at once, some out-of-band 2179 * data may creep in... ick. 2180 */ 2181 if (th->th_urp <= (u_long)tlen && 2182 !(so->so_options & SO_OOBINLINE)) { 2183 /* hdr drop is delayed */ 2184 tcp_pulloutofband(so, th, m, drop_hdrlen); 2185 } 2186 } else { 2187 /* 2188 * If no out of band data is expected, 2189 * pull receive urgent pointer along 2190 * with the receive window. 2191 */ 2192 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2193 tp->rcv_up = tp->rcv_nxt; 2194 } 2195dodata: /* XXX */ 2196 KASSERT(headlocked, ("headlocked")); 2197 /* 2198 * Process the segment text, merging it into the TCP sequencing queue, 2199 * and arranging for acknowledgment of receipt if necessary. 2200 * This process logically involves adjusting tp->rcv_wnd as data 2201 * is presented to the user (this happens in tcp_usrreq.c, 2202 * case PRU_RCVD). If a FIN has already been received on this 2203 * connection then we just ignore the text. 2204 */ 2205 if ((tlen || (thflags & TH_FIN)) && 2206 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2207 m_adj(m, drop_hdrlen); /* delayed header drop */ 2208 /* 2209 * Insert segment which includes th into TCP reassembly queue 2210 * with control block tp. Set thflags to whether reassembly now 2211 * includes a segment with FIN. This handles the common case 2212 * inline (segment is the next to be received on an established 2213 * connection, and the queue is empty), avoiding linkage into 2214 * and removal from the queue and repetition of various 2215 * conversions. 2216 * Set DELACK for segments received in order, but ack 2217 * immediately when segments are out of order (so 2218 * fast retransmit can work). 2219 */ 2220 if (th->th_seq == tp->rcv_nxt && 2221 LIST_EMPTY(&tp->t_segq) && 2222 TCPS_HAVEESTABLISHED(tp->t_state)) { 2223 if (DELAY_ACK(tp)) 2224 tp->t_flags |= TF_DELACK; 2225 else 2226 tp->t_flags |= TF_ACKNOW; 2227 tp->rcv_nxt += tlen; 2228 thflags = th->th_flags & TH_FIN; 2229 tcpstat.tcps_rcvpack++; 2230 tcpstat.tcps_rcvbyte += tlen; 2231 ND6_HINT(tp); 2232 if (so->so_state & SS_CANTRCVMORE) 2233 m_freem(m); 2234 else 2235 sbappendstream(&so->so_rcv, m); 2236 sorwakeup(so); 2237 } else { 2238 thflags = tcp_reass(tp, th, &tlen, m); 2239 tp->t_flags |= TF_ACKNOW; 2240 } 2241 2242 /* 2243 * Note the amount of data that peer has sent into 2244 * our window, in order to estimate the sender's 2245 * buffer size. 2246 */ 2247 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2248 } else { 2249 m_freem(m); 2250 thflags &= ~TH_FIN; 2251 } 2252 2253 /* 2254 * If FIN is received ACK the FIN and let the user know 2255 * that the connection is closing. 2256 */ 2257 if (thflags & TH_FIN) { 2258 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2259 socantrcvmore(so); 2260 /* 2261 * If connection is half-synchronized 2262 * (ie NEEDSYN flag on) then delay ACK, 2263 * so it may be piggybacked when SYN is sent. 2264 * Otherwise, since we received a FIN then no 2265 * more input can be expected, send ACK now. 2266 */ 2267 if (tp->t_flags & TF_NEEDSYN) 2268 tp->t_flags |= TF_DELACK; 2269 else 2270 tp->t_flags |= TF_ACKNOW; 2271 tp->rcv_nxt++; 2272 } 2273 switch (tp->t_state) { 2274 2275 /* 2276 * In SYN_RECEIVED and ESTABLISHED STATES 2277 * enter the CLOSE_WAIT state. 2278 */ 2279 case TCPS_SYN_RECEIVED: 2280 tp->t_starttime = ticks; 2281 /*FALLTHROUGH*/ 2282 case TCPS_ESTABLISHED: 2283 tp->t_state = TCPS_CLOSE_WAIT; 2284 break; 2285 2286 /* 2287 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2288 * enter the CLOSING state. 2289 */ 2290 case TCPS_FIN_WAIT_1: 2291 tp->t_state = TCPS_CLOSING; 2292 break; 2293 2294 /* 2295 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2296 * starting the time-wait timer, turning off the other 2297 * standard timers. 2298 */ 2299 case TCPS_FIN_WAIT_2: 2300 KASSERT(headlocked == 1, ("headlocked should be 1")); 2301 tcp_twstart(tp); 2302 INP_INFO_WUNLOCK(&tcbinfo); 2303 return; 2304 2305 /* 2306 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2307 */ 2308 case TCPS_TIME_WAIT: 2309 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2310 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2311 tcp_timer_2msl, tp); 2312 break; 2313 } 2314 } 2315 INP_INFO_WUNLOCK(&tcbinfo); 2316#ifdef TCPDEBUG 2317 if (so->so_options & SO_DEBUG) 2318 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2319 &tcp_savetcp, 0); 2320#endif 2321 2322 /* 2323 * Return any desired output. 2324 */ 2325 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2326 (void) tcp_output(tp); 2327 2328check_delack: 2329 if (tp->t_flags & TF_DELACK) { 2330 tp->t_flags &= ~TF_DELACK; 2331 callout_reset(tp->tt_delack, tcp_delacktime, 2332 tcp_timer_delack, tp); 2333 } 2334 INP_UNLOCK(inp); 2335 return; 2336 2337dropafterack: 2338 /* 2339 * Generate an ACK dropping incoming segment if it occupies 2340 * sequence space, where the ACK reflects our state. 2341 * 2342 * We can now skip the test for the RST flag since all 2343 * paths to this code happen after packets containing 2344 * RST have been dropped. 2345 * 2346 * In the SYN-RECEIVED state, don't send an ACK unless the 2347 * segment we received passes the SYN-RECEIVED ACK test. 2348 * If it fails send a RST. This breaks the loop in the 2349 * "LAND" DoS attack, and also prevents an ACK storm 2350 * between two listening ports that have been sent forged 2351 * SYN segments, each with the source address of the other. 2352 */ 2353 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2354 (SEQ_GT(tp->snd_una, th->th_ack) || 2355 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2356 rstreason = BANDLIM_RST_OPENPORT; 2357 goto dropwithreset; 2358 } 2359#ifdef TCPDEBUG 2360 if (so->so_options & SO_DEBUG) 2361 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2362 &tcp_savetcp, 0); 2363#endif 2364 KASSERT(headlocked, ("headlocked should be 1")); 2365 INP_INFO_WUNLOCK(&tcbinfo); 2366 m_freem(m); 2367 tp->t_flags |= TF_ACKNOW; 2368 (void) tcp_output(tp); 2369 INP_UNLOCK(inp); 2370 return; 2371 2372dropwithreset: 2373 /* 2374 * Generate a RST, dropping incoming segment. 2375 * Make ACK acceptable to originator of segment. 2376 * Don't bother to respond if destination was broadcast/multicast. 2377 */ 2378 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2379 goto drop; 2380 if (isipv6) { 2381 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2382 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2383 goto drop; 2384 } else { 2385 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2386 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2387 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2388 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2389 goto drop; 2390 } 2391 /* IPv6 anycast check is done at tcp6_input() */ 2392 2393 /* 2394 * Perform bandwidth limiting. 2395 */ 2396 if (badport_bandlim(rstreason) < 0) 2397 goto drop; 2398 2399#ifdef TCPDEBUG 2400 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2401 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2402 &tcp_savetcp, 0); 2403#endif 2404 2405 if (thflags & TH_ACK) 2406 /* mtod() below is safe as long as hdr dropping is delayed */ 2407 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2408 TH_RST); 2409 else { 2410 if (thflags & TH_SYN) 2411 tlen++; 2412 /* mtod() below is safe as long as hdr dropping is delayed */ 2413 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2414 (tcp_seq)0, TH_RST|TH_ACK); 2415 } 2416 2417 if (tp) 2418 INP_UNLOCK(inp); 2419 if (headlocked) 2420 INP_INFO_WUNLOCK(&tcbinfo); 2421 return; 2422 2423drop: 2424 /* 2425 * Drop space held by incoming segment and return. 2426 */ 2427#ifdef TCPDEBUG 2428 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2429 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2430 &tcp_savetcp, 0); 2431#endif 2432 if (tp) 2433 INP_UNLOCK(inp); 2434 m_freem(m); 2435 if (headlocked) 2436 INP_INFO_WUNLOCK(&tcbinfo); 2437 return; 2438} 2439 2440/* 2441 * Parse TCP options and place in tcpopt. 2442 */ 2443static void 2444tcp_dooptions(to, cp, cnt, is_syn) 2445 struct tcpopt *to; 2446 u_char *cp; 2447 int cnt; 2448 int is_syn; 2449{ 2450 int opt, optlen; 2451 2452 to->to_flags = 0; 2453 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2454 opt = cp[0]; 2455 if (opt == TCPOPT_EOL) 2456 break; 2457 if (opt == TCPOPT_NOP) 2458 optlen = 1; 2459 else { 2460 if (cnt < 2) 2461 break; 2462 optlen = cp[1]; 2463 if (optlen < 2 || optlen > cnt) 2464 break; 2465 } 2466 switch (opt) { 2467 case TCPOPT_MAXSEG: 2468 if (optlen != TCPOLEN_MAXSEG) 2469 continue; 2470 if (!is_syn) 2471 continue; 2472 to->to_flags |= TOF_MSS; 2473 bcopy((char *)cp + 2, 2474 (char *)&to->to_mss, sizeof(to->to_mss)); 2475 to->to_mss = ntohs(to->to_mss); 2476 break; 2477 case TCPOPT_WINDOW: 2478 if (optlen != TCPOLEN_WINDOW) 2479 continue; 2480 if (! is_syn) 2481 continue; 2482 to->to_flags |= TOF_SCALE; 2483 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2484 break; 2485 case TCPOPT_TIMESTAMP: 2486 if (optlen != TCPOLEN_TIMESTAMP) 2487 continue; 2488 to->to_flags |= TOF_TS; 2489 bcopy((char *)cp + 2, 2490 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2491 to->to_tsval = ntohl(to->to_tsval); 2492 bcopy((char *)cp + 6, 2493 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2494 to->to_tsecr = ntohl(to->to_tsecr); 2495 break; 2496 case TCPOPT_CC: 2497 if (optlen != TCPOLEN_CC) 2498 continue; 2499 to->to_flags |= TOF_CC; 2500 bcopy((char *)cp + 2, 2501 (char *)&to->to_cc, sizeof(to->to_cc)); 2502 to->to_cc = ntohl(to->to_cc); 2503 break; 2504 case TCPOPT_CCNEW: 2505 if (optlen != TCPOLEN_CC) 2506 continue; 2507 if (!is_syn) 2508 continue; 2509 to->to_flags |= TOF_CCNEW; 2510 bcopy((char *)cp + 2, 2511 (char *)&to->to_cc, sizeof(to->to_cc)); 2512 to->to_cc = ntohl(to->to_cc); 2513 break; 2514 case TCPOPT_CCECHO: 2515 if (optlen != TCPOLEN_CC) 2516 continue; 2517 if (!is_syn) 2518 continue; 2519 to->to_flags |= TOF_CCECHO; 2520 bcopy((char *)cp + 2, 2521 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2522 to->to_ccecho = ntohl(to->to_ccecho); 2523 break; 2524#ifdef TCP_SIGNATURE 2525 /* 2526 * XXX In order to reply to a host which has set the 2527 * TCP_SIGNATURE option in its initial SYN, we have to 2528 * record the fact that the option was observed here 2529 * for the syncache code to perform the correct response. 2530 */ 2531 case TCPOPT_SIGNATURE: 2532 if (optlen != TCPOLEN_SIGNATURE) 2533 continue; 2534 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2535 break; 2536#endif 2537 default: 2538 continue; 2539 } 2540 } 2541} 2542 2543/* 2544 * Pull out of band byte out of a segment so 2545 * it doesn't appear in the user's data queue. 2546 * It is still reflected in the segment length for 2547 * sequencing purposes. 2548 */ 2549static void 2550tcp_pulloutofband(so, th, m, off) 2551 struct socket *so; 2552 struct tcphdr *th; 2553 register struct mbuf *m; 2554 int off; /* delayed to be droped hdrlen */ 2555{ 2556 int cnt = off + th->th_urp - 1; 2557 2558 while (cnt >= 0) { 2559 if (m->m_len > cnt) { 2560 char *cp = mtod(m, caddr_t) + cnt; 2561 struct tcpcb *tp = sototcpcb(so); 2562 2563 tp->t_iobc = *cp; 2564 tp->t_oobflags |= TCPOOB_HAVEDATA; 2565 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2566 m->m_len--; 2567 if (m->m_flags & M_PKTHDR) 2568 m->m_pkthdr.len--; 2569 return; 2570 } 2571 cnt -= m->m_len; 2572 m = m->m_next; 2573 if (m == 0) 2574 break; 2575 } 2576 panic("tcp_pulloutofband"); 2577} 2578 2579/* 2580 * Collect new round-trip time estimate 2581 * and update averages and current timeout. 2582 */ 2583static void 2584tcp_xmit_timer(tp, rtt) 2585 register struct tcpcb *tp; 2586 int rtt; 2587{ 2588 register int delta; 2589 2590 tcpstat.tcps_rttupdated++; 2591 tp->t_rttupdated++; 2592 if (tp->t_srtt != 0) { 2593 /* 2594 * srtt is stored as fixed point with 5 bits after the 2595 * binary point (i.e., scaled by 8). The following magic 2596 * is equivalent to the smoothing algorithm in rfc793 with 2597 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2598 * point). Adjust rtt to origin 0. 2599 */ 2600 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2601 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2602 2603 if ((tp->t_srtt += delta) <= 0) 2604 tp->t_srtt = 1; 2605 2606 /* 2607 * We accumulate a smoothed rtt variance (actually, a 2608 * smoothed mean difference), then set the retransmit 2609 * timer to smoothed rtt + 4 times the smoothed variance. 2610 * rttvar is stored as fixed point with 4 bits after the 2611 * binary point (scaled by 16). The following is 2612 * equivalent to rfc793 smoothing with an alpha of .75 2613 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2614 * rfc793's wired-in beta. 2615 */ 2616 if (delta < 0) 2617 delta = -delta; 2618 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2619 if ((tp->t_rttvar += delta) <= 0) 2620 tp->t_rttvar = 1; 2621 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2622 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2623 } else { 2624 /* 2625 * No rtt measurement yet - use the unsmoothed rtt. 2626 * Set the variance to half the rtt (so our first 2627 * retransmit happens at 3*rtt). 2628 */ 2629 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2630 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2631 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2632 } 2633 tp->t_rtttime = 0; 2634 tp->t_rxtshift = 0; 2635 2636 /* 2637 * the retransmit should happen at rtt + 4 * rttvar. 2638 * Because of the way we do the smoothing, srtt and rttvar 2639 * will each average +1/2 tick of bias. When we compute 2640 * the retransmit timer, we want 1/2 tick of rounding and 2641 * 1 extra tick because of +-1/2 tick uncertainty in the 2642 * firing of the timer. The bias will give us exactly the 2643 * 1.5 tick we need. But, because the bias is 2644 * statistical, we have to test that we don't drop below 2645 * the minimum feasible timer (which is 2 ticks). 2646 */ 2647 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2648 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2649 2650 /* 2651 * We received an ack for a packet that wasn't retransmitted; 2652 * it is probably safe to discard any error indications we've 2653 * received recently. This isn't quite right, but close enough 2654 * for now (a route might have failed after we sent a segment, 2655 * and the return path might not be symmetrical). 2656 */ 2657 tp->t_softerror = 0; 2658} 2659 2660/* 2661 * Determine a reasonable value for maxseg size. 2662 * If the route is known, check route for mtu. 2663 * If none, use an mss that can be handled on the outgoing 2664 * interface without forcing IP to fragment; if bigger than 2665 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2666 * to utilize large mbufs. If no route is found, route has no mtu, 2667 * or the destination isn't local, use a default, hopefully conservative 2668 * size (usually 512 or the default IP max size, but no more than the mtu 2669 * of the interface), as we can't discover anything about intervening 2670 * gateways or networks. We also initialize the congestion/slow start 2671 * window to be a single segment if the destination isn't local. 2672 * While looking at the routing entry, we also initialize other path-dependent 2673 * parameters from pre-set or cached values in the routing entry. 2674 * 2675 * Also take into account the space needed for options that we 2676 * send regularly. Make maxseg shorter by that amount to assure 2677 * that we can send maxseg amount of data even when the options 2678 * are present. Store the upper limit of the length of options plus 2679 * data in maxopd. 2680 * 2681 * 2682 * In case of T/TCP, we call this routine during implicit connection 2683 * setup as well (offer = -1), to initialize maxseg from the cached 2684 * MSS of our peer. 2685 * 2686 * NOTE that this routine is only called when we process an incoming 2687 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2688 */ 2689void 2690tcp_mss(tp, offer) 2691 struct tcpcb *tp; 2692 int offer; 2693{ 2694 int rtt, mss; 2695 u_long bufsize; 2696 u_long maxmtu; 2697 struct inpcb *inp = tp->t_inpcb; 2698 struct socket *so; 2699 struct hc_metrics_lite metrics; 2700 struct rmxp_tao tao; 2701 int origoffer = offer; 2702#ifdef INET6 2703 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2704 size_t min_protoh = isipv6 ? 2705 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2706 sizeof (struct tcpiphdr); 2707#else 2708 const size_t min_protoh = sizeof(struct tcpiphdr); 2709#endif 2710 bzero(&tao, sizeof(tao)); 2711 2712 /* initialize */ 2713#ifdef INET6 2714 if (isipv6) { 2715 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2716 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2717 } else 2718#endif 2719 { 2720 maxmtu = tcp_maxmtu(&inp->inp_inc); 2721 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2722 } 2723 so = inp->inp_socket; 2724 2725 /* 2726 * no route to sender, take default mss and return 2727 */ 2728 if (maxmtu == 0) 2729 return; 2730 2731 /* what have we got? */ 2732 switch (offer) { 2733 case 0: 2734 /* 2735 * Offer == 0 means that there was no MSS on the SYN 2736 * segment, in this case we use tcp_mssdflt. 2737 */ 2738 offer = 2739#ifdef INET6 2740 isipv6 ? tcp_v6mssdflt : 2741#endif 2742 tcp_mssdflt; 2743 break; 2744 2745 case -1: 2746 /* 2747 * Offer == -1 means that we didn't receive SYN yet, 2748 * use cached value in that case; 2749 */ 2750 if (tcp_do_rfc1644) 2751 tcp_hc_gettao(&inp->inp_inc, &tao); 2752 if (tao.tao_mssopt != 0) 2753 offer = tao.tao_mssopt; 2754 /* FALLTHROUGH */ 2755 2756 default: 2757 /* 2758 * Prevent DoS attack with too small MSS. Round up 2759 * to at least minmss. 2760 */ 2761 offer = max(offer, tcp_minmss); 2762 /* 2763 * Sanity check: make sure that maxopd will be large 2764 * enough to allow some data on segments even if the 2765 * all the option space is used (40bytes). Otherwise 2766 * funny things may happen in tcp_output. 2767 */ 2768 offer = max(offer, 64); 2769 if (tcp_do_rfc1644) 2770 tcp_hc_updatetao(&inp->inp_inc, 2771 TCP_HC_TAO_MSSOPT, 0, offer); 2772 } 2773 2774 /* 2775 * rmx information is now retrieved from tcp_hostcache 2776 */ 2777 tcp_hc_get(&inp->inp_inc, &metrics); 2778 2779 /* 2780 * if there's a discovered mtu int tcp hostcache, use it 2781 * else, use the link mtu. 2782 */ 2783 if (metrics.rmx_mtu) 2784 mss = metrics.rmx_mtu - min_protoh; 2785 else { 2786#ifdef INET6 2787 if (isipv6) { 2788 mss = maxmtu - min_protoh; 2789 if (!path_mtu_discovery && 2790 !in6_localaddr(&inp->in6p_faddr)) 2791 mss = min(mss, tcp_v6mssdflt); 2792 } else 2793#endif 2794 { 2795 mss = maxmtu - min_protoh; 2796 if (!path_mtu_discovery && 2797 !in_localaddr(inp->inp_faddr)) 2798 mss = min(mss, tcp_mssdflt); 2799 } 2800 } 2801 mss = min(mss, offer); 2802 2803 /* 2804 * maxopd stores the maximum length of data AND options 2805 * in a segment; maxseg is the amount of data in a normal 2806 * segment. We need to store this value (maxopd) apart 2807 * from maxseg, because now every segment carries options 2808 * and thus we normally have somewhat less data in segments. 2809 */ 2810 tp->t_maxopd = mss; 2811 2812 /* 2813 * In case of T/TCP, origoffer==-1 indicates, that no segments 2814 * were received yet. In this case we just guess, otherwise 2815 * we do the same as before T/TCP. 2816 */ 2817 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2818 (origoffer == -1 || 2819 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2820 mss -= TCPOLEN_TSTAMP_APPA; 2821 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2822 (origoffer == -1 || 2823 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2824 mss -= TCPOLEN_CC_APPA; 2825 tp->t_maxseg = mss; 2826 2827#if (MCLBYTES & (MCLBYTES - 1)) == 0 2828 if (mss > MCLBYTES) 2829 mss &= ~(MCLBYTES-1); 2830#else 2831 if (mss > MCLBYTES) 2832 mss = mss / MCLBYTES * MCLBYTES; 2833#endif 2834 tp->t_maxseg = mss; 2835 2836 /* 2837 * If there's a pipesize, change the socket buffer to that size, 2838 * don't change if sb_hiwat is different than default (then it 2839 * has been changed on purpose with setsockopt). 2840 * Make the socket buffers an integral number of mss units; 2841 * if the mss is larger than the socket buffer, decrease the mss. 2842 */ 2843 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2844 bufsize = metrics.rmx_sendpipe; 2845 else 2846 bufsize = so->so_snd.sb_hiwat; 2847 if (bufsize < mss) 2848 mss = bufsize; 2849 else { 2850 bufsize = roundup(bufsize, mss); 2851 if (bufsize > sb_max) 2852 bufsize = sb_max; 2853 if (bufsize > so->so_snd.sb_hiwat) 2854 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2855 } 2856 tp->t_maxseg = mss; 2857 2858 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2859 bufsize = metrics.rmx_recvpipe; 2860 else 2861 bufsize = so->so_rcv.sb_hiwat; 2862 if (bufsize > mss) { 2863 bufsize = roundup(bufsize, mss); 2864 if (bufsize > sb_max) 2865 bufsize = sb_max; 2866 if (bufsize > so->so_rcv.sb_hiwat) 2867 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2868 } 2869 /* 2870 * While we're here, check the others too 2871 */ 2872 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2873 tp->t_srtt = rtt; 2874 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2875 tcpstat.tcps_usedrtt++; 2876 if (metrics.rmx_rttvar) { 2877 tp->t_rttvar = metrics.rmx_rttvar; 2878 tcpstat.tcps_usedrttvar++; 2879 } else { 2880 /* default variation is +- 1 rtt */ 2881 tp->t_rttvar = 2882 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2883 } 2884 TCPT_RANGESET(tp->t_rxtcur, 2885 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2886 tp->t_rttmin, TCPTV_REXMTMAX); 2887 } 2888 if (metrics.rmx_ssthresh) { 2889 /* 2890 * There's some sort of gateway or interface 2891 * buffer limit on the path. Use this to set 2892 * the slow start threshhold, but set the 2893 * threshold to no less than 2*mss. 2894 */ 2895 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2896 tcpstat.tcps_usedssthresh++; 2897 } 2898 if (metrics.rmx_bandwidth) 2899 tp->snd_bandwidth = metrics.rmx_bandwidth; 2900 2901 /* 2902 * Set the slow-start flight size depending on whether this 2903 * is a local network or not. 2904 * 2905 * Extend this so we cache the cwnd too and retrieve it here. 2906 * Make cwnd even bigger than RFC3390 suggests but only if we 2907 * have previous experience with the remote host. Be careful 2908 * not make cwnd bigger than remote receive window or our own 2909 * send socket buffer. Maybe put some additional upper bound 2910 * on the retrieved cwnd. Should do incremental updates to 2911 * hostcache when cwnd collapses so next connection doesn't 2912 * overloads the path again. 2913 * 2914 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2915 * We currently check only in syncache_socket for that. 2916 */ 2917#define TCP_METRICS_CWND 2918#ifdef TCP_METRICS_CWND 2919 if (metrics.rmx_cwnd) 2920 tp->snd_cwnd = max(mss, 2921 min(metrics.rmx_cwnd / 2, 2922 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2923 else 2924#endif 2925 if (tcp_do_rfc3390) 2926 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2927#ifdef INET6 2928 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2929 (!isipv6 && in_localaddr(inp->inp_faddr))) 2930#else 2931 else if (in_localaddr(inp->inp_faddr)) 2932#endif 2933 tp->snd_cwnd = mss * ss_fltsz_local; 2934 else 2935 tp->snd_cwnd = mss * ss_fltsz; 2936} 2937 2938/* 2939 * Determine the MSS option to send on an outgoing SYN. 2940 */ 2941int 2942tcp_mssopt(inc) 2943 struct in_conninfo *inc; 2944{ 2945 int mss = 0; 2946 u_long maxmtu = 0; 2947 u_long thcmtu = 0; 2948 size_t min_protoh; 2949#ifdef INET6 2950 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2951#endif 2952 2953 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2954 2955#ifdef INET6 2956 if (isipv6) { 2957 mss = tcp_v6mssdflt; 2958 maxmtu = tcp_maxmtu6(inc); 2959 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2960 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2961 } else 2962#endif 2963 { 2964 mss = tcp_mssdflt; 2965 maxmtu = tcp_maxmtu(inc); 2966 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2967 min_protoh = sizeof(struct tcpiphdr); 2968 } 2969 if (maxmtu && thcmtu) 2970 mss = min(maxmtu, thcmtu) - min_protoh; 2971 else if (maxmtu || thcmtu) 2972 mss = max(maxmtu, thcmtu) - min_protoh; 2973 2974 return (mss); 2975} 2976 2977 2978/* 2979 * On a partial ack arrives, force the retransmission of the 2980 * next unacknowledged segment. Do not clear tp->t_dupacks. 2981 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2982 * be started again. 2983 */ 2984static void 2985tcp_newreno_partial_ack(tp, th) 2986 struct tcpcb *tp; 2987 struct tcphdr *th; 2988{ 2989 tcp_seq onxt = tp->snd_nxt; 2990 u_long ocwnd = tp->snd_cwnd; 2991 2992 callout_stop(tp->tt_rexmt); 2993 tp->t_rtttime = 0; 2994 tp->snd_nxt = th->th_ack; 2995 /* 2996 * Set snd_cwnd to one segment beyond acknowledged offset. 2997 * (tp->snd_una has not yet been updated when this function is called.) 2998 */ 2999 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3000 tp->t_flags |= TF_ACKNOW; 3001 (void) tcp_output(tp); 3002 tp->snd_cwnd = ocwnd; 3003 if (SEQ_GT(onxt, tp->snd_nxt)) 3004 tp->snd_nxt = onxt; 3005 /* 3006 * Partial window deflation. Relies on fact that tp->snd_una 3007 * not updated yet. 3008 */ 3009 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3010} 3011 3012/* 3013 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3014 * looking for a pcb in the listen state. Returns 0 otherwise. 3015 */ 3016static int 3017tcp_timewait(tw, to, th, m, tlen) 3018 struct tcptw *tw; 3019 struct tcpopt *to; 3020 struct tcphdr *th; 3021 struct mbuf *m; 3022 int tlen; 3023{ 3024 int thflags; 3025 tcp_seq seq; 3026#ifdef INET6 3027 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3028#else 3029 const int isipv6 = 0; 3030#endif 3031 3032 thflags = th->th_flags; 3033 3034 /* 3035 * NOTE: for FIN_WAIT_2 (to be added later), 3036 * must validate sequence number before accepting RST 3037 */ 3038 3039 /* 3040 * If the segment contains RST: 3041 * Drop the segment - see Stevens, vol. 2, p. 964 and 3042 * RFC 1337. 3043 */ 3044 if (thflags & TH_RST) 3045 goto drop; 3046 3047 /* 3048 * If segment contains a SYN and CC [not CC.NEW] option: 3049 * if connection duration > MSL, drop packet and send RST; 3050 * 3051 * if SEG.CC > CCrecv then is new SYN. 3052 * Complete close and delete TCPCB. Then reprocess 3053 * segment, hoping to find new TCPCB in LISTEN state; 3054 * 3055 * else must be old SYN; drop it. 3056 * else do normal processing. 3057 */ 3058 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3059 if ((ticks - tw->t_starttime) > tcp_msl) 3060 goto reset; 3061 if (CC_GT(to->to_cc, tw->cc_recv)) { 3062 (void) tcp_twclose(tw, 0); 3063 return (1); 3064 } 3065 goto drop; 3066 } 3067 3068#if 0 3069/* PAWS not needed at the moment */ 3070 /* 3071 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3072 * and it's less than ts_recent, drop it. 3073 */ 3074 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3075 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3076 if ((thflags & TH_ACK) == 0) 3077 goto drop; 3078 goto ack; 3079 } 3080 /* 3081 * ts_recent is never updated because we never accept new segments. 3082 */ 3083#endif 3084 3085 /* 3086 * If a new connection request is received 3087 * while in TIME_WAIT, drop the old connection 3088 * and start over if the sequence numbers 3089 * are above the previous ones. 3090 */ 3091 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3092 (void) tcp_twclose(tw, 0); 3093 return (1); 3094 } 3095 3096 /* 3097 * Drop the the segment if it does not contain an ACK. 3098 */ 3099 if ((thflags & TH_ACK) == 0) 3100 goto drop; 3101 3102 /* 3103 * Reset the 2MSL timer if this is a duplicate FIN. 3104 */ 3105 if (thflags & TH_FIN) { 3106 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3107 if (seq + 1 == tw->rcv_nxt) 3108 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3109 } 3110 3111 /* 3112 * Acknowledge the segment if it has data or is not a duplicate ACK. 3113 */ 3114 if (thflags != TH_ACK || tlen != 0 || 3115 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3116 tcp_twrespond(tw, NULL, m, TH_ACK); 3117 goto drop; 3118 3119reset: 3120 /* 3121 * Generate a RST, dropping incoming segment. 3122 * Make ACK acceptable to originator of segment. 3123 * Don't bother to respond if destination was broadcast/multicast. 3124 */ 3125 if (m->m_flags & (M_BCAST|M_MCAST)) 3126 goto drop; 3127 if (isipv6) { 3128 struct ip6_hdr *ip6; 3129 3130 /* IPv6 anycast check is done at tcp6_input() */ 3131 ip6 = mtod(m, struct ip6_hdr *); 3132 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3133 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3134 goto drop; 3135 } else { 3136 struct ip *ip; 3137 3138 ip = mtod(m, struct ip *); 3139 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3140 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3141 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3142 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3143 goto drop; 3144 } 3145 if (thflags & TH_ACK) { 3146 tcp_respond(NULL, 3147 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3148 } else { 3149 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3150 tcp_respond(NULL, 3151 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3152 } 3153 INP_UNLOCK(tw->tw_inpcb); 3154 return (0); 3155 3156drop: 3157 INP_UNLOCK(tw->tw_inpcb); 3158 m_freem(m); 3159 return (0); 3160} 3161