tcp_input.c revision 82122
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $FreeBSD: head/sys/netinet/tcp_input.c 82122 2001-08-22 00:58:16Z silby $ 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 123int tcp_lq_overflow = 1; 124SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW, 125 &tcp_lq_overflow, 0, 126 "Listen Queue Overflow"); 127 128#ifdef TCP_DROP_SYNFIN 129static int drop_synfin = 0; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 131 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 132#endif 133 134struct inpcbhead tcb; 135#define tcb6 tcb /* for KAME src sync over BSD*'s */ 136struct inpcbinfo tcbinfo; 137 138static void tcp_dooptions __P((struct tcpcb *, 139 u_char *, int, struct tcphdr *, struct tcpopt *)); 140static void tcp_pulloutofband __P((struct socket *, 141 struct tcphdr *, struct mbuf *, int)); 142static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *, 143 struct mbuf *)); 144static void tcp_xmit_timer __P((struct tcpcb *, int)); 145static int tcp_newreno __P((struct tcpcb *, struct tcphdr *)); 146 147/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 148#ifdef INET6 149#define ND6_HINT(tp) \ 150do { \ 151 if ((tp) && (tp)->t_inpcb && \ 152 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 153 (tp)->t_inpcb->in6p_route.ro_rt) \ 154 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 155} while (0) 156#else 157#define ND6_HINT(tp) 158#endif 159 160/* 161 * Indicate whether this ack should be delayed. 162 */ 163#define DELAY_ACK(tp) \ 164 (tcp_delack_enabled && !callout_pending(tp->tt_delack)) 165 166static int 167tcp_reass(tp, th, tlenp, m) 168 register struct tcpcb *tp; 169 register struct tcphdr *th; 170 int *tlenp; 171 struct mbuf *m; 172{ 173 struct tseg_qent *q; 174 struct tseg_qent *p = NULL; 175 struct tseg_qent *nq; 176 struct tseg_qent *te; 177 struct socket *so = tp->t_inpcb->inp_socket; 178 int flags; 179 180 /* 181 * Call with th==0 after become established to 182 * force pre-ESTABLISHED data up to user socket. 183 */ 184 if (th == 0) 185 goto present; 186 187 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 188 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 189 M_NOWAIT); 190 if (te == NULL) { 191 tcpstat.tcps_rcvmemdrop++; 192 m_freem(m); 193 return (0); 194 } 195 196 /* 197 * Find a segment which begins after this one does. 198 */ 199 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 200 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 201 break; 202 p = q; 203 } 204 205 /* 206 * If there is a preceding segment, it may provide some of 207 * our data already. If so, drop the data from the incoming 208 * segment. If it provides all of our data, drop us. 209 */ 210 if (p != NULL) { 211 register int i; 212 /* conversion to int (in i) handles seq wraparound */ 213 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 214 if (i > 0) { 215 if (i >= *tlenp) { 216 tcpstat.tcps_rcvduppack++; 217 tcpstat.tcps_rcvdupbyte += *tlenp; 218 m_freem(m); 219 FREE(te, M_TSEGQ); 220 /* 221 * Try to present any queued data 222 * at the left window edge to the user. 223 * This is needed after the 3-WHS 224 * completes. 225 */ 226 goto present; /* ??? */ 227 } 228 m_adj(m, i); 229 *tlenp -= i; 230 th->th_seq += i; 231 } 232 } 233 tcpstat.tcps_rcvoopack++; 234 tcpstat.tcps_rcvoobyte += *tlenp; 235 236 /* 237 * While we overlap succeeding segments trim them or, 238 * if they are completely covered, dequeue them. 239 */ 240 while (q) { 241 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 242 if (i <= 0) 243 break; 244 if (i < q->tqe_len) { 245 q->tqe_th->th_seq += i; 246 q->tqe_len -= i; 247 m_adj(q->tqe_m, i); 248 break; 249 } 250 251 nq = LIST_NEXT(q, tqe_q); 252 LIST_REMOVE(q, tqe_q); 253 m_freem(q->tqe_m); 254 FREE(q, M_TSEGQ); 255 q = nq; 256 } 257 258 /* Insert the new segment queue entry into place. */ 259 te->tqe_m = m; 260 te->tqe_th = th; 261 te->tqe_len = *tlenp; 262 263 if (p == NULL) { 264 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 265 } else { 266 LIST_INSERT_AFTER(p, te, tqe_q); 267 } 268 269present: 270 /* 271 * Present data to user, advancing rcv_nxt through 272 * completed sequence space. 273 */ 274 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 275 return (0); 276 q = LIST_FIRST(&tp->t_segq); 277 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 278 return (0); 279 do { 280 tp->rcv_nxt += q->tqe_len; 281 flags = q->tqe_th->th_flags & TH_FIN; 282 nq = LIST_NEXT(q, tqe_q); 283 LIST_REMOVE(q, tqe_q); 284 if (so->so_state & SS_CANTRCVMORE) 285 m_freem(q->tqe_m); 286 else 287 sbappend(&so->so_rcv, q->tqe_m); 288 FREE(q, M_TSEGQ); 289 q = nq; 290 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 291 ND6_HINT(tp); 292 sorwakeup(so); 293 return (flags); 294} 295 296/* 297 * TCP input routine, follows pages 65-76 of the 298 * protocol specification dated September, 1981 very closely. 299 */ 300#ifdef INET6 301int 302tcp6_input(mp, offp, proto) 303 struct mbuf **mp; 304 int *offp, proto; 305{ 306 register struct mbuf *m = *mp; 307 struct in6_ifaddr *ia6; 308 309 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 310 311 /* 312 * draft-itojun-ipv6-tcp-to-anycast 313 * better place to put this in? 314 */ 315 ia6 = ip6_getdstifaddr(m); 316 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 317 struct ip6_hdr *ip6; 318 319 ip6 = mtod(m, struct ip6_hdr *); 320 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 321 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 322 return IPPROTO_DONE; 323 } 324 325 tcp_input(m, *offp, proto); 326 return IPPROTO_DONE; 327} 328#endif 329 330void 331tcp_input(m, off0, proto) 332 register struct mbuf *m; 333 int off0, proto; 334{ 335 register struct tcphdr *th; 336 register struct ip *ip = NULL; 337 register struct ipovly *ipov; 338 register struct inpcb *inp; 339 u_char *optp = NULL; 340 int optlen = 0; 341 int len, tlen, off; 342 int drop_hdrlen; 343 register struct tcpcb *tp = 0; 344 register int thflags; 345 struct socket *so = 0; 346 int todrop, acked, ourfinisacked, needoutput = 0; 347 struct in_addr laddr; 348#ifdef INET6 349 struct in6_addr laddr6; 350#endif 351 int dropsocket = 0; 352 int iss = 0; 353 u_long tiwin; 354 struct tcpopt to; /* options in this segment */ 355 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 356 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 357#ifdef TCPDEBUG 358 short ostate = 0; 359#endif 360#ifdef INET6 361 struct ip6_hdr *ip6 = NULL; 362 int isipv6; 363#endif /* INET6 */ 364 int rstreason; /* For badport_bandlim accounting purposes */ 365 366#ifdef INET6 367 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 368#endif 369 bzero((char *)&to, sizeof(to)); 370 371 tcpstat.tcps_rcvtotal++; 372 373#ifdef INET6 374 if (isipv6) { 375 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 376 ip6 = mtod(m, struct ip6_hdr *); 377 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 378 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 379 tcpstat.tcps_rcvbadsum++; 380 goto drop; 381 } 382 th = (struct tcphdr *)((caddr_t)ip6 + off0); 383 384 /* 385 * Be proactive about unspecified IPv6 address in source. 386 * As we use all-zero to indicate unbounded/unconnected pcb, 387 * unspecified IPv6 address can be used to confuse us. 388 * 389 * Note that packets with unspecified IPv6 destination is 390 * already dropped in ip6_input. 391 */ 392 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 393 /* XXX stat */ 394 goto drop; 395 } 396 } else 397#endif /* INET6 */ 398 { 399 /* 400 * Get IP and TCP header together in first mbuf. 401 * Note: IP leaves IP header in first mbuf. 402 */ 403 if (off0 > sizeof (struct ip)) { 404 ip_stripoptions(m, (struct mbuf *)0); 405 off0 = sizeof(struct ip); 406 } 407 if (m->m_len < sizeof (struct tcpiphdr)) { 408 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 409 tcpstat.tcps_rcvshort++; 410 return; 411 } 412 } 413 ip = mtod(m, struct ip *); 414 ipov = (struct ipovly *)ip; 415 th = (struct tcphdr *)((caddr_t)ip + off0); 416 tlen = ip->ip_len; 417 418 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 419 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 420 th->th_sum = m->m_pkthdr.csum_data; 421 else 422 th->th_sum = in_pseudo(ip->ip_src.s_addr, 423 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 424 ip->ip_len + IPPROTO_TCP)); 425 th->th_sum ^= 0xffff; 426 } else { 427 /* 428 * Checksum extended TCP header and data. 429 */ 430 len = sizeof (struct ip) + tlen; 431 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 432 ipov->ih_len = (u_short)tlen; 433 HTONS(ipov->ih_len); 434 th->th_sum = in_cksum(m, len); 435 } 436 if (th->th_sum) { 437 tcpstat.tcps_rcvbadsum++; 438 goto drop; 439 } 440#ifdef INET6 441 /* Re-initialization for later version check */ 442 ip->ip_v = IPVERSION; 443#endif 444 } 445 446 /* 447 * Check that TCP offset makes sense, 448 * pull out TCP options and adjust length. XXX 449 */ 450 off = th->th_off << 2; 451 if (off < sizeof (struct tcphdr) || off > tlen) { 452 tcpstat.tcps_rcvbadoff++; 453 goto drop; 454 } 455 tlen -= off; /* tlen is used instead of ti->ti_len */ 456 if (off > sizeof (struct tcphdr)) { 457#ifdef INET6 458 if (isipv6) { 459 IP6_EXTHDR_CHECK(m, off0, off, ); 460 ip6 = mtod(m, struct ip6_hdr *); 461 th = (struct tcphdr *)((caddr_t)ip6 + off0); 462 } else 463#endif /* INET6 */ 464 { 465 if (m->m_len < sizeof(struct ip) + off) { 466 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 467 tcpstat.tcps_rcvshort++; 468 return; 469 } 470 ip = mtod(m, struct ip *); 471 ipov = (struct ipovly *)ip; 472 th = (struct tcphdr *)((caddr_t)ip + off0); 473 } 474 } 475 optlen = off - sizeof (struct tcphdr); 476 optp = (u_char *)(th + 1); 477 } 478 thflags = th->th_flags; 479 480#ifdef TCP_DROP_SYNFIN 481 /* 482 * If the drop_synfin option is enabled, drop all packets with 483 * both the SYN and FIN bits set. This prevents e.g. nmap from 484 * identifying the TCP/IP stack. 485 * 486 * This is a violation of the TCP specification. 487 */ 488 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 489 goto drop; 490#endif 491 492 /* 493 * Convert TCP protocol specific fields to host format. 494 */ 495 NTOHL(th->th_seq); 496 NTOHL(th->th_ack); 497 NTOHS(th->th_win); 498 NTOHS(th->th_urp); 499 500 /* 501 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 502 * until after ip6_savecontrol() is called and before other functions 503 * which don't want those proto headers. 504 * Because ip6_savecontrol() is going to parse the mbuf to 505 * search for data to be passed up to user-land, it wants mbuf 506 * parameters to be unchanged. 507 */ 508 drop_hdrlen = off0 + off; 509 510 /* 511 * Locate pcb for segment. 512 */ 513findpcb: 514#ifdef IPFIREWALL_FORWARD 515 if (ip_fw_fwd_addr != NULL 516#ifdef INET6 517 && isipv6 == NULL /* IPv6 support is not yet */ 518#endif /* INET6 */ 519 ) { 520 /* 521 * Diverted. Pretend to be the destination. 522 * already got one like this? 523 */ 524 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 525 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 526 if (!inp) { 527 /* 528 * No, then it's new. Try find the ambushing socket 529 */ 530 if (!ip_fw_fwd_addr->sin_port) { 531 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 532 th->th_sport, ip_fw_fwd_addr->sin_addr, 533 th->th_dport, 1, m->m_pkthdr.rcvif); 534 } else { 535 inp = in_pcblookup_hash(&tcbinfo, 536 ip->ip_src, th->th_sport, 537 ip_fw_fwd_addr->sin_addr, 538 ntohs(ip_fw_fwd_addr->sin_port), 1, 539 m->m_pkthdr.rcvif); 540 } 541 } 542 ip_fw_fwd_addr = NULL; 543 } else 544#endif /* IPFIREWALL_FORWARD */ 545 { 546#ifdef INET6 547 if (isipv6) 548 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 549 &ip6->ip6_dst, th->th_dport, 1, 550 m->m_pkthdr.rcvif); 551 else 552#endif /* INET6 */ 553 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 554 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 555 } 556 557#ifdef IPSEC 558#ifdef INET6 559 if (isipv6) { 560 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 561 ipsec6stat.in_polvio++; 562 goto drop; 563 } 564 } else 565#endif /* INET6 */ 566 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 567 ipsecstat.in_polvio++; 568 goto drop; 569 } 570#endif /*IPSEC*/ 571 572 /* 573 * If the state is CLOSED (i.e., TCB does not exist) then 574 * all data in the incoming segment is discarded. 575 * If the TCB exists but is in CLOSED state, it is embryonic, 576 * but should either do a listen or a connect soon. 577 */ 578 if (inp == NULL) { 579 if (log_in_vain) { 580#ifdef INET6 581 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 582#else /* INET6 */ 583 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 584#endif /* INET6 */ 585 586#ifdef INET6 587 if (isipv6) { 588 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 589 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 590 } else 591#endif 592 { 593 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 594 strcpy(sbuf, inet_ntoa(ip->ip_src)); 595 } 596 switch (log_in_vain) { 597 case 1: 598 if(thflags & TH_SYN) 599 log(LOG_INFO, 600 "Connection attempt to TCP %s:%d from %s:%d\n", 601 dbuf, ntohs(th->th_dport), 602 sbuf, 603 ntohs(th->th_sport)); 604 break; 605 case 2: 606 log(LOG_INFO, 607 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 608 dbuf, ntohs(th->th_dport), sbuf, 609 ntohs(th->th_sport), thflags); 610 break; 611 default: 612 break; 613 } 614 } 615 if (blackhole) { 616 switch (blackhole) { 617 case 1: 618 if (thflags & TH_SYN) 619 goto drop; 620 break; 621 case 2: 622 goto drop; 623 default: 624 goto drop; 625 } 626 } 627 rstreason = BANDLIM_RST_CLOSEDPORT; 628 goto dropwithreset; 629 } 630 tp = intotcpcb(inp); 631 if (tp == 0) { 632 rstreason = BANDLIM_RST_CLOSEDPORT; 633 goto dropwithreset; 634 } 635 if (tp->t_state == TCPS_CLOSED) 636 goto drop; 637 638 /* Unscale the window into a 32-bit value. */ 639 if ((thflags & TH_SYN) == 0) 640 tiwin = th->th_win << tp->snd_scale; 641 else 642 tiwin = th->th_win; 643 644 so = inp->inp_socket; 645 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 646#ifdef TCPDEBUG 647 if (so->so_options & SO_DEBUG) { 648 ostate = tp->t_state; 649#ifdef INET6 650 if (isipv6) 651 bcopy((char *)ip6, (char *)tcp_saveipgen, 652 sizeof(*ip6)); 653 else 654#endif /* INET6 */ 655 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 656 tcp_savetcp = *th; 657 } 658#endif 659 if (so->so_options & SO_ACCEPTCONN) { 660 register struct tcpcb *tp0 = tp; 661 struct socket *so2; 662#ifdef IPSEC 663 struct socket *oso; 664#endif 665#ifdef INET6 666 struct inpcb *oinp = sotoinpcb(so); 667#endif /* INET6 */ 668 669#ifndef IPSEC 670 /* 671 * Current IPsec implementation makes incorrect IPsec 672 * cache if this check is done here. 673 * So delay this until duplicated socket is created. 674 */ 675 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 676 /* 677 * Note: dropwithreset makes sure we don't 678 * send a RST in response to a RST. 679 */ 680 if (thflags & TH_ACK) { 681 tcpstat.tcps_badsyn++; 682 rstreason = BANDLIM_RST_OPENPORT; 683 goto dropwithreset; 684 } 685 goto drop; 686 } 687#endif 688 689#ifdef INET6 690 /* 691 * If deprecated address is forbidden, 692 * we do not accept SYN to deprecated interface 693 * address to prevent any new inbound connection from 694 * getting established. 695 * When we do not accept SYN, we send a TCP RST, 696 * with deprecated source address (instead of dropping 697 * it). We compromise it as it is much better for peer 698 * to send a RST, and RST will be the final packet 699 * for the exchange. 700 * 701 * If we do not forbid deprecated addresses, we accept 702 * the SYN packet. RFC2462 does not suggest dropping 703 * SYN in this case. 704 * If we decipher RFC2462 5.5.4, it says like this: 705 * 1. use of deprecated addr with existing 706 * communication is okay - "SHOULD continue to be 707 * used" 708 * 2. use of it with new communication: 709 * (2a) "SHOULD NOT be used if alternate address 710 * with sufficient scope is available" 711 * (2b) nothing mentioned otherwise. 712 * Here we fall into (2b) case as we have no choice in 713 * our source address selection - we must obey the peer. 714 * 715 * The wording in RFC2462 is confusing, and there are 716 * multiple description text for deprecated address 717 * handling - worse, they are not exactly the same. 718 * I believe 5.5.4 is the best one, so we follow 5.5.4. 719 */ 720 if (isipv6 && !ip6_use_deprecated) { 721 struct in6_ifaddr *ia6; 722 723 if ((ia6 = ip6_getdstifaddr(m)) && 724 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 725 tp = NULL; 726 rstreason = BANDLIM_RST_OPENPORT; 727 goto dropwithreset; 728 } 729 } 730#endif 731 732 so2 = sonewconn(so, 0); 733 if (so2 == 0) { 734 /* 735 * If we were unable to create a new socket 736 * for this SYN, we call sodropablereq to 737 * see if there are any other sockets we 738 * can kick out of the listen queue. If 739 * so, we'll silently drop the socket 740 * sodropablereq told us to drop and 741 * create a new one. 742 * 743 * If sodropablereq returns 0, we'll 744 * simply drop the incoming SYN, as we 745 * can not allocate a socket for it. 746 */ 747 tcpstat.tcps_listendrop++; 748 so2 = sodropablereq(so); 749 if (so2) { 750 if (tcp_lq_overflow) 751 sototcpcb(so2)->t_flags |= 752 TF_LQ_OVERFLOW; 753 tcp_close(sototcpcb(so2)); 754 so2 = sonewconn(so, 0); 755 } 756 if (!so2) 757 goto drop; 758 } 759#ifdef IPSEC 760 oso = so; 761#endif 762 so = so2; 763 /* 764 * This is ugly, but .... 765 * 766 * Mark socket as temporary until we're 767 * committed to keeping it. The code at 768 * ``drop'' and ``dropwithreset'' check the 769 * flag dropsocket to see if the temporary 770 * socket created here should be discarded. 771 * We mark the socket as discardable until 772 * we're committed to it below in TCPS_LISTEN. 773 */ 774 dropsocket++; 775 inp = (struct inpcb *)so->so_pcb; 776#ifdef INET6 777 if (isipv6) 778 inp->in6p_laddr = ip6->ip6_dst; 779 else { 780 inp->inp_vflag &= ~INP_IPV6; 781 inp->inp_vflag |= INP_IPV4; 782#endif /* INET6 */ 783 inp->inp_laddr = ip->ip_dst; 784#ifdef INET6 785 } 786#endif /* INET6 */ 787 inp->inp_lport = th->th_dport; 788 if (in_pcbinshash(inp) != 0) { 789 /* 790 * Undo the assignments above if we failed to 791 * put the PCB on the hash lists. 792 */ 793#ifdef INET6 794 if (isipv6) 795 inp->in6p_laddr = in6addr_any; 796 else 797#endif /* INET6 */ 798 inp->inp_laddr.s_addr = INADDR_ANY; 799 inp->inp_lport = 0; 800 goto drop; 801 } 802#ifdef IPSEC 803 /* 804 * To avoid creating incorrectly cached IPsec 805 * association, this is need to be done here. 806 * 807 * Subject: (KAME-snap 748) 808 * From: Wayne Knowles <w.knowles@niwa.cri.nz> 809 * ftp://ftp.kame.net/pub/mail-list/snap-users/748 810 */ 811 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 812 /* 813 * Note: dropwithreset makes sure we don't 814 * send a RST in response to a RST. 815 */ 816 if (thflags & TH_ACK) { 817 tcpstat.tcps_badsyn++; 818 rstreason = BANDLIM_RST_OPENPORT; 819 goto dropwithreset; 820 } 821 goto drop; 822 } 823#endif 824#ifdef INET6 825 if (isipv6) { 826 /* 827 * Inherit socket options from the listening 828 * socket. 829 * Note that in6p_inputopts are not (even 830 * should not be) copied, since it stores 831 * previously received options and is used to 832 * detect if each new option is different than 833 * the previous one and hence should be passed 834 * to a user. 835 * If we copied in6p_inputopts, a user would 836 * not be able to receive options just after 837 * calling the accept system call. 838 */ 839 inp->inp_flags |= 840 oinp->inp_flags & INP_CONTROLOPTS; 841 if (oinp->in6p_outputopts) 842 inp->in6p_outputopts = 843 ip6_copypktopts(oinp->in6p_outputopts, 844 M_NOWAIT); 845 } else 846#endif /* INET6 */ 847 inp->inp_options = ip_srcroute(); 848#ifdef IPSEC 849 /* copy old policy into new socket's */ 850 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, 851 inp->inp_sp)) 852 printf("tcp_input: could not copy policy\n"); 853#endif 854 tp = intotcpcb(inp); 855 tp->t_state = TCPS_LISTEN; 856 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 857 858 /* Compute proper scaling value from buffer space */ 859 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 860 TCP_MAXWIN << tp->request_r_scale < 861 so->so_rcv.sb_hiwat) 862 tp->request_r_scale++; 863 } 864 } 865 866 /* 867 * Segment received on connection. 868 * Reset idle time and keep-alive timer. 869 */ 870 tp->t_rcvtime = ticks; 871 if (TCPS_HAVEESTABLISHED(tp->t_state)) 872 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 873 874 /* 875 * Process options if not in LISTEN state, 876 * else do it below (after getting remote address). 877 */ 878 if (tp->t_state != TCPS_LISTEN) 879 tcp_dooptions(tp, optp, optlen, th, &to); 880 881 /* 882 * Header prediction: check for the two common cases 883 * of a uni-directional data xfer. If the packet has 884 * no control flags, is in-sequence, the window didn't 885 * change and we're not retransmitting, it's a 886 * candidate. If the length is zero and the ack moved 887 * forward, we're the sender side of the xfer. Just 888 * free the data acked & wake any higher level process 889 * that was blocked waiting for space. If the length 890 * is non-zero and the ack didn't move, we're the 891 * receiver side. If we're getting packets in-order 892 * (the reassembly queue is empty), add the data to 893 * the socket buffer and note that we need a delayed ack. 894 * Make sure that the hidden state-flags are also off. 895 * Since we check for TCPS_ESTABLISHED above, it can only 896 * be TH_NEEDSYN. 897 */ 898 if (tp->t_state == TCPS_ESTABLISHED && 899 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 900 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 901 ((to.to_flag & TOF_TS) == 0 || 902 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 903 /* 904 * Using the CC option is compulsory if once started: 905 * the segment is OK if no T/TCP was negotiated or 906 * if the segment has a CC option equal to CCrecv 907 */ 908 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 909 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 910 th->th_seq == tp->rcv_nxt && 911 tiwin && tiwin == tp->snd_wnd && 912 tp->snd_nxt == tp->snd_max) { 913 914 /* 915 * If last ACK falls within this segment's sequence numbers, 916 * record the timestamp. 917 * NOTE that the test is modified according to the latest 918 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 919 */ 920 if ((to.to_flag & TOF_TS) != 0 && 921 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 922 tp->ts_recent_age = ticks; 923 tp->ts_recent = to.to_tsval; 924 } 925 926 if (tlen == 0) { 927 if (SEQ_GT(th->th_ack, tp->snd_una) && 928 SEQ_LEQ(th->th_ack, tp->snd_max) && 929 tp->snd_cwnd >= tp->snd_wnd && 930 tp->t_dupacks < tcprexmtthresh) { 931 /* 932 * this is a pure ack for outstanding data. 933 */ 934 ++tcpstat.tcps_predack; 935 /* 936 * "bad retransmit" recovery 937 */ 938 if (tp->t_rxtshift == 1 && 939 ticks < tp->t_badrxtwin) { 940 tp->snd_cwnd = tp->snd_cwnd_prev; 941 tp->snd_ssthresh = 942 tp->snd_ssthresh_prev; 943 tp->snd_nxt = tp->snd_max; 944 tp->t_badrxtwin = 0; 945 } 946 if ((to.to_flag & TOF_TS) != 0) 947 tcp_xmit_timer(tp, 948 ticks - to.to_tsecr + 1); 949 else if (tp->t_rtttime && 950 SEQ_GT(th->th_ack, tp->t_rtseq)) 951 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 952 acked = th->th_ack - tp->snd_una; 953 tcpstat.tcps_rcvackpack++; 954 tcpstat.tcps_rcvackbyte += acked; 955 sbdrop(&so->so_snd, acked); 956 tp->snd_una = th->th_ack; 957 m_freem(m); 958 ND6_HINT(tp); /* some progress has been done */ 959 960 /* 961 * If all outstanding data are acked, stop 962 * retransmit timer, otherwise restart timer 963 * using current (possibly backed-off) value. 964 * If process is waiting for space, 965 * wakeup/selwakeup/signal. If data 966 * are ready to send, let tcp_output 967 * decide between more output or persist. 968 */ 969 if (tp->snd_una == tp->snd_max) 970 callout_stop(tp->tt_rexmt); 971 else if (!callout_active(tp->tt_persist)) 972 callout_reset(tp->tt_rexmt, 973 tp->t_rxtcur, 974 tcp_timer_rexmt, tp); 975 976 sowwakeup(so); 977 if (so->so_snd.sb_cc) 978 (void) tcp_output(tp); 979 return; 980 } 981 } else if (th->th_ack == tp->snd_una && 982 LIST_EMPTY(&tp->t_segq) && 983 tlen <= sbspace(&so->so_rcv)) { 984 /* 985 * this is a pure, in-sequence data packet 986 * with nothing on the reassembly queue and 987 * we have enough buffer space to take it. 988 */ 989 ++tcpstat.tcps_preddat; 990 tp->rcv_nxt += tlen; 991 tcpstat.tcps_rcvpack++; 992 tcpstat.tcps_rcvbyte += tlen; 993 ND6_HINT(tp); /* some progress has been done */ 994 /* 995 * Add data to socket buffer. 996 */ 997 m_adj(m, drop_hdrlen); /* delayed header drop */ 998 sbappend(&so->so_rcv, m); 999 sorwakeup(so); 1000 if (DELAY_ACK(tp)) { 1001 callout_reset(tp->tt_delack, tcp_delacktime, 1002 tcp_timer_delack, tp); 1003 } else { 1004 tp->t_flags |= TF_ACKNOW; 1005 tcp_output(tp); 1006 } 1007 return; 1008 } 1009 } 1010 1011 /* 1012 * Calculate amount of space in receive window, 1013 * and then do TCP input processing. 1014 * Receive window is amount of space in rcv queue, 1015 * but not less than advertised window. 1016 */ 1017 { int win; 1018 1019 win = sbspace(&so->so_rcv); 1020 if (win < 0) 1021 win = 0; 1022 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1023 } 1024 1025 switch (tp->t_state) { 1026 1027 /* 1028 * If the state is LISTEN then ignore segment if it contains an RST. 1029 * If the segment contains an ACK then it is bad and send a RST. 1030 * If it does not contain a SYN then it is not interesting; drop it. 1031 * If it is from this socket, drop it, it must be forged. 1032 * Don't bother responding if the destination was a broadcast. 1033 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 1034 * tp->iss, and send a segment: 1035 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 1036 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 1037 * Fill in remote peer address fields if not previously specified. 1038 * Enter SYN_RECEIVED state, and process any other fields of this 1039 * segment in this state. 1040 */ 1041 case TCPS_LISTEN: { 1042 register struct sockaddr_in *sin; 1043#ifdef INET6 1044 register struct sockaddr_in6 *sin6; 1045#endif 1046 1047 if (thflags & TH_RST) 1048 goto drop; 1049 if (thflags & TH_ACK) { 1050 rstreason = BANDLIM_RST_OPENPORT; 1051 goto dropwithreset; 1052 } 1053 if ((thflags & TH_SYN) == 0) 1054 goto drop; 1055 if (th->th_dport == th->th_sport) { 1056#ifdef INET6 1057 if (isipv6) { 1058 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 1059 &ip6->ip6_src)) 1060 goto drop; 1061 } else 1062#endif /* INET6 */ 1063 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1064 goto drop; 1065 } 1066 /* 1067 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1068 * in_broadcast() should never return true on a received 1069 * packet with M_BCAST not set. 1070 * 1071 * Packets with a multicast source address should also 1072 * be discarded. 1073 */ 1074 if (m->m_flags & (M_BCAST|M_MCAST)) 1075 goto drop; 1076#ifdef INET6 1077 if (isipv6) { 1078 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1079 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1080 goto drop; 1081 } else 1082#endif 1083 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1084 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1085 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 1086 goto drop; 1087#ifdef INET6 1088 if (isipv6) { 1089 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, 1090 M_SONAME, M_NOWAIT | M_ZERO); 1091 if (sin6 == NULL) 1092 goto drop; 1093 sin6->sin6_family = AF_INET6; 1094 sin6->sin6_len = sizeof(*sin6); 1095 sin6->sin6_addr = ip6->ip6_src; 1096 sin6->sin6_port = th->th_sport; 1097 laddr6 = inp->in6p_laddr; 1098 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1099 inp->in6p_laddr = ip6->ip6_dst; 1100 if (in6_pcbconnect(inp, (struct sockaddr *)sin6, 1101 &proc0)) { 1102 inp->in6p_laddr = laddr6; 1103 FREE(sin6, M_SONAME); 1104 goto drop; 1105 } 1106 FREE(sin6, M_SONAME); 1107 } else 1108#endif 1109 { 1110 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 1111 M_NOWAIT); 1112 if (sin == NULL) 1113 goto drop; 1114 sin->sin_family = AF_INET; 1115 sin->sin_len = sizeof(*sin); 1116 sin->sin_addr = ip->ip_src; 1117 sin->sin_port = th->th_sport; 1118 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 1119 laddr = inp->inp_laddr; 1120 if (inp->inp_laddr.s_addr == INADDR_ANY) 1121 inp->inp_laddr = ip->ip_dst; 1122 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 1123 inp->inp_laddr = laddr; 1124 FREE(sin, M_SONAME); 1125 goto drop; 1126 } 1127 FREE(sin, M_SONAME); 1128 } 1129 if ((taop = tcp_gettaocache(inp)) == NULL) { 1130 taop = &tao_noncached; 1131 bzero(taop, sizeof(*taop)); 1132 } 1133 tcp_dooptions(tp, optp, optlen, th, &to); 1134 if (iss) 1135 tp->iss = iss; 1136 else { 1137 tp->iss = tcp_new_isn(tp); 1138 } 1139 tp->irs = th->th_seq; 1140 tcp_sendseqinit(tp); 1141 tcp_rcvseqinit(tp); 1142 tp->snd_recover = tp->snd_una; 1143 /* 1144 * Initialization of the tcpcb for transaction; 1145 * set SND.WND = SEG.WND, 1146 * initialize CCsend and CCrecv. 1147 */ 1148 tp->snd_wnd = tiwin; /* initial send-window */ 1149 tp->cc_send = CC_INC(tcp_ccgen); 1150 tp->cc_recv = to.to_cc; 1151 /* 1152 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1153 * - compare SEG.CC against cached CC from the same host, 1154 * if any. 1155 * - if SEG.CC > chached value, SYN must be new and is accepted 1156 * immediately: save new CC in the cache, mark the socket 1157 * connected, enter ESTABLISHED state, turn on flag to 1158 * send a SYN in the next segment. 1159 * A virtual advertised window is set in rcv_adv to 1160 * initialize SWS prevention. Then enter normal segment 1161 * processing: drop SYN, process data and FIN. 1162 * - otherwise do a normal 3-way handshake. 1163 */ 1164 if ((to.to_flag & TOF_CC) != 0) { 1165 if (((tp->t_flags & TF_NOPUSH) != 0) && 1166 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 1167 1168 taop->tao_cc = to.to_cc; 1169 tp->t_starttime = ticks; 1170 tp->t_state = TCPS_ESTABLISHED; 1171 1172 /* 1173 * If there is a FIN, or if there is data and the 1174 * connection is local, then delay SYN,ACK(SYN) in 1175 * the hope of piggy-backing it on a response 1176 * segment. Otherwise must send ACK now in case 1177 * the other side is slow starting. 1178 */ 1179 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 1180 (tlen != 0 && 1181#ifdef INET6 1182 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 1183 || 1184 (!isipv6 && 1185#endif 1186 in_localaddr(inp->inp_faddr) 1187#ifdef INET6 1188 )) 1189#endif 1190 ))) { 1191 callout_reset(tp->tt_delack, tcp_delacktime, 1192 tcp_timer_delack, tp); 1193 tp->t_flags |= TF_NEEDSYN; 1194 } else 1195 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1196 1197 /* 1198 * Limit the `virtual advertised window' to TCP_MAXWIN 1199 * here. Even if we requested window scaling, it will 1200 * become effective only later when our SYN is acked. 1201 */ 1202 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 1203 tcpstat.tcps_connects++; 1204 soisconnected(so); 1205 callout_reset(tp->tt_keep, tcp_keepinit, 1206 tcp_timer_keep, tp); 1207 dropsocket = 0; /* committed to socket */ 1208 tcpstat.tcps_accepts++; 1209 goto trimthenstep6; 1210 } 1211 /* else do standard 3-way handshake */ 1212 } else { 1213 /* 1214 * No CC option, but maybe CC.NEW: 1215 * invalidate cached value. 1216 */ 1217 taop->tao_cc = 0; 1218 } 1219 /* 1220 * TAO test failed or there was no CC option, 1221 * do a standard 3-way handshake. 1222 */ 1223 tp->t_flags |= TF_ACKNOW; 1224 tp->t_state = TCPS_SYN_RECEIVED; 1225 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 1226 dropsocket = 0; /* committed to socket */ 1227 tcpstat.tcps_accepts++; 1228 goto trimthenstep6; 1229 } 1230 1231 /* 1232 * If the state is SYN_RECEIVED: 1233 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1234 */ 1235 case TCPS_SYN_RECEIVED: 1236 if ((thflags & TH_ACK) && 1237 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1238 SEQ_GT(th->th_ack, tp->snd_max))) { 1239 rstreason = BANDLIM_RST_OPENPORT; 1240 goto dropwithreset; 1241 } 1242 break; 1243 1244 /* 1245 * If the state is SYN_SENT: 1246 * if seg contains an ACK, but not for our SYN, drop the input. 1247 * if seg contains a RST, then drop the connection. 1248 * if seg does not contain SYN, then drop it. 1249 * Otherwise this is an acceptable SYN segment 1250 * initialize tp->rcv_nxt and tp->irs 1251 * if seg contains ack then advance tp->snd_una 1252 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1253 * arrange for segment to be acked (eventually) 1254 * continue processing rest of data/controls, beginning with URG 1255 */ 1256 case TCPS_SYN_SENT: 1257 if ((taop = tcp_gettaocache(inp)) == NULL) { 1258 taop = &tao_noncached; 1259 bzero(taop, sizeof(*taop)); 1260 } 1261 1262 if ((thflags & TH_ACK) && 1263 (SEQ_LEQ(th->th_ack, tp->iss) || 1264 SEQ_GT(th->th_ack, tp->snd_max))) { 1265 /* 1266 * If we have a cached CCsent for the remote host, 1267 * hence we haven't just crashed and restarted, 1268 * do not send a RST. This may be a retransmission 1269 * from the other side after our earlier ACK was lost. 1270 * Our new SYN, when it arrives, will serve as the 1271 * needed ACK. 1272 */ 1273 if (taop->tao_ccsent != 0) 1274 goto drop; 1275 else { 1276 rstreason = BANDLIM_UNLIMITED; 1277 goto dropwithreset; 1278 } 1279 } 1280 if (thflags & TH_RST) { 1281 if (thflags & TH_ACK) 1282 tp = tcp_drop(tp, ECONNREFUSED); 1283 goto drop; 1284 } 1285 if ((thflags & TH_SYN) == 0) 1286 goto drop; 1287 tp->snd_wnd = th->th_win; /* initial send window */ 1288 tp->cc_recv = to.to_cc; /* foreign CC */ 1289 1290 tp->irs = th->th_seq; 1291 tcp_rcvseqinit(tp); 1292 if (thflags & TH_ACK) { 1293 /* 1294 * Our SYN was acked. If segment contains CC.ECHO 1295 * option, check it to make sure this segment really 1296 * matches our SYN. If not, just drop it as old 1297 * duplicate, but send an RST if we're still playing 1298 * by the old rules. If no CC.ECHO option, make sure 1299 * we don't get fooled into using T/TCP. 1300 */ 1301 if (to.to_flag & TOF_CCECHO) { 1302 if (tp->cc_send != to.to_ccecho) { 1303 if (taop->tao_ccsent != 0) 1304 goto drop; 1305 else { 1306 rstreason = BANDLIM_UNLIMITED; 1307 goto dropwithreset; 1308 } 1309 } 1310 } else 1311 tp->t_flags &= ~TF_RCVD_CC; 1312 tcpstat.tcps_connects++; 1313 soisconnected(so); 1314 /* Do window scaling on this connection? */ 1315 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1316 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1317 tp->snd_scale = tp->requested_s_scale; 1318 tp->rcv_scale = tp->request_r_scale; 1319 } 1320 /* Segment is acceptable, update cache if undefined. */ 1321 if (taop->tao_ccsent == 0) 1322 taop->tao_ccsent = to.to_ccecho; 1323 1324 tp->rcv_adv += tp->rcv_wnd; 1325 tp->snd_una++; /* SYN is acked */ 1326 /* 1327 * If there's data, delay ACK; if there's also a FIN 1328 * ACKNOW will be turned on later. 1329 */ 1330 if (DELAY_ACK(tp) && tlen != 0) 1331 callout_reset(tp->tt_delack, tcp_delacktime, 1332 tcp_timer_delack, tp); 1333 else 1334 tp->t_flags |= TF_ACKNOW; 1335 /* 1336 * Received <SYN,ACK> in SYN_SENT[*] state. 1337 * Transitions: 1338 * SYN_SENT --> ESTABLISHED 1339 * SYN_SENT* --> FIN_WAIT_1 1340 */ 1341 tp->t_starttime = ticks; 1342 if (tp->t_flags & TF_NEEDFIN) { 1343 tp->t_state = TCPS_FIN_WAIT_1; 1344 tp->t_flags &= ~TF_NEEDFIN; 1345 thflags &= ~TH_SYN; 1346 } else { 1347 tp->t_state = TCPS_ESTABLISHED; 1348 callout_reset(tp->tt_keep, tcp_keepidle, 1349 tcp_timer_keep, tp); 1350 } 1351 } else { 1352 /* 1353 * Received initial SYN in SYN-SENT[*] state => simul- 1354 * taneous open. If segment contains CC option and there is 1355 * a cached CC, apply TAO test; if it succeeds, connection is 1356 * half-synchronized. Otherwise, do 3-way handshake: 1357 * SYN-SENT -> SYN-RECEIVED 1358 * SYN-SENT* -> SYN-RECEIVED* 1359 * If there was no CC option, clear cached CC value. 1360 */ 1361 tp->t_flags |= TF_ACKNOW; 1362 callout_stop(tp->tt_rexmt); 1363 if (to.to_flag & TOF_CC) { 1364 if (taop->tao_cc != 0 && 1365 CC_GT(to.to_cc, taop->tao_cc)) { 1366 /* 1367 * update cache and make transition: 1368 * SYN-SENT -> ESTABLISHED* 1369 * SYN-SENT* -> FIN-WAIT-1* 1370 */ 1371 taop->tao_cc = to.to_cc; 1372 tp->t_starttime = ticks; 1373 if (tp->t_flags & TF_NEEDFIN) { 1374 tp->t_state = TCPS_FIN_WAIT_1; 1375 tp->t_flags &= ~TF_NEEDFIN; 1376 } else { 1377 tp->t_state = TCPS_ESTABLISHED; 1378 callout_reset(tp->tt_keep, 1379 tcp_keepidle, 1380 tcp_timer_keep, 1381 tp); 1382 } 1383 tp->t_flags |= TF_NEEDSYN; 1384 } else 1385 tp->t_state = TCPS_SYN_RECEIVED; 1386 } else { 1387 /* CC.NEW or no option => invalidate cache */ 1388 taop->tao_cc = 0; 1389 tp->t_state = TCPS_SYN_RECEIVED; 1390 } 1391 } 1392 1393trimthenstep6: 1394 /* 1395 * Advance th->th_seq to correspond to first data byte. 1396 * If data, trim to stay within window, 1397 * dropping FIN if necessary. 1398 */ 1399 th->th_seq++; 1400 if (tlen > tp->rcv_wnd) { 1401 todrop = tlen - tp->rcv_wnd; 1402 m_adj(m, -todrop); 1403 tlen = tp->rcv_wnd; 1404 thflags &= ~TH_FIN; 1405 tcpstat.tcps_rcvpackafterwin++; 1406 tcpstat.tcps_rcvbyteafterwin += todrop; 1407 } 1408 tp->snd_wl1 = th->th_seq - 1; 1409 tp->rcv_up = th->th_seq; 1410 /* 1411 * Client side of transaction: already sent SYN and data. 1412 * If the remote host used T/TCP to validate the SYN, 1413 * our data will be ACK'd; if so, enter normal data segment 1414 * processing in the middle of step 5, ack processing. 1415 * Otherwise, goto step 6. 1416 */ 1417 if (thflags & TH_ACK) 1418 goto process_ACK; 1419 goto step6; 1420 /* 1421 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1422 * if segment contains a SYN and CC [not CC.NEW] option: 1423 * if state == TIME_WAIT and connection duration > MSL, 1424 * drop packet and send RST; 1425 * 1426 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1427 * ack the FIN (and data) in retransmission queue. 1428 * Complete close and delete TCPCB. Then reprocess 1429 * segment, hoping to find new TCPCB in LISTEN state; 1430 * 1431 * else must be old SYN; drop it. 1432 * else do normal processing. 1433 */ 1434 case TCPS_LAST_ACK: 1435 case TCPS_CLOSING: 1436 case TCPS_TIME_WAIT: 1437 if ((thflags & TH_SYN) && 1438 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 1439 if (tp->t_state == TCPS_TIME_WAIT && 1440 (ticks - tp->t_starttime) > tcp_msl) { 1441 rstreason = BANDLIM_UNLIMITED; 1442 goto dropwithreset; 1443 } 1444 if (CC_GT(to.to_cc, tp->cc_recv)) { 1445 tp = tcp_close(tp); 1446 goto findpcb; 1447 } 1448 else 1449 goto drop; 1450 } 1451 break; /* continue normal processing */ 1452 } 1453 1454 /* 1455 * States other than LISTEN or SYN_SENT. 1456 * First check the RST flag and sequence number since reset segments 1457 * are exempt from the timestamp and connection count tests. This 1458 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1459 * below which allowed reset segments in half the sequence space 1460 * to fall though and be processed (which gives forged reset 1461 * segments with a random sequence number a 50 percent chance of 1462 * killing a connection). 1463 * Then check timestamp, if present. 1464 * Then check the connection count, if present. 1465 * Then check that at least some bytes of segment are within 1466 * receive window. If segment begins before rcv_nxt, 1467 * drop leading data (and SYN); if nothing left, just ack. 1468 * 1469 * 1470 * If the RST bit is set, check the sequence number to see 1471 * if this is a valid reset segment. 1472 * RFC 793 page 37: 1473 * In all states except SYN-SENT, all reset (RST) segments 1474 * are validated by checking their SEQ-fields. A reset is 1475 * valid if its sequence number is in the window. 1476 * Note: this does not take into account delayed ACKs, so 1477 * we should test against last_ack_sent instead of rcv_nxt. 1478 * The sequence number in the reset segment is normally an 1479 * echo of our outgoing acknowlegement numbers, but some hosts 1480 * send a reset with the sequence number at the rightmost edge 1481 * of our receive window, and we have to handle this case. 1482 * If we have multiple segments in flight, the intial reset 1483 * segment sequence numbers will be to the left of last_ack_sent, 1484 * but they will eventually catch up. 1485 * In any case, it never made sense to trim reset segments to 1486 * fit the receive window since RFC 1122 says: 1487 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1488 * 1489 * A TCP SHOULD allow a received RST segment to include data. 1490 * 1491 * DISCUSSION 1492 * It has been suggested that a RST segment could contain 1493 * ASCII text that encoded and explained the cause of the 1494 * RST. No standard has yet been established for such 1495 * data. 1496 * 1497 * If the reset segment passes the sequence number test examine 1498 * the state: 1499 * SYN_RECEIVED STATE: 1500 * If passive open, return to LISTEN state. 1501 * If active open, inform user that connection was refused. 1502 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1503 * Inform user that connection was reset, and close tcb. 1504 * CLOSING, LAST_ACK STATES: 1505 * Close the tcb. 1506 * TIME_WAIT STATE: 1507 * Drop the segment - see Stevens, vol. 2, p. 964 and 1508 * RFC 1337. 1509 */ 1510 if (thflags & TH_RST) { 1511 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1512 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1513 switch (tp->t_state) { 1514 1515 case TCPS_SYN_RECEIVED: 1516 so->so_error = ECONNREFUSED; 1517 goto close; 1518 1519 case TCPS_ESTABLISHED: 1520 case TCPS_FIN_WAIT_1: 1521 case TCPS_FIN_WAIT_2: 1522 case TCPS_CLOSE_WAIT: 1523 so->so_error = ECONNRESET; 1524 close: 1525 tp->t_state = TCPS_CLOSED; 1526 tcpstat.tcps_drops++; 1527 tp = tcp_close(tp); 1528 break; 1529 1530 case TCPS_CLOSING: 1531 case TCPS_LAST_ACK: 1532 tp = tcp_close(tp); 1533 break; 1534 1535 case TCPS_TIME_WAIT: 1536 break; 1537 } 1538 } 1539 goto drop; 1540 } 1541 1542 /* 1543 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1544 * and it's less than ts_recent, drop it. 1545 */ 1546 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1547 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1548 1549 /* Check to see if ts_recent is over 24 days old. */ 1550 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1551 /* 1552 * Invalidate ts_recent. If this segment updates 1553 * ts_recent, the age will be reset later and ts_recent 1554 * will get a valid value. If it does not, setting 1555 * ts_recent to zero will at least satisfy the 1556 * requirement that zero be placed in the timestamp 1557 * echo reply when ts_recent isn't valid. The 1558 * age isn't reset until we get a valid ts_recent 1559 * because we don't want out-of-order segments to be 1560 * dropped when ts_recent is old. 1561 */ 1562 tp->ts_recent = 0; 1563 } else { 1564 tcpstat.tcps_rcvduppack++; 1565 tcpstat.tcps_rcvdupbyte += tlen; 1566 tcpstat.tcps_pawsdrop++; 1567 goto dropafterack; 1568 } 1569 } 1570 1571 /* 1572 * T/TCP mechanism 1573 * If T/TCP was negotiated and the segment doesn't have CC, 1574 * or if its CC is wrong then drop the segment. 1575 * RST segments do not have to comply with this. 1576 */ 1577 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1578 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1579 goto dropafterack; 1580 1581 /* 1582 * In the SYN-RECEIVED state, validate that the packet belongs to 1583 * this connection before trimming the data to fit the receive 1584 * window. Check the sequence number versus IRS since we know 1585 * the sequence numbers haven't wrapped. This is a partial fix 1586 * for the "LAND" DoS attack. 1587 */ 1588 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1589 rstreason = BANDLIM_RST_OPENPORT; 1590 goto dropwithreset; 1591 } 1592 1593 todrop = tp->rcv_nxt - th->th_seq; 1594 if (todrop > 0) { 1595 if (thflags & TH_SYN) { 1596 thflags &= ~TH_SYN; 1597 th->th_seq++; 1598 if (th->th_urp > 1) 1599 th->th_urp--; 1600 else 1601 thflags &= ~TH_URG; 1602 todrop--; 1603 } 1604 /* 1605 * Following if statement from Stevens, vol. 2, p. 960. 1606 */ 1607 if (todrop > tlen 1608 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1609 /* 1610 * Any valid FIN must be to the left of the window. 1611 * At this point the FIN must be a duplicate or out 1612 * of sequence; drop it. 1613 */ 1614 thflags &= ~TH_FIN; 1615 1616 /* 1617 * Send an ACK to resynchronize and drop any data. 1618 * But keep on processing for RST or ACK. 1619 */ 1620 tp->t_flags |= TF_ACKNOW; 1621 todrop = tlen; 1622 tcpstat.tcps_rcvduppack++; 1623 tcpstat.tcps_rcvdupbyte += todrop; 1624 } else { 1625 tcpstat.tcps_rcvpartduppack++; 1626 tcpstat.tcps_rcvpartdupbyte += todrop; 1627 } 1628 drop_hdrlen += todrop; /* drop from the top afterwards */ 1629 th->th_seq += todrop; 1630 tlen -= todrop; 1631 if (th->th_urp > todrop) 1632 th->th_urp -= todrop; 1633 else { 1634 thflags &= ~TH_URG; 1635 th->th_urp = 0; 1636 } 1637 } 1638 1639 /* 1640 * If new data are received on a connection after the 1641 * user processes are gone, then RST the other end. 1642 */ 1643 if ((so->so_state & SS_NOFDREF) && 1644 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1645 tp = tcp_close(tp); 1646 tcpstat.tcps_rcvafterclose++; 1647 rstreason = BANDLIM_UNLIMITED; 1648 goto dropwithreset; 1649 } 1650 1651 /* 1652 * If segment ends after window, drop trailing data 1653 * (and PUSH and FIN); if nothing left, just ACK. 1654 */ 1655 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1656 if (todrop > 0) { 1657 tcpstat.tcps_rcvpackafterwin++; 1658 if (todrop >= tlen) { 1659 tcpstat.tcps_rcvbyteafterwin += tlen; 1660 /* 1661 * If a new connection request is received 1662 * while in TIME_WAIT, drop the old connection 1663 * and start over if the sequence numbers 1664 * are above the previous ones. 1665 */ 1666 if (thflags & TH_SYN && 1667 tp->t_state == TCPS_TIME_WAIT && 1668 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1669 iss = tcp_new_isn(tp); 1670 tp = tcp_close(tp); 1671 goto findpcb; 1672 } 1673 /* 1674 * If window is closed can only take segments at 1675 * window edge, and have to drop data and PUSH from 1676 * incoming segments. Continue processing, but 1677 * remember to ack. Otherwise, drop segment 1678 * and ack. 1679 */ 1680 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1681 tp->t_flags |= TF_ACKNOW; 1682 tcpstat.tcps_rcvwinprobe++; 1683 } else 1684 goto dropafterack; 1685 } else 1686 tcpstat.tcps_rcvbyteafterwin += todrop; 1687 m_adj(m, -todrop); 1688 tlen -= todrop; 1689 thflags &= ~(TH_PUSH|TH_FIN); 1690 } 1691 1692 /* 1693 * If last ACK falls within this segment's sequence numbers, 1694 * record its timestamp. 1695 * NOTE that the test is modified according to the latest 1696 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1697 */ 1698 if ((to.to_flag & TOF_TS) != 0 && 1699 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1700 tp->ts_recent_age = ticks; 1701 tp->ts_recent = to.to_tsval; 1702 } 1703 1704 /* 1705 * If a SYN is in the window, then this is an 1706 * error and we send an RST and drop the connection. 1707 */ 1708 if (thflags & TH_SYN) { 1709 tp = tcp_drop(tp, ECONNRESET); 1710 rstreason = BANDLIM_UNLIMITED; 1711 goto dropwithreset; 1712 } 1713 1714 /* 1715 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1716 * flag is on (half-synchronized state), then queue data for 1717 * later processing; else drop segment and return. 1718 */ 1719 if ((thflags & TH_ACK) == 0) { 1720 if (tp->t_state == TCPS_SYN_RECEIVED || 1721 (tp->t_flags & TF_NEEDSYN)) 1722 goto step6; 1723 else 1724 goto drop; 1725 } 1726 1727 /* 1728 * Ack processing. 1729 */ 1730 switch (tp->t_state) { 1731 1732 /* 1733 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1734 * ESTABLISHED state and continue processing. 1735 * The ACK was checked above. 1736 */ 1737 case TCPS_SYN_RECEIVED: 1738 1739 tcpstat.tcps_connects++; 1740 soisconnected(so); 1741 /* Do window scaling? */ 1742 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1743 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1744 tp->snd_scale = tp->requested_s_scale; 1745 tp->rcv_scale = tp->request_r_scale; 1746 } 1747 /* 1748 * Upon successful completion of 3-way handshake, 1749 * update cache.CC if it was undefined, pass any queued 1750 * data to the user, and advance state appropriately. 1751 */ 1752 if ((taop = tcp_gettaocache(inp)) != NULL && 1753 taop->tao_cc == 0) 1754 taop->tao_cc = tp->cc_recv; 1755 1756 /* 1757 * Make transitions: 1758 * SYN-RECEIVED -> ESTABLISHED 1759 * SYN-RECEIVED* -> FIN-WAIT-1 1760 */ 1761 tp->t_starttime = ticks; 1762 if (tp->t_flags & TF_NEEDFIN) { 1763 tp->t_state = TCPS_FIN_WAIT_1; 1764 tp->t_flags &= ~TF_NEEDFIN; 1765 } else { 1766 tp->t_state = TCPS_ESTABLISHED; 1767 callout_reset(tp->tt_keep, tcp_keepidle, 1768 tcp_timer_keep, tp); 1769 } 1770 /* 1771 * If segment contains data or ACK, will call tcp_reass() 1772 * later; if not, do so now to pass queued data to user. 1773 */ 1774 if (tlen == 0 && (thflags & TH_FIN) == 0) 1775 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1776 (struct mbuf *)0); 1777 tp->snd_wl1 = th->th_seq - 1; 1778 /* fall into ... */ 1779 1780 /* 1781 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1782 * ACKs. If the ack is in the range 1783 * tp->snd_una < th->th_ack <= tp->snd_max 1784 * then advance tp->snd_una to th->th_ack and drop 1785 * data from the retransmission queue. If this ACK reflects 1786 * more up to date window information we update our window information. 1787 */ 1788 case TCPS_ESTABLISHED: 1789 case TCPS_FIN_WAIT_1: 1790 case TCPS_FIN_WAIT_2: 1791 case TCPS_CLOSE_WAIT: 1792 case TCPS_CLOSING: 1793 case TCPS_LAST_ACK: 1794 case TCPS_TIME_WAIT: 1795 1796 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1797 if (tlen == 0 && tiwin == tp->snd_wnd) { 1798 tcpstat.tcps_rcvdupack++; 1799 /* 1800 * If we have outstanding data (other than 1801 * a window probe), this is a completely 1802 * duplicate ack (ie, window info didn't 1803 * change), the ack is the biggest we've 1804 * seen and we've seen exactly our rexmt 1805 * threshhold of them, assume a packet 1806 * has been dropped and retransmit it. 1807 * Kludge snd_nxt & the congestion 1808 * window so we send only this one 1809 * packet. 1810 * 1811 * We know we're losing at the current 1812 * window size so do congestion avoidance 1813 * (set ssthresh to half the current window 1814 * and pull our congestion window back to 1815 * the new ssthresh). 1816 * 1817 * Dup acks mean that packets have left the 1818 * network (they're now cached at the receiver) 1819 * so bump cwnd by the amount in the receiver 1820 * to keep a constant cwnd packets in the 1821 * network. 1822 */ 1823 if (!callout_active(tp->tt_rexmt) || 1824 th->th_ack != tp->snd_una) 1825 tp->t_dupacks = 0; 1826 else if (++tp->t_dupacks == tcprexmtthresh) { 1827 tcp_seq onxt = tp->snd_nxt; 1828 u_int win = 1829 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1830 tp->t_maxseg; 1831 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1832 tp->snd_recover)) { 1833 /* False retransmit, should not 1834 * cut window 1835 */ 1836 tp->snd_cwnd += tp->t_maxseg; 1837 tp->t_dupacks = 0; 1838 (void) tcp_output(tp); 1839 goto drop; 1840 } 1841 if (win < 2) 1842 win = 2; 1843 tp->snd_ssthresh = win * tp->t_maxseg; 1844 tp->snd_recover = tp->snd_max; 1845 callout_stop(tp->tt_rexmt); 1846 tp->t_rtttime = 0; 1847 tp->snd_nxt = th->th_ack; 1848 tp->snd_cwnd = tp->t_maxseg; 1849 (void) tcp_output(tp); 1850 tp->snd_cwnd = tp->snd_ssthresh + 1851 tp->t_maxseg * tp->t_dupacks; 1852 if (SEQ_GT(onxt, tp->snd_nxt)) 1853 tp->snd_nxt = onxt; 1854 goto drop; 1855 } else if (tp->t_dupacks > tcprexmtthresh) { 1856 tp->snd_cwnd += tp->t_maxseg; 1857 (void) tcp_output(tp); 1858 goto drop; 1859 } 1860 } else 1861 tp->t_dupacks = 0; 1862 break; 1863 } 1864 /* 1865 * If the congestion window was inflated to account 1866 * for the other side's cached packets, retract it. 1867 */ 1868 if (tcp_do_newreno == 0) { 1869 if (tp->t_dupacks >= tcprexmtthresh && 1870 tp->snd_cwnd > tp->snd_ssthresh) 1871 tp->snd_cwnd = tp->snd_ssthresh; 1872 tp->t_dupacks = 0; 1873 } else if (tp->t_dupacks >= tcprexmtthresh && 1874 !tcp_newreno(tp, th)) { 1875 /* 1876 * Window inflation should have left us with approx. 1877 * snd_ssthresh outstanding data. But in case we 1878 * would be inclined to send a burst, better to do 1879 * it via the slow start mechanism. 1880 */ 1881 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1882 tp->snd_cwnd = 1883 tp->snd_max - th->th_ack + tp->t_maxseg; 1884 else 1885 tp->snd_cwnd = tp->snd_ssthresh; 1886 tp->t_dupacks = 0; 1887 } 1888 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1889 tcpstat.tcps_rcvacktoomuch++; 1890 goto dropafterack; 1891 } 1892 /* 1893 * If we reach this point, ACK is not a duplicate, 1894 * i.e., it ACKs something we sent. 1895 */ 1896 if (tp->t_flags & TF_NEEDSYN) { 1897 /* 1898 * T/TCP: Connection was half-synchronized, and our 1899 * SYN has been ACK'd (so connection is now fully 1900 * synchronized). Go to non-starred state, 1901 * increment snd_una for ACK of SYN, and check if 1902 * we can do window scaling. 1903 */ 1904 tp->t_flags &= ~TF_NEEDSYN; 1905 tp->snd_una++; 1906 /* Do window scaling? */ 1907 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1908 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1909 tp->snd_scale = tp->requested_s_scale; 1910 tp->rcv_scale = tp->request_r_scale; 1911 } 1912 } 1913 1914process_ACK: 1915 acked = th->th_ack - tp->snd_una; 1916 tcpstat.tcps_rcvackpack++; 1917 tcpstat.tcps_rcvackbyte += acked; 1918 1919 /* 1920 * If we just performed our first retransmit, and the ACK 1921 * arrives within our recovery window, then it was a mistake 1922 * to do the retransmit in the first place. Recover our 1923 * original cwnd and ssthresh, and proceed to transmit where 1924 * we left off. 1925 */ 1926 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1927 tp->snd_cwnd = tp->snd_cwnd_prev; 1928 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1929 tp->snd_nxt = tp->snd_max; 1930 tp->t_badrxtwin = 0; /* XXX probably not required */ 1931 } 1932 1933 /* 1934 * If we have a timestamp reply, update smoothed 1935 * round trip time. If no timestamp is present but 1936 * transmit timer is running and timed sequence 1937 * number was acked, update smoothed round trip time. 1938 * Since we now have an rtt measurement, cancel the 1939 * timer backoff (cf., Phil Karn's retransmit alg.). 1940 * Recompute the initial retransmit timer. 1941 */ 1942 if (to.to_flag & TOF_TS) 1943 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1944 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1945 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1946 1947 /* 1948 * If all outstanding data is acked, stop retransmit 1949 * timer and remember to restart (more output or persist). 1950 * If there is more data to be acked, restart retransmit 1951 * timer, using current (possibly backed-off) value. 1952 */ 1953 if (th->th_ack == tp->snd_max) { 1954 callout_stop(tp->tt_rexmt); 1955 needoutput = 1; 1956 } else if (!callout_active(tp->tt_persist)) 1957 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1958 tcp_timer_rexmt, tp); 1959 1960 /* 1961 * If no data (only SYN) was ACK'd, 1962 * skip rest of ACK processing. 1963 */ 1964 if (acked == 0) 1965 goto step6; 1966 1967 /* 1968 * When new data is acked, open the congestion window. 1969 * If the window gives us less than ssthresh packets 1970 * in flight, open exponentially (maxseg per packet). 1971 * Otherwise open linearly: maxseg per window 1972 * (maxseg^2 / cwnd per packet). 1973 */ 1974 { 1975 register u_int cw = tp->snd_cwnd; 1976 register u_int incr = tp->t_maxseg; 1977 1978 if (cw > tp->snd_ssthresh) 1979 incr = incr * incr / cw; 1980 /* 1981 * If t_dupacks != 0 here, it indicates that we are still 1982 * in NewReno fast recovery mode, so we leave the congestion 1983 * window alone. 1984 */ 1985 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1986 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1987 } 1988 if (acked > so->so_snd.sb_cc) { 1989 tp->snd_wnd -= so->so_snd.sb_cc; 1990 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1991 ourfinisacked = 1; 1992 } else { 1993 sbdrop(&so->so_snd, acked); 1994 tp->snd_wnd -= acked; 1995 ourfinisacked = 0; 1996 } 1997 sowwakeup(so); 1998 tp->snd_una = th->th_ack; 1999 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2000 tp->snd_nxt = tp->snd_una; 2001 2002 switch (tp->t_state) { 2003 2004 /* 2005 * In FIN_WAIT_1 STATE in addition to the processing 2006 * for the ESTABLISHED state if our FIN is now acknowledged 2007 * then enter FIN_WAIT_2. 2008 */ 2009 case TCPS_FIN_WAIT_1: 2010 if (ourfinisacked) { 2011 /* 2012 * If we can't receive any more 2013 * data, then closing user can proceed. 2014 * Starting the timer is contrary to the 2015 * specification, but if we don't get a FIN 2016 * we'll hang forever. 2017 */ 2018 if (so->so_state & SS_CANTRCVMORE) { 2019 soisdisconnected(so); 2020 callout_reset(tp->tt_2msl, tcp_maxidle, 2021 tcp_timer_2msl, tp); 2022 } 2023 tp->t_state = TCPS_FIN_WAIT_2; 2024 } 2025 break; 2026 2027 /* 2028 * In CLOSING STATE in addition to the processing for 2029 * the ESTABLISHED state if the ACK acknowledges our FIN 2030 * then enter the TIME-WAIT state, otherwise ignore 2031 * the segment. 2032 */ 2033 case TCPS_CLOSING: 2034 if (ourfinisacked) { 2035 tp->t_state = TCPS_TIME_WAIT; 2036 tcp_canceltimers(tp); 2037 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2038 if (tp->cc_recv != 0 && 2039 (ticks - tp->t_starttime) < tcp_msl) 2040 callout_reset(tp->tt_2msl, 2041 tp->t_rxtcur * 2042 TCPTV_TWTRUNC, 2043 tcp_timer_2msl, tp); 2044 else 2045 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2046 tcp_timer_2msl, tp); 2047 soisdisconnected(so); 2048 } 2049 break; 2050 2051 /* 2052 * In LAST_ACK, we may still be waiting for data to drain 2053 * and/or to be acked, as well as for the ack of our FIN. 2054 * If our FIN is now acknowledged, delete the TCB, 2055 * enter the closed state and return. 2056 */ 2057 case TCPS_LAST_ACK: 2058 if (ourfinisacked) { 2059 tp = tcp_close(tp); 2060 goto drop; 2061 } 2062 break; 2063 2064 /* 2065 * In TIME_WAIT state the only thing that should arrive 2066 * is a retransmission of the remote FIN. Acknowledge 2067 * it and restart the finack timer. 2068 */ 2069 case TCPS_TIME_WAIT: 2070 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2071 tcp_timer_2msl, tp); 2072 goto dropafterack; 2073 } 2074 } 2075 2076step6: 2077 /* 2078 * Update window information. 2079 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2080 */ 2081 if ((thflags & TH_ACK) && 2082 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2083 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2084 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2085 /* keep track of pure window updates */ 2086 if (tlen == 0 && 2087 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2088 tcpstat.tcps_rcvwinupd++; 2089 tp->snd_wnd = tiwin; 2090 tp->snd_wl1 = th->th_seq; 2091 tp->snd_wl2 = th->th_ack; 2092 if (tp->snd_wnd > tp->max_sndwnd) 2093 tp->max_sndwnd = tp->snd_wnd; 2094 needoutput = 1; 2095 } 2096 2097 /* 2098 * Process segments with URG. 2099 */ 2100 if ((thflags & TH_URG) && th->th_urp && 2101 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2102 /* 2103 * This is a kludge, but if we receive and accept 2104 * random urgent pointers, we'll crash in 2105 * soreceive. It's hard to imagine someone 2106 * actually wanting to send this much urgent data. 2107 */ 2108 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2109 th->th_urp = 0; /* XXX */ 2110 thflags &= ~TH_URG; /* XXX */ 2111 goto dodata; /* XXX */ 2112 } 2113 /* 2114 * If this segment advances the known urgent pointer, 2115 * then mark the data stream. This should not happen 2116 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2117 * a FIN has been received from the remote side. 2118 * In these states we ignore the URG. 2119 * 2120 * According to RFC961 (Assigned Protocols), 2121 * the urgent pointer points to the last octet 2122 * of urgent data. We continue, however, 2123 * to consider it to indicate the first octet 2124 * of data past the urgent section as the original 2125 * spec states (in one of two places). 2126 */ 2127 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2128 tp->rcv_up = th->th_seq + th->th_urp; 2129 so->so_oobmark = so->so_rcv.sb_cc + 2130 (tp->rcv_up - tp->rcv_nxt) - 1; 2131 if (so->so_oobmark == 0) 2132 so->so_state |= SS_RCVATMARK; 2133 sohasoutofband(so); 2134 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2135 } 2136 /* 2137 * Remove out of band data so doesn't get presented to user. 2138 * This can happen independent of advancing the URG pointer, 2139 * but if two URG's are pending at once, some out-of-band 2140 * data may creep in... ick. 2141 */ 2142 if (th->th_urp <= (u_long)tlen 2143#ifdef SO_OOBINLINE 2144 && (so->so_options & SO_OOBINLINE) == 0 2145#endif 2146 ) 2147 tcp_pulloutofband(so, th, m, 2148 drop_hdrlen); /* hdr drop is delayed */ 2149 } else 2150 /* 2151 * If no out of band data is expected, 2152 * pull receive urgent pointer along 2153 * with the receive window. 2154 */ 2155 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2156 tp->rcv_up = tp->rcv_nxt; 2157dodata: /* XXX */ 2158 2159 /* 2160 * Process the segment text, merging it into the TCP sequencing queue, 2161 * and arranging for acknowledgment of receipt if necessary. 2162 * This process logically involves adjusting tp->rcv_wnd as data 2163 * is presented to the user (this happens in tcp_usrreq.c, 2164 * case PRU_RCVD). If a FIN has already been received on this 2165 * connection then we just ignore the text. 2166 */ 2167 if ((tlen || (thflags&TH_FIN)) && 2168 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2169 m_adj(m, drop_hdrlen); /* delayed header drop */ 2170 /* 2171 * Insert segment which inludes th into reassembly queue of tcp with 2172 * control block tp. Return TH_FIN if reassembly now includes 2173 * a segment with FIN. This handle the common case inline (segment 2174 * is the next to be received on an established connection, and the 2175 * queue is empty), avoiding linkage into and removal from the queue 2176 * and repetition of various conversions. 2177 * Set DELACK for segments received in order, but ack immediately 2178 * when segments are out of order (so fast retransmit can work). 2179 */ 2180 if (th->th_seq == tp->rcv_nxt && 2181 LIST_EMPTY(&tp->t_segq) && 2182 TCPS_HAVEESTABLISHED(tp->t_state)) { 2183 if (DELAY_ACK(tp)) 2184 callout_reset(tp->tt_delack, tcp_delacktime, 2185 tcp_timer_delack, tp); 2186 else 2187 tp->t_flags |= TF_ACKNOW; 2188 tp->rcv_nxt += tlen; 2189 thflags = th->th_flags & TH_FIN; 2190 tcpstat.tcps_rcvpack++; 2191 tcpstat.tcps_rcvbyte += tlen; 2192 ND6_HINT(tp); 2193 sbappend(&so->so_rcv, m); 2194 sorwakeup(so); 2195 } else { 2196 thflags = tcp_reass(tp, th, &tlen, m); 2197 tp->t_flags |= TF_ACKNOW; 2198 } 2199 2200 /* 2201 * Note the amount of data that peer has sent into 2202 * our window, in order to estimate the sender's 2203 * buffer size. 2204 */ 2205 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2206 } else { 2207 m_freem(m); 2208 thflags &= ~TH_FIN; 2209 } 2210 2211 /* 2212 * If FIN is received ACK the FIN and let the user know 2213 * that the connection is closing. 2214 */ 2215 if (thflags & TH_FIN) { 2216 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2217 socantrcvmore(so); 2218 /* 2219 * If connection is half-synchronized 2220 * (ie NEEDSYN flag on) then delay ACK, 2221 * so it may be piggybacked when SYN is sent. 2222 * Otherwise, since we received a FIN then no 2223 * more input can be expected, send ACK now. 2224 */ 2225 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2226 callout_reset(tp->tt_delack, tcp_delacktime, 2227 tcp_timer_delack, tp); 2228 else 2229 tp->t_flags |= TF_ACKNOW; 2230 tp->rcv_nxt++; 2231 } 2232 switch (tp->t_state) { 2233 2234 /* 2235 * In SYN_RECEIVED and ESTABLISHED STATES 2236 * enter the CLOSE_WAIT state. 2237 */ 2238 case TCPS_SYN_RECEIVED: 2239 tp->t_starttime = ticks; 2240 /*FALLTHROUGH*/ 2241 case TCPS_ESTABLISHED: 2242 tp->t_state = TCPS_CLOSE_WAIT; 2243 break; 2244 2245 /* 2246 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2247 * enter the CLOSING state. 2248 */ 2249 case TCPS_FIN_WAIT_1: 2250 tp->t_state = TCPS_CLOSING; 2251 break; 2252 2253 /* 2254 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2255 * starting the time-wait timer, turning off the other 2256 * standard timers. 2257 */ 2258 case TCPS_FIN_WAIT_2: 2259 tp->t_state = TCPS_TIME_WAIT; 2260 tcp_canceltimers(tp); 2261 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2262 if (tp->cc_recv != 0 && 2263 (ticks - tp->t_starttime) < tcp_msl) { 2264 callout_reset(tp->tt_2msl, 2265 tp->t_rxtcur * TCPTV_TWTRUNC, 2266 tcp_timer_2msl, tp); 2267 /* For transaction client, force ACK now. */ 2268 tp->t_flags |= TF_ACKNOW; 2269 } 2270 else 2271 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2272 tcp_timer_2msl, tp); 2273 soisdisconnected(so); 2274 break; 2275 2276 /* 2277 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2278 */ 2279 case TCPS_TIME_WAIT: 2280 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2281 tcp_timer_2msl, tp); 2282 break; 2283 } 2284 } 2285#ifdef TCPDEBUG 2286 if (so->so_options & SO_DEBUG) 2287 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2288 &tcp_savetcp, 0); 2289#endif 2290 2291 /* 2292 * Return any desired output. 2293 */ 2294 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2295 (void) tcp_output(tp); 2296 return; 2297 2298dropafterack: 2299 /* 2300 * Generate an ACK dropping incoming segment if it occupies 2301 * sequence space, where the ACK reflects our state. 2302 * 2303 * We can now skip the test for the RST flag since all 2304 * paths to this code happen after packets containing 2305 * RST have been dropped. 2306 * 2307 * In the SYN-RECEIVED state, don't send an ACK unless the 2308 * segment we received passes the SYN-RECEIVED ACK test. 2309 * If it fails send a RST. This breaks the loop in the 2310 * "LAND" DoS attack, and also prevents an ACK storm 2311 * between two listening ports that have been sent forged 2312 * SYN segments, each with the source address of the other. 2313 */ 2314 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2315 (SEQ_GT(tp->snd_una, th->th_ack) || 2316 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2317 rstreason = BANDLIM_RST_OPENPORT; 2318 goto dropwithreset; 2319 } 2320#ifdef TCPDEBUG 2321 if (so->so_options & SO_DEBUG) 2322 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2323 &tcp_savetcp, 0); 2324#endif 2325 m_freem(m); 2326 tp->t_flags |= TF_ACKNOW; 2327 (void) tcp_output(tp); 2328 return; 2329 2330dropwithreset: 2331 /* 2332 * Generate a RST, dropping incoming segment. 2333 * Make ACK acceptable to originator of segment. 2334 * Don't bother to respond if destination was broadcast/multicast. 2335 */ 2336 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2337 goto drop; 2338#ifdef INET6 2339 if (isipv6) { 2340 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2341 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2342 goto drop; 2343 } else 2344#endif /* INET6 */ 2345 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2346 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2347 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 2348 goto drop; 2349 /* IPv6 anycast check is done at tcp6_input() */ 2350 2351 /* 2352 * Perform bandwidth limiting. 2353 */ 2354 if (badport_bandlim(rstreason) < 0) 2355 goto drop; 2356 2357#ifdef TCPDEBUG 2358 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2359 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2360 &tcp_savetcp, 0); 2361#endif 2362 if (thflags & TH_ACK) 2363 /* mtod() below is safe as long as hdr dropping is delayed */ 2364 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2365 TH_RST); 2366 else { 2367 if (thflags & TH_SYN) 2368 tlen++; 2369 /* mtod() below is safe as long as hdr dropping is delayed */ 2370 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2371 (tcp_seq)0, TH_RST|TH_ACK); 2372 } 2373 /* destroy temporarily created socket */ 2374 if (dropsocket) 2375 (void) soabort(so); 2376 return; 2377 2378drop: 2379 /* 2380 * Drop space held by incoming segment and return. 2381 */ 2382#ifdef TCPDEBUG 2383 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2384 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2385 &tcp_savetcp, 0); 2386#endif 2387 m_freem(m); 2388 /* destroy temporarily created socket */ 2389 if (dropsocket) 2390 (void) soabort(so); 2391 return; 2392} 2393 2394static void 2395tcp_dooptions(tp, cp, cnt, th, to) 2396 struct tcpcb *tp; 2397 u_char *cp; 2398 int cnt; 2399 struct tcphdr *th; 2400 struct tcpopt *to; 2401{ 2402 u_short mss = 0; 2403 int opt, optlen; 2404 2405 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2406 opt = cp[0]; 2407 if (opt == TCPOPT_EOL) 2408 break; 2409 if (opt == TCPOPT_NOP) 2410 optlen = 1; 2411 else { 2412 if (cnt < 2) 2413 break; 2414 optlen = cp[1]; 2415 if (optlen < 2 || optlen > cnt) 2416 break; 2417 } 2418 switch (opt) { 2419 2420 default: 2421 continue; 2422 2423 case TCPOPT_MAXSEG: 2424 if (optlen != TCPOLEN_MAXSEG) 2425 continue; 2426 if (!(th->th_flags & TH_SYN)) 2427 continue; 2428 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2429 NTOHS(mss); 2430 break; 2431 2432 case TCPOPT_WINDOW: 2433 if (optlen != TCPOLEN_WINDOW) 2434 continue; 2435 if (!(th->th_flags & TH_SYN)) 2436 continue; 2437 tp->t_flags |= TF_RCVD_SCALE; 2438 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2439 break; 2440 2441 case TCPOPT_TIMESTAMP: 2442 if (optlen != TCPOLEN_TIMESTAMP) 2443 continue; 2444 to->to_flag |= TOF_TS; 2445 bcopy((char *)cp + 2, 2446 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2447 NTOHL(to->to_tsval); 2448 bcopy((char *)cp + 6, 2449 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2450 NTOHL(to->to_tsecr); 2451 2452 /* 2453 * A timestamp received in a SYN makes 2454 * it ok to send timestamp requests and replies. 2455 */ 2456 if (th->th_flags & TH_SYN) { 2457 tp->t_flags |= TF_RCVD_TSTMP; 2458 tp->ts_recent = to->to_tsval; 2459 tp->ts_recent_age = ticks; 2460 } 2461 break; 2462 case TCPOPT_CC: 2463 if (optlen != TCPOLEN_CC) 2464 continue; 2465 to->to_flag |= TOF_CC; 2466 bcopy((char *)cp + 2, 2467 (char *)&to->to_cc, sizeof(to->to_cc)); 2468 NTOHL(to->to_cc); 2469 /* 2470 * A CC or CC.new option received in a SYN makes 2471 * it ok to send CC in subsequent segments. 2472 */ 2473 if (th->th_flags & TH_SYN) 2474 tp->t_flags |= TF_RCVD_CC; 2475 break; 2476 case TCPOPT_CCNEW: 2477 if (optlen != TCPOLEN_CC) 2478 continue; 2479 if (!(th->th_flags & TH_SYN)) 2480 continue; 2481 to->to_flag |= TOF_CCNEW; 2482 bcopy((char *)cp + 2, 2483 (char *)&to->to_cc, sizeof(to->to_cc)); 2484 NTOHL(to->to_cc); 2485 /* 2486 * A CC or CC.new option received in a SYN makes 2487 * it ok to send CC in subsequent segments. 2488 */ 2489 tp->t_flags |= TF_RCVD_CC; 2490 break; 2491 case TCPOPT_CCECHO: 2492 if (optlen != TCPOLEN_CC) 2493 continue; 2494 if (!(th->th_flags & TH_SYN)) 2495 continue; 2496 to->to_flag |= TOF_CCECHO; 2497 bcopy((char *)cp + 2, 2498 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2499 NTOHL(to->to_ccecho); 2500 break; 2501 } 2502 } 2503 if (th->th_flags & TH_SYN) 2504 tcp_mss(tp, mss); /* sets t_maxseg */ 2505} 2506 2507/* 2508 * Pull out of band byte out of a segment so 2509 * it doesn't appear in the user's data queue. 2510 * It is still reflected in the segment length for 2511 * sequencing purposes. 2512 */ 2513static void 2514tcp_pulloutofband(so, th, m, off) 2515 struct socket *so; 2516 struct tcphdr *th; 2517 register struct mbuf *m; 2518 int off; /* delayed to be droped hdrlen */ 2519{ 2520 int cnt = off + th->th_urp - 1; 2521 2522 while (cnt >= 0) { 2523 if (m->m_len > cnt) { 2524 char *cp = mtod(m, caddr_t) + cnt; 2525 struct tcpcb *tp = sototcpcb(so); 2526 2527 tp->t_iobc = *cp; 2528 tp->t_oobflags |= TCPOOB_HAVEDATA; 2529 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2530 m->m_len--; 2531 if (m->m_flags & M_PKTHDR) 2532 m->m_pkthdr.len--; 2533 return; 2534 } 2535 cnt -= m->m_len; 2536 m = m->m_next; 2537 if (m == 0) 2538 break; 2539 } 2540 panic("tcp_pulloutofband"); 2541} 2542 2543/* 2544 * Collect new round-trip time estimate 2545 * and update averages and current timeout. 2546 */ 2547static void 2548tcp_xmit_timer(tp, rtt) 2549 register struct tcpcb *tp; 2550 int rtt; 2551{ 2552 register int delta; 2553 2554 tcpstat.tcps_rttupdated++; 2555 tp->t_rttupdated++; 2556 if (tp->t_srtt != 0) { 2557 /* 2558 * srtt is stored as fixed point with 5 bits after the 2559 * binary point (i.e., scaled by 8). The following magic 2560 * is equivalent to the smoothing algorithm in rfc793 with 2561 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2562 * point). Adjust rtt to origin 0. 2563 */ 2564 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2565 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2566 2567 if ((tp->t_srtt += delta) <= 0) 2568 tp->t_srtt = 1; 2569 2570 /* 2571 * We accumulate a smoothed rtt variance (actually, a 2572 * smoothed mean difference), then set the retransmit 2573 * timer to smoothed rtt + 4 times the smoothed variance. 2574 * rttvar is stored as fixed point with 4 bits after the 2575 * binary point (scaled by 16). The following is 2576 * equivalent to rfc793 smoothing with an alpha of .75 2577 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2578 * rfc793's wired-in beta. 2579 */ 2580 if (delta < 0) 2581 delta = -delta; 2582 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2583 if ((tp->t_rttvar += delta) <= 0) 2584 tp->t_rttvar = 1; 2585 } else { 2586 /* 2587 * No rtt measurement yet - use the unsmoothed rtt. 2588 * Set the variance to half the rtt (so our first 2589 * retransmit happens at 3*rtt). 2590 */ 2591 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2592 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2593 } 2594 tp->t_rtttime = 0; 2595 tp->t_rxtshift = 0; 2596 2597 /* 2598 * the retransmit should happen at rtt + 4 * rttvar. 2599 * Because of the way we do the smoothing, srtt and rttvar 2600 * will each average +1/2 tick of bias. When we compute 2601 * the retransmit timer, we want 1/2 tick of rounding and 2602 * 1 extra tick because of +-1/2 tick uncertainty in the 2603 * firing of the timer. The bias will give us exactly the 2604 * 1.5 tick we need. But, because the bias is 2605 * statistical, we have to test that we don't drop below 2606 * the minimum feasible timer (which is 2 ticks). 2607 */ 2608 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2609 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2610 2611 /* 2612 * We received an ack for a packet that wasn't retransmitted; 2613 * it is probably safe to discard any error indications we've 2614 * received recently. This isn't quite right, but close enough 2615 * for now (a route might have failed after we sent a segment, 2616 * and the return path might not be symmetrical). 2617 */ 2618 tp->t_softerror = 0; 2619} 2620 2621/* 2622 * Determine a reasonable value for maxseg size. 2623 * If the route is known, check route for mtu. 2624 * If none, use an mss that can be handled on the outgoing 2625 * interface without forcing IP to fragment; if bigger than 2626 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2627 * to utilize large mbufs. If no route is found, route has no mtu, 2628 * or the destination isn't local, use a default, hopefully conservative 2629 * size (usually 512 or the default IP max size, but no more than the mtu 2630 * of the interface), as we can't discover anything about intervening 2631 * gateways or networks. We also initialize the congestion/slow start 2632 * window to be a single segment if the destination isn't local. 2633 * While looking at the routing entry, we also initialize other path-dependent 2634 * parameters from pre-set or cached values in the routing entry. 2635 * 2636 * Also take into account the space needed for options that we 2637 * send regularly. Make maxseg shorter by that amount to assure 2638 * that we can send maxseg amount of data even when the options 2639 * are present. Store the upper limit of the length of options plus 2640 * data in maxopd. 2641 * 2642 * NOTE that this routine is only called when we process an incoming 2643 * segment, for outgoing segments only tcp_mssopt is called. 2644 * 2645 * In case of T/TCP, we call this routine during implicit connection 2646 * setup as well (offer = -1), to initialize maxseg from the cached 2647 * MSS of our peer. 2648 */ 2649void 2650tcp_mss(tp, offer) 2651 struct tcpcb *tp; 2652 int offer; 2653{ 2654 register struct rtentry *rt; 2655 struct ifnet *ifp; 2656 register int rtt, mss; 2657 u_long bufsize; 2658 struct inpcb *inp; 2659 struct socket *so; 2660 struct rmxp_tao *taop; 2661 int origoffer = offer; 2662#ifdef INET6 2663 int isipv6; 2664 int min_protoh; 2665#endif 2666 2667 inp = tp->t_inpcb; 2668#ifdef INET6 2669 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2670 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2671 : sizeof (struct tcpiphdr); 2672#else 2673#define min_protoh (sizeof (struct tcpiphdr)) 2674#endif 2675#ifdef INET6 2676 if (isipv6) 2677 rt = tcp_rtlookup6(inp); 2678 else 2679#endif 2680 rt = tcp_rtlookup(inp); 2681 if (rt == NULL) { 2682 tp->t_maxopd = tp->t_maxseg = 2683#ifdef INET6 2684 isipv6 ? tcp_v6mssdflt : 2685#endif /* INET6 */ 2686 tcp_mssdflt; 2687 return; 2688 } 2689 ifp = rt->rt_ifp; 2690 so = inp->inp_socket; 2691 2692 taop = rmx_taop(rt->rt_rmx); 2693 /* 2694 * Offer == -1 means that we didn't receive SYN yet, 2695 * use cached value in that case; 2696 */ 2697 if (offer == -1) 2698 offer = taop->tao_mssopt; 2699 /* 2700 * Offer == 0 means that there was no MSS on the SYN segment, 2701 * in this case we use tcp_mssdflt. 2702 */ 2703 if (offer == 0) 2704 offer = 2705#ifdef INET6 2706 isipv6 ? tcp_v6mssdflt : 2707#endif /* INET6 */ 2708 tcp_mssdflt; 2709 else 2710 /* 2711 * Sanity check: make sure that maxopd will be large 2712 * enough to allow some data on segments even is the 2713 * all the option space is used (40bytes). Otherwise 2714 * funny things may happen in tcp_output. 2715 */ 2716 offer = max(offer, 64); 2717 taop->tao_mssopt = offer; 2718 2719 /* 2720 * While we're here, check if there's an initial rtt 2721 * or rttvar. Convert from the route-table units 2722 * to scaled multiples of the slow timeout timer. 2723 */ 2724 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2725 /* 2726 * XXX the lock bit for RTT indicates that the value 2727 * is also a minimum value; this is subject to time. 2728 */ 2729 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2730 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2731 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2732 tcpstat.tcps_usedrtt++; 2733 if (rt->rt_rmx.rmx_rttvar) { 2734 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2735 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2736 tcpstat.tcps_usedrttvar++; 2737 } else { 2738 /* default variation is +- 1 rtt */ 2739 tp->t_rttvar = 2740 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2741 } 2742 TCPT_RANGESET(tp->t_rxtcur, 2743 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2744 tp->t_rttmin, TCPTV_REXMTMAX); 2745 } 2746 /* 2747 * if there's an mtu associated with the route, use it 2748 * else, use the link mtu. 2749 */ 2750 if (rt->rt_rmx.rmx_mtu) 2751 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2752 else 2753 { 2754 mss = 2755#ifdef INET6 2756 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2757#endif 2758 ifp->if_mtu 2759#ifdef INET6 2760 ) 2761#endif 2762 - min_protoh; 2763#ifdef INET6 2764 if (isipv6) { 2765 if (!in6_localaddr(&inp->in6p_faddr)) 2766 mss = min(mss, tcp_v6mssdflt); 2767 } else 2768#endif 2769 if (!in_localaddr(inp->inp_faddr)) 2770 mss = min(mss, tcp_mssdflt); 2771 } 2772 mss = min(mss, offer); 2773 /* 2774 * maxopd stores the maximum length of data AND options 2775 * in a segment; maxseg is the amount of data in a normal 2776 * segment. We need to store this value (maxopd) apart 2777 * from maxseg, because now every segment carries options 2778 * and thus we normally have somewhat less data in segments. 2779 */ 2780 tp->t_maxopd = mss; 2781 2782 /* 2783 * In case of T/TCP, origoffer==-1 indicates, that no segments 2784 * were received yet. In this case we just guess, otherwise 2785 * we do the same as before T/TCP. 2786 */ 2787 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2788 (origoffer == -1 || 2789 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2790 mss -= TCPOLEN_TSTAMP_APPA; 2791 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2792 (origoffer == -1 || 2793 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2794 mss -= TCPOLEN_CC_APPA; 2795 2796#if (MCLBYTES & (MCLBYTES - 1)) == 0 2797 if (mss > MCLBYTES) 2798 mss &= ~(MCLBYTES-1); 2799#else 2800 if (mss > MCLBYTES) 2801 mss = mss / MCLBYTES * MCLBYTES; 2802#endif 2803 /* 2804 * If there's a pipesize, change the socket buffer 2805 * to that size. Make the socket buffers an integral 2806 * number of mss units; if the mss is larger than 2807 * the socket buffer, decrease the mss. 2808 */ 2809#ifdef RTV_SPIPE 2810 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2811#endif 2812 bufsize = so->so_snd.sb_hiwat; 2813 if (bufsize < mss) 2814 mss = bufsize; 2815 else { 2816 bufsize = roundup(bufsize, mss); 2817 if (bufsize > sb_max) 2818 bufsize = sb_max; 2819 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2820 } 2821 tp->t_maxseg = mss; 2822 2823#ifdef RTV_RPIPE 2824 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2825#endif 2826 bufsize = so->so_rcv.sb_hiwat; 2827 if (bufsize > mss) { 2828 bufsize = roundup(bufsize, mss); 2829 if (bufsize > sb_max) 2830 bufsize = sb_max; 2831 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2832 } 2833 2834 /* 2835 * Set the slow-start flight size depending on whether this 2836 * is a local network or not. 2837 */ 2838 if ( 2839#ifdef INET6 2840 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2841 (!isipv6 && 2842#endif 2843 in_localaddr(inp->inp_faddr) 2844#ifdef INET6 2845 ) 2846#endif 2847 ) 2848 tp->snd_cwnd = mss * ss_fltsz_local; 2849 else 2850 tp->snd_cwnd = mss * ss_fltsz; 2851 2852 if (rt->rt_rmx.rmx_ssthresh) { 2853 /* 2854 * There's some sort of gateway or interface 2855 * buffer limit on the path. Use this to set 2856 * the slow start threshhold, but set the 2857 * threshold to no less than 2*mss. 2858 */ 2859 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2860 tcpstat.tcps_usedssthresh++; 2861 } 2862} 2863 2864/* 2865 * Determine the MSS option to send on an outgoing SYN. 2866 */ 2867int 2868tcp_mssopt(tp) 2869 struct tcpcb *tp; 2870{ 2871 struct rtentry *rt; 2872#ifdef INET6 2873 int isipv6; 2874 int min_protoh; 2875#endif 2876 2877#ifdef INET6 2878 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2879 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2880 : sizeof (struct tcpiphdr); 2881#else 2882#define min_protoh (sizeof (struct tcpiphdr)) 2883#endif 2884#ifdef INET6 2885 if (isipv6) 2886 rt = tcp_rtlookup6(tp->t_inpcb); 2887 else 2888#endif /* INET6 */ 2889 rt = tcp_rtlookup(tp->t_inpcb); 2890 if (rt == NULL) 2891 return 2892#ifdef INET6 2893 isipv6 ? tcp_v6mssdflt : 2894#endif /* INET6 */ 2895 tcp_mssdflt; 2896 2897 return rt->rt_ifp->if_mtu - min_protoh; 2898} 2899 2900 2901/* 2902 * Checks for partial ack. If partial ack arrives, force the retransmission 2903 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2904 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2905 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2906 */ 2907static int 2908tcp_newreno(tp, th) 2909 struct tcpcb *tp; 2910 struct tcphdr *th; 2911{ 2912 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2913 tcp_seq onxt = tp->snd_nxt; 2914 u_long ocwnd = tp->snd_cwnd; 2915 2916 callout_stop(tp->tt_rexmt); 2917 tp->t_rtttime = 0; 2918 tp->snd_nxt = th->th_ack; 2919 /* 2920 * Set snd_cwnd to one segment beyond acknowledged offset 2921 * (tp->snd_una has not yet been updated when this function 2922 * is called) 2923 */ 2924 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2925 (void) tcp_output(tp); 2926 tp->snd_cwnd = ocwnd; 2927 if (SEQ_GT(onxt, tp->snd_nxt)) 2928 tp->snd_nxt = onxt; 2929 /* 2930 * Partial window deflation. Relies on fact that tp->snd_una 2931 * not updated yet. 2932 */ 2933 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2934 return (1); 2935 } 2936 return (0); 2937} 2938