tcp_input.c revision 130584
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 130584 2004-06-16 09:35:07Z bms $ 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 41#include <sys/param.h> 42#include <sys/kernel.h> 43#include <sys/mac.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/signalvar.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/sysctl.h> 52#include <sys/syslog.h> 53#include <sys/systm.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <vm/uma.h> 58 59#include <net/if.h> 60#include <net/route.h> 61 62#include <netinet/in.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_systm.h> 65#include <netinet/in_var.h> 66#include <netinet/ip.h> 67#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69#include <netinet/ip_var.h> 70#include <netinet/ip6.h> 71#include <netinet/icmp6.h> 72#include <netinet6/in6_pcb.h> 73#include <netinet6/ip6_var.h> 74#include <netinet6/nd6.h> 75#include <netinet/tcp.h> 76#include <netinet/tcp_fsm.h> 77#include <netinet/tcp_seq.h> 78#include <netinet/tcp_timer.h> 79#include <netinet/tcp_var.h> 80#include <netinet6/tcp6_var.h> 81#include <netinet/tcpip.h> 82#ifdef TCPDEBUG 83#include <netinet/tcp_debug.h> 84#endif /* TCPDEBUG */ 85 86#ifdef FAST_IPSEC 87#include <netipsec/ipsec.h> 88#include <netipsec/ipsec6.h> 89#endif /*FAST_IPSEC*/ 90 91#ifdef IPSEC 92#include <netinet6/ipsec.h> 93#include <netinet6/ipsec6.h> 94#include <netkey/key.h> 95#endif /*IPSEC*/ 96 97#include <machine/in_cksum.h> 98 99static const int tcprexmtthresh = 3; 100tcp_cc tcp_ccgen; 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 157struct inpcbhead tcb; 158#define tcb6 tcb /* for KAME src sync over BSD*'s */ 159struct inpcbinfo tcbinfo; 160struct mtx *tcbinfo_mtx; 161 162static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 163static void tcp_pulloutofband(struct socket *, 164 struct tcphdr *, struct mbuf *, int); 165static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 166 struct mbuf *); 167static void tcp_xmit_timer(struct tcpcb *, int); 168static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 169static int tcp_timewait(struct tcptw *, struct tcpopt *, 170 struct tcphdr *, struct mbuf *, int); 171 172/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 173#ifdef INET6 174#define ND6_HINT(tp) \ 175do { \ 176 if ((tp) && (tp)->t_inpcb && \ 177 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 178 nd6_nud_hint(NULL, NULL, 0); \ 179} while (0) 180#else 181#define ND6_HINT(tp) 182#endif 183 184/* 185 * Indicate whether this ack should be delayed. We can delay the ack if 186 * - there is no delayed ack timer in progress and 187 * - our last ack wasn't a 0-sized window. We never want to delay 188 * the ack that opens up a 0-sized window and 189 * - delayed acks are enabled or 190 * - this is a half-synchronized T/TCP connection. 191 */ 192#define DELAY_ACK(tp) \ 193 ((!callout_active(tp->tt_delack) && \ 194 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 195 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 196 197/* Initialize TCP reassembly queue */ 198uma_zone_t tcp_reass_zone; 199void 200tcp_reass_init() 201{ 202 tcp_reass_maxseg = nmbclusters / 16; 203 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 204 &tcp_reass_maxseg); 205 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 207 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 208} 209 210static int 211tcp_reass(tp, th, tlenp, m) 212 register struct tcpcb *tp; 213 register struct tcphdr *th; 214 int *tlenp; 215 struct mbuf *m; 216{ 217 struct tseg_qent *q; 218 struct tseg_qent *p = NULL; 219 struct tseg_qent *nq; 220 struct tseg_qent *te = NULL; 221 struct socket *so = tp->t_inpcb->inp_socket; 222 int flags; 223 224 /* 225 * XXX: tcp_reass() is rather inefficient with its data structures 226 * and should be rewritten (see NetBSD for optimizations). While 227 * doing that it should move to its own file tcp_reass.c. 228 */ 229 230 /* 231 * Call with th==0 after become established to 232 * force pre-ESTABLISHED data up to user socket. 233 */ 234 if (th == 0) 235 goto present; 236 237 /* 238 * Limit the number of segments in the reassembly queue to prevent 239 * holding on to too many segments (and thus running out of mbufs). 240 * Make sure to let the missing segment through which caused this 241 * queue. Always keep one global queue entry spare to be able to 242 * process the missing segment. 243 */ 244 if (th->th_seq != tp->rcv_nxt && 245 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 246 tp->t_segqlen >= tcp_reass_maxqlen)) { 247 tcp_reass_overflows++; 248 tcpstat.tcps_rcvmemdrop++; 249 m_freem(m); 250 return (0); 251 } 252 253 /* 254 * Allocate a new queue entry. If we can't, or hit the zone limit 255 * just drop the pkt. 256 */ 257 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 258 if (te == NULL) { 259 tcpstat.tcps_rcvmemdrop++; 260 m_freem(m); 261 return (0); 262 } 263 tp->t_segqlen++; 264 tcp_reass_qsize++; 265 266 /* 267 * Find a segment which begins after this one does. 268 */ 269 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 270 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 271 break; 272 p = q; 273 } 274 275 /* 276 * If there is a preceding segment, it may provide some of 277 * our data already. If so, drop the data from the incoming 278 * segment. If it provides all of our data, drop us. 279 */ 280 if (p != NULL) { 281 register int i; 282 /* conversion to int (in i) handles seq wraparound */ 283 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 284 if (i > 0) { 285 if (i >= *tlenp) { 286 tcpstat.tcps_rcvduppack++; 287 tcpstat.tcps_rcvdupbyte += *tlenp; 288 m_freem(m); 289 uma_zfree(tcp_reass_zone, te); 290 tp->t_segqlen--; 291 tcp_reass_qsize--; 292 /* 293 * Try to present any queued data 294 * at the left window edge to the user. 295 * This is needed after the 3-WHS 296 * completes. 297 */ 298 goto present; /* ??? */ 299 } 300 m_adj(m, i); 301 *tlenp -= i; 302 th->th_seq += i; 303 } 304 } 305 tcpstat.tcps_rcvoopack++; 306 tcpstat.tcps_rcvoobyte += *tlenp; 307 308 /* 309 * While we overlap succeeding segments trim them or, 310 * if they are completely covered, dequeue them. 311 */ 312 while (q) { 313 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 314 if (i <= 0) 315 break; 316 if (i < q->tqe_len) { 317 q->tqe_th->th_seq += i; 318 q->tqe_len -= i; 319 m_adj(q->tqe_m, i); 320 break; 321 } 322 323 nq = LIST_NEXT(q, tqe_q); 324 LIST_REMOVE(q, tqe_q); 325 m_freem(q->tqe_m); 326 uma_zfree(tcp_reass_zone, q); 327 tp->t_segqlen--; 328 tcp_reass_qsize--; 329 q = nq; 330 } 331 332 /* Insert the new segment queue entry into place. */ 333 te->tqe_m = m; 334 te->tqe_th = th; 335 te->tqe_len = *tlenp; 336 337 if (p == NULL) { 338 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 339 } else { 340 LIST_INSERT_AFTER(p, te, tqe_q); 341 } 342 343present: 344 /* 345 * Present data to user, advancing rcv_nxt through 346 * completed sequence space. 347 */ 348 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 349 return (0); 350 q = LIST_FIRST(&tp->t_segq); 351 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 352 return (0); 353 do { 354 tp->rcv_nxt += q->tqe_len; 355 flags = q->tqe_th->th_flags & TH_FIN; 356 nq = LIST_NEXT(q, tqe_q); 357 LIST_REMOVE(q, tqe_q); 358 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 359 m_freem(q->tqe_m); 360 else 361 sbappendstream(&so->so_rcv, q->tqe_m); 362 uma_zfree(tcp_reass_zone, q); 363 tp->t_segqlen--; 364 tcp_reass_qsize--; 365 q = nq; 366 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 367 ND6_HINT(tp); 368 sorwakeup(so); 369 return (flags); 370} 371 372/* 373 * TCP input routine, follows pages 65-76 of the 374 * protocol specification dated September, 1981 very closely. 375 */ 376#ifdef INET6 377int 378tcp6_input(mp, offp, proto) 379 struct mbuf **mp; 380 int *offp, proto; 381{ 382 register struct mbuf *m = *mp; 383 struct in6_ifaddr *ia6; 384 385 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 386 387 /* 388 * draft-itojun-ipv6-tcp-to-anycast 389 * better place to put this in? 390 */ 391 ia6 = ip6_getdstifaddr(m); 392 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 393 struct ip6_hdr *ip6; 394 395 ip6 = mtod(m, struct ip6_hdr *); 396 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 397 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 398 return IPPROTO_DONE; 399 } 400 401 tcp_input(m, *offp); 402 return IPPROTO_DONE; 403} 404#endif 405 406void 407tcp_input(m, off0) 408 register struct mbuf *m; 409 int off0; 410{ 411 register struct tcphdr *th; 412 register struct ip *ip = NULL; 413 register struct ipovly *ipov; 414 register struct inpcb *inp = NULL; 415 u_char *optp = NULL; 416 int optlen = 0; 417 int len, tlen, off; 418 int drop_hdrlen; 419 register struct tcpcb *tp = 0; 420 register int thflags; 421 struct socket *so = 0; 422 int todrop, acked, ourfinisacked, needoutput = 0; 423 u_long tiwin; 424 struct tcpopt to; /* options in this segment */ 425 struct rmxp_tao tao; /* our TAO cache entry */ 426 int headlocked = 0; 427 struct sockaddr_in *next_hop = NULL; 428 int rstreason; /* For badport_bandlim accounting purposes */ 429 430 struct ip6_hdr *ip6 = NULL; 431#ifdef INET6 432 int isipv6; 433#else 434 const int isipv6 = 0; 435#endif 436 437#ifdef TCPDEBUG 438 /* 439 * The size of tcp_saveipgen must be the size of the max ip header, 440 * now IPv6. 441 */ 442 u_char tcp_saveipgen[40]; 443 struct tcphdr tcp_savetcp; 444 short ostate = 0; 445#endif 446 447 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 448 next_hop = m_claim_next(m, PACKET_TAG_IPFORWARD); 449#ifdef INET6 450 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 451#endif 452 bzero(&tao, sizeof(tao)); 453 bzero((char *)&to, sizeof(to)); 454 455 tcpstat.tcps_rcvtotal++; 456 457 if (isipv6) { 458#ifdef INET6 459 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 460 ip6 = mtod(m, struct ip6_hdr *); 461 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 462 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 463 tcpstat.tcps_rcvbadsum++; 464 goto drop; 465 } 466 th = (struct tcphdr *)((caddr_t)ip6 + off0); 467 468 /* 469 * Be proactive about unspecified IPv6 address in source. 470 * As we use all-zero to indicate unbounded/unconnected pcb, 471 * unspecified IPv6 address can be used to confuse us. 472 * 473 * Note that packets with unspecified IPv6 destination is 474 * already dropped in ip6_input. 475 */ 476 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 477 /* XXX stat */ 478 goto drop; 479 } 480#else 481 th = NULL; /* XXX: avoid compiler warning */ 482#endif 483 } else { 484 /* 485 * Get IP and TCP header together in first mbuf. 486 * Note: IP leaves IP header in first mbuf. 487 */ 488 if (off0 > sizeof (struct ip)) { 489 ip_stripoptions(m, (struct mbuf *)0); 490 off0 = sizeof(struct ip); 491 } 492 if (m->m_len < sizeof (struct tcpiphdr)) { 493 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 494 tcpstat.tcps_rcvshort++; 495 return; 496 } 497 } 498 ip = mtod(m, struct ip *); 499 ipov = (struct ipovly *)ip; 500 th = (struct tcphdr *)((caddr_t)ip + off0); 501 tlen = ip->ip_len; 502 503 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 504 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 505 th->th_sum = m->m_pkthdr.csum_data; 506 else 507 th->th_sum = in_pseudo(ip->ip_src.s_addr, 508 ip->ip_dst.s_addr, 509 htonl(m->m_pkthdr.csum_data + 510 ip->ip_len + 511 IPPROTO_TCP)); 512 th->th_sum ^= 0xffff; 513#ifdef TCPDEBUG 514 ipov->ih_len = (u_short)tlen; 515 ipov->ih_len = htons(ipov->ih_len); 516#endif 517 } else { 518 /* 519 * Checksum extended TCP header and data. 520 */ 521 len = sizeof (struct ip) + tlen; 522 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525 th->th_sum = in_cksum(m, len); 526 } 527 if (th->th_sum) { 528 tcpstat.tcps_rcvbadsum++; 529 goto drop; 530 } 531#ifdef INET6 532 /* Re-initialization for later version check */ 533 ip->ip_v = IPVERSION; 534#endif 535 } 536 537 /* 538 * Check that TCP offset makes sense, 539 * pull out TCP options and adjust length. XXX 540 */ 541 off = th->th_off << 2; 542 if (off < sizeof (struct tcphdr) || off > tlen) { 543 tcpstat.tcps_rcvbadoff++; 544 goto drop; 545 } 546 tlen -= off; /* tlen is used instead of ti->ti_len */ 547 if (off > sizeof (struct tcphdr)) { 548 if (isipv6) { 549#ifdef INET6 550 IP6_EXTHDR_CHECK(m, off0, off, ); 551 ip6 = mtod(m, struct ip6_hdr *); 552 th = (struct tcphdr *)((caddr_t)ip6 + off0); 553#endif 554 } else { 555 if (m->m_len < sizeof(struct ip) + off) { 556 if ((m = m_pullup(m, sizeof (struct ip) + off)) 557 == 0) { 558 tcpstat.tcps_rcvshort++; 559 return; 560 } 561 ip = mtod(m, struct ip *); 562 ipov = (struct ipovly *)ip; 563 th = (struct tcphdr *)((caddr_t)ip + off0); 564 } 565 } 566 optlen = off - sizeof (struct tcphdr); 567 optp = (u_char *)(th + 1); 568 } 569 thflags = th->th_flags; 570 571#ifdef TCP_DROP_SYNFIN 572 /* 573 * If the drop_synfin option is enabled, drop all packets with 574 * both the SYN and FIN bits set. This prevents e.g. nmap from 575 * identifying the TCP/IP stack. 576 * 577 * This is a violation of the TCP specification. 578 */ 579 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 580 goto drop; 581#endif 582 583 /* 584 * Convert TCP protocol specific fields to host format. 585 */ 586 th->th_seq = ntohl(th->th_seq); 587 th->th_ack = ntohl(th->th_ack); 588 th->th_win = ntohs(th->th_win); 589 th->th_urp = ntohs(th->th_urp); 590 591 /* 592 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 593 * until after ip6_savecontrol() is called and before other functions 594 * which don't want those proto headers. 595 * Because ip6_savecontrol() is going to parse the mbuf to 596 * search for data to be passed up to user-land, it wants mbuf 597 * parameters to be unchanged. 598 * XXX: the call of ip6_savecontrol() has been obsoleted based on 599 * latest version of the advanced API (20020110). 600 */ 601 drop_hdrlen = off0 + off; 602 603 /* 604 * Locate pcb for segment. 605 */ 606 INP_INFO_WLOCK(&tcbinfo); 607 headlocked = 1; 608findpcb: 609 /* IPFIREWALL_FORWARD section */ 610 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 611 /* 612 * Transparently forwarded. Pretend to be the destination. 613 * already got one like this? 614 */ 615 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 616 ip->ip_dst, th->th_dport, 617 0, m->m_pkthdr.rcvif); 618 if (!inp) { 619 /* It's new. Try find the ambushing socket. */ 620 inp = in_pcblookup_hash(&tcbinfo, 621 ip->ip_src, th->th_sport, 622 next_hop->sin_addr, 623 next_hop->sin_port ? 624 ntohs(next_hop->sin_port) : 625 th->th_dport, 626 1, m->m_pkthdr.rcvif); 627 } 628 } else { 629 if (isipv6) { 630#ifdef INET6 631 inp = in6_pcblookup_hash(&tcbinfo, 632 &ip6->ip6_src, th->th_sport, 633 &ip6->ip6_dst, th->th_dport, 634 1, m->m_pkthdr.rcvif); 635#endif 636 } else 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 ip->ip_dst, th->th_dport, 640 1, m->m_pkthdr.rcvif); 641 } 642 643#if defined(IPSEC) || defined(FAST_IPSEC) 644#ifdef INET6 645 if (isipv6) { 646 if (inp != NULL && ipsec6_in_reject(m, inp)) { 647#ifdef IPSEC 648 ipsec6stat.in_polvio++; 649#endif 650 goto drop; 651 } 652 } else 653#endif /* INET6 */ 654 if (inp != NULL && ipsec4_in_reject(m, inp)) { 655#ifdef IPSEC 656 ipsecstat.in_polvio++; 657#endif 658 goto drop; 659 } 660#endif /*IPSEC || FAST_IPSEC*/ 661 662 /* 663 * If the state is CLOSED (i.e., TCB does not exist) then 664 * all data in the incoming segment is discarded. 665 * If the TCB exists but is in CLOSED state, it is embryonic, 666 * but should either do a listen or a connect soon. 667 */ 668 if (inp == NULL) { 669 if (log_in_vain) { 670#ifdef INET6 671 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 672#else 673 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 674#endif 675 676 if (isipv6) { 677#ifdef INET6 678 strcpy(dbuf, "["); 679 strcpy(sbuf, "["); 680 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 681 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 682 strcat(dbuf, "]"); 683 strcat(sbuf, "]"); 684#endif 685 } else { 686 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 687 strcpy(sbuf, inet_ntoa(ip->ip_src)); 688 } 689 switch (log_in_vain) { 690 case 1: 691 if ((thflags & TH_SYN) == 0) 692 break; 693 /* FALLTHROUGH */ 694 case 2: 695 log(LOG_INFO, 696 "Connection attempt to TCP %s:%d " 697 "from %s:%d flags:0x%02x\n", 698 dbuf, ntohs(th->th_dport), sbuf, 699 ntohs(th->th_sport), thflags); 700 break; 701 default: 702 break; 703 } 704 } 705 if (blackhole) { 706 switch (blackhole) { 707 case 1: 708 if (thflags & TH_SYN) 709 goto drop; 710 break; 711 case 2: 712 goto drop; 713 default: 714 goto drop; 715 } 716 } 717 rstreason = BANDLIM_RST_CLOSEDPORT; 718 goto dropwithreset; 719 } 720 INP_LOCK(inp); 721 if (inp->inp_vflag & INP_TIMEWAIT) { 722 /* 723 * The only option of relevance is TOF_CC, and only if 724 * present in a SYN segment. See tcp_timewait(). 725 */ 726 if (thflags & TH_SYN) 727 tcp_dooptions(&to, optp, optlen, 1); 728 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 729 &to, th, m, tlen)) 730 goto findpcb; 731 /* 732 * tcp_timewait unlocks inp. 733 */ 734 INP_INFO_WUNLOCK(&tcbinfo); 735 return; 736 } 737 tp = intotcpcb(inp); 738 if (tp == 0) { 739 INP_UNLOCK(inp); 740 rstreason = BANDLIM_RST_CLOSEDPORT; 741 goto dropwithreset; 742 } 743 if (tp->t_state == TCPS_CLOSED) 744 goto drop; 745 746 /* Unscale the window into a 32-bit value. */ 747 if ((thflags & TH_SYN) == 0) 748 tiwin = th->th_win << tp->snd_scale; 749 else 750 tiwin = th->th_win; 751 752#ifdef MAC 753 if (mac_check_inpcb_deliver(inp, m)) 754 goto drop; 755#endif 756 so = inp->inp_socket; 757#ifdef TCPDEBUG 758 if (so->so_options & SO_DEBUG) { 759 ostate = tp->t_state; 760 if (isipv6) 761 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 762 else 763 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 764 tcp_savetcp = *th; 765 } 766#endif 767 if (so->so_options & SO_ACCEPTCONN) { 768 struct in_conninfo inc; 769 770#ifdef INET6 771 inc.inc_isipv6 = isipv6; 772#endif 773 if (isipv6) { 774 inc.inc6_faddr = ip6->ip6_src; 775 inc.inc6_laddr = ip6->ip6_dst; 776 } else { 777 inc.inc_faddr = ip->ip_src; 778 inc.inc_laddr = ip->ip_dst; 779 } 780 inc.inc_fport = th->th_sport; 781 inc.inc_lport = th->th_dport; 782 783 /* 784 * If the state is LISTEN then ignore segment if it contains 785 * a RST. If the segment contains an ACK then it is bad and 786 * send a RST. If it does not contain a SYN then it is not 787 * interesting; drop it. 788 * 789 * If the state is SYN_RECEIVED (syncache) and seg contains 790 * an ACK, but not for our SYN/ACK, send a RST. If the seg 791 * contains a RST, check the sequence number to see if it 792 * is a valid reset segment. 793 */ 794 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 795 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 796 if (!syncache_expand(&inc, th, &so, m)) { 797 /* 798 * No syncache entry, or ACK was not 799 * for our SYN/ACK. Send a RST. 800 */ 801 tcpstat.tcps_badsyn++; 802 rstreason = BANDLIM_RST_OPENPORT; 803 goto dropwithreset; 804 } 805 if (so == NULL) { 806 /* 807 * Could not complete 3-way handshake, 808 * connection is being closed down, and 809 * syncache will free mbuf. 810 */ 811 INP_UNLOCK(inp); 812 INP_INFO_WUNLOCK(&tcbinfo); 813 return; 814 } 815 /* 816 * Socket is created in state SYN_RECEIVED. 817 * Continue processing segment. 818 */ 819 INP_UNLOCK(inp); 820 inp = sotoinpcb(so); 821 INP_LOCK(inp); 822 tp = intotcpcb(inp); 823 /* 824 * This is what would have happened in 825 * tcp_output() when the SYN,ACK was sent. 826 */ 827 tp->snd_up = tp->snd_una; 828 tp->snd_max = tp->snd_nxt = tp->iss + 1; 829 tp->last_ack_sent = tp->rcv_nxt; 830 /* 831 * RFC1323: The window in SYN & SYN/ACK 832 * segments is never scaled. 833 */ 834 tp->snd_wnd = tiwin; /* unscaled */ 835 goto after_listen; 836 } 837 if (thflags & TH_RST) { 838 syncache_chkrst(&inc, th); 839 goto drop; 840 } 841 if (thflags & TH_ACK) { 842 syncache_badack(&inc); 843 tcpstat.tcps_badsyn++; 844 rstreason = BANDLIM_RST_OPENPORT; 845 goto dropwithreset; 846 } 847 goto drop; 848 } 849 850 /* 851 * Segment's flags are (SYN) or (SYN|FIN). 852 */ 853#ifdef INET6 854 /* 855 * If deprecated address is forbidden, 856 * we do not accept SYN to deprecated interface 857 * address to prevent any new inbound connection from 858 * getting established. 859 * When we do not accept SYN, we send a TCP RST, 860 * with deprecated source address (instead of dropping 861 * it). We compromise it as it is much better for peer 862 * to send a RST, and RST will be the final packet 863 * for the exchange. 864 * 865 * If we do not forbid deprecated addresses, we accept 866 * the SYN packet. RFC2462 does not suggest dropping 867 * SYN in this case. 868 * If we decipher RFC2462 5.5.4, it says like this: 869 * 1. use of deprecated addr with existing 870 * communication is okay - "SHOULD continue to be 871 * used" 872 * 2. use of it with new communication: 873 * (2a) "SHOULD NOT be used if alternate address 874 * with sufficient scope is available" 875 * (2b) nothing mentioned otherwise. 876 * Here we fall into (2b) case as we have no choice in 877 * our source address selection - we must obey the peer. 878 * 879 * The wording in RFC2462 is confusing, and there are 880 * multiple description text for deprecated address 881 * handling - worse, they are not exactly the same. 882 * I believe 5.5.4 is the best one, so we follow 5.5.4. 883 */ 884 if (isipv6 && !ip6_use_deprecated) { 885 struct in6_ifaddr *ia6; 886 887 if ((ia6 = ip6_getdstifaddr(m)) && 888 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 889 INP_UNLOCK(inp); 890 tp = NULL; 891 rstreason = BANDLIM_RST_OPENPORT; 892 goto dropwithreset; 893 } 894 } 895#endif 896 /* 897 * If it is from this socket, drop it, it must be forged. 898 * Don't bother responding if the destination was a broadcast. 899 */ 900 if (th->th_dport == th->th_sport) { 901 if (isipv6) { 902 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 903 &ip6->ip6_src)) 904 goto drop; 905 } else { 906 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 907 goto drop; 908 } 909 } 910 /* 911 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 912 * 913 * Note that it is quite possible to receive unicast 914 * link-layer packets with a broadcast IP address. Use 915 * in_broadcast() to find them. 916 */ 917 if (m->m_flags & (M_BCAST|M_MCAST)) 918 goto drop; 919 if (isipv6) { 920 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 921 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 922 goto drop; 923 } else { 924 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 925 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 926 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 927 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 928 goto drop; 929 } 930 /* 931 * SYN appears to be valid; create compressed TCP state 932 * for syncache, or perform t/tcp connection. 933 */ 934 if (so->so_qlen <= so->so_qlimit) { 935#ifdef TCPDEBUG 936 if (so->so_options & SO_DEBUG) 937 tcp_trace(TA_INPUT, ostate, tp, 938 (void *)tcp_saveipgen, &tcp_savetcp, 0); 939#endif 940 tcp_dooptions(&to, optp, optlen, 1); 941 if (!syncache_add(&inc, &to, th, &so, m)) 942 goto drop; 943 if (so == NULL) { 944 /* 945 * Entry added to syncache, mbuf used to 946 * send SYN,ACK packet. 947 */ 948 KASSERT(headlocked, ("headlocked")); 949 INP_UNLOCK(inp); 950 INP_INFO_WUNLOCK(&tcbinfo); 951 return; 952 } 953 /* 954 * Segment passed TAO tests. 955 */ 956 INP_UNLOCK(inp); 957 inp = sotoinpcb(so); 958 INP_LOCK(inp); 959 tp = intotcpcb(inp); 960 tp->snd_wnd = tiwin; 961 tp->t_starttime = ticks; 962 tp->t_state = TCPS_ESTABLISHED; 963 964 /* 965 * T/TCP logic: 966 * If there is a FIN or if there is data, then 967 * delay SYN,ACK(SYN) in the hope of piggy-backing 968 * it on a response segment. Otherwise must send 969 * ACK now in case the other side is slow starting. 970 */ 971 if (thflags & TH_FIN || tlen != 0) 972 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 973 else 974 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 975 tcpstat.tcps_connects++; 976 soisconnected(so); 977 goto trimthenstep6; 978 } 979 goto drop; 980 } 981after_listen: 982 983 /* XXX temp debugging */ 984 /* should not happen - syncache should pick up these connections */ 985 if (tp->t_state == TCPS_LISTEN) 986 panic("tcp_input: TCPS_LISTEN"); 987 988 /* 989 * This is the second part of the MSS DoS prevention code (after 990 * minmss on the sending side) and it deals with too many too small 991 * tcp packets in a too short timeframe (1 second). 992 * 993 * For every full second we count the number of received packets 994 * and bytes. If we get a lot of packets per second for this connection 995 * (tcp_minmssoverload) we take a closer look at it and compute the 996 * average packet size for the past second. If that is less than 997 * tcp_minmss we get too many packets with very small payload which 998 * is not good and burdens our system (and every packet generates 999 * a wakeup to the process connected to our socket). We can reasonable 1000 * expect this to be small packet DoS attack to exhaust our CPU 1001 * cycles. 1002 * 1003 * Care has to be taken for the minimum packet overload value. This 1004 * value defines the minimum number of packets per second before we 1005 * start to worry. This must not be too low to avoid killing for 1006 * example interactive connections with many small packets like 1007 * telnet or SSH. 1008 * 1009 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1010 * this check. 1011 * 1012 * Account for packet if payload packet, skip over ACK, etc. 1013 */ 1014 if (tcp_minmss && tcp_minmssoverload && 1015 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1016 if (tp->rcv_second > ticks) { 1017 tp->rcv_pps++; 1018 tp->rcv_byps += tlen + off; 1019 if (tp->rcv_pps > tcp_minmssoverload) { 1020 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1021 printf("too many small tcp packets from " 1022 "%s:%u, av. %lubyte/packet, " 1023 "dropping connection\n", 1024#ifdef INET6 1025 isipv6 ? 1026 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1027#endif 1028 inet_ntoa(inp->inp_inc.inc_faddr), 1029 inp->inp_inc.inc_fport, 1030 tp->rcv_byps / tp->rcv_pps); 1031 tp = tcp_drop(tp, ECONNRESET); 1032 tcpstat.tcps_minmssdrops++; 1033 goto drop; 1034 } 1035 } 1036 } else { 1037 tp->rcv_second = ticks + hz; 1038 tp->rcv_pps = 1; 1039 tp->rcv_byps = tlen + off; 1040 } 1041 } 1042 1043 /* 1044 * Segment received on connection. 1045 * Reset idle time and keep-alive timer. 1046 */ 1047 tp->t_rcvtime = ticks; 1048 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1049 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1050 1051 /* 1052 * Process options only when we get SYN/ACK back. The SYN case 1053 * for incoming connections is handled in tcp_syncache. 1054 * XXX this is traditional behavior, may need to be cleaned up. 1055 */ 1056 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 1057 if (thflags & TH_SYN) { 1058 if (to.to_flags & TOF_SCALE) { 1059 tp->t_flags |= TF_RCVD_SCALE; 1060 tp->requested_s_scale = to.to_requested_s_scale; 1061 } 1062 if (to.to_flags & TOF_TS) { 1063 tp->t_flags |= TF_RCVD_TSTMP; 1064 tp->ts_recent = to.to_tsval; 1065 tp->ts_recent_age = ticks; 1066 } 1067 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1068 tp->t_flags |= TF_RCVD_CC; 1069 if (to.to_flags & TOF_MSS) 1070 tcp_mss(tp, to.to_mss); 1071 } 1072 1073 /* 1074 * Header prediction: check for the two common cases 1075 * of a uni-directional data xfer. If the packet has 1076 * no control flags, is in-sequence, the window didn't 1077 * change and we're not retransmitting, it's a 1078 * candidate. If the length is zero and the ack moved 1079 * forward, we're the sender side of the xfer. Just 1080 * free the data acked & wake any higher level process 1081 * that was blocked waiting for space. If the length 1082 * is non-zero and the ack didn't move, we're the 1083 * receiver side. If we're getting packets in-order 1084 * (the reassembly queue is empty), add the data to 1085 * the socket buffer and note that we need a delayed ack. 1086 * Make sure that the hidden state-flags are also off. 1087 * Since we check for TCPS_ESTABLISHED above, it can only 1088 * be TH_NEEDSYN. 1089 */ 1090 if (tp->t_state == TCPS_ESTABLISHED && 1091 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1092 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1093 ((to.to_flags & TOF_TS) == 0 || 1094 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1095 /* 1096 * Using the CC option is compulsory if once started: 1097 * the segment is OK if no T/TCP was negotiated or 1098 * if the segment has a CC option equal to CCrecv 1099 */ 1100 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1101 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1102 th->th_seq == tp->rcv_nxt && 1103 tiwin && tiwin == tp->snd_wnd && 1104 tp->snd_nxt == tp->snd_max) { 1105 1106 /* 1107 * If last ACK falls within this segment's sequence numbers, 1108 * record the timestamp. 1109 * NOTE that the test is modified according to the latest 1110 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1111 */ 1112 if ((to.to_flags & TOF_TS) != 0 && 1113 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1114 tp->ts_recent_age = ticks; 1115 tp->ts_recent = to.to_tsval; 1116 } 1117 1118 if (tlen == 0) { 1119 if (SEQ_GT(th->th_ack, tp->snd_una) && 1120 SEQ_LEQ(th->th_ack, tp->snd_max) && 1121 tp->snd_cwnd >= tp->snd_wnd && 1122 ((!tcp_do_newreno && 1123 tp->t_dupacks < tcprexmtthresh) || 1124 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1125 KASSERT(headlocked, ("headlocked")); 1126 INP_INFO_WUNLOCK(&tcbinfo); 1127 /* 1128 * this is a pure ack for outstanding data. 1129 */ 1130 ++tcpstat.tcps_predack; 1131 /* 1132 * "bad retransmit" recovery 1133 */ 1134 if (tp->t_rxtshift == 1 && 1135 ticks < tp->t_badrxtwin) { 1136 ++tcpstat.tcps_sndrexmitbad; 1137 tp->snd_cwnd = tp->snd_cwnd_prev; 1138 tp->snd_ssthresh = 1139 tp->snd_ssthresh_prev; 1140 tp->snd_recover = tp->snd_recover_prev; 1141 if (tp->t_flags & TF_WASFRECOVERY) 1142 ENTER_FASTRECOVERY(tp); 1143 tp->snd_nxt = tp->snd_max; 1144 tp->t_badrxtwin = 0; 1145 } 1146 1147 /* 1148 * Recalculate the transmit timer / rtt. 1149 * 1150 * Some boxes send broken timestamp replies 1151 * during the SYN+ACK phase, ignore 1152 * timestamps of 0 or we could calculate a 1153 * huge RTT and blow up the retransmit timer. 1154 */ 1155 if ((to.to_flags & TOF_TS) != 0 && 1156 to.to_tsecr) { 1157 tcp_xmit_timer(tp, 1158 ticks - to.to_tsecr + 1); 1159 } else if (tp->t_rtttime && 1160 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1161 tcp_xmit_timer(tp, 1162 ticks - tp->t_rtttime); 1163 } 1164 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1165 acked = th->th_ack - tp->snd_una; 1166 tcpstat.tcps_rcvackpack++; 1167 tcpstat.tcps_rcvackbyte += acked; 1168 sbdrop(&so->so_snd, acked); 1169 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1170 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1171 tp->snd_recover = th->th_ack - 1; 1172 tp->snd_una = th->th_ack; 1173 /* 1174 * pull snd_wl2 up to prevent seq wrap relative 1175 * to th_ack. 1176 */ 1177 tp->snd_wl2 = th->th_ack; 1178 tp->t_dupacks = 0; 1179 m_freem(m); 1180 ND6_HINT(tp); /* some progress has been done */ 1181 1182 /* 1183 * If all outstanding data are acked, stop 1184 * retransmit timer, otherwise restart timer 1185 * using current (possibly backed-off) value. 1186 * If process is waiting for space, 1187 * wakeup/selwakeup/signal. If data 1188 * are ready to send, let tcp_output 1189 * decide between more output or persist. 1190 1191#ifdef TCPDEBUG 1192 if (so->so_options & SO_DEBUG) 1193 tcp_trace(TA_INPUT, ostate, tp, 1194 (void *)tcp_saveipgen, 1195 &tcp_savetcp, 0); 1196#endif 1197 */ 1198 if (tp->snd_una == tp->snd_max) 1199 callout_stop(tp->tt_rexmt); 1200 else if (!callout_active(tp->tt_persist)) 1201 callout_reset(tp->tt_rexmt, 1202 tp->t_rxtcur, 1203 tcp_timer_rexmt, tp); 1204 1205 sowwakeup(so); 1206 if (so->so_snd.sb_cc) 1207 (void) tcp_output(tp); 1208 goto check_delack; 1209 } 1210 } else if (th->th_ack == tp->snd_una && 1211 LIST_EMPTY(&tp->t_segq) && 1212 tlen <= sbspace(&so->so_rcv)) { 1213 KASSERT(headlocked, ("headlocked")); 1214 INP_INFO_WUNLOCK(&tcbinfo); 1215 /* 1216 * this is a pure, in-sequence data packet 1217 * with nothing on the reassembly queue and 1218 * we have enough buffer space to take it. 1219 */ 1220 ++tcpstat.tcps_preddat; 1221 tp->rcv_nxt += tlen; 1222 /* 1223 * Pull snd_wl1 up to prevent seq wrap relative to 1224 * th_seq. 1225 */ 1226 tp->snd_wl1 = th->th_seq; 1227 /* 1228 * Pull rcv_up up to prevent seq wrap relative to 1229 * rcv_nxt. 1230 */ 1231 tp->rcv_up = tp->rcv_nxt; 1232 tcpstat.tcps_rcvpack++; 1233 tcpstat.tcps_rcvbyte += tlen; 1234 ND6_HINT(tp); /* some progress has been done */ 1235 /* 1236#ifdef TCPDEBUG 1237 if (so->so_options & SO_DEBUG) 1238 tcp_trace(TA_INPUT, ostate, tp, 1239 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1240#endif 1241 * Add data to socket buffer. 1242 */ 1243 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1244 m_freem(m); 1245 } else { 1246 m_adj(m, drop_hdrlen); /* delayed header drop */ 1247 sbappendstream(&so->so_rcv, m); 1248 } 1249 sorwakeup(so); 1250 if (DELAY_ACK(tp)) { 1251 tp->t_flags |= TF_DELACK; 1252 } else { 1253 tp->t_flags |= TF_ACKNOW; 1254 tcp_output(tp); 1255 } 1256 goto check_delack; 1257 } 1258 } 1259 1260 /* 1261 * Calculate amount of space in receive window, 1262 * and then do TCP input processing. 1263 * Receive window is amount of space in rcv queue, 1264 * but not less than advertised window. 1265 */ 1266 { int win; 1267 1268 win = sbspace(&so->so_rcv); 1269 if (win < 0) 1270 win = 0; 1271 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1272 } 1273 1274 switch (tp->t_state) { 1275 1276 /* 1277 * If the state is SYN_RECEIVED: 1278 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1279 */ 1280 case TCPS_SYN_RECEIVED: 1281 if ((thflags & TH_ACK) && 1282 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1283 SEQ_GT(th->th_ack, tp->snd_max))) { 1284 rstreason = BANDLIM_RST_OPENPORT; 1285 goto dropwithreset; 1286 } 1287 break; 1288 1289 /* 1290 * If the state is SYN_SENT: 1291 * if seg contains an ACK, but not for our SYN, drop the input. 1292 * if seg contains a RST, then drop the connection. 1293 * if seg does not contain SYN, then drop it. 1294 * Otherwise this is an acceptable SYN segment 1295 * initialize tp->rcv_nxt and tp->irs 1296 * if seg contains ack then advance tp->snd_una 1297 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1298 * arrange for segment to be acked (eventually) 1299 * continue processing rest of data/controls, beginning with URG 1300 */ 1301 case TCPS_SYN_SENT: 1302 if (tcp_do_rfc1644) 1303 tcp_hc_gettao(&inp->inp_inc, &tao); 1304 1305 if ((thflags & TH_ACK) && 1306 (SEQ_LEQ(th->th_ack, tp->iss) || 1307 SEQ_GT(th->th_ack, tp->snd_max))) { 1308 /* 1309 * If we have a cached CCsent for the remote host, 1310 * hence we haven't just crashed and restarted, 1311 * do not send a RST. This may be a retransmission 1312 * from the other side after our earlier ACK was lost. 1313 * Our new SYN, when it arrives, will serve as the 1314 * needed ACK. 1315 */ 1316 if (tao.tao_ccsent != 0) 1317 goto drop; 1318 else { 1319 rstreason = BANDLIM_UNLIMITED; 1320 goto dropwithreset; 1321 } 1322 } 1323 if (thflags & TH_RST) { 1324 if (thflags & TH_ACK) 1325 tp = tcp_drop(tp, ECONNREFUSED); 1326 goto drop; 1327 } 1328 if ((thflags & TH_SYN) == 0) 1329 goto drop; 1330 tp->snd_wnd = th->th_win; /* initial send window */ 1331 tp->cc_recv = to.to_cc; /* foreign CC */ 1332 1333 tp->irs = th->th_seq; 1334 tcp_rcvseqinit(tp); 1335 if (thflags & TH_ACK) { 1336 /* 1337 * Our SYN was acked. If segment contains CC.ECHO 1338 * option, check it to make sure this segment really 1339 * matches our SYN. If not, just drop it as old 1340 * duplicate, but send an RST if we're still playing 1341 * by the old rules. If no CC.ECHO option, make sure 1342 * we don't get fooled into using T/TCP. 1343 */ 1344 if (to.to_flags & TOF_CCECHO) { 1345 if (tp->cc_send != to.to_ccecho) { 1346 if (tao.tao_ccsent != 0) 1347 goto drop; 1348 else { 1349 rstreason = BANDLIM_UNLIMITED; 1350 goto dropwithreset; 1351 } 1352 } 1353 } else 1354 tp->t_flags &= ~TF_RCVD_CC; 1355 tcpstat.tcps_connects++; 1356 soisconnected(so); 1357#ifdef MAC 1358 SOCK_LOCK(so); 1359 mac_set_socket_peer_from_mbuf(m, so); 1360 SOCK_UNLOCK(so); 1361#endif 1362 /* Do window scaling on this connection? */ 1363 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1364 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1365 tp->snd_scale = tp->requested_s_scale; 1366 tp->rcv_scale = tp->request_r_scale; 1367 } 1368 /* Segment is acceptable, update cache if undefined. */ 1369 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1370 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1371 1372 tp->rcv_adv += tp->rcv_wnd; 1373 tp->snd_una++; /* SYN is acked */ 1374 /* 1375 * If there's data, delay ACK; if there's also a FIN 1376 * ACKNOW will be turned on later. 1377 */ 1378 if (DELAY_ACK(tp) && tlen != 0) 1379 callout_reset(tp->tt_delack, tcp_delacktime, 1380 tcp_timer_delack, tp); 1381 else 1382 tp->t_flags |= TF_ACKNOW; 1383 /* 1384 * Received <SYN,ACK> in SYN_SENT[*] state. 1385 * Transitions: 1386 * SYN_SENT --> ESTABLISHED 1387 * SYN_SENT* --> FIN_WAIT_1 1388 */ 1389 tp->t_starttime = ticks; 1390 if (tp->t_flags & TF_NEEDFIN) { 1391 tp->t_state = TCPS_FIN_WAIT_1; 1392 tp->t_flags &= ~TF_NEEDFIN; 1393 thflags &= ~TH_SYN; 1394 } else { 1395 tp->t_state = TCPS_ESTABLISHED; 1396 callout_reset(tp->tt_keep, tcp_keepidle, 1397 tcp_timer_keep, tp); 1398 } 1399 } else { 1400 /* 1401 * Received initial SYN in SYN-SENT[*] state => 1402 * simultaneous open. If segment contains CC option 1403 * and there is a cached CC, apply TAO test. 1404 * If it succeeds, connection is * half-synchronized. 1405 * Otherwise, do 3-way handshake: 1406 * SYN-SENT -> SYN-RECEIVED 1407 * SYN-SENT* -> SYN-RECEIVED* 1408 * If there was no CC option, clear cached CC value. 1409 */ 1410 tp->t_flags |= TF_ACKNOW; 1411 callout_stop(tp->tt_rexmt); 1412 if (to.to_flags & TOF_CC) { 1413 if (tao.tao_cc != 0 && 1414 CC_GT(to.to_cc, tao.tao_cc)) { 1415 /* 1416 * update cache and make transition: 1417 * SYN-SENT -> ESTABLISHED* 1418 * SYN-SENT* -> FIN-WAIT-1* 1419 */ 1420 tao.tao_cc = to.to_cc; 1421 tcp_hc_updatetao(&inp->inp_inc, 1422 TCP_HC_TAO_CC, to.to_cc, 0); 1423 tp->t_starttime = ticks; 1424 if (tp->t_flags & TF_NEEDFIN) { 1425 tp->t_state = TCPS_FIN_WAIT_1; 1426 tp->t_flags &= ~TF_NEEDFIN; 1427 } else { 1428 tp->t_state = TCPS_ESTABLISHED; 1429 callout_reset(tp->tt_keep, 1430 tcp_keepidle, 1431 tcp_timer_keep, 1432 tp); 1433 } 1434 tp->t_flags |= TF_NEEDSYN; 1435 } else 1436 tp->t_state = TCPS_SYN_RECEIVED; 1437 } else { 1438 if (tcp_do_rfc1644) { 1439 /* CC.NEW or no option => invalidate cache */ 1440 tao.tao_cc = 0; 1441 tcp_hc_updatetao(&inp->inp_inc, 1442 TCP_HC_TAO_CC, to.to_cc, 0); 1443 } 1444 tp->t_state = TCPS_SYN_RECEIVED; 1445 } 1446 } 1447 1448trimthenstep6: 1449 /* 1450 * Advance th->th_seq to correspond to first data byte. 1451 * If data, trim to stay within window, 1452 * dropping FIN if necessary. 1453 */ 1454 th->th_seq++; 1455 if (tlen > tp->rcv_wnd) { 1456 todrop = tlen - tp->rcv_wnd; 1457 m_adj(m, -todrop); 1458 tlen = tp->rcv_wnd; 1459 thflags &= ~TH_FIN; 1460 tcpstat.tcps_rcvpackafterwin++; 1461 tcpstat.tcps_rcvbyteafterwin += todrop; 1462 } 1463 tp->snd_wl1 = th->th_seq - 1; 1464 tp->rcv_up = th->th_seq; 1465 /* 1466 * Client side of transaction: already sent SYN and data. 1467 * If the remote host used T/TCP to validate the SYN, 1468 * our data will be ACK'd; if so, enter normal data segment 1469 * processing in the middle of step 5, ack processing. 1470 * Otherwise, goto step 6. 1471 */ 1472 if (thflags & TH_ACK) 1473 goto process_ACK; 1474 1475 goto step6; 1476 1477 /* 1478 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1479 * if segment contains a SYN and CC [not CC.NEW] option: 1480 * if state == TIME_WAIT and connection duration > MSL, 1481 * drop packet and send RST; 1482 * 1483 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1484 * ack the FIN (and data) in retransmission queue. 1485 * Complete close and delete TCPCB. Then reprocess 1486 * segment, hoping to find new TCPCB in LISTEN state; 1487 * 1488 * else must be old SYN; drop it. 1489 * else do normal processing. 1490 */ 1491 case TCPS_LAST_ACK: 1492 case TCPS_CLOSING: 1493 case TCPS_TIME_WAIT: 1494 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1495 if ((thflags & TH_SYN) && 1496 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1497 if (tp->t_state == TCPS_TIME_WAIT && 1498 (ticks - tp->t_starttime) > tcp_msl) { 1499 rstreason = BANDLIM_UNLIMITED; 1500 goto dropwithreset; 1501 } 1502 if (CC_GT(to.to_cc, tp->cc_recv)) { 1503 tp = tcp_close(tp); 1504 goto findpcb; 1505 } 1506 else 1507 goto drop; 1508 } 1509 break; /* continue normal processing */ 1510 } 1511 1512 /* 1513 * States other than LISTEN or SYN_SENT. 1514 * First check the RST flag and sequence number since reset segments 1515 * are exempt from the timestamp and connection count tests. This 1516 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1517 * below which allowed reset segments in half the sequence space 1518 * to fall though and be processed (which gives forged reset 1519 * segments with a random sequence number a 50 percent chance of 1520 * killing a connection). 1521 * Then check timestamp, if present. 1522 * Then check the connection count, if present. 1523 * Then check that at least some bytes of segment are within 1524 * receive window. If segment begins before rcv_nxt, 1525 * drop leading data (and SYN); if nothing left, just ack. 1526 * 1527 * 1528 * If the RST bit is set, check the sequence number to see 1529 * if this is a valid reset segment. 1530 * RFC 793 page 37: 1531 * In all states except SYN-SENT, all reset (RST) segments 1532 * are validated by checking their SEQ-fields. A reset is 1533 * valid if its sequence number is in the window. 1534 * Note: this does not take into account delayed ACKs, so 1535 * we should test against last_ack_sent instead of rcv_nxt. 1536 * The sequence number in the reset segment is normally an 1537 * echo of our outgoing acknowlegement numbers, but some hosts 1538 * send a reset with the sequence number at the rightmost edge 1539 * of our receive window, and we have to handle this case. 1540 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1541 * that brute force RST attacks are possible. To combat this, 1542 * we use a much stricter check while in the ESTABLISHED state, 1543 * only accepting RSTs where the sequence number is equal to 1544 * last_ack_sent. In all other states (the states in which a 1545 * RST is more likely), the more permissive check is used. 1546 * If we have multiple segments in flight, the intial reset 1547 * segment sequence numbers will be to the left of last_ack_sent, 1548 * but they will eventually catch up. 1549 * In any case, it never made sense to trim reset segments to 1550 * fit the receive window since RFC 1122 says: 1551 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1552 * 1553 * A TCP SHOULD allow a received RST segment to include data. 1554 * 1555 * DISCUSSION 1556 * It has been suggested that a RST segment could contain 1557 * ASCII text that encoded and explained the cause of the 1558 * RST. No standard has yet been established for such 1559 * data. 1560 * 1561 * If the reset segment passes the sequence number test examine 1562 * the state: 1563 * SYN_RECEIVED STATE: 1564 * If passive open, return to LISTEN state. 1565 * If active open, inform user that connection was refused. 1566 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1567 * Inform user that connection was reset, and close tcb. 1568 * CLOSING, LAST_ACK STATES: 1569 * Close the tcb. 1570 * TIME_WAIT STATE: 1571 * Drop the segment - see Stevens, vol. 2, p. 964 and 1572 * RFC 1337. 1573 */ 1574 if (thflags & TH_RST) { 1575 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1576 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1577 switch (tp->t_state) { 1578 1579 case TCPS_SYN_RECEIVED: 1580 so->so_error = ECONNREFUSED; 1581 goto close; 1582 1583 case TCPS_ESTABLISHED: 1584 if (tp->last_ack_sent != th->th_seq) { 1585 tcpstat.tcps_badrst++; 1586 goto drop; 1587 } 1588 case TCPS_FIN_WAIT_1: 1589 case TCPS_FIN_WAIT_2: 1590 case TCPS_CLOSE_WAIT: 1591 so->so_error = ECONNRESET; 1592 close: 1593 tp->t_state = TCPS_CLOSED; 1594 tcpstat.tcps_drops++; 1595 tp = tcp_close(tp); 1596 break; 1597 1598 case TCPS_CLOSING: 1599 case TCPS_LAST_ACK: 1600 tp = tcp_close(tp); 1601 break; 1602 1603 case TCPS_TIME_WAIT: 1604 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1605 ("timewait")); 1606 break; 1607 } 1608 } 1609 goto drop; 1610 } 1611 1612 /* 1613 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1614 * and it's less than ts_recent, drop it. 1615 */ 1616 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1617 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1618 1619 /* Check to see if ts_recent is over 24 days old. */ 1620 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1621 /* 1622 * Invalidate ts_recent. If this segment updates 1623 * ts_recent, the age will be reset later and ts_recent 1624 * will get a valid value. If it does not, setting 1625 * ts_recent to zero will at least satisfy the 1626 * requirement that zero be placed in the timestamp 1627 * echo reply when ts_recent isn't valid. The 1628 * age isn't reset until we get a valid ts_recent 1629 * because we don't want out-of-order segments to be 1630 * dropped when ts_recent is old. 1631 */ 1632 tp->ts_recent = 0; 1633 } else { 1634 tcpstat.tcps_rcvduppack++; 1635 tcpstat.tcps_rcvdupbyte += tlen; 1636 tcpstat.tcps_pawsdrop++; 1637 if (tlen) 1638 goto dropafterack; 1639 goto drop; 1640 } 1641 } 1642 1643 /* 1644 * T/TCP mechanism 1645 * If T/TCP was negotiated and the segment doesn't have CC, 1646 * or if its CC is wrong then drop the segment. 1647 * RST segments do not have to comply with this. 1648 */ 1649 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1650 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1651 goto dropafterack; 1652 1653 /* 1654 * In the SYN-RECEIVED state, validate that the packet belongs to 1655 * this connection before trimming the data to fit the receive 1656 * window. Check the sequence number versus IRS since we know 1657 * the sequence numbers haven't wrapped. This is a partial fix 1658 * for the "LAND" DoS attack. 1659 */ 1660 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1661 rstreason = BANDLIM_RST_OPENPORT; 1662 goto dropwithreset; 1663 } 1664 1665 todrop = tp->rcv_nxt - th->th_seq; 1666 if (todrop > 0) { 1667 if (thflags & TH_SYN) { 1668 thflags &= ~TH_SYN; 1669 th->th_seq++; 1670 if (th->th_urp > 1) 1671 th->th_urp--; 1672 else 1673 thflags &= ~TH_URG; 1674 todrop--; 1675 } 1676 /* 1677 * Following if statement from Stevens, vol. 2, p. 960. 1678 */ 1679 if (todrop > tlen 1680 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1681 /* 1682 * Any valid FIN must be to the left of the window. 1683 * At this point the FIN must be a duplicate or out 1684 * of sequence; drop it. 1685 */ 1686 thflags &= ~TH_FIN; 1687 1688 /* 1689 * Send an ACK to resynchronize and drop any data. 1690 * But keep on processing for RST or ACK. 1691 */ 1692 tp->t_flags |= TF_ACKNOW; 1693 todrop = tlen; 1694 tcpstat.tcps_rcvduppack++; 1695 tcpstat.tcps_rcvdupbyte += todrop; 1696 } else { 1697 tcpstat.tcps_rcvpartduppack++; 1698 tcpstat.tcps_rcvpartdupbyte += todrop; 1699 } 1700 drop_hdrlen += todrop; /* drop from the top afterwards */ 1701 th->th_seq += todrop; 1702 tlen -= todrop; 1703 if (th->th_urp > todrop) 1704 th->th_urp -= todrop; 1705 else { 1706 thflags &= ~TH_URG; 1707 th->th_urp = 0; 1708 } 1709 } 1710 1711 /* 1712 * If new data are received on a connection after the 1713 * user processes are gone, then RST the other end. 1714 */ 1715 if ((so->so_state & SS_NOFDREF) && 1716 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1717 tp = tcp_close(tp); 1718 tcpstat.tcps_rcvafterclose++; 1719 rstreason = BANDLIM_UNLIMITED; 1720 goto dropwithreset; 1721 } 1722 1723 /* 1724 * If segment ends after window, drop trailing data 1725 * (and PUSH and FIN); if nothing left, just ACK. 1726 */ 1727 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1728 if (todrop > 0) { 1729 tcpstat.tcps_rcvpackafterwin++; 1730 if (todrop >= tlen) { 1731 tcpstat.tcps_rcvbyteafterwin += tlen; 1732 /* 1733 * If a new connection request is received 1734 * while in TIME_WAIT, drop the old connection 1735 * and start over if the sequence numbers 1736 * are above the previous ones. 1737 */ 1738 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1739 if (thflags & TH_SYN && 1740 tp->t_state == TCPS_TIME_WAIT && 1741 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1742 tp = tcp_close(tp); 1743 goto findpcb; 1744 } 1745 /* 1746 * If window is closed can only take segments at 1747 * window edge, and have to drop data and PUSH from 1748 * incoming segments. Continue processing, but 1749 * remember to ack. Otherwise, drop segment 1750 * and ack. 1751 */ 1752 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1753 tp->t_flags |= TF_ACKNOW; 1754 tcpstat.tcps_rcvwinprobe++; 1755 } else 1756 goto dropafterack; 1757 } else 1758 tcpstat.tcps_rcvbyteafterwin += todrop; 1759 m_adj(m, -todrop); 1760 tlen -= todrop; 1761 thflags &= ~(TH_PUSH|TH_FIN); 1762 } 1763 1764 /* 1765 * If last ACK falls within this segment's sequence numbers, 1766 * record its timestamp. 1767 * NOTE that the test is modified according to the latest 1768 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1769 */ 1770 if ((to.to_flags & TOF_TS) != 0 && 1771 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1772 tp->ts_recent_age = ticks; 1773 tp->ts_recent = to.to_tsval; 1774 } 1775 1776 /* 1777 * If a SYN is in the window, then this is an 1778 * error and we send an RST and drop the connection. 1779 */ 1780 if (thflags & TH_SYN) { 1781 tp = tcp_drop(tp, ECONNRESET); 1782 rstreason = BANDLIM_UNLIMITED; 1783 goto drop; 1784 } 1785 1786 /* 1787 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1788 * flag is on (half-synchronized state), then queue data for 1789 * later processing; else drop segment and return. 1790 */ 1791 if ((thflags & TH_ACK) == 0) { 1792 if (tp->t_state == TCPS_SYN_RECEIVED || 1793 (tp->t_flags & TF_NEEDSYN)) 1794 goto step6; 1795 else 1796 goto drop; 1797 } 1798 1799 /* 1800 * Ack processing. 1801 */ 1802 switch (tp->t_state) { 1803 1804 /* 1805 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1806 * ESTABLISHED state and continue processing. 1807 * The ACK was checked above. 1808 */ 1809 case TCPS_SYN_RECEIVED: 1810 1811 tcpstat.tcps_connects++; 1812 soisconnected(so); 1813 /* Do window scaling? */ 1814 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1815 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1816 tp->snd_scale = tp->requested_s_scale; 1817 tp->rcv_scale = tp->request_r_scale; 1818 } 1819 /* 1820 * Upon successful completion of 3-way handshake, 1821 * update cache.CC, pass any queued data to the user, 1822 * and advance state appropriately. 1823 */ 1824 if (tcp_do_rfc1644) { 1825 tao.tao_cc = tp->cc_recv; 1826 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1827 tp->cc_recv, 0); 1828 } 1829 /* 1830 * Make transitions: 1831 * SYN-RECEIVED -> ESTABLISHED 1832 * SYN-RECEIVED* -> FIN-WAIT-1 1833 */ 1834 tp->t_starttime = ticks; 1835 if (tp->t_flags & TF_NEEDFIN) { 1836 tp->t_state = TCPS_FIN_WAIT_1; 1837 tp->t_flags &= ~TF_NEEDFIN; 1838 } else { 1839 tp->t_state = TCPS_ESTABLISHED; 1840 callout_reset(tp->tt_keep, tcp_keepidle, 1841 tcp_timer_keep, tp); 1842 } 1843 /* 1844 * If segment contains data or ACK, will call tcp_reass() 1845 * later; if not, do so now to pass queued data to user. 1846 */ 1847 if (tlen == 0 && (thflags & TH_FIN) == 0) 1848 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1849 (struct mbuf *)0); 1850 tp->snd_wl1 = th->th_seq - 1; 1851 /* FALLTHROUGH */ 1852 1853 /* 1854 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1855 * ACKs. If the ack is in the range 1856 * tp->snd_una < th->th_ack <= tp->snd_max 1857 * then advance tp->snd_una to th->th_ack and drop 1858 * data from the retransmission queue. If this ACK reflects 1859 * more up to date window information we update our window information. 1860 */ 1861 case TCPS_ESTABLISHED: 1862 case TCPS_FIN_WAIT_1: 1863 case TCPS_FIN_WAIT_2: 1864 case TCPS_CLOSE_WAIT: 1865 case TCPS_CLOSING: 1866 case TCPS_LAST_ACK: 1867 case TCPS_TIME_WAIT: 1868 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1869 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1870 if (tlen == 0 && tiwin == tp->snd_wnd) { 1871 tcpstat.tcps_rcvdupack++; 1872 /* 1873 * If we have outstanding data (other than 1874 * a window probe), this is a completely 1875 * duplicate ack (ie, window info didn't 1876 * change), the ack is the biggest we've 1877 * seen and we've seen exactly our rexmt 1878 * threshhold of them, assume a packet 1879 * has been dropped and retransmit it. 1880 * Kludge snd_nxt & the congestion 1881 * window so we send only this one 1882 * packet. 1883 * 1884 * We know we're losing at the current 1885 * window size so do congestion avoidance 1886 * (set ssthresh to half the current window 1887 * and pull our congestion window back to 1888 * the new ssthresh). 1889 * 1890 * Dup acks mean that packets have left the 1891 * network (they're now cached at the receiver) 1892 * so bump cwnd by the amount in the receiver 1893 * to keep a constant cwnd packets in the 1894 * network. 1895 */ 1896 if (!callout_active(tp->tt_rexmt) || 1897 th->th_ack != tp->snd_una) 1898 tp->t_dupacks = 0; 1899 else if (++tp->t_dupacks > tcprexmtthresh || 1900 (tcp_do_newreno && 1901 IN_FASTRECOVERY(tp))) { 1902 tp->snd_cwnd += tp->t_maxseg; 1903 (void) tcp_output(tp); 1904 goto drop; 1905 } else if (tp->t_dupacks == tcprexmtthresh) { 1906 tcp_seq onxt = tp->snd_nxt; 1907 u_int win; 1908 if (tcp_do_newreno && 1909 SEQ_LEQ(th->th_ack, 1910 tp->snd_recover)) { 1911 tp->t_dupacks = 0; 1912 break; 1913 } 1914 win = min(tp->snd_wnd, tp->snd_cwnd) / 1915 2 / tp->t_maxseg; 1916 if (win < 2) 1917 win = 2; 1918 tp->snd_ssthresh = win * tp->t_maxseg; 1919 ENTER_FASTRECOVERY(tp); 1920 tp->snd_recover = tp->snd_max; 1921 callout_stop(tp->tt_rexmt); 1922 tp->t_rtttime = 0; 1923 tp->snd_nxt = th->th_ack; 1924 tp->snd_cwnd = tp->t_maxseg; 1925 (void) tcp_output(tp); 1926 KASSERT(tp->snd_limited <= 2, 1927 ("tp->snd_limited too big")); 1928 tp->snd_cwnd = tp->snd_ssthresh + 1929 tp->t_maxseg * 1930 (tp->t_dupacks - tp->snd_limited); 1931 if (SEQ_GT(onxt, tp->snd_nxt)) 1932 tp->snd_nxt = onxt; 1933 goto drop; 1934 } else if (tcp_do_rfc3042) { 1935 u_long oldcwnd = tp->snd_cwnd; 1936 tcp_seq oldsndmax = tp->snd_max; 1937 u_int sent; 1938 1939 KASSERT(tp->t_dupacks == 1 || 1940 tp->t_dupacks == 2, 1941 ("dupacks not 1 or 2")); 1942 if (tp->t_dupacks == 1) 1943 tp->snd_limited = 0; 1944 tp->snd_cwnd = 1945 (tp->snd_nxt - tp->snd_una) + 1946 (tp->t_dupacks - tp->snd_limited) * 1947 tp->t_maxseg; 1948 (void) tcp_output(tp); 1949 sent = tp->snd_max - oldsndmax; 1950 if (sent > tp->t_maxseg) { 1951 KASSERT((tp->t_dupacks == 2 && 1952 tp->snd_limited == 0) || 1953 (sent == tp->t_maxseg + 1 && 1954 tp->t_flags & TF_SENTFIN), 1955 ("sent too much")); 1956 tp->snd_limited = 2; 1957 } else if (sent > 0) 1958 ++tp->snd_limited; 1959 tp->snd_cwnd = oldcwnd; 1960 goto drop; 1961 } 1962 } else 1963 tp->t_dupacks = 0; 1964 break; 1965 } 1966 1967 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1968 1969 /* 1970 * If the congestion window was inflated to account 1971 * for the other side's cached packets, retract it. 1972 */ 1973 if (tcp_do_newreno) { 1974 if (IN_FASTRECOVERY(tp)) { 1975 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1976 tcp_newreno_partial_ack(tp, th); 1977 } else { 1978 /* 1979 * Window inflation should have left us 1980 * with approximately snd_ssthresh 1981 * outstanding data. 1982 * But in case we would be inclined to 1983 * send a burst, better to do it via 1984 * the slow start mechanism. 1985 */ 1986 if (SEQ_GT(th->th_ack + 1987 tp->snd_ssthresh, 1988 tp->snd_max)) 1989 tp->snd_cwnd = tp->snd_max - 1990 th->th_ack + 1991 tp->t_maxseg; 1992 else 1993 tp->snd_cwnd = tp->snd_ssthresh; 1994 } 1995 } 1996 } else { 1997 if (tp->t_dupacks >= tcprexmtthresh && 1998 tp->snd_cwnd > tp->snd_ssthresh) 1999 tp->snd_cwnd = tp->snd_ssthresh; 2000 } 2001 tp->t_dupacks = 0; 2002 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2003 tcpstat.tcps_rcvacktoomuch++; 2004 goto dropafterack; 2005 } 2006 /* 2007 * If we reach this point, ACK is not a duplicate, 2008 * i.e., it ACKs something we sent. 2009 */ 2010 if (tp->t_flags & TF_NEEDSYN) { 2011 /* 2012 * T/TCP: Connection was half-synchronized, and our 2013 * SYN has been ACK'd (so connection is now fully 2014 * synchronized). Go to non-starred state, 2015 * increment snd_una for ACK of SYN, and check if 2016 * we can do window scaling. 2017 */ 2018 tp->t_flags &= ~TF_NEEDSYN; 2019 tp->snd_una++; 2020 /* Do window scaling? */ 2021 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2022 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2023 tp->snd_scale = tp->requested_s_scale; 2024 tp->rcv_scale = tp->request_r_scale; 2025 } 2026 } 2027 2028process_ACK: 2029 acked = th->th_ack - tp->snd_una; 2030 tcpstat.tcps_rcvackpack++; 2031 tcpstat.tcps_rcvackbyte += acked; 2032 2033 /* 2034 * If we just performed our first retransmit, and the ACK 2035 * arrives within our recovery window, then it was a mistake 2036 * to do the retransmit in the first place. Recover our 2037 * original cwnd and ssthresh, and proceed to transmit where 2038 * we left off. 2039 */ 2040 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2041 ++tcpstat.tcps_sndrexmitbad; 2042 tp->snd_cwnd = tp->snd_cwnd_prev; 2043 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2044 tp->snd_recover = tp->snd_recover_prev; 2045 if (tp->t_flags & TF_WASFRECOVERY) 2046 ENTER_FASTRECOVERY(tp); 2047 tp->snd_nxt = tp->snd_max; 2048 tp->t_badrxtwin = 0; /* XXX probably not required */ 2049 } 2050 2051 /* 2052 * If we have a timestamp reply, update smoothed 2053 * round trip time. If no timestamp is present but 2054 * transmit timer is running and timed sequence 2055 * number was acked, update smoothed round trip time. 2056 * Since we now have an rtt measurement, cancel the 2057 * timer backoff (cf., Phil Karn's retransmit alg.). 2058 * Recompute the initial retransmit timer. 2059 * 2060 * Some boxes send broken timestamp replies 2061 * during the SYN+ACK phase, ignore 2062 * timestamps of 0 or we could calculate a 2063 * huge RTT and blow up the retransmit timer. 2064 */ 2065 if ((to.to_flags & TOF_TS) != 0 && 2066 to.to_tsecr) { 2067 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2068 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2069 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2070 } 2071 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2072 2073 /* 2074 * If all outstanding data is acked, stop retransmit 2075 * timer and remember to restart (more output or persist). 2076 * If there is more data to be acked, restart retransmit 2077 * timer, using current (possibly backed-off) value. 2078 */ 2079 if (th->th_ack == tp->snd_max) { 2080 callout_stop(tp->tt_rexmt); 2081 needoutput = 1; 2082 } else if (!callout_active(tp->tt_persist)) 2083 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2084 tcp_timer_rexmt, tp); 2085 2086 /* 2087 * If no data (only SYN) was ACK'd, 2088 * skip rest of ACK processing. 2089 */ 2090 if (acked == 0) 2091 goto step6; 2092 2093 /* 2094 * When new data is acked, open the congestion window. 2095 * If the window gives us less than ssthresh packets 2096 * in flight, open exponentially (maxseg per packet). 2097 * Otherwise open linearly: maxseg per window 2098 * (maxseg^2 / cwnd per packet). 2099 */ 2100 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 2101 register u_int cw = tp->snd_cwnd; 2102 register u_int incr = tp->t_maxseg; 2103 if (cw > tp->snd_ssthresh) 2104 incr = incr * incr / cw; 2105 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2106 } 2107 if (acked > so->so_snd.sb_cc) { 2108 tp->snd_wnd -= so->so_snd.sb_cc; 2109 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2110 ourfinisacked = 1; 2111 } else { 2112 sbdrop(&so->so_snd, acked); 2113 tp->snd_wnd -= acked; 2114 ourfinisacked = 0; 2115 } 2116 sowwakeup(so); 2117 /* detect una wraparound */ 2118 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 2119 SEQ_GT(tp->snd_una, tp->snd_recover) && 2120 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2121 tp->snd_recover = th->th_ack - 1; 2122 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 2123 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2124 EXIT_FASTRECOVERY(tp); 2125 tp->snd_una = th->th_ack; 2126 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2127 tp->snd_nxt = tp->snd_una; 2128 2129 switch (tp->t_state) { 2130 2131 /* 2132 * In FIN_WAIT_1 STATE in addition to the processing 2133 * for the ESTABLISHED state if our FIN is now acknowledged 2134 * then enter FIN_WAIT_2. 2135 */ 2136 case TCPS_FIN_WAIT_1: 2137 if (ourfinisacked) { 2138 /* 2139 * If we can't receive any more 2140 * data, then closing user can proceed. 2141 * Starting the timer is contrary to the 2142 * specification, but if we don't get a FIN 2143 * we'll hang forever. 2144 */ 2145 /* XXXjl 2146 * we should release the tp also, and use a 2147 * compressed state. 2148 */ 2149 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2150 soisdisconnected(so); 2151 callout_reset(tp->tt_2msl, tcp_maxidle, 2152 tcp_timer_2msl, tp); 2153 } 2154 tp->t_state = TCPS_FIN_WAIT_2; 2155 } 2156 break; 2157 2158 /* 2159 * In CLOSING STATE in addition to the processing for 2160 * the ESTABLISHED state if the ACK acknowledges our FIN 2161 * then enter the TIME-WAIT state, otherwise ignore 2162 * the segment. 2163 */ 2164 case TCPS_CLOSING: 2165 if (ourfinisacked) { 2166 KASSERT(headlocked, ("headlocked")); 2167 tcp_twstart(tp); 2168 INP_INFO_WUNLOCK(&tcbinfo); 2169 m_freem(m); 2170 return; 2171 } 2172 break; 2173 2174 /* 2175 * In LAST_ACK, we may still be waiting for data to drain 2176 * and/or to be acked, as well as for the ack of our FIN. 2177 * If our FIN is now acknowledged, delete the TCB, 2178 * enter the closed state and return. 2179 */ 2180 case TCPS_LAST_ACK: 2181 if (ourfinisacked) { 2182 tp = tcp_close(tp); 2183 goto drop; 2184 } 2185 break; 2186 2187 /* 2188 * In TIME_WAIT state the only thing that should arrive 2189 * is a retransmission of the remote FIN. Acknowledge 2190 * it and restart the finack timer. 2191 */ 2192 case TCPS_TIME_WAIT: 2193 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2194 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2195 tcp_timer_2msl, tp); 2196 goto dropafterack; 2197 } 2198 } 2199 2200step6: 2201 /* 2202 * Update window information. 2203 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2204 */ 2205 if ((thflags & TH_ACK) && 2206 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2207 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2208 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2209 /* keep track of pure window updates */ 2210 if (tlen == 0 && 2211 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2212 tcpstat.tcps_rcvwinupd++; 2213 tp->snd_wnd = tiwin; 2214 tp->snd_wl1 = th->th_seq; 2215 tp->snd_wl2 = th->th_ack; 2216 if (tp->snd_wnd > tp->max_sndwnd) 2217 tp->max_sndwnd = tp->snd_wnd; 2218 needoutput = 1; 2219 } 2220 2221 /* 2222 * Process segments with URG. 2223 */ 2224 if ((thflags & TH_URG) && th->th_urp && 2225 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2226 /* 2227 * This is a kludge, but if we receive and accept 2228 * random urgent pointers, we'll crash in 2229 * soreceive. It's hard to imagine someone 2230 * actually wanting to send this much urgent data. 2231 */ 2232 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2233 th->th_urp = 0; /* XXX */ 2234 thflags &= ~TH_URG; /* XXX */ 2235 goto dodata; /* XXX */ 2236 } 2237 /* 2238 * If this segment advances the known urgent pointer, 2239 * then mark the data stream. This should not happen 2240 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2241 * a FIN has been received from the remote side. 2242 * In these states we ignore the URG. 2243 * 2244 * According to RFC961 (Assigned Protocols), 2245 * the urgent pointer points to the last octet 2246 * of urgent data. We continue, however, 2247 * to consider it to indicate the first octet 2248 * of data past the urgent section as the original 2249 * spec states (in one of two places). 2250 */ 2251 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2252 tp->rcv_up = th->th_seq + th->th_urp; 2253 so->so_oobmark = so->so_rcv.sb_cc + 2254 (tp->rcv_up - tp->rcv_nxt) - 1; 2255 if (so->so_oobmark == 0) { 2256 SOCKBUF_LOCK(&so->so_rcv); 2257 so->so_rcv.sb_state |= SBS_RCVATMARK; 2258 SOCKBUF_UNLOCK(&so->so_rcv); 2259 } 2260 sohasoutofband(so); 2261 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2262 } 2263 /* 2264 * Remove out of band data so doesn't get presented to user. 2265 * This can happen independent of advancing the URG pointer, 2266 * but if two URG's are pending at once, some out-of-band 2267 * data may creep in... ick. 2268 */ 2269 if (th->th_urp <= (u_long)tlen && 2270 !(so->so_options & SO_OOBINLINE)) { 2271 /* hdr drop is delayed */ 2272 tcp_pulloutofband(so, th, m, drop_hdrlen); 2273 } 2274 } else { 2275 /* 2276 * If no out of band data is expected, 2277 * pull receive urgent pointer along 2278 * with the receive window. 2279 */ 2280 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2281 tp->rcv_up = tp->rcv_nxt; 2282 } 2283dodata: /* XXX */ 2284 KASSERT(headlocked, ("headlocked")); 2285 /* 2286 * Process the segment text, merging it into the TCP sequencing queue, 2287 * and arranging for acknowledgment of receipt if necessary. 2288 * This process logically involves adjusting tp->rcv_wnd as data 2289 * is presented to the user (this happens in tcp_usrreq.c, 2290 * case PRU_RCVD). If a FIN has already been received on this 2291 * connection then we just ignore the text. 2292 */ 2293 if ((tlen || (thflags & TH_FIN)) && 2294 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2295 m_adj(m, drop_hdrlen); /* delayed header drop */ 2296 /* 2297 * Insert segment which includes th into TCP reassembly queue 2298 * with control block tp. Set thflags to whether reassembly now 2299 * includes a segment with FIN. This handles the common case 2300 * inline (segment is the next to be received on an established 2301 * connection, and the queue is empty), avoiding linkage into 2302 * and removal from the queue and repetition of various 2303 * conversions. 2304 * Set DELACK for segments received in order, but ack 2305 * immediately when segments are out of order (so 2306 * fast retransmit can work). 2307 */ 2308 if (th->th_seq == tp->rcv_nxt && 2309 LIST_EMPTY(&tp->t_segq) && 2310 TCPS_HAVEESTABLISHED(tp->t_state)) { 2311 if (DELAY_ACK(tp)) 2312 tp->t_flags |= TF_DELACK; 2313 else 2314 tp->t_flags |= TF_ACKNOW; 2315 tp->rcv_nxt += tlen; 2316 thflags = th->th_flags & TH_FIN; 2317 tcpstat.tcps_rcvpack++; 2318 tcpstat.tcps_rcvbyte += tlen; 2319 ND6_HINT(tp); 2320 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2321 m_freem(m); 2322 else 2323 sbappendstream(&so->so_rcv, m); 2324 sorwakeup(so); 2325 } else { 2326 thflags = tcp_reass(tp, th, &tlen, m); 2327 tp->t_flags |= TF_ACKNOW; 2328 } 2329 2330 /* 2331 * Note the amount of data that peer has sent into 2332 * our window, in order to estimate the sender's 2333 * buffer size. 2334 */ 2335 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2336 } else { 2337 m_freem(m); 2338 thflags &= ~TH_FIN; 2339 } 2340 2341 /* 2342 * If FIN is received ACK the FIN and let the user know 2343 * that the connection is closing. 2344 */ 2345 if (thflags & TH_FIN) { 2346 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2347 socantrcvmore(so); 2348 /* 2349 * If connection is half-synchronized 2350 * (ie NEEDSYN flag on) then delay ACK, 2351 * so it may be piggybacked when SYN is sent. 2352 * Otherwise, since we received a FIN then no 2353 * more input can be expected, send ACK now. 2354 */ 2355 if (tp->t_flags & TF_NEEDSYN) 2356 tp->t_flags |= TF_DELACK; 2357 else 2358 tp->t_flags |= TF_ACKNOW; 2359 tp->rcv_nxt++; 2360 } 2361 switch (tp->t_state) { 2362 2363 /* 2364 * In SYN_RECEIVED and ESTABLISHED STATES 2365 * enter the CLOSE_WAIT state. 2366 */ 2367 case TCPS_SYN_RECEIVED: 2368 tp->t_starttime = ticks; 2369 /*FALLTHROUGH*/ 2370 case TCPS_ESTABLISHED: 2371 tp->t_state = TCPS_CLOSE_WAIT; 2372 break; 2373 2374 /* 2375 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2376 * enter the CLOSING state. 2377 */ 2378 case TCPS_FIN_WAIT_1: 2379 tp->t_state = TCPS_CLOSING; 2380 break; 2381 2382 /* 2383 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2384 * starting the time-wait timer, turning off the other 2385 * standard timers. 2386 */ 2387 case TCPS_FIN_WAIT_2: 2388 KASSERT(headlocked == 1, ("headlocked should be 1")); 2389 tcp_twstart(tp); 2390 INP_INFO_WUNLOCK(&tcbinfo); 2391 return; 2392 2393 /* 2394 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2395 */ 2396 case TCPS_TIME_WAIT: 2397 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2398 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2399 tcp_timer_2msl, tp); 2400 break; 2401 } 2402 } 2403 INP_INFO_WUNLOCK(&tcbinfo); 2404#ifdef TCPDEBUG 2405 if (so->so_options & SO_DEBUG) 2406 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2407 &tcp_savetcp, 0); 2408#endif 2409 2410 /* 2411 * Return any desired output. 2412 */ 2413 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2414 (void) tcp_output(tp); 2415 2416check_delack: 2417 if (tp->t_flags & TF_DELACK) { 2418 tp->t_flags &= ~TF_DELACK; 2419 callout_reset(tp->tt_delack, tcp_delacktime, 2420 tcp_timer_delack, tp); 2421 } 2422 INP_UNLOCK(inp); 2423 return; 2424 2425dropafterack: 2426 /* 2427 * Generate an ACK dropping incoming segment if it occupies 2428 * sequence space, where the ACK reflects our state. 2429 * 2430 * We can now skip the test for the RST flag since all 2431 * paths to this code happen after packets containing 2432 * RST have been dropped. 2433 * 2434 * In the SYN-RECEIVED state, don't send an ACK unless the 2435 * segment we received passes the SYN-RECEIVED ACK test. 2436 * If it fails send a RST. This breaks the loop in the 2437 * "LAND" DoS attack, and also prevents an ACK storm 2438 * between two listening ports that have been sent forged 2439 * SYN segments, each with the source address of the other. 2440 */ 2441 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2442 (SEQ_GT(tp->snd_una, th->th_ack) || 2443 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2444 rstreason = BANDLIM_RST_OPENPORT; 2445 goto dropwithreset; 2446 } 2447#ifdef TCPDEBUG 2448 if (so->so_options & SO_DEBUG) 2449 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2450 &tcp_savetcp, 0); 2451#endif 2452 KASSERT(headlocked, ("headlocked should be 1")); 2453 INP_INFO_WUNLOCK(&tcbinfo); 2454 m_freem(m); 2455 tp->t_flags |= TF_ACKNOW; 2456 (void) tcp_output(tp); 2457 INP_UNLOCK(inp); 2458 return; 2459 2460dropwithreset: 2461 /* 2462 * Generate a RST, dropping incoming segment. 2463 * Make ACK acceptable to originator of segment. 2464 * Don't bother to respond if destination was broadcast/multicast. 2465 */ 2466 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2467 goto drop; 2468 if (isipv6) { 2469 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2470 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2471 goto drop; 2472 } else { 2473 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2474 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2475 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2476 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2477 goto drop; 2478 } 2479 /* IPv6 anycast check is done at tcp6_input() */ 2480 2481 /* 2482 * Perform bandwidth limiting. 2483 */ 2484 if (badport_bandlim(rstreason) < 0) 2485 goto drop; 2486 2487#ifdef TCPDEBUG 2488 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2489 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2490 &tcp_savetcp, 0); 2491#endif 2492 2493 if (thflags & TH_ACK) 2494 /* mtod() below is safe as long as hdr dropping is delayed */ 2495 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2496 TH_RST); 2497 else { 2498 if (thflags & TH_SYN) 2499 tlen++; 2500 /* mtod() below is safe as long as hdr dropping is delayed */ 2501 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2502 (tcp_seq)0, TH_RST|TH_ACK); 2503 } 2504 2505 if (tp) 2506 INP_UNLOCK(inp); 2507 if (headlocked) 2508 INP_INFO_WUNLOCK(&tcbinfo); 2509 return; 2510 2511drop: 2512 /* 2513 * Drop space held by incoming segment and return. 2514 */ 2515#ifdef TCPDEBUG 2516 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2517 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2518 &tcp_savetcp, 0); 2519#endif 2520 if (tp) 2521 INP_UNLOCK(inp); 2522 m_freem(m); 2523 if (headlocked) 2524 INP_INFO_WUNLOCK(&tcbinfo); 2525 return; 2526} 2527 2528/* 2529 * Parse TCP options and place in tcpopt. 2530 */ 2531static void 2532tcp_dooptions(to, cp, cnt, is_syn) 2533 struct tcpopt *to; 2534 u_char *cp; 2535 int cnt; 2536 int is_syn; 2537{ 2538 int opt, optlen; 2539 2540 to->to_flags = 0; 2541 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2542 opt = cp[0]; 2543 if (opt == TCPOPT_EOL) 2544 break; 2545 if (opt == TCPOPT_NOP) 2546 optlen = 1; 2547 else { 2548 if (cnt < 2) 2549 break; 2550 optlen = cp[1]; 2551 if (optlen < 2 || optlen > cnt) 2552 break; 2553 } 2554 switch (opt) { 2555 case TCPOPT_MAXSEG: 2556 if (optlen != TCPOLEN_MAXSEG) 2557 continue; 2558 if (!is_syn) 2559 continue; 2560 to->to_flags |= TOF_MSS; 2561 bcopy((char *)cp + 2, 2562 (char *)&to->to_mss, sizeof(to->to_mss)); 2563 to->to_mss = ntohs(to->to_mss); 2564 break; 2565 case TCPOPT_WINDOW: 2566 if (optlen != TCPOLEN_WINDOW) 2567 continue; 2568 if (! is_syn) 2569 continue; 2570 to->to_flags |= TOF_SCALE; 2571 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2572 break; 2573 case TCPOPT_TIMESTAMP: 2574 if (optlen != TCPOLEN_TIMESTAMP) 2575 continue; 2576 to->to_flags |= TOF_TS; 2577 bcopy((char *)cp + 2, 2578 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2579 to->to_tsval = ntohl(to->to_tsval); 2580 bcopy((char *)cp + 6, 2581 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2582 to->to_tsecr = ntohl(to->to_tsecr); 2583 break; 2584 case TCPOPT_CC: 2585 if (optlen != TCPOLEN_CC) 2586 continue; 2587 to->to_flags |= TOF_CC; 2588 bcopy((char *)cp + 2, 2589 (char *)&to->to_cc, sizeof(to->to_cc)); 2590 to->to_cc = ntohl(to->to_cc); 2591 break; 2592 case TCPOPT_CCNEW: 2593 if (optlen != TCPOLEN_CC) 2594 continue; 2595 if (!is_syn) 2596 continue; 2597 to->to_flags |= TOF_CCNEW; 2598 bcopy((char *)cp + 2, 2599 (char *)&to->to_cc, sizeof(to->to_cc)); 2600 to->to_cc = ntohl(to->to_cc); 2601 break; 2602 case TCPOPT_CCECHO: 2603 if (optlen != TCPOLEN_CC) 2604 continue; 2605 if (!is_syn) 2606 continue; 2607 to->to_flags |= TOF_CCECHO; 2608 bcopy((char *)cp + 2, 2609 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2610 to->to_ccecho = ntohl(to->to_ccecho); 2611 break; 2612#ifdef TCP_SIGNATURE 2613 /* 2614 * XXX In order to reply to a host which has set the 2615 * TCP_SIGNATURE option in its initial SYN, we have to 2616 * record the fact that the option was observed here 2617 * for the syncache code to perform the correct response. 2618 */ 2619 case TCPOPT_SIGNATURE: 2620 if (optlen != TCPOLEN_SIGNATURE) 2621 continue; 2622 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2623 break; 2624#endif 2625 default: 2626 continue; 2627 } 2628 } 2629} 2630 2631/* 2632 * Pull out of band byte out of a segment so 2633 * it doesn't appear in the user's data queue. 2634 * It is still reflected in the segment length for 2635 * sequencing purposes. 2636 */ 2637static void 2638tcp_pulloutofband(so, th, m, off) 2639 struct socket *so; 2640 struct tcphdr *th; 2641 register struct mbuf *m; 2642 int off; /* delayed to be droped hdrlen */ 2643{ 2644 int cnt = off + th->th_urp - 1; 2645 2646 while (cnt >= 0) { 2647 if (m->m_len > cnt) { 2648 char *cp = mtod(m, caddr_t) + cnt; 2649 struct tcpcb *tp = sototcpcb(so); 2650 2651 tp->t_iobc = *cp; 2652 tp->t_oobflags |= TCPOOB_HAVEDATA; 2653 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2654 m->m_len--; 2655 if (m->m_flags & M_PKTHDR) 2656 m->m_pkthdr.len--; 2657 return; 2658 } 2659 cnt -= m->m_len; 2660 m = m->m_next; 2661 if (m == 0) 2662 break; 2663 } 2664 panic("tcp_pulloutofband"); 2665} 2666 2667/* 2668 * Collect new round-trip time estimate 2669 * and update averages and current timeout. 2670 */ 2671static void 2672tcp_xmit_timer(tp, rtt) 2673 register struct tcpcb *tp; 2674 int rtt; 2675{ 2676 register int delta; 2677 2678 tcpstat.tcps_rttupdated++; 2679 tp->t_rttupdated++; 2680 if (tp->t_srtt != 0) { 2681 /* 2682 * srtt is stored as fixed point with 5 bits after the 2683 * binary point (i.e., scaled by 8). The following magic 2684 * is equivalent to the smoothing algorithm in rfc793 with 2685 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2686 * point). Adjust rtt to origin 0. 2687 */ 2688 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2689 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2690 2691 if ((tp->t_srtt += delta) <= 0) 2692 tp->t_srtt = 1; 2693 2694 /* 2695 * We accumulate a smoothed rtt variance (actually, a 2696 * smoothed mean difference), then set the retransmit 2697 * timer to smoothed rtt + 4 times the smoothed variance. 2698 * rttvar is stored as fixed point with 4 bits after the 2699 * binary point (scaled by 16). The following is 2700 * equivalent to rfc793 smoothing with an alpha of .75 2701 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2702 * rfc793's wired-in beta. 2703 */ 2704 if (delta < 0) 2705 delta = -delta; 2706 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2707 if ((tp->t_rttvar += delta) <= 0) 2708 tp->t_rttvar = 1; 2709 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2710 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2711 } else { 2712 /* 2713 * No rtt measurement yet - use the unsmoothed rtt. 2714 * Set the variance to half the rtt (so our first 2715 * retransmit happens at 3*rtt). 2716 */ 2717 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2718 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2719 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2720 } 2721 tp->t_rtttime = 0; 2722 tp->t_rxtshift = 0; 2723 2724 /* 2725 * the retransmit should happen at rtt + 4 * rttvar. 2726 * Because of the way we do the smoothing, srtt and rttvar 2727 * will each average +1/2 tick of bias. When we compute 2728 * the retransmit timer, we want 1/2 tick of rounding and 2729 * 1 extra tick because of +-1/2 tick uncertainty in the 2730 * firing of the timer. The bias will give us exactly the 2731 * 1.5 tick we need. But, because the bias is 2732 * statistical, we have to test that we don't drop below 2733 * the minimum feasible timer (which is 2 ticks). 2734 */ 2735 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2736 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2737 2738 /* 2739 * We received an ack for a packet that wasn't retransmitted; 2740 * it is probably safe to discard any error indications we've 2741 * received recently. This isn't quite right, but close enough 2742 * for now (a route might have failed after we sent a segment, 2743 * and the return path might not be symmetrical). 2744 */ 2745 tp->t_softerror = 0; 2746} 2747 2748/* 2749 * Determine a reasonable value for maxseg size. 2750 * If the route is known, check route for mtu. 2751 * If none, use an mss that can be handled on the outgoing 2752 * interface without forcing IP to fragment; if bigger than 2753 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2754 * to utilize large mbufs. If no route is found, route has no mtu, 2755 * or the destination isn't local, use a default, hopefully conservative 2756 * size (usually 512 or the default IP max size, but no more than the mtu 2757 * of the interface), as we can't discover anything about intervening 2758 * gateways or networks. We also initialize the congestion/slow start 2759 * window to be a single segment if the destination isn't local. 2760 * While looking at the routing entry, we also initialize other path-dependent 2761 * parameters from pre-set or cached values in the routing entry. 2762 * 2763 * Also take into account the space needed for options that we 2764 * send regularly. Make maxseg shorter by that amount to assure 2765 * that we can send maxseg amount of data even when the options 2766 * are present. Store the upper limit of the length of options plus 2767 * data in maxopd. 2768 * 2769 * 2770 * In case of T/TCP, we call this routine during implicit connection 2771 * setup as well (offer = -1), to initialize maxseg from the cached 2772 * MSS of our peer. 2773 * 2774 * NOTE that this routine is only called when we process an incoming 2775 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2776 */ 2777void 2778tcp_mss(tp, offer) 2779 struct tcpcb *tp; 2780 int offer; 2781{ 2782 int rtt, mss; 2783 u_long bufsize; 2784 u_long maxmtu; 2785 struct inpcb *inp = tp->t_inpcb; 2786 struct socket *so; 2787 struct hc_metrics_lite metrics; 2788 struct rmxp_tao tao; 2789 int origoffer = offer; 2790#ifdef INET6 2791 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2792 size_t min_protoh = isipv6 ? 2793 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2794 sizeof (struct tcpiphdr); 2795#else 2796 const size_t min_protoh = sizeof(struct tcpiphdr); 2797#endif 2798 bzero(&tao, sizeof(tao)); 2799 2800 /* initialize */ 2801#ifdef INET6 2802 if (isipv6) { 2803 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2804 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2805 } else 2806#endif 2807 { 2808 maxmtu = tcp_maxmtu(&inp->inp_inc); 2809 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2810 } 2811 so = inp->inp_socket; 2812 2813 /* 2814 * no route to sender, stay with default mss and return 2815 */ 2816 if (maxmtu == 0) 2817 return; 2818 2819 /* what have we got? */ 2820 switch (offer) { 2821 case 0: 2822 /* 2823 * Offer == 0 means that there was no MSS on the SYN 2824 * segment, in this case we use tcp_mssdflt. 2825 */ 2826 offer = 2827#ifdef INET6 2828 isipv6 ? tcp_v6mssdflt : 2829#endif 2830 tcp_mssdflt; 2831 break; 2832 2833 case -1: 2834 /* 2835 * Offer == -1 means that we didn't receive SYN yet, 2836 * use cached value in that case; 2837 */ 2838 if (tcp_do_rfc1644) 2839 tcp_hc_gettao(&inp->inp_inc, &tao); 2840 if (tao.tao_mssopt != 0) 2841 offer = tao.tao_mssopt; 2842 /* FALLTHROUGH */ 2843 2844 default: 2845 /* 2846 * Prevent DoS attack with too small MSS. Round up 2847 * to at least minmss. 2848 */ 2849 offer = max(offer, tcp_minmss); 2850 /* 2851 * Sanity check: make sure that maxopd will be large 2852 * enough to allow some data on segments even if the 2853 * all the option space is used (40bytes). Otherwise 2854 * funny things may happen in tcp_output. 2855 */ 2856 offer = max(offer, 64); 2857 if (tcp_do_rfc1644) 2858 tcp_hc_updatetao(&inp->inp_inc, 2859 TCP_HC_TAO_MSSOPT, 0, offer); 2860 } 2861 2862 /* 2863 * rmx information is now retrieved from tcp_hostcache 2864 */ 2865 tcp_hc_get(&inp->inp_inc, &metrics); 2866 2867 /* 2868 * if there's a discovered mtu int tcp hostcache, use it 2869 * else, use the link mtu. 2870 */ 2871 if (metrics.rmx_mtu) 2872 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2873 else { 2874#ifdef INET6 2875 if (isipv6) { 2876 mss = maxmtu - min_protoh; 2877 if (!path_mtu_discovery && 2878 !in6_localaddr(&inp->in6p_faddr)) 2879 mss = min(mss, tcp_v6mssdflt); 2880 } else 2881#endif 2882 { 2883 mss = maxmtu - min_protoh; 2884 if (!path_mtu_discovery && 2885 !in_localaddr(inp->inp_faddr)) 2886 mss = min(mss, tcp_mssdflt); 2887 } 2888 } 2889 mss = min(mss, offer); 2890 2891 /* 2892 * maxopd stores the maximum length of data AND options 2893 * in a segment; maxseg is the amount of data in a normal 2894 * segment. We need to store this value (maxopd) apart 2895 * from maxseg, because now every segment carries options 2896 * and thus we normally have somewhat less data in segments. 2897 */ 2898 tp->t_maxopd = mss; 2899 2900 /* 2901 * In case of T/TCP, origoffer==-1 indicates, that no segments 2902 * were received yet. In this case we just guess, otherwise 2903 * we do the same as before T/TCP. 2904 */ 2905 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2906 (origoffer == -1 || 2907 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2908 mss -= TCPOLEN_TSTAMP_APPA; 2909 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2910 (origoffer == -1 || 2911 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2912 mss -= TCPOLEN_CC_APPA; 2913 tp->t_maxseg = mss; 2914 2915#if (MCLBYTES & (MCLBYTES - 1)) == 0 2916 if (mss > MCLBYTES) 2917 mss &= ~(MCLBYTES-1); 2918#else 2919 if (mss > MCLBYTES) 2920 mss = mss / MCLBYTES * MCLBYTES; 2921#endif 2922 tp->t_maxseg = mss; 2923 2924 /* 2925 * If there's a pipesize, change the socket buffer to that size, 2926 * don't change if sb_hiwat is different than default (then it 2927 * has been changed on purpose with setsockopt). 2928 * Make the socket buffers an integral number of mss units; 2929 * if the mss is larger than the socket buffer, decrease the mss. 2930 */ 2931 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2932 bufsize = metrics.rmx_sendpipe; 2933 else 2934 bufsize = so->so_snd.sb_hiwat; 2935 if (bufsize < mss) 2936 mss = bufsize; 2937 else { 2938 bufsize = roundup(bufsize, mss); 2939 if (bufsize > sb_max) 2940 bufsize = sb_max; 2941 if (bufsize > so->so_snd.sb_hiwat) 2942 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2943 } 2944 tp->t_maxseg = mss; 2945 2946 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2947 bufsize = metrics.rmx_recvpipe; 2948 else 2949 bufsize = so->so_rcv.sb_hiwat; 2950 if (bufsize > mss) { 2951 bufsize = roundup(bufsize, mss); 2952 if (bufsize > sb_max) 2953 bufsize = sb_max; 2954 if (bufsize > so->so_rcv.sb_hiwat) 2955 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2956 } 2957 /* 2958 * While we're here, check the others too 2959 */ 2960 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2961 tp->t_srtt = rtt; 2962 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2963 tcpstat.tcps_usedrtt++; 2964 if (metrics.rmx_rttvar) { 2965 tp->t_rttvar = metrics.rmx_rttvar; 2966 tcpstat.tcps_usedrttvar++; 2967 } else { 2968 /* default variation is +- 1 rtt */ 2969 tp->t_rttvar = 2970 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2971 } 2972 TCPT_RANGESET(tp->t_rxtcur, 2973 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2974 tp->t_rttmin, TCPTV_REXMTMAX); 2975 } 2976 if (metrics.rmx_ssthresh) { 2977 /* 2978 * There's some sort of gateway or interface 2979 * buffer limit on the path. Use this to set 2980 * the slow start threshhold, but set the 2981 * threshold to no less than 2*mss. 2982 */ 2983 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2984 tcpstat.tcps_usedssthresh++; 2985 } 2986 if (metrics.rmx_bandwidth) 2987 tp->snd_bandwidth = metrics.rmx_bandwidth; 2988 2989 /* 2990 * Set the slow-start flight size depending on whether this 2991 * is a local network or not. 2992 * 2993 * Extend this so we cache the cwnd too and retrieve it here. 2994 * Make cwnd even bigger than RFC3390 suggests but only if we 2995 * have previous experience with the remote host. Be careful 2996 * not make cwnd bigger than remote receive window or our own 2997 * send socket buffer. Maybe put some additional upper bound 2998 * on the retrieved cwnd. Should do incremental updates to 2999 * hostcache when cwnd collapses so next connection doesn't 3000 * overloads the path again. 3001 * 3002 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3003 * We currently check only in syncache_socket for that. 3004 */ 3005#define TCP_METRICS_CWND 3006#ifdef TCP_METRICS_CWND 3007 if (metrics.rmx_cwnd) 3008 tp->snd_cwnd = max(mss, 3009 min(metrics.rmx_cwnd / 2, 3010 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3011 else 3012#endif 3013 if (tcp_do_rfc3390) 3014 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3015#ifdef INET6 3016 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3017 (!isipv6 && in_localaddr(inp->inp_faddr))) 3018#else 3019 else if (in_localaddr(inp->inp_faddr)) 3020#endif 3021 tp->snd_cwnd = mss * ss_fltsz_local; 3022 else 3023 tp->snd_cwnd = mss * ss_fltsz; 3024} 3025 3026/* 3027 * Determine the MSS option to send on an outgoing SYN. 3028 */ 3029int 3030tcp_mssopt(inc) 3031 struct in_conninfo *inc; 3032{ 3033 int mss = 0; 3034 u_long maxmtu = 0; 3035 u_long thcmtu = 0; 3036 size_t min_protoh; 3037#ifdef INET6 3038 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3039#endif 3040 3041 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3042 3043#ifdef INET6 3044 if (isipv6) { 3045 mss = tcp_v6mssdflt; 3046 maxmtu = tcp_maxmtu6(inc); 3047 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3048 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3049 } else 3050#endif 3051 { 3052 mss = tcp_mssdflt; 3053 maxmtu = tcp_maxmtu(inc); 3054 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3055 min_protoh = sizeof(struct tcpiphdr); 3056 } 3057 if (maxmtu && thcmtu) 3058 mss = min(maxmtu, thcmtu) - min_protoh; 3059 else if (maxmtu || thcmtu) 3060 mss = max(maxmtu, thcmtu) - min_protoh; 3061 3062 return (mss); 3063} 3064 3065 3066/* 3067 * On a partial ack arrives, force the retransmission of the 3068 * next unacknowledged segment. Do not clear tp->t_dupacks. 3069 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3070 * be started again. 3071 */ 3072static void 3073tcp_newreno_partial_ack(tp, th) 3074 struct tcpcb *tp; 3075 struct tcphdr *th; 3076{ 3077 tcp_seq onxt = tp->snd_nxt; 3078 u_long ocwnd = tp->snd_cwnd; 3079 3080 callout_stop(tp->tt_rexmt); 3081 tp->t_rtttime = 0; 3082 tp->snd_nxt = th->th_ack; 3083 /* 3084 * Set snd_cwnd to one segment beyond acknowledged offset. 3085 * (tp->snd_una has not yet been updated when this function is called.) 3086 */ 3087 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3088 tp->t_flags |= TF_ACKNOW; 3089 (void) tcp_output(tp); 3090 tp->snd_cwnd = ocwnd; 3091 if (SEQ_GT(onxt, tp->snd_nxt)) 3092 tp->snd_nxt = onxt; 3093 /* 3094 * Partial window deflation. Relies on fact that tp->snd_una 3095 * not updated yet. 3096 */ 3097 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3098} 3099 3100/* 3101 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3102 * looking for a pcb in the listen state. Returns 0 otherwise. 3103 */ 3104static int 3105tcp_timewait(tw, to, th, m, tlen) 3106 struct tcptw *tw; 3107 struct tcpopt *to; 3108 struct tcphdr *th; 3109 struct mbuf *m; 3110 int tlen; 3111{ 3112 int thflags; 3113 tcp_seq seq; 3114#ifdef INET6 3115 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3116#else 3117 const int isipv6 = 0; 3118#endif 3119 3120 thflags = th->th_flags; 3121 3122 /* 3123 * NOTE: for FIN_WAIT_2 (to be added later), 3124 * must validate sequence number before accepting RST 3125 */ 3126 3127 /* 3128 * If the segment contains RST: 3129 * Drop the segment - see Stevens, vol. 2, p. 964 and 3130 * RFC 1337. 3131 */ 3132 if (thflags & TH_RST) 3133 goto drop; 3134 3135 /* 3136 * If segment contains a SYN and CC [not CC.NEW] option: 3137 * if connection duration > MSL, drop packet and send RST; 3138 * 3139 * if SEG.CC > CCrecv then is new SYN. 3140 * Complete close and delete TCPCB. Then reprocess 3141 * segment, hoping to find new TCPCB in LISTEN state; 3142 * 3143 * else must be old SYN; drop it. 3144 * else do normal processing. 3145 */ 3146 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3147 if ((ticks - tw->t_starttime) > tcp_msl) 3148 goto reset; 3149 if (CC_GT(to->to_cc, tw->cc_recv)) { 3150 (void) tcp_twclose(tw, 0); 3151 return (1); 3152 } 3153 goto drop; 3154 } 3155 3156#if 0 3157/* PAWS not needed at the moment */ 3158 /* 3159 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3160 * and it's less than ts_recent, drop it. 3161 */ 3162 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3163 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3164 if ((thflags & TH_ACK) == 0) 3165 goto drop; 3166 goto ack; 3167 } 3168 /* 3169 * ts_recent is never updated because we never accept new segments. 3170 */ 3171#endif 3172 3173 /* 3174 * If a new connection request is received 3175 * while in TIME_WAIT, drop the old connection 3176 * and start over if the sequence numbers 3177 * are above the previous ones. 3178 */ 3179 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3180 (void) tcp_twclose(tw, 0); 3181 return (1); 3182 } 3183 3184 /* 3185 * Drop the the segment if it does not contain an ACK. 3186 */ 3187 if ((thflags & TH_ACK) == 0) 3188 goto drop; 3189 3190 /* 3191 * Reset the 2MSL timer if this is a duplicate FIN. 3192 */ 3193 if (thflags & TH_FIN) { 3194 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3195 if (seq + 1 == tw->rcv_nxt) 3196 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3197 } 3198 3199 /* 3200 * Acknowledge the segment if it has data or is not a duplicate ACK. 3201 */ 3202 if (thflags != TH_ACK || tlen != 0 || 3203 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3204 tcp_twrespond(tw, TH_ACK); 3205 goto drop; 3206 3207reset: 3208 /* 3209 * Generate a RST, dropping incoming segment. 3210 * Make ACK acceptable to originator of segment. 3211 * Don't bother to respond if destination was broadcast/multicast. 3212 */ 3213 if (m->m_flags & (M_BCAST|M_MCAST)) 3214 goto drop; 3215 if (isipv6) { 3216 struct ip6_hdr *ip6; 3217 3218 /* IPv6 anycast check is done at tcp6_input() */ 3219 ip6 = mtod(m, struct ip6_hdr *); 3220 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3221 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3222 goto drop; 3223 } else { 3224 struct ip *ip; 3225 3226 ip = mtod(m, struct ip *); 3227 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3228 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3229 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3230 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3231 goto drop; 3232 } 3233 if (thflags & TH_ACK) { 3234 tcp_respond(NULL, 3235 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3236 } else { 3237 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3238 tcp_respond(NULL, 3239 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3240 } 3241 INP_UNLOCK(tw->tw_inpcb); 3242 return (0); 3243 3244drop: 3245 INP_UNLOCK(tw->tw_inpcb); 3246 m_freem(m); 3247 return (0); 3248} 3249