35 */ 36 37#include "opt_ipfw.h" /* for ipfw_fwd */ 38#include "opt_inet6.h" 39#include "opt_ipsec.h" 40#include "opt_tcpdebug.h" 41#include "opt_tcp_input.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/proc.h> /* for proc0 declaration */ 50#include <sys/protosw.h> 51#include <sys/socket.h> 52#include <sys/socketvar.h> 53#include <sys/syslog.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/ip.h> 63#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 64#include <netinet/in_var.h> 65#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 66#include <netinet/in_pcb.h> 67#include <netinet/ip_var.h> 68#ifdef INET6 69#include <netinet/ip6.h> 70#include <netinet/icmp6.h> 71#include <netinet6/nd6.h> 72#include <netinet6/ip6_var.h> 73#include <netinet6/in6_pcb.h> 74#endif 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#ifdef INET6 81#include <netinet6/tcp6_var.h> 82#endif 83#include <netinet/tcpip.h> 84#ifdef TCPDEBUG 85#include <netinet/tcp_debug.h> 86 87u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 88struct tcphdr tcp_savetcp; 89#endif /* TCPDEBUG */ 90 91#ifdef IPSEC 92#include <netinet6/ipsec.h> 93#ifdef INET6 94#include <netinet6/ipsec6.h> 95#endif 96#include <netkey/key.h> 97#endif /*IPSEC*/ 98 99#include <machine/in_cksum.h> 100 101MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 102 103static int tcprexmtthresh = 3; 104tcp_cc tcp_ccgen; 105 106struct tcpstat tcpstat; 107SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 109 110static int log_in_vain = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 112 &log_in_vain, 0, "Log all incoming TCP connections"); 113 114static int blackhole = 0; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 116 &blackhole, 0, "Do not send RST when dropping refused connections"); 117 118int tcp_delack_enabled = 1; 119SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 120 &tcp_delack_enabled, 0, 121 "Delay ACK to try and piggyback it onto a data packet"); 122 123#ifdef TCP_DROP_SYNFIN 124static int drop_synfin = 0; 125SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 127#endif 128 129struct inpcbhead tcb; 130#define tcb6 tcb /* for KAME src sync over BSD*'s */ 131struct inpcbinfo tcbinfo; 132 133static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 134static void tcp_pulloutofband(struct socket *, 135 struct tcphdr *, struct mbuf *, int); 136static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 137 struct mbuf *); 138static void tcp_xmit_timer(struct tcpcb *, int); 139static int tcp_newreno(struct tcpcb *, struct tcphdr *); 140 141/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 142#ifdef INET6 143#define ND6_HINT(tp) \ 144do { \ 145 if ((tp) && (tp)->t_inpcb && \ 146 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 147 (tp)->t_inpcb->in6p_route.ro_rt) \ 148 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 149} while (0) 150#else 151#define ND6_HINT(tp) 152#endif 153 154/* 155 * Indicate whether this ack should be delayed. We can delay the ack if 156 * - delayed acks are enabled and 157 * - there is no delayed ack timer in progress and 158 * - our last ack wasn't a 0-sized window. We never want to delay 159 * the ack that opens up a 0-sized window. 160 */ 161#define DELAY_ACK(tp) \ 162 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 163 (tp->t_flags & TF_RXWIN0SENT) == 0) 164 165static int 166tcp_reass(tp, th, tlenp, m) 167 register struct tcpcb *tp; 168 register struct tcphdr *th; 169 int *tlenp; 170 struct mbuf *m; 171{ 172 struct tseg_qent *q; 173 struct tseg_qent *p = NULL; 174 struct tseg_qent *nq; 175 struct tseg_qent *te; 176 struct socket *so = tp->t_inpcb->inp_socket; 177 int flags; 178 179 /* 180 * Call with th==0 after become established to 181 * force pre-ESTABLISHED data up to user socket. 182 */ 183 if (th == 0) 184 goto present; 185 186 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 187 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 188 M_NOWAIT); 189 if (te == NULL) { 190 tcpstat.tcps_rcvmemdrop++; 191 m_freem(m); 192 return (0); 193 } 194 195 /* 196 * Find a segment which begins after this one does. 197 */ 198 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 199 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 200 break; 201 p = q; 202 } 203 204 /* 205 * If there is a preceding segment, it may provide some of 206 * our data already. If so, drop the data from the incoming 207 * segment. If it provides all of our data, drop us. 208 */ 209 if (p != NULL) { 210 register int i; 211 /* conversion to int (in i) handles seq wraparound */ 212 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 213 if (i > 0) { 214 if (i >= *tlenp) { 215 tcpstat.tcps_rcvduppack++; 216 tcpstat.tcps_rcvdupbyte += *tlenp; 217 m_freem(m); 218 FREE(te, M_TSEGQ); 219 /* 220 * Try to present any queued data 221 * at the left window edge to the user. 222 * This is needed after the 3-WHS 223 * completes. 224 */ 225 goto present; /* ??? */ 226 } 227 m_adj(m, i); 228 *tlenp -= i; 229 th->th_seq += i; 230 } 231 } 232 tcpstat.tcps_rcvoopack++; 233 tcpstat.tcps_rcvoobyte += *tlenp; 234 235 /* 236 * While we overlap succeeding segments trim them or, 237 * if they are completely covered, dequeue them. 238 */ 239 while (q) { 240 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 241 if (i <= 0) 242 break; 243 if (i < q->tqe_len) { 244 q->tqe_th->th_seq += i; 245 q->tqe_len -= i; 246 m_adj(q->tqe_m, i); 247 break; 248 } 249 250 nq = LIST_NEXT(q, tqe_q); 251 LIST_REMOVE(q, tqe_q); 252 m_freem(q->tqe_m); 253 FREE(q, M_TSEGQ); 254 q = nq; 255 } 256 257 /* Insert the new segment queue entry into place. */ 258 te->tqe_m = m; 259 te->tqe_th = th; 260 te->tqe_len = *tlenp; 261 262 if (p == NULL) { 263 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 264 } else { 265 LIST_INSERT_AFTER(p, te, tqe_q); 266 } 267 268present: 269 /* 270 * Present data to user, advancing rcv_nxt through 271 * completed sequence space. 272 */ 273 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 274 return (0); 275 q = LIST_FIRST(&tp->t_segq); 276 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 277 return (0); 278 do { 279 tp->rcv_nxt += q->tqe_len; 280 flags = q->tqe_th->th_flags & TH_FIN; 281 nq = LIST_NEXT(q, tqe_q); 282 LIST_REMOVE(q, tqe_q); 283 if (so->so_state & SS_CANTRCVMORE) 284 m_freem(q->tqe_m); 285 else 286 sbappend(&so->so_rcv, q->tqe_m); 287 FREE(q, M_TSEGQ); 288 q = nq; 289 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 290 ND6_HINT(tp); 291 sorwakeup(so); 292 return (flags); 293} 294 295/* 296 * TCP input routine, follows pages 65-76 of the 297 * protocol specification dated September, 1981 very closely. 298 */ 299#ifdef INET6 300int 301tcp6_input(mp, offp, proto) 302 struct mbuf **mp; 303 int *offp, proto; 304{ 305 register struct mbuf *m = *mp; 306 struct in6_ifaddr *ia6; 307 308 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 309 310 /* 311 * draft-itojun-ipv6-tcp-to-anycast 312 * better place to put this in? 313 */ 314 ia6 = ip6_getdstifaddr(m); 315 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 316 struct ip6_hdr *ip6; 317 318 ip6 = mtod(m, struct ip6_hdr *); 319 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 320 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 321 return IPPROTO_DONE; 322 } 323 324 tcp_input(m, *offp); 325 return IPPROTO_DONE; 326} 327#endif 328 329void 330tcp_input(m, off0) 331 register struct mbuf *m; 332 int off0; 333{ 334 register struct tcphdr *th; 335 register struct ip *ip = NULL; 336 register struct ipovly *ipov; 337 register struct inpcb *inp; 338 u_char *optp = NULL; 339 int optlen = 0; 340 int len, tlen, off; 341 int drop_hdrlen; 342 register struct tcpcb *tp = 0; 343 register int thflags; 344 struct socket *so = 0; 345 int todrop, acked, ourfinisacked, needoutput = 0; 346 u_long tiwin; 347 struct tcpopt to; /* options in this segment */ 348 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 349 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 350#ifdef TCPDEBUG 351 short ostate = 0; 352#endif 353#ifdef INET6 354 struct ip6_hdr *ip6 = NULL; 355 int isipv6; 356#endif /* INET6 */ 357 int rstreason; /* For badport_bandlim accounting purposes */ 358 359#ifdef INET6 360 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 361#endif 362 bzero((char *)&to, sizeof(to)); 363 364 tcpstat.tcps_rcvtotal++; 365 366#ifdef INET6 367 if (isipv6) { 368 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 369 ip6 = mtod(m, struct ip6_hdr *); 370 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 371 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 372 tcpstat.tcps_rcvbadsum++; 373 goto drop; 374 } 375 th = (struct tcphdr *)((caddr_t)ip6 + off0); 376 377 /* 378 * Be proactive about unspecified IPv6 address in source. 379 * As we use all-zero to indicate unbounded/unconnected pcb, 380 * unspecified IPv6 address can be used to confuse us. 381 * 382 * Note that packets with unspecified IPv6 destination is 383 * already dropped in ip6_input. 384 */ 385 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 386 /* XXX stat */ 387 goto drop; 388 } 389 } else 390#endif /* INET6 */ 391 { 392 /* 393 * Get IP and TCP header together in first mbuf. 394 * Note: IP leaves IP header in first mbuf. 395 */ 396 if (off0 > sizeof (struct ip)) { 397 ip_stripoptions(m, (struct mbuf *)0); 398 off0 = sizeof(struct ip); 399 } 400 if (m->m_len < sizeof (struct tcpiphdr)) { 401 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 402 tcpstat.tcps_rcvshort++; 403 return; 404 } 405 } 406 ip = mtod(m, struct ip *); 407 ipov = (struct ipovly *)ip; 408 th = (struct tcphdr *)((caddr_t)ip + off0); 409 tlen = ip->ip_len; 410 411 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 412 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 413 th->th_sum = m->m_pkthdr.csum_data; 414 else 415 th->th_sum = in_pseudo(ip->ip_src.s_addr, 416 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 417 ip->ip_len + IPPROTO_TCP)); 418 th->th_sum ^= 0xffff; 419 } else { 420 /* 421 * Checksum extended TCP header and data. 422 */ 423 len = sizeof (struct ip) + tlen; 424 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 425 ipov->ih_len = (u_short)tlen; 426 ipov->ih_len = htons(ipov->ih_len); 427 th->th_sum = in_cksum(m, len); 428 } 429 if (th->th_sum) { 430 tcpstat.tcps_rcvbadsum++; 431 goto drop; 432 } 433#ifdef INET6 434 /* Re-initialization for later version check */ 435 ip->ip_v = IPVERSION; 436#endif 437 } 438 439 /* 440 * Check that TCP offset makes sense, 441 * pull out TCP options and adjust length. XXX 442 */ 443 off = th->th_off << 2; 444 if (off < sizeof (struct tcphdr) || off > tlen) { 445 tcpstat.tcps_rcvbadoff++; 446 goto drop; 447 } 448 tlen -= off; /* tlen is used instead of ti->ti_len */ 449 if (off > sizeof (struct tcphdr)) { 450#ifdef INET6 451 if (isipv6) { 452 IP6_EXTHDR_CHECK(m, off0, off, ); 453 ip6 = mtod(m, struct ip6_hdr *); 454 th = (struct tcphdr *)((caddr_t)ip6 + off0); 455 } else 456#endif /* INET6 */ 457 { 458 if (m->m_len < sizeof(struct ip) + off) { 459 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 460 tcpstat.tcps_rcvshort++; 461 return; 462 } 463 ip = mtod(m, struct ip *); 464 ipov = (struct ipovly *)ip; 465 th = (struct tcphdr *)((caddr_t)ip + off0); 466 } 467 } 468 optlen = off - sizeof (struct tcphdr); 469 optp = (u_char *)(th + 1); 470 } 471 thflags = th->th_flags; 472 473#ifdef TCP_DROP_SYNFIN 474 /* 475 * If the drop_synfin option is enabled, drop all packets with 476 * both the SYN and FIN bits set. This prevents e.g. nmap from 477 * identifying the TCP/IP stack. 478 * 479 * This is a violation of the TCP specification. 480 */ 481 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 482 goto drop; 483#endif 484 485 /* 486 * Convert TCP protocol specific fields to host format. 487 */ 488 th->th_seq = ntohl(th->th_seq); 489 th->th_ack = ntohl(th->th_ack); 490 th->th_win = ntohs(th->th_win); 491 th->th_urp = ntohs(th->th_urp); 492 493 /* 494 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 495 * until after ip6_savecontrol() is called and before other functions 496 * which don't want those proto headers. 497 * Because ip6_savecontrol() is going to parse the mbuf to 498 * search for data to be passed up to user-land, it wants mbuf 499 * parameters to be unchanged. 500 * XXX: the call of ip6_savecontrol() has been obsoleted based on 501 * latest version of the advanced API (20020110). 502 */ 503 drop_hdrlen = off0 + off; 504 505 /* 506 * Locate pcb for segment. 507 */ 508findpcb: 509#ifdef IPFIREWALL_FORWARD 510 if (ip_fw_fwd_addr != NULL 511#ifdef INET6 512 && isipv6 == NULL /* IPv6 support is not yet */ 513#endif /* INET6 */ 514 ) { 515 /* 516 * Diverted. Pretend to be the destination. 517 * already got one like this? 518 */ 519 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 520 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 521 if (!inp) { 522 /* 523 * No, then it's new. Try find the ambushing socket 524 */ 525 if (!ip_fw_fwd_addr->sin_port) { 526 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 527 th->th_sport, ip_fw_fwd_addr->sin_addr, 528 th->th_dport, 1, m->m_pkthdr.rcvif); 529 } else { 530 inp = in_pcblookup_hash(&tcbinfo, 531 ip->ip_src, th->th_sport, 532 ip_fw_fwd_addr->sin_addr, 533 ntohs(ip_fw_fwd_addr->sin_port), 1, 534 m->m_pkthdr.rcvif); 535 } 536 } 537 ip_fw_fwd_addr = NULL; 538 } else 539#endif /* IPFIREWALL_FORWARD */ 540 { 541#ifdef INET6 542 if (isipv6) 543 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 544 &ip6->ip6_dst, th->th_dport, 1, 545 m->m_pkthdr.rcvif); 546 else 547#endif /* INET6 */ 548 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 549 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 550 } 551 552#ifdef IPSEC 553#ifdef INET6 554 if (isipv6) { 555 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 556 ipsec6stat.in_polvio++; 557 goto drop; 558 } 559 } else 560#endif /* INET6 */ 561 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 562 ipsecstat.in_polvio++; 563 goto drop; 564 } 565#endif /*IPSEC*/ 566 567 /* 568 * If the state is CLOSED (i.e., TCB does not exist) then 569 * all data in the incoming segment is discarded. 570 * If the TCB exists but is in CLOSED state, it is embryonic, 571 * but should either do a listen or a connect soon. 572 */ 573 if (inp == NULL) { 574 if (log_in_vain) { 575#ifdef INET6 576 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 577#else /* INET6 */ 578 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 579#endif /* INET6 */ 580 581#ifdef INET6 582 if (isipv6) { 583 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 584 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 585 } else 586#endif 587 { 588 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 589 strcpy(sbuf, inet_ntoa(ip->ip_src)); 590 } 591 switch (log_in_vain) { 592 case 1: 593 if(thflags & TH_SYN) 594 log(LOG_INFO, 595 "Connection attempt to TCP %s:%d from %s:%d\n", 596 dbuf, ntohs(th->th_dport), 597 sbuf, 598 ntohs(th->th_sport)); 599 break; 600 case 2: 601 log(LOG_INFO, 602 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 603 dbuf, ntohs(th->th_dport), sbuf, 604 ntohs(th->th_sport), thflags); 605 break; 606 default: 607 break; 608 } 609 } 610 if (blackhole) { 611 switch (blackhole) { 612 case 1: 613 if (thflags & TH_SYN) 614 goto drop; 615 break; 616 case 2: 617 goto drop; 618 default: 619 goto drop; 620 } 621 } 622 rstreason = BANDLIM_RST_CLOSEDPORT; 623 goto dropwithreset; 624 } 625 tp = intotcpcb(inp); 626 if (tp == 0) { 627 rstreason = BANDLIM_RST_CLOSEDPORT; 628 goto dropwithreset; 629 } 630 if (tp->t_state == TCPS_CLOSED) 631 goto drop; 632 633 /* Unscale the window into a 32-bit value. */ 634 if ((thflags & TH_SYN) == 0) 635 tiwin = th->th_win << tp->snd_scale; 636 else 637 tiwin = th->th_win; 638 639 so = inp->inp_socket; 640 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 641 struct in_conninfo inc; 642#ifdef TCPDEBUG 643 if (so->so_options & SO_DEBUG) { 644 ostate = tp->t_state; 645#ifdef INET6 646 if (isipv6) 647 bcopy((char *)ip6, (char *)tcp_saveipgen, 648 sizeof(*ip6)); 649 else 650#endif /* INET6 */ 651 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 652 tcp_savetcp = *th; 653 } 654#endif 655 /* skip if this isn't a listen socket */ 656 if ((so->so_options & SO_ACCEPTCONN) == 0) 657 goto after_listen; 658#ifdef INET6 659 inc.inc_isipv6 = isipv6; 660 if (isipv6) { 661 inc.inc6_faddr = ip6->ip6_src; 662 inc.inc6_laddr = ip6->ip6_dst; 663 inc.inc6_route.ro_rt = NULL; /* XXX */ 664 665 } else 666#endif /* INET6 */ 667 { 668 inc.inc_faddr = ip->ip_src; 669 inc.inc_laddr = ip->ip_dst; 670 inc.inc_route.ro_rt = NULL; /* XXX */ 671 } 672 inc.inc_fport = th->th_sport; 673 inc.inc_lport = th->th_dport; 674 675 /* 676 * If the state is LISTEN then ignore segment if it contains 677 * a RST. If the segment contains an ACK then it is bad and 678 * send a RST. If it does not contain a SYN then it is not 679 * interesting; drop it. 680 * 681 * If the state is SYN_RECEIVED (syncache) and seg contains 682 * an ACK, but not for our SYN/ACK, send a RST. If the seg 683 * contains a RST, check the sequence number to see if it 684 * is a valid reset segment. 685 */ 686 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 687 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 688 if (!syncache_expand(&inc, th, &so, m)) { 689 /* 690 * No syncache entry, or ACK was not 691 * for our SYN/ACK. Send a RST. 692 */ 693 tcpstat.tcps_badsyn++; 694 rstreason = BANDLIM_RST_OPENPORT; 695 goto dropwithreset; 696 } 697 if (so == NULL) 698 /* 699 * Could not complete 3-way handshake, 700 * connection is being closed down, and 701 * syncache will free mbuf. 702 */ 703 return; 704 /* 705 * Socket is created in state SYN_RECEIVED. 706 * Continue processing segment. 707 */ 708 inp = sotoinpcb(so); 709 tp = intotcpcb(inp); 710 /* 711 * This is what would have happened in 712 * tcp_output() when the SYN,ACK was sent. 713 */ 714 tp->snd_up = tp->snd_una; 715 tp->snd_max = tp->snd_nxt = tp->iss + 1; 716 tp->last_ack_sent = tp->rcv_nxt; 717/* 718 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 719 * until the _second_ ACK is received: 720 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 721 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 722 * move to ESTAB, set snd_wnd to tiwin. 723 */ 724 tp->snd_wnd = tiwin; /* unscaled */ 725 goto after_listen; 726 } 727 if (thflags & TH_RST) { 728 syncache_chkrst(&inc, th); 729 goto drop; 730 } 731 if (thflags & TH_ACK) { 732 syncache_badack(&inc); 733 tcpstat.tcps_badsyn++; 734 rstreason = BANDLIM_RST_OPENPORT; 735 goto dropwithreset; 736 } 737 goto drop; 738 } 739 740 /* 741 * Segment's flags are (SYN) or (SYN|FIN). 742 */ 743#ifdef INET6 744 /* 745 * If deprecated address is forbidden, 746 * we do not accept SYN to deprecated interface 747 * address to prevent any new inbound connection from 748 * getting established. 749 * When we do not accept SYN, we send a TCP RST, 750 * with deprecated source address (instead of dropping 751 * it). We compromise it as it is much better for peer 752 * to send a RST, and RST will be the final packet 753 * for the exchange. 754 * 755 * If we do not forbid deprecated addresses, we accept 756 * the SYN packet. RFC2462 does not suggest dropping 757 * SYN in this case. 758 * If we decipher RFC2462 5.5.4, it says like this: 759 * 1. use of deprecated addr with existing 760 * communication is okay - "SHOULD continue to be 761 * used" 762 * 2. use of it with new communication: 763 * (2a) "SHOULD NOT be used if alternate address 764 * with sufficient scope is available" 765 * (2b) nothing mentioned otherwise. 766 * Here we fall into (2b) case as we have no choice in 767 * our source address selection - we must obey the peer. 768 * 769 * The wording in RFC2462 is confusing, and there are 770 * multiple description text for deprecated address 771 * handling - worse, they are not exactly the same. 772 * I believe 5.5.4 is the best one, so we follow 5.5.4. 773 */ 774 if (isipv6 && !ip6_use_deprecated) { 775 struct in6_ifaddr *ia6; 776 777 if ((ia6 = ip6_getdstifaddr(m)) && 778 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 779 tp = NULL; 780 rstreason = BANDLIM_RST_OPENPORT; 781 goto dropwithreset; 782 } 783 } 784#endif 785 /* 786 * If it is from this socket, drop it, it must be forged. 787 * Don't bother responding if the destination was a broadcast. 788 */ 789 if (th->th_dport == th->th_sport) { 790#ifdef INET6 791 if (isipv6) { 792 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 793 &ip6->ip6_src)) 794 goto drop; 795 } else 796#endif /* INET6 */ 797 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 798 goto drop; 799 } 800 /* 801 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 802 * 803 * Note that it is quite possible to receive unicast 804 * link-layer packets with a broadcast IP address. Use 805 * in_broadcast() to find them. 806 */ 807 if (m->m_flags & (M_BCAST|M_MCAST)) 808 goto drop; 809#ifdef INET6 810 if (isipv6) { 811 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 812 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 813 goto drop; 814 } else 815#endif 816 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 817 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 818 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 819 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 820 goto drop; 821 /* 822 * SYN appears to be valid; create compressed TCP state 823 * for syncache, or perform t/tcp connection. 824 */ 825 if (so->so_qlen <= so->so_qlimit) { 826 tcp_dooptions(&to, optp, optlen, 1); 827 if (!syncache_add(&inc, &to, th, &so, m)) 828 goto drop; 829 if (so == NULL) 830 /* 831 * Entry added to syncache, mbuf used to 832 * send SYN,ACK packet. 833 */ 834 return; 835 /* 836 * Segment passed TAO tests. 837 */ 838 inp = sotoinpcb(so); 839 tp = intotcpcb(inp); 840 tp->snd_wnd = tiwin; 841 tp->t_starttime = ticks; 842 tp->t_state = TCPS_ESTABLISHED; 843 844 /* 845 * If there is a FIN, or if there is data and the 846 * connection is local, then delay SYN,ACK(SYN) in 847 * the hope of piggy-backing it on a response 848 * segment. Otherwise must send ACK now in case 849 * the other side is slow starting. 850 */ 851 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 852 (tlen != 0 && 853#ifdef INET6 854 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 855 || 856 (!isipv6 && 857#endif 858 in_localaddr(inp->inp_faddr) 859#ifdef INET6 860 )) 861#endif 862 ))) { 863 callout_reset(tp->tt_delack, tcp_delacktime, 864 tcp_timer_delack, tp); 865 tp->t_flags |= TF_NEEDSYN; 866 } else 867 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 868 869 tcpstat.tcps_connects++; 870 soisconnected(so); 871 goto trimthenstep6; 872 } 873 goto drop; 874 } 875after_listen: 876 877/* XXX temp debugging */ 878 /* should not happen - syncache should pick up these connections */ 879 if (tp->t_state == TCPS_LISTEN) 880 panic("tcp_input: TCPS_LISTEN"); 881 882 /* 883 * Segment received on connection. 884 * Reset idle time and keep-alive timer. 885 */ 886 tp->t_rcvtime = ticks; 887 if (TCPS_HAVEESTABLISHED(tp->t_state)) 888 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 889 890 /* 891 * Process options. 892 * XXX this is tradtitional behavior, may need to be cleaned up. 893 */ 894 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 895 if (thflags & TH_SYN) { 896 if (to.to_flags & TOF_SCALE) { 897 tp->t_flags |= TF_RCVD_SCALE; 898 tp->requested_s_scale = to.to_requested_s_scale; 899 } 900 if (to.to_flags & TOF_TS) { 901 tp->t_flags |= TF_RCVD_TSTMP; 902 tp->ts_recent = to.to_tsval; 903 tp->ts_recent_age = ticks; 904 } 905 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 906 tp->t_flags |= TF_RCVD_CC; 907 if (to.to_flags & TOF_MSS) 908 tcp_mss(tp, to.to_mss); 909 } 910 911 /* 912 * Header prediction: check for the two common cases 913 * of a uni-directional data xfer. If the packet has 914 * no control flags, is in-sequence, the window didn't 915 * change and we're not retransmitting, it's a 916 * candidate. If the length is zero and the ack moved 917 * forward, we're the sender side of the xfer. Just 918 * free the data acked & wake any higher level process 919 * that was blocked waiting for space. If the length 920 * is non-zero and the ack didn't move, we're the 921 * receiver side. If we're getting packets in-order 922 * (the reassembly queue is empty), add the data to 923 * the socket buffer and note that we need a delayed ack. 924 * Make sure that the hidden state-flags are also off. 925 * Since we check for TCPS_ESTABLISHED above, it can only 926 * be TH_NEEDSYN. 927 */ 928 if (tp->t_state == TCPS_ESTABLISHED && 929 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 930 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 931 ((to.to_flags & TOF_TS) == 0 || 932 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 933 /* 934 * Using the CC option is compulsory if once started: 935 * the segment is OK if no T/TCP was negotiated or 936 * if the segment has a CC option equal to CCrecv 937 */ 938 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 939 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 940 th->th_seq == tp->rcv_nxt && 941 tiwin && tiwin == tp->snd_wnd && 942 tp->snd_nxt == tp->snd_max) { 943 944 /* 945 * If last ACK falls within this segment's sequence numbers, 946 * record the timestamp. 947 * NOTE that the test is modified according to the latest 948 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 949 */ 950 if ((to.to_flags & TOF_TS) != 0 && 951 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 952 tp->ts_recent_age = ticks; 953 tp->ts_recent = to.to_tsval; 954 } 955 956 if (tlen == 0) { 957 if (SEQ_GT(th->th_ack, tp->snd_una) && 958 SEQ_LEQ(th->th_ack, tp->snd_max) && 959 tp->snd_cwnd >= tp->snd_wnd && 960 tp->t_dupacks < tcprexmtthresh) { 961 /* 962 * this is a pure ack for outstanding data. 963 */ 964 ++tcpstat.tcps_predack; 965 /* 966 * "bad retransmit" recovery 967 */ 968 if (tp->t_rxtshift == 1 && 969 ticks < tp->t_badrxtwin) { 970 tp->snd_cwnd = tp->snd_cwnd_prev; 971 tp->snd_ssthresh = 972 tp->snd_ssthresh_prev; 973 tp->snd_nxt = tp->snd_max; 974 tp->t_badrxtwin = 0; 975 } 976 if ((to.to_flags & TOF_TS) != 0) 977 tcp_xmit_timer(tp, 978 ticks - to.to_tsecr + 1); 979 else if (tp->t_rtttime && 980 SEQ_GT(th->th_ack, tp->t_rtseq)) 981 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 982 acked = th->th_ack - tp->snd_una; 983 tcpstat.tcps_rcvackpack++; 984 tcpstat.tcps_rcvackbyte += acked; 985 sbdrop(&so->so_snd, acked); 986 tp->snd_una = th->th_ack; 987 m_freem(m); 988 ND6_HINT(tp); /* some progress has been done */ 989 990 /* 991 * If all outstanding data are acked, stop 992 * retransmit timer, otherwise restart timer 993 * using current (possibly backed-off) value. 994 * If process is waiting for space, 995 * wakeup/selwakeup/signal. If data 996 * are ready to send, let tcp_output 997 * decide between more output or persist. 998 */ 999 if (tp->snd_una == tp->snd_max) 1000 callout_stop(tp->tt_rexmt); 1001 else if (!callout_active(tp->tt_persist)) 1002 callout_reset(tp->tt_rexmt, 1003 tp->t_rxtcur, 1004 tcp_timer_rexmt, tp); 1005 1006 sowwakeup(so); 1007 if (so->so_snd.sb_cc) 1008 (void) tcp_output(tp); 1009 return; 1010 } 1011 } else if (th->th_ack == tp->snd_una && 1012 LIST_EMPTY(&tp->t_segq) && 1013 tlen <= sbspace(&so->so_rcv)) { 1014 /* 1015 * this is a pure, in-sequence data packet 1016 * with nothing on the reassembly queue and 1017 * we have enough buffer space to take it. 1018 */ 1019 ++tcpstat.tcps_preddat; 1020 tp->rcv_nxt += tlen; 1021 tcpstat.tcps_rcvpack++; 1022 tcpstat.tcps_rcvbyte += tlen; 1023 ND6_HINT(tp); /* some progress has been done */ 1024 /* 1025 * Add data to socket buffer. 1026 */ 1027 m_adj(m, drop_hdrlen); /* delayed header drop */ 1028 sbappend(&so->so_rcv, m); 1029 sorwakeup(so); 1030 if (DELAY_ACK(tp)) { 1031 callout_reset(tp->tt_delack, tcp_delacktime, 1032 tcp_timer_delack, tp); 1033 } else { 1034 tp->t_flags |= TF_ACKNOW; 1035 tcp_output(tp); 1036 } 1037 return; 1038 } 1039 } 1040 1041 /* 1042 * Calculate amount of space in receive window, 1043 * and then do TCP input processing. 1044 * Receive window is amount of space in rcv queue, 1045 * but not less than advertised window. 1046 */ 1047 { int win; 1048 1049 win = sbspace(&so->so_rcv); 1050 if (win < 0) 1051 win = 0; 1052 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1053 } 1054 1055 switch (tp->t_state) { 1056 1057 /* 1058 * If the state is SYN_RECEIVED: 1059 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1060 */ 1061 case TCPS_SYN_RECEIVED: 1062 if ((thflags & TH_ACK) && 1063 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1064 SEQ_GT(th->th_ack, tp->snd_max))) { 1065 rstreason = BANDLIM_RST_OPENPORT; 1066 goto dropwithreset; 1067 } 1068 break; 1069 1070 /* 1071 * If the state is SYN_SENT: 1072 * if seg contains an ACK, but not for our SYN, drop the input. 1073 * if seg contains a RST, then drop the connection. 1074 * if seg does not contain SYN, then drop it. 1075 * Otherwise this is an acceptable SYN segment 1076 * initialize tp->rcv_nxt and tp->irs 1077 * if seg contains ack then advance tp->snd_una 1078 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1079 * arrange for segment to be acked (eventually) 1080 * continue processing rest of data/controls, beginning with URG 1081 */ 1082 case TCPS_SYN_SENT: 1083 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1084 taop = &tao_noncached; 1085 bzero(taop, sizeof(*taop)); 1086 } 1087 1088 if ((thflags & TH_ACK) && 1089 (SEQ_LEQ(th->th_ack, tp->iss) || 1090 SEQ_GT(th->th_ack, tp->snd_max))) { 1091 /* 1092 * If we have a cached CCsent for the remote host, 1093 * hence we haven't just crashed and restarted, 1094 * do not send a RST. This may be a retransmission 1095 * from the other side after our earlier ACK was lost. 1096 * Our new SYN, when it arrives, will serve as the 1097 * needed ACK. 1098 */ 1099 if (taop->tao_ccsent != 0) 1100 goto drop; 1101 else { 1102 rstreason = BANDLIM_UNLIMITED; 1103 goto dropwithreset; 1104 } 1105 } 1106 if (thflags & TH_RST) { 1107 if (thflags & TH_ACK) 1108 tp = tcp_drop(tp, ECONNREFUSED); 1109 goto drop; 1110 } 1111 if ((thflags & TH_SYN) == 0) 1112 goto drop; 1113 tp->snd_wnd = th->th_win; /* initial send window */ 1114 tp->cc_recv = to.to_cc; /* foreign CC */ 1115 1116 tp->irs = th->th_seq; 1117 tcp_rcvseqinit(tp); 1118 if (thflags & TH_ACK) { 1119 /* 1120 * Our SYN was acked. If segment contains CC.ECHO 1121 * option, check it to make sure this segment really 1122 * matches our SYN. If not, just drop it as old 1123 * duplicate, but send an RST if we're still playing 1124 * by the old rules. If no CC.ECHO option, make sure 1125 * we don't get fooled into using T/TCP. 1126 */ 1127 if (to.to_flags & TOF_CCECHO) { 1128 if (tp->cc_send != to.to_ccecho) { 1129 if (taop->tao_ccsent != 0) 1130 goto drop; 1131 else { 1132 rstreason = BANDLIM_UNLIMITED; 1133 goto dropwithreset; 1134 } 1135 } 1136 } else 1137 tp->t_flags &= ~TF_RCVD_CC; 1138 tcpstat.tcps_connects++; 1139 soisconnected(so); 1140 /* Do window scaling on this connection? */ 1141 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1142 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1143 tp->snd_scale = tp->requested_s_scale; 1144 tp->rcv_scale = tp->request_r_scale; 1145 } 1146 /* Segment is acceptable, update cache if undefined. */ 1147 if (taop->tao_ccsent == 0) 1148 taop->tao_ccsent = to.to_ccecho; 1149 1150 tp->rcv_adv += tp->rcv_wnd; 1151 tp->snd_una++; /* SYN is acked */ 1152 /* 1153 * If there's data, delay ACK; if there's also a FIN 1154 * ACKNOW will be turned on later. 1155 */ 1156 if (DELAY_ACK(tp) && tlen != 0) 1157 callout_reset(tp->tt_delack, tcp_delacktime, 1158 tcp_timer_delack, tp); 1159 else 1160 tp->t_flags |= TF_ACKNOW; 1161 /* 1162 * Received <SYN,ACK> in SYN_SENT[*] state. 1163 * Transitions: 1164 * SYN_SENT --> ESTABLISHED 1165 * SYN_SENT* --> FIN_WAIT_1 1166 */ 1167 tp->t_starttime = ticks; 1168 if (tp->t_flags & TF_NEEDFIN) { 1169 tp->t_state = TCPS_FIN_WAIT_1; 1170 tp->t_flags &= ~TF_NEEDFIN; 1171 thflags &= ~TH_SYN; 1172 } else { 1173 tp->t_state = TCPS_ESTABLISHED; 1174 callout_reset(tp->tt_keep, tcp_keepidle, 1175 tcp_timer_keep, tp); 1176 } 1177 } else { 1178 /* 1179 * Received initial SYN in SYN-SENT[*] state => simul- 1180 * taneous open. If segment contains CC option and there is 1181 * a cached CC, apply TAO test; if it succeeds, connection is 1182 * half-synchronized. Otherwise, do 3-way handshake: 1183 * SYN-SENT -> SYN-RECEIVED 1184 * SYN-SENT* -> SYN-RECEIVED* 1185 * If there was no CC option, clear cached CC value. 1186 */ 1187 tp->t_flags |= TF_ACKNOW; 1188 callout_stop(tp->tt_rexmt); 1189 if (to.to_flags & TOF_CC) { 1190 if (taop->tao_cc != 0 && 1191 CC_GT(to.to_cc, taop->tao_cc)) { 1192 /* 1193 * update cache and make transition: 1194 * SYN-SENT -> ESTABLISHED* 1195 * SYN-SENT* -> FIN-WAIT-1* 1196 */ 1197 taop->tao_cc = to.to_cc; 1198 tp->t_starttime = ticks; 1199 if (tp->t_flags & TF_NEEDFIN) { 1200 tp->t_state = TCPS_FIN_WAIT_1; 1201 tp->t_flags &= ~TF_NEEDFIN; 1202 } else { 1203 tp->t_state = TCPS_ESTABLISHED; 1204 callout_reset(tp->tt_keep, 1205 tcp_keepidle, 1206 tcp_timer_keep, 1207 tp); 1208 } 1209 tp->t_flags |= TF_NEEDSYN; 1210 } else 1211 tp->t_state = TCPS_SYN_RECEIVED; 1212 } else { 1213 /* CC.NEW or no option => invalidate cache */ 1214 taop->tao_cc = 0; 1215 tp->t_state = TCPS_SYN_RECEIVED; 1216 } 1217 } 1218 1219trimthenstep6: 1220 /* 1221 * Advance th->th_seq to correspond to first data byte. 1222 * If data, trim to stay within window, 1223 * dropping FIN if necessary. 1224 */ 1225 th->th_seq++; 1226 if (tlen > tp->rcv_wnd) { 1227 todrop = tlen - tp->rcv_wnd; 1228 m_adj(m, -todrop); 1229 tlen = tp->rcv_wnd; 1230 thflags &= ~TH_FIN; 1231 tcpstat.tcps_rcvpackafterwin++; 1232 tcpstat.tcps_rcvbyteafterwin += todrop; 1233 } 1234 tp->snd_wl1 = th->th_seq - 1; 1235 tp->rcv_up = th->th_seq; 1236 /* 1237 * Client side of transaction: already sent SYN and data. 1238 * If the remote host used T/TCP to validate the SYN, 1239 * our data will be ACK'd; if so, enter normal data segment 1240 * processing in the middle of step 5, ack processing. 1241 * Otherwise, goto step 6. 1242 */ 1243 if (thflags & TH_ACK) 1244 goto process_ACK; 1245 goto step6; 1246 /* 1247 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1248 * if segment contains a SYN and CC [not CC.NEW] option: 1249 * if state == TIME_WAIT and connection duration > MSL, 1250 * drop packet and send RST; 1251 * 1252 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1253 * ack the FIN (and data) in retransmission queue. 1254 * Complete close and delete TCPCB. Then reprocess 1255 * segment, hoping to find new TCPCB in LISTEN state; 1256 * 1257 * else must be old SYN; drop it. 1258 * else do normal processing. 1259 */ 1260 case TCPS_LAST_ACK: 1261 case TCPS_CLOSING: 1262 case TCPS_TIME_WAIT: 1263 if ((thflags & TH_SYN) && 1264 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1265 if (tp->t_state == TCPS_TIME_WAIT && 1266 (ticks - tp->t_starttime) > tcp_msl) { 1267 rstreason = BANDLIM_UNLIMITED; 1268 goto dropwithreset; 1269 } 1270 if (CC_GT(to.to_cc, tp->cc_recv)) { 1271 tp = tcp_close(tp); 1272 goto findpcb; 1273 } 1274 else 1275 goto drop; 1276 } 1277 break; /* continue normal processing */ 1278 } 1279 1280 /* 1281 * States other than LISTEN or SYN_SENT. 1282 * First check the RST flag and sequence number since reset segments 1283 * are exempt from the timestamp and connection count tests. This 1284 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1285 * below which allowed reset segments in half the sequence space 1286 * to fall though and be processed (which gives forged reset 1287 * segments with a random sequence number a 50 percent chance of 1288 * killing a connection). 1289 * Then check timestamp, if present. 1290 * Then check the connection count, if present. 1291 * Then check that at least some bytes of segment are within 1292 * receive window. If segment begins before rcv_nxt, 1293 * drop leading data (and SYN); if nothing left, just ack. 1294 * 1295 * 1296 * If the RST bit is set, check the sequence number to see 1297 * if this is a valid reset segment. 1298 * RFC 793 page 37: 1299 * In all states except SYN-SENT, all reset (RST) segments 1300 * are validated by checking their SEQ-fields. A reset is 1301 * valid if its sequence number is in the window. 1302 * Note: this does not take into account delayed ACKs, so 1303 * we should test against last_ack_sent instead of rcv_nxt. 1304 * The sequence number in the reset segment is normally an 1305 * echo of our outgoing acknowlegement numbers, but some hosts 1306 * send a reset with the sequence number at the rightmost edge 1307 * of our receive window, and we have to handle this case. 1308 * If we have multiple segments in flight, the intial reset 1309 * segment sequence numbers will be to the left of last_ack_sent, 1310 * but they will eventually catch up. 1311 * In any case, it never made sense to trim reset segments to 1312 * fit the receive window since RFC 1122 says: 1313 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1314 * 1315 * A TCP SHOULD allow a received RST segment to include data. 1316 * 1317 * DISCUSSION 1318 * It has been suggested that a RST segment could contain 1319 * ASCII text that encoded and explained the cause of the 1320 * RST. No standard has yet been established for such 1321 * data. 1322 * 1323 * If the reset segment passes the sequence number test examine 1324 * the state: 1325 * SYN_RECEIVED STATE: 1326 * If passive open, return to LISTEN state. 1327 * If active open, inform user that connection was refused. 1328 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1329 * Inform user that connection was reset, and close tcb. 1330 * CLOSING, LAST_ACK STATES: 1331 * Close the tcb. 1332 * TIME_WAIT STATE: 1333 * Drop the segment - see Stevens, vol. 2, p. 964 and 1334 * RFC 1337. 1335 */ 1336 if (thflags & TH_RST) { 1337 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1338 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1339 switch (tp->t_state) { 1340 1341 case TCPS_SYN_RECEIVED: 1342 so->so_error = ECONNREFUSED; 1343 goto close; 1344 1345 case TCPS_ESTABLISHED: 1346 case TCPS_FIN_WAIT_1: 1347 case TCPS_FIN_WAIT_2: 1348 case TCPS_CLOSE_WAIT: 1349 so->so_error = ECONNRESET; 1350 close: 1351 tp->t_state = TCPS_CLOSED; 1352 tcpstat.tcps_drops++; 1353 tp = tcp_close(tp); 1354 break; 1355 1356 case TCPS_CLOSING: 1357 case TCPS_LAST_ACK: 1358 tp = tcp_close(tp); 1359 break; 1360 1361 case TCPS_TIME_WAIT: 1362 break; 1363 } 1364 } 1365 goto drop; 1366 } 1367 1368 /* 1369 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1370 * and it's less than ts_recent, drop it. 1371 */ 1372 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1373 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1374 1375 /* Check to see if ts_recent is over 24 days old. */ 1376 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1377 /* 1378 * Invalidate ts_recent. If this segment updates 1379 * ts_recent, the age will be reset later and ts_recent 1380 * will get a valid value. If it does not, setting 1381 * ts_recent to zero will at least satisfy the 1382 * requirement that zero be placed in the timestamp 1383 * echo reply when ts_recent isn't valid. The 1384 * age isn't reset until we get a valid ts_recent 1385 * because we don't want out-of-order segments to be 1386 * dropped when ts_recent is old. 1387 */ 1388 tp->ts_recent = 0; 1389 } else { 1390 tcpstat.tcps_rcvduppack++; 1391 tcpstat.tcps_rcvdupbyte += tlen; 1392 tcpstat.tcps_pawsdrop++; 1393 goto dropafterack; 1394 } 1395 } 1396 1397 /* 1398 * T/TCP mechanism 1399 * If T/TCP was negotiated and the segment doesn't have CC, 1400 * or if its CC is wrong then drop the segment. 1401 * RST segments do not have to comply with this. 1402 */ 1403 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1404 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1405 goto dropafterack; 1406 1407 /* 1408 * In the SYN-RECEIVED state, validate that the packet belongs to 1409 * this connection before trimming the data to fit the receive 1410 * window. Check the sequence number versus IRS since we know 1411 * the sequence numbers haven't wrapped. This is a partial fix 1412 * for the "LAND" DoS attack. 1413 */ 1414 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1415 rstreason = BANDLIM_RST_OPENPORT; 1416 goto dropwithreset; 1417 } 1418 1419 todrop = tp->rcv_nxt - th->th_seq; 1420 if (todrop > 0) { 1421 if (thflags & TH_SYN) { 1422 thflags &= ~TH_SYN; 1423 th->th_seq++; 1424 if (th->th_urp > 1) 1425 th->th_urp--; 1426 else 1427 thflags &= ~TH_URG; 1428 todrop--; 1429 } 1430 /* 1431 * Following if statement from Stevens, vol. 2, p. 960. 1432 */ 1433 if (todrop > tlen 1434 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1435 /* 1436 * Any valid FIN must be to the left of the window. 1437 * At this point the FIN must be a duplicate or out 1438 * of sequence; drop it. 1439 */ 1440 thflags &= ~TH_FIN; 1441 1442 /* 1443 * Send an ACK to resynchronize and drop any data. 1444 * But keep on processing for RST or ACK. 1445 */ 1446 tp->t_flags |= TF_ACKNOW; 1447 todrop = tlen; 1448 tcpstat.tcps_rcvduppack++; 1449 tcpstat.tcps_rcvdupbyte += todrop; 1450 } else { 1451 tcpstat.tcps_rcvpartduppack++; 1452 tcpstat.tcps_rcvpartdupbyte += todrop; 1453 } 1454 drop_hdrlen += todrop; /* drop from the top afterwards */ 1455 th->th_seq += todrop; 1456 tlen -= todrop; 1457 if (th->th_urp > todrop) 1458 th->th_urp -= todrop; 1459 else { 1460 thflags &= ~TH_URG; 1461 th->th_urp = 0; 1462 } 1463 } 1464 1465 /* 1466 * If new data are received on a connection after the 1467 * user processes are gone, then RST the other end. 1468 */ 1469 if ((so->so_state & SS_NOFDREF) && 1470 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1471 tp = tcp_close(tp); 1472 tcpstat.tcps_rcvafterclose++; 1473 rstreason = BANDLIM_UNLIMITED; 1474 goto dropwithreset; 1475 } 1476 1477 /* 1478 * If segment ends after window, drop trailing data 1479 * (and PUSH and FIN); if nothing left, just ACK. 1480 */ 1481 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1482 if (todrop > 0) { 1483 tcpstat.tcps_rcvpackafterwin++; 1484 if (todrop >= tlen) { 1485 tcpstat.tcps_rcvbyteafterwin += tlen; 1486 /* 1487 * If a new connection request is received 1488 * while in TIME_WAIT, drop the old connection 1489 * and start over if the sequence numbers 1490 * are above the previous ones. 1491 */ 1492 if (thflags & TH_SYN && 1493 tp->t_state == TCPS_TIME_WAIT && 1494 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1495 tp = tcp_close(tp); 1496 goto findpcb; 1497 } 1498 /* 1499 * If window is closed can only take segments at 1500 * window edge, and have to drop data and PUSH from 1501 * incoming segments. Continue processing, but 1502 * remember to ack. Otherwise, drop segment 1503 * and ack. 1504 */ 1505 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1506 tp->t_flags |= TF_ACKNOW; 1507 tcpstat.tcps_rcvwinprobe++; 1508 } else 1509 goto dropafterack; 1510 } else 1511 tcpstat.tcps_rcvbyteafterwin += todrop; 1512 m_adj(m, -todrop); 1513 tlen -= todrop; 1514 thflags &= ~(TH_PUSH|TH_FIN); 1515 } 1516 1517 /* 1518 * If last ACK falls within this segment's sequence numbers, 1519 * record its timestamp. 1520 * NOTE that the test is modified according to the latest 1521 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1522 */ 1523 if ((to.to_flags & TOF_TS) != 0 && 1524 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1525 tp->ts_recent_age = ticks; 1526 tp->ts_recent = to.to_tsval; 1527 } 1528 1529 /* 1530 * If a SYN is in the window, then this is an 1531 * error and we send an RST and drop the connection. 1532 */ 1533 if (thflags & TH_SYN) { 1534 tp = tcp_drop(tp, ECONNRESET); 1535 rstreason = BANDLIM_UNLIMITED; 1536 goto dropwithreset; 1537 } 1538 1539 /* 1540 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1541 * flag is on (half-synchronized state), then queue data for 1542 * later processing; else drop segment and return. 1543 */ 1544 if ((thflags & TH_ACK) == 0) { 1545 if (tp->t_state == TCPS_SYN_RECEIVED || 1546 (tp->t_flags & TF_NEEDSYN)) 1547 goto step6; 1548 else 1549 goto drop; 1550 } 1551 1552 /* 1553 * Ack processing. 1554 */ 1555 switch (tp->t_state) { 1556 1557 /* 1558 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1559 * ESTABLISHED state and continue processing. 1560 * The ACK was checked above. 1561 */ 1562 case TCPS_SYN_RECEIVED: 1563 1564 tcpstat.tcps_connects++; 1565 soisconnected(so); 1566 /* Do window scaling? */ 1567 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1568 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1569 tp->snd_scale = tp->requested_s_scale; 1570 tp->rcv_scale = tp->request_r_scale; 1571 } 1572 /* 1573 * Upon successful completion of 3-way handshake, 1574 * update cache.CC if it was undefined, pass any queued 1575 * data to the user, and advance state appropriately. 1576 */ 1577 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1578 taop->tao_cc == 0) 1579 taop->tao_cc = tp->cc_recv; 1580 1581 /* 1582 * Make transitions: 1583 * SYN-RECEIVED -> ESTABLISHED 1584 * SYN-RECEIVED* -> FIN-WAIT-1 1585 */ 1586 tp->t_starttime = ticks; 1587 if (tp->t_flags & TF_NEEDFIN) { 1588 tp->t_state = TCPS_FIN_WAIT_1; 1589 tp->t_flags &= ~TF_NEEDFIN; 1590 } else { 1591 tp->t_state = TCPS_ESTABLISHED; 1592 callout_reset(tp->tt_keep, tcp_keepidle, 1593 tcp_timer_keep, tp); 1594 } 1595 /* 1596 * If segment contains data or ACK, will call tcp_reass() 1597 * later; if not, do so now to pass queued data to user. 1598 */ 1599 if (tlen == 0 && (thflags & TH_FIN) == 0) 1600 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1601 (struct mbuf *)0); 1602 tp->snd_wl1 = th->th_seq - 1; 1603 /* fall into ... */ 1604 1605 /* 1606 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1607 * ACKs. If the ack is in the range 1608 * tp->snd_una < th->th_ack <= tp->snd_max 1609 * then advance tp->snd_una to th->th_ack and drop 1610 * data from the retransmission queue. If this ACK reflects 1611 * more up to date window information we update our window information. 1612 */ 1613 case TCPS_ESTABLISHED: 1614 case TCPS_FIN_WAIT_1: 1615 case TCPS_FIN_WAIT_2: 1616 case TCPS_CLOSE_WAIT: 1617 case TCPS_CLOSING: 1618 case TCPS_LAST_ACK: 1619 case TCPS_TIME_WAIT: 1620 1621 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1622 if (tlen == 0 && tiwin == tp->snd_wnd) { 1623 tcpstat.tcps_rcvdupack++; 1624 /* 1625 * If we have outstanding data (other than 1626 * a window probe), this is a completely 1627 * duplicate ack (ie, window info didn't 1628 * change), the ack is the biggest we've 1629 * seen and we've seen exactly our rexmt 1630 * threshhold of them, assume a packet 1631 * has been dropped and retransmit it. 1632 * Kludge snd_nxt & the congestion 1633 * window so we send only this one 1634 * packet. 1635 * 1636 * We know we're losing at the current 1637 * window size so do congestion avoidance 1638 * (set ssthresh to half the current window 1639 * and pull our congestion window back to 1640 * the new ssthresh). 1641 * 1642 * Dup acks mean that packets have left the 1643 * network (they're now cached at the receiver) 1644 * so bump cwnd by the amount in the receiver 1645 * to keep a constant cwnd packets in the 1646 * network. 1647 */ 1648 if (!callout_active(tp->tt_rexmt) || 1649 th->th_ack != tp->snd_una) 1650 tp->t_dupacks = 0; 1651 else if (++tp->t_dupacks == tcprexmtthresh) { 1652 tcp_seq onxt = tp->snd_nxt; 1653 u_int win = 1654 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1655 tp->t_maxseg; 1656 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1657 tp->snd_recover)) { 1658 /* False retransmit, should not 1659 * cut window 1660 */ 1661 tp->snd_cwnd += tp->t_maxseg; 1662 tp->t_dupacks = 0; 1663 (void) tcp_output(tp); 1664 goto drop; 1665 } 1666 if (win < 2) 1667 win = 2; 1668 tp->snd_ssthresh = win * tp->t_maxseg; 1669 tp->snd_recover = tp->snd_max; 1670 callout_stop(tp->tt_rexmt); 1671 tp->t_rtttime = 0; 1672 tp->snd_nxt = th->th_ack; 1673 tp->snd_cwnd = tp->t_maxseg; 1674 (void) tcp_output(tp); 1675 tp->snd_cwnd = tp->snd_ssthresh + 1676 tp->t_maxseg * tp->t_dupacks; 1677 if (SEQ_GT(onxt, tp->snd_nxt)) 1678 tp->snd_nxt = onxt; 1679 goto drop; 1680 } else if (tp->t_dupacks > tcprexmtthresh) { 1681 tp->snd_cwnd += tp->t_maxseg; 1682 (void) tcp_output(tp); 1683 goto drop; 1684 } 1685 } else 1686 tp->t_dupacks = 0; 1687 break; 1688 } 1689 /* 1690 * If the congestion window was inflated to account 1691 * for the other side's cached packets, retract it. 1692 */ 1693 if (tcp_do_newreno == 0) { 1694 if (tp->t_dupacks >= tcprexmtthresh && 1695 tp->snd_cwnd > tp->snd_ssthresh) 1696 tp->snd_cwnd = tp->snd_ssthresh; 1697 tp->t_dupacks = 0; 1698 } else if (tp->t_dupacks >= tcprexmtthresh && 1699 !tcp_newreno(tp, th)) { 1700 /* 1701 * Window inflation should have left us with approx. 1702 * snd_ssthresh outstanding data. But in case we 1703 * would be inclined to send a burst, better to do 1704 * it via the slow start mechanism. 1705 */ 1706 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1707 tp->snd_cwnd = 1708 tp->snd_max - th->th_ack + tp->t_maxseg; 1709 else 1710 tp->snd_cwnd = tp->snd_ssthresh; 1711 tp->t_dupacks = 0; 1712 } 1713 if (tp->t_dupacks < tcprexmtthresh) 1714 tp->t_dupacks = 0; 1715 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1716 tcpstat.tcps_rcvacktoomuch++; 1717 goto dropafterack; 1718 } 1719 /* 1720 * If we reach this point, ACK is not a duplicate, 1721 * i.e., it ACKs something we sent. 1722 */ 1723 if (tp->t_flags & TF_NEEDSYN) { 1724 /* 1725 * T/TCP: Connection was half-synchronized, and our 1726 * SYN has been ACK'd (so connection is now fully 1727 * synchronized). Go to non-starred state, 1728 * increment snd_una for ACK of SYN, and check if 1729 * we can do window scaling. 1730 */ 1731 tp->t_flags &= ~TF_NEEDSYN; 1732 tp->snd_una++; 1733 /* Do window scaling? */ 1734 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1735 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1736 tp->snd_scale = tp->requested_s_scale; 1737 tp->rcv_scale = tp->request_r_scale; 1738 } 1739 } 1740 1741process_ACK: 1742 acked = th->th_ack - tp->snd_una; 1743 tcpstat.tcps_rcvackpack++; 1744 tcpstat.tcps_rcvackbyte += acked; 1745 1746 /* 1747 * If we just performed our first retransmit, and the ACK 1748 * arrives within our recovery window, then it was a mistake 1749 * to do the retransmit in the first place. Recover our 1750 * original cwnd and ssthresh, and proceed to transmit where 1751 * we left off. 1752 */ 1753 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1754 tp->snd_cwnd = tp->snd_cwnd_prev; 1755 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1756 tp->snd_nxt = tp->snd_max; 1757 tp->t_badrxtwin = 0; /* XXX probably not required */ 1758 } 1759 1760 /* 1761 * If we have a timestamp reply, update smoothed 1762 * round trip time. If no timestamp is present but 1763 * transmit timer is running and timed sequence 1764 * number was acked, update smoothed round trip time. 1765 * Since we now have an rtt measurement, cancel the 1766 * timer backoff (cf., Phil Karn's retransmit alg.). 1767 * Recompute the initial retransmit timer. 1768 */ 1769 if (to.to_flags & TOF_TS) 1770 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1771 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1772 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1773 1774 /* 1775 * If all outstanding data is acked, stop retransmit 1776 * timer and remember to restart (more output or persist). 1777 * If there is more data to be acked, restart retransmit 1778 * timer, using current (possibly backed-off) value. 1779 */ 1780 if (th->th_ack == tp->snd_max) { 1781 callout_stop(tp->tt_rexmt); 1782 needoutput = 1; 1783 } else if (!callout_active(tp->tt_persist)) 1784 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1785 tcp_timer_rexmt, tp); 1786 1787 /* 1788 * If no data (only SYN) was ACK'd, 1789 * skip rest of ACK processing. 1790 */ 1791 if (acked == 0) 1792 goto step6; 1793 1794 /* 1795 * When new data is acked, open the congestion window. 1796 * If the window gives us less than ssthresh packets 1797 * in flight, open exponentially (maxseg per packet). 1798 * Otherwise open linearly: maxseg per window 1799 * (maxseg^2 / cwnd per packet). 1800 */ 1801 { 1802 register u_int cw = tp->snd_cwnd; 1803 register u_int incr = tp->t_maxseg; 1804 1805 if (cw > tp->snd_ssthresh) 1806 incr = incr * incr / cw; 1807 /* 1808 * If t_dupacks != 0 here, it indicates that we are still 1809 * in NewReno fast recovery mode, so we leave the congestion 1810 * window alone. 1811 */ 1812 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1813 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1814 } 1815 if (acked > so->so_snd.sb_cc) { 1816 tp->snd_wnd -= so->so_snd.sb_cc; 1817 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1818 ourfinisacked = 1; 1819 } else { 1820 sbdrop(&so->so_snd, acked); 1821 tp->snd_wnd -= acked; 1822 ourfinisacked = 0; 1823 } 1824 sowwakeup(so); 1825 tp->snd_una = th->th_ack; 1826 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1827 tp->snd_nxt = tp->snd_una; 1828 1829 switch (tp->t_state) { 1830 1831 /* 1832 * In FIN_WAIT_1 STATE in addition to the processing 1833 * for the ESTABLISHED state if our FIN is now acknowledged 1834 * then enter FIN_WAIT_2. 1835 */ 1836 case TCPS_FIN_WAIT_1: 1837 if (ourfinisacked) { 1838 /* 1839 * If we can't receive any more 1840 * data, then closing user can proceed. 1841 * Starting the timer is contrary to the 1842 * specification, but if we don't get a FIN 1843 * we'll hang forever. 1844 */
| 35 */ 36 37#include "opt_ipfw.h" /* for ipfw_fwd */ 38#include "opt_inet6.h" 39#include "opt_ipsec.h" 40#include "opt_tcpdebug.h" 41#include "opt_tcp_input.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/proc.h> /* for proc0 declaration */ 50#include <sys/protosw.h> 51#include <sys/socket.h> 52#include <sys/socketvar.h> 53#include <sys/syslog.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/ip.h> 63#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 64#include <netinet/in_var.h> 65#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 66#include <netinet/in_pcb.h> 67#include <netinet/ip_var.h> 68#ifdef INET6 69#include <netinet/ip6.h> 70#include <netinet/icmp6.h> 71#include <netinet6/nd6.h> 72#include <netinet6/ip6_var.h> 73#include <netinet6/in6_pcb.h> 74#endif 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#ifdef INET6 81#include <netinet6/tcp6_var.h> 82#endif 83#include <netinet/tcpip.h> 84#ifdef TCPDEBUG 85#include <netinet/tcp_debug.h> 86 87u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 88struct tcphdr tcp_savetcp; 89#endif /* TCPDEBUG */ 90 91#ifdef IPSEC 92#include <netinet6/ipsec.h> 93#ifdef INET6 94#include <netinet6/ipsec6.h> 95#endif 96#include <netkey/key.h> 97#endif /*IPSEC*/ 98 99#include <machine/in_cksum.h> 100 101MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 102 103static int tcprexmtthresh = 3; 104tcp_cc tcp_ccgen; 105 106struct tcpstat tcpstat; 107SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 109 110static int log_in_vain = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 112 &log_in_vain, 0, "Log all incoming TCP connections"); 113 114static int blackhole = 0; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 116 &blackhole, 0, "Do not send RST when dropping refused connections"); 117 118int tcp_delack_enabled = 1; 119SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 120 &tcp_delack_enabled, 0, 121 "Delay ACK to try and piggyback it onto a data packet"); 122 123#ifdef TCP_DROP_SYNFIN 124static int drop_synfin = 0; 125SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 127#endif 128 129struct inpcbhead tcb; 130#define tcb6 tcb /* for KAME src sync over BSD*'s */ 131struct inpcbinfo tcbinfo; 132 133static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 134static void tcp_pulloutofband(struct socket *, 135 struct tcphdr *, struct mbuf *, int); 136static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 137 struct mbuf *); 138static void tcp_xmit_timer(struct tcpcb *, int); 139static int tcp_newreno(struct tcpcb *, struct tcphdr *); 140 141/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 142#ifdef INET6 143#define ND6_HINT(tp) \ 144do { \ 145 if ((tp) && (tp)->t_inpcb && \ 146 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 147 (tp)->t_inpcb->in6p_route.ro_rt) \ 148 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 149} while (0) 150#else 151#define ND6_HINT(tp) 152#endif 153 154/* 155 * Indicate whether this ack should be delayed. We can delay the ack if 156 * - delayed acks are enabled and 157 * - there is no delayed ack timer in progress and 158 * - our last ack wasn't a 0-sized window. We never want to delay 159 * the ack that opens up a 0-sized window. 160 */ 161#define DELAY_ACK(tp) \ 162 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 163 (tp->t_flags & TF_RXWIN0SENT) == 0) 164 165static int 166tcp_reass(tp, th, tlenp, m) 167 register struct tcpcb *tp; 168 register struct tcphdr *th; 169 int *tlenp; 170 struct mbuf *m; 171{ 172 struct tseg_qent *q; 173 struct tseg_qent *p = NULL; 174 struct tseg_qent *nq; 175 struct tseg_qent *te; 176 struct socket *so = tp->t_inpcb->inp_socket; 177 int flags; 178 179 /* 180 * Call with th==0 after become established to 181 * force pre-ESTABLISHED data up to user socket. 182 */ 183 if (th == 0) 184 goto present; 185 186 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 187 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 188 M_NOWAIT); 189 if (te == NULL) { 190 tcpstat.tcps_rcvmemdrop++; 191 m_freem(m); 192 return (0); 193 } 194 195 /* 196 * Find a segment which begins after this one does. 197 */ 198 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 199 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 200 break; 201 p = q; 202 } 203 204 /* 205 * If there is a preceding segment, it may provide some of 206 * our data already. If so, drop the data from the incoming 207 * segment. If it provides all of our data, drop us. 208 */ 209 if (p != NULL) { 210 register int i; 211 /* conversion to int (in i) handles seq wraparound */ 212 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 213 if (i > 0) { 214 if (i >= *tlenp) { 215 tcpstat.tcps_rcvduppack++; 216 tcpstat.tcps_rcvdupbyte += *tlenp; 217 m_freem(m); 218 FREE(te, M_TSEGQ); 219 /* 220 * Try to present any queued data 221 * at the left window edge to the user. 222 * This is needed after the 3-WHS 223 * completes. 224 */ 225 goto present; /* ??? */ 226 } 227 m_adj(m, i); 228 *tlenp -= i; 229 th->th_seq += i; 230 } 231 } 232 tcpstat.tcps_rcvoopack++; 233 tcpstat.tcps_rcvoobyte += *tlenp; 234 235 /* 236 * While we overlap succeeding segments trim them or, 237 * if they are completely covered, dequeue them. 238 */ 239 while (q) { 240 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 241 if (i <= 0) 242 break; 243 if (i < q->tqe_len) { 244 q->tqe_th->th_seq += i; 245 q->tqe_len -= i; 246 m_adj(q->tqe_m, i); 247 break; 248 } 249 250 nq = LIST_NEXT(q, tqe_q); 251 LIST_REMOVE(q, tqe_q); 252 m_freem(q->tqe_m); 253 FREE(q, M_TSEGQ); 254 q = nq; 255 } 256 257 /* Insert the new segment queue entry into place. */ 258 te->tqe_m = m; 259 te->tqe_th = th; 260 te->tqe_len = *tlenp; 261 262 if (p == NULL) { 263 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 264 } else { 265 LIST_INSERT_AFTER(p, te, tqe_q); 266 } 267 268present: 269 /* 270 * Present data to user, advancing rcv_nxt through 271 * completed sequence space. 272 */ 273 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 274 return (0); 275 q = LIST_FIRST(&tp->t_segq); 276 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 277 return (0); 278 do { 279 tp->rcv_nxt += q->tqe_len; 280 flags = q->tqe_th->th_flags & TH_FIN; 281 nq = LIST_NEXT(q, tqe_q); 282 LIST_REMOVE(q, tqe_q); 283 if (so->so_state & SS_CANTRCVMORE) 284 m_freem(q->tqe_m); 285 else 286 sbappend(&so->so_rcv, q->tqe_m); 287 FREE(q, M_TSEGQ); 288 q = nq; 289 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 290 ND6_HINT(tp); 291 sorwakeup(so); 292 return (flags); 293} 294 295/* 296 * TCP input routine, follows pages 65-76 of the 297 * protocol specification dated September, 1981 very closely. 298 */ 299#ifdef INET6 300int 301tcp6_input(mp, offp, proto) 302 struct mbuf **mp; 303 int *offp, proto; 304{ 305 register struct mbuf *m = *mp; 306 struct in6_ifaddr *ia6; 307 308 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 309 310 /* 311 * draft-itojun-ipv6-tcp-to-anycast 312 * better place to put this in? 313 */ 314 ia6 = ip6_getdstifaddr(m); 315 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 316 struct ip6_hdr *ip6; 317 318 ip6 = mtod(m, struct ip6_hdr *); 319 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 320 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 321 return IPPROTO_DONE; 322 } 323 324 tcp_input(m, *offp); 325 return IPPROTO_DONE; 326} 327#endif 328 329void 330tcp_input(m, off0) 331 register struct mbuf *m; 332 int off0; 333{ 334 register struct tcphdr *th; 335 register struct ip *ip = NULL; 336 register struct ipovly *ipov; 337 register struct inpcb *inp; 338 u_char *optp = NULL; 339 int optlen = 0; 340 int len, tlen, off; 341 int drop_hdrlen; 342 register struct tcpcb *tp = 0; 343 register int thflags; 344 struct socket *so = 0; 345 int todrop, acked, ourfinisacked, needoutput = 0; 346 u_long tiwin; 347 struct tcpopt to; /* options in this segment */ 348 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 349 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 350#ifdef TCPDEBUG 351 short ostate = 0; 352#endif 353#ifdef INET6 354 struct ip6_hdr *ip6 = NULL; 355 int isipv6; 356#endif /* INET6 */ 357 int rstreason; /* For badport_bandlim accounting purposes */ 358 359#ifdef INET6 360 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 361#endif 362 bzero((char *)&to, sizeof(to)); 363 364 tcpstat.tcps_rcvtotal++; 365 366#ifdef INET6 367 if (isipv6) { 368 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 369 ip6 = mtod(m, struct ip6_hdr *); 370 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 371 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 372 tcpstat.tcps_rcvbadsum++; 373 goto drop; 374 } 375 th = (struct tcphdr *)((caddr_t)ip6 + off0); 376 377 /* 378 * Be proactive about unspecified IPv6 address in source. 379 * As we use all-zero to indicate unbounded/unconnected pcb, 380 * unspecified IPv6 address can be used to confuse us. 381 * 382 * Note that packets with unspecified IPv6 destination is 383 * already dropped in ip6_input. 384 */ 385 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 386 /* XXX stat */ 387 goto drop; 388 } 389 } else 390#endif /* INET6 */ 391 { 392 /* 393 * Get IP and TCP header together in first mbuf. 394 * Note: IP leaves IP header in first mbuf. 395 */ 396 if (off0 > sizeof (struct ip)) { 397 ip_stripoptions(m, (struct mbuf *)0); 398 off0 = sizeof(struct ip); 399 } 400 if (m->m_len < sizeof (struct tcpiphdr)) { 401 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 402 tcpstat.tcps_rcvshort++; 403 return; 404 } 405 } 406 ip = mtod(m, struct ip *); 407 ipov = (struct ipovly *)ip; 408 th = (struct tcphdr *)((caddr_t)ip + off0); 409 tlen = ip->ip_len; 410 411 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 412 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 413 th->th_sum = m->m_pkthdr.csum_data; 414 else 415 th->th_sum = in_pseudo(ip->ip_src.s_addr, 416 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 417 ip->ip_len + IPPROTO_TCP)); 418 th->th_sum ^= 0xffff; 419 } else { 420 /* 421 * Checksum extended TCP header and data. 422 */ 423 len = sizeof (struct ip) + tlen; 424 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 425 ipov->ih_len = (u_short)tlen; 426 ipov->ih_len = htons(ipov->ih_len); 427 th->th_sum = in_cksum(m, len); 428 } 429 if (th->th_sum) { 430 tcpstat.tcps_rcvbadsum++; 431 goto drop; 432 } 433#ifdef INET6 434 /* Re-initialization for later version check */ 435 ip->ip_v = IPVERSION; 436#endif 437 } 438 439 /* 440 * Check that TCP offset makes sense, 441 * pull out TCP options and adjust length. XXX 442 */ 443 off = th->th_off << 2; 444 if (off < sizeof (struct tcphdr) || off > tlen) { 445 tcpstat.tcps_rcvbadoff++; 446 goto drop; 447 } 448 tlen -= off; /* tlen is used instead of ti->ti_len */ 449 if (off > sizeof (struct tcphdr)) { 450#ifdef INET6 451 if (isipv6) { 452 IP6_EXTHDR_CHECK(m, off0, off, ); 453 ip6 = mtod(m, struct ip6_hdr *); 454 th = (struct tcphdr *)((caddr_t)ip6 + off0); 455 } else 456#endif /* INET6 */ 457 { 458 if (m->m_len < sizeof(struct ip) + off) { 459 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 460 tcpstat.tcps_rcvshort++; 461 return; 462 } 463 ip = mtod(m, struct ip *); 464 ipov = (struct ipovly *)ip; 465 th = (struct tcphdr *)((caddr_t)ip + off0); 466 } 467 } 468 optlen = off - sizeof (struct tcphdr); 469 optp = (u_char *)(th + 1); 470 } 471 thflags = th->th_flags; 472 473#ifdef TCP_DROP_SYNFIN 474 /* 475 * If the drop_synfin option is enabled, drop all packets with 476 * both the SYN and FIN bits set. This prevents e.g. nmap from 477 * identifying the TCP/IP stack. 478 * 479 * This is a violation of the TCP specification. 480 */ 481 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 482 goto drop; 483#endif 484 485 /* 486 * Convert TCP protocol specific fields to host format. 487 */ 488 th->th_seq = ntohl(th->th_seq); 489 th->th_ack = ntohl(th->th_ack); 490 th->th_win = ntohs(th->th_win); 491 th->th_urp = ntohs(th->th_urp); 492 493 /* 494 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 495 * until after ip6_savecontrol() is called and before other functions 496 * which don't want those proto headers. 497 * Because ip6_savecontrol() is going to parse the mbuf to 498 * search for data to be passed up to user-land, it wants mbuf 499 * parameters to be unchanged. 500 * XXX: the call of ip6_savecontrol() has been obsoleted based on 501 * latest version of the advanced API (20020110). 502 */ 503 drop_hdrlen = off0 + off; 504 505 /* 506 * Locate pcb for segment. 507 */ 508findpcb: 509#ifdef IPFIREWALL_FORWARD 510 if (ip_fw_fwd_addr != NULL 511#ifdef INET6 512 && isipv6 == NULL /* IPv6 support is not yet */ 513#endif /* INET6 */ 514 ) { 515 /* 516 * Diverted. Pretend to be the destination. 517 * already got one like this? 518 */ 519 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 520 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 521 if (!inp) { 522 /* 523 * No, then it's new. Try find the ambushing socket 524 */ 525 if (!ip_fw_fwd_addr->sin_port) { 526 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 527 th->th_sport, ip_fw_fwd_addr->sin_addr, 528 th->th_dport, 1, m->m_pkthdr.rcvif); 529 } else { 530 inp = in_pcblookup_hash(&tcbinfo, 531 ip->ip_src, th->th_sport, 532 ip_fw_fwd_addr->sin_addr, 533 ntohs(ip_fw_fwd_addr->sin_port), 1, 534 m->m_pkthdr.rcvif); 535 } 536 } 537 ip_fw_fwd_addr = NULL; 538 } else 539#endif /* IPFIREWALL_FORWARD */ 540 { 541#ifdef INET6 542 if (isipv6) 543 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 544 &ip6->ip6_dst, th->th_dport, 1, 545 m->m_pkthdr.rcvif); 546 else 547#endif /* INET6 */ 548 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 549 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 550 } 551 552#ifdef IPSEC 553#ifdef INET6 554 if (isipv6) { 555 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 556 ipsec6stat.in_polvio++; 557 goto drop; 558 } 559 } else 560#endif /* INET6 */ 561 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 562 ipsecstat.in_polvio++; 563 goto drop; 564 } 565#endif /*IPSEC*/ 566 567 /* 568 * If the state is CLOSED (i.e., TCB does not exist) then 569 * all data in the incoming segment is discarded. 570 * If the TCB exists but is in CLOSED state, it is embryonic, 571 * but should either do a listen or a connect soon. 572 */ 573 if (inp == NULL) { 574 if (log_in_vain) { 575#ifdef INET6 576 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 577#else /* INET6 */ 578 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 579#endif /* INET6 */ 580 581#ifdef INET6 582 if (isipv6) { 583 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 584 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 585 } else 586#endif 587 { 588 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 589 strcpy(sbuf, inet_ntoa(ip->ip_src)); 590 } 591 switch (log_in_vain) { 592 case 1: 593 if(thflags & TH_SYN) 594 log(LOG_INFO, 595 "Connection attempt to TCP %s:%d from %s:%d\n", 596 dbuf, ntohs(th->th_dport), 597 sbuf, 598 ntohs(th->th_sport)); 599 break; 600 case 2: 601 log(LOG_INFO, 602 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 603 dbuf, ntohs(th->th_dport), sbuf, 604 ntohs(th->th_sport), thflags); 605 break; 606 default: 607 break; 608 } 609 } 610 if (blackhole) { 611 switch (blackhole) { 612 case 1: 613 if (thflags & TH_SYN) 614 goto drop; 615 break; 616 case 2: 617 goto drop; 618 default: 619 goto drop; 620 } 621 } 622 rstreason = BANDLIM_RST_CLOSEDPORT; 623 goto dropwithreset; 624 } 625 tp = intotcpcb(inp); 626 if (tp == 0) { 627 rstreason = BANDLIM_RST_CLOSEDPORT; 628 goto dropwithreset; 629 } 630 if (tp->t_state == TCPS_CLOSED) 631 goto drop; 632 633 /* Unscale the window into a 32-bit value. */ 634 if ((thflags & TH_SYN) == 0) 635 tiwin = th->th_win << tp->snd_scale; 636 else 637 tiwin = th->th_win; 638 639 so = inp->inp_socket; 640 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 641 struct in_conninfo inc; 642#ifdef TCPDEBUG 643 if (so->so_options & SO_DEBUG) { 644 ostate = tp->t_state; 645#ifdef INET6 646 if (isipv6) 647 bcopy((char *)ip6, (char *)tcp_saveipgen, 648 sizeof(*ip6)); 649 else 650#endif /* INET6 */ 651 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 652 tcp_savetcp = *th; 653 } 654#endif 655 /* skip if this isn't a listen socket */ 656 if ((so->so_options & SO_ACCEPTCONN) == 0) 657 goto after_listen; 658#ifdef INET6 659 inc.inc_isipv6 = isipv6; 660 if (isipv6) { 661 inc.inc6_faddr = ip6->ip6_src; 662 inc.inc6_laddr = ip6->ip6_dst; 663 inc.inc6_route.ro_rt = NULL; /* XXX */ 664 665 } else 666#endif /* INET6 */ 667 { 668 inc.inc_faddr = ip->ip_src; 669 inc.inc_laddr = ip->ip_dst; 670 inc.inc_route.ro_rt = NULL; /* XXX */ 671 } 672 inc.inc_fport = th->th_sport; 673 inc.inc_lport = th->th_dport; 674 675 /* 676 * If the state is LISTEN then ignore segment if it contains 677 * a RST. If the segment contains an ACK then it is bad and 678 * send a RST. If it does not contain a SYN then it is not 679 * interesting; drop it. 680 * 681 * If the state is SYN_RECEIVED (syncache) and seg contains 682 * an ACK, but not for our SYN/ACK, send a RST. If the seg 683 * contains a RST, check the sequence number to see if it 684 * is a valid reset segment. 685 */ 686 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 687 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 688 if (!syncache_expand(&inc, th, &so, m)) { 689 /* 690 * No syncache entry, or ACK was not 691 * for our SYN/ACK. Send a RST. 692 */ 693 tcpstat.tcps_badsyn++; 694 rstreason = BANDLIM_RST_OPENPORT; 695 goto dropwithreset; 696 } 697 if (so == NULL) 698 /* 699 * Could not complete 3-way handshake, 700 * connection is being closed down, and 701 * syncache will free mbuf. 702 */ 703 return; 704 /* 705 * Socket is created in state SYN_RECEIVED. 706 * Continue processing segment. 707 */ 708 inp = sotoinpcb(so); 709 tp = intotcpcb(inp); 710 /* 711 * This is what would have happened in 712 * tcp_output() when the SYN,ACK was sent. 713 */ 714 tp->snd_up = tp->snd_una; 715 tp->snd_max = tp->snd_nxt = tp->iss + 1; 716 tp->last_ack_sent = tp->rcv_nxt; 717/* 718 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 719 * until the _second_ ACK is received: 720 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 721 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 722 * move to ESTAB, set snd_wnd to tiwin. 723 */ 724 tp->snd_wnd = tiwin; /* unscaled */ 725 goto after_listen; 726 } 727 if (thflags & TH_RST) { 728 syncache_chkrst(&inc, th); 729 goto drop; 730 } 731 if (thflags & TH_ACK) { 732 syncache_badack(&inc); 733 tcpstat.tcps_badsyn++; 734 rstreason = BANDLIM_RST_OPENPORT; 735 goto dropwithreset; 736 } 737 goto drop; 738 } 739 740 /* 741 * Segment's flags are (SYN) or (SYN|FIN). 742 */ 743#ifdef INET6 744 /* 745 * If deprecated address is forbidden, 746 * we do not accept SYN to deprecated interface 747 * address to prevent any new inbound connection from 748 * getting established. 749 * When we do not accept SYN, we send a TCP RST, 750 * with deprecated source address (instead of dropping 751 * it). We compromise it as it is much better for peer 752 * to send a RST, and RST will be the final packet 753 * for the exchange. 754 * 755 * If we do not forbid deprecated addresses, we accept 756 * the SYN packet. RFC2462 does not suggest dropping 757 * SYN in this case. 758 * If we decipher RFC2462 5.5.4, it says like this: 759 * 1. use of deprecated addr with existing 760 * communication is okay - "SHOULD continue to be 761 * used" 762 * 2. use of it with new communication: 763 * (2a) "SHOULD NOT be used if alternate address 764 * with sufficient scope is available" 765 * (2b) nothing mentioned otherwise. 766 * Here we fall into (2b) case as we have no choice in 767 * our source address selection - we must obey the peer. 768 * 769 * The wording in RFC2462 is confusing, and there are 770 * multiple description text for deprecated address 771 * handling - worse, they are not exactly the same. 772 * I believe 5.5.4 is the best one, so we follow 5.5.4. 773 */ 774 if (isipv6 && !ip6_use_deprecated) { 775 struct in6_ifaddr *ia6; 776 777 if ((ia6 = ip6_getdstifaddr(m)) && 778 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 779 tp = NULL; 780 rstreason = BANDLIM_RST_OPENPORT; 781 goto dropwithreset; 782 } 783 } 784#endif 785 /* 786 * If it is from this socket, drop it, it must be forged. 787 * Don't bother responding if the destination was a broadcast. 788 */ 789 if (th->th_dport == th->th_sport) { 790#ifdef INET6 791 if (isipv6) { 792 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 793 &ip6->ip6_src)) 794 goto drop; 795 } else 796#endif /* INET6 */ 797 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 798 goto drop; 799 } 800 /* 801 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 802 * 803 * Note that it is quite possible to receive unicast 804 * link-layer packets with a broadcast IP address. Use 805 * in_broadcast() to find them. 806 */ 807 if (m->m_flags & (M_BCAST|M_MCAST)) 808 goto drop; 809#ifdef INET6 810 if (isipv6) { 811 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 812 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 813 goto drop; 814 } else 815#endif 816 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 817 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 818 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 819 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 820 goto drop; 821 /* 822 * SYN appears to be valid; create compressed TCP state 823 * for syncache, or perform t/tcp connection. 824 */ 825 if (so->so_qlen <= so->so_qlimit) { 826 tcp_dooptions(&to, optp, optlen, 1); 827 if (!syncache_add(&inc, &to, th, &so, m)) 828 goto drop; 829 if (so == NULL) 830 /* 831 * Entry added to syncache, mbuf used to 832 * send SYN,ACK packet. 833 */ 834 return; 835 /* 836 * Segment passed TAO tests. 837 */ 838 inp = sotoinpcb(so); 839 tp = intotcpcb(inp); 840 tp->snd_wnd = tiwin; 841 tp->t_starttime = ticks; 842 tp->t_state = TCPS_ESTABLISHED; 843 844 /* 845 * If there is a FIN, or if there is data and the 846 * connection is local, then delay SYN,ACK(SYN) in 847 * the hope of piggy-backing it on a response 848 * segment. Otherwise must send ACK now in case 849 * the other side is slow starting. 850 */ 851 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 852 (tlen != 0 && 853#ifdef INET6 854 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 855 || 856 (!isipv6 && 857#endif 858 in_localaddr(inp->inp_faddr) 859#ifdef INET6 860 )) 861#endif 862 ))) { 863 callout_reset(tp->tt_delack, tcp_delacktime, 864 tcp_timer_delack, tp); 865 tp->t_flags |= TF_NEEDSYN; 866 } else 867 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 868 869 tcpstat.tcps_connects++; 870 soisconnected(so); 871 goto trimthenstep6; 872 } 873 goto drop; 874 } 875after_listen: 876 877/* XXX temp debugging */ 878 /* should not happen - syncache should pick up these connections */ 879 if (tp->t_state == TCPS_LISTEN) 880 panic("tcp_input: TCPS_LISTEN"); 881 882 /* 883 * Segment received on connection. 884 * Reset idle time and keep-alive timer. 885 */ 886 tp->t_rcvtime = ticks; 887 if (TCPS_HAVEESTABLISHED(tp->t_state)) 888 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 889 890 /* 891 * Process options. 892 * XXX this is tradtitional behavior, may need to be cleaned up. 893 */ 894 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 895 if (thflags & TH_SYN) { 896 if (to.to_flags & TOF_SCALE) { 897 tp->t_flags |= TF_RCVD_SCALE; 898 tp->requested_s_scale = to.to_requested_s_scale; 899 } 900 if (to.to_flags & TOF_TS) { 901 tp->t_flags |= TF_RCVD_TSTMP; 902 tp->ts_recent = to.to_tsval; 903 tp->ts_recent_age = ticks; 904 } 905 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 906 tp->t_flags |= TF_RCVD_CC; 907 if (to.to_flags & TOF_MSS) 908 tcp_mss(tp, to.to_mss); 909 } 910 911 /* 912 * Header prediction: check for the two common cases 913 * of a uni-directional data xfer. If the packet has 914 * no control flags, is in-sequence, the window didn't 915 * change and we're not retransmitting, it's a 916 * candidate. If the length is zero and the ack moved 917 * forward, we're the sender side of the xfer. Just 918 * free the data acked & wake any higher level process 919 * that was blocked waiting for space. If the length 920 * is non-zero and the ack didn't move, we're the 921 * receiver side. If we're getting packets in-order 922 * (the reassembly queue is empty), add the data to 923 * the socket buffer and note that we need a delayed ack. 924 * Make sure that the hidden state-flags are also off. 925 * Since we check for TCPS_ESTABLISHED above, it can only 926 * be TH_NEEDSYN. 927 */ 928 if (tp->t_state == TCPS_ESTABLISHED && 929 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 930 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 931 ((to.to_flags & TOF_TS) == 0 || 932 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 933 /* 934 * Using the CC option is compulsory if once started: 935 * the segment is OK if no T/TCP was negotiated or 936 * if the segment has a CC option equal to CCrecv 937 */ 938 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 939 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 940 th->th_seq == tp->rcv_nxt && 941 tiwin && tiwin == tp->snd_wnd && 942 tp->snd_nxt == tp->snd_max) { 943 944 /* 945 * If last ACK falls within this segment's sequence numbers, 946 * record the timestamp. 947 * NOTE that the test is modified according to the latest 948 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 949 */ 950 if ((to.to_flags & TOF_TS) != 0 && 951 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 952 tp->ts_recent_age = ticks; 953 tp->ts_recent = to.to_tsval; 954 } 955 956 if (tlen == 0) { 957 if (SEQ_GT(th->th_ack, tp->snd_una) && 958 SEQ_LEQ(th->th_ack, tp->snd_max) && 959 tp->snd_cwnd >= tp->snd_wnd && 960 tp->t_dupacks < tcprexmtthresh) { 961 /* 962 * this is a pure ack for outstanding data. 963 */ 964 ++tcpstat.tcps_predack; 965 /* 966 * "bad retransmit" recovery 967 */ 968 if (tp->t_rxtshift == 1 && 969 ticks < tp->t_badrxtwin) { 970 tp->snd_cwnd = tp->snd_cwnd_prev; 971 tp->snd_ssthresh = 972 tp->snd_ssthresh_prev; 973 tp->snd_nxt = tp->snd_max; 974 tp->t_badrxtwin = 0; 975 } 976 if ((to.to_flags & TOF_TS) != 0) 977 tcp_xmit_timer(tp, 978 ticks - to.to_tsecr + 1); 979 else if (tp->t_rtttime && 980 SEQ_GT(th->th_ack, tp->t_rtseq)) 981 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 982 acked = th->th_ack - tp->snd_una; 983 tcpstat.tcps_rcvackpack++; 984 tcpstat.tcps_rcvackbyte += acked; 985 sbdrop(&so->so_snd, acked); 986 tp->snd_una = th->th_ack; 987 m_freem(m); 988 ND6_HINT(tp); /* some progress has been done */ 989 990 /* 991 * If all outstanding data are acked, stop 992 * retransmit timer, otherwise restart timer 993 * using current (possibly backed-off) value. 994 * If process is waiting for space, 995 * wakeup/selwakeup/signal. If data 996 * are ready to send, let tcp_output 997 * decide between more output or persist. 998 */ 999 if (tp->snd_una == tp->snd_max) 1000 callout_stop(tp->tt_rexmt); 1001 else if (!callout_active(tp->tt_persist)) 1002 callout_reset(tp->tt_rexmt, 1003 tp->t_rxtcur, 1004 tcp_timer_rexmt, tp); 1005 1006 sowwakeup(so); 1007 if (so->so_snd.sb_cc) 1008 (void) tcp_output(tp); 1009 return; 1010 } 1011 } else if (th->th_ack == tp->snd_una && 1012 LIST_EMPTY(&tp->t_segq) && 1013 tlen <= sbspace(&so->so_rcv)) { 1014 /* 1015 * this is a pure, in-sequence data packet 1016 * with nothing on the reassembly queue and 1017 * we have enough buffer space to take it. 1018 */ 1019 ++tcpstat.tcps_preddat; 1020 tp->rcv_nxt += tlen; 1021 tcpstat.tcps_rcvpack++; 1022 tcpstat.tcps_rcvbyte += tlen; 1023 ND6_HINT(tp); /* some progress has been done */ 1024 /* 1025 * Add data to socket buffer. 1026 */ 1027 m_adj(m, drop_hdrlen); /* delayed header drop */ 1028 sbappend(&so->so_rcv, m); 1029 sorwakeup(so); 1030 if (DELAY_ACK(tp)) { 1031 callout_reset(tp->tt_delack, tcp_delacktime, 1032 tcp_timer_delack, tp); 1033 } else { 1034 tp->t_flags |= TF_ACKNOW; 1035 tcp_output(tp); 1036 } 1037 return; 1038 } 1039 } 1040 1041 /* 1042 * Calculate amount of space in receive window, 1043 * and then do TCP input processing. 1044 * Receive window is amount of space in rcv queue, 1045 * but not less than advertised window. 1046 */ 1047 { int win; 1048 1049 win = sbspace(&so->so_rcv); 1050 if (win < 0) 1051 win = 0; 1052 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1053 } 1054 1055 switch (tp->t_state) { 1056 1057 /* 1058 * If the state is SYN_RECEIVED: 1059 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1060 */ 1061 case TCPS_SYN_RECEIVED: 1062 if ((thflags & TH_ACK) && 1063 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1064 SEQ_GT(th->th_ack, tp->snd_max))) { 1065 rstreason = BANDLIM_RST_OPENPORT; 1066 goto dropwithreset; 1067 } 1068 break; 1069 1070 /* 1071 * If the state is SYN_SENT: 1072 * if seg contains an ACK, but not for our SYN, drop the input. 1073 * if seg contains a RST, then drop the connection. 1074 * if seg does not contain SYN, then drop it. 1075 * Otherwise this is an acceptable SYN segment 1076 * initialize tp->rcv_nxt and tp->irs 1077 * if seg contains ack then advance tp->snd_una 1078 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1079 * arrange for segment to be acked (eventually) 1080 * continue processing rest of data/controls, beginning with URG 1081 */ 1082 case TCPS_SYN_SENT: 1083 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1084 taop = &tao_noncached; 1085 bzero(taop, sizeof(*taop)); 1086 } 1087 1088 if ((thflags & TH_ACK) && 1089 (SEQ_LEQ(th->th_ack, tp->iss) || 1090 SEQ_GT(th->th_ack, tp->snd_max))) { 1091 /* 1092 * If we have a cached CCsent for the remote host, 1093 * hence we haven't just crashed and restarted, 1094 * do not send a RST. This may be a retransmission 1095 * from the other side after our earlier ACK was lost. 1096 * Our new SYN, when it arrives, will serve as the 1097 * needed ACK. 1098 */ 1099 if (taop->tao_ccsent != 0) 1100 goto drop; 1101 else { 1102 rstreason = BANDLIM_UNLIMITED; 1103 goto dropwithreset; 1104 } 1105 } 1106 if (thflags & TH_RST) { 1107 if (thflags & TH_ACK) 1108 tp = tcp_drop(tp, ECONNREFUSED); 1109 goto drop; 1110 } 1111 if ((thflags & TH_SYN) == 0) 1112 goto drop; 1113 tp->snd_wnd = th->th_win; /* initial send window */ 1114 tp->cc_recv = to.to_cc; /* foreign CC */ 1115 1116 tp->irs = th->th_seq; 1117 tcp_rcvseqinit(tp); 1118 if (thflags & TH_ACK) { 1119 /* 1120 * Our SYN was acked. If segment contains CC.ECHO 1121 * option, check it to make sure this segment really 1122 * matches our SYN. If not, just drop it as old 1123 * duplicate, but send an RST if we're still playing 1124 * by the old rules. If no CC.ECHO option, make sure 1125 * we don't get fooled into using T/TCP. 1126 */ 1127 if (to.to_flags & TOF_CCECHO) { 1128 if (tp->cc_send != to.to_ccecho) { 1129 if (taop->tao_ccsent != 0) 1130 goto drop; 1131 else { 1132 rstreason = BANDLIM_UNLIMITED; 1133 goto dropwithreset; 1134 } 1135 } 1136 } else 1137 tp->t_flags &= ~TF_RCVD_CC; 1138 tcpstat.tcps_connects++; 1139 soisconnected(so); 1140 /* Do window scaling on this connection? */ 1141 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1142 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1143 tp->snd_scale = tp->requested_s_scale; 1144 tp->rcv_scale = tp->request_r_scale; 1145 } 1146 /* Segment is acceptable, update cache if undefined. */ 1147 if (taop->tao_ccsent == 0) 1148 taop->tao_ccsent = to.to_ccecho; 1149 1150 tp->rcv_adv += tp->rcv_wnd; 1151 tp->snd_una++; /* SYN is acked */ 1152 /* 1153 * If there's data, delay ACK; if there's also a FIN 1154 * ACKNOW will be turned on later. 1155 */ 1156 if (DELAY_ACK(tp) && tlen != 0) 1157 callout_reset(tp->tt_delack, tcp_delacktime, 1158 tcp_timer_delack, tp); 1159 else 1160 tp->t_flags |= TF_ACKNOW; 1161 /* 1162 * Received <SYN,ACK> in SYN_SENT[*] state. 1163 * Transitions: 1164 * SYN_SENT --> ESTABLISHED 1165 * SYN_SENT* --> FIN_WAIT_1 1166 */ 1167 tp->t_starttime = ticks; 1168 if (tp->t_flags & TF_NEEDFIN) { 1169 tp->t_state = TCPS_FIN_WAIT_1; 1170 tp->t_flags &= ~TF_NEEDFIN; 1171 thflags &= ~TH_SYN; 1172 } else { 1173 tp->t_state = TCPS_ESTABLISHED; 1174 callout_reset(tp->tt_keep, tcp_keepidle, 1175 tcp_timer_keep, tp); 1176 } 1177 } else { 1178 /* 1179 * Received initial SYN in SYN-SENT[*] state => simul- 1180 * taneous open. If segment contains CC option and there is 1181 * a cached CC, apply TAO test; if it succeeds, connection is 1182 * half-synchronized. Otherwise, do 3-way handshake: 1183 * SYN-SENT -> SYN-RECEIVED 1184 * SYN-SENT* -> SYN-RECEIVED* 1185 * If there was no CC option, clear cached CC value. 1186 */ 1187 tp->t_flags |= TF_ACKNOW; 1188 callout_stop(tp->tt_rexmt); 1189 if (to.to_flags & TOF_CC) { 1190 if (taop->tao_cc != 0 && 1191 CC_GT(to.to_cc, taop->tao_cc)) { 1192 /* 1193 * update cache and make transition: 1194 * SYN-SENT -> ESTABLISHED* 1195 * SYN-SENT* -> FIN-WAIT-1* 1196 */ 1197 taop->tao_cc = to.to_cc; 1198 tp->t_starttime = ticks; 1199 if (tp->t_flags & TF_NEEDFIN) { 1200 tp->t_state = TCPS_FIN_WAIT_1; 1201 tp->t_flags &= ~TF_NEEDFIN; 1202 } else { 1203 tp->t_state = TCPS_ESTABLISHED; 1204 callout_reset(tp->tt_keep, 1205 tcp_keepidle, 1206 tcp_timer_keep, 1207 tp); 1208 } 1209 tp->t_flags |= TF_NEEDSYN; 1210 } else 1211 tp->t_state = TCPS_SYN_RECEIVED; 1212 } else { 1213 /* CC.NEW or no option => invalidate cache */ 1214 taop->tao_cc = 0; 1215 tp->t_state = TCPS_SYN_RECEIVED; 1216 } 1217 } 1218 1219trimthenstep6: 1220 /* 1221 * Advance th->th_seq to correspond to first data byte. 1222 * If data, trim to stay within window, 1223 * dropping FIN if necessary. 1224 */ 1225 th->th_seq++; 1226 if (tlen > tp->rcv_wnd) { 1227 todrop = tlen - tp->rcv_wnd; 1228 m_adj(m, -todrop); 1229 tlen = tp->rcv_wnd; 1230 thflags &= ~TH_FIN; 1231 tcpstat.tcps_rcvpackafterwin++; 1232 tcpstat.tcps_rcvbyteafterwin += todrop; 1233 } 1234 tp->snd_wl1 = th->th_seq - 1; 1235 tp->rcv_up = th->th_seq; 1236 /* 1237 * Client side of transaction: already sent SYN and data. 1238 * If the remote host used T/TCP to validate the SYN, 1239 * our data will be ACK'd; if so, enter normal data segment 1240 * processing in the middle of step 5, ack processing. 1241 * Otherwise, goto step 6. 1242 */ 1243 if (thflags & TH_ACK) 1244 goto process_ACK; 1245 goto step6; 1246 /* 1247 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1248 * if segment contains a SYN and CC [not CC.NEW] option: 1249 * if state == TIME_WAIT and connection duration > MSL, 1250 * drop packet and send RST; 1251 * 1252 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1253 * ack the FIN (and data) in retransmission queue. 1254 * Complete close and delete TCPCB. Then reprocess 1255 * segment, hoping to find new TCPCB in LISTEN state; 1256 * 1257 * else must be old SYN; drop it. 1258 * else do normal processing. 1259 */ 1260 case TCPS_LAST_ACK: 1261 case TCPS_CLOSING: 1262 case TCPS_TIME_WAIT: 1263 if ((thflags & TH_SYN) && 1264 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1265 if (tp->t_state == TCPS_TIME_WAIT && 1266 (ticks - tp->t_starttime) > tcp_msl) { 1267 rstreason = BANDLIM_UNLIMITED; 1268 goto dropwithreset; 1269 } 1270 if (CC_GT(to.to_cc, tp->cc_recv)) { 1271 tp = tcp_close(tp); 1272 goto findpcb; 1273 } 1274 else 1275 goto drop; 1276 } 1277 break; /* continue normal processing */ 1278 } 1279 1280 /* 1281 * States other than LISTEN or SYN_SENT. 1282 * First check the RST flag and sequence number since reset segments 1283 * are exempt from the timestamp and connection count tests. This 1284 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1285 * below which allowed reset segments in half the sequence space 1286 * to fall though and be processed (which gives forged reset 1287 * segments with a random sequence number a 50 percent chance of 1288 * killing a connection). 1289 * Then check timestamp, if present. 1290 * Then check the connection count, if present. 1291 * Then check that at least some bytes of segment are within 1292 * receive window. If segment begins before rcv_nxt, 1293 * drop leading data (and SYN); if nothing left, just ack. 1294 * 1295 * 1296 * If the RST bit is set, check the sequence number to see 1297 * if this is a valid reset segment. 1298 * RFC 793 page 37: 1299 * In all states except SYN-SENT, all reset (RST) segments 1300 * are validated by checking their SEQ-fields. A reset is 1301 * valid if its sequence number is in the window. 1302 * Note: this does not take into account delayed ACKs, so 1303 * we should test against last_ack_sent instead of rcv_nxt. 1304 * The sequence number in the reset segment is normally an 1305 * echo of our outgoing acknowlegement numbers, but some hosts 1306 * send a reset with the sequence number at the rightmost edge 1307 * of our receive window, and we have to handle this case. 1308 * If we have multiple segments in flight, the intial reset 1309 * segment sequence numbers will be to the left of last_ack_sent, 1310 * but they will eventually catch up. 1311 * In any case, it never made sense to trim reset segments to 1312 * fit the receive window since RFC 1122 says: 1313 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1314 * 1315 * A TCP SHOULD allow a received RST segment to include data. 1316 * 1317 * DISCUSSION 1318 * It has been suggested that a RST segment could contain 1319 * ASCII text that encoded and explained the cause of the 1320 * RST. No standard has yet been established for such 1321 * data. 1322 * 1323 * If the reset segment passes the sequence number test examine 1324 * the state: 1325 * SYN_RECEIVED STATE: 1326 * If passive open, return to LISTEN state. 1327 * If active open, inform user that connection was refused. 1328 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1329 * Inform user that connection was reset, and close tcb. 1330 * CLOSING, LAST_ACK STATES: 1331 * Close the tcb. 1332 * TIME_WAIT STATE: 1333 * Drop the segment - see Stevens, vol. 2, p. 964 and 1334 * RFC 1337. 1335 */ 1336 if (thflags & TH_RST) { 1337 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1338 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1339 switch (tp->t_state) { 1340 1341 case TCPS_SYN_RECEIVED: 1342 so->so_error = ECONNREFUSED; 1343 goto close; 1344 1345 case TCPS_ESTABLISHED: 1346 case TCPS_FIN_WAIT_1: 1347 case TCPS_FIN_WAIT_2: 1348 case TCPS_CLOSE_WAIT: 1349 so->so_error = ECONNRESET; 1350 close: 1351 tp->t_state = TCPS_CLOSED; 1352 tcpstat.tcps_drops++; 1353 tp = tcp_close(tp); 1354 break; 1355 1356 case TCPS_CLOSING: 1357 case TCPS_LAST_ACK: 1358 tp = tcp_close(tp); 1359 break; 1360 1361 case TCPS_TIME_WAIT: 1362 break; 1363 } 1364 } 1365 goto drop; 1366 } 1367 1368 /* 1369 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1370 * and it's less than ts_recent, drop it. 1371 */ 1372 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1373 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1374 1375 /* Check to see if ts_recent is over 24 days old. */ 1376 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1377 /* 1378 * Invalidate ts_recent. If this segment updates 1379 * ts_recent, the age will be reset later and ts_recent 1380 * will get a valid value. If it does not, setting 1381 * ts_recent to zero will at least satisfy the 1382 * requirement that zero be placed in the timestamp 1383 * echo reply when ts_recent isn't valid. The 1384 * age isn't reset until we get a valid ts_recent 1385 * because we don't want out-of-order segments to be 1386 * dropped when ts_recent is old. 1387 */ 1388 tp->ts_recent = 0; 1389 } else { 1390 tcpstat.tcps_rcvduppack++; 1391 tcpstat.tcps_rcvdupbyte += tlen; 1392 tcpstat.tcps_pawsdrop++; 1393 goto dropafterack; 1394 } 1395 } 1396 1397 /* 1398 * T/TCP mechanism 1399 * If T/TCP was negotiated and the segment doesn't have CC, 1400 * or if its CC is wrong then drop the segment. 1401 * RST segments do not have to comply with this. 1402 */ 1403 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1404 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1405 goto dropafterack; 1406 1407 /* 1408 * In the SYN-RECEIVED state, validate that the packet belongs to 1409 * this connection before trimming the data to fit the receive 1410 * window. Check the sequence number versus IRS since we know 1411 * the sequence numbers haven't wrapped. This is a partial fix 1412 * for the "LAND" DoS attack. 1413 */ 1414 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1415 rstreason = BANDLIM_RST_OPENPORT; 1416 goto dropwithreset; 1417 } 1418 1419 todrop = tp->rcv_nxt - th->th_seq; 1420 if (todrop > 0) { 1421 if (thflags & TH_SYN) { 1422 thflags &= ~TH_SYN; 1423 th->th_seq++; 1424 if (th->th_urp > 1) 1425 th->th_urp--; 1426 else 1427 thflags &= ~TH_URG; 1428 todrop--; 1429 } 1430 /* 1431 * Following if statement from Stevens, vol. 2, p. 960. 1432 */ 1433 if (todrop > tlen 1434 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1435 /* 1436 * Any valid FIN must be to the left of the window. 1437 * At this point the FIN must be a duplicate or out 1438 * of sequence; drop it. 1439 */ 1440 thflags &= ~TH_FIN; 1441 1442 /* 1443 * Send an ACK to resynchronize and drop any data. 1444 * But keep on processing for RST or ACK. 1445 */ 1446 tp->t_flags |= TF_ACKNOW; 1447 todrop = tlen; 1448 tcpstat.tcps_rcvduppack++; 1449 tcpstat.tcps_rcvdupbyte += todrop; 1450 } else { 1451 tcpstat.tcps_rcvpartduppack++; 1452 tcpstat.tcps_rcvpartdupbyte += todrop; 1453 } 1454 drop_hdrlen += todrop; /* drop from the top afterwards */ 1455 th->th_seq += todrop; 1456 tlen -= todrop; 1457 if (th->th_urp > todrop) 1458 th->th_urp -= todrop; 1459 else { 1460 thflags &= ~TH_URG; 1461 th->th_urp = 0; 1462 } 1463 } 1464 1465 /* 1466 * If new data are received on a connection after the 1467 * user processes are gone, then RST the other end. 1468 */ 1469 if ((so->so_state & SS_NOFDREF) && 1470 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1471 tp = tcp_close(tp); 1472 tcpstat.tcps_rcvafterclose++; 1473 rstreason = BANDLIM_UNLIMITED; 1474 goto dropwithreset; 1475 } 1476 1477 /* 1478 * If segment ends after window, drop trailing data 1479 * (and PUSH and FIN); if nothing left, just ACK. 1480 */ 1481 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1482 if (todrop > 0) { 1483 tcpstat.tcps_rcvpackafterwin++; 1484 if (todrop >= tlen) { 1485 tcpstat.tcps_rcvbyteafterwin += tlen; 1486 /* 1487 * If a new connection request is received 1488 * while in TIME_WAIT, drop the old connection 1489 * and start over if the sequence numbers 1490 * are above the previous ones. 1491 */ 1492 if (thflags & TH_SYN && 1493 tp->t_state == TCPS_TIME_WAIT && 1494 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1495 tp = tcp_close(tp); 1496 goto findpcb; 1497 } 1498 /* 1499 * If window is closed can only take segments at 1500 * window edge, and have to drop data and PUSH from 1501 * incoming segments. Continue processing, but 1502 * remember to ack. Otherwise, drop segment 1503 * and ack. 1504 */ 1505 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1506 tp->t_flags |= TF_ACKNOW; 1507 tcpstat.tcps_rcvwinprobe++; 1508 } else 1509 goto dropafterack; 1510 } else 1511 tcpstat.tcps_rcvbyteafterwin += todrop; 1512 m_adj(m, -todrop); 1513 tlen -= todrop; 1514 thflags &= ~(TH_PUSH|TH_FIN); 1515 } 1516 1517 /* 1518 * If last ACK falls within this segment's sequence numbers, 1519 * record its timestamp. 1520 * NOTE that the test is modified according to the latest 1521 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1522 */ 1523 if ((to.to_flags & TOF_TS) != 0 && 1524 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1525 tp->ts_recent_age = ticks; 1526 tp->ts_recent = to.to_tsval; 1527 } 1528 1529 /* 1530 * If a SYN is in the window, then this is an 1531 * error and we send an RST and drop the connection. 1532 */ 1533 if (thflags & TH_SYN) { 1534 tp = tcp_drop(tp, ECONNRESET); 1535 rstreason = BANDLIM_UNLIMITED; 1536 goto dropwithreset; 1537 } 1538 1539 /* 1540 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1541 * flag is on (half-synchronized state), then queue data for 1542 * later processing; else drop segment and return. 1543 */ 1544 if ((thflags & TH_ACK) == 0) { 1545 if (tp->t_state == TCPS_SYN_RECEIVED || 1546 (tp->t_flags & TF_NEEDSYN)) 1547 goto step6; 1548 else 1549 goto drop; 1550 } 1551 1552 /* 1553 * Ack processing. 1554 */ 1555 switch (tp->t_state) { 1556 1557 /* 1558 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1559 * ESTABLISHED state and continue processing. 1560 * The ACK was checked above. 1561 */ 1562 case TCPS_SYN_RECEIVED: 1563 1564 tcpstat.tcps_connects++; 1565 soisconnected(so); 1566 /* Do window scaling? */ 1567 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1568 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1569 tp->snd_scale = tp->requested_s_scale; 1570 tp->rcv_scale = tp->request_r_scale; 1571 } 1572 /* 1573 * Upon successful completion of 3-way handshake, 1574 * update cache.CC if it was undefined, pass any queued 1575 * data to the user, and advance state appropriately. 1576 */ 1577 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1578 taop->tao_cc == 0) 1579 taop->tao_cc = tp->cc_recv; 1580 1581 /* 1582 * Make transitions: 1583 * SYN-RECEIVED -> ESTABLISHED 1584 * SYN-RECEIVED* -> FIN-WAIT-1 1585 */ 1586 tp->t_starttime = ticks; 1587 if (tp->t_flags & TF_NEEDFIN) { 1588 tp->t_state = TCPS_FIN_WAIT_1; 1589 tp->t_flags &= ~TF_NEEDFIN; 1590 } else { 1591 tp->t_state = TCPS_ESTABLISHED; 1592 callout_reset(tp->tt_keep, tcp_keepidle, 1593 tcp_timer_keep, tp); 1594 } 1595 /* 1596 * If segment contains data or ACK, will call tcp_reass() 1597 * later; if not, do so now to pass queued data to user. 1598 */ 1599 if (tlen == 0 && (thflags & TH_FIN) == 0) 1600 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1601 (struct mbuf *)0); 1602 tp->snd_wl1 = th->th_seq - 1; 1603 /* fall into ... */ 1604 1605 /* 1606 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1607 * ACKs. If the ack is in the range 1608 * tp->snd_una < th->th_ack <= tp->snd_max 1609 * then advance tp->snd_una to th->th_ack and drop 1610 * data from the retransmission queue. If this ACK reflects 1611 * more up to date window information we update our window information. 1612 */ 1613 case TCPS_ESTABLISHED: 1614 case TCPS_FIN_WAIT_1: 1615 case TCPS_FIN_WAIT_2: 1616 case TCPS_CLOSE_WAIT: 1617 case TCPS_CLOSING: 1618 case TCPS_LAST_ACK: 1619 case TCPS_TIME_WAIT: 1620 1621 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1622 if (tlen == 0 && tiwin == tp->snd_wnd) { 1623 tcpstat.tcps_rcvdupack++; 1624 /* 1625 * If we have outstanding data (other than 1626 * a window probe), this is a completely 1627 * duplicate ack (ie, window info didn't 1628 * change), the ack is the biggest we've 1629 * seen and we've seen exactly our rexmt 1630 * threshhold of them, assume a packet 1631 * has been dropped and retransmit it. 1632 * Kludge snd_nxt & the congestion 1633 * window so we send only this one 1634 * packet. 1635 * 1636 * We know we're losing at the current 1637 * window size so do congestion avoidance 1638 * (set ssthresh to half the current window 1639 * and pull our congestion window back to 1640 * the new ssthresh). 1641 * 1642 * Dup acks mean that packets have left the 1643 * network (they're now cached at the receiver) 1644 * so bump cwnd by the amount in the receiver 1645 * to keep a constant cwnd packets in the 1646 * network. 1647 */ 1648 if (!callout_active(tp->tt_rexmt) || 1649 th->th_ack != tp->snd_una) 1650 tp->t_dupacks = 0; 1651 else if (++tp->t_dupacks == tcprexmtthresh) { 1652 tcp_seq onxt = tp->snd_nxt; 1653 u_int win = 1654 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1655 tp->t_maxseg; 1656 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1657 tp->snd_recover)) { 1658 /* False retransmit, should not 1659 * cut window 1660 */ 1661 tp->snd_cwnd += tp->t_maxseg; 1662 tp->t_dupacks = 0; 1663 (void) tcp_output(tp); 1664 goto drop; 1665 } 1666 if (win < 2) 1667 win = 2; 1668 tp->snd_ssthresh = win * tp->t_maxseg; 1669 tp->snd_recover = tp->snd_max; 1670 callout_stop(tp->tt_rexmt); 1671 tp->t_rtttime = 0; 1672 tp->snd_nxt = th->th_ack; 1673 tp->snd_cwnd = tp->t_maxseg; 1674 (void) tcp_output(tp); 1675 tp->snd_cwnd = tp->snd_ssthresh + 1676 tp->t_maxseg * tp->t_dupacks; 1677 if (SEQ_GT(onxt, tp->snd_nxt)) 1678 tp->snd_nxt = onxt; 1679 goto drop; 1680 } else if (tp->t_dupacks > tcprexmtthresh) { 1681 tp->snd_cwnd += tp->t_maxseg; 1682 (void) tcp_output(tp); 1683 goto drop; 1684 } 1685 } else 1686 tp->t_dupacks = 0; 1687 break; 1688 } 1689 /* 1690 * If the congestion window was inflated to account 1691 * for the other side's cached packets, retract it. 1692 */ 1693 if (tcp_do_newreno == 0) { 1694 if (tp->t_dupacks >= tcprexmtthresh && 1695 tp->snd_cwnd > tp->snd_ssthresh) 1696 tp->snd_cwnd = tp->snd_ssthresh; 1697 tp->t_dupacks = 0; 1698 } else if (tp->t_dupacks >= tcprexmtthresh && 1699 !tcp_newreno(tp, th)) { 1700 /* 1701 * Window inflation should have left us with approx. 1702 * snd_ssthresh outstanding data. But in case we 1703 * would be inclined to send a burst, better to do 1704 * it via the slow start mechanism. 1705 */ 1706 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1707 tp->snd_cwnd = 1708 tp->snd_max - th->th_ack + tp->t_maxseg; 1709 else 1710 tp->snd_cwnd = tp->snd_ssthresh; 1711 tp->t_dupacks = 0; 1712 } 1713 if (tp->t_dupacks < tcprexmtthresh) 1714 tp->t_dupacks = 0; 1715 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1716 tcpstat.tcps_rcvacktoomuch++; 1717 goto dropafterack; 1718 } 1719 /* 1720 * If we reach this point, ACK is not a duplicate, 1721 * i.e., it ACKs something we sent. 1722 */ 1723 if (tp->t_flags & TF_NEEDSYN) { 1724 /* 1725 * T/TCP: Connection was half-synchronized, and our 1726 * SYN has been ACK'd (so connection is now fully 1727 * synchronized). Go to non-starred state, 1728 * increment snd_una for ACK of SYN, and check if 1729 * we can do window scaling. 1730 */ 1731 tp->t_flags &= ~TF_NEEDSYN; 1732 tp->snd_una++; 1733 /* Do window scaling? */ 1734 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1735 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1736 tp->snd_scale = tp->requested_s_scale; 1737 tp->rcv_scale = tp->request_r_scale; 1738 } 1739 } 1740 1741process_ACK: 1742 acked = th->th_ack - tp->snd_una; 1743 tcpstat.tcps_rcvackpack++; 1744 tcpstat.tcps_rcvackbyte += acked; 1745 1746 /* 1747 * If we just performed our first retransmit, and the ACK 1748 * arrives within our recovery window, then it was a mistake 1749 * to do the retransmit in the first place. Recover our 1750 * original cwnd and ssthresh, and proceed to transmit where 1751 * we left off. 1752 */ 1753 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1754 tp->snd_cwnd = tp->snd_cwnd_prev; 1755 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1756 tp->snd_nxt = tp->snd_max; 1757 tp->t_badrxtwin = 0; /* XXX probably not required */ 1758 } 1759 1760 /* 1761 * If we have a timestamp reply, update smoothed 1762 * round trip time. If no timestamp is present but 1763 * transmit timer is running and timed sequence 1764 * number was acked, update smoothed round trip time. 1765 * Since we now have an rtt measurement, cancel the 1766 * timer backoff (cf., Phil Karn's retransmit alg.). 1767 * Recompute the initial retransmit timer. 1768 */ 1769 if (to.to_flags & TOF_TS) 1770 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1771 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1772 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1773 1774 /* 1775 * If all outstanding data is acked, stop retransmit 1776 * timer and remember to restart (more output or persist). 1777 * If there is more data to be acked, restart retransmit 1778 * timer, using current (possibly backed-off) value. 1779 */ 1780 if (th->th_ack == tp->snd_max) { 1781 callout_stop(tp->tt_rexmt); 1782 needoutput = 1; 1783 } else if (!callout_active(tp->tt_persist)) 1784 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1785 tcp_timer_rexmt, tp); 1786 1787 /* 1788 * If no data (only SYN) was ACK'd, 1789 * skip rest of ACK processing. 1790 */ 1791 if (acked == 0) 1792 goto step6; 1793 1794 /* 1795 * When new data is acked, open the congestion window. 1796 * If the window gives us less than ssthresh packets 1797 * in flight, open exponentially (maxseg per packet). 1798 * Otherwise open linearly: maxseg per window 1799 * (maxseg^2 / cwnd per packet). 1800 */ 1801 { 1802 register u_int cw = tp->snd_cwnd; 1803 register u_int incr = tp->t_maxseg; 1804 1805 if (cw > tp->snd_ssthresh) 1806 incr = incr * incr / cw; 1807 /* 1808 * If t_dupacks != 0 here, it indicates that we are still 1809 * in NewReno fast recovery mode, so we leave the congestion 1810 * window alone. 1811 */ 1812 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1813 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1814 } 1815 if (acked > so->so_snd.sb_cc) { 1816 tp->snd_wnd -= so->so_snd.sb_cc; 1817 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1818 ourfinisacked = 1; 1819 } else { 1820 sbdrop(&so->so_snd, acked); 1821 tp->snd_wnd -= acked; 1822 ourfinisacked = 0; 1823 } 1824 sowwakeup(so); 1825 tp->snd_una = th->th_ack; 1826 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1827 tp->snd_nxt = tp->snd_una; 1828 1829 switch (tp->t_state) { 1830 1831 /* 1832 * In FIN_WAIT_1 STATE in addition to the processing 1833 * for the ESTABLISHED state if our FIN is now acknowledged 1834 * then enter FIN_WAIT_2. 1835 */ 1836 case TCPS_FIN_WAIT_1: 1837 if (ourfinisacked) { 1838 /* 1839 * If we can't receive any more 1840 * data, then closing user can proceed. 1841 * Starting the timer is contrary to the 1842 * specification, but if we don't get a FIN 1843 * we'll hang forever. 1844 */
|
1849 } 1850 tp->t_state = TCPS_FIN_WAIT_2; 1851 } 1852 break; 1853 1854 /* 1855 * In CLOSING STATE in addition to the processing for 1856 * the ESTABLISHED state if the ACK acknowledges our FIN 1857 * then enter the TIME-WAIT state, otherwise ignore 1858 * the segment. 1859 */ 1860 case TCPS_CLOSING: 1861 if (ourfinisacked) { 1862 tp->t_state = TCPS_TIME_WAIT; 1863 tcp_canceltimers(tp); 1864 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1865 if (tp->cc_recv != 0 && 1866 (ticks - tp->t_starttime) < tcp_msl) 1867 callout_reset(tp->tt_2msl, 1868 tp->t_rxtcur * 1869 TCPTV_TWTRUNC, 1870 tcp_timer_2msl, tp); 1871 else 1872 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1873 tcp_timer_2msl, tp); 1874 soisdisconnected(so); 1875 } 1876 break; 1877 1878 /* 1879 * In LAST_ACK, we may still be waiting for data to drain 1880 * and/or to be acked, as well as for the ack of our FIN. 1881 * If our FIN is now acknowledged, delete the TCB, 1882 * enter the closed state and return. 1883 */ 1884 case TCPS_LAST_ACK: 1885 if (ourfinisacked) { 1886 tp = tcp_close(tp); 1887 goto drop; 1888 } 1889 break; 1890 1891 /* 1892 * In TIME_WAIT state the only thing that should arrive 1893 * is a retransmission of the remote FIN. Acknowledge 1894 * it and restart the finack timer. 1895 */ 1896 case TCPS_TIME_WAIT: 1897 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1898 tcp_timer_2msl, tp); 1899 goto dropafterack; 1900 } 1901 } 1902 1903step6: 1904 /* 1905 * Update window information. 1906 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1907 */ 1908 if ((thflags & TH_ACK) && 1909 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1910 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1911 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1912 /* keep track of pure window updates */ 1913 if (tlen == 0 && 1914 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1915 tcpstat.tcps_rcvwinupd++; 1916 tp->snd_wnd = tiwin; 1917 tp->snd_wl1 = th->th_seq; 1918 tp->snd_wl2 = th->th_ack; 1919 if (tp->snd_wnd > tp->max_sndwnd) 1920 tp->max_sndwnd = tp->snd_wnd; 1921 needoutput = 1; 1922 } 1923 1924 /* 1925 * Process segments with URG. 1926 */ 1927 if ((thflags & TH_URG) && th->th_urp && 1928 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1929 /* 1930 * This is a kludge, but if we receive and accept 1931 * random urgent pointers, we'll crash in 1932 * soreceive. It's hard to imagine someone 1933 * actually wanting to send this much urgent data. 1934 */ 1935 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1936 th->th_urp = 0; /* XXX */ 1937 thflags &= ~TH_URG; /* XXX */ 1938 goto dodata; /* XXX */ 1939 } 1940 /* 1941 * If this segment advances the known urgent pointer, 1942 * then mark the data stream. This should not happen 1943 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1944 * a FIN has been received from the remote side. 1945 * In these states we ignore the URG. 1946 * 1947 * According to RFC961 (Assigned Protocols), 1948 * the urgent pointer points to the last octet 1949 * of urgent data. We continue, however, 1950 * to consider it to indicate the first octet 1951 * of data past the urgent section as the original 1952 * spec states (in one of two places). 1953 */ 1954 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1955 tp->rcv_up = th->th_seq + th->th_urp; 1956 so->so_oobmark = so->so_rcv.sb_cc + 1957 (tp->rcv_up - tp->rcv_nxt) - 1; 1958 if (so->so_oobmark == 0) 1959 so->so_state |= SS_RCVATMARK; 1960 sohasoutofband(so); 1961 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1962 } 1963 /* 1964 * Remove out of band data so doesn't get presented to user. 1965 * This can happen independent of advancing the URG pointer, 1966 * but if two URG's are pending at once, some out-of-band 1967 * data may creep in... ick. 1968 */ 1969 if (th->th_urp <= (u_long)tlen 1970#ifdef SO_OOBINLINE 1971 && (so->so_options & SO_OOBINLINE) == 0 1972#endif 1973 ) 1974 tcp_pulloutofband(so, th, m, 1975 drop_hdrlen); /* hdr drop is delayed */ 1976 } else 1977 /* 1978 * If no out of band data is expected, 1979 * pull receive urgent pointer along 1980 * with the receive window. 1981 */ 1982 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1983 tp->rcv_up = tp->rcv_nxt; 1984dodata: /* XXX */ 1985 1986 /* 1987 * Process the segment text, merging it into the TCP sequencing queue, 1988 * and arranging for acknowledgment of receipt if necessary. 1989 * This process logically involves adjusting tp->rcv_wnd as data 1990 * is presented to the user (this happens in tcp_usrreq.c, 1991 * case PRU_RCVD). If a FIN has already been received on this 1992 * connection then we just ignore the text. 1993 */ 1994 if ((tlen || (thflags&TH_FIN)) && 1995 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1996 m_adj(m, drop_hdrlen); /* delayed header drop */ 1997 /* 1998 * Insert segment which inludes th into reassembly queue of tcp with 1999 * control block tp. Return TH_FIN if reassembly now includes 2000 * a segment with FIN. This handle the common case inline (segment 2001 * is the next to be received on an established connection, and the 2002 * queue is empty), avoiding linkage into and removal from the queue 2003 * and repetition of various conversions. 2004 * Set DELACK for segments received in order, but ack immediately 2005 * when segments are out of order (so fast retransmit can work). 2006 */ 2007 if (th->th_seq == tp->rcv_nxt && 2008 LIST_EMPTY(&tp->t_segq) && 2009 TCPS_HAVEESTABLISHED(tp->t_state)) { 2010 if (DELAY_ACK(tp)) 2011 callout_reset(tp->tt_delack, tcp_delacktime, 2012 tcp_timer_delack, tp); 2013 else 2014 tp->t_flags |= TF_ACKNOW; 2015 tp->rcv_nxt += tlen; 2016 thflags = th->th_flags & TH_FIN; 2017 tcpstat.tcps_rcvpack++; 2018 tcpstat.tcps_rcvbyte += tlen; 2019 ND6_HINT(tp); 2020 sbappend(&so->so_rcv, m); 2021 sorwakeup(so); 2022 } else { 2023 thflags = tcp_reass(tp, th, &tlen, m); 2024 tp->t_flags |= TF_ACKNOW; 2025 } 2026 2027 /* 2028 * Note the amount of data that peer has sent into 2029 * our window, in order to estimate the sender's 2030 * buffer size. 2031 */ 2032 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2033 } else { 2034 m_freem(m); 2035 thflags &= ~TH_FIN; 2036 } 2037 2038 /* 2039 * If FIN is received ACK the FIN and let the user know 2040 * that the connection is closing. 2041 */ 2042 if (thflags & TH_FIN) { 2043 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2044 socantrcvmore(so); 2045 /* 2046 * If connection is half-synchronized 2047 * (ie NEEDSYN flag on) then delay ACK, 2048 * so it may be piggybacked when SYN is sent. 2049 * Otherwise, since we received a FIN then no 2050 * more input can be expected, send ACK now. 2051 */ 2052 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2053 callout_reset(tp->tt_delack, tcp_delacktime, 2054 tcp_timer_delack, tp); 2055 else 2056 tp->t_flags |= TF_ACKNOW; 2057 tp->rcv_nxt++; 2058 } 2059 switch (tp->t_state) { 2060 2061 /* 2062 * In SYN_RECEIVED and ESTABLISHED STATES 2063 * enter the CLOSE_WAIT state. 2064 */ 2065 case TCPS_SYN_RECEIVED: 2066 tp->t_starttime = ticks; 2067 /*FALLTHROUGH*/ 2068 case TCPS_ESTABLISHED: 2069 tp->t_state = TCPS_CLOSE_WAIT; 2070 break; 2071 2072 /* 2073 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2074 * enter the CLOSING state. 2075 */ 2076 case TCPS_FIN_WAIT_1: 2077 tp->t_state = TCPS_CLOSING; 2078 break; 2079 2080 /* 2081 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2082 * starting the time-wait timer, turning off the other 2083 * standard timers. 2084 */ 2085 case TCPS_FIN_WAIT_2: 2086 tp->t_state = TCPS_TIME_WAIT; 2087 tcp_canceltimers(tp); 2088 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2089 if (tp->cc_recv != 0 && 2090 (ticks - tp->t_starttime) < tcp_msl) { 2091 callout_reset(tp->tt_2msl, 2092 tp->t_rxtcur * TCPTV_TWTRUNC, 2093 tcp_timer_2msl, tp); 2094 /* For transaction client, force ACK now. */ 2095 tp->t_flags |= TF_ACKNOW; 2096 } 2097 else 2098 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2099 tcp_timer_2msl, tp); 2100 soisdisconnected(so); 2101 break; 2102 2103 /* 2104 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2105 */ 2106 case TCPS_TIME_WAIT: 2107 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2108 tcp_timer_2msl, tp); 2109 break; 2110 } 2111 } 2112#ifdef TCPDEBUG 2113 if (so->so_options & SO_DEBUG) 2114 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2115 &tcp_savetcp, 0); 2116#endif 2117 2118 /* 2119 * Return any desired output. 2120 */ 2121 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2122 (void) tcp_output(tp); 2123 return; 2124 2125dropafterack: 2126 /* 2127 * Generate an ACK dropping incoming segment if it occupies 2128 * sequence space, where the ACK reflects our state. 2129 * 2130 * We can now skip the test for the RST flag since all 2131 * paths to this code happen after packets containing 2132 * RST have been dropped. 2133 * 2134 * In the SYN-RECEIVED state, don't send an ACK unless the 2135 * segment we received passes the SYN-RECEIVED ACK test. 2136 * If it fails send a RST. This breaks the loop in the 2137 * "LAND" DoS attack, and also prevents an ACK storm 2138 * between two listening ports that have been sent forged 2139 * SYN segments, each with the source address of the other. 2140 */ 2141 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2142 (SEQ_GT(tp->snd_una, th->th_ack) || 2143 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2144 rstreason = BANDLIM_RST_OPENPORT; 2145 goto dropwithreset; 2146 } 2147#ifdef TCPDEBUG 2148 if (so->so_options & SO_DEBUG) 2149 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2150 &tcp_savetcp, 0); 2151#endif 2152 m_freem(m); 2153 tp->t_flags |= TF_ACKNOW; 2154 (void) tcp_output(tp); 2155 return; 2156 2157dropwithreset: 2158 /* 2159 * Generate a RST, dropping incoming segment. 2160 * Make ACK acceptable to originator of segment. 2161 * Don't bother to respond if destination was broadcast/multicast. 2162 */ 2163 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2164 goto drop; 2165#ifdef INET6 2166 if (isipv6) { 2167 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2168 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2169 goto drop; 2170 } else 2171#endif /* INET6 */ 2172 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2173 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2174 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2175 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2176 goto drop; 2177 /* IPv6 anycast check is done at tcp6_input() */ 2178 2179 /* 2180 * Perform bandwidth limiting. 2181 */ 2182 if (badport_bandlim(rstreason) < 0) 2183 goto drop; 2184 2185#ifdef TCPDEBUG 2186 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2187 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2188 &tcp_savetcp, 0); 2189#endif 2190 if (thflags & TH_ACK) 2191 /* mtod() below is safe as long as hdr dropping is delayed */ 2192 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2193 TH_RST); 2194 else { 2195 if (thflags & TH_SYN) 2196 tlen++; 2197 /* mtod() below is safe as long as hdr dropping is delayed */ 2198 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2199 (tcp_seq)0, TH_RST|TH_ACK); 2200 } 2201 return; 2202 2203drop: 2204 /* 2205 * Drop space held by incoming segment and return. 2206 */ 2207#ifdef TCPDEBUG 2208 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2209 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2210 &tcp_savetcp, 0); 2211#endif 2212 m_freem(m); 2213 return; 2214} 2215 2216/* 2217 * Parse TCP options and place in tcpopt. 2218 */ 2219static void 2220tcp_dooptions(to, cp, cnt, is_syn) 2221 struct tcpopt *to; 2222 u_char *cp; 2223 int cnt; 2224{ 2225 int opt, optlen; 2226 2227 to->to_flags = 0; 2228 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2229 opt = cp[0]; 2230 if (opt == TCPOPT_EOL) 2231 break; 2232 if (opt == TCPOPT_NOP) 2233 optlen = 1; 2234 else { 2235 if (cnt < 2) 2236 break; 2237 optlen = cp[1]; 2238 if (optlen < 2 || optlen > cnt) 2239 break; 2240 } 2241 switch (opt) { 2242 case TCPOPT_MAXSEG: 2243 if (optlen != TCPOLEN_MAXSEG) 2244 continue; 2245 if (!is_syn) 2246 continue; 2247 to->to_flags |= TOF_MSS; 2248 bcopy((char *)cp + 2, 2249 (char *)&to->to_mss, sizeof(to->to_mss)); 2250 to->to_mss = ntohs(to->to_mss); 2251 break; 2252 case TCPOPT_WINDOW: 2253 if (optlen != TCPOLEN_WINDOW) 2254 continue; 2255 if (! is_syn) 2256 continue; 2257 to->to_flags |= TOF_SCALE; 2258 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2259 break; 2260 case TCPOPT_TIMESTAMP: 2261 if (optlen != TCPOLEN_TIMESTAMP) 2262 continue; 2263 to->to_flags |= TOF_TS; 2264 bcopy((char *)cp + 2, 2265 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2266 to->to_tsval = ntohl(to->to_tsval); 2267 bcopy((char *)cp + 6, 2268 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2269 to->to_tsecr = ntohl(to->to_tsecr); 2270 break; 2271 case TCPOPT_CC: 2272 if (optlen != TCPOLEN_CC) 2273 continue; 2274 to->to_flags |= TOF_CC; 2275 bcopy((char *)cp + 2, 2276 (char *)&to->to_cc, sizeof(to->to_cc)); 2277 to->to_cc = ntohl(to->to_cc); 2278 break; 2279 case TCPOPT_CCNEW: 2280 if (optlen != TCPOLEN_CC) 2281 continue; 2282 if (!is_syn) 2283 continue; 2284 to->to_flags |= TOF_CCNEW; 2285 bcopy((char *)cp + 2, 2286 (char *)&to->to_cc, sizeof(to->to_cc)); 2287 to->to_cc = ntohl(to->to_cc); 2288 break; 2289 case TCPOPT_CCECHO: 2290 if (optlen != TCPOLEN_CC) 2291 continue; 2292 if (!is_syn) 2293 continue; 2294 to->to_flags |= TOF_CCECHO; 2295 bcopy((char *)cp + 2, 2296 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2297 to->to_ccecho = ntohl(to->to_ccecho); 2298 break; 2299 default: 2300 continue; 2301 } 2302 } 2303} 2304 2305/* 2306 * Pull out of band byte out of a segment so 2307 * it doesn't appear in the user's data queue. 2308 * It is still reflected in the segment length for 2309 * sequencing purposes. 2310 */ 2311static void 2312tcp_pulloutofband(so, th, m, off) 2313 struct socket *so; 2314 struct tcphdr *th; 2315 register struct mbuf *m; 2316 int off; /* delayed to be droped hdrlen */ 2317{ 2318 int cnt = off + th->th_urp - 1; 2319 2320 while (cnt >= 0) { 2321 if (m->m_len > cnt) { 2322 char *cp = mtod(m, caddr_t) + cnt; 2323 struct tcpcb *tp = sototcpcb(so); 2324 2325 tp->t_iobc = *cp; 2326 tp->t_oobflags |= TCPOOB_HAVEDATA; 2327 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2328 m->m_len--; 2329 if (m->m_flags & M_PKTHDR) 2330 m->m_pkthdr.len--; 2331 return; 2332 } 2333 cnt -= m->m_len; 2334 m = m->m_next; 2335 if (m == 0) 2336 break; 2337 } 2338 panic("tcp_pulloutofband"); 2339} 2340 2341/* 2342 * Collect new round-trip time estimate 2343 * and update averages and current timeout. 2344 */ 2345static void 2346tcp_xmit_timer(tp, rtt) 2347 register struct tcpcb *tp; 2348 int rtt; 2349{ 2350 register int delta; 2351 2352 tcpstat.tcps_rttupdated++; 2353 tp->t_rttupdated++; 2354 if (tp->t_srtt != 0) { 2355 /* 2356 * srtt is stored as fixed point with 5 bits after the 2357 * binary point (i.e., scaled by 8). The following magic 2358 * is equivalent to the smoothing algorithm in rfc793 with 2359 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2360 * point). Adjust rtt to origin 0. 2361 */ 2362 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2363 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2364 2365 if ((tp->t_srtt += delta) <= 0) 2366 tp->t_srtt = 1; 2367 2368 /* 2369 * We accumulate a smoothed rtt variance (actually, a 2370 * smoothed mean difference), then set the retransmit 2371 * timer to smoothed rtt + 4 times the smoothed variance. 2372 * rttvar is stored as fixed point with 4 bits after the 2373 * binary point (scaled by 16). The following is 2374 * equivalent to rfc793 smoothing with an alpha of .75 2375 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2376 * rfc793's wired-in beta. 2377 */ 2378 if (delta < 0) 2379 delta = -delta; 2380 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2381 if ((tp->t_rttvar += delta) <= 0) 2382 tp->t_rttvar = 1; 2383 } else { 2384 /* 2385 * No rtt measurement yet - use the unsmoothed rtt. 2386 * Set the variance to half the rtt (so our first 2387 * retransmit happens at 3*rtt). 2388 */ 2389 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2390 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2391 } 2392 tp->t_rtttime = 0; 2393 tp->t_rxtshift = 0; 2394 2395 /* 2396 * the retransmit should happen at rtt + 4 * rttvar. 2397 * Because of the way we do the smoothing, srtt and rttvar 2398 * will each average +1/2 tick of bias. When we compute 2399 * the retransmit timer, we want 1/2 tick of rounding and 2400 * 1 extra tick because of +-1/2 tick uncertainty in the 2401 * firing of the timer. The bias will give us exactly the 2402 * 1.5 tick we need. But, because the bias is 2403 * statistical, we have to test that we don't drop below 2404 * the minimum feasible timer (which is 2 ticks). 2405 */ 2406 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2407 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2408 2409 /* 2410 * We received an ack for a packet that wasn't retransmitted; 2411 * it is probably safe to discard any error indications we've 2412 * received recently. This isn't quite right, but close enough 2413 * for now (a route might have failed after we sent a segment, 2414 * and the return path might not be symmetrical). 2415 */ 2416 tp->t_softerror = 0; 2417} 2418 2419/* 2420 * Determine a reasonable value for maxseg size. 2421 * If the route is known, check route for mtu. 2422 * If none, use an mss that can be handled on the outgoing 2423 * interface without forcing IP to fragment; if bigger than 2424 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2425 * to utilize large mbufs. If no route is found, route has no mtu, 2426 * or the destination isn't local, use a default, hopefully conservative 2427 * size (usually 512 or the default IP max size, but no more than the mtu 2428 * of the interface), as we can't discover anything about intervening 2429 * gateways or networks. We also initialize the congestion/slow start 2430 * window to be a single segment if the destination isn't local. 2431 * While looking at the routing entry, we also initialize other path-dependent 2432 * parameters from pre-set or cached values in the routing entry. 2433 * 2434 * Also take into account the space needed for options that we 2435 * send regularly. Make maxseg shorter by that amount to assure 2436 * that we can send maxseg amount of data even when the options 2437 * are present. Store the upper limit of the length of options plus 2438 * data in maxopd. 2439 * 2440 * NOTE that this routine is only called when we process an incoming 2441 * segment, for outgoing segments only tcp_mssopt is called. 2442 * 2443 * In case of T/TCP, we call this routine during implicit connection 2444 * setup as well (offer = -1), to initialize maxseg from the cached 2445 * MSS of our peer. 2446 */ 2447void 2448tcp_mss(tp, offer) 2449 struct tcpcb *tp; 2450 int offer; 2451{ 2452 register struct rtentry *rt; 2453 struct ifnet *ifp; 2454 register int rtt, mss; 2455 u_long bufsize; 2456 struct inpcb *inp; 2457 struct socket *so; 2458 struct rmxp_tao *taop; 2459 int origoffer = offer; 2460#ifdef INET6 2461 int isipv6; 2462 int min_protoh; 2463#endif 2464 2465 inp = tp->t_inpcb; 2466#ifdef INET6 2467 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2468 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2469 : sizeof (struct tcpiphdr); 2470#else 2471#define min_protoh (sizeof (struct tcpiphdr)) 2472#endif 2473#ifdef INET6 2474 if (isipv6) 2475 rt = tcp_rtlookup6(&inp->inp_inc); 2476 else 2477#endif 2478 rt = tcp_rtlookup(&inp->inp_inc); 2479 if (rt == NULL) { 2480 tp->t_maxopd = tp->t_maxseg = 2481#ifdef INET6 2482 isipv6 ? tcp_v6mssdflt : 2483#endif /* INET6 */ 2484 tcp_mssdflt; 2485 return; 2486 } 2487 ifp = rt->rt_ifp; 2488 so = inp->inp_socket; 2489 2490 taop = rmx_taop(rt->rt_rmx); 2491 /* 2492 * Offer == -1 means that we didn't receive SYN yet, 2493 * use cached value in that case; 2494 */ 2495 if (offer == -1) 2496 offer = taop->tao_mssopt; 2497 /* 2498 * Offer == 0 means that there was no MSS on the SYN segment, 2499 * in this case we use tcp_mssdflt. 2500 */ 2501 if (offer == 0) 2502 offer = 2503#ifdef INET6 2504 isipv6 ? tcp_v6mssdflt : 2505#endif /* INET6 */ 2506 tcp_mssdflt; 2507 else 2508 /* 2509 * Sanity check: make sure that maxopd will be large 2510 * enough to allow some data on segments even is the 2511 * all the option space is used (40bytes). Otherwise 2512 * funny things may happen in tcp_output. 2513 */ 2514 offer = max(offer, 64); 2515 taop->tao_mssopt = offer; 2516 2517 /* 2518 * While we're here, check if there's an initial rtt 2519 * or rttvar. Convert from the route-table units 2520 * to scaled multiples of the slow timeout timer. 2521 */ 2522 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2523 /* 2524 * XXX the lock bit for RTT indicates that the value 2525 * is also a minimum value; this is subject to time. 2526 */ 2527 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2528 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2529 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2530 tcpstat.tcps_usedrtt++; 2531 if (rt->rt_rmx.rmx_rttvar) { 2532 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2533 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2534 tcpstat.tcps_usedrttvar++; 2535 } else { 2536 /* default variation is +- 1 rtt */ 2537 tp->t_rttvar = 2538 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2539 } 2540 TCPT_RANGESET(tp->t_rxtcur, 2541 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2542 tp->t_rttmin, TCPTV_REXMTMAX); 2543 } 2544 /* 2545 * if there's an mtu associated with the route, use it 2546 * else, use the link mtu. 2547 */ 2548 if (rt->rt_rmx.rmx_mtu) 2549 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2550 else 2551 { 2552 mss = 2553#ifdef INET6 2554 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2555#endif 2556 ifp->if_mtu 2557#ifdef INET6 2558 ) 2559#endif 2560 - min_protoh; 2561#ifdef INET6 2562 if (isipv6) { 2563 if (!in6_localaddr(&inp->in6p_faddr)) 2564 mss = min(mss, tcp_v6mssdflt); 2565 } else 2566#endif 2567 if (!in_localaddr(inp->inp_faddr)) 2568 mss = min(mss, tcp_mssdflt); 2569 } 2570 mss = min(mss, offer); 2571 /* 2572 * maxopd stores the maximum length of data AND options 2573 * in a segment; maxseg is the amount of data in a normal 2574 * segment. We need to store this value (maxopd) apart 2575 * from maxseg, because now every segment carries options 2576 * and thus we normally have somewhat less data in segments. 2577 */ 2578 tp->t_maxopd = mss; 2579 2580 /* 2581 * In case of T/TCP, origoffer==-1 indicates, that no segments 2582 * were received yet. In this case we just guess, otherwise 2583 * we do the same as before T/TCP. 2584 */ 2585 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2586 (origoffer == -1 || 2587 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2588 mss -= TCPOLEN_TSTAMP_APPA; 2589 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2590 (origoffer == -1 || 2591 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2592 mss -= TCPOLEN_CC_APPA; 2593 2594#if (MCLBYTES & (MCLBYTES - 1)) == 0 2595 if (mss > MCLBYTES) 2596 mss &= ~(MCLBYTES-1); 2597#else 2598 if (mss > MCLBYTES) 2599 mss = mss / MCLBYTES * MCLBYTES; 2600#endif 2601 /* 2602 * If there's a pipesize, change the socket buffer 2603 * to that size. Make the socket buffers an integral 2604 * number of mss units; if the mss is larger than 2605 * the socket buffer, decrease the mss. 2606 */ 2607#ifdef RTV_SPIPE 2608 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2609#endif 2610 bufsize = so->so_snd.sb_hiwat; 2611 if (bufsize < mss) 2612 mss = bufsize; 2613 else { 2614 bufsize = roundup(bufsize, mss); 2615 if (bufsize > sb_max) 2616 bufsize = sb_max; 2617 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2618 } 2619 tp->t_maxseg = mss; 2620 2621#ifdef RTV_RPIPE 2622 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2623#endif 2624 bufsize = so->so_rcv.sb_hiwat; 2625 if (bufsize > mss) { 2626 bufsize = roundup(bufsize, mss); 2627 if (bufsize > sb_max) 2628 bufsize = sb_max; 2629 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2630 } 2631 2632 /* 2633 * Set the slow-start flight size depending on whether this 2634 * is a local network or not. 2635 */ 2636 if ( 2637#ifdef INET6 2638 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2639 (!isipv6 && 2640#endif 2641 in_localaddr(inp->inp_faddr) 2642#ifdef INET6 2643 ) 2644#endif 2645 ) 2646 tp->snd_cwnd = mss * ss_fltsz_local; 2647 else 2648 tp->snd_cwnd = mss * ss_fltsz; 2649 2650 if (rt->rt_rmx.rmx_ssthresh) { 2651 /* 2652 * There's some sort of gateway or interface 2653 * buffer limit on the path. Use this to set 2654 * the slow start threshhold, but set the 2655 * threshold to no less than 2*mss. 2656 */ 2657 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2658 tcpstat.tcps_usedssthresh++; 2659 } 2660} 2661 2662/* 2663 * Determine the MSS option to send on an outgoing SYN. 2664 */ 2665int 2666tcp_mssopt(tp) 2667 struct tcpcb *tp; 2668{ 2669 struct rtentry *rt; 2670#ifdef INET6 2671 int isipv6; 2672 int min_protoh; 2673#endif 2674 2675#ifdef INET6 2676 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2677 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2678 : sizeof (struct tcpiphdr); 2679#else 2680#define min_protoh (sizeof (struct tcpiphdr)) 2681#endif 2682#ifdef INET6 2683 if (isipv6) 2684 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2685 else 2686#endif /* INET6 */ 2687 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2688 if (rt == NULL) 2689 return 2690#ifdef INET6 2691 isipv6 ? tcp_v6mssdflt : 2692#endif /* INET6 */ 2693 tcp_mssdflt; 2694 2695 return rt->rt_ifp->if_mtu - min_protoh; 2696} 2697 2698 2699/* 2700 * Checks for partial ack. If partial ack arrives, force the retransmission 2701 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2702 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2703 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2704 */ 2705static int 2706tcp_newreno(tp, th) 2707 struct tcpcb *tp; 2708 struct tcphdr *th; 2709{ 2710 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2711 tcp_seq onxt = tp->snd_nxt; 2712 u_long ocwnd = tp->snd_cwnd; 2713 2714 callout_stop(tp->tt_rexmt); 2715 tp->t_rtttime = 0; 2716 tp->snd_nxt = th->th_ack; 2717 /* 2718 * Set snd_cwnd to one segment beyond acknowledged offset 2719 * (tp->snd_una has not yet been updated when this function 2720 * is called) 2721 */ 2722 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2723 (void) tcp_output(tp); 2724 tp->snd_cwnd = ocwnd; 2725 if (SEQ_GT(onxt, tp->snd_nxt)) 2726 tp->snd_nxt = onxt; 2727 /* 2728 * Partial window deflation. Relies on fact that tp->snd_una 2729 * not updated yet. 2730 */ 2731 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2732 return (1); 2733 } 2734 return (0); 2735}
| 1853 } 1854 tp->t_state = TCPS_FIN_WAIT_2; 1855 } 1856 break; 1857 1858 /* 1859 * In CLOSING STATE in addition to the processing for 1860 * the ESTABLISHED state if the ACK acknowledges our FIN 1861 * then enter the TIME-WAIT state, otherwise ignore 1862 * the segment. 1863 */ 1864 case TCPS_CLOSING: 1865 if (ourfinisacked) { 1866 tp->t_state = TCPS_TIME_WAIT; 1867 tcp_canceltimers(tp); 1868 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1869 if (tp->cc_recv != 0 && 1870 (ticks - tp->t_starttime) < tcp_msl) 1871 callout_reset(tp->tt_2msl, 1872 tp->t_rxtcur * 1873 TCPTV_TWTRUNC, 1874 tcp_timer_2msl, tp); 1875 else 1876 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1877 tcp_timer_2msl, tp); 1878 soisdisconnected(so); 1879 } 1880 break; 1881 1882 /* 1883 * In LAST_ACK, we may still be waiting for data to drain 1884 * and/or to be acked, as well as for the ack of our FIN. 1885 * If our FIN is now acknowledged, delete the TCB, 1886 * enter the closed state and return. 1887 */ 1888 case TCPS_LAST_ACK: 1889 if (ourfinisacked) { 1890 tp = tcp_close(tp); 1891 goto drop; 1892 } 1893 break; 1894 1895 /* 1896 * In TIME_WAIT state the only thing that should arrive 1897 * is a retransmission of the remote FIN. Acknowledge 1898 * it and restart the finack timer. 1899 */ 1900 case TCPS_TIME_WAIT: 1901 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1902 tcp_timer_2msl, tp); 1903 goto dropafterack; 1904 } 1905 } 1906 1907step6: 1908 /* 1909 * Update window information. 1910 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1911 */ 1912 if ((thflags & TH_ACK) && 1913 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1914 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1915 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1916 /* keep track of pure window updates */ 1917 if (tlen == 0 && 1918 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1919 tcpstat.tcps_rcvwinupd++; 1920 tp->snd_wnd = tiwin; 1921 tp->snd_wl1 = th->th_seq; 1922 tp->snd_wl2 = th->th_ack; 1923 if (tp->snd_wnd > tp->max_sndwnd) 1924 tp->max_sndwnd = tp->snd_wnd; 1925 needoutput = 1; 1926 } 1927 1928 /* 1929 * Process segments with URG. 1930 */ 1931 if ((thflags & TH_URG) && th->th_urp && 1932 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1933 /* 1934 * This is a kludge, but if we receive and accept 1935 * random urgent pointers, we'll crash in 1936 * soreceive. It's hard to imagine someone 1937 * actually wanting to send this much urgent data. 1938 */ 1939 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1940 th->th_urp = 0; /* XXX */ 1941 thflags &= ~TH_URG; /* XXX */ 1942 goto dodata; /* XXX */ 1943 } 1944 /* 1945 * If this segment advances the known urgent pointer, 1946 * then mark the data stream. This should not happen 1947 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1948 * a FIN has been received from the remote side. 1949 * In these states we ignore the URG. 1950 * 1951 * According to RFC961 (Assigned Protocols), 1952 * the urgent pointer points to the last octet 1953 * of urgent data. We continue, however, 1954 * to consider it to indicate the first octet 1955 * of data past the urgent section as the original 1956 * spec states (in one of two places). 1957 */ 1958 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1959 tp->rcv_up = th->th_seq + th->th_urp; 1960 so->so_oobmark = so->so_rcv.sb_cc + 1961 (tp->rcv_up - tp->rcv_nxt) - 1; 1962 if (so->so_oobmark == 0) 1963 so->so_state |= SS_RCVATMARK; 1964 sohasoutofband(so); 1965 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1966 } 1967 /* 1968 * Remove out of band data so doesn't get presented to user. 1969 * This can happen independent of advancing the URG pointer, 1970 * but if two URG's are pending at once, some out-of-band 1971 * data may creep in... ick. 1972 */ 1973 if (th->th_urp <= (u_long)tlen 1974#ifdef SO_OOBINLINE 1975 && (so->so_options & SO_OOBINLINE) == 0 1976#endif 1977 ) 1978 tcp_pulloutofband(so, th, m, 1979 drop_hdrlen); /* hdr drop is delayed */ 1980 } else 1981 /* 1982 * If no out of band data is expected, 1983 * pull receive urgent pointer along 1984 * with the receive window. 1985 */ 1986 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1987 tp->rcv_up = tp->rcv_nxt; 1988dodata: /* XXX */ 1989 1990 /* 1991 * Process the segment text, merging it into the TCP sequencing queue, 1992 * and arranging for acknowledgment of receipt if necessary. 1993 * This process logically involves adjusting tp->rcv_wnd as data 1994 * is presented to the user (this happens in tcp_usrreq.c, 1995 * case PRU_RCVD). If a FIN has already been received on this 1996 * connection then we just ignore the text. 1997 */ 1998 if ((tlen || (thflags&TH_FIN)) && 1999 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2000 m_adj(m, drop_hdrlen); /* delayed header drop */ 2001 /* 2002 * Insert segment which inludes th into reassembly queue of tcp with 2003 * control block tp. Return TH_FIN if reassembly now includes 2004 * a segment with FIN. This handle the common case inline (segment 2005 * is the next to be received on an established connection, and the 2006 * queue is empty), avoiding linkage into and removal from the queue 2007 * and repetition of various conversions. 2008 * Set DELACK for segments received in order, but ack immediately 2009 * when segments are out of order (so fast retransmit can work). 2010 */ 2011 if (th->th_seq == tp->rcv_nxt && 2012 LIST_EMPTY(&tp->t_segq) && 2013 TCPS_HAVEESTABLISHED(tp->t_state)) { 2014 if (DELAY_ACK(tp)) 2015 callout_reset(tp->tt_delack, tcp_delacktime, 2016 tcp_timer_delack, tp); 2017 else 2018 tp->t_flags |= TF_ACKNOW; 2019 tp->rcv_nxt += tlen; 2020 thflags = th->th_flags & TH_FIN; 2021 tcpstat.tcps_rcvpack++; 2022 tcpstat.tcps_rcvbyte += tlen; 2023 ND6_HINT(tp); 2024 sbappend(&so->so_rcv, m); 2025 sorwakeup(so); 2026 } else { 2027 thflags = tcp_reass(tp, th, &tlen, m); 2028 tp->t_flags |= TF_ACKNOW; 2029 } 2030 2031 /* 2032 * Note the amount of data that peer has sent into 2033 * our window, in order to estimate the sender's 2034 * buffer size. 2035 */ 2036 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2037 } else { 2038 m_freem(m); 2039 thflags &= ~TH_FIN; 2040 } 2041 2042 /* 2043 * If FIN is received ACK the FIN and let the user know 2044 * that the connection is closing. 2045 */ 2046 if (thflags & TH_FIN) { 2047 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2048 socantrcvmore(so); 2049 /* 2050 * If connection is half-synchronized 2051 * (ie NEEDSYN flag on) then delay ACK, 2052 * so it may be piggybacked when SYN is sent. 2053 * Otherwise, since we received a FIN then no 2054 * more input can be expected, send ACK now. 2055 */ 2056 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2057 callout_reset(tp->tt_delack, tcp_delacktime, 2058 tcp_timer_delack, tp); 2059 else 2060 tp->t_flags |= TF_ACKNOW; 2061 tp->rcv_nxt++; 2062 } 2063 switch (tp->t_state) { 2064 2065 /* 2066 * In SYN_RECEIVED and ESTABLISHED STATES 2067 * enter the CLOSE_WAIT state. 2068 */ 2069 case TCPS_SYN_RECEIVED: 2070 tp->t_starttime = ticks; 2071 /*FALLTHROUGH*/ 2072 case TCPS_ESTABLISHED: 2073 tp->t_state = TCPS_CLOSE_WAIT; 2074 break; 2075 2076 /* 2077 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2078 * enter the CLOSING state. 2079 */ 2080 case TCPS_FIN_WAIT_1: 2081 tp->t_state = TCPS_CLOSING; 2082 break; 2083 2084 /* 2085 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2086 * starting the time-wait timer, turning off the other 2087 * standard timers. 2088 */ 2089 case TCPS_FIN_WAIT_2: 2090 tp->t_state = TCPS_TIME_WAIT; 2091 tcp_canceltimers(tp); 2092 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2093 if (tp->cc_recv != 0 && 2094 (ticks - tp->t_starttime) < tcp_msl) { 2095 callout_reset(tp->tt_2msl, 2096 tp->t_rxtcur * TCPTV_TWTRUNC, 2097 tcp_timer_2msl, tp); 2098 /* For transaction client, force ACK now. */ 2099 tp->t_flags |= TF_ACKNOW; 2100 } 2101 else 2102 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2103 tcp_timer_2msl, tp); 2104 soisdisconnected(so); 2105 break; 2106 2107 /* 2108 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2109 */ 2110 case TCPS_TIME_WAIT: 2111 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2112 tcp_timer_2msl, tp); 2113 break; 2114 } 2115 } 2116#ifdef TCPDEBUG 2117 if (so->so_options & SO_DEBUG) 2118 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2119 &tcp_savetcp, 0); 2120#endif 2121 2122 /* 2123 * Return any desired output. 2124 */ 2125 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2126 (void) tcp_output(tp); 2127 return; 2128 2129dropafterack: 2130 /* 2131 * Generate an ACK dropping incoming segment if it occupies 2132 * sequence space, where the ACK reflects our state. 2133 * 2134 * We can now skip the test for the RST flag since all 2135 * paths to this code happen after packets containing 2136 * RST have been dropped. 2137 * 2138 * In the SYN-RECEIVED state, don't send an ACK unless the 2139 * segment we received passes the SYN-RECEIVED ACK test. 2140 * If it fails send a RST. This breaks the loop in the 2141 * "LAND" DoS attack, and also prevents an ACK storm 2142 * between two listening ports that have been sent forged 2143 * SYN segments, each with the source address of the other. 2144 */ 2145 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2146 (SEQ_GT(tp->snd_una, th->th_ack) || 2147 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2148 rstreason = BANDLIM_RST_OPENPORT; 2149 goto dropwithreset; 2150 } 2151#ifdef TCPDEBUG 2152 if (so->so_options & SO_DEBUG) 2153 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2154 &tcp_savetcp, 0); 2155#endif 2156 m_freem(m); 2157 tp->t_flags |= TF_ACKNOW; 2158 (void) tcp_output(tp); 2159 return; 2160 2161dropwithreset: 2162 /* 2163 * Generate a RST, dropping incoming segment. 2164 * Make ACK acceptable to originator of segment. 2165 * Don't bother to respond if destination was broadcast/multicast. 2166 */ 2167 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2168 goto drop; 2169#ifdef INET6 2170 if (isipv6) { 2171 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2172 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2173 goto drop; 2174 } else 2175#endif /* INET6 */ 2176 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2177 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2178 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2179 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2180 goto drop; 2181 /* IPv6 anycast check is done at tcp6_input() */ 2182 2183 /* 2184 * Perform bandwidth limiting. 2185 */ 2186 if (badport_bandlim(rstreason) < 0) 2187 goto drop; 2188 2189#ifdef TCPDEBUG 2190 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2191 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2192 &tcp_savetcp, 0); 2193#endif 2194 if (thflags & TH_ACK) 2195 /* mtod() below is safe as long as hdr dropping is delayed */ 2196 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2197 TH_RST); 2198 else { 2199 if (thflags & TH_SYN) 2200 tlen++; 2201 /* mtod() below is safe as long as hdr dropping is delayed */ 2202 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2203 (tcp_seq)0, TH_RST|TH_ACK); 2204 } 2205 return; 2206 2207drop: 2208 /* 2209 * Drop space held by incoming segment and return. 2210 */ 2211#ifdef TCPDEBUG 2212 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2213 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2214 &tcp_savetcp, 0); 2215#endif 2216 m_freem(m); 2217 return; 2218} 2219 2220/* 2221 * Parse TCP options and place in tcpopt. 2222 */ 2223static void 2224tcp_dooptions(to, cp, cnt, is_syn) 2225 struct tcpopt *to; 2226 u_char *cp; 2227 int cnt; 2228{ 2229 int opt, optlen; 2230 2231 to->to_flags = 0; 2232 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2233 opt = cp[0]; 2234 if (opt == TCPOPT_EOL) 2235 break; 2236 if (opt == TCPOPT_NOP) 2237 optlen = 1; 2238 else { 2239 if (cnt < 2) 2240 break; 2241 optlen = cp[1]; 2242 if (optlen < 2 || optlen > cnt) 2243 break; 2244 } 2245 switch (opt) { 2246 case TCPOPT_MAXSEG: 2247 if (optlen != TCPOLEN_MAXSEG) 2248 continue; 2249 if (!is_syn) 2250 continue; 2251 to->to_flags |= TOF_MSS; 2252 bcopy((char *)cp + 2, 2253 (char *)&to->to_mss, sizeof(to->to_mss)); 2254 to->to_mss = ntohs(to->to_mss); 2255 break; 2256 case TCPOPT_WINDOW: 2257 if (optlen != TCPOLEN_WINDOW) 2258 continue; 2259 if (! is_syn) 2260 continue; 2261 to->to_flags |= TOF_SCALE; 2262 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2263 break; 2264 case TCPOPT_TIMESTAMP: 2265 if (optlen != TCPOLEN_TIMESTAMP) 2266 continue; 2267 to->to_flags |= TOF_TS; 2268 bcopy((char *)cp + 2, 2269 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2270 to->to_tsval = ntohl(to->to_tsval); 2271 bcopy((char *)cp + 6, 2272 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2273 to->to_tsecr = ntohl(to->to_tsecr); 2274 break; 2275 case TCPOPT_CC: 2276 if (optlen != TCPOLEN_CC) 2277 continue; 2278 to->to_flags |= TOF_CC; 2279 bcopy((char *)cp + 2, 2280 (char *)&to->to_cc, sizeof(to->to_cc)); 2281 to->to_cc = ntohl(to->to_cc); 2282 break; 2283 case TCPOPT_CCNEW: 2284 if (optlen != TCPOLEN_CC) 2285 continue; 2286 if (!is_syn) 2287 continue; 2288 to->to_flags |= TOF_CCNEW; 2289 bcopy((char *)cp + 2, 2290 (char *)&to->to_cc, sizeof(to->to_cc)); 2291 to->to_cc = ntohl(to->to_cc); 2292 break; 2293 case TCPOPT_CCECHO: 2294 if (optlen != TCPOLEN_CC) 2295 continue; 2296 if (!is_syn) 2297 continue; 2298 to->to_flags |= TOF_CCECHO; 2299 bcopy((char *)cp + 2, 2300 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2301 to->to_ccecho = ntohl(to->to_ccecho); 2302 break; 2303 default: 2304 continue; 2305 } 2306 } 2307} 2308 2309/* 2310 * Pull out of band byte out of a segment so 2311 * it doesn't appear in the user's data queue. 2312 * It is still reflected in the segment length for 2313 * sequencing purposes. 2314 */ 2315static void 2316tcp_pulloutofband(so, th, m, off) 2317 struct socket *so; 2318 struct tcphdr *th; 2319 register struct mbuf *m; 2320 int off; /* delayed to be droped hdrlen */ 2321{ 2322 int cnt = off + th->th_urp - 1; 2323 2324 while (cnt >= 0) { 2325 if (m->m_len > cnt) { 2326 char *cp = mtod(m, caddr_t) + cnt; 2327 struct tcpcb *tp = sototcpcb(so); 2328 2329 tp->t_iobc = *cp; 2330 tp->t_oobflags |= TCPOOB_HAVEDATA; 2331 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2332 m->m_len--; 2333 if (m->m_flags & M_PKTHDR) 2334 m->m_pkthdr.len--; 2335 return; 2336 } 2337 cnt -= m->m_len; 2338 m = m->m_next; 2339 if (m == 0) 2340 break; 2341 } 2342 panic("tcp_pulloutofband"); 2343} 2344 2345/* 2346 * Collect new round-trip time estimate 2347 * and update averages and current timeout. 2348 */ 2349static void 2350tcp_xmit_timer(tp, rtt) 2351 register struct tcpcb *tp; 2352 int rtt; 2353{ 2354 register int delta; 2355 2356 tcpstat.tcps_rttupdated++; 2357 tp->t_rttupdated++; 2358 if (tp->t_srtt != 0) { 2359 /* 2360 * srtt is stored as fixed point with 5 bits after the 2361 * binary point (i.e., scaled by 8). The following magic 2362 * is equivalent to the smoothing algorithm in rfc793 with 2363 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2364 * point). Adjust rtt to origin 0. 2365 */ 2366 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2367 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2368 2369 if ((tp->t_srtt += delta) <= 0) 2370 tp->t_srtt = 1; 2371 2372 /* 2373 * We accumulate a smoothed rtt variance (actually, a 2374 * smoothed mean difference), then set the retransmit 2375 * timer to smoothed rtt + 4 times the smoothed variance. 2376 * rttvar is stored as fixed point with 4 bits after the 2377 * binary point (scaled by 16). The following is 2378 * equivalent to rfc793 smoothing with an alpha of .75 2379 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2380 * rfc793's wired-in beta. 2381 */ 2382 if (delta < 0) 2383 delta = -delta; 2384 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2385 if ((tp->t_rttvar += delta) <= 0) 2386 tp->t_rttvar = 1; 2387 } else { 2388 /* 2389 * No rtt measurement yet - use the unsmoothed rtt. 2390 * Set the variance to half the rtt (so our first 2391 * retransmit happens at 3*rtt). 2392 */ 2393 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2394 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2395 } 2396 tp->t_rtttime = 0; 2397 tp->t_rxtshift = 0; 2398 2399 /* 2400 * the retransmit should happen at rtt + 4 * rttvar. 2401 * Because of the way we do the smoothing, srtt and rttvar 2402 * will each average +1/2 tick of bias. When we compute 2403 * the retransmit timer, we want 1/2 tick of rounding and 2404 * 1 extra tick because of +-1/2 tick uncertainty in the 2405 * firing of the timer. The bias will give us exactly the 2406 * 1.5 tick we need. But, because the bias is 2407 * statistical, we have to test that we don't drop below 2408 * the minimum feasible timer (which is 2 ticks). 2409 */ 2410 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2411 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2412 2413 /* 2414 * We received an ack for a packet that wasn't retransmitted; 2415 * it is probably safe to discard any error indications we've 2416 * received recently. This isn't quite right, but close enough 2417 * for now (a route might have failed after we sent a segment, 2418 * and the return path might not be symmetrical). 2419 */ 2420 tp->t_softerror = 0; 2421} 2422 2423/* 2424 * Determine a reasonable value for maxseg size. 2425 * If the route is known, check route for mtu. 2426 * If none, use an mss that can be handled on the outgoing 2427 * interface without forcing IP to fragment; if bigger than 2428 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2429 * to utilize large mbufs. If no route is found, route has no mtu, 2430 * or the destination isn't local, use a default, hopefully conservative 2431 * size (usually 512 or the default IP max size, but no more than the mtu 2432 * of the interface), as we can't discover anything about intervening 2433 * gateways or networks. We also initialize the congestion/slow start 2434 * window to be a single segment if the destination isn't local. 2435 * While looking at the routing entry, we also initialize other path-dependent 2436 * parameters from pre-set or cached values in the routing entry. 2437 * 2438 * Also take into account the space needed for options that we 2439 * send regularly. Make maxseg shorter by that amount to assure 2440 * that we can send maxseg amount of data even when the options 2441 * are present. Store the upper limit of the length of options plus 2442 * data in maxopd. 2443 * 2444 * NOTE that this routine is only called when we process an incoming 2445 * segment, for outgoing segments only tcp_mssopt is called. 2446 * 2447 * In case of T/TCP, we call this routine during implicit connection 2448 * setup as well (offer = -1), to initialize maxseg from the cached 2449 * MSS of our peer. 2450 */ 2451void 2452tcp_mss(tp, offer) 2453 struct tcpcb *tp; 2454 int offer; 2455{ 2456 register struct rtentry *rt; 2457 struct ifnet *ifp; 2458 register int rtt, mss; 2459 u_long bufsize; 2460 struct inpcb *inp; 2461 struct socket *so; 2462 struct rmxp_tao *taop; 2463 int origoffer = offer; 2464#ifdef INET6 2465 int isipv6; 2466 int min_protoh; 2467#endif 2468 2469 inp = tp->t_inpcb; 2470#ifdef INET6 2471 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2472 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2473 : sizeof (struct tcpiphdr); 2474#else 2475#define min_protoh (sizeof (struct tcpiphdr)) 2476#endif 2477#ifdef INET6 2478 if (isipv6) 2479 rt = tcp_rtlookup6(&inp->inp_inc); 2480 else 2481#endif 2482 rt = tcp_rtlookup(&inp->inp_inc); 2483 if (rt == NULL) { 2484 tp->t_maxopd = tp->t_maxseg = 2485#ifdef INET6 2486 isipv6 ? tcp_v6mssdflt : 2487#endif /* INET6 */ 2488 tcp_mssdflt; 2489 return; 2490 } 2491 ifp = rt->rt_ifp; 2492 so = inp->inp_socket; 2493 2494 taop = rmx_taop(rt->rt_rmx); 2495 /* 2496 * Offer == -1 means that we didn't receive SYN yet, 2497 * use cached value in that case; 2498 */ 2499 if (offer == -1) 2500 offer = taop->tao_mssopt; 2501 /* 2502 * Offer == 0 means that there was no MSS on the SYN segment, 2503 * in this case we use tcp_mssdflt. 2504 */ 2505 if (offer == 0) 2506 offer = 2507#ifdef INET6 2508 isipv6 ? tcp_v6mssdflt : 2509#endif /* INET6 */ 2510 tcp_mssdflt; 2511 else 2512 /* 2513 * Sanity check: make sure that maxopd will be large 2514 * enough to allow some data on segments even is the 2515 * all the option space is used (40bytes). Otherwise 2516 * funny things may happen in tcp_output. 2517 */ 2518 offer = max(offer, 64); 2519 taop->tao_mssopt = offer; 2520 2521 /* 2522 * While we're here, check if there's an initial rtt 2523 * or rttvar. Convert from the route-table units 2524 * to scaled multiples of the slow timeout timer. 2525 */ 2526 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2527 /* 2528 * XXX the lock bit for RTT indicates that the value 2529 * is also a minimum value; this is subject to time. 2530 */ 2531 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2532 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2533 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2534 tcpstat.tcps_usedrtt++; 2535 if (rt->rt_rmx.rmx_rttvar) { 2536 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2537 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2538 tcpstat.tcps_usedrttvar++; 2539 } else { 2540 /* default variation is +- 1 rtt */ 2541 tp->t_rttvar = 2542 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2543 } 2544 TCPT_RANGESET(tp->t_rxtcur, 2545 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2546 tp->t_rttmin, TCPTV_REXMTMAX); 2547 } 2548 /* 2549 * if there's an mtu associated with the route, use it 2550 * else, use the link mtu. 2551 */ 2552 if (rt->rt_rmx.rmx_mtu) 2553 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2554 else 2555 { 2556 mss = 2557#ifdef INET6 2558 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2559#endif 2560 ifp->if_mtu 2561#ifdef INET6 2562 ) 2563#endif 2564 - min_protoh; 2565#ifdef INET6 2566 if (isipv6) { 2567 if (!in6_localaddr(&inp->in6p_faddr)) 2568 mss = min(mss, tcp_v6mssdflt); 2569 } else 2570#endif 2571 if (!in_localaddr(inp->inp_faddr)) 2572 mss = min(mss, tcp_mssdflt); 2573 } 2574 mss = min(mss, offer); 2575 /* 2576 * maxopd stores the maximum length of data AND options 2577 * in a segment; maxseg is the amount of data in a normal 2578 * segment. We need to store this value (maxopd) apart 2579 * from maxseg, because now every segment carries options 2580 * and thus we normally have somewhat less data in segments. 2581 */ 2582 tp->t_maxopd = mss; 2583 2584 /* 2585 * In case of T/TCP, origoffer==-1 indicates, that no segments 2586 * were received yet. In this case we just guess, otherwise 2587 * we do the same as before T/TCP. 2588 */ 2589 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2590 (origoffer == -1 || 2591 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2592 mss -= TCPOLEN_TSTAMP_APPA; 2593 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2594 (origoffer == -1 || 2595 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2596 mss -= TCPOLEN_CC_APPA; 2597 2598#if (MCLBYTES & (MCLBYTES - 1)) == 0 2599 if (mss > MCLBYTES) 2600 mss &= ~(MCLBYTES-1); 2601#else 2602 if (mss > MCLBYTES) 2603 mss = mss / MCLBYTES * MCLBYTES; 2604#endif 2605 /* 2606 * If there's a pipesize, change the socket buffer 2607 * to that size. Make the socket buffers an integral 2608 * number of mss units; if the mss is larger than 2609 * the socket buffer, decrease the mss. 2610 */ 2611#ifdef RTV_SPIPE 2612 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2613#endif 2614 bufsize = so->so_snd.sb_hiwat; 2615 if (bufsize < mss) 2616 mss = bufsize; 2617 else { 2618 bufsize = roundup(bufsize, mss); 2619 if (bufsize > sb_max) 2620 bufsize = sb_max; 2621 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2622 } 2623 tp->t_maxseg = mss; 2624 2625#ifdef RTV_RPIPE 2626 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2627#endif 2628 bufsize = so->so_rcv.sb_hiwat; 2629 if (bufsize > mss) { 2630 bufsize = roundup(bufsize, mss); 2631 if (bufsize > sb_max) 2632 bufsize = sb_max; 2633 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2634 } 2635 2636 /* 2637 * Set the slow-start flight size depending on whether this 2638 * is a local network or not. 2639 */ 2640 if ( 2641#ifdef INET6 2642 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2643 (!isipv6 && 2644#endif 2645 in_localaddr(inp->inp_faddr) 2646#ifdef INET6 2647 ) 2648#endif 2649 ) 2650 tp->snd_cwnd = mss * ss_fltsz_local; 2651 else 2652 tp->snd_cwnd = mss * ss_fltsz; 2653 2654 if (rt->rt_rmx.rmx_ssthresh) { 2655 /* 2656 * There's some sort of gateway or interface 2657 * buffer limit on the path. Use this to set 2658 * the slow start threshhold, but set the 2659 * threshold to no less than 2*mss. 2660 */ 2661 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2662 tcpstat.tcps_usedssthresh++; 2663 } 2664} 2665 2666/* 2667 * Determine the MSS option to send on an outgoing SYN. 2668 */ 2669int 2670tcp_mssopt(tp) 2671 struct tcpcb *tp; 2672{ 2673 struct rtentry *rt; 2674#ifdef INET6 2675 int isipv6; 2676 int min_protoh; 2677#endif 2678 2679#ifdef INET6 2680 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2681 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2682 : sizeof (struct tcpiphdr); 2683#else 2684#define min_protoh (sizeof (struct tcpiphdr)) 2685#endif 2686#ifdef INET6 2687 if (isipv6) 2688 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2689 else 2690#endif /* INET6 */ 2691 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2692 if (rt == NULL) 2693 return 2694#ifdef INET6 2695 isipv6 ? tcp_v6mssdflt : 2696#endif /* INET6 */ 2697 tcp_mssdflt; 2698 2699 return rt->rt_ifp->if_mtu - min_protoh; 2700} 2701 2702 2703/* 2704 * Checks for partial ack. If partial ack arrives, force the retransmission 2705 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2706 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2707 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2708 */ 2709static int 2710tcp_newreno(tp, th) 2711 struct tcpcb *tp; 2712 struct tcphdr *th; 2713{ 2714 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2715 tcp_seq onxt = tp->snd_nxt; 2716 u_long ocwnd = tp->snd_cwnd; 2717 2718 callout_stop(tp->tt_rexmt); 2719 tp->t_rtttime = 0; 2720 tp->snd_nxt = th->th_ack; 2721 /* 2722 * Set snd_cwnd to one segment beyond acknowledged offset 2723 * (tp->snd_una has not yet been updated when this function 2724 * is called) 2725 */ 2726 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2727 (void) tcp_output(tp); 2728 tp->snd_cwnd = ocwnd; 2729 if (SEQ_GT(onxt, tp->snd_nxt)) 2730 tp->snd_nxt = onxt; 2731 /* 2732 * Partial window deflation. Relies on fact that tp->snd_una 2733 * not updated yet. 2734 */ 2735 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2736 return (1); 2737 } 2738 return (0); 2739}
|