tcp_input.c revision 142031
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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD: head/sys/netinet/tcp_input.c 142031 2005-02-17 23:04:56Z ps $ 31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_input.h" 40#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/kernel.h> 44#include <sys/mac.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/proc.h> /* for proc0 declaration */ 48#include <sys/protosw.h> 49#include <sys/signalvar.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53#include <sys/syslog.h> 54#include <sys/systm.h> 55 56#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58#include <vm/uma.h> 59 60#include <net/if.h> 61#include <net/route.h> 62 63#include <netinet/in.h> 64#include <netinet/in_pcb.h> 65#include <netinet/in_systm.h> 66#include <netinet/in_var.h> 67#include <netinet/ip.h> 68#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70#include <netinet/ip_var.h> 71#include <netinet/ip6.h> 72#include <netinet/icmp6.h> 73#include <netinet6/in6_pcb.h> 74#include <netinet6/ip6_var.h> 75#include <netinet6/nd6.h> 76#include <netinet/tcp.h> 77#include <netinet/tcp_fsm.h> 78#include <netinet/tcp_seq.h> 79#include <netinet/tcp_timer.h> 80#include <netinet/tcp_var.h> 81#include <netinet6/tcp6_var.h> 82#include <netinet/tcpip.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif /* TCPDEBUG */ 86 87#ifdef FAST_IPSEC 88#include <netipsec/ipsec.h> 89#include <netipsec/ipsec6.h> 90#endif /*FAST_IPSEC*/ 91 92#ifdef IPSEC 93#include <netinet6/ipsec.h> 94#include <netinet6/ipsec6.h> 95#include <netkey/key.h> 96#endif /*IPSEC*/ 97 98#include <machine/in_cksum.h> 99 100static const int tcprexmtthresh = 3; 101 102struct tcpstat tcpstat; 103SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106static int log_in_vain = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &log_in_vain, 0, "Log all incoming TCP connections"); 109 110static int blackhole = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST when dropping refused connections"); 113 114int tcp_delack_enabled = 1; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119#ifdef TCP_DROP_SYNFIN 120static int drop_synfin = 0; 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 123#endif 124 125static int tcp_do_rfc3042 = 1; 126SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 127 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 128 129static int tcp_do_rfc3390 = 1; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 131 &tcp_do_rfc3390, 0, 132 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 133 134static int tcp_insecure_rst = 0; 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 136 &tcp_insecure_rst, 0, 137 "Follow the old (insecure) criteria for accepting RST packets."); 138 139SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 140 "TCP Segment Reassembly Queue"); 141 142static int tcp_reass_maxseg = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 144 &tcp_reass_maxseg, 0, 145 "Global maximum number of TCP Segments in Reassembly Queue"); 146 147int tcp_reass_qsize = 0; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 149 &tcp_reass_qsize, 0, 150 "Global number of TCP Segments currently in Reassembly Queue"); 151 152static int tcp_reass_maxqlen = 48; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 154 &tcp_reass_maxqlen, 0, 155 "Maximum number of TCP Segments per individual Reassembly Queue"); 156 157static int tcp_reass_overflows = 0; 158SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 159 &tcp_reass_overflows, 0, 160 "Global number of TCP Segment Reassembly Queue Overflows"); 161 162struct inpcbhead tcb; 163#define tcb6 tcb /* for KAME src sync over BSD*'s */ 164struct inpcbinfo tcbinfo; 165struct mtx *tcbinfo_mtx; 166 167static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *, 168 int, int, struct tcphdr *); 169 170static void tcp_pulloutofband(struct socket *, 171 struct tcphdr *, struct mbuf *, int); 172static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 173 struct mbuf *); 174static void tcp_xmit_timer(struct tcpcb *, int); 175static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 176static int tcp_timewait(struct tcptw *, struct tcpopt *, 177 struct tcphdr *, struct mbuf *, int); 178 179/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 180#ifdef INET6 181#define ND6_HINT(tp) \ 182do { \ 183 if ((tp) && (tp)->t_inpcb && \ 184 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 185 nd6_nud_hint(NULL, NULL, 0); \ 186} while (0) 187#else 188#define ND6_HINT(tp) 189#endif 190 191/* 192 * Indicate whether this ack should be delayed. We can delay the ack if 193 * - there is no delayed ack timer in progress and 194 * - our last ack wasn't a 0-sized window. We never want to delay 195 * the ack that opens up a 0-sized window and 196 * - delayed acks are enabled or 197 * - this is a half-synchronized T/TCP connection. 198 */ 199#define DELAY_ACK(tp) \ 200 ((!callout_active(tp->tt_delack) && \ 201 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 202 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 203 204/* Initialize TCP reassembly queue */ 205uma_zone_t tcp_reass_zone; 206void 207tcp_reass_init() 208{ 209 tcp_reass_maxseg = nmbclusters / 16; 210 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 211 &tcp_reass_maxseg); 212 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 213 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 214 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 215} 216 217static int 218tcp_reass(tp, th, tlenp, m) 219 register struct tcpcb *tp; 220 register struct tcphdr *th; 221 int *tlenp; 222 struct mbuf *m; 223{ 224 struct tseg_qent *q; 225 struct tseg_qent *p = NULL; 226 struct tseg_qent *nq; 227 struct tseg_qent *te = NULL; 228 struct socket *so = tp->t_inpcb->inp_socket; 229 int flags; 230 231 INP_LOCK_ASSERT(tp->t_inpcb); 232 233 /* 234 * XXX: tcp_reass() is rather inefficient with its data structures 235 * and should be rewritten (see NetBSD for optimizations). While 236 * doing that it should move to its own file tcp_reass.c. 237 */ 238 239 /* 240 * Call with th==NULL after become established to 241 * force pre-ESTABLISHED data up to user socket. 242 */ 243 if (th == NULL) 244 goto present; 245 246 /* 247 * Limit the number of segments in the reassembly queue to prevent 248 * holding on to too many segments (and thus running out of mbufs). 249 * Make sure to let the missing segment through which caused this 250 * queue. Always keep one global queue entry spare to be able to 251 * process the missing segment. 252 */ 253 if (th->th_seq != tp->rcv_nxt && 254 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 255 tp->t_segqlen >= tcp_reass_maxqlen)) { 256 tcp_reass_overflows++; 257 tcpstat.tcps_rcvmemdrop++; 258 m_freem(m); 259 return (0); 260 } 261 262 /* 263 * Allocate a new queue entry. If we can't, or hit the zone limit 264 * just drop the pkt. 265 */ 266 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 267 if (te == NULL) { 268 tcpstat.tcps_rcvmemdrop++; 269 m_freem(m); 270 return (0); 271 } 272 tp->t_segqlen++; 273 tcp_reass_qsize++; 274 275 /* 276 * Find a segment which begins after this one does. 277 */ 278 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 279 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 280 break; 281 p = q; 282 } 283 284 /* 285 * If there is a preceding segment, it may provide some of 286 * our data already. If so, drop the data from the incoming 287 * segment. If it provides all of our data, drop us. 288 */ 289 if (p != NULL) { 290 register int i; 291 /* conversion to int (in i) handles seq wraparound */ 292 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 293 if (i > 0) { 294 if (i >= *tlenp) { 295 tcpstat.tcps_rcvduppack++; 296 tcpstat.tcps_rcvdupbyte += *tlenp; 297 m_freem(m); 298 uma_zfree(tcp_reass_zone, te); 299 tp->t_segqlen--; 300 tcp_reass_qsize--; 301 /* 302 * Try to present any queued data 303 * at the left window edge to the user. 304 * This is needed after the 3-WHS 305 * completes. 306 */ 307 goto present; /* ??? */ 308 } 309 m_adj(m, i); 310 *tlenp -= i; 311 th->th_seq += i; 312 } 313 } 314 tcpstat.tcps_rcvoopack++; 315 tcpstat.tcps_rcvoobyte += *tlenp; 316 317 /* 318 * While we overlap succeeding segments trim them or, 319 * if they are completely covered, dequeue them. 320 */ 321 while (q) { 322 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 323 if (i <= 0) 324 break; 325 if (i < q->tqe_len) { 326 q->tqe_th->th_seq += i; 327 q->tqe_len -= i; 328 m_adj(q->tqe_m, i); 329 break; 330 } 331 332 nq = LIST_NEXT(q, tqe_q); 333 LIST_REMOVE(q, tqe_q); 334 m_freem(q->tqe_m); 335 uma_zfree(tcp_reass_zone, q); 336 tp->t_segqlen--; 337 tcp_reass_qsize--; 338 q = nq; 339 } 340 341 /* Insert the new segment queue entry into place. */ 342 te->tqe_m = m; 343 te->tqe_th = th; 344 te->tqe_len = *tlenp; 345 346 if (p == NULL) { 347 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 348 } else { 349 LIST_INSERT_AFTER(p, te, tqe_q); 350 } 351 352present: 353 /* 354 * Present data to user, advancing rcv_nxt through 355 * completed sequence space. 356 */ 357 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 358 return (0); 359 q = LIST_FIRST(&tp->t_segq); 360 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 361 return (0); 362 SOCKBUF_LOCK(&so->so_rcv); 363 do { 364 tp->rcv_nxt += q->tqe_len; 365 flags = q->tqe_th->th_flags & TH_FIN; 366 nq = LIST_NEXT(q, tqe_q); 367 LIST_REMOVE(q, tqe_q); 368 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 369 m_freem(q->tqe_m); 370 else 371 sbappendstream_locked(&so->so_rcv, q->tqe_m); 372 uma_zfree(tcp_reass_zone, q); 373 tp->t_segqlen--; 374 tcp_reass_qsize--; 375 q = nq; 376 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 377 ND6_HINT(tp); 378 sorwakeup_locked(so); 379 return (flags); 380} 381 382/* 383 * TCP input routine, follows pages 65-76 of the 384 * protocol specification dated September, 1981 very closely. 385 */ 386#ifdef INET6 387int 388tcp6_input(mp, offp, proto) 389 struct mbuf **mp; 390 int *offp, proto; 391{ 392 register struct mbuf *m = *mp; 393 struct in6_ifaddr *ia6; 394 395 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 396 397 /* 398 * draft-itojun-ipv6-tcp-to-anycast 399 * better place to put this in? 400 */ 401 ia6 = ip6_getdstifaddr(m); 402 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 403 struct ip6_hdr *ip6; 404 405 ip6 = mtod(m, struct ip6_hdr *); 406 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 407 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 408 return IPPROTO_DONE; 409 } 410 411 tcp_input(m, *offp); 412 return IPPROTO_DONE; 413} 414#endif 415 416void 417tcp_input(m, off0) 418 register struct mbuf *m; 419 int off0; 420{ 421 register struct tcphdr *th; 422 register struct ip *ip = NULL; 423 register struct ipovly *ipov; 424 register struct inpcb *inp = NULL; 425 u_char *optp = NULL; 426 int optlen = 0; 427 int len, tlen, off; 428 int drop_hdrlen; 429 register struct tcpcb *tp = 0; 430 register int thflags; 431 struct socket *so = 0; 432 int todrop, acked, ourfinisacked, needoutput = 0; 433 u_long tiwin; 434 struct tcpopt to; /* options in this segment */ 435 int headlocked = 0; 436#ifdef IPFIREWALL_FORWARD 437 struct m_tag *fwd_tag; 438#endif 439 int rstreason; /* For badport_bandlim accounting purposes */ 440 441 struct ip6_hdr *ip6 = NULL; 442#ifdef INET6 443 int isipv6; 444#else 445 const int isipv6 = 0; 446#endif 447 448#ifdef TCPDEBUG 449 /* 450 * The size of tcp_saveipgen must be the size of the max ip header, 451 * now IPv6. 452 */ 453 u_char tcp_saveipgen[40]; 454 struct tcphdr tcp_savetcp; 455 short ostate = 0; 456#endif 457 458#ifdef INET6 459 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 460#endif 461 bzero((char *)&to, sizeof(to)); 462 463 tcpstat.tcps_rcvtotal++; 464 465 if (isipv6) { 466#ifdef INET6 467 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 468 ip6 = mtod(m, struct ip6_hdr *); 469 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 470 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 471 tcpstat.tcps_rcvbadsum++; 472 goto drop; 473 } 474 th = (struct tcphdr *)((caddr_t)ip6 + off0); 475 476 /* 477 * Be proactive about unspecified IPv6 address in source. 478 * As we use all-zero to indicate unbounded/unconnected pcb, 479 * unspecified IPv6 address can be used to confuse us. 480 * 481 * Note that packets with unspecified IPv6 destination is 482 * already dropped in ip6_input. 483 */ 484 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 485 /* XXX stat */ 486 goto drop; 487 } 488#else 489 th = NULL; /* XXX: avoid compiler warning */ 490#endif 491 } else { 492 /* 493 * Get IP and TCP header together in first mbuf. 494 * Note: IP leaves IP header in first mbuf. 495 */ 496 if (off0 > sizeof (struct ip)) { 497 ip_stripoptions(m, (struct mbuf *)0); 498 off0 = sizeof(struct ip); 499 } 500 if (m->m_len < sizeof (struct tcpiphdr)) { 501 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 502 tcpstat.tcps_rcvshort++; 503 return; 504 } 505 } 506 ip = mtod(m, struct ip *); 507 ipov = (struct ipovly *)ip; 508 th = (struct tcphdr *)((caddr_t)ip + off0); 509 tlen = ip->ip_len; 510 511 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 512 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 513 th->th_sum = m->m_pkthdr.csum_data; 514 else 515 th->th_sum = in_pseudo(ip->ip_src.s_addr, 516 ip->ip_dst.s_addr, 517 htonl(m->m_pkthdr.csum_data + 518 ip->ip_len + 519 IPPROTO_TCP)); 520 th->th_sum ^= 0xffff; 521#ifdef TCPDEBUG 522 ipov->ih_len = (u_short)tlen; 523 ipov->ih_len = htons(ipov->ih_len); 524#endif 525 } else { 526 /* 527 * Checksum extended TCP header and data. 528 */ 529 len = sizeof (struct ip) + tlen; 530 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 531 ipov->ih_len = (u_short)tlen; 532 ipov->ih_len = htons(ipov->ih_len); 533 th->th_sum = in_cksum(m, len); 534 } 535 if (th->th_sum) { 536 tcpstat.tcps_rcvbadsum++; 537 goto drop; 538 } 539#ifdef INET6 540 /* Re-initialization for later version check */ 541 ip->ip_v = IPVERSION; 542#endif 543 } 544 545 /* 546 * Check that TCP offset makes sense, 547 * pull out TCP options and adjust length. XXX 548 */ 549 off = th->th_off << 2; 550 if (off < sizeof (struct tcphdr) || off > tlen) { 551 tcpstat.tcps_rcvbadoff++; 552 goto drop; 553 } 554 tlen -= off; /* tlen is used instead of ti->ti_len */ 555 if (off > sizeof (struct tcphdr)) { 556 if (isipv6) { 557#ifdef INET6 558 IP6_EXTHDR_CHECK(m, off0, off, ); 559 ip6 = mtod(m, struct ip6_hdr *); 560 th = (struct tcphdr *)((caddr_t)ip6 + off0); 561#endif 562 } else { 563 if (m->m_len < sizeof(struct ip) + off) { 564 if ((m = m_pullup(m, sizeof (struct ip) + off)) 565 == 0) { 566 tcpstat.tcps_rcvshort++; 567 return; 568 } 569 ip = mtod(m, struct ip *); 570 ipov = (struct ipovly *)ip; 571 th = (struct tcphdr *)((caddr_t)ip + off0); 572 } 573 } 574 optlen = off - sizeof (struct tcphdr); 575 optp = (u_char *)(th + 1); 576 } 577 thflags = th->th_flags; 578 579#ifdef TCP_DROP_SYNFIN 580 /* 581 * If the drop_synfin option is enabled, drop all packets with 582 * both the SYN and FIN bits set. This prevents e.g. nmap from 583 * identifying the TCP/IP stack. 584 * 585 * This is a violation of the TCP specification. 586 */ 587 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 588 goto drop; 589#endif 590 591 /* 592 * Convert TCP protocol specific fields to host format. 593 */ 594 th->th_seq = ntohl(th->th_seq); 595 th->th_ack = ntohl(th->th_ack); 596 th->th_win = ntohs(th->th_win); 597 th->th_urp = ntohs(th->th_urp); 598 599 /* 600 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 601 * until after ip6_savecontrol() is called and before other functions 602 * which don't want those proto headers. 603 * Because ip6_savecontrol() is going to parse the mbuf to 604 * search for data to be passed up to user-land, it wants mbuf 605 * parameters to be unchanged. 606 * XXX: the call of ip6_savecontrol() has been obsoleted based on 607 * latest version of the advanced API (20020110). 608 */ 609 drop_hdrlen = off0 + off; 610 611 /* 612 * Locate pcb for segment. 613 */ 614 INP_INFO_WLOCK(&tcbinfo); 615 headlocked = 1; 616findpcb: 617 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 618#ifdef IPFIREWALL_FORWARD 619 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 620 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 621 622 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 623 struct sockaddr_in *next_hop; 624 625 next_hop = (struct sockaddr_in *)(fwd_tag+1); 626 /* 627 * Transparently forwarded. Pretend to be the destination. 628 * already got one like this? 629 */ 630 inp = in_pcblookup_hash(&tcbinfo, 631 ip->ip_src, th->th_sport, 632 ip->ip_dst, th->th_dport, 633 0, m->m_pkthdr.rcvif); 634 if (!inp) { 635 /* It's new. Try to find the ambushing socket. */ 636 inp = in_pcblookup_hash(&tcbinfo, 637 ip->ip_src, th->th_sport, 638 next_hop->sin_addr, 639 next_hop->sin_port ? 640 ntohs(next_hop->sin_port) : 641 th->th_dport, 642 1, m->m_pkthdr.rcvif); 643 } 644 /* Remove the tag from the packet. We don't need it anymore. */ 645 m_tag_delete(m, fwd_tag); 646 } else { 647#endif /* IPFIREWALL_FORWARD */ 648 if (isipv6) { 649#ifdef INET6 650 inp = in6_pcblookup_hash(&tcbinfo, 651 &ip6->ip6_src, th->th_sport, 652 &ip6->ip6_dst, th->th_dport, 653 1, m->m_pkthdr.rcvif); 654#endif 655 } else 656 inp = in_pcblookup_hash(&tcbinfo, 657 ip->ip_src, th->th_sport, 658 ip->ip_dst, th->th_dport, 659 1, m->m_pkthdr.rcvif); 660#ifdef IPFIREWALL_FORWARD 661 } 662#endif /* IPFIREWALL_FORWARD */ 663 664#if defined(IPSEC) || defined(FAST_IPSEC) 665#ifdef INET6 666 if (isipv6) { 667 if (inp != NULL && ipsec6_in_reject(m, inp)) { 668#ifdef IPSEC 669 ipsec6stat.in_polvio++; 670#endif 671 goto drop; 672 } 673 } else 674#endif /* INET6 */ 675 if (inp != NULL && ipsec4_in_reject(m, inp)) { 676#ifdef IPSEC 677 ipsecstat.in_polvio++; 678#endif 679 goto drop; 680 } 681#endif /*IPSEC || FAST_IPSEC*/ 682 683 /* 684 * If the state is CLOSED (i.e., TCB does not exist) then 685 * all data in the incoming segment is discarded. 686 * If the TCB exists but is in CLOSED state, it is embryonic, 687 * but should either do a listen or a connect soon. 688 */ 689 if (inp == NULL) { 690 if (log_in_vain) { 691#ifdef INET6 692 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 693#else 694 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 695#endif 696 697 if (isipv6) { 698#ifdef INET6 699 strcpy(dbuf, "["); 700 strcpy(sbuf, "["); 701 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 702 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 703 strcat(dbuf, "]"); 704 strcat(sbuf, "]"); 705#endif 706 } else { 707 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 708 strcpy(sbuf, inet_ntoa(ip->ip_src)); 709 } 710 switch (log_in_vain) { 711 case 1: 712 if ((thflags & TH_SYN) == 0) 713 break; 714 /* FALLTHROUGH */ 715 case 2: 716 log(LOG_INFO, 717 "Connection attempt to TCP %s:%d " 718 "from %s:%d flags:0x%02x\n", 719 dbuf, ntohs(th->th_dport), sbuf, 720 ntohs(th->th_sport), thflags); 721 break; 722 default: 723 break; 724 } 725 } 726 if (blackhole) { 727 switch (blackhole) { 728 case 1: 729 if (thflags & TH_SYN) 730 goto drop; 731 break; 732 case 2: 733 goto drop; 734 default: 735 goto drop; 736 } 737 } 738 rstreason = BANDLIM_RST_CLOSEDPORT; 739 goto dropwithreset; 740 } 741 INP_LOCK(inp); 742 if (inp->inp_vflag & INP_TIMEWAIT) { 743 /* 744 * The only option of relevance is TOF_CC, and only if 745 * present in a SYN segment. See tcp_timewait(). 746 */ 747 if (thflags & TH_SYN) 748 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th); 749 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 750 &to, th, m, tlen)) 751 goto findpcb; 752 /* 753 * tcp_timewait unlocks inp. 754 */ 755 INP_INFO_WUNLOCK(&tcbinfo); 756 return; 757 } 758 tp = intotcpcb(inp); 759 if (tp == 0) { 760 INP_UNLOCK(inp); 761 rstreason = BANDLIM_RST_CLOSEDPORT; 762 goto dropwithreset; 763 } 764 if (tp->t_state == TCPS_CLOSED) 765 goto drop; 766 767 /* Unscale the window into a 32-bit value. */ 768 if ((thflags & TH_SYN) == 0) 769 tiwin = th->th_win << tp->snd_scale; 770 else 771 tiwin = th->th_win; 772 773#ifdef MAC 774 INP_LOCK_ASSERT(inp); 775 if (mac_check_inpcb_deliver(inp, m)) 776 goto drop; 777#endif 778 so = inp->inp_socket; 779#ifdef TCPDEBUG 780 if (so->so_options & SO_DEBUG) { 781 ostate = tp->t_state; 782 if (isipv6) 783 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 784 else 785 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 786 tcp_savetcp = *th; 787 } 788#endif 789 if (so->so_options & SO_ACCEPTCONN) { 790 struct in_conninfo inc; 791 792#ifdef INET6 793 inc.inc_isipv6 = isipv6; 794#endif 795 if (isipv6) { 796 inc.inc6_faddr = ip6->ip6_src; 797 inc.inc6_laddr = ip6->ip6_dst; 798 } else { 799 inc.inc_faddr = ip->ip_src; 800 inc.inc_laddr = ip->ip_dst; 801 } 802 inc.inc_fport = th->th_sport; 803 inc.inc_lport = th->th_dport; 804 805 /* 806 * If the state is LISTEN then ignore segment if it contains 807 * a RST. If the segment contains an ACK then it is bad and 808 * send a RST. If it does not contain a SYN then it is not 809 * interesting; drop it. 810 * 811 * If the state is SYN_RECEIVED (syncache) and seg contains 812 * an ACK, but not for our SYN/ACK, send a RST. If the seg 813 * contains a RST, check the sequence number to see if it 814 * is a valid reset segment. 815 */ 816 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 817 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 818 if (!syncache_expand(&inc, th, &so, m)) { 819 /* 820 * No syncache entry, or ACK was not 821 * for our SYN/ACK. Send a RST. 822 */ 823 tcpstat.tcps_badsyn++; 824 rstreason = BANDLIM_RST_OPENPORT; 825 goto dropwithreset; 826 } 827 if (so == NULL) { 828 /* 829 * Could not complete 3-way handshake, 830 * connection is being closed down, and 831 * syncache will free mbuf. 832 */ 833 INP_UNLOCK(inp); 834 INP_INFO_WUNLOCK(&tcbinfo); 835 return; 836 } 837 /* 838 * Socket is created in state SYN_RECEIVED. 839 * Continue processing segment. 840 */ 841 INP_UNLOCK(inp); 842 inp = sotoinpcb(so); 843 INP_LOCK(inp); 844 tp = intotcpcb(inp); 845 /* 846 * This is what would have happened in 847 * tcp_output() when the SYN,ACK was sent. 848 */ 849 tp->snd_up = tp->snd_una; 850 tp->snd_max = tp->snd_nxt = tp->iss + 1; 851 tp->last_ack_sent = tp->rcv_nxt; 852 /* 853 * RFC1323: The window in SYN & SYN/ACK 854 * segments is never scaled. 855 */ 856 tp->snd_wnd = tiwin; /* unscaled */ 857 goto after_listen; 858 } 859 if (thflags & TH_RST) { 860 syncache_chkrst(&inc, th); 861 goto drop; 862 } 863 if (thflags & TH_ACK) { 864 syncache_badack(&inc); 865 tcpstat.tcps_badsyn++; 866 rstreason = BANDLIM_RST_OPENPORT; 867 goto dropwithreset; 868 } 869 goto drop; 870 } 871 872 /* 873 * Segment's flags are (SYN) or (SYN|FIN). 874 */ 875#ifdef INET6 876 /* 877 * If deprecated address is forbidden, 878 * we do not accept SYN to deprecated interface 879 * address to prevent any new inbound connection from 880 * getting established. 881 * When we do not accept SYN, we send a TCP RST, 882 * with deprecated source address (instead of dropping 883 * it). We compromise it as it is much better for peer 884 * to send a RST, and RST will be the final packet 885 * for the exchange. 886 * 887 * If we do not forbid deprecated addresses, we accept 888 * the SYN packet. RFC2462 does not suggest dropping 889 * SYN in this case. 890 * If we decipher RFC2462 5.5.4, it says like this: 891 * 1. use of deprecated addr with existing 892 * communication is okay - "SHOULD continue to be 893 * used" 894 * 2. use of it with new communication: 895 * (2a) "SHOULD NOT be used if alternate address 896 * with sufficient scope is available" 897 * (2b) nothing mentioned otherwise. 898 * Here we fall into (2b) case as we have no choice in 899 * our source address selection - we must obey the peer. 900 * 901 * The wording in RFC2462 is confusing, and there are 902 * multiple description text for deprecated address 903 * handling - worse, they are not exactly the same. 904 * I believe 5.5.4 is the best one, so we follow 5.5.4. 905 */ 906 if (isipv6 && !ip6_use_deprecated) { 907 struct in6_ifaddr *ia6; 908 909 if ((ia6 = ip6_getdstifaddr(m)) && 910 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 911 INP_UNLOCK(inp); 912 tp = NULL; 913 rstreason = BANDLIM_RST_OPENPORT; 914 goto dropwithreset; 915 } 916 } 917#endif 918 /* 919 * If it is from this socket, drop it, it must be forged. 920 * Don't bother responding if the destination was a broadcast. 921 */ 922 if (th->th_dport == th->th_sport) { 923 if (isipv6) { 924 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 925 &ip6->ip6_src)) 926 goto drop; 927 } else { 928 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 929 goto drop; 930 } 931 } 932 /* 933 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 934 * 935 * Note that it is quite possible to receive unicast 936 * link-layer packets with a broadcast IP address. Use 937 * in_broadcast() to find them. 938 */ 939 if (m->m_flags & (M_BCAST|M_MCAST)) 940 goto drop; 941 if (isipv6) { 942 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 943 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 944 goto drop; 945 } else { 946 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 947 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 948 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 949 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 950 goto drop; 951 } 952 /* 953 * SYN appears to be valid; create compressed TCP state 954 * for syncache, or perform t/tcp connection. 955 */ 956 if (so->so_qlen <= so->so_qlimit) { 957#ifdef TCPDEBUG 958 if (so->so_options & SO_DEBUG) 959 tcp_trace(TA_INPUT, ostate, tp, 960 (void *)tcp_saveipgen, &tcp_savetcp, 0); 961#endif 962 tcp_dooptions(tp, &to, optp, optlen, 1, th); 963 if (!syncache_add(&inc, &to, th, &so, m)) 964 goto drop; 965 if (so == NULL) { 966 /* 967 * Entry added to syncache, mbuf used to 968 * send SYN,ACK packet. 969 */ 970 KASSERT(headlocked, ("headlocked")); 971 INP_UNLOCK(inp); 972 INP_INFO_WUNLOCK(&tcbinfo); 973 return; 974 } 975 /* 976 * Segment passed TAO tests. 977 */ 978 INP_UNLOCK(inp); 979 inp = sotoinpcb(so); 980 INP_LOCK(inp); 981 tp = intotcpcb(inp); 982 tp->snd_wnd = tiwin; 983 tp->t_starttime = ticks; 984 tp->t_state = TCPS_ESTABLISHED; 985 986 /* 987 * T/TCP logic: 988 * If there is a FIN or if there is data, then 989 * delay SYN,ACK(SYN) in the hope of piggy-backing 990 * it on a response segment. Otherwise must send 991 * ACK now in case the other side is slow starting. 992 */ 993 if (thflags & TH_FIN || tlen != 0) 994 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 995 else 996 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 997 tcpstat.tcps_connects++; 998 soisconnected(so); 999 goto trimthenstep6; 1000 } 1001 goto drop; 1002 } 1003after_listen: 1004 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1005 INP_LOCK_ASSERT(inp); 1006 1007 /* XXX temp debugging */ 1008 /* should not happen - syncache should pick up these connections */ 1009 if (tp->t_state == TCPS_LISTEN) 1010 panic("tcp_input: TCPS_LISTEN"); 1011 1012 /* 1013 * This is the second part of the MSS DoS prevention code (after 1014 * minmss on the sending side) and it deals with too many too small 1015 * tcp packets in a too short timeframe (1 second). 1016 * 1017 * For every full second we count the number of received packets 1018 * and bytes. If we get a lot of packets per second for this connection 1019 * (tcp_minmssoverload) we take a closer look at it and compute the 1020 * average packet size for the past second. If that is less than 1021 * tcp_minmss we get too many packets with very small payload which 1022 * is not good and burdens our system (and every packet generates 1023 * a wakeup to the process connected to our socket). We can reasonable 1024 * expect this to be small packet DoS attack to exhaust our CPU 1025 * cycles. 1026 * 1027 * Care has to be taken for the minimum packet overload value. This 1028 * value defines the minimum number of packets per second before we 1029 * start to worry. This must not be too low to avoid killing for 1030 * example interactive connections with many small packets like 1031 * telnet or SSH. 1032 * 1033 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1034 * this check. 1035 * 1036 * Account for packet if payload packet, skip over ACK, etc. 1037 */ 1038 if (tcp_minmss && tcp_minmssoverload && 1039 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1040 if (tp->rcv_second > ticks) { 1041 tp->rcv_pps++; 1042 tp->rcv_byps += tlen + off; 1043 if (tp->rcv_pps > tcp_minmssoverload) { 1044 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1045 printf("too many small tcp packets from " 1046 "%s:%u, av. %lubyte/packet, " 1047 "dropping connection\n", 1048#ifdef INET6 1049 isipv6 ? 1050 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1051#endif 1052 inet_ntoa(inp->inp_inc.inc_faddr), 1053 inp->inp_inc.inc_fport, 1054 tp->rcv_byps / tp->rcv_pps); 1055 tp = tcp_drop(tp, ECONNRESET); 1056 tcpstat.tcps_minmssdrops++; 1057 goto drop; 1058 } 1059 } 1060 } else { 1061 tp->rcv_second = ticks + hz; 1062 tp->rcv_pps = 1; 1063 tp->rcv_byps = tlen + off; 1064 } 1065 } 1066 1067 /* 1068 * Segment received on connection. 1069 * Reset idle time and keep-alive timer. 1070 */ 1071 tp->t_rcvtime = ticks; 1072 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1073 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1074 1075 /* 1076 * Process options only when we get SYN/ACK back. The SYN case 1077 * for incoming connections is handled in tcp_syncache. 1078 * XXX this is traditional behavior, may need to be cleaned up. 1079 */ 1080 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th); 1081 if (thflags & TH_SYN) { 1082 if (to.to_flags & TOF_SCALE) { 1083 tp->t_flags |= TF_RCVD_SCALE; 1084 tp->requested_s_scale = to.to_requested_s_scale; 1085 } 1086 if (to.to_flags & TOF_TS) { 1087 tp->t_flags |= TF_RCVD_TSTMP; 1088 tp->ts_recent = to.to_tsval; 1089 tp->ts_recent_age = ticks; 1090 } 1091 if (to.to_flags & TOF_MSS) 1092 tcp_mss(tp, to.to_mss); 1093 if (tp->sack_enable) { 1094 if (!(to.to_flags & TOF_SACK)) 1095 tp->sack_enable = 0; 1096 else 1097 tp->t_flags |= TF_SACK_PERMIT; 1098 } 1099 1100 } 1101 1102 if (tp->sack_enable) { 1103 /* Delete stale (cumulatively acked) SACK holes */ 1104 tcp_del_sackholes(tp, th); 1105 } 1106 1107 /* 1108 * Header prediction: check for the two common cases 1109 * of a uni-directional data xfer. If the packet has 1110 * no control flags, is in-sequence, the window didn't 1111 * change and we're not retransmitting, it's a 1112 * candidate. If the length is zero and the ack moved 1113 * forward, we're the sender side of the xfer. Just 1114 * free the data acked & wake any higher level process 1115 * that was blocked waiting for space. If the length 1116 * is non-zero and the ack didn't move, we're the 1117 * receiver side. If we're getting packets in-order 1118 * (the reassembly queue is empty), add the data to 1119 * the socket buffer and note that we need a delayed ack. 1120 * Make sure that the hidden state-flags are also off. 1121 * Since we check for TCPS_ESTABLISHED above, it can only 1122 * be TH_NEEDSYN. 1123 */ 1124 if (tp->t_state == TCPS_ESTABLISHED && 1125 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1126 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1127 ((to.to_flags & TOF_TS) == 0 || 1128 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1129 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1130 tp->snd_nxt == tp->snd_max) { 1131 1132 /* 1133 * If last ACK falls within this segment's sequence numbers, 1134 * record the timestamp. 1135 * NOTE that the test is modified according to the latest 1136 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1137 */ 1138 if ((to.to_flags & TOF_TS) != 0 && 1139 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1140 tp->ts_recent_age = ticks; 1141 tp->ts_recent = to.to_tsval; 1142 } 1143 1144 if (tlen == 0) { 1145 if (SEQ_GT(th->th_ack, tp->snd_una) && 1146 SEQ_LEQ(th->th_ack, tp->snd_max) && 1147 tp->snd_cwnd >= tp->snd_wnd && 1148 ((!tcp_do_newreno && !tp->sack_enable && 1149 tp->t_dupacks < tcprexmtthresh) || 1150 ((tcp_do_newreno || tp->sack_enable) && 1151 !IN_FASTRECOVERY(tp)))) { 1152 KASSERT(headlocked, ("headlocked")); 1153 INP_INFO_WUNLOCK(&tcbinfo); 1154 headlocked = 0; 1155 /* 1156 * this is a pure ack for outstanding data. 1157 */ 1158 ++tcpstat.tcps_predack; 1159 /* 1160 * "bad retransmit" recovery 1161 */ 1162 if (tp->t_rxtshift == 1 && 1163 ticks < tp->t_badrxtwin) { 1164 ++tcpstat.tcps_sndrexmitbad; 1165 tp->snd_cwnd = tp->snd_cwnd_prev; 1166 tp->snd_ssthresh = 1167 tp->snd_ssthresh_prev; 1168 tp->snd_recover = tp->snd_recover_prev; 1169 if (tp->t_flags & TF_WASFRECOVERY) 1170 ENTER_FASTRECOVERY(tp); 1171 tp->snd_nxt = tp->snd_max; 1172 tp->t_badrxtwin = 0; 1173 } 1174 1175 /* 1176 * Recalculate the transmit timer / rtt. 1177 * 1178 * Some boxes send broken timestamp replies 1179 * during the SYN+ACK phase, ignore 1180 * timestamps of 0 or we could calculate a 1181 * huge RTT and blow up the retransmit timer. 1182 */ 1183 if ((to.to_flags & TOF_TS) != 0 && 1184 to.to_tsecr) { 1185 tcp_xmit_timer(tp, 1186 ticks - to.to_tsecr + 1); 1187 } else if (tp->t_rtttime && 1188 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1189 tcp_xmit_timer(tp, 1190 ticks - tp->t_rtttime); 1191 } 1192 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1193 acked = th->th_ack - tp->snd_una; 1194 tcpstat.tcps_rcvackpack++; 1195 tcpstat.tcps_rcvackbyte += acked; 1196 sbdrop(&so->so_snd, acked); 1197 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1198 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1199 tp->snd_recover = th->th_ack - 1; 1200 tp->snd_una = th->th_ack; 1201 /* 1202 * pull snd_wl2 up to prevent seq wrap relative 1203 * to th_ack. 1204 */ 1205 tp->snd_wl2 = th->th_ack; 1206 tp->t_dupacks = 0; 1207 m_freem(m); 1208 ND6_HINT(tp); /* some progress has been done */ 1209 1210 /* 1211 * If all outstanding data are acked, stop 1212 * retransmit timer, otherwise restart timer 1213 * using current (possibly backed-off) value. 1214 * If process is waiting for space, 1215 * wakeup/selwakeup/signal. If data 1216 * are ready to send, let tcp_output 1217 * decide between more output or persist. 1218 1219#ifdef TCPDEBUG 1220 if (so->so_options & SO_DEBUG) 1221 tcp_trace(TA_INPUT, ostate, tp, 1222 (void *)tcp_saveipgen, 1223 &tcp_savetcp, 0); 1224#endif 1225 */ 1226 if (tp->snd_una == tp->snd_max) 1227 callout_stop(tp->tt_rexmt); 1228 else if (!callout_active(tp->tt_persist)) 1229 callout_reset(tp->tt_rexmt, 1230 tp->t_rxtcur, 1231 tcp_timer_rexmt, tp); 1232 1233 sowwakeup(so); 1234 if (so->so_snd.sb_cc) 1235 (void) tcp_output(tp); 1236 goto check_delack; 1237 } 1238 } else if (th->th_ack == tp->snd_una && 1239 LIST_EMPTY(&tp->t_segq) && 1240 tlen <= sbspace(&so->so_rcv)) { 1241 KASSERT(headlocked, ("headlocked")); 1242 INP_INFO_WUNLOCK(&tcbinfo); 1243 headlocked = 0; 1244 /* 1245 * this is a pure, in-sequence data packet 1246 * with nothing on the reassembly queue and 1247 * we have enough buffer space to take it. 1248 */ 1249 /* Clean receiver SACK report if present */ 1250 if (tp->sack_enable && tp->rcv_numsacks) 1251 tcp_clean_sackreport(tp); 1252 ++tcpstat.tcps_preddat; 1253 tp->rcv_nxt += tlen; 1254 /* 1255 * Pull snd_wl1 up to prevent seq wrap relative to 1256 * th_seq. 1257 */ 1258 tp->snd_wl1 = th->th_seq; 1259 /* 1260 * Pull rcv_up up to prevent seq wrap relative to 1261 * rcv_nxt. 1262 */ 1263 tp->rcv_up = tp->rcv_nxt; 1264 tcpstat.tcps_rcvpack++; 1265 tcpstat.tcps_rcvbyte += tlen; 1266 ND6_HINT(tp); /* some progress has been done */ 1267 /* 1268#ifdef TCPDEBUG 1269 if (so->so_options & SO_DEBUG) 1270 tcp_trace(TA_INPUT, ostate, tp, 1271 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1272#endif 1273 * Add data to socket buffer. 1274 */ 1275 SOCKBUF_LOCK(&so->so_rcv); 1276 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1277 m_freem(m); 1278 } else { 1279 m_adj(m, drop_hdrlen); /* delayed header drop */ 1280 sbappendstream_locked(&so->so_rcv, m); 1281 } 1282 sorwakeup_locked(so); 1283 if (DELAY_ACK(tp)) { 1284 tp->t_flags |= TF_DELACK; 1285 } else { 1286 tp->t_flags |= TF_ACKNOW; 1287 tcp_output(tp); 1288 } 1289 goto check_delack; 1290 } 1291 } 1292 1293 /* 1294 * Calculate amount of space in receive window, 1295 * and then do TCP input processing. 1296 * Receive window is amount of space in rcv queue, 1297 * but not less than advertised window. 1298 */ 1299 { int win; 1300 1301 win = sbspace(&so->so_rcv); 1302 if (win < 0) 1303 win = 0; 1304 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1305 } 1306 1307 switch (tp->t_state) { 1308 1309 /* 1310 * If the state is SYN_RECEIVED: 1311 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1312 */ 1313 case TCPS_SYN_RECEIVED: 1314 if ((thflags & TH_ACK) && 1315 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1316 SEQ_GT(th->th_ack, tp->snd_max))) { 1317 rstreason = BANDLIM_RST_OPENPORT; 1318 goto dropwithreset; 1319 } 1320 break; 1321 1322 /* 1323 * If the state is SYN_SENT: 1324 * if seg contains an ACK, but not for our SYN, drop the input. 1325 * if seg contains a RST, then drop the connection. 1326 * if seg does not contain SYN, then drop it. 1327 * Otherwise this is an acceptable SYN segment 1328 * initialize tp->rcv_nxt and tp->irs 1329 * if seg contains ack then advance tp->snd_una 1330 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1331 * arrange for segment to be acked (eventually) 1332 * continue processing rest of data/controls, beginning with URG 1333 */ 1334 case TCPS_SYN_SENT: 1335 if ((thflags & TH_ACK) && 1336 (SEQ_LEQ(th->th_ack, tp->iss) || 1337 SEQ_GT(th->th_ack, tp->snd_max))) { 1338 rstreason = BANDLIM_UNLIMITED; 1339 goto dropwithreset; 1340 } 1341 if (thflags & TH_RST) { 1342 if (thflags & TH_ACK) 1343 tp = tcp_drop(tp, ECONNREFUSED); 1344 goto drop; 1345 } 1346 if ((thflags & TH_SYN) == 0) 1347 goto drop; 1348 tp->snd_wnd = th->th_win; /* initial send window */ 1349 1350 tp->irs = th->th_seq; 1351 tcp_rcvseqinit(tp); 1352 if (thflags & TH_ACK) { 1353 tcpstat.tcps_connects++; 1354 soisconnected(so); 1355#ifdef MAC 1356 SOCK_LOCK(so); 1357 mac_set_socket_peer_from_mbuf(m, so); 1358 SOCK_UNLOCK(so); 1359#endif 1360 /* Do window scaling on this connection? */ 1361 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1362 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1363 tp->snd_scale = tp->requested_s_scale; 1364 tp->rcv_scale = tp->request_r_scale; 1365 } 1366 tp->rcv_adv += tp->rcv_wnd; 1367 tp->snd_una++; /* SYN is acked */ 1368 /* 1369 * If there's data, delay ACK; if there's also a FIN 1370 * ACKNOW will be turned on later. 1371 */ 1372 if (DELAY_ACK(tp) && tlen != 0) 1373 callout_reset(tp->tt_delack, tcp_delacktime, 1374 tcp_timer_delack, tp); 1375 else 1376 tp->t_flags |= TF_ACKNOW; 1377 /* 1378 * Received <SYN,ACK> in SYN_SENT[*] state. 1379 * Transitions: 1380 * SYN_SENT --> ESTABLISHED 1381 * SYN_SENT* --> FIN_WAIT_1 1382 */ 1383 tp->t_starttime = ticks; 1384 if (tp->t_flags & TF_NEEDFIN) { 1385 tp->t_state = TCPS_FIN_WAIT_1; 1386 tp->t_flags &= ~TF_NEEDFIN; 1387 thflags &= ~TH_SYN; 1388 } else { 1389 tp->t_state = TCPS_ESTABLISHED; 1390 callout_reset(tp->tt_keep, tcp_keepidle, 1391 tcp_timer_keep, tp); 1392 } 1393 } else { 1394 /* 1395 * Received initial SYN in SYN-SENT[*] state => 1396 * simultaneous open. If segment contains CC option 1397 * and there is a cached CC, apply TAO test. 1398 * If it succeeds, connection is * half-synchronized. 1399 * Otherwise, do 3-way handshake: 1400 * SYN-SENT -> SYN-RECEIVED 1401 * SYN-SENT* -> SYN-RECEIVED* 1402 * If there was no CC option, clear cached CC value. 1403 */ 1404 tp->t_flags |= TF_ACKNOW; 1405 callout_stop(tp->tt_rexmt); 1406 tp->t_state = TCPS_SYN_RECEIVED; 1407 } 1408 1409trimthenstep6: 1410 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1411 "locked")); 1412 INP_LOCK_ASSERT(inp); 1413 1414 /* 1415 * Advance th->th_seq to correspond to first data byte. 1416 * If data, trim to stay within window, 1417 * dropping FIN if necessary. 1418 */ 1419 th->th_seq++; 1420 if (tlen > tp->rcv_wnd) { 1421 todrop = tlen - tp->rcv_wnd; 1422 m_adj(m, -todrop); 1423 tlen = tp->rcv_wnd; 1424 thflags &= ~TH_FIN; 1425 tcpstat.tcps_rcvpackafterwin++; 1426 tcpstat.tcps_rcvbyteafterwin += todrop; 1427 } 1428 tp->snd_wl1 = th->th_seq - 1; 1429 tp->rcv_up = th->th_seq; 1430 /* 1431 * Client side of transaction: already sent SYN and data. 1432 * If the remote host used T/TCP to validate the SYN, 1433 * our data will be ACK'd; if so, enter normal data segment 1434 * processing in the middle of step 5, ack processing. 1435 * Otherwise, goto step 6. 1436 */ 1437 if (thflags & TH_ACK) 1438 goto process_ACK; 1439 1440 goto step6; 1441 1442 /* 1443 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1444 * do normal processing. 1445 * 1446 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1447 */ 1448 case TCPS_LAST_ACK: 1449 case TCPS_CLOSING: 1450 case TCPS_TIME_WAIT: 1451 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1452 break; /* continue normal processing */ 1453 } 1454 1455 /* 1456 * States other than LISTEN or SYN_SENT. 1457 * First check the RST flag and sequence number since reset segments 1458 * are exempt from the timestamp and connection count tests. This 1459 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1460 * below which allowed reset segments in half the sequence space 1461 * to fall though and be processed (which gives forged reset 1462 * segments with a random sequence number a 50 percent chance of 1463 * killing a connection). 1464 * Then check timestamp, if present. 1465 * Then check the connection count, if present. 1466 * Then check that at least some bytes of segment are within 1467 * receive window. If segment begins before rcv_nxt, 1468 * drop leading data (and SYN); if nothing left, just ack. 1469 * 1470 * 1471 * If the RST bit is set, check the sequence number to see 1472 * if this is a valid reset segment. 1473 * RFC 793 page 37: 1474 * In all states except SYN-SENT, all reset (RST) segments 1475 * are validated by checking their SEQ-fields. A reset is 1476 * valid if its sequence number is in the window. 1477 * Note: this does not take into account delayed ACKs, so 1478 * we should test against last_ack_sent instead of rcv_nxt. 1479 * The sequence number in the reset segment is normally an 1480 * echo of our outgoing acknowlegement numbers, but some hosts 1481 * send a reset with the sequence number at the rightmost edge 1482 * of our receive window, and we have to handle this case. 1483 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1484 * that brute force RST attacks are possible. To combat this, 1485 * we use a much stricter check while in the ESTABLISHED state, 1486 * only accepting RSTs where the sequence number is equal to 1487 * last_ack_sent. In all other states (the states in which a 1488 * RST is more likely), the more permissive check is used. 1489 * If we have multiple segments in flight, the intial reset 1490 * segment sequence numbers will be to the left of last_ack_sent, 1491 * but they will eventually catch up. 1492 * In any case, it never made sense to trim reset segments to 1493 * fit the receive window since RFC 1122 says: 1494 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1495 * 1496 * A TCP SHOULD allow a received RST segment to include data. 1497 * 1498 * DISCUSSION 1499 * It has been suggested that a RST segment could contain 1500 * ASCII text that encoded and explained the cause of the 1501 * RST. No standard has yet been established for such 1502 * data. 1503 * 1504 * If the reset segment passes the sequence number test examine 1505 * the state: 1506 * SYN_RECEIVED STATE: 1507 * If passive open, return to LISTEN state. 1508 * If active open, inform user that connection was refused. 1509 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1510 * Inform user that connection was reset, and close tcb. 1511 * CLOSING, LAST_ACK STATES: 1512 * Close the tcb. 1513 * TIME_WAIT STATE: 1514 * Drop the segment - see Stevens, vol. 2, p. 964 and 1515 * RFC 1337. 1516 */ 1517 if (thflags & TH_RST) { 1518 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1519 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1520 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1521 switch (tp->t_state) { 1522 1523 case TCPS_SYN_RECEIVED: 1524 so->so_error = ECONNREFUSED; 1525 goto close; 1526 1527 case TCPS_ESTABLISHED: 1528 if (tp->last_ack_sent != th->th_seq && 1529 tcp_insecure_rst == 0) { 1530 tcpstat.tcps_badrst++; 1531 goto drop; 1532 } 1533 case TCPS_FIN_WAIT_1: 1534 case TCPS_FIN_WAIT_2: 1535 case TCPS_CLOSE_WAIT: 1536 so->so_error = ECONNRESET; 1537 close: 1538 tp->t_state = TCPS_CLOSED; 1539 tcpstat.tcps_drops++; 1540 tp = tcp_close(tp); 1541 break; 1542 1543 case TCPS_CLOSING: 1544 case TCPS_LAST_ACK: 1545 tp = tcp_close(tp); 1546 break; 1547 1548 case TCPS_TIME_WAIT: 1549 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1550 ("timewait")); 1551 break; 1552 } 1553 } 1554 goto drop; 1555 } 1556 1557 /* 1558 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1559 * and it's less than ts_recent, drop it. 1560 */ 1561 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1562 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1563 1564 /* Check to see if ts_recent is over 24 days old. */ 1565 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1566 /* 1567 * Invalidate ts_recent. If this segment updates 1568 * ts_recent, the age will be reset later and ts_recent 1569 * will get a valid value. If it does not, setting 1570 * ts_recent to zero will at least satisfy the 1571 * requirement that zero be placed in the timestamp 1572 * echo reply when ts_recent isn't valid. The 1573 * age isn't reset until we get a valid ts_recent 1574 * because we don't want out-of-order segments to be 1575 * dropped when ts_recent is old. 1576 */ 1577 tp->ts_recent = 0; 1578 } else { 1579 tcpstat.tcps_rcvduppack++; 1580 tcpstat.tcps_rcvdupbyte += tlen; 1581 tcpstat.tcps_pawsdrop++; 1582 if (tlen) 1583 goto dropafterack; 1584 goto drop; 1585 } 1586 } 1587 1588 /* 1589 * In the SYN-RECEIVED state, validate that the packet belongs to 1590 * this connection before trimming the data to fit the receive 1591 * window. Check the sequence number versus IRS since we know 1592 * the sequence numbers haven't wrapped. This is a partial fix 1593 * for the "LAND" DoS attack. 1594 */ 1595 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1596 rstreason = BANDLIM_RST_OPENPORT; 1597 goto dropwithreset; 1598 } 1599 1600 todrop = tp->rcv_nxt - th->th_seq; 1601 if (todrop > 0) { 1602 if (thflags & TH_SYN) { 1603 thflags &= ~TH_SYN; 1604 th->th_seq++; 1605 if (th->th_urp > 1) 1606 th->th_urp--; 1607 else 1608 thflags &= ~TH_URG; 1609 todrop--; 1610 } 1611 /* 1612 * Following if statement from Stevens, vol. 2, p. 960. 1613 */ 1614 if (todrop > tlen 1615 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1616 /* 1617 * Any valid FIN must be to the left of the window. 1618 * At this point the FIN must be a duplicate or out 1619 * of sequence; drop it. 1620 */ 1621 thflags &= ~TH_FIN; 1622 1623 /* 1624 * Send an ACK to resynchronize and drop any data. 1625 * But keep on processing for RST or ACK. 1626 */ 1627 tp->t_flags |= TF_ACKNOW; 1628 todrop = tlen; 1629 tcpstat.tcps_rcvduppack++; 1630 tcpstat.tcps_rcvdupbyte += todrop; 1631 } else { 1632 tcpstat.tcps_rcvpartduppack++; 1633 tcpstat.tcps_rcvpartdupbyte += todrop; 1634 } 1635 drop_hdrlen += todrop; /* drop from the top afterwards */ 1636 th->th_seq += todrop; 1637 tlen -= todrop; 1638 if (th->th_urp > todrop) 1639 th->th_urp -= todrop; 1640 else { 1641 thflags &= ~TH_URG; 1642 th->th_urp = 0; 1643 } 1644 } 1645 1646 /* 1647 * If new data are received on a connection after the 1648 * user processes are gone, then RST the other end. 1649 */ 1650 if ((so->so_state & SS_NOFDREF) && 1651 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1652 tp = tcp_close(tp); 1653 tcpstat.tcps_rcvafterclose++; 1654 rstreason = BANDLIM_UNLIMITED; 1655 goto dropwithreset; 1656 } 1657 1658 /* 1659 * If segment ends after window, drop trailing data 1660 * (and PUSH and FIN); if nothing left, just ACK. 1661 */ 1662 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1663 if (todrop > 0) { 1664 tcpstat.tcps_rcvpackafterwin++; 1665 if (todrop >= tlen) { 1666 tcpstat.tcps_rcvbyteafterwin += tlen; 1667 /* 1668 * If a new connection request is received 1669 * while in TIME_WAIT, drop the old connection 1670 * and start over if the sequence numbers 1671 * are above the previous ones. 1672 */ 1673 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1674 if (thflags & TH_SYN && 1675 tp->t_state == TCPS_TIME_WAIT && 1676 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1677 tp = tcp_close(tp); 1678 goto findpcb; 1679 } 1680 /* 1681 * If window is closed can only take segments at 1682 * window edge, and have to drop data and PUSH from 1683 * incoming segments. Continue processing, but 1684 * remember to ack. Otherwise, drop segment 1685 * and ack. 1686 */ 1687 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1688 tp->t_flags |= TF_ACKNOW; 1689 tcpstat.tcps_rcvwinprobe++; 1690 } else 1691 goto dropafterack; 1692 } else 1693 tcpstat.tcps_rcvbyteafterwin += todrop; 1694 m_adj(m, -todrop); 1695 tlen -= todrop; 1696 thflags &= ~(TH_PUSH|TH_FIN); 1697 } 1698 1699 /* 1700 * If last ACK falls within this segment's sequence numbers, 1701 * record its timestamp. 1702 * NOTE that the test is modified according to the latest 1703 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1704 */ 1705 if ((to.to_flags & TOF_TS) != 0 && 1706 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1707 tp->ts_recent_age = ticks; 1708 tp->ts_recent = to.to_tsval; 1709 } 1710 1711 /* 1712 * If a SYN is in the window, then this is an 1713 * error and we send an RST and drop the connection. 1714 */ 1715 if (thflags & TH_SYN) { 1716 tp = tcp_drop(tp, ECONNRESET); 1717 rstreason = BANDLIM_UNLIMITED; 1718 goto drop; 1719 } 1720 1721 /* 1722 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1723 * flag is on (half-synchronized state), then queue data for 1724 * later processing; else drop segment and return. 1725 */ 1726 if ((thflags & TH_ACK) == 0) { 1727 if (tp->t_state == TCPS_SYN_RECEIVED || 1728 (tp->t_flags & TF_NEEDSYN)) 1729 goto step6; 1730 else 1731 goto drop; 1732 } 1733 1734 /* 1735 * Ack processing. 1736 */ 1737 switch (tp->t_state) { 1738 1739 /* 1740 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1741 * ESTABLISHED state and continue processing. 1742 * The ACK was checked above. 1743 */ 1744 case TCPS_SYN_RECEIVED: 1745 1746 tcpstat.tcps_connects++; 1747 soisconnected(so); 1748 /* Do window scaling? */ 1749 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1750 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1751 tp->snd_scale = tp->requested_s_scale; 1752 tp->rcv_scale = tp->request_r_scale; 1753 } 1754 /* 1755 * Make transitions: 1756 * SYN-RECEIVED -> ESTABLISHED 1757 * SYN-RECEIVED* -> FIN-WAIT-1 1758 */ 1759 tp->t_starttime = ticks; 1760 if (tp->t_flags & TF_NEEDFIN) { 1761 tp->t_state = TCPS_FIN_WAIT_1; 1762 tp->t_flags &= ~TF_NEEDFIN; 1763 } else { 1764 tp->t_state = TCPS_ESTABLISHED; 1765 callout_reset(tp->tt_keep, tcp_keepidle, 1766 tcp_timer_keep, tp); 1767 } 1768 /* 1769 * If segment contains data or ACK, will call tcp_reass() 1770 * later; if not, do so now to pass queued data to user. 1771 */ 1772 if (tlen == 0 && (thflags & TH_FIN) == 0) 1773 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1774 (struct mbuf *)0); 1775 tp->snd_wl1 = th->th_seq - 1; 1776 /* FALLTHROUGH */ 1777 1778 /* 1779 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1780 * ACKs. If the ack is in the range 1781 * tp->snd_una < th->th_ack <= tp->snd_max 1782 * then advance tp->snd_una to th->th_ack and drop 1783 * data from the retransmission queue. If this ACK reflects 1784 * more up to date window information we update our window information. 1785 */ 1786 case TCPS_ESTABLISHED: 1787 case TCPS_FIN_WAIT_1: 1788 case TCPS_FIN_WAIT_2: 1789 case TCPS_CLOSE_WAIT: 1790 case TCPS_CLOSING: 1791 case TCPS_LAST_ACK: 1792 case TCPS_TIME_WAIT: 1793 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1794 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1795 if (tlen == 0 && tiwin == tp->snd_wnd) { 1796 tcpstat.tcps_rcvdupack++; 1797 /* 1798 * If we have outstanding data (other than 1799 * a window probe), this is a completely 1800 * duplicate ack (ie, window info didn't 1801 * change), the ack is the biggest we've 1802 * seen and we've seen exactly our rexmt 1803 * threshhold of them, assume a packet 1804 * has been dropped and retransmit it. 1805 * Kludge snd_nxt & the congestion 1806 * window so we send only this one 1807 * packet. 1808 * 1809 * We know we're losing at the current 1810 * window size so do congestion avoidance 1811 * (set ssthresh to half the current window 1812 * and pull our congestion window back to 1813 * the new ssthresh). 1814 * 1815 * Dup acks mean that packets have left the 1816 * network (they're now cached at the receiver) 1817 * so bump cwnd by the amount in the receiver 1818 * to keep a constant cwnd packets in the 1819 * network. 1820 */ 1821 if (!callout_active(tp->tt_rexmt) || 1822 th->th_ack != tp->snd_una) 1823 tp->t_dupacks = 0; 1824 else if (++tp->t_dupacks > tcprexmtthresh || 1825 ((tcp_do_newreno || tp->sack_enable) && 1826 IN_FASTRECOVERY(tp))) { 1827 tp->snd_cwnd += tp->t_maxseg; 1828 (void) tcp_output(tp); 1829 goto drop; 1830 } else if (tp->t_dupacks == tcprexmtthresh) { 1831 tcp_seq onxt = tp->snd_nxt; 1832 u_int win; 1833 1834 /* 1835 * If we're doing sack, check to 1836 * see if we're already in sack 1837 * recovery. If we're not doing sack, 1838 * check to see if we're in newreno 1839 * recovery. 1840 */ 1841 if (tp->sack_enable) { 1842 if (IN_FASTRECOVERY(tp)) { 1843 tp->t_dupacks = 0; 1844 break; 1845 } 1846 } else if (tcp_do_newreno) { 1847 if (SEQ_LEQ(th->th_ack, 1848 tp->snd_recover)) { 1849 tp->t_dupacks = 0; 1850 break; 1851 } 1852 } 1853 win = min(tp->snd_wnd, tp->snd_cwnd) / 1854 2 / tp->t_maxseg; 1855 if (win < 2) 1856 win = 2; 1857 tp->snd_ssthresh = win * tp->t_maxseg; 1858 ENTER_FASTRECOVERY(tp); 1859 tp->snd_recover = tp->snd_max; 1860 callout_stop(tp->tt_rexmt); 1861 tp->t_rtttime = 0; 1862 if (tp->sack_enable) { 1863 tcpstat.tcps_sack_recovery_episode++; 1864 tp->sack_newdata = tp->snd_nxt; 1865 tp->snd_cwnd = tp->t_maxseg; 1866 (void) tcp_output(tp); 1867 goto drop; 1868 } 1869 tp->snd_nxt = th->th_ack; 1870 tp->snd_cwnd = tp->t_maxseg; 1871 (void) tcp_output(tp); 1872 KASSERT(tp->snd_limited <= 2, 1873 ("tp->snd_limited too big")); 1874 tp->snd_cwnd = tp->snd_ssthresh + 1875 tp->t_maxseg * 1876 (tp->t_dupacks - tp->snd_limited); 1877 if (SEQ_GT(onxt, tp->snd_nxt)) 1878 tp->snd_nxt = onxt; 1879 goto drop; 1880 } else if (tcp_do_rfc3042) { 1881 u_long oldcwnd = tp->snd_cwnd; 1882 tcp_seq oldsndmax = tp->snd_max; 1883 u_int sent; 1884 1885 KASSERT(tp->t_dupacks == 1 || 1886 tp->t_dupacks == 2, 1887 ("dupacks not 1 or 2")); 1888 if (tp->t_dupacks == 1) 1889 tp->snd_limited = 0; 1890 tp->snd_cwnd = 1891 (tp->snd_nxt - tp->snd_una) + 1892 (tp->t_dupacks - tp->snd_limited) * 1893 tp->t_maxseg; 1894 (void) tcp_output(tp); 1895 sent = tp->snd_max - oldsndmax; 1896 if (sent > tp->t_maxseg) { 1897 KASSERT((tp->t_dupacks == 2 && 1898 tp->snd_limited == 0) || 1899 (sent == tp->t_maxseg + 1 && 1900 tp->t_flags & TF_SENTFIN), 1901 ("sent too much")); 1902 tp->snd_limited = 2; 1903 } else if (sent > 0) 1904 ++tp->snd_limited; 1905 tp->snd_cwnd = oldcwnd; 1906 goto drop; 1907 } 1908 } else 1909 tp->t_dupacks = 0; 1910 break; 1911 } 1912 1913 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1914 1915 /* 1916 * If the congestion window was inflated to account 1917 * for the other side's cached packets, retract it. 1918 */ 1919 if (tcp_do_newreno || tp->sack_enable) { 1920 if (IN_FASTRECOVERY(tp)) { 1921 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1922 if (tp->sack_enable) 1923 tcp_sack_partialack(tp, th); 1924 else 1925 tcp_newreno_partial_ack(tp, th); 1926 } else { 1927 /* 1928 * Out of fast recovery. 1929 * Window inflation should have left us 1930 * with approximately snd_ssthresh 1931 * outstanding data. 1932 * But in case we would be inclined to 1933 * send a burst, better to do it via 1934 * the slow start mechanism. 1935 */ 1936 if (SEQ_GT(th->th_ack + 1937 tp->snd_ssthresh, 1938 tp->snd_max)) 1939 tp->snd_cwnd = tp->snd_max - 1940 th->th_ack + 1941 tp->t_maxseg; 1942 else 1943 tp->snd_cwnd = tp->snd_ssthresh; 1944 } 1945 } 1946 } else { 1947 if (tp->t_dupacks >= tcprexmtthresh && 1948 tp->snd_cwnd > tp->snd_ssthresh) 1949 tp->snd_cwnd = tp->snd_ssthresh; 1950 } 1951 tp->t_dupacks = 0; 1952 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1953 tcpstat.tcps_rcvacktoomuch++; 1954 goto dropafterack; 1955 } 1956 /* 1957 * If we reach this point, ACK is not a duplicate, 1958 * i.e., it ACKs something we sent. 1959 */ 1960 if (tp->t_flags & TF_NEEDSYN) { 1961 /* 1962 * T/TCP: Connection was half-synchronized, and our 1963 * SYN has been ACK'd (so connection is now fully 1964 * synchronized). Go to non-starred state, 1965 * increment snd_una for ACK of SYN, and check if 1966 * we can do window scaling. 1967 */ 1968 tp->t_flags &= ~TF_NEEDSYN; 1969 tp->snd_una++; 1970 /* Do window scaling? */ 1971 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1972 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1973 tp->snd_scale = tp->requested_s_scale; 1974 tp->rcv_scale = tp->request_r_scale; 1975 } 1976 } 1977 1978process_ACK: 1979 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 1980 "locked")); 1981 INP_LOCK_ASSERT(inp); 1982 1983 acked = th->th_ack - tp->snd_una; 1984 tcpstat.tcps_rcvackpack++; 1985 tcpstat.tcps_rcvackbyte += acked; 1986 1987 /* 1988 * If we just performed our first retransmit, and the ACK 1989 * arrives within our recovery window, then it was a mistake 1990 * to do the retransmit in the first place. Recover our 1991 * original cwnd and ssthresh, and proceed to transmit where 1992 * we left off. 1993 */ 1994 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1995 ++tcpstat.tcps_sndrexmitbad; 1996 tp->snd_cwnd = tp->snd_cwnd_prev; 1997 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1998 tp->snd_recover = tp->snd_recover_prev; 1999 if (tp->t_flags & TF_WASFRECOVERY) 2000 ENTER_FASTRECOVERY(tp); 2001 tp->snd_nxt = tp->snd_max; 2002 tp->t_badrxtwin = 0; /* XXX probably not required */ 2003 } 2004 2005 /* 2006 * If we have a timestamp reply, update smoothed 2007 * round trip time. If no timestamp is present but 2008 * transmit timer is running and timed sequence 2009 * number was acked, update smoothed round trip time. 2010 * Since we now have an rtt measurement, cancel the 2011 * timer backoff (cf., Phil Karn's retransmit alg.). 2012 * Recompute the initial retransmit timer. 2013 * 2014 * Some boxes send broken timestamp replies 2015 * during the SYN+ACK phase, ignore 2016 * timestamps of 0 or we could calculate a 2017 * huge RTT and blow up the retransmit timer. 2018 */ 2019 if ((to.to_flags & TOF_TS) != 0 && 2020 to.to_tsecr) { 2021 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2022 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2023 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2024 } 2025 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2026 2027 /* 2028 * If all outstanding data is acked, stop retransmit 2029 * timer and remember to restart (more output or persist). 2030 * If there is more data to be acked, restart retransmit 2031 * timer, using current (possibly backed-off) value. 2032 */ 2033 if (th->th_ack == tp->snd_max) { 2034 callout_stop(tp->tt_rexmt); 2035 needoutput = 1; 2036 } else if (!callout_active(tp->tt_persist)) 2037 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2038 tcp_timer_rexmt, tp); 2039 2040 /* 2041 * If no data (only SYN) was ACK'd, 2042 * skip rest of ACK processing. 2043 */ 2044 if (acked == 0) 2045 goto step6; 2046 2047 /* 2048 * When new data is acked, open the congestion window. 2049 * If the window gives us less than ssthresh packets 2050 * in flight, open exponentially (maxseg per packet). 2051 * Otherwise open linearly: maxseg per window 2052 * (maxseg^2 / cwnd per packet). 2053 */ 2054 if ((!tcp_do_newreno && !tp->sack_enable) || 2055 !IN_FASTRECOVERY(tp)) { 2056 register u_int cw = tp->snd_cwnd; 2057 register u_int incr = tp->t_maxseg; 2058 if (cw > tp->snd_ssthresh) 2059 incr = incr * incr / cw; 2060 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2061 } 2062 SOCKBUF_LOCK(&so->so_snd); 2063 if (acked > so->so_snd.sb_cc) { 2064 tp->snd_wnd -= so->so_snd.sb_cc; 2065 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2066 ourfinisacked = 1; 2067 } else { 2068 sbdrop_locked(&so->so_snd, acked); 2069 tp->snd_wnd -= acked; 2070 ourfinisacked = 0; 2071 } 2072 sowwakeup_locked(so); 2073 /* detect una wraparound */ 2074 if ((tcp_do_newreno || tp->sack_enable) && 2075 !IN_FASTRECOVERY(tp) && 2076 SEQ_GT(tp->snd_una, tp->snd_recover) && 2077 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2078 tp->snd_recover = th->th_ack - 1; 2079 if ((tcp_do_newreno || tp->sack_enable) && 2080 IN_FASTRECOVERY(tp) && 2081 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2082 EXIT_FASTRECOVERY(tp); 2083 tp->snd_una = th->th_ack; 2084 if (tp->sack_enable) { 2085 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2086 tp->snd_recover = tp->snd_una; 2087 } 2088 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2089 tp->snd_nxt = tp->snd_una; 2090 2091 switch (tp->t_state) { 2092 2093 /* 2094 * In FIN_WAIT_1 STATE in addition to the processing 2095 * for the ESTABLISHED state if our FIN is now acknowledged 2096 * then enter FIN_WAIT_2. 2097 */ 2098 case TCPS_FIN_WAIT_1: 2099 if (ourfinisacked) { 2100 /* 2101 * If we can't receive any more 2102 * data, then closing user can proceed. 2103 * Starting the timer is contrary to the 2104 * specification, but if we don't get a FIN 2105 * we'll hang forever. 2106 */ 2107 /* XXXjl 2108 * we should release the tp also, and use a 2109 * compressed state. 2110 */ 2111 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2112 soisdisconnected(so); 2113 callout_reset(tp->tt_2msl, tcp_maxidle, 2114 tcp_timer_2msl, tp); 2115 } 2116 tp->t_state = TCPS_FIN_WAIT_2; 2117 } 2118 break; 2119 2120 /* 2121 * In CLOSING STATE in addition to the processing for 2122 * the ESTABLISHED state if the ACK acknowledges our FIN 2123 * then enter the TIME-WAIT state, otherwise ignore 2124 * the segment. 2125 */ 2126 case TCPS_CLOSING: 2127 if (ourfinisacked) { 2128 KASSERT(headlocked, ("tcp_input: process_ACK: " 2129 "head not locked")); 2130 tcp_twstart(tp); 2131 INP_INFO_WUNLOCK(&tcbinfo); 2132 m_freem(m); 2133 return; 2134 } 2135 break; 2136 2137 /* 2138 * In LAST_ACK, we may still be waiting for data to drain 2139 * and/or to be acked, as well as for the ack of our FIN. 2140 * If our FIN is now acknowledged, delete the TCB, 2141 * enter the closed state and return. 2142 */ 2143 case TCPS_LAST_ACK: 2144 if (ourfinisacked) { 2145 tp = tcp_close(tp); 2146 goto drop; 2147 } 2148 break; 2149 2150 /* 2151 * In TIME_WAIT state the only thing that should arrive 2152 * is a retransmission of the remote FIN. Acknowledge 2153 * it and restart the finack timer. 2154 */ 2155 case TCPS_TIME_WAIT: 2156 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2157 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2158 tcp_timer_2msl, tp); 2159 goto dropafterack; 2160 } 2161 } 2162 2163step6: 2164 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2165 INP_LOCK_ASSERT(inp); 2166 2167 /* 2168 * Update window information. 2169 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2170 */ 2171 if ((thflags & TH_ACK) && 2172 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2173 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2174 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2175 /* keep track of pure window updates */ 2176 if (tlen == 0 && 2177 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2178 tcpstat.tcps_rcvwinupd++; 2179 tp->snd_wnd = tiwin; 2180 tp->snd_wl1 = th->th_seq; 2181 tp->snd_wl2 = th->th_ack; 2182 if (tp->snd_wnd > tp->max_sndwnd) 2183 tp->max_sndwnd = tp->snd_wnd; 2184 needoutput = 1; 2185 } 2186 2187 /* 2188 * Process segments with URG. 2189 */ 2190 if ((thflags & TH_URG) && th->th_urp && 2191 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2192 /* 2193 * This is a kludge, but if we receive and accept 2194 * random urgent pointers, we'll crash in 2195 * soreceive. It's hard to imagine someone 2196 * actually wanting to send this much urgent data. 2197 */ 2198 SOCKBUF_LOCK(&so->so_rcv); 2199 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2200 th->th_urp = 0; /* XXX */ 2201 thflags &= ~TH_URG; /* XXX */ 2202 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2203 goto dodata; /* XXX */ 2204 } 2205 /* 2206 * If this segment advances the known urgent pointer, 2207 * then mark the data stream. This should not happen 2208 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2209 * a FIN has been received from the remote side. 2210 * In these states we ignore the URG. 2211 * 2212 * According to RFC961 (Assigned Protocols), 2213 * the urgent pointer points to the last octet 2214 * of urgent data. We continue, however, 2215 * to consider it to indicate the first octet 2216 * of data past the urgent section as the original 2217 * spec states (in one of two places). 2218 */ 2219 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2220 tp->rcv_up = th->th_seq + th->th_urp; 2221 so->so_oobmark = so->so_rcv.sb_cc + 2222 (tp->rcv_up - tp->rcv_nxt) - 1; 2223 if (so->so_oobmark == 0) 2224 so->so_rcv.sb_state |= SBS_RCVATMARK; 2225 sohasoutofband(so); 2226 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2227 } 2228 SOCKBUF_UNLOCK(&so->so_rcv); 2229 /* 2230 * Remove out of band data so doesn't get presented to user. 2231 * This can happen independent of advancing the URG pointer, 2232 * but if two URG's are pending at once, some out-of-band 2233 * data may creep in... ick. 2234 */ 2235 if (th->th_urp <= (u_long)tlen && 2236 !(so->so_options & SO_OOBINLINE)) { 2237 /* hdr drop is delayed */ 2238 tcp_pulloutofband(so, th, m, drop_hdrlen); 2239 } 2240 } else { 2241 /* 2242 * If no out of band data is expected, 2243 * pull receive urgent pointer along 2244 * with the receive window. 2245 */ 2246 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2247 tp->rcv_up = tp->rcv_nxt; 2248 } 2249dodata: /* XXX */ 2250 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2251 INP_LOCK_ASSERT(inp); 2252 2253 /* 2254 * Process the segment text, merging it into the TCP sequencing queue, 2255 * and arranging for acknowledgment of receipt if necessary. 2256 * This process logically involves adjusting tp->rcv_wnd as data 2257 * is presented to the user (this happens in tcp_usrreq.c, 2258 * case PRU_RCVD). If a FIN has already been received on this 2259 * connection then we just ignore the text. 2260 */ 2261 if ((tlen || (thflags & TH_FIN)) && 2262 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2263 m_adj(m, drop_hdrlen); /* delayed header drop */ 2264 /* 2265 * Insert segment which includes th into TCP reassembly queue 2266 * with control block tp. Set thflags to whether reassembly now 2267 * includes a segment with FIN. This handles the common case 2268 * inline (segment is the next to be received on an established 2269 * connection, and the queue is empty), avoiding linkage into 2270 * and removal from the queue and repetition of various 2271 * conversions. 2272 * Set DELACK for segments received in order, but ack 2273 * immediately when segments are out of order (so 2274 * fast retransmit can work). 2275 */ 2276 if (th->th_seq == tp->rcv_nxt && 2277 LIST_EMPTY(&tp->t_segq) && 2278 TCPS_HAVEESTABLISHED(tp->t_state)) { 2279 if (DELAY_ACK(tp)) 2280 tp->t_flags |= TF_DELACK; 2281 else 2282 tp->t_flags |= TF_ACKNOW; 2283 tp->rcv_nxt += tlen; 2284 thflags = th->th_flags & TH_FIN; 2285 tcpstat.tcps_rcvpack++; 2286 tcpstat.tcps_rcvbyte += tlen; 2287 ND6_HINT(tp); 2288 SOCKBUF_LOCK(&so->so_rcv); 2289 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2290 m_freem(m); 2291 else 2292 sbappendstream_locked(&so->so_rcv, m); 2293 sorwakeup_locked(so); 2294 } else { 2295 thflags = tcp_reass(tp, th, &tlen, m); 2296 tp->t_flags |= TF_ACKNOW; 2297 } 2298 if (tp->sack_enable) 2299 tcp_update_sack_list(tp, th->th_seq, th->th_seq + tlen); 2300 /* 2301 * Note the amount of data that peer has sent into 2302 * our window, in order to estimate the sender's 2303 * buffer size. 2304 */ 2305 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2306 } else { 2307 m_freem(m); 2308 thflags &= ~TH_FIN; 2309 } 2310 2311 /* 2312 * If FIN is received ACK the FIN and let the user know 2313 * that the connection is closing. 2314 */ 2315 if (thflags & TH_FIN) { 2316 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2317 socantrcvmore(so); 2318 /* 2319 * If connection is half-synchronized 2320 * (ie NEEDSYN flag on) then delay ACK, 2321 * so it may be piggybacked when SYN is sent. 2322 * Otherwise, since we received a FIN then no 2323 * more input can be expected, send ACK now. 2324 */ 2325 if (tp->t_flags & TF_NEEDSYN) 2326 tp->t_flags |= TF_DELACK; 2327 else 2328 tp->t_flags |= TF_ACKNOW; 2329 tp->rcv_nxt++; 2330 } 2331 switch (tp->t_state) { 2332 2333 /* 2334 * In SYN_RECEIVED and ESTABLISHED STATES 2335 * enter the CLOSE_WAIT state. 2336 */ 2337 case TCPS_SYN_RECEIVED: 2338 tp->t_starttime = ticks; 2339 /*FALLTHROUGH*/ 2340 case TCPS_ESTABLISHED: 2341 tp->t_state = TCPS_CLOSE_WAIT; 2342 break; 2343 2344 /* 2345 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2346 * enter the CLOSING state. 2347 */ 2348 case TCPS_FIN_WAIT_1: 2349 tp->t_state = TCPS_CLOSING; 2350 break; 2351 2352 /* 2353 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2354 * starting the time-wait timer, turning off the other 2355 * standard timers. 2356 */ 2357 case TCPS_FIN_WAIT_2: 2358 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2359 "TCP_FIN_WAIT_2: head not locked")); 2360 tcp_twstart(tp); 2361 INP_INFO_WUNLOCK(&tcbinfo); 2362 return; 2363 2364 /* 2365 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2366 */ 2367 case TCPS_TIME_WAIT: 2368 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2369 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2370 tcp_timer_2msl, tp); 2371 break; 2372 } 2373 } 2374 INP_INFO_WUNLOCK(&tcbinfo); 2375 headlocked = 0; 2376#ifdef TCPDEBUG 2377 if (so->so_options & SO_DEBUG) 2378 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2379 &tcp_savetcp, 0); 2380#endif 2381 2382 /* 2383 * Return any desired output. 2384 */ 2385 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2386 (void) tcp_output(tp); 2387 2388check_delack: 2389 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2390 INP_LOCK_ASSERT(inp); 2391 if (tp->t_flags & TF_DELACK) { 2392 tp->t_flags &= ~TF_DELACK; 2393 callout_reset(tp->tt_delack, tcp_delacktime, 2394 tcp_timer_delack, tp); 2395 } 2396 INP_UNLOCK(inp); 2397 return; 2398 2399dropafterack: 2400 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2401 /* 2402 * Generate an ACK dropping incoming segment if it occupies 2403 * sequence space, where the ACK reflects our state. 2404 * 2405 * We can now skip the test for the RST flag since all 2406 * paths to this code happen after packets containing 2407 * RST have been dropped. 2408 * 2409 * In the SYN-RECEIVED state, don't send an ACK unless the 2410 * segment we received passes the SYN-RECEIVED ACK test. 2411 * If it fails send a RST. This breaks the loop in the 2412 * "LAND" DoS attack, and also prevents an ACK storm 2413 * between two listening ports that have been sent forged 2414 * SYN segments, each with the source address of the other. 2415 */ 2416 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2417 (SEQ_GT(tp->snd_una, th->th_ack) || 2418 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2419 rstreason = BANDLIM_RST_OPENPORT; 2420 goto dropwithreset; 2421 } 2422#ifdef TCPDEBUG 2423 if (so->so_options & SO_DEBUG) 2424 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2425 &tcp_savetcp, 0); 2426#endif 2427 KASSERT(headlocked, ("headlocked should be 1")); 2428 INP_INFO_WUNLOCK(&tcbinfo); 2429 tp->t_flags |= TF_ACKNOW; 2430 (void) tcp_output(tp); 2431 INP_UNLOCK(inp); 2432 m_freem(m); 2433 return; 2434 2435dropwithreset: 2436 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2437 /* 2438 * Generate a RST, dropping incoming segment. 2439 * Make ACK acceptable to originator of segment. 2440 * Don't bother to respond if destination was broadcast/multicast. 2441 */ 2442 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2443 goto drop; 2444 if (isipv6) { 2445 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2446 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2447 goto drop; 2448 } else { 2449 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2450 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2451 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2452 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2453 goto drop; 2454 } 2455 /* IPv6 anycast check is done at tcp6_input() */ 2456 2457 /* 2458 * Perform bandwidth limiting. 2459 */ 2460 if (badport_bandlim(rstreason) < 0) 2461 goto drop; 2462 2463#ifdef TCPDEBUG 2464 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2465 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2466 &tcp_savetcp, 0); 2467#endif 2468 2469 if (thflags & TH_ACK) 2470 /* mtod() below is safe as long as hdr dropping is delayed */ 2471 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2472 TH_RST); 2473 else { 2474 if (thflags & TH_SYN) 2475 tlen++; 2476 /* mtod() below is safe as long as hdr dropping is delayed */ 2477 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2478 (tcp_seq)0, TH_RST|TH_ACK); 2479 } 2480 2481 if (tp) 2482 INP_UNLOCK(inp); 2483 if (headlocked) 2484 INP_INFO_WUNLOCK(&tcbinfo); 2485 return; 2486 2487drop: 2488 /* 2489 * Drop space held by incoming segment and return. 2490 */ 2491#ifdef TCPDEBUG 2492 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2493 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2494 &tcp_savetcp, 0); 2495#endif 2496 if (tp) 2497 INP_UNLOCK(inp); 2498 if (headlocked) 2499 INP_INFO_WUNLOCK(&tcbinfo); 2500 m_freem(m); 2501 return; 2502} 2503 2504/* 2505 * Parse TCP options and place in tcpopt. 2506 */ 2507static void 2508tcp_dooptions(tp, to, cp, cnt, is_syn, th) 2509 struct tcpcb *tp; 2510 struct tcpopt *to; 2511 u_char *cp; 2512 int cnt; 2513 int is_syn; 2514 struct tcphdr *th; 2515{ 2516 int opt, optlen; 2517 2518 to->to_flags = 0; 2519 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2520 opt = cp[0]; 2521 if (opt == TCPOPT_EOL) 2522 break; 2523 if (opt == TCPOPT_NOP) 2524 optlen = 1; 2525 else { 2526 if (cnt < 2) 2527 break; 2528 optlen = cp[1]; 2529 if (optlen < 2 || optlen > cnt) 2530 break; 2531 } 2532 switch (opt) { 2533 case TCPOPT_MAXSEG: 2534 if (optlen != TCPOLEN_MAXSEG) 2535 continue; 2536 if (!is_syn) 2537 continue; 2538 to->to_flags |= TOF_MSS; 2539 bcopy((char *)cp + 2, 2540 (char *)&to->to_mss, sizeof(to->to_mss)); 2541 to->to_mss = ntohs(to->to_mss); 2542 break; 2543 case TCPOPT_WINDOW: 2544 if (optlen != TCPOLEN_WINDOW) 2545 continue; 2546 if (! is_syn) 2547 continue; 2548 to->to_flags |= TOF_SCALE; 2549 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2550 break; 2551 case TCPOPT_TIMESTAMP: 2552 if (optlen != TCPOLEN_TIMESTAMP) 2553 continue; 2554 to->to_flags |= TOF_TS; 2555 bcopy((char *)cp + 2, 2556 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2557 to->to_tsval = ntohl(to->to_tsval); 2558 bcopy((char *)cp + 6, 2559 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2560 to->to_tsecr = ntohl(to->to_tsecr); 2561 break; 2562#ifdef TCP_SIGNATURE 2563 /* 2564 * XXX In order to reply to a host which has set the 2565 * TCP_SIGNATURE option in its initial SYN, we have to 2566 * record the fact that the option was observed here 2567 * for the syncache code to perform the correct response. 2568 */ 2569 case TCPOPT_SIGNATURE: 2570 if (optlen != TCPOLEN_SIGNATURE) 2571 continue; 2572 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2573 break; 2574#endif 2575 case TCPOPT_SACK_PERMITTED: 2576 if (!tcp_do_sack || 2577 optlen != TCPOLEN_SACK_PERMITTED) 2578 continue; 2579 if (is_syn) { 2580 /* MUST only be set on SYN */ 2581 to->to_flags |= TOF_SACK; 2582 } 2583 break; 2584 2585 case TCPOPT_SACK: 2586 if (!tp || tcp_sack_option(tp, th, cp, optlen)) 2587 continue; 2588 break; 2589 default: 2590 continue; 2591 } 2592 } 2593} 2594 2595/* 2596 * Pull out of band byte out of a segment so 2597 * it doesn't appear in the user's data queue. 2598 * It is still reflected in the segment length for 2599 * sequencing purposes. 2600 */ 2601static void 2602tcp_pulloutofband(so, th, m, off) 2603 struct socket *so; 2604 struct tcphdr *th; 2605 register struct mbuf *m; 2606 int off; /* delayed to be droped hdrlen */ 2607{ 2608 int cnt = off + th->th_urp - 1; 2609 2610 while (cnt >= 0) { 2611 if (m->m_len > cnt) { 2612 char *cp = mtod(m, caddr_t) + cnt; 2613 struct tcpcb *tp = sototcpcb(so); 2614 2615 tp->t_iobc = *cp; 2616 tp->t_oobflags |= TCPOOB_HAVEDATA; 2617 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2618 m->m_len--; 2619 if (m->m_flags & M_PKTHDR) 2620 m->m_pkthdr.len--; 2621 return; 2622 } 2623 cnt -= m->m_len; 2624 m = m->m_next; 2625 if (m == 0) 2626 break; 2627 } 2628 panic("tcp_pulloutofband"); 2629} 2630 2631/* 2632 * Collect new round-trip time estimate 2633 * and update averages and current timeout. 2634 */ 2635static void 2636tcp_xmit_timer(tp, rtt) 2637 register struct tcpcb *tp; 2638 int rtt; 2639{ 2640 register int delta; 2641 2642 INP_LOCK_ASSERT(tp->t_inpcb); 2643 2644 tcpstat.tcps_rttupdated++; 2645 tp->t_rttupdated++; 2646 if (tp->t_srtt != 0) { 2647 /* 2648 * srtt is stored as fixed point with 5 bits after the 2649 * binary point (i.e., scaled by 8). The following magic 2650 * is equivalent to the smoothing algorithm in rfc793 with 2651 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2652 * point). Adjust rtt to origin 0. 2653 */ 2654 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2655 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2656 2657 if ((tp->t_srtt += delta) <= 0) 2658 tp->t_srtt = 1; 2659 2660 /* 2661 * We accumulate a smoothed rtt variance (actually, a 2662 * smoothed mean difference), then set the retransmit 2663 * timer to smoothed rtt + 4 times the smoothed variance. 2664 * rttvar is stored as fixed point with 4 bits after the 2665 * binary point (scaled by 16). The following is 2666 * equivalent to rfc793 smoothing with an alpha of .75 2667 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2668 * rfc793's wired-in beta. 2669 */ 2670 if (delta < 0) 2671 delta = -delta; 2672 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2673 if ((tp->t_rttvar += delta) <= 0) 2674 tp->t_rttvar = 1; 2675 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2676 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2677 } else { 2678 /* 2679 * No rtt measurement yet - use the unsmoothed rtt. 2680 * Set the variance to half the rtt (so our first 2681 * retransmit happens at 3*rtt). 2682 */ 2683 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2684 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2685 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2686 } 2687 tp->t_rtttime = 0; 2688 tp->t_rxtshift = 0; 2689 2690 /* 2691 * the retransmit should happen at rtt + 4 * rttvar. 2692 * Because of the way we do the smoothing, srtt and rttvar 2693 * will each average +1/2 tick of bias. When we compute 2694 * the retransmit timer, we want 1/2 tick of rounding and 2695 * 1 extra tick because of +-1/2 tick uncertainty in the 2696 * firing of the timer. The bias will give us exactly the 2697 * 1.5 tick we need. But, because the bias is 2698 * statistical, we have to test that we don't drop below 2699 * the minimum feasible timer (which is 2 ticks). 2700 */ 2701 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2702 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2703 2704 /* 2705 * We received an ack for a packet that wasn't retransmitted; 2706 * it is probably safe to discard any error indications we've 2707 * received recently. This isn't quite right, but close enough 2708 * for now (a route might have failed after we sent a segment, 2709 * and the return path might not be symmetrical). 2710 */ 2711 tp->t_softerror = 0; 2712} 2713 2714/* 2715 * Determine a reasonable value for maxseg size. 2716 * If the route is known, check route for mtu. 2717 * If none, use an mss that can be handled on the outgoing 2718 * interface without forcing IP to fragment; if bigger than 2719 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2720 * to utilize large mbufs. If no route is found, route has no mtu, 2721 * or the destination isn't local, use a default, hopefully conservative 2722 * size (usually 512 or the default IP max size, but no more than the mtu 2723 * of the interface), as we can't discover anything about intervening 2724 * gateways or networks. We also initialize the congestion/slow start 2725 * window to be a single segment if the destination isn't local. 2726 * While looking at the routing entry, we also initialize other path-dependent 2727 * parameters from pre-set or cached values in the routing entry. 2728 * 2729 * Also take into account the space needed for options that we 2730 * send regularly. Make maxseg shorter by that amount to assure 2731 * that we can send maxseg amount of data even when the options 2732 * are present. Store the upper limit of the length of options plus 2733 * data in maxopd. 2734 * 2735 * 2736 * In case of T/TCP, we call this routine during implicit connection 2737 * setup as well (offer = -1), to initialize maxseg from the cached 2738 * MSS of our peer. 2739 * 2740 * NOTE that this routine is only called when we process an incoming 2741 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2742 */ 2743void 2744tcp_mss(tp, offer) 2745 struct tcpcb *tp; 2746 int offer; 2747{ 2748 int rtt, mss; 2749 u_long bufsize; 2750 u_long maxmtu; 2751 struct inpcb *inp = tp->t_inpcb; 2752 struct socket *so; 2753 struct hc_metrics_lite metrics; 2754 int origoffer = offer; 2755#ifdef INET6 2756 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2757 size_t min_protoh = isipv6 ? 2758 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2759 sizeof (struct tcpiphdr); 2760#else 2761 const size_t min_protoh = sizeof(struct tcpiphdr); 2762#endif 2763 2764 /* initialize */ 2765#ifdef INET6 2766 if (isipv6) { 2767 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2768 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2769 } else 2770#endif 2771 { 2772 maxmtu = tcp_maxmtu(&inp->inp_inc); 2773 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2774 } 2775 so = inp->inp_socket; 2776 2777 /* 2778 * no route to sender, stay with default mss and return 2779 */ 2780 if (maxmtu == 0) 2781 return; 2782 2783 /* what have we got? */ 2784 switch (offer) { 2785 case 0: 2786 /* 2787 * Offer == 0 means that there was no MSS on the SYN 2788 * segment, in this case we use tcp_mssdflt. 2789 */ 2790 offer = 2791#ifdef INET6 2792 isipv6 ? tcp_v6mssdflt : 2793#endif 2794 tcp_mssdflt; 2795 break; 2796 2797 case -1: 2798 /* 2799 * Offer == -1 means that we didn't receive SYN yet. 2800 */ 2801 /* FALLTHROUGH */ 2802 2803 default: 2804 /* 2805 * Prevent DoS attack with too small MSS. Round up 2806 * to at least minmss. 2807 */ 2808 offer = max(offer, tcp_minmss); 2809 /* 2810 * Sanity check: make sure that maxopd will be large 2811 * enough to allow some data on segments even if the 2812 * all the option space is used (40bytes). Otherwise 2813 * funny things may happen in tcp_output. 2814 */ 2815 offer = max(offer, 64); 2816 } 2817 2818 /* 2819 * rmx information is now retrieved from tcp_hostcache 2820 */ 2821 tcp_hc_get(&inp->inp_inc, &metrics); 2822 2823 /* 2824 * if there's a discovered mtu int tcp hostcache, use it 2825 * else, use the link mtu. 2826 */ 2827 if (metrics.rmx_mtu) 2828 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2829 else { 2830#ifdef INET6 2831 if (isipv6) { 2832 mss = maxmtu - min_protoh; 2833 if (!path_mtu_discovery && 2834 !in6_localaddr(&inp->in6p_faddr)) 2835 mss = min(mss, tcp_v6mssdflt); 2836 } else 2837#endif 2838 { 2839 mss = maxmtu - min_protoh; 2840 if (!path_mtu_discovery && 2841 !in_localaddr(inp->inp_faddr)) 2842 mss = min(mss, tcp_mssdflt); 2843 } 2844 } 2845 mss = min(mss, offer); 2846 2847 /* 2848 * maxopd stores the maximum length of data AND options 2849 * in a segment; maxseg is the amount of data in a normal 2850 * segment. We need to store this value (maxopd) apart 2851 * from maxseg, because now every segment carries options 2852 * and thus we normally have somewhat less data in segments. 2853 */ 2854 tp->t_maxopd = mss; 2855 2856 /* 2857 * origoffer==-1 indicates, that no segments were received yet. 2858 * In this case we just guess. 2859 */ 2860 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2861 (origoffer == -1 || 2862 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2863 mss -= TCPOLEN_TSTAMP_APPA; 2864 tp->t_maxseg = mss; 2865 2866#if (MCLBYTES & (MCLBYTES - 1)) == 0 2867 if (mss > MCLBYTES) 2868 mss &= ~(MCLBYTES-1); 2869#else 2870 if (mss > MCLBYTES) 2871 mss = mss / MCLBYTES * MCLBYTES; 2872#endif 2873 tp->t_maxseg = mss; 2874 2875 /* 2876 * If there's a pipesize, change the socket buffer to that size, 2877 * don't change if sb_hiwat is different than default (then it 2878 * has been changed on purpose with setsockopt). 2879 * Make the socket buffers an integral number of mss units; 2880 * if the mss is larger than the socket buffer, decrease the mss. 2881 */ 2882 SOCKBUF_LOCK(&so->so_snd); 2883 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2884 bufsize = metrics.rmx_sendpipe; 2885 else 2886 bufsize = so->so_snd.sb_hiwat; 2887 if (bufsize < mss) 2888 mss = bufsize; 2889 else { 2890 bufsize = roundup(bufsize, mss); 2891 if (bufsize > sb_max) 2892 bufsize = sb_max; 2893 if (bufsize > so->so_snd.sb_hiwat) 2894 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2895 } 2896 SOCKBUF_UNLOCK(&so->so_snd); 2897 tp->t_maxseg = mss; 2898 2899 SOCKBUF_LOCK(&so->so_rcv); 2900 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2901 bufsize = metrics.rmx_recvpipe; 2902 else 2903 bufsize = so->so_rcv.sb_hiwat; 2904 if (bufsize > mss) { 2905 bufsize = roundup(bufsize, mss); 2906 if (bufsize > sb_max) 2907 bufsize = sb_max; 2908 if (bufsize > so->so_rcv.sb_hiwat) 2909 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2910 } 2911 SOCKBUF_UNLOCK(&so->so_rcv); 2912 /* 2913 * While we're here, check the others too 2914 */ 2915 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2916 tp->t_srtt = rtt; 2917 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2918 tcpstat.tcps_usedrtt++; 2919 if (metrics.rmx_rttvar) { 2920 tp->t_rttvar = metrics.rmx_rttvar; 2921 tcpstat.tcps_usedrttvar++; 2922 } else { 2923 /* default variation is +- 1 rtt */ 2924 tp->t_rttvar = 2925 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2926 } 2927 TCPT_RANGESET(tp->t_rxtcur, 2928 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2929 tp->t_rttmin, TCPTV_REXMTMAX); 2930 } 2931 if (metrics.rmx_ssthresh) { 2932 /* 2933 * There's some sort of gateway or interface 2934 * buffer limit on the path. Use this to set 2935 * the slow start threshhold, but set the 2936 * threshold to no less than 2*mss. 2937 */ 2938 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2939 tcpstat.tcps_usedssthresh++; 2940 } 2941 if (metrics.rmx_bandwidth) 2942 tp->snd_bandwidth = metrics.rmx_bandwidth; 2943 2944 /* 2945 * Set the slow-start flight size depending on whether this 2946 * is a local network or not. 2947 * 2948 * Extend this so we cache the cwnd too and retrieve it here. 2949 * Make cwnd even bigger than RFC3390 suggests but only if we 2950 * have previous experience with the remote host. Be careful 2951 * not make cwnd bigger than remote receive window or our own 2952 * send socket buffer. Maybe put some additional upper bound 2953 * on the retrieved cwnd. Should do incremental updates to 2954 * hostcache when cwnd collapses so next connection doesn't 2955 * overloads the path again. 2956 * 2957 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2958 * We currently check only in syncache_socket for that. 2959 */ 2960#define TCP_METRICS_CWND 2961#ifdef TCP_METRICS_CWND 2962 if (metrics.rmx_cwnd) 2963 tp->snd_cwnd = max(mss, 2964 min(metrics.rmx_cwnd / 2, 2965 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2966 else 2967#endif 2968 if (tcp_do_rfc3390) 2969 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2970#ifdef INET6 2971 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2972 (!isipv6 && in_localaddr(inp->inp_faddr))) 2973#else 2974 else if (in_localaddr(inp->inp_faddr)) 2975#endif 2976 tp->snd_cwnd = mss * ss_fltsz_local; 2977 else 2978 tp->snd_cwnd = mss * ss_fltsz; 2979} 2980 2981/* 2982 * Determine the MSS option to send on an outgoing SYN. 2983 */ 2984int 2985tcp_mssopt(inc) 2986 struct in_conninfo *inc; 2987{ 2988 int mss = 0; 2989 u_long maxmtu = 0; 2990 u_long thcmtu = 0; 2991 size_t min_protoh; 2992#ifdef INET6 2993 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2994#endif 2995 2996 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2997 2998#ifdef INET6 2999 if (isipv6) { 3000 mss = tcp_v6mssdflt; 3001 maxmtu = tcp_maxmtu6(inc); 3002 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3003 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3004 } else 3005#endif 3006 { 3007 mss = tcp_mssdflt; 3008 maxmtu = tcp_maxmtu(inc); 3009 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3010 min_protoh = sizeof(struct tcpiphdr); 3011 } 3012 if (maxmtu && thcmtu) 3013 mss = min(maxmtu, thcmtu) - min_protoh; 3014 else if (maxmtu || thcmtu) 3015 mss = max(maxmtu, thcmtu) - min_protoh; 3016 3017 return (mss); 3018} 3019 3020 3021/* 3022 * On a partial ack arrives, force the retransmission of the 3023 * next unacknowledged segment. Do not clear tp->t_dupacks. 3024 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3025 * be started again. 3026 */ 3027static void 3028tcp_newreno_partial_ack(tp, th) 3029 struct tcpcb *tp; 3030 struct tcphdr *th; 3031{ 3032 tcp_seq onxt = tp->snd_nxt; 3033 u_long ocwnd = tp->snd_cwnd; 3034 3035 callout_stop(tp->tt_rexmt); 3036 tp->t_rtttime = 0; 3037 tp->snd_nxt = th->th_ack; 3038 /* 3039 * Set snd_cwnd to one segment beyond acknowledged offset. 3040 * (tp->snd_una has not yet been updated when this function is called.) 3041 */ 3042 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3043 tp->t_flags |= TF_ACKNOW; 3044 (void) tcp_output(tp); 3045 tp->snd_cwnd = ocwnd; 3046 if (SEQ_GT(onxt, tp->snd_nxt)) 3047 tp->snd_nxt = onxt; 3048 /* 3049 * Partial window deflation. Relies on fact that tp->snd_una 3050 * not updated yet. 3051 */ 3052 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3053} 3054 3055/* 3056 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3057 * looking for a pcb in the listen state. Returns 0 otherwise. 3058 */ 3059static int 3060tcp_timewait(tw, to, th, m, tlen) 3061 struct tcptw *tw; 3062 struct tcpopt *to; 3063 struct tcphdr *th; 3064 struct mbuf *m; 3065 int tlen; 3066{ 3067 int thflags; 3068 tcp_seq seq; 3069#ifdef INET6 3070 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3071#else 3072 const int isipv6 = 0; 3073#endif 3074 3075 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3076 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3077 INP_LOCK_ASSERT(tw->tw_inpcb); 3078 3079 thflags = th->th_flags; 3080 3081 /* 3082 * NOTE: for FIN_WAIT_2 (to be added later), 3083 * must validate sequence number before accepting RST 3084 */ 3085 3086 /* 3087 * If the segment contains RST: 3088 * Drop the segment - see Stevens, vol. 2, p. 964 and 3089 * RFC 1337. 3090 */ 3091 if (thflags & TH_RST) 3092 goto drop; 3093 3094#if 0 3095/* PAWS not needed at the moment */ 3096 /* 3097 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3098 * and it's less than ts_recent, drop it. 3099 */ 3100 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3101 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3102 if ((thflags & TH_ACK) == 0) 3103 goto drop; 3104 goto ack; 3105 } 3106 /* 3107 * ts_recent is never updated because we never accept new segments. 3108 */ 3109#endif 3110 3111 /* 3112 * If a new connection request is received 3113 * while in TIME_WAIT, drop the old connection 3114 * and start over if the sequence numbers 3115 * are above the previous ones. 3116 */ 3117 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3118 (void) tcp_twclose(tw, 0); 3119 return (1); 3120 } 3121 3122 /* 3123 * Drop the the segment if it does not contain an ACK. 3124 */ 3125 if ((thflags & TH_ACK) == 0) 3126 goto drop; 3127 3128 /* 3129 * Reset the 2MSL timer if this is a duplicate FIN. 3130 */ 3131 if (thflags & TH_FIN) { 3132 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3133 if (seq + 1 == tw->rcv_nxt) 3134 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3135 } 3136 3137 /* 3138 * Acknowledge the segment if it has data or is not a duplicate ACK. 3139 */ 3140 if (thflags != TH_ACK || tlen != 0 || 3141 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3142 tcp_twrespond(tw, TH_ACK); 3143 goto drop; 3144 3145 /* 3146 * Generate a RST, dropping incoming segment. 3147 * Make ACK acceptable to originator of segment. 3148 * Don't bother to respond if destination was broadcast/multicast. 3149 */ 3150 if (m->m_flags & (M_BCAST|M_MCAST)) 3151 goto drop; 3152 if (isipv6) { 3153 struct ip6_hdr *ip6; 3154 3155 /* IPv6 anycast check is done at tcp6_input() */ 3156 ip6 = mtod(m, struct ip6_hdr *); 3157 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3158 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3159 goto drop; 3160 } else { 3161 struct ip *ip; 3162 3163 ip = mtod(m, struct ip *); 3164 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3165 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3166 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3167 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3168 goto drop; 3169 } 3170 if (thflags & TH_ACK) { 3171 tcp_respond(NULL, 3172 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3173 } else { 3174 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3175 tcp_respond(NULL, 3176 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3177 } 3178 INP_UNLOCK(tw->tw_inpcb); 3179 return (0); 3180 3181drop: 3182 INP_UNLOCK(tw->tw_inpcb); 3183 m_freem(m); 3184 return (0); 3185} 3186