tcp_input.c revision 138040
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 138040 2004-11-23 23:41:20Z rwatson $ 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 134SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 135 "TCP Segment Reassembly Queue"); 136 137static int tcp_reass_maxseg = 0; 138SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 139 &tcp_reass_maxseg, 0, 140 "Global maximum number of TCP Segments in Reassembly Queue"); 141 142int tcp_reass_qsize = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 144 &tcp_reass_qsize, 0, 145 "Global number of TCP Segments currently in Reassembly Queue"); 146 147static int tcp_reass_maxqlen = 48; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 149 &tcp_reass_maxqlen, 0, 150 "Maximum number of TCP Segments per individual Reassembly Queue"); 151 152static int tcp_reass_overflows = 0; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 154 &tcp_reass_overflows, 0, 155 "Global number of TCP Segment Reassembly Queue Overflows"); 156 157static int tcp_sack_recovery_initburst = 3; 158SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, 159 initburst, CTLFLAG_RW, 160 &tcp_sack_recovery_initburst, 0, 161 "Initial Number of Rexmits when sack recovery is set up"); 162 163struct inpcbhead tcb; 164#define tcb6 tcb /* for KAME src sync over BSD*'s */ 165struct inpcbinfo tcbinfo; 166struct mtx *tcbinfo_mtx; 167 168static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *, 169 int, int, struct tcphdr *); 170 171static void tcp_pulloutofband(struct socket *, 172 struct tcphdr *, struct mbuf *, int); 173static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 174 struct mbuf *); 175static void tcp_xmit_timer(struct tcpcb *, int); 176static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 177static int tcp_timewait(struct tcptw *, struct tcpopt *, 178 struct tcphdr *, struct mbuf *, int); 179 180/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 181#ifdef INET6 182#define ND6_HINT(tp) \ 183do { \ 184 if ((tp) && (tp)->t_inpcb && \ 185 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 186 nd6_nud_hint(NULL, NULL, 0); \ 187} while (0) 188#else 189#define ND6_HINT(tp) 190#endif 191 192/* 193 * Indicate whether this ack should be delayed. We can delay the ack if 194 * - there is no delayed ack timer in progress and 195 * - our last ack wasn't a 0-sized window. We never want to delay 196 * the ack that opens up a 0-sized window and 197 * - delayed acks are enabled or 198 * - this is a half-synchronized T/TCP connection. 199 */ 200#define DELAY_ACK(tp) \ 201 ((!callout_active(tp->tt_delack) && \ 202 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 203 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 204 205/* Initialize TCP reassembly queue */ 206uma_zone_t tcp_reass_zone; 207void 208tcp_reass_init() 209{ 210 tcp_reass_maxseg = nmbclusters / 16; 211 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 212 &tcp_reass_maxseg); 213 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 215 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 216} 217 218static int 219tcp_reass(tp, th, tlenp, m) 220 register struct tcpcb *tp; 221 register struct tcphdr *th; 222 int *tlenp; 223 struct mbuf *m; 224{ 225 struct tseg_qent *q; 226 struct tseg_qent *p = NULL; 227 struct tseg_qent *nq; 228 struct tseg_qent *te = NULL; 229 struct socket *so = tp->t_inpcb->inp_socket; 230 int flags; 231 232 INP_LOCK_ASSERT(tp->t_inpcb); 233 234 /* 235 * XXX: tcp_reass() is rather inefficient with its data structures 236 * and should be rewritten (see NetBSD for optimizations). While 237 * doing that it should move to its own file tcp_reass.c. 238 */ 239 240 /* 241 * Call with th==NULL after become established to 242 * force pre-ESTABLISHED data up to user socket. 243 */ 244 if (th == NULL) 245 goto present; 246 247 /* 248 * Limit the number of segments in the reassembly queue to prevent 249 * holding on to too many segments (and thus running out of mbufs). 250 * Make sure to let the missing segment through which caused this 251 * queue. Always keep one global queue entry spare to be able to 252 * process the missing segment. 253 */ 254 if (th->th_seq != tp->rcv_nxt && 255 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 256 tp->t_segqlen >= tcp_reass_maxqlen)) { 257 tcp_reass_overflows++; 258 tcpstat.tcps_rcvmemdrop++; 259 m_freem(m); 260 return (0); 261 } 262 263 /* 264 * Allocate a new queue entry. If we can't, or hit the zone limit 265 * just drop the pkt. 266 */ 267 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 268 if (te == NULL) { 269 tcpstat.tcps_rcvmemdrop++; 270 m_freem(m); 271 return (0); 272 } 273 tp->t_segqlen++; 274 tcp_reass_qsize++; 275 276 /* 277 * Find a segment which begins after this one does. 278 */ 279 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 280 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 281 break; 282 p = q; 283 } 284 285 /* 286 * If there is a preceding segment, it may provide some of 287 * our data already. If so, drop the data from the incoming 288 * segment. If it provides all of our data, drop us. 289 */ 290 if (p != NULL) { 291 register int i; 292 /* conversion to int (in i) handles seq wraparound */ 293 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 294 if (i > 0) { 295 if (i >= *tlenp) { 296 tcpstat.tcps_rcvduppack++; 297 tcpstat.tcps_rcvdupbyte += *tlenp; 298 m_freem(m); 299 uma_zfree(tcp_reass_zone, te); 300 tp->t_segqlen--; 301 tcp_reass_qsize--; 302 /* 303 * Try to present any queued data 304 * at the left window edge to the user. 305 * This is needed after the 3-WHS 306 * completes. 307 */ 308 goto present; /* ??? */ 309 } 310 m_adj(m, i); 311 *tlenp -= i; 312 th->th_seq += i; 313 } 314 } 315 tcpstat.tcps_rcvoopack++; 316 tcpstat.tcps_rcvoobyte += *tlenp; 317 318 /* 319 * While we overlap succeeding segments trim them or, 320 * if they are completely covered, dequeue them. 321 */ 322 while (q) { 323 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 324 if (i <= 0) 325 break; 326 if (i < q->tqe_len) { 327 q->tqe_th->th_seq += i; 328 q->tqe_len -= i; 329 m_adj(q->tqe_m, i); 330 break; 331 } 332 333 nq = LIST_NEXT(q, tqe_q); 334 LIST_REMOVE(q, tqe_q); 335 m_freem(q->tqe_m); 336 uma_zfree(tcp_reass_zone, q); 337 tp->t_segqlen--; 338 tcp_reass_qsize--; 339 q = nq; 340 } 341 342 /* Insert the new segment queue entry into place. */ 343 te->tqe_m = m; 344 te->tqe_th = th; 345 te->tqe_len = *tlenp; 346 347 if (p == NULL) { 348 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 349 } else { 350 LIST_INSERT_AFTER(p, te, tqe_q); 351 } 352 353present: 354 /* 355 * Present data to user, advancing rcv_nxt through 356 * completed sequence space. 357 */ 358 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 359 return (0); 360 q = LIST_FIRST(&tp->t_segq); 361 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 362 return (0); 363 SOCKBUF_LOCK(&so->so_rcv); 364 do { 365 tp->rcv_nxt += q->tqe_len; 366 flags = q->tqe_th->th_flags & TH_FIN; 367 nq = LIST_NEXT(q, tqe_q); 368 LIST_REMOVE(q, tqe_q); 369 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 370 m_freem(q->tqe_m); 371 else 372 sbappendstream_locked(&so->so_rcv, q->tqe_m); 373 uma_zfree(tcp_reass_zone, q); 374 tp->t_segqlen--; 375 tcp_reass_qsize--; 376 q = nq; 377 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 378 ND6_HINT(tp); 379 sorwakeup_locked(so); 380 return (flags); 381} 382 383/* 384 * TCP input routine, follows pages 65-76 of the 385 * protocol specification dated September, 1981 very closely. 386 */ 387#ifdef INET6 388int 389tcp6_input(mp, offp, proto) 390 struct mbuf **mp; 391 int *offp, proto; 392{ 393 register struct mbuf *m = *mp; 394 struct in6_ifaddr *ia6; 395 396 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 397 398 /* 399 * draft-itojun-ipv6-tcp-to-anycast 400 * better place to put this in? 401 */ 402 ia6 = ip6_getdstifaddr(m); 403 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 404 struct ip6_hdr *ip6; 405 406 ip6 = mtod(m, struct ip6_hdr *); 407 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 408 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 409 return IPPROTO_DONE; 410 } 411 412 tcp_input(m, *offp); 413 return IPPROTO_DONE; 414} 415#endif 416 417void 418tcp_input(m, off0) 419 register struct mbuf *m; 420 int off0; 421{ 422 register struct tcphdr *th; 423 register struct ip *ip = NULL; 424 register struct ipovly *ipov; 425 register struct inpcb *inp = NULL; 426 u_char *optp = NULL; 427 int optlen = 0; 428 int len, tlen, off; 429 int drop_hdrlen; 430 register struct tcpcb *tp = 0; 431 register int thflags; 432 struct socket *so = 0; 433 int todrop, acked, ourfinisacked, needoutput = 0; 434 u_long tiwin; 435 struct tcpopt to; /* options in this segment */ 436 int headlocked = 0; 437#ifdef IPFIREWALL_FORWARD 438 struct m_tag *fwd_tag; 439#endif 440 int rstreason; /* For badport_bandlim accounting purposes */ 441 442 struct ip6_hdr *ip6 = NULL; 443#ifdef INET6 444 int isipv6; 445#else 446 const int isipv6 = 0; 447#endif 448 449#ifdef TCPDEBUG 450 /* 451 * The size of tcp_saveipgen must be the size of the max ip header, 452 * now IPv6. 453 */ 454 u_char tcp_saveipgen[40]; 455 struct tcphdr tcp_savetcp; 456 short ostate = 0; 457#endif 458 459#ifdef INET6 460 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 461#endif 462 bzero((char *)&to, sizeof(to)); 463 464 tcpstat.tcps_rcvtotal++; 465 466 if (isipv6) { 467#ifdef INET6 468 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 469 ip6 = mtod(m, struct ip6_hdr *); 470 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 471 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 472 tcpstat.tcps_rcvbadsum++; 473 goto drop; 474 } 475 th = (struct tcphdr *)((caddr_t)ip6 + off0); 476 477 /* 478 * Be proactive about unspecified IPv6 address in source. 479 * As we use all-zero to indicate unbounded/unconnected pcb, 480 * unspecified IPv6 address can be used to confuse us. 481 * 482 * Note that packets with unspecified IPv6 destination is 483 * already dropped in ip6_input. 484 */ 485 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 486 /* XXX stat */ 487 goto drop; 488 } 489#else 490 th = NULL; /* XXX: avoid compiler warning */ 491#endif 492 } else { 493 /* 494 * Get IP and TCP header together in first mbuf. 495 * Note: IP leaves IP header in first mbuf. 496 */ 497 if (off0 > sizeof (struct ip)) { 498 ip_stripoptions(m, (struct mbuf *)0); 499 off0 = sizeof(struct ip); 500 } 501 if (m->m_len < sizeof (struct tcpiphdr)) { 502 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 503 tcpstat.tcps_rcvshort++; 504 return; 505 } 506 } 507 ip = mtod(m, struct ip *); 508 ipov = (struct ipovly *)ip; 509 th = (struct tcphdr *)((caddr_t)ip + off0); 510 tlen = ip->ip_len; 511 512 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 513 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 514 th->th_sum = m->m_pkthdr.csum_data; 515 else 516 th->th_sum = in_pseudo(ip->ip_src.s_addr, 517 ip->ip_dst.s_addr, 518 htonl(m->m_pkthdr.csum_data + 519 ip->ip_len + 520 IPPROTO_TCP)); 521 th->th_sum ^= 0xffff; 522#ifdef TCPDEBUG 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525#endif 526 } else { 527 /* 528 * Checksum extended TCP header and data. 529 */ 530 len = sizeof (struct ip) + tlen; 531 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 532 ipov->ih_len = (u_short)tlen; 533 ipov->ih_len = htons(ipov->ih_len); 534 th->th_sum = in_cksum(m, len); 535 } 536 if (th->th_sum) { 537 tcpstat.tcps_rcvbadsum++; 538 goto drop; 539 } 540#ifdef INET6 541 /* Re-initialization for later version check */ 542 ip->ip_v = IPVERSION; 543#endif 544 } 545 546 /* 547 * Check that TCP offset makes sense, 548 * pull out TCP options and adjust length. XXX 549 */ 550 off = th->th_off << 2; 551 if (off < sizeof (struct tcphdr) || off > tlen) { 552 tcpstat.tcps_rcvbadoff++; 553 goto drop; 554 } 555 tlen -= off; /* tlen is used instead of ti->ti_len */ 556 if (off > sizeof (struct tcphdr)) { 557 if (isipv6) { 558#ifdef INET6 559 IP6_EXTHDR_CHECK(m, off0, off, ); 560 ip6 = mtod(m, struct ip6_hdr *); 561 th = (struct tcphdr *)((caddr_t)ip6 + off0); 562#endif 563 } else { 564 if (m->m_len < sizeof(struct ip) + off) { 565 if ((m = m_pullup(m, sizeof (struct ip) + off)) 566 == 0) { 567 tcpstat.tcps_rcvshort++; 568 return; 569 } 570 ip = mtod(m, struct ip *); 571 ipov = (struct ipovly *)ip; 572 th = (struct tcphdr *)((caddr_t)ip + off0); 573 } 574 } 575 optlen = off - sizeof (struct tcphdr); 576 optp = (u_char *)(th + 1); 577 } 578 thflags = th->th_flags; 579 580#ifdef TCP_DROP_SYNFIN 581 /* 582 * If the drop_synfin option is enabled, drop all packets with 583 * both the SYN and FIN bits set. This prevents e.g. nmap from 584 * identifying the TCP/IP stack. 585 * 586 * This is a violation of the TCP specification. 587 */ 588 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 589 goto drop; 590#endif 591 592 /* 593 * Convert TCP protocol specific fields to host format. 594 */ 595 th->th_seq = ntohl(th->th_seq); 596 th->th_ack = ntohl(th->th_ack); 597 th->th_win = ntohs(th->th_win); 598 th->th_urp = ntohs(th->th_urp); 599 600 /* 601 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 602 * until after ip6_savecontrol() is called and before other functions 603 * which don't want those proto headers. 604 * Because ip6_savecontrol() is going to parse the mbuf to 605 * search for data to be passed up to user-land, it wants mbuf 606 * parameters to be unchanged. 607 * XXX: the call of ip6_savecontrol() has been obsoleted based on 608 * latest version of the advanced API (20020110). 609 */ 610 drop_hdrlen = off0 + off; 611 612 /* 613 * Locate pcb for segment. 614 */ 615 INP_INFO_WLOCK(&tcbinfo); 616 headlocked = 1; 617findpcb: 618 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 619#ifdef IPFIREWALL_FORWARD 620 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 621 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 622 623 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 624 struct sockaddr_in *next_hop; 625 626 next_hop = (struct sockaddr_in *)(fwd_tag+1); 627 /* 628 * Transparently forwarded. Pretend to be the destination. 629 * already got one like this? 630 */ 631 inp = in_pcblookup_hash(&tcbinfo, 632 ip->ip_src, th->th_sport, 633 ip->ip_dst, th->th_dport, 634 0, m->m_pkthdr.rcvif); 635 if (!inp) { 636 /* It's new. Try to find the ambushing socket. */ 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 next_hop->sin_addr, 640 next_hop->sin_port ? 641 ntohs(next_hop->sin_port) : 642 th->th_dport, 643 1, m->m_pkthdr.rcvif); 644 } 645 /* Remove the tag from the packet. We don't need it anymore. */ 646 m_tag_delete(m, fwd_tag); 647 } else { 648#endif /* IPFIREWALL_FORWARD */ 649 if (isipv6) { 650#ifdef INET6 651 inp = in6_pcblookup_hash(&tcbinfo, 652 &ip6->ip6_src, th->th_sport, 653 &ip6->ip6_dst, th->th_dport, 654 1, m->m_pkthdr.rcvif); 655#endif 656 } else 657 inp = in_pcblookup_hash(&tcbinfo, 658 ip->ip_src, th->th_sport, 659 ip->ip_dst, th->th_dport, 660 1, m->m_pkthdr.rcvif); 661#ifdef IPFIREWALL_FORWARD 662 } 663#endif /* IPFIREWALL_FORWARD */ 664 665#if defined(IPSEC) || defined(FAST_IPSEC) 666#ifdef INET6 667 if (isipv6) { 668 if (inp != NULL && ipsec6_in_reject(m, inp)) { 669#ifdef IPSEC 670 ipsec6stat.in_polvio++; 671#endif 672 goto drop; 673 } 674 } else 675#endif /* INET6 */ 676 if (inp != NULL && ipsec4_in_reject(m, inp)) { 677#ifdef IPSEC 678 ipsecstat.in_polvio++; 679#endif 680 goto drop; 681 } 682#endif /*IPSEC || FAST_IPSEC*/ 683 684 /* 685 * If the state is CLOSED (i.e., TCB does not exist) then 686 * all data in the incoming segment is discarded. 687 * If the TCB exists but is in CLOSED state, it is embryonic, 688 * but should either do a listen or a connect soon. 689 */ 690 if (inp == NULL) { 691 if (log_in_vain) { 692#ifdef INET6 693 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 694#else 695 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 696#endif 697 698 if (isipv6) { 699#ifdef INET6 700 strcpy(dbuf, "["); 701 strcpy(sbuf, "["); 702 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 703 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 704 strcat(dbuf, "]"); 705 strcat(sbuf, "]"); 706#endif 707 } else { 708 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 709 strcpy(sbuf, inet_ntoa(ip->ip_src)); 710 } 711 switch (log_in_vain) { 712 case 1: 713 if ((thflags & TH_SYN) == 0) 714 break; 715 /* FALLTHROUGH */ 716 case 2: 717 log(LOG_INFO, 718 "Connection attempt to TCP %s:%d " 719 "from %s:%d flags:0x%02x\n", 720 dbuf, ntohs(th->th_dport), sbuf, 721 ntohs(th->th_sport), thflags); 722 break; 723 default: 724 break; 725 } 726 } 727 if (blackhole) { 728 switch (blackhole) { 729 case 1: 730 if (thflags & TH_SYN) 731 goto drop; 732 break; 733 case 2: 734 goto drop; 735 default: 736 goto drop; 737 } 738 } 739 rstreason = BANDLIM_RST_CLOSEDPORT; 740 goto dropwithreset; 741 } 742 INP_LOCK(inp); 743 if (inp->inp_vflag & INP_TIMEWAIT) { 744 /* 745 * The only option of relevance is TOF_CC, and only if 746 * present in a SYN segment. See tcp_timewait(). 747 */ 748 if (thflags & TH_SYN) 749 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th); 750 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 751 &to, th, m, tlen)) 752 goto findpcb; 753 /* 754 * tcp_timewait unlocks inp. 755 */ 756 INP_INFO_WUNLOCK(&tcbinfo); 757 return; 758 } 759 tp = intotcpcb(inp); 760 if (tp == 0) { 761 INP_UNLOCK(inp); 762 rstreason = BANDLIM_RST_CLOSEDPORT; 763 goto dropwithreset; 764 } 765 if (tp->t_state == TCPS_CLOSED) 766 goto drop; 767 768 /* Unscale the window into a 32-bit value. */ 769 if ((thflags & TH_SYN) == 0) 770 tiwin = th->th_win << tp->snd_scale; 771 else 772 tiwin = th->th_win; 773 774#ifdef MAC 775 INP_LOCK_ASSERT(inp); 776 if (mac_check_inpcb_deliver(inp, m)) 777 goto drop; 778#endif 779 so = inp->inp_socket; 780#ifdef TCPDEBUG 781 if (so->so_options & SO_DEBUG) { 782 ostate = tp->t_state; 783 if (isipv6) 784 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 785 else 786 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 787 tcp_savetcp = *th; 788 } 789#endif 790 if (so->so_options & SO_ACCEPTCONN) { 791 struct in_conninfo inc; 792 793#ifdef INET6 794 inc.inc_isipv6 = isipv6; 795#endif 796 if (isipv6) { 797 inc.inc6_faddr = ip6->ip6_src; 798 inc.inc6_laddr = ip6->ip6_dst; 799 } else { 800 inc.inc_faddr = ip->ip_src; 801 inc.inc_laddr = ip->ip_dst; 802 } 803 inc.inc_fport = th->th_sport; 804 inc.inc_lport = th->th_dport; 805 806 /* 807 * If the state is LISTEN then ignore segment if it contains 808 * a RST. If the segment contains an ACK then it is bad and 809 * send a RST. If it does not contain a SYN then it is not 810 * interesting; drop it. 811 * 812 * If the state is SYN_RECEIVED (syncache) and seg contains 813 * an ACK, but not for our SYN/ACK, send a RST. If the seg 814 * contains a RST, check the sequence number to see if it 815 * is a valid reset segment. 816 */ 817 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 818 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 819 if (!syncache_expand(&inc, th, &so, m)) { 820 /* 821 * No syncache entry, or ACK was not 822 * for our SYN/ACK. Send a RST. 823 */ 824 tcpstat.tcps_badsyn++; 825 rstreason = BANDLIM_RST_OPENPORT; 826 goto dropwithreset; 827 } 828 if (so == NULL) { 829 /* 830 * Could not complete 3-way handshake, 831 * connection is being closed down, and 832 * syncache will free mbuf. 833 */ 834 INP_UNLOCK(inp); 835 INP_INFO_WUNLOCK(&tcbinfo); 836 return; 837 } 838 /* 839 * Socket is created in state SYN_RECEIVED. 840 * Continue processing segment. 841 */ 842 INP_UNLOCK(inp); 843 inp = sotoinpcb(so); 844 INP_LOCK(inp); 845 tp = intotcpcb(inp); 846 /* 847 * This is what would have happened in 848 * tcp_output() when the SYN,ACK was sent. 849 */ 850 tp->snd_up = tp->snd_una; 851 tp->snd_max = tp->snd_nxt = tp->iss + 1; 852 tp->last_ack_sent = tp->rcv_nxt; 853 /* 854 * RFC1323: The window in SYN & SYN/ACK 855 * segments is never scaled. 856 */ 857 tp->snd_wnd = tiwin; /* unscaled */ 858 goto after_listen; 859 } 860 if (thflags & TH_RST) { 861 syncache_chkrst(&inc, th); 862 goto drop; 863 } 864 if (thflags & TH_ACK) { 865 syncache_badack(&inc); 866 tcpstat.tcps_badsyn++; 867 rstreason = BANDLIM_RST_OPENPORT; 868 goto dropwithreset; 869 } 870 goto drop; 871 } 872 873 /* 874 * Segment's flags are (SYN) or (SYN|FIN). 875 */ 876#ifdef INET6 877 /* 878 * If deprecated address is forbidden, 879 * we do not accept SYN to deprecated interface 880 * address to prevent any new inbound connection from 881 * getting established. 882 * When we do not accept SYN, we send a TCP RST, 883 * with deprecated source address (instead of dropping 884 * it). We compromise it as it is much better for peer 885 * to send a RST, and RST will be the final packet 886 * for the exchange. 887 * 888 * If we do not forbid deprecated addresses, we accept 889 * the SYN packet. RFC2462 does not suggest dropping 890 * SYN in this case. 891 * If we decipher RFC2462 5.5.4, it says like this: 892 * 1. use of deprecated addr with existing 893 * communication is okay - "SHOULD continue to be 894 * used" 895 * 2. use of it with new communication: 896 * (2a) "SHOULD NOT be used if alternate address 897 * with sufficient scope is available" 898 * (2b) nothing mentioned otherwise. 899 * Here we fall into (2b) case as we have no choice in 900 * our source address selection - we must obey the peer. 901 * 902 * The wording in RFC2462 is confusing, and there are 903 * multiple description text for deprecated address 904 * handling - worse, they are not exactly the same. 905 * I believe 5.5.4 is the best one, so we follow 5.5.4. 906 */ 907 if (isipv6 && !ip6_use_deprecated) { 908 struct in6_ifaddr *ia6; 909 910 if ((ia6 = ip6_getdstifaddr(m)) && 911 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 912 INP_UNLOCK(inp); 913 tp = NULL; 914 rstreason = BANDLIM_RST_OPENPORT; 915 goto dropwithreset; 916 } 917 } 918#endif 919 /* 920 * If it is from this socket, drop it, it must be forged. 921 * Don't bother responding if the destination was a broadcast. 922 */ 923 if (th->th_dport == th->th_sport) { 924 if (isipv6) { 925 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 926 &ip6->ip6_src)) 927 goto drop; 928 } else { 929 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 930 goto drop; 931 } 932 } 933 /* 934 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 935 * 936 * Note that it is quite possible to receive unicast 937 * link-layer packets with a broadcast IP address. Use 938 * in_broadcast() to find them. 939 */ 940 if (m->m_flags & (M_BCAST|M_MCAST)) 941 goto drop; 942 if (isipv6) { 943 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 944 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 945 goto drop; 946 } else { 947 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 948 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 949 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 950 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 951 goto drop; 952 } 953 /* 954 * SYN appears to be valid; create compressed TCP state 955 * for syncache, or perform t/tcp connection. 956 */ 957 if (so->so_qlen <= so->so_qlimit) { 958#ifdef TCPDEBUG 959 if (so->so_options & SO_DEBUG) 960 tcp_trace(TA_INPUT, ostate, tp, 961 (void *)tcp_saveipgen, &tcp_savetcp, 0); 962#endif 963 tcp_dooptions(tp, &to, optp, optlen, 1, th); 964 if (!syncache_add(&inc, &to, th, &so, m)) 965 goto drop; 966 if (so == NULL) { 967 /* 968 * Entry added to syncache, mbuf used to 969 * send SYN,ACK packet. 970 */ 971 KASSERT(headlocked, ("headlocked")); 972 INP_UNLOCK(inp); 973 INP_INFO_WUNLOCK(&tcbinfo); 974 return; 975 } 976 /* 977 * Segment passed TAO tests. 978 */ 979 INP_UNLOCK(inp); 980 inp = sotoinpcb(so); 981 INP_LOCK(inp); 982 tp = intotcpcb(inp); 983 tp->snd_wnd = tiwin; 984 tp->t_starttime = ticks; 985 tp->t_state = TCPS_ESTABLISHED; 986 987 /* 988 * T/TCP logic: 989 * If there is a FIN or if there is data, then 990 * delay SYN,ACK(SYN) in the hope of piggy-backing 991 * it on a response segment. Otherwise must send 992 * ACK now in case the other side is slow starting. 993 */ 994 if (thflags & TH_FIN || tlen != 0) 995 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 996 else 997 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 998 tcpstat.tcps_connects++; 999 soisconnected(so); 1000 goto trimthenstep6; 1001 } 1002 goto drop; 1003 } 1004after_listen: 1005 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1006 INP_LOCK_ASSERT(inp); 1007 1008 /* XXX temp debugging */ 1009 /* should not happen - syncache should pick up these connections */ 1010 if (tp->t_state == TCPS_LISTEN) 1011 panic("tcp_input: TCPS_LISTEN"); 1012 1013 /* 1014 * This is the second part of the MSS DoS prevention code (after 1015 * minmss on the sending side) and it deals with too many too small 1016 * tcp packets in a too short timeframe (1 second). 1017 * 1018 * For every full second we count the number of received packets 1019 * and bytes. If we get a lot of packets per second for this connection 1020 * (tcp_minmssoverload) we take a closer look at it and compute the 1021 * average packet size for the past second. If that is less than 1022 * tcp_minmss we get too many packets with very small payload which 1023 * is not good and burdens our system (and every packet generates 1024 * a wakeup to the process connected to our socket). We can reasonable 1025 * expect this to be small packet DoS attack to exhaust our CPU 1026 * cycles. 1027 * 1028 * Care has to be taken for the minimum packet overload value. This 1029 * value defines the minimum number of packets per second before we 1030 * start to worry. This must not be too low to avoid killing for 1031 * example interactive connections with many small packets like 1032 * telnet or SSH. 1033 * 1034 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1035 * this check. 1036 * 1037 * Account for packet if payload packet, skip over ACK, etc. 1038 */ 1039 if (tcp_minmss && tcp_minmssoverload && 1040 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1041 if (tp->rcv_second > ticks) { 1042 tp->rcv_pps++; 1043 tp->rcv_byps += tlen + off; 1044 if (tp->rcv_pps > tcp_minmssoverload) { 1045 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1046 printf("too many small tcp packets from " 1047 "%s:%u, av. %lubyte/packet, " 1048 "dropping connection\n", 1049#ifdef INET6 1050 isipv6 ? 1051 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1052#endif 1053 inet_ntoa(inp->inp_inc.inc_faddr), 1054 inp->inp_inc.inc_fport, 1055 tp->rcv_byps / tp->rcv_pps); 1056 tp = tcp_drop(tp, ECONNRESET); 1057 tcpstat.tcps_minmssdrops++; 1058 goto drop; 1059 } 1060 } 1061 } else { 1062 tp->rcv_second = ticks + hz; 1063 tp->rcv_pps = 1; 1064 tp->rcv_byps = tlen + off; 1065 } 1066 } 1067 1068 /* 1069 * Segment received on connection. 1070 * Reset idle time and keep-alive timer. 1071 */ 1072 tp->t_rcvtime = ticks; 1073 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1074 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1075 1076 /* 1077 * Process options only when we get SYN/ACK back. The SYN case 1078 * for incoming connections is handled in tcp_syncache. 1079 * XXX this is traditional behavior, may need to be cleaned up. 1080 */ 1081 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th); 1082 if (thflags & TH_SYN) { 1083 if (to.to_flags & TOF_SCALE) { 1084 tp->t_flags |= TF_RCVD_SCALE; 1085 tp->requested_s_scale = to.to_requested_s_scale; 1086 } 1087 if (to.to_flags & TOF_TS) { 1088 tp->t_flags |= TF_RCVD_TSTMP; 1089 tp->ts_recent = to.to_tsval; 1090 tp->ts_recent_age = ticks; 1091 } 1092 if (to.to_flags & TOF_MSS) 1093 tcp_mss(tp, to.to_mss); 1094 if (tp->sack_enable) { 1095 if (!(to.to_flags & TOF_SACK)) 1096 tp->sack_enable = 0; 1097 else 1098 tp->t_flags |= TF_SACK_PERMIT; 1099 } 1100 1101 } 1102 1103 if (tp->sack_enable) { 1104 /* Delete stale (cumulatively acked) SACK holes */ 1105 tcp_del_sackholes(tp, th); 1106 tp->rcv_laststart = th->th_seq; /* last recv'd segment*/ 1107 tp->rcv_lastend = th->th_seq + tlen; 1108 } 1109 1110 /* 1111 * Header prediction: check for the two common cases 1112 * of a uni-directional data xfer. If the packet has 1113 * no control flags, is in-sequence, the window didn't 1114 * change and we're not retransmitting, it's a 1115 * candidate. If the length is zero and the ack moved 1116 * forward, we're the sender side of the xfer. Just 1117 * free the data acked & wake any higher level process 1118 * that was blocked waiting for space. If the length 1119 * is non-zero and the ack didn't move, we're the 1120 * receiver side. If we're getting packets in-order 1121 * (the reassembly queue is empty), add the data to 1122 * the socket buffer and note that we need a delayed ack. 1123 * Make sure that the hidden state-flags are also off. 1124 * Since we check for TCPS_ESTABLISHED above, it can only 1125 * be TH_NEEDSYN. 1126 */ 1127 if (tp->t_state == TCPS_ESTABLISHED && 1128 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1129 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1130 ((to.to_flags & TOF_TS) == 0 || 1131 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1132 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1133 tp->snd_nxt == tp->snd_max) { 1134 1135 /* 1136 * If last ACK falls within this segment's sequence numbers, 1137 * record the timestamp. 1138 * NOTE that the test is modified according to the latest 1139 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1140 */ 1141 if ((to.to_flags & TOF_TS) != 0 && 1142 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1143 tp->ts_recent_age = ticks; 1144 tp->ts_recent = to.to_tsval; 1145 } 1146 1147 if (tlen == 0) { 1148 if (SEQ_GT(th->th_ack, tp->snd_una) && 1149 SEQ_LEQ(th->th_ack, tp->snd_max) && 1150 tp->snd_cwnd >= tp->snd_wnd && 1151 ((!tcp_do_newreno && !tp->sack_enable && 1152 tp->t_dupacks < tcprexmtthresh) || 1153 ((tcp_do_newreno || tp->sack_enable) && 1154 !IN_FASTRECOVERY(tp)))) { 1155 KASSERT(headlocked, ("headlocked")); 1156 /* 1157 * this is a pure ack for outstanding data. 1158 */ 1159 ++tcpstat.tcps_predack; 1160 /* 1161 * "bad retransmit" recovery 1162 */ 1163 if (tp->t_rxtshift == 1 && 1164 ticks < tp->t_badrxtwin) { 1165 ++tcpstat.tcps_sndrexmitbad; 1166 tp->snd_cwnd = tp->snd_cwnd_prev; 1167 tp->snd_ssthresh = 1168 tp->snd_ssthresh_prev; 1169 tp->snd_recover = tp->snd_recover_prev; 1170 if (tp->t_flags & TF_WASFRECOVERY) 1171 ENTER_FASTRECOVERY(tp); 1172 tp->snd_nxt = tp->snd_max; 1173 tp->t_badrxtwin = 0; 1174 } 1175 1176 /* 1177 * Recalculate the transmit timer / rtt. 1178 * 1179 * Some boxes send broken timestamp replies 1180 * during the SYN+ACK phase, ignore 1181 * timestamps of 0 or we could calculate a 1182 * huge RTT and blow up the retransmit timer. 1183 */ 1184 if ((to.to_flags & TOF_TS) != 0 && 1185 to.to_tsecr) { 1186 tcp_xmit_timer(tp, 1187 ticks - to.to_tsecr + 1); 1188 } else if (tp->t_rtttime && 1189 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1190 tcp_xmit_timer(tp, 1191 ticks - tp->t_rtttime); 1192 } 1193 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1194 acked = th->th_ack - tp->snd_una; 1195 tcpstat.tcps_rcvackpack++; 1196 tcpstat.tcps_rcvackbyte += acked; 1197 sbdrop(&so->so_snd, acked); 1198 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1199 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1200 tp->snd_recover = th->th_ack - 1; 1201 tp->snd_una = th->th_ack; 1202 /* 1203 * pull snd_wl2 up to prevent seq wrap relative 1204 * to th_ack. 1205 */ 1206 tp->snd_wl2 = th->th_ack; 1207 tp->t_dupacks = 0; 1208 m_freem(m); 1209 ND6_HINT(tp); /* some progress has been done */ 1210 1211 /* 1212 * If all outstanding data are acked, stop 1213 * retransmit timer, otherwise restart timer 1214 * using current (possibly backed-off) value. 1215 * If process is waiting for space, 1216 * wakeup/selwakeup/signal. If data 1217 * are ready to send, let tcp_output 1218 * decide between more output or persist. 1219 1220#ifdef TCPDEBUG 1221 if (so->so_options & SO_DEBUG) 1222 tcp_trace(TA_INPUT, ostate, tp, 1223 (void *)tcp_saveipgen, 1224 &tcp_savetcp, 0); 1225#endif 1226 */ 1227 if (tp->snd_una == tp->snd_max) 1228 callout_stop(tp->tt_rexmt); 1229 else if (!callout_active(tp->tt_persist)) 1230 callout_reset(tp->tt_rexmt, 1231 tp->t_rxtcur, 1232 tcp_timer_rexmt, tp); 1233 1234 sowwakeup(so); 1235 if (so->so_snd.sb_cc) 1236 (void) tcp_output(tp); 1237 goto check_delack; 1238 } 1239 } else if (th->th_ack == tp->snd_una && 1240 LIST_EMPTY(&tp->t_segq) && 1241 tlen <= sbspace(&so->so_rcv)) { 1242 KASSERT(headlocked, ("headlocked")); 1243 /* 1244 * this is a pure, in-sequence data packet 1245 * with nothing on the reassembly queue and 1246 * we have enough buffer space to take it. 1247 */ 1248 /* Clean receiver SACK report if present */ 1249 if (tp->sack_enable && tp->rcv_numsacks) 1250 tcp_clean_sackreport(tp); 1251 ++tcpstat.tcps_preddat; 1252 tp->rcv_nxt += tlen; 1253 /* 1254 * Pull snd_wl1 up to prevent seq wrap relative to 1255 * th_seq. 1256 */ 1257 tp->snd_wl1 = th->th_seq; 1258 /* 1259 * Pull rcv_up up to prevent seq wrap relative to 1260 * rcv_nxt. 1261 */ 1262 tp->rcv_up = tp->rcv_nxt; 1263 tcpstat.tcps_rcvpack++; 1264 tcpstat.tcps_rcvbyte += tlen; 1265 ND6_HINT(tp); /* some progress has been done */ 1266 /* 1267#ifdef TCPDEBUG 1268 if (so->so_options & SO_DEBUG) 1269 tcp_trace(TA_INPUT, ostate, tp, 1270 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1271#endif 1272 * Add data to socket buffer. 1273 */ 1274 SOCKBUF_LOCK(&so->so_rcv); 1275 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1276 m_freem(m); 1277 } else { 1278 m_adj(m, drop_hdrlen); /* delayed header drop */ 1279 sbappendstream_locked(&so->so_rcv, m); 1280 } 1281 sorwakeup_locked(so); 1282 if (DELAY_ACK(tp)) { 1283 tp->t_flags |= TF_DELACK; 1284 } else { 1285 tp->t_flags |= TF_ACKNOW; 1286 tcp_output(tp); 1287 } 1288 goto check_delack; 1289 } 1290 } 1291 1292 /* 1293 * Calculate amount of space in receive window, 1294 * and then do TCP input processing. 1295 * Receive window is amount of space in rcv queue, 1296 * but not less than advertised window. 1297 */ 1298 { int win; 1299 1300 win = sbspace(&so->so_rcv); 1301 if (win < 0) 1302 win = 0; 1303 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1304 } 1305 1306 switch (tp->t_state) { 1307 1308 /* 1309 * If the state is SYN_RECEIVED: 1310 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1311 */ 1312 case TCPS_SYN_RECEIVED: 1313 if ((thflags & TH_ACK) && 1314 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1315 SEQ_GT(th->th_ack, tp->snd_max))) { 1316 rstreason = BANDLIM_RST_OPENPORT; 1317 goto dropwithreset; 1318 } 1319 break; 1320 1321 /* 1322 * If the state is SYN_SENT: 1323 * if seg contains an ACK, but not for our SYN, drop the input. 1324 * if seg contains a RST, then drop the connection. 1325 * if seg does not contain SYN, then drop it. 1326 * Otherwise this is an acceptable SYN segment 1327 * initialize tp->rcv_nxt and tp->irs 1328 * if seg contains ack then advance tp->snd_una 1329 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1330 * arrange for segment to be acked (eventually) 1331 * continue processing rest of data/controls, beginning with URG 1332 */ 1333 case TCPS_SYN_SENT: 1334 if ((thflags & TH_ACK) && 1335 (SEQ_LEQ(th->th_ack, tp->iss) || 1336 SEQ_GT(th->th_ack, tp->snd_max))) { 1337 rstreason = BANDLIM_UNLIMITED; 1338 goto dropwithreset; 1339 } 1340 if (thflags & TH_RST) { 1341 if (thflags & TH_ACK) 1342 tp = tcp_drop(tp, ECONNREFUSED); 1343 goto drop; 1344 } 1345 if ((thflags & TH_SYN) == 0) 1346 goto drop; 1347 tp->snd_wnd = th->th_win; /* initial send window */ 1348 1349 tp->irs = th->th_seq; 1350 tcp_rcvseqinit(tp); 1351 if (thflags & TH_ACK) { 1352 tcpstat.tcps_connects++; 1353 soisconnected(so); 1354#ifdef MAC 1355 SOCK_LOCK(so); 1356 mac_set_socket_peer_from_mbuf(m, so); 1357 SOCK_UNLOCK(so); 1358#endif 1359 /* Do window scaling on this connection? */ 1360 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1361 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1362 tp->snd_scale = tp->requested_s_scale; 1363 tp->rcv_scale = tp->request_r_scale; 1364 } 1365 tp->rcv_adv += tp->rcv_wnd; 1366 tp->snd_una++; /* SYN is acked */ 1367 /* 1368 * If there's data, delay ACK; if there's also a FIN 1369 * ACKNOW will be turned on later. 1370 */ 1371 if (DELAY_ACK(tp) && tlen != 0) 1372 callout_reset(tp->tt_delack, tcp_delacktime, 1373 tcp_timer_delack, tp); 1374 else 1375 tp->t_flags |= TF_ACKNOW; 1376 /* 1377 * Received <SYN,ACK> in SYN_SENT[*] state. 1378 * Transitions: 1379 * SYN_SENT --> ESTABLISHED 1380 * SYN_SENT* --> FIN_WAIT_1 1381 */ 1382 tp->t_starttime = ticks; 1383 if (tp->t_flags & TF_NEEDFIN) { 1384 tp->t_state = TCPS_FIN_WAIT_1; 1385 tp->t_flags &= ~TF_NEEDFIN; 1386 thflags &= ~TH_SYN; 1387 } else { 1388 tp->t_state = TCPS_ESTABLISHED; 1389 callout_reset(tp->tt_keep, tcp_keepidle, 1390 tcp_timer_keep, tp); 1391 } 1392 } else { 1393 /* 1394 * Received initial SYN in SYN-SENT[*] state => 1395 * simultaneous open. If segment contains CC option 1396 * and there is a cached CC, apply TAO test. 1397 * If it succeeds, connection is * half-synchronized. 1398 * Otherwise, do 3-way handshake: 1399 * SYN-SENT -> SYN-RECEIVED 1400 * SYN-SENT* -> SYN-RECEIVED* 1401 * If there was no CC option, clear cached CC value. 1402 */ 1403 tp->t_flags |= TF_ACKNOW; 1404 callout_stop(tp->tt_rexmt); 1405 tp->t_state = TCPS_SYN_RECEIVED; 1406 } 1407 1408trimthenstep6: 1409 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1410 "locked")); 1411 INP_LOCK_ASSERT(inp); 1412 1413 /* 1414 * Advance th->th_seq to correspond to first data byte. 1415 * If data, trim to stay within window, 1416 * dropping FIN if necessary. 1417 */ 1418 th->th_seq++; 1419 if (tlen > tp->rcv_wnd) { 1420 todrop = tlen - tp->rcv_wnd; 1421 m_adj(m, -todrop); 1422 tlen = tp->rcv_wnd; 1423 thflags &= ~TH_FIN; 1424 tcpstat.tcps_rcvpackafterwin++; 1425 tcpstat.tcps_rcvbyteafterwin += todrop; 1426 } 1427 tp->snd_wl1 = th->th_seq - 1; 1428 tp->rcv_up = th->th_seq; 1429 /* 1430 * Client side of transaction: already sent SYN and data. 1431 * If the remote host used T/TCP to validate the SYN, 1432 * our data will be ACK'd; if so, enter normal data segment 1433 * processing in the middle of step 5, ack processing. 1434 * Otherwise, goto step 6. 1435 */ 1436 if (thflags & TH_ACK) 1437 goto process_ACK; 1438 1439 goto step6; 1440 1441 /* 1442 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1443 * do normal processing. 1444 * 1445 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1446 */ 1447 case TCPS_LAST_ACK: 1448 case TCPS_CLOSING: 1449 case TCPS_TIME_WAIT: 1450 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1451 break; /* continue normal processing */ 1452 } 1453 1454 /* 1455 * States other than LISTEN or SYN_SENT. 1456 * First check the RST flag and sequence number since reset segments 1457 * are exempt from the timestamp and connection count tests. This 1458 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1459 * below which allowed reset segments in half the sequence space 1460 * to fall though and be processed (which gives forged reset 1461 * segments with a random sequence number a 50 percent chance of 1462 * killing a connection). 1463 * Then check timestamp, if present. 1464 * Then check the connection count, if present. 1465 * Then check that at least some bytes of segment are within 1466 * receive window. If segment begins before rcv_nxt, 1467 * drop leading data (and SYN); if nothing left, just ack. 1468 * 1469 * 1470 * If the RST bit is set, check the sequence number to see 1471 * if this is a valid reset segment. 1472 * RFC 793 page 37: 1473 * In all states except SYN-SENT, all reset (RST) segments 1474 * are validated by checking their SEQ-fields. A reset is 1475 * valid if its sequence number is in the window. 1476 * Note: this does not take into account delayed ACKs, so 1477 * we should test against last_ack_sent instead of rcv_nxt. 1478 * The sequence number in the reset segment is normally an 1479 * echo of our outgoing acknowlegement numbers, but some hosts 1480 * send a reset with the sequence number at the rightmost edge 1481 * of our receive window, and we have to handle this case. 1482 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1483 * that brute force RST attacks are possible. To combat this, 1484 * we use a much stricter check while in the ESTABLISHED state, 1485 * only accepting RSTs where the sequence number is equal to 1486 * last_ack_sent. In all other states (the states in which a 1487 * RST is more likely), the more permissive check is used. 1488 * If we have multiple segments in flight, the intial reset 1489 * segment sequence numbers will be to the left of last_ack_sent, 1490 * but they will eventually catch up. 1491 * In any case, it never made sense to trim reset segments to 1492 * fit the receive window since RFC 1122 says: 1493 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1494 * 1495 * A TCP SHOULD allow a received RST segment to include data. 1496 * 1497 * DISCUSSION 1498 * It has been suggested that a RST segment could contain 1499 * ASCII text that encoded and explained the cause of the 1500 * RST. No standard has yet been established for such 1501 * data. 1502 * 1503 * If the reset segment passes the sequence number test examine 1504 * the state: 1505 * SYN_RECEIVED STATE: 1506 * If passive open, return to LISTEN state. 1507 * If active open, inform user that connection was refused. 1508 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1509 * Inform user that connection was reset, and close tcb. 1510 * CLOSING, LAST_ACK STATES: 1511 * Close the tcb. 1512 * TIME_WAIT STATE: 1513 * Drop the segment - see Stevens, vol. 2, p. 964 and 1514 * RFC 1337. 1515 */ 1516 if (thflags & TH_RST) { 1517 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1518 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1519 switch (tp->t_state) { 1520 1521 case TCPS_SYN_RECEIVED: 1522 so->so_error = ECONNREFUSED; 1523 goto close; 1524 1525 case TCPS_ESTABLISHED: 1526 if (tp->last_ack_sent != th->th_seq) { 1527 tcpstat.tcps_badrst++; 1528 goto drop; 1529 } 1530 case TCPS_FIN_WAIT_1: 1531 case TCPS_FIN_WAIT_2: 1532 case TCPS_CLOSE_WAIT: 1533 so->so_error = ECONNRESET; 1534 close: 1535 tp->t_state = TCPS_CLOSED; 1536 tcpstat.tcps_drops++; 1537 tp = tcp_close(tp); 1538 break; 1539 1540 case TCPS_CLOSING: 1541 case TCPS_LAST_ACK: 1542 tp = tcp_close(tp); 1543 break; 1544 1545 case TCPS_TIME_WAIT: 1546 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1547 ("timewait")); 1548 break; 1549 } 1550 } 1551 goto drop; 1552 } 1553 1554 /* 1555 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1556 * and it's less than ts_recent, drop it. 1557 */ 1558 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1559 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1560 1561 /* Check to see if ts_recent is over 24 days old. */ 1562 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1563 /* 1564 * Invalidate ts_recent. If this segment updates 1565 * ts_recent, the age will be reset later and ts_recent 1566 * will get a valid value. If it does not, setting 1567 * ts_recent to zero will at least satisfy the 1568 * requirement that zero be placed in the timestamp 1569 * echo reply when ts_recent isn't valid. The 1570 * age isn't reset until we get a valid ts_recent 1571 * because we don't want out-of-order segments to be 1572 * dropped when ts_recent is old. 1573 */ 1574 tp->ts_recent = 0; 1575 } else { 1576 tcpstat.tcps_rcvduppack++; 1577 tcpstat.tcps_rcvdupbyte += tlen; 1578 tcpstat.tcps_pawsdrop++; 1579 if (tlen) 1580 goto dropafterack; 1581 goto drop; 1582 } 1583 } 1584 1585 /* 1586 * In the SYN-RECEIVED state, validate that the packet belongs to 1587 * this connection before trimming the data to fit the receive 1588 * window. Check the sequence number versus IRS since we know 1589 * the sequence numbers haven't wrapped. This is a partial fix 1590 * for the "LAND" DoS attack. 1591 */ 1592 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1593 rstreason = BANDLIM_RST_OPENPORT; 1594 goto dropwithreset; 1595 } 1596 1597 todrop = tp->rcv_nxt - th->th_seq; 1598 if (todrop > 0) { 1599 if (thflags & TH_SYN) { 1600 thflags &= ~TH_SYN; 1601 th->th_seq++; 1602 if (th->th_urp > 1) 1603 th->th_urp--; 1604 else 1605 thflags &= ~TH_URG; 1606 todrop--; 1607 } 1608 /* 1609 * Following if statement from Stevens, vol. 2, p. 960. 1610 */ 1611 if (todrop > tlen 1612 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1613 /* 1614 * Any valid FIN must be to the left of the window. 1615 * At this point the FIN must be a duplicate or out 1616 * of sequence; drop it. 1617 */ 1618 thflags &= ~TH_FIN; 1619 1620 /* 1621 * Send an ACK to resynchronize and drop any data. 1622 * But keep on processing for RST or ACK. 1623 */ 1624 tp->t_flags |= TF_ACKNOW; 1625 todrop = tlen; 1626 tcpstat.tcps_rcvduppack++; 1627 tcpstat.tcps_rcvdupbyte += todrop; 1628 } else { 1629 tcpstat.tcps_rcvpartduppack++; 1630 tcpstat.tcps_rcvpartdupbyte += todrop; 1631 } 1632 drop_hdrlen += todrop; /* drop from the top afterwards */ 1633 th->th_seq += todrop; 1634 tlen -= todrop; 1635 if (th->th_urp > todrop) 1636 th->th_urp -= todrop; 1637 else { 1638 thflags &= ~TH_URG; 1639 th->th_urp = 0; 1640 } 1641 } 1642 1643 /* 1644 * If new data are received on a connection after the 1645 * user processes are gone, then RST the other end. 1646 */ 1647 if ((so->so_state & SS_NOFDREF) && 1648 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1649 tp = tcp_close(tp); 1650 tcpstat.tcps_rcvafterclose++; 1651 rstreason = BANDLIM_UNLIMITED; 1652 goto dropwithreset; 1653 } 1654 1655 /* 1656 * If segment ends after window, drop trailing data 1657 * (and PUSH and FIN); if nothing left, just ACK. 1658 */ 1659 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1660 if (todrop > 0) { 1661 tcpstat.tcps_rcvpackafterwin++; 1662 if (todrop >= tlen) { 1663 tcpstat.tcps_rcvbyteafterwin += tlen; 1664 /* 1665 * If a new connection request is received 1666 * while in TIME_WAIT, drop the old connection 1667 * and start over if the sequence numbers 1668 * are above the previous ones. 1669 */ 1670 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1671 if (thflags & TH_SYN && 1672 tp->t_state == TCPS_TIME_WAIT && 1673 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1674 tp = tcp_close(tp); 1675 goto findpcb; 1676 } 1677 /* 1678 * If window is closed can only take segments at 1679 * window edge, and have to drop data and PUSH from 1680 * incoming segments. Continue processing, but 1681 * remember to ack. Otherwise, drop segment 1682 * and ack. 1683 */ 1684 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1685 tp->t_flags |= TF_ACKNOW; 1686 tcpstat.tcps_rcvwinprobe++; 1687 } else 1688 goto dropafterack; 1689 } else 1690 tcpstat.tcps_rcvbyteafterwin += todrop; 1691 m_adj(m, -todrop); 1692 tlen -= todrop; 1693 thflags &= ~(TH_PUSH|TH_FIN); 1694 } 1695 1696 /* 1697 * If last ACK falls within this segment's sequence numbers, 1698 * record its timestamp. 1699 * NOTE that the test is modified according to the latest 1700 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1701 */ 1702 if ((to.to_flags & TOF_TS) != 0 && 1703 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1704 tp->ts_recent_age = ticks; 1705 tp->ts_recent = to.to_tsval; 1706 } 1707 1708 /* 1709 * If a SYN is in the window, then this is an 1710 * error and we send an RST and drop the connection. 1711 */ 1712 if (thflags & TH_SYN) { 1713 tp = tcp_drop(tp, ECONNRESET); 1714 rstreason = BANDLIM_UNLIMITED; 1715 goto drop; 1716 } 1717 1718 /* 1719 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1720 * flag is on (half-synchronized state), then queue data for 1721 * later processing; else drop segment and return. 1722 */ 1723 if ((thflags & TH_ACK) == 0) { 1724 if (tp->t_state == TCPS_SYN_RECEIVED || 1725 (tp->t_flags & TF_NEEDSYN)) 1726 goto step6; 1727 else 1728 goto drop; 1729 } 1730 1731 /* 1732 * Ack processing. 1733 */ 1734 switch (tp->t_state) { 1735 1736 /* 1737 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1738 * ESTABLISHED state and continue processing. 1739 * The ACK was checked above. 1740 */ 1741 case TCPS_SYN_RECEIVED: 1742 1743 tcpstat.tcps_connects++; 1744 soisconnected(so); 1745 /* Do window scaling? */ 1746 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1747 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1748 tp->snd_scale = tp->requested_s_scale; 1749 tp->rcv_scale = tp->request_r_scale; 1750 } 1751 /* 1752 * Make transitions: 1753 * SYN-RECEIVED -> ESTABLISHED 1754 * SYN-RECEIVED* -> FIN-WAIT-1 1755 */ 1756 tp->t_starttime = ticks; 1757 if (tp->t_flags & TF_NEEDFIN) { 1758 tp->t_state = TCPS_FIN_WAIT_1; 1759 tp->t_flags &= ~TF_NEEDFIN; 1760 } else { 1761 tp->t_state = TCPS_ESTABLISHED; 1762 callout_reset(tp->tt_keep, tcp_keepidle, 1763 tcp_timer_keep, tp); 1764 } 1765 /* 1766 * If segment contains data or ACK, will call tcp_reass() 1767 * later; if not, do so now to pass queued data to user. 1768 */ 1769 if (tlen == 0 && (thflags & TH_FIN) == 0) 1770 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1771 (struct mbuf *)0); 1772 tp->snd_wl1 = th->th_seq - 1; 1773 /* FALLTHROUGH */ 1774 1775 /* 1776 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1777 * ACKs. If the ack is in the range 1778 * tp->snd_una < th->th_ack <= tp->snd_max 1779 * then advance tp->snd_una to th->th_ack and drop 1780 * data from the retransmission queue. If this ACK reflects 1781 * more up to date window information we update our window information. 1782 */ 1783 case TCPS_ESTABLISHED: 1784 case TCPS_FIN_WAIT_1: 1785 case TCPS_FIN_WAIT_2: 1786 case TCPS_CLOSE_WAIT: 1787 case TCPS_CLOSING: 1788 case TCPS_LAST_ACK: 1789 case TCPS_TIME_WAIT: 1790 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1791 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1792 if (tlen == 0 && tiwin == tp->snd_wnd) { 1793 tcpstat.tcps_rcvdupack++; 1794 /* 1795 * If we have outstanding data (other than 1796 * a window probe), this is a completely 1797 * duplicate ack (ie, window info didn't 1798 * change), the ack is the biggest we've 1799 * seen and we've seen exactly our rexmt 1800 * threshhold of them, assume a packet 1801 * has been dropped and retransmit it. 1802 * Kludge snd_nxt & the congestion 1803 * window so we send only this one 1804 * packet. 1805 * 1806 * We know we're losing at the current 1807 * window size so do congestion avoidance 1808 * (set ssthresh to half the current window 1809 * and pull our congestion window back to 1810 * the new ssthresh). 1811 * 1812 * Dup acks mean that packets have left the 1813 * network (they're now cached at the receiver) 1814 * so bump cwnd by the amount in the receiver 1815 * to keep a constant cwnd packets in the 1816 * network. 1817 */ 1818 if (!callout_active(tp->tt_rexmt) || 1819 th->th_ack != tp->snd_una) 1820 tp->t_dupacks = 0; 1821 else if (++tp->t_dupacks > tcprexmtthresh || 1822 ((tcp_do_newreno || tp->sack_enable) && 1823 IN_FASTRECOVERY(tp))) { 1824 tp->snd_cwnd += tp->t_maxseg; 1825 (void) tcp_output(tp); 1826 goto drop; 1827 } else if (tp->t_dupacks == tcprexmtthresh) { 1828 tcp_seq onxt = tp->snd_nxt; 1829 u_int win; 1830 1831 /* 1832 * If we're doing sack, check to 1833 * see if we're already in sack 1834 * recovery. If we're not doing sack, 1835 * check to see if we're in newreno 1836 * recovery. 1837 */ 1838 if (tp->sack_enable) { 1839 if (IN_FASTRECOVERY(tp)) { 1840 tp->t_dupacks = 0; 1841 break; 1842 } 1843 } else if (tcp_do_newreno) { 1844 if (SEQ_LEQ(th->th_ack, 1845 tp->snd_recover)) { 1846 tp->t_dupacks = 0; 1847 break; 1848 } 1849 } 1850 win = min(tp->snd_wnd, tp->snd_cwnd) / 1851 2 / tp->t_maxseg; 1852 if (win < 2) 1853 win = 2; 1854 tp->snd_ssthresh = win * tp->t_maxseg; 1855 ENTER_FASTRECOVERY(tp); 1856 tp->snd_recover = tp->snd_max; 1857 callout_stop(tp->tt_rexmt); 1858 tp->t_rtttime = 0; 1859 if (tp->sack_enable) { 1860 tcpstat.tcps_sack_recovery_episode++; 1861 tp->sack_newdata = tp->snd_nxt; 1862 tp->snd_cwnd = 1863 tp->t_maxseg * tcp_sack_recovery_initburst; 1864 (void) tcp_output(tp); 1865 tp->snd_cwnd += 1866 tp->snd_ssthresh; 1867 goto drop; 1868 } 1869 1870 tp->snd_nxt = th->th_ack; 1871 tp->snd_cwnd = tp->t_maxseg; 1872 (void) tcp_output(tp); 1873 KASSERT(tp->snd_limited <= 2, 1874 ("tp->snd_limited too big")); 1875 tp->snd_cwnd = tp->snd_ssthresh + 1876 tp->t_maxseg * 1877 (tp->t_dupacks - tp->snd_limited); 1878 if (SEQ_GT(onxt, tp->snd_nxt)) 1879 tp->snd_nxt = onxt; 1880 goto drop; 1881 } else if (tcp_do_rfc3042) { 1882 u_long oldcwnd = tp->snd_cwnd; 1883 tcp_seq oldsndmax = tp->snd_max; 1884 u_int sent; 1885 1886 KASSERT(tp->t_dupacks == 1 || 1887 tp->t_dupacks == 2, 1888 ("dupacks not 1 or 2")); 1889 if (tp->t_dupacks == 1) 1890 tp->snd_limited = 0; 1891 tp->snd_cwnd = 1892 (tp->snd_nxt - tp->snd_una) + 1893 (tp->t_dupacks - tp->snd_limited) * 1894 tp->t_maxseg; 1895 (void) tcp_output(tp); 1896 sent = tp->snd_max - oldsndmax; 1897 if (sent > tp->t_maxseg) { 1898 KASSERT((tp->t_dupacks == 2 && 1899 tp->snd_limited == 0) || 1900 (sent == tp->t_maxseg + 1 && 1901 tp->t_flags & TF_SENTFIN), 1902 ("sent too much")); 1903 tp->snd_limited = 2; 1904 } else if (sent > 0) 1905 ++tp->snd_limited; 1906 tp->snd_cwnd = oldcwnd; 1907 goto drop; 1908 } 1909 } else 1910 tp->t_dupacks = 0; 1911 break; 1912 } 1913 1914 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1915 1916 /* 1917 * If the congestion window was inflated to account 1918 * for the other side's cached packets, retract it. 1919 */ 1920 if (tcp_do_newreno || tp->sack_enable) { 1921 if (IN_FASTRECOVERY(tp)) { 1922 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1923 if (tp->sack_enable) 1924 tcp_sack_partialack(tp, th); 1925 else 1926 tcp_newreno_partial_ack(tp, th); 1927 } else { 1928 /* 1929 * Out of fast recovery. 1930 * Window inflation should have left us 1931 * with approximately snd_ssthresh 1932 * outstanding data. 1933 * But in case we would be inclined to 1934 * send a burst, better to do it via 1935 * the slow start mechanism. 1936 */ 1937 if (SEQ_GT(th->th_ack + 1938 tp->snd_ssthresh, 1939 tp->snd_max)) 1940 tp->snd_cwnd = tp->snd_max - 1941 th->th_ack + 1942 tp->t_maxseg; 1943 else 1944 tp->snd_cwnd = tp->snd_ssthresh; 1945 } 1946 } 1947 } else { 1948 if (tp->t_dupacks >= tcprexmtthresh && 1949 tp->snd_cwnd > tp->snd_ssthresh) 1950 tp->snd_cwnd = tp->snd_ssthresh; 1951 } 1952 tp->t_dupacks = 0; 1953 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1954 tcpstat.tcps_rcvacktoomuch++; 1955 goto dropafterack; 1956 } 1957 /* 1958 * If we reach this point, ACK is not a duplicate, 1959 * i.e., it ACKs something we sent. 1960 */ 1961 if (tp->t_flags & TF_NEEDSYN) { 1962 /* 1963 * T/TCP: Connection was half-synchronized, and our 1964 * SYN has been ACK'd (so connection is now fully 1965 * synchronized). Go to non-starred state, 1966 * increment snd_una for ACK of SYN, and check if 1967 * we can do window scaling. 1968 */ 1969 tp->t_flags &= ~TF_NEEDSYN; 1970 tp->snd_una++; 1971 /* Do window scaling? */ 1972 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1973 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1974 tp->snd_scale = tp->requested_s_scale; 1975 tp->rcv_scale = tp->request_r_scale; 1976 } 1977 } 1978 1979process_ACK: 1980 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 1981 "locked")); 1982 INP_LOCK_ASSERT(inp); 1983 1984 acked = th->th_ack - tp->snd_una; 1985 tcpstat.tcps_rcvackpack++; 1986 tcpstat.tcps_rcvackbyte += acked; 1987 1988 /* 1989 * If we just performed our first retransmit, and the ACK 1990 * arrives within our recovery window, then it was a mistake 1991 * to do the retransmit in the first place. Recover our 1992 * original cwnd and ssthresh, and proceed to transmit where 1993 * we left off. 1994 */ 1995 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1996 ++tcpstat.tcps_sndrexmitbad; 1997 tp->snd_cwnd = tp->snd_cwnd_prev; 1998 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1999 tp->snd_recover = tp->snd_recover_prev; 2000 if (tp->t_flags & TF_WASFRECOVERY) 2001 ENTER_FASTRECOVERY(tp); 2002 tp->snd_nxt = tp->snd_max; 2003 tp->t_badrxtwin = 0; /* XXX probably not required */ 2004 } 2005 2006 /* 2007 * If we have a timestamp reply, update smoothed 2008 * round trip time. If no timestamp is present but 2009 * transmit timer is running and timed sequence 2010 * number was acked, update smoothed round trip time. 2011 * Since we now have an rtt measurement, cancel the 2012 * timer backoff (cf., Phil Karn's retransmit alg.). 2013 * Recompute the initial retransmit timer. 2014 * 2015 * Some boxes send broken timestamp replies 2016 * during the SYN+ACK phase, ignore 2017 * timestamps of 0 or we could calculate a 2018 * huge RTT and blow up the retransmit timer. 2019 */ 2020 if ((to.to_flags & TOF_TS) != 0 && 2021 to.to_tsecr) { 2022 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2023 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2024 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2025 } 2026 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2027 2028 /* 2029 * If all outstanding data is acked, stop retransmit 2030 * timer and remember to restart (more output or persist). 2031 * If there is more data to be acked, restart retransmit 2032 * timer, using current (possibly backed-off) value. 2033 */ 2034 if (th->th_ack == tp->snd_max) { 2035 callout_stop(tp->tt_rexmt); 2036 needoutput = 1; 2037 } else if (!callout_active(tp->tt_persist)) 2038 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2039 tcp_timer_rexmt, tp); 2040 2041 /* 2042 * If no data (only SYN) was ACK'd, 2043 * skip rest of ACK processing. 2044 */ 2045 if (acked == 0) 2046 goto step6; 2047 2048 /* 2049 * When new data is acked, open the congestion window. 2050 * If the window gives us less than ssthresh packets 2051 * in flight, open exponentially (maxseg per packet). 2052 * Otherwise open linearly: maxseg per window 2053 * (maxseg^2 / cwnd per packet). 2054 */ 2055 if ((!tcp_do_newreno && !tp->sack_enable) || 2056 !IN_FASTRECOVERY(tp)) { 2057 register u_int cw = tp->snd_cwnd; 2058 register u_int incr = tp->t_maxseg; 2059 if (cw > tp->snd_ssthresh) 2060 incr = incr * incr / cw; 2061 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2062 } 2063 SOCKBUF_LOCK(&so->so_snd); 2064 if (acked > so->so_snd.sb_cc) { 2065 tp->snd_wnd -= so->so_snd.sb_cc; 2066 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2067 ourfinisacked = 1; 2068 } else { 2069 sbdrop_locked(&so->so_snd, acked); 2070 tp->snd_wnd -= acked; 2071 ourfinisacked = 0; 2072 } 2073 sowwakeup_locked(so); 2074 /* detect una wraparound */ 2075 if ((tcp_do_newreno || tp->sack_enable) && 2076 !IN_FASTRECOVERY(tp) && 2077 SEQ_GT(tp->snd_una, tp->snd_recover) && 2078 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2079 tp->snd_recover = th->th_ack - 1; 2080 if ((tcp_do_newreno || tp->sack_enable) && 2081 IN_FASTRECOVERY(tp) && 2082 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2083 EXIT_FASTRECOVERY(tp); 2084 tp->snd_una = th->th_ack; 2085 if (tp->sack_enable) { 2086 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2087 tp->snd_recover = tp->snd_una; 2088 } 2089 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2090 tp->snd_nxt = tp->snd_una; 2091 2092 switch (tp->t_state) { 2093 2094 /* 2095 * In FIN_WAIT_1 STATE in addition to the processing 2096 * for the ESTABLISHED state if our FIN is now acknowledged 2097 * then enter FIN_WAIT_2. 2098 */ 2099 case TCPS_FIN_WAIT_1: 2100 if (ourfinisacked) { 2101 /* 2102 * If we can't receive any more 2103 * data, then closing user can proceed. 2104 * Starting the timer is contrary to the 2105 * specification, but if we don't get a FIN 2106 * we'll hang forever. 2107 */ 2108 /* XXXjl 2109 * we should release the tp also, and use a 2110 * compressed state. 2111 */ 2112 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2113 soisdisconnected(so); 2114 callout_reset(tp->tt_2msl, tcp_maxidle, 2115 tcp_timer_2msl, tp); 2116 } 2117 tp->t_state = TCPS_FIN_WAIT_2; 2118 } 2119 break; 2120 2121 /* 2122 * In CLOSING STATE in addition to the processing for 2123 * the ESTABLISHED state if the ACK acknowledges our FIN 2124 * then enter the TIME-WAIT state, otherwise ignore 2125 * the segment. 2126 */ 2127 case TCPS_CLOSING: 2128 if (ourfinisacked) { 2129 KASSERT(headlocked, ("tcp_input: process_ACK: " 2130 "head not locked")); 2131 tcp_twstart(tp); 2132 INP_INFO_WUNLOCK(&tcbinfo); 2133 m_freem(m); 2134 return; 2135 } 2136 break; 2137 2138 /* 2139 * In LAST_ACK, we may still be waiting for data to drain 2140 * and/or to be acked, as well as for the ack of our FIN. 2141 * If our FIN is now acknowledged, delete the TCB, 2142 * enter the closed state and return. 2143 */ 2144 case TCPS_LAST_ACK: 2145 if (ourfinisacked) { 2146 tp = tcp_close(tp); 2147 goto drop; 2148 } 2149 break; 2150 2151 /* 2152 * In TIME_WAIT state the only thing that should arrive 2153 * is a retransmission of the remote FIN. Acknowledge 2154 * it and restart the finack timer. 2155 */ 2156 case TCPS_TIME_WAIT: 2157 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2158 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2159 tcp_timer_2msl, tp); 2160 goto dropafterack; 2161 } 2162 } 2163 2164step6: 2165 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2166 INP_LOCK_ASSERT(inp); 2167 2168 /* 2169 * Update window information. 2170 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2171 */ 2172 if ((thflags & TH_ACK) && 2173 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2174 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2175 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2176 /* keep track of pure window updates */ 2177 if (tlen == 0 && 2178 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2179 tcpstat.tcps_rcvwinupd++; 2180 tp->snd_wnd = tiwin; 2181 tp->snd_wl1 = th->th_seq; 2182 tp->snd_wl2 = th->th_ack; 2183 if (tp->snd_wnd > tp->max_sndwnd) 2184 tp->max_sndwnd = tp->snd_wnd; 2185 needoutput = 1; 2186 } 2187 2188 /* 2189 * Process segments with URG. 2190 */ 2191 if ((thflags & TH_URG) && th->th_urp && 2192 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2193 /* 2194 * This is a kludge, but if we receive and accept 2195 * random urgent pointers, we'll crash in 2196 * soreceive. It's hard to imagine someone 2197 * actually wanting to send this much urgent data. 2198 */ 2199 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2200 th->th_urp = 0; /* XXX */ 2201 thflags &= ~TH_URG; /* XXX */ 2202 goto dodata; /* XXX */ 2203 } 2204 /* 2205 * If this segment advances the known urgent pointer, 2206 * then mark the data stream. This should not happen 2207 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2208 * a FIN has been received from the remote side. 2209 * In these states we ignore the URG. 2210 * 2211 * According to RFC961 (Assigned Protocols), 2212 * the urgent pointer points to the last octet 2213 * of urgent data. We continue, however, 2214 * to consider it to indicate the first octet 2215 * of data past the urgent section as the original 2216 * spec states (in one of two places). 2217 */ 2218 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2219 tp->rcv_up = th->th_seq + th->th_urp; 2220 SOCKBUF_LOCK(&so->so_rcv); 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 SOCKBUF_UNLOCK(&so->so_rcv); 2226 sohasoutofband(so); 2227 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2228 } 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); 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#ifdef TCPDEBUG 2375 if (so->so_options & SO_DEBUG) 2376 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2377 &tcp_savetcp, 0); 2378#endif 2379 2380 /* 2381 * Return any desired output. 2382 */ 2383 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2384 (void) tcp_output(tp); 2385 2386check_delack: 2387 KASSERT(headlocked == 1, ("tcp_input: check_delack: head not locked")); 2388 INP_LOCK_ASSERT(inp); 2389 if (tp->t_flags & TF_DELACK) { 2390 tp->t_flags &= ~TF_DELACK; 2391 callout_reset(tp->tt_delack, tcp_delacktime, 2392 tcp_timer_delack, tp); 2393 } 2394 INP_UNLOCK(inp); 2395 INP_INFO_WUNLOCK(&tcbinfo); 2396 return; 2397 2398dropafterack: 2399 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2400 /* 2401 * Generate an ACK dropping incoming segment if it occupies 2402 * sequence space, where the ACK reflects our state. 2403 * 2404 * We can now skip the test for the RST flag since all 2405 * paths to this code happen after packets containing 2406 * RST have been dropped. 2407 * 2408 * In the SYN-RECEIVED state, don't send an ACK unless the 2409 * segment we received passes the SYN-RECEIVED ACK test. 2410 * If it fails send a RST. This breaks the loop in the 2411 * "LAND" DoS attack, and also prevents an ACK storm 2412 * between two listening ports that have been sent forged 2413 * SYN segments, each with the source address of the other. 2414 */ 2415 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2416 (SEQ_GT(tp->snd_una, th->th_ack) || 2417 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2418 rstreason = BANDLIM_RST_OPENPORT; 2419 goto dropwithreset; 2420 } 2421#ifdef TCPDEBUG 2422 if (so->so_options & SO_DEBUG) 2423 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2424 &tcp_savetcp, 0); 2425#endif 2426 KASSERT(headlocked, ("headlocked should be 1")); 2427 tp->t_flags |= TF_ACKNOW; 2428 (void) tcp_output(tp); 2429 INP_UNLOCK(inp); 2430 INP_INFO_WUNLOCK(&tcbinfo); 2431 m_freem(m); 2432 return; 2433 2434dropwithreset: 2435 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2436 /* 2437 * Generate a RST, dropping incoming segment. 2438 * Make ACK acceptable to originator of segment. 2439 * Don't bother to respond if destination was broadcast/multicast. 2440 */ 2441 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2442 goto drop; 2443 if (isipv6) { 2444 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2445 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2446 goto drop; 2447 } else { 2448 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2449 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2450 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2451 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2452 goto drop; 2453 } 2454 /* IPv6 anycast check is done at tcp6_input() */ 2455 2456 /* 2457 * Perform bandwidth limiting. 2458 */ 2459 if (badport_bandlim(rstreason) < 0) 2460 goto drop; 2461 2462#ifdef TCPDEBUG 2463 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2464 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2465 &tcp_savetcp, 0); 2466#endif 2467 2468 if (thflags & TH_ACK) 2469 /* mtod() below is safe as long as hdr dropping is delayed */ 2470 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2471 TH_RST); 2472 else { 2473 if (thflags & TH_SYN) 2474 tlen++; 2475 /* mtod() below is safe as long as hdr dropping is delayed */ 2476 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2477 (tcp_seq)0, TH_RST|TH_ACK); 2478 } 2479 2480 if (tp) 2481 INP_UNLOCK(inp); 2482 if (headlocked) 2483 INP_INFO_WUNLOCK(&tcbinfo); 2484 return; 2485 2486drop: 2487 /* 2488 * Drop space held by incoming segment and return. 2489 */ 2490#ifdef TCPDEBUG 2491 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2492 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2493 &tcp_savetcp, 0); 2494#endif 2495 if (tp) 2496 INP_UNLOCK(inp); 2497 if (headlocked) 2498 INP_INFO_WUNLOCK(&tcbinfo); 2499 m_freem(m); 2500 return; 2501} 2502 2503/* 2504 * Parse TCP options and place in tcpopt. 2505 */ 2506static void 2507tcp_dooptions(tp, to, cp, cnt, is_syn, th) 2508 struct tcpcb *tp; 2509 struct tcpopt *to; 2510 u_char *cp; 2511 int cnt; 2512 int is_syn; 2513 struct tcphdr *th; 2514{ 2515 int opt, optlen; 2516 2517 to->to_flags = 0; 2518 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2519 opt = cp[0]; 2520 if (opt == TCPOPT_EOL) 2521 break; 2522 if (opt == TCPOPT_NOP) 2523 optlen = 1; 2524 else { 2525 if (cnt < 2) 2526 break; 2527 optlen = cp[1]; 2528 if (optlen < 2 || optlen > cnt) 2529 break; 2530 } 2531 switch (opt) { 2532 case TCPOPT_MAXSEG: 2533 if (optlen != TCPOLEN_MAXSEG) 2534 continue; 2535 if (!is_syn) 2536 continue; 2537 to->to_flags |= TOF_MSS; 2538 bcopy((char *)cp + 2, 2539 (char *)&to->to_mss, sizeof(to->to_mss)); 2540 to->to_mss = ntohs(to->to_mss); 2541 break; 2542 case TCPOPT_WINDOW: 2543 if (optlen != TCPOLEN_WINDOW) 2544 continue; 2545 if (! is_syn) 2546 continue; 2547 to->to_flags |= TOF_SCALE; 2548 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2549 break; 2550 case TCPOPT_TIMESTAMP: 2551 if (optlen != TCPOLEN_TIMESTAMP) 2552 continue; 2553 to->to_flags |= TOF_TS; 2554 bcopy((char *)cp + 2, 2555 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2556 to->to_tsval = ntohl(to->to_tsval); 2557 bcopy((char *)cp + 6, 2558 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2559 to->to_tsecr = ntohl(to->to_tsecr); 2560 break; 2561#ifdef TCP_SIGNATURE 2562 /* 2563 * XXX In order to reply to a host which has set the 2564 * TCP_SIGNATURE option in its initial SYN, we have to 2565 * record the fact that the option was observed here 2566 * for the syncache code to perform the correct response. 2567 */ 2568 case TCPOPT_SIGNATURE: 2569 if (optlen != TCPOLEN_SIGNATURE) 2570 continue; 2571 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2572 break; 2573#endif 2574 case TCPOPT_SACK_PERMITTED: 2575 if (!tcp_do_sack || 2576 optlen != TCPOLEN_SACK_PERMITTED) 2577 continue; 2578 if (is_syn) { 2579 /* MUST only be set on SYN */ 2580 to->to_flags |= TOF_SACK; 2581 } 2582 break; 2583 2584 case TCPOPT_SACK: 2585 if (!tp || tcp_sack_option(tp, th, cp, optlen)) 2586 continue; 2587 break; 2588 default: 2589 continue; 2590 } 2591 } 2592} 2593 2594/* 2595 * Pull out of band byte out of a segment so 2596 * it doesn't appear in the user's data queue. 2597 * It is still reflected in the segment length for 2598 * sequencing purposes. 2599 */ 2600static void 2601tcp_pulloutofband(so, th, m, off) 2602 struct socket *so; 2603 struct tcphdr *th; 2604 register struct mbuf *m; 2605 int off; /* delayed to be droped hdrlen */ 2606{ 2607 int cnt = off + th->th_urp - 1; 2608 2609 while (cnt >= 0) { 2610 if (m->m_len > cnt) { 2611 char *cp = mtod(m, caddr_t) + cnt; 2612 struct tcpcb *tp = sototcpcb(so); 2613 2614 tp->t_iobc = *cp; 2615 tp->t_oobflags |= TCPOOB_HAVEDATA; 2616 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2617 m->m_len--; 2618 if (m->m_flags & M_PKTHDR) 2619 m->m_pkthdr.len--; 2620 return; 2621 } 2622 cnt -= m->m_len; 2623 m = m->m_next; 2624 if (m == 0) 2625 break; 2626 } 2627 panic("tcp_pulloutofband"); 2628} 2629 2630/* 2631 * Collect new round-trip time estimate 2632 * and update averages and current timeout. 2633 */ 2634static void 2635tcp_xmit_timer(tp, rtt) 2636 register struct tcpcb *tp; 2637 int rtt; 2638{ 2639 register int delta; 2640 2641 tcpstat.tcps_rttupdated++; 2642 tp->t_rttupdated++; 2643 if (tp->t_srtt != 0) { 2644 /* 2645 * srtt is stored as fixed point with 5 bits after the 2646 * binary point (i.e., scaled by 8). The following magic 2647 * is equivalent to the smoothing algorithm in rfc793 with 2648 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2649 * point). Adjust rtt to origin 0. 2650 */ 2651 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2652 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2653 2654 if ((tp->t_srtt += delta) <= 0) 2655 tp->t_srtt = 1; 2656 2657 /* 2658 * We accumulate a smoothed rtt variance (actually, a 2659 * smoothed mean difference), then set the retransmit 2660 * timer to smoothed rtt + 4 times the smoothed variance. 2661 * rttvar is stored as fixed point with 4 bits after the 2662 * binary point (scaled by 16). The following is 2663 * equivalent to rfc793 smoothing with an alpha of .75 2664 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2665 * rfc793's wired-in beta. 2666 */ 2667 if (delta < 0) 2668 delta = -delta; 2669 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2670 if ((tp->t_rttvar += delta) <= 0) 2671 tp->t_rttvar = 1; 2672 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2673 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2674 } else { 2675 /* 2676 * No rtt measurement yet - use the unsmoothed rtt. 2677 * Set the variance to half the rtt (so our first 2678 * retransmit happens at 3*rtt). 2679 */ 2680 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2681 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2682 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2683 } 2684 tp->t_rtttime = 0; 2685 tp->t_rxtshift = 0; 2686 2687 /* 2688 * the retransmit should happen at rtt + 4 * rttvar. 2689 * Because of the way we do the smoothing, srtt and rttvar 2690 * will each average +1/2 tick of bias. When we compute 2691 * the retransmit timer, we want 1/2 tick of rounding and 2692 * 1 extra tick because of +-1/2 tick uncertainty in the 2693 * firing of the timer. The bias will give us exactly the 2694 * 1.5 tick we need. But, because the bias is 2695 * statistical, we have to test that we don't drop below 2696 * the minimum feasible timer (which is 2 ticks). 2697 */ 2698 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2699 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2700 2701 /* 2702 * We received an ack for a packet that wasn't retransmitted; 2703 * it is probably safe to discard any error indications we've 2704 * received recently. This isn't quite right, but close enough 2705 * for now (a route might have failed after we sent a segment, 2706 * and the return path might not be symmetrical). 2707 */ 2708 tp->t_softerror = 0; 2709} 2710 2711/* 2712 * Determine a reasonable value for maxseg size. 2713 * If the route is known, check route for mtu. 2714 * If none, use an mss that can be handled on the outgoing 2715 * interface without forcing IP to fragment; if bigger than 2716 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2717 * to utilize large mbufs. If no route is found, route has no mtu, 2718 * or the destination isn't local, use a default, hopefully conservative 2719 * size (usually 512 or the default IP max size, but no more than the mtu 2720 * of the interface), as we can't discover anything about intervening 2721 * gateways or networks. We also initialize the congestion/slow start 2722 * window to be a single segment if the destination isn't local. 2723 * While looking at the routing entry, we also initialize other path-dependent 2724 * parameters from pre-set or cached values in the routing entry. 2725 * 2726 * Also take into account the space needed for options that we 2727 * send regularly. Make maxseg shorter by that amount to assure 2728 * that we can send maxseg amount of data even when the options 2729 * are present. Store the upper limit of the length of options plus 2730 * data in maxopd. 2731 * 2732 * 2733 * In case of T/TCP, we call this routine during implicit connection 2734 * setup as well (offer = -1), to initialize maxseg from the cached 2735 * MSS of our peer. 2736 * 2737 * NOTE that this routine is only called when we process an incoming 2738 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2739 */ 2740void 2741tcp_mss(tp, offer) 2742 struct tcpcb *tp; 2743 int offer; 2744{ 2745 int rtt, mss; 2746 u_long bufsize; 2747 u_long maxmtu; 2748 struct inpcb *inp = tp->t_inpcb; 2749 struct socket *so; 2750 struct hc_metrics_lite metrics; 2751 int origoffer = offer; 2752#ifdef INET6 2753 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2754 size_t min_protoh = isipv6 ? 2755 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2756 sizeof (struct tcpiphdr); 2757#else 2758 const size_t min_protoh = sizeof(struct tcpiphdr); 2759#endif 2760 2761 /* initialize */ 2762#ifdef INET6 2763 if (isipv6) { 2764 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2765 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2766 } else 2767#endif 2768 { 2769 maxmtu = tcp_maxmtu(&inp->inp_inc); 2770 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2771 } 2772 so = inp->inp_socket; 2773 2774 /* 2775 * no route to sender, stay with default mss and return 2776 */ 2777 if (maxmtu == 0) 2778 return; 2779 2780 /* what have we got? */ 2781 switch (offer) { 2782 case 0: 2783 /* 2784 * Offer == 0 means that there was no MSS on the SYN 2785 * segment, in this case we use tcp_mssdflt. 2786 */ 2787 offer = 2788#ifdef INET6 2789 isipv6 ? tcp_v6mssdflt : 2790#endif 2791 tcp_mssdflt; 2792 break; 2793 2794 case -1: 2795 /* 2796 * Offer == -1 means that we didn't receive SYN yet. 2797 */ 2798 /* FALLTHROUGH */ 2799 2800 default: 2801 /* 2802 * Prevent DoS attack with too small MSS. Round up 2803 * to at least minmss. 2804 */ 2805 offer = max(offer, tcp_minmss); 2806 /* 2807 * Sanity check: make sure that maxopd will be large 2808 * enough to allow some data on segments even if the 2809 * all the option space is used (40bytes). Otherwise 2810 * funny things may happen in tcp_output. 2811 */ 2812 offer = max(offer, 64); 2813 } 2814 2815 /* 2816 * rmx information is now retrieved from tcp_hostcache 2817 */ 2818 tcp_hc_get(&inp->inp_inc, &metrics); 2819 2820 /* 2821 * if there's a discovered mtu int tcp hostcache, use it 2822 * else, use the link mtu. 2823 */ 2824 if (metrics.rmx_mtu) 2825 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2826 else { 2827#ifdef INET6 2828 if (isipv6) { 2829 mss = maxmtu - min_protoh; 2830 if (!path_mtu_discovery && 2831 !in6_localaddr(&inp->in6p_faddr)) 2832 mss = min(mss, tcp_v6mssdflt); 2833 } else 2834#endif 2835 { 2836 mss = maxmtu - min_protoh; 2837 if (!path_mtu_discovery && 2838 !in_localaddr(inp->inp_faddr)) 2839 mss = min(mss, tcp_mssdflt); 2840 } 2841 } 2842 mss = min(mss, offer); 2843 2844 /* 2845 * maxopd stores the maximum length of data AND options 2846 * in a segment; maxseg is the amount of data in a normal 2847 * segment. We need to store this value (maxopd) apart 2848 * from maxseg, because now every segment carries options 2849 * and thus we normally have somewhat less data in segments. 2850 */ 2851 tp->t_maxopd = mss; 2852 2853 /* 2854 * origoffer==-1 indicates, that no segments were received yet. 2855 * In this case we just guess. 2856 */ 2857 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2858 (origoffer == -1 || 2859 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2860 mss -= TCPOLEN_TSTAMP_APPA; 2861 tp->t_maxseg = mss; 2862 2863#if (MCLBYTES & (MCLBYTES - 1)) == 0 2864 if (mss > MCLBYTES) 2865 mss &= ~(MCLBYTES-1); 2866#else 2867 if (mss > MCLBYTES) 2868 mss = mss / MCLBYTES * MCLBYTES; 2869#endif 2870 tp->t_maxseg = mss; 2871 2872 /* 2873 * If there's a pipesize, change the socket buffer to that size, 2874 * don't change if sb_hiwat is different than default (then it 2875 * has been changed on purpose with setsockopt). 2876 * Make the socket buffers an integral number of mss units; 2877 * if the mss is larger than the socket buffer, decrease the mss. 2878 */ 2879 SOCKBUF_LOCK(&so->so_snd); 2880 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2881 bufsize = metrics.rmx_sendpipe; 2882 else 2883 bufsize = so->so_snd.sb_hiwat; 2884 if (bufsize < mss) 2885 mss = bufsize; 2886 else { 2887 bufsize = roundup(bufsize, mss); 2888 if (bufsize > sb_max) 2889 bufsize = sb_max; 2890 if (bufsize > so->so_snd.sb_hiwat) 2891 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2892 } 2893 SOCKBUF_UNLOCK(&so->so_snd); 2894 tp->t_maxseg = mss; 2895 2896 SOCKBUF_LOCK(&so->so_rcv); 2897 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2898 bufsize = metrics.rmx_recvpipe; 2899 else 2900 bufsize = so->so_rcv.sb_hiwat; 2901 if (bufsize > mss) { 2902 bufsize = roundup(bufsize, mss); 2903 if (bufsize > sb_max) 2904 bufsize = sb_max; 2905 if (bufsize > so->so_rcv.sb_hiwat) 2906 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2907 } 2908 SOCKBUF_UNLOCK(&so->so_rcv); 2909 /* 2910 * While we're here, check the others too 2911 */ 2912 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2913 tp->t_srtt = rtt; 2914 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2915 tcpstat.tcps_usedrtt++; 2916 if (metrics.rmx_rttvar) { 2917 tp->t_rttvar = metrics.rmx_rttvar; 2918 tcpstat.tcps_usedrttvar++; 2919 } else { 2920 /* default variation is +- 1 rtt */ 2921 tp->t_rttvar = 2922 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2923 } 2924 TCPT_RANGESET(tp->t_rxtcur, 2925 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2926 tp->t_rttmin, TCPTV_REXMTMAX); 2927 } 2928 if (metrics.rmx_ssthresh) { 2929 /* 2930 * There's some sort of gateway or interface 2931 * buffer limit on the path. Use this to set 2932 * the slow start threshhold, but set the 2933 * threshold to no less than 2*mss. 2934 */ 2935 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2936 tcpstat.tcps_usedssthresh++; 2937 } 2938 if (metrics.rmx_bandwidth) 2939 tp->snd_bandwidth = metrics.rmx_bandwidth; 2940 2941 /* 2942 * Set the slow-start flight size depending on whether this 2943 * is a local network or not. 2944 * 2945 * Extend this so we cache the cwnd too and retrieve it here. 2946 * Make cwnd even bigger than RFC3390 suggests but only if we 2947 * have previous experience with the remote host. Be careful 2948 * not make cwnd bigger than remote receive window or our own 2949 * send socket buffer. Maybe put some additional upper bound 2950 * on the retrieved cwnd. Should do incremental updates to 2951 * hostcache when cwnd collapses so next connection doesn't 2952 * overloads the path again. 2953 * 2954 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2955 * We currently check only in syncache_socket for that. 2956 */ 2957#define TCP_METRICS_CWND 2958#ifdef TCP_METRICS_CWND 2959 if (metrics.rmx_cwnd) 2960 tp->snd_cwnd = max(mss, 2961 min(metrics.rmx_cwnd / 2, 2962 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2963 else 2964#endif 2965 if (tcp_do_rfc3390) 2966 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2967#ifdef INET6 2968 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2969 (!isipv6 && in_localaddr(inp->inp_faddr))) 2970#else 2971 else if (in_localaddr(inp->inp_faddr)) 2972#endif 2973 tp->snd_cwnd = mss * ss_fltsz_local; 2974 else 2975 tp->snd_cwnd = mss * ss_fltsz; 2976} 2977 2978/* 2979 * Determine the MSS option to send on an outgoing SYN. 2980 */ 2981int 2982tcp_mssopt(inc) 2983 struct in_conninfo *inc; 2984{ 2985 int mss = 0; 2986 u_long maxmtu = 0; 2987 u_long thcmtu = 0; 2988 size_t min_protoh; 2989#ifdef INET6 2990 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2991#endif 2992 2993 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2994 2995#ifdef INET6 2996 if (isipv6) { 2997 mss = tcp_v6mssdflt; 2998 maxmtu = tcp_maxmtu6(inc); 2999 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3000 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3001 } else 3002#endif 3003 { 3004 mss = tcp_mssdflt; 3005 maxmtu = tcp_maxmtu(inc); 3006 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3007 min_protoh = sizeof(struct tcpiphdr); 3008 } 3009 if (maxmtu && thcmtu) 3010 mss = min(maxmtu, thcmtu) - min_protoh; 3011 else if (maxmtu || thcmtu) 3012 mss = max(maxmtu, thcmtu) - min_protoh; 3013 3014 return (mss); 3015} 3016 3017 3018/* 3019 * On a partial ack arrives, force the retransmission of the 3020 * next unacknowledged segment. Do not clear tp->t_dupacks. 3021 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3022 * be started again. 3023 */ 3024static void 3025tcp_newreno_partial_ack(tp, th) 3026 struct tcpcb *tp; 3027 struct tcphdr *th; 3028{ 3029 tcp_seq onxt = tp->snd_nxt; 3030 u_long ocwnd = tp->snd_cwnd; 3031 3032 callout_stop(tp->tt_rexmt); 3033 tp->t_rtttime = 0; 3034 tp->snd_nxt = th->th_ack; 3035 /* 3036 * Set snd_cwnd to one segment beyond acknowledged offset. 3037 * (tp->snd_una has not yet been updated when this function is called.) 3038 */ 3039 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3040 tp->t_flags |= TF_ACKNOW; 3041 (void) tcp_output(tp); 3042 tp->snd_cwnd = ocwnd; 3043 if (SEQ_GT(onxt, tp->snd_nxt)) 3044 tp->snd_nxt = onxt; 3045 /* 3046 * Partial window deflation. Relies on fact that tp->snd_una 3047 * not updated yet. 3048 */ 3049 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3050} 3051 3052/* 3053 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3054 * looking for a pcb in the listen state. Returns 0 otherwise. 3055 */ 3056static int 3057tcp_timewait(tw, to, th, m, tlen) 3058 struct tcptw *tw; 3059 struct tcpopt *to; 3060 struct tcphdr *th; 3061 struct mbuf *m; 3062 int tlen; 3063{ 3064 int thflags; 3065 tcp_seq seq; 3066#ifdef INET6 3067 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3068#else 3069 const int isipv6 = 0; 3070#endif 3071 3072 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3073 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3074 INP_LOCK_ASSERT(tw->tw_inpcb); 3075 3076 thflags = th->th_flags; 3077 3078 /* 3079 * NOTE: for FIN_WAIT_2 (to be added later), 3080 * must validate sequence number before accepting RST 3081 */ 3082 3083 /* 3084 * If the segment contains RST: 3085 * Drop the segment - see Stevens, vol. 2, p. 964 and 3086 * RFC 1337. 3087 */ 3088 if (thflags & TH_RST) 3089 goto drop; 3090 3091#if 0 3092/* PAWS not needed at the moment */ 3093 /* 3094 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3095 * and it's less than ts_recent, drop it. 3096 */ 3097 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3098 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3099 if ((thflags & TH_ACK) == 0) 3100 goto drop; 3101 goto ack; 3102 } 3103 /* 3104 * ts_recent is never updated because we never accept new segments. 3105 */ 3106#endif 3107 3108 /* 3109 * If a new connection request is received 3110 * while in TIME_WAIT, drop the old connection 3111 * and start over if the sequence numbers 3112 * are above the previous ones. 3113 */ 3114 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3115 (void) tcp_twclose(tw, 0); 3116 return (1); 3117 } 3118 3119 /* 3120 * Drop the the segment if it does not contain an ACK. 3121 */ 3122 if ((thflags & TH_ACK) == 0) 3123 goto drop; 3124 3125 /* 3126 * Reset the 2MSL timer if this is a duplicate FIN. 3127 */ 3128 if (thflags & TH_FIN) { 3129 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3130 if (seq + 1 == tw->rcv_nxt) 3131 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3132 } 3133 3134 /* 3135 * Acknowledge the segment if it has data or is not a duplicate ACK. 3136 */ 3137 if (thflags != TH_ACK || tlen != 0 || 3138 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3139 tcp_twrespond(tw, TH_ACK); 3140 goto drop; 3141 3142 /* 3143 * Generate a RST, dropping incoming segment. 3144 * Make ACK acceptable to originator of segment. 3145 * Don't bother to respond if destination was broadcast/multicast. 3146 */ 3147 if (m->m_flags & (M_BCAST|M_MCAST)) 3148 goto drop; 3149 if (isipv6) { 3150 struct ip6_hdr *ip6; 3151 3152 /* IPv6 anycast check is done at tcp6_input() */ 3153 ip6 = mtod(m, struct ip6_hdr *); 3154 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3155 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3156 goto drop; 3157 } else { 3158 struct ip *ip; 3159 3160 ip = mtod(m, struct ip *); 3161 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3162 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3163 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3164 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3165 goto drop; 3166 } 3167 if (thflags & TH_ACK) { 3168 tcp_respond(NULL, 3169 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3170 } else { 3171 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3172 tcp_respond(NULL, 3173 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3174 } 3175 INP_UNLOCK(tw->tw_inpcb); 3176 return (0); 3177 3178drop: 3179 INP_UNLOCK(tw->tw_inpcb); 3180 m_freem(m); 3181 return (0); 3182} 3183