tcp_input.c revision 78064
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 78064 2001-06-11 12:39:29Z ume $ 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_seq tcp_iss; 105tcp_cc tcp_ccgen; 106 107struct tcpstat tcpstat; 108SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, 109 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 110 111static int log_in_vain = 0; 112SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 113 &log_in_vain, 0, "Log all incoming TCP connections"); 114 115static int blackhole = 0; 116SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 117 &blackhole, 0, "Do not send RST when dropping refused connections"); 118 119int tcp_delack_enabled = 1; 120SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 121 &tcp_delack_enabled, 0, 122 "Delay ACK to try and piggyback it onto a data packet"); 123 124int tcp_lq_overflow = 1; 125SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW, 126 &tcp_lq_overflow, 0, 127 "Listen Queue Overflow"); 128 129#ifdef TCP_DROP_SYNFIN 130static int drop_synfin = 0; 131SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 132 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 133#endif 134 135struct inpcbhead tcb; 136#define tcb6 tcb /* for KAME src sync over BSD*'s */ 137struct inpcbinfo tcbinfo; 138 139static void tcp_dooptions __P((struct tcpcb *, 140 u_char *, int, struct tcphdr *, struct tcpopt *)); 141static void tcp_pulloutofband __P((struct socket *, 142 struct tcphdr *, struct mbuf *, int)); 143static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *, 144 struct mbuf *)); 145static void tcp_xmit_timer __P((struct tcpcb *, int)); 146static int tcp_newreno __P((struct tcpcb *, struct tcphdr *)); 147 148/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 149#ifdef INET6 150#define ND6_HINT(tp) \ 151do { \ 152 if ((tp) && (tp)->t_inpcb && \ 153 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 154 (tp)->t_inpcb->in6p_route.ro_rt) \ 155 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 156} while (0) 157#else 158#define ND6_HINT(tp) 159#endif 160 161/* 162 * Indicate whether this ack should be delayed. 163 */ 164#define DELAY_ACK(tp) \ 165 (tcp_delack_enabled && !callout_pending(tp->tt_delack)) 166 167static int 168tcp_reass(tp, th, tlenp, m) 169 register struct tcpcb *tp; 170 register struct tcphdr *th; 171 int *tlenp; 172 struct mbuf *m; 173{ 174 struct tseg_qent *q; 175 struct tseg_qent *p = NULL; 176 struct tseg_qent *nq; 177 struct tseg_qent *te; 178 struct socket *so = tp->t_inpcb->inp_socket; 179 int flags; 180 181 /* 182 * Call with th==0 after become established to 183 * force pre-ESTABLISHED data up to user socket. 184 */ 185 if (th == 0) 186 goto present; 187 188 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 189 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 190 M_NOWAIT); 191 if (te == NULL) { 192 tcpstat.tcps_rcvmemdrop++; 193 m_freem(m); 194 return (0); 195 } 196 197 /* 198 * Find a segment which begins after this one does. 199 */ 200 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 201 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 202 break; 203 p = q; 204 } 205 206 /* 207 * If there is a preceding segment, it may provide some of 208 * our data already. If so, drop the data from the incoming 209 * segment. If it provides all of our data, drop us. 210 */ 211 if (p != NULL) { 212 register int i; 213 /* conversion to int (in i) handles seq wraparound */ 214 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 215 if (i > 0) { 216 if (i >= *tlenp) { 217 tcpstat.tcps_rcvduppack++; 218 tcpstat.tcps_rcvdupbyte += *tlenp; 219 m_freem(m); 220 FREE(te, M_TSEGQ); 221 /* 222 * Try to present any queued data 223 * at the left window edge to the user. 224 * This is needed after the 3-WHS 225 * completes. 226 */ 227 goto present; /* ??? */ 228 } 229 m_adj(m, i); 230 *tlenp -= i; 231 th->th_seq += i; 232 } 233 } 234 tcpstat.tcps_rcvoopack++; 235 tcpstat.tcps_rcvoobyte += *tlenp; 236 237 /* 238 * While we overlap succeeding segments trim them or, 239 * if they are completely covered, dequeue them. 240 */ 241 while (q) { 242 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 243 if (i <= 0) 244 break; 245 if (i < q->tqe_len) { 246 q->tqe_th->th_seq += i; 247 q->tqe_len -= i; 248 m_adj(q->tqe_m, i); 249 break; 250 } 251 252 nq = LIST_NEXT(q, tqe_q); 253 LIST_REMOVE(q, tqe_q); 254 m_freem(q->tqe_m); 255 FREE(q, M_TSEGQ); 256 q = nq; 257 } 258 259 /* Insert the new segment queue entry into place. */ 260 te->tqe_m = m; 261 te->tqe_th = th; 262 te->tqe_len = *tlenp; 263 264 if (p == NULL) { 265 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 266 } else { 267 LIST_INSERT_AFTER(p, te, tqe_q); 268 } 269 270present: 271 /* 272 * Present data to user, advancing rcv_nxt through 273 * completed sequence space. 274 */ 275 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 276 return (0); 277 q = LIST_FIRST(&tp->t_segq); 278 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 279 return (0); 280 do { 281 tp->rcv_nxt += q->tqe_len; 282 flags = q->tqe_th->th_flags & TH_FIN; 283 nq = LIST_NEXT(q, tqe_q); 284 LIST_REMOVE(q, tqe_q); 285 if (so->so_state & SS_CANTRCVMORE) 286 m_freem(q->tqe_m); 287 else 288 sbappend(&so->so_rcv, q->tqe_m); 289 FREE(q, M_TSEGQ); 290 q = nq; 291 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 292 ND6_HINT(tp); 293 sorwakeup(so); 294 return (flags); 295} 296 297/* 298 * TCP input routine, follows pages 65-76 of the 299 * protocol specification dated September, 1981 very closely. 300 */ 301#ifdef INET6 302int 303tcp6_input(mp, offp, proto) 304 struct mbuf **mp; 305 int *offp, proto; 306{ 307 register struct mbuf *m = *mp; 308 struct in6_ifaddr *ia6; 309 310 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 311 312 /* 313 * draft-itojun-ipv6-tcp-to-anycast 314 * better place to put this in? 315 */ 316 ia6 = ip6_getdstifaddr(m); 317 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 318 struct ip6_hdr *ip6; 319 320 ip6 = mtod(m, struct ip6_hdr *); 321 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 322 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 323 return IPPROTO_DONE; 324 } 325 326 tcp_input(m, *offp, proto); 327 return IPPROTO_DONE; 328} 329#endif 330 331void 332tcp_input(m, off0, proto) 333 register struct mbuf *m; 334 int off0, proto; 335{ 336 register struct tcphdr *th; 337 register struct ip *ip = NULL; 338 register struct ipovly *ipov; 339 register struct inpcb *inp; 340 u_char *optp = NULL; 341 int optlen = 0; 342 int len, tlen, off; 343 int drop_hdrlen; 344 register struct tcpcb *tp = 0; 345 register int thflags; 346 struct socket *so = 0; 347 int todrop, acked, ourfinisacked, needoutput = 0; 348 struct in_addr laddr; 349#ifdef INET6 350 struct in6_addr laddr6; 351#endif 352 int dropsocket = 0; 353 int iss = 0; 354 u_long tiwin; 355 struct tcpopt to; /* options in this segment */ 356 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 357 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 358#ifdef TCPDEBUG 359 short ostate = 0; 360#endif 361#ifdef INET6 362 struct ip6_hdr *ip6 = NULL; 363 int isipv6; 364#endif /* INET6 */ 365 int rstreason; /* For badport_bandlim accounting purposes */ 366 367#ifdef INET6 368 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 369#endif 370 bzero((char *)&to, sizeof(to)); 371 372 tcpstat.tcps_rcvtotal++; 373 374#ifdef INET6 375 if (isipv6) { 376 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 377 ip6 = mtod(m, struct ip6_hdr *); 378 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 379 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 380 tcpstat.tcps_rcvbadsum++; 381 goto drop; 382 } 383 th = (struct tcphdr *)((caddr_t)ip6 + off0); 384 385 /* 386 * Be proactive about unspecified IPv6 address in source. 387 * As we use all-zero to indicate unbounded/unconnected pcb, 388 * unspecified IPv6 address can be used to confuse us. 389 * 390 * Note that packets with unspecified IPv6 destination is 391 * already dropped in ip6_input. 392 */ 393 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 394 /* XXX stat */ 395 goto drop; 396 } 397 } else 398#endif /* INET6 */ 399 { 400 /* 401 * Get IP and TCP header together in first mbuf. 402 * Note: IP leaves IP header in first mbuf. 403 */ 404 if (off0 > sizeof (struct ip)) { 405 ip_stripoptions(m, (struct mbuf *)0); 406 off0 = sizeof(struct ip); 407 } 408 if (m->m_len < sizeof (struct tcpiphdr)) { 409 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 410 tcpstat.tcps_rcvshort++; 411 return; 412 } 413 } 414 ip = mtod(m, struct ip *); 415 ipov = (struct ipovly *)ip; 416 th = (struct tcphdr *)((caddr_t)ip + off0); 417 tlen = ip->ip_len; 418 419 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 420 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 421 th->th_sum = m->m_pkthdr.csum_data; 422 else 423 th->th_sum = in_pseudo(ip->ip_src.s_addr, 424 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 425 ip->ip_len + IPPROTO_TCP)); 426 th->th_sum ^= 0xffff; 427 } else { 428 /* 429 * Checksum extended TCP header and data. 430 */ 431 len = sizeof (struct ip) + tlen; 432 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 433 ipov->ih_len = (u_short)tlen; 434 HTONS(ipov->ih_len); 435 th->th_sum = in_cksum(m, len); 436 } 437 if (th->th_sum) { 438 tcpstat.tcps_rcvbadsum++; 439 goto drop; 440 } 441#ifdef INET6 442 /* Re-initialization for later version check */ 443 ip->ip_v = IPVERSION; 444#endif 445 } 446 447 /* 448 * Check that TCP offset makes sense, 449 * pull out TCP options and adjust length. XXX 450 */ 451 off = th->th_off << 2; 452 if (off < sizeof (struct tcphdr) || off > tlen) { 453 tcpstat.tcps_rcvbadoff++; 454 goto drop; 455 } 456 tlen -= off; /* tlen is used instead of ti->ti_len */ 457 if (off > sizeof (struct tcphdr)) { 458#ifdef INET6 459 if (isipv6) { 460 IP6_EXTHDR_CHECK(m, off0, off, ); 461 ip6 = mtod(m, struct ip6_hdr *); 462 th = (struct tcphdr *)((caddr_t)ip6 + off0); 463 } else 464#endif /* INET6 */ 465 { 466 if (m->m_len < sizeof(struct ip) + off) { 467 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 468 tcpstat.tcps_rcvshort++; 469 return; 470 } 471 ip = mtod(m, struct ip *); 472 ipov = (struct ipovly *)ip; 473 th = (struct tcphdr *)((caddr_t)ip + off0); 474 } 475 } 476 optlen = off - sizeof (struct tcphdr); 477 optp = (u_char *)(th + 1); 478 } 479 thflags = th->th_flags; 480 481#ifdef TCP_DROP_SYNFIN 482 /* 483 * If the drop_synfin option is enabled, drop all packets with 484 * both the SYN and FIN bits set. This prevents e.g. nmap from 485 * identifying the TCP/IP stack. 486 * 487 * This is a violation of the TCP specification. 488 */ 489 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 490 goto drop; 491#endif 492 493 /* 494 * Convert TCP protocol specific fields to host format. 495 */ 496 NTOHL(th->th_seq); 497 NTOHL(th->th_ack); 498 NTOHS(th->th_win); 499 NTOHS(th->th_urp); 500 501 /* 502 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 503 * until after ip6_savecontrol() is called and before other functions 504 * which don't want those proto headers. 505 * Because ip6_savecontrol() is going to parse the mbuf to 506 * search for data to be passed up to user-land, it wants mbuf 507 * parameters to be unchanged. 508 */ 509 drop_hdrlen = off0 + off; 510 511 /* 512 * Locate pcb for segment. 513 */ 514findpcb: 515#ifdef IPFIREWALL_FORWARD 516 if (ip_fw_fwd_addr != NULL 517#ifdef INET6 518 && isipv6 == NULL /* IPv6 support is not yet */ 519#endif /* INET6 */ 520 ) { 521 /* 522 * Diverted. Pretend to be the destination. 523 * already got one like this? 524 */ 525 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 526 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 527 if (!inp) { 528 /* 529 * No, then it's new. Try find the ambushing socket 530 */ 531 if (!ip_fw_fwd_addr->sin_port) { 532 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 533 th->th_sport, ip_fw_fwd_addr->sin_addr, 534 th->th_dport, 1, m->m_pkthdr.rcvif); 535 } else { 536 inp = in_pcblookup_hash(&tcbinfo, 537 ip->ip_src, th->th_sport, 538 ip_fw_fwd_addr->sin_addr, 539 ntohs(ip_fw_fwd_addr->sin_port), 1, 540 m->m_pkthdr.rcvif); 541 } 542 } 543 ip_fw_fwd_addr = NULL; 544 } else 545#endif /* IPFIREWALL_FORWARD */ 546 { 547#ifdef INET6 548 if (isipv6) 549 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 550 &ip6->ip6_dst, th->th_dport, 1, 551 m->m_pkthdr.rcvif); 552 else 553#endif /* INET6 */ 554 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 555 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 556 } 557 558#ifdef IPSEC 559#ifdef INET6 560 if (isipv6) { 561 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 562 ipsec6stat.in_polvio++; 563 goto drop; 564 } 565 } else 566#endif /* INET6 */ 567 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 568 ipsecstat.in_polvio++; 569 goto drop; 570 } 571#endif /*IPSEC*/ 572 573 /* 574 * If the state is CLOSED (i.e., TCB does not exist) then 575 * all data in the incoming segment is discarded. 576 * If the TCB exists but is in CLOSED state, it is embryonic, 577 * but should either do a listen or a connect soon. 578 */ 579 if (inp == NULL) { 580 if (log_in_vain) { 581#ifdef INET6 582 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 583#else /* INET6 */ 584 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 585#endif /* INET6 */ 586 587#ifdef INET6 588 if (isipv6) { 589 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 590 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 591 } else 592#endif 593 { 594 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 595 strcpy(sbuf, inet_ntoa(ip->ip_src)); 596 } 597 switch (log_in_vain) { 598 case 1: 599 if(thflags & TH_SYN) 600 log(LOG_INFO, 601 "Connection attempt to TCP %s:%d from %s:%d\n", 602 dbuf, ntohs(th->th_dport), 603 sbuf, 604 ntohs(th->th_sport)); 605 break; 606 case 2: 607 log(LOG_INFO, 608 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 609 dbuf, ntohs(th->th_dport), sbuf, 610 ntohs(th->th_sport), thflags); 611 break; 612 default: 613 break; 614 } 615 } 616 if (blackhole) { 617 switch (blackhole) { 618 case 1: 619 if (thflags & TH_SYN) 620 goto drop; 621 break; 622 case 2: 623 goto drop; 624 default: 625 goto drop; 626 } 627 } 628 rstreason = BANDLIM_RST_CLOSEDPORT; 629 goto dropwithreset; 630 } 631 tp = intotcpcb(inp); 632 if (tp == 0) { 633 rstreason = BANDLIM_RST_CLOSEDPORT; 634 goto dropwithreset; 635 } 636 if (tp->t_state == TCPS_CLOSED) 637 goto drop; 638 639 /* Unscale the window into a 32-bit value. */ 640 if ((thflags & TH_SYN) == 0) 641 tiwin = th->th_win << tp->snd_scale; 642 else 643 tiwin = th->th_win; 644 645 so = inp->inp_socket; 646 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 647#ifdef TCPDEBUG 648 if (so->so_options & SO_DEBUG) { 649 ostate = tp->t_state; 650#ifdef INET6 651 if (isipv6) 652 bcopy((char *)ip6, (char *)tcp_saveipgen, 653 sizeof(*ip6)); 654 else 655#endif /* INET6 */ 656 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 657 tcp_savetcp = *th; 658 } 659#endif 660 if (so->so_options & SO_ACCEPTCONN) { 661 register struct tcpcb *tp0 = tp; 662 struct socket *so2; 663#ifdef IPSEC 664 struct socket *oso; 665#endif 666#ifdef INET6 667 struct inpcb *oinp = sotoinpcb(so); 668#endif /* INET6 */ 669 670#ifndef IPSEC 671 /* 672 * Current IPsec implementation makes incorrect IPsec 673 * cache if this check is done here. 674 * So delay this until duplicated socket is created. 675 */ 676 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 677 /* 678 * Note: dropwithreset makes sure we don't 679 * send a RST in response to a RST. 680 */ 681 if (thflags & TH_ACK) { 682 tcpstat.tcps_badsyn++; 683 rstreason = BANDLIM_RST_OPENPORT; 684 goto dropwithreset; 685 } 686 goto drop; 687 } 688#endif 689 690#ifdef INET6 691 /* 692 * If deprecated address is forbidden, 693 * we do not accept SYN to deprecated interface 694 * address to prevent any new inbound connection from 695 * getting established. 696 * When we do not accept SYN, we send a TCP RST, 697 * with deprecated source address (instead of dropping 698 * it). We compromise it as it is much better for peer 699 * to send a RST, and RST will be the final packet 700 * for the exchange. 701 * 702 * If we do not forbid deprecated addresses, we accept 703 * the SYN packet. RFC2462 does not suggest dropping 704 * SYN in this case. 705 * If we decipher RFC2462 5.5.4, it says like this: 706 * 1. use of deprecated addr with existing 707 * communication is okay - "SHOULD continue to be 708 * used" 709 * 2. use of it with new communication: 710 * (2a) "SHOULD NOT be used if alternate address 711 * with sufficient scope is available" 712 * (2b) nothing mentioned otherwise. 713 * Here we fall into (2b) case as we have no choice in 714 * our source address selection - we must obey the peer. 715 * 716 * The wording in RFC2462 is confusing, and there are 717 * multiple description text for deprecated address 718 * handling - worse, they are not exactly the same. 719 * I believe 5.5.4 is the best one, so we follow 5.5.4. 720 */ 721 if (isipv6 && !ip6_use_deprecated) { 722 struct in6_ifaddr *ia6; 723 724 if ((ia6 = ip6_getdstifaddr(m)) && 725 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 726 tp = NULL; 727 rstreason = BANDLIM_RST_OPENPORT; 728 goto dropwithreset; 729 } 730 } 731#endif 732 733 so2 = sonewconn(so, 0); 734 if (so2 == 0) { 735 /* 736 * If we were unable to create a new socket 737 * for this SYN, we call sodropablereq to 738 * see if there are any other sockets we 739 * can kick out of the listen queue. If 740 * so, we'll silently drop the socket 741 * sodropablereq told us to drop and 742 * create a new one. 743 * 744 * If sodropablereq returns 0, we'll 745 * simply drop the incoming SYN, as we 746 * can not allocate a socket for it. 747 */ 748 tcpstat.tcps_listendrop++; 749 so2 = sodropablereq(so); 750 if (so2) { 751 if (tcp_lq_overflow) 752 sototcpcb(so2)->t_flags |= 753 TF_LQ_OVERFLOW; 754 tcp_close(sototcpcb(so2)); 755 so2 = sonewconn(so, 0); 756 } 757 if (!so2) 758 goto drop; 759 } 760#ifdef IPSEC 761 oso = so; 762#endif 763 so = so2; 764 /* 765 * This is ugly, but .... 766 * 767 * Mark socket as temporary until we're 768 * committed to keeping it. The code at 769 * ``drop'' and ``dropwithreset'' check the 770 * flag dropsocket to see if the temporary 771 * socket created here should be discarded. 772 * We mark the socket as discardable until 773 * we're committed to it below in TCPS_LISTEN. 774 */ 775 dropsocket++; 776 inp = (struct inpcb *)so->so_pcb; 777#ifdef INET6 778 if (isipv6) 779 inp->in6p_laddr = ip6->ip6_dst; 780 else { 781 inp->inp_vflag &= ~INP_IPV6; 782 inp->inp_vflag |= INP_IPV4; 783#endif /* INET6 */ 784 inp->inp_laddr = ip->ip_dst; 785#ifdef INET6 786 } 787#endif /* INET6 */ 788 inp->inp_lport = th->th_dport; 789 if (in_pcbinshash(inp) != 0) { 790 /* 791 * Undo the assignments above if we failed to 792 * put the PCB on the hash lists. 793 */ 794#ifdef INET6 795 if (isipv6) 796 inp->in6p_laddr = in6addr_any; 797 else 798#endif /* INET6 */ 799 inp->inp_laddr.s_addr = INADDR_ANY; 800 inp->inp_lport = 0; 801 goto drop; 802 } 803#ifdef IPSEC 804 /* 805 * To avoid creating incorrectly cached IPsec 806 * association, this is need to be done here. 807 * 808 * Subject: (KAME-snap 748) 809 * From: Wayne Knowles <w.knowles@niwa.cri.nz> 810 * ftp://ftp.kame.net/pub/mail-list/snap-users/748 811 */ 812 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 813 /* 814 * Note: dropwithreset makes sure we don't 815 * send a RST in response to a RST. 816 */ 817 if (thflags & TH_ACK) { 818 tcpstat.tcps_badsyn++; 819 rstreason = BANDLIM_RST_OPENPORT; 820 goto dropwithreset; 821 } 822 goto drop; 823 } 824#endif 825#ifdef INET6 826 if (isipv6) { 827 /* 828 * Inherit socket options from the listening 829 * socket. 830 * Note that in6p_inputopts are not (even 831 * should not be) copied, since it stores 832 * previously received options and is used to 833 * detect if each new option is different than 834 * the previous one and hence should be passed 835 * to a user. 836 * If we copied in6p_inputopts, a user would 837 * not be able to receive options just after 838 * calling the accept system call. 839 */ 840 inp->inp_flags |= 841 oinp->inp_flags & INP_CONTROLOPTS; 842 if (oinp->in6p_outputopts) 843 inp->in6p_outputopts = 844 ip6_copypktopts(oinp->in6p_outputopts, 845 M_NOWAIT); 846 } else 847#endif /* INET6 */ 848 inp->inp_options = ip_srcroute(); 849#ifdef IPSEC 850 /* copy old policy into new socket's */ 851 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, 852 inp->inp_sp)) 853 printf("tcp_input: could not copy policy\n"); 854#endif 855 tp = intotcpcb(inp); 856 tp->t_state = TCPS_LISTEN; 857 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 858 859 /* Compute proper scaling value from buffer space */ 860 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 861 TCP_MAXWIN << tp->request_r_scale < 862 so->so_rcv.sb_hiwat) 863 tp->request_r_scale++; 864 } 865 } 866 867 /* 868 * Segment received on connection. 869 * Reset idle time and keep-alive timer. 870 */ 871 tp->t_rcvtime = ticks; 872 if (TCPS_HAVEESTABLISHED(tp->t_state)) 873 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 874 875 /* 876 * Process options if not in LISTEN state, 877 * else do it below (after getting remote address). 878 */ 879 if (tp->t_state != TCPS_LISTEN) 880 tcp_dooptions(tp, optp, optlen, th, &to); 881 882 /* 883 * Header prediction: check for the two common cases 884 * of a uni-directional data xfer. If the packet has 885 * no control flags, is in-sequence, the window didn't 886 * change and we're not retransmitting, it's a 887 * candidate. If the length is zero and the ack moved 888 * forward, we're the sender side of the xfer. Just 889 * free the data acked & wake any higher level process 890 * that was blocked waiting for space. If the length 891 * is non-zero and the ack didn't move, we're the 892 * receiver side. If we're getting packets in-order 893 * (the reassembly queue is empty), add the data to 894 * the socket buffer and note that we need a delayed ack. 895 * Make sure that the hidden state-flags are also off. 896 * Since we check for TCPS_ESTABLISHED above, it can only 897 * be TH_NEEDSYN. 898 */ 899 if (tp->t_state == TCPS_ESTABLISHED && 900 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 901 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 902 ((to.to_flag & TOF_TS) == 0 || 903 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 904 /* 905 * Using the CC option is compulsory if once started: 906 * the segment is OK if no T/TCP was negotiated or 907 * if the segment has a CC option equal to CCrecv 908 */ 909 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 910 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 911 th->th_seq == tp->rcv_nxt && 912 tiwin && tiwin == tp->snd_wnd && 913 tp->snd_nxt == tp->snd_max) { 914 915 /* 916 * If last ACK falls within this segment's sequence numbers, 917 * record the timestamp. 918 * NOTE that the test is modified according to the latest 919 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 920 */ 921 if ((to.to_flag & TOF_TS) != 0 && 922 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 923 tp->ts_recent_age = ticks; 924 tp->ts_recent = to.to_tsval; 925 } 926 927 if (tlen == 0) { 928 if (SEQ_GT(th->th_ack, tp->snd_una) && 929 SEQ_LEQ(th->th_ack, tp->snd_max) && 930 tp->snd_cwnd >= tp->snd_wnd && 931 tp->t_dupacks < tcprexmtthresh) { 932 /* 933 * this is a pure ack for outstanding data. 934 */ 935 ++tcpstat.tcps_predack; 936 /* 937 * "bad retransmit" recovery 938 */ 939 if (tp->t_rxtshift == 1 && 940 ticks < tp->t_badrxtwin) { 941 tp->snd_cwnd = tp->snd_cwnd_prev; 942 tp->snd_ssthresh = 943 tp->snd_ssthresh_prev; 944 tp->snd_nxt = tp->snd_max; 945 tp->t_badrxtwin = 0; 946 } 947 if ((to.to_flag & TOF_TS) != 0) 948 tcp_xmit_timer(tp, 949 ticks - to.to_tsecr + 1); 950 else if (tp->t_rtttime && 951 SEQ_GT(th->th_ack, tp->t_rtseq)) 952 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 953 acked = th->th_ack - tp->snd_una; 954 tcpstat.tcps_rcvackpack++; 955 tcpstat.tcps_rcvackbyte += acked; 956 sbdrop(&so->so_snd, acked); 957 tp->snd_una = th->th_ack; 958 m_freem(m); 959 ND6_HINT(tp); /* some progress has been done */ 960 961 /* 962 * If all outstanding data are acked, stop 963 * retransmit timer, otherwise restart timer 964 * using current (possibly backed-off) value. 965 * If process is waiting for space, 966 * wakeup/selwakeup/signal. If data 967 * are ready to send, let tcp_output 968 * decide between more output or persist. 969 */ 970 if (tp->snd_una == tp->snd_max) 971 callout_stop(tp->tt_rexmt); 972 else if (!callout_active(tp->tt_persist)) 973 callout_reset(tp->tt_rexmt, 974 tp->t_rxtcur, 975 tcp_timer_rexmt, tp); 976 977 sowwakeup(so); 978 if (so->so_snd.sb_cc) 979 (void) tcp_output(tp); 980 return; 981 } 982 } else if (th->th_ack == tp->snd_una && 983 LIST_EMPTY(&tp->t_segq) && 984 tlen <= sbspace(&so->so_rcv)) { 985 /* 986 * this is a pure, in-sequence data packet 987 * with nothing on the reassembly queue and 988 * we have enough buffer space to take it. 989 */ 990 ++tcpstat.tcps_preddat; 991 tp->rcv_nxt += tlen; 992 tcpstat.tcps_rcvpack++; 993 tcpstat.tcps_rcvbyte += tlen; 994 ND6_HINT(tp); /* some progress has been done */ 995 /* 996 * Add data to socket buffer. 997 */ 998 m_adj(m, drop_hdrlen); /* delayed header drop */ 999 sbappend(&so->so_rcv, m); 1000 sorwakeup(so); 1001 if (DELAY_ACK(tp)) { 1002 callout_reset(tp->tt_delack, tcp_delacktime, 1003 tcp_timer_delack, tp); 1004 } else { 1005 tp->t_flags |= TF_ACKNOW; 1006 tcp_output(tp); 1007 } 1008 return; 1009 } 1010 } 1011 1012 /* 1013 * Calculate amount of space in receive window, 1014 * and then do TCP input processing. 1015 * Receive window is amount of space in rcv queue, 1016 * but not less than advertised window. 1017 */ 1018 { int win; 1019 1020 win = sbspace(&so->so_rcv); 1021 if (win < 0) 1022 win = 0; 1023 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1024 } 1025 1026 switch (tp->t_state) { 1027 1028 /* 1029 * If the state is LISTEN then ignore segment if it contains an RST. 1030 * If the segment contains an ACK then it is bad and send a RST. 1031 * If it does not contain a SYN then it is not interesting; drop it. 1032 * If it is from this socket, drop it, it must be forged. 1033 * Don't bother responding if the destination was a broadcast. 1034 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 1035 * tp->iss, and send a segment: 1036 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 1037 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 1038 * Fill in remote peer address fields if not previously specified. 1039 * Enter SYN_RECEIVED state, and process any other fields of this 1040 * segment in this state. 1041 */ 1042 case TCPS_LISTEN: { 1043 register struct sockaddr_in *sin; 1044#ifdef INET6 1045 register struct sockaddr_in6 *sin6; 1046#endif 1047 1048 if (thflags & TH_RST) 1049 goto drop; 1050 if (thflags & TH_ACK) { 1051 rstreason = BANDLIM_RST_OPENPORT; 1052 goto dropwithreset; 1053 } 1054 if ((thflags & TH_SYN) == 0) 1055 goto drop; 1056 if (th->th_dport == th->th_sport) { 1057#ifdef INET6 1058 if (isipv6) { 1059 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 1060 &ip6->ip6_src)) 1061 goto drop; 1062 } else 1063#endif /* INET6 */ 1064 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1065 goto drop; 1066 } 1067 /* 1068 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1069 * in_broadcast() should never return true on a received 1070 * packet with M_BCAST not set. 1071 * 1072 * Packets with a multicast source address should also 1073 * be discarded. 1074 */ 1075 if (m->m_flags & (M_BCAST|M_MCAST)) 1076 goto drop; 1077#ifdef INET6 1078 if (isipv6) { 1079 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1080 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1081 goto drop; 1082 } else 1083#endif 1084 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1085 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1086 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 1087 goto drop; 1088#ifdef INET6 1089 if (isipv6) { 1090 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, 1091 M_SONAME, M_NOWAIT | M_ZERO); 1092 if (sin6 == NULL) 1093 goto drop; 1094 sin6->sin6_family = AF_INET6; 1095 sin6->sin6_len = sizeof(*sin6); 1096 sin6->sin6_addr = ip6->ip6_src; 1097 sin6->sin6_port = th->th_sport; 1098 laddr6 = inp->in6p_laddr; 1099 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1100 inp->in6p_laddr = ip6->ip6_dst; 1101 if (in6_pcbconnect(inp, (struct sockaddr *)sin6, 1102 &proc0)) { 1103 inp->in6p_laddr = laddr6; 1104 FREE(sin6, M_SONAME); 1105 goto drop; 1106 } 1107 FREE(sin6, M_SONAME); 1108 } else 1109#endif 1110 { 1111 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 1112 M_NOWAIT); 1113 if (sin == NULL) 1114 goto drop; 1115 sin->sin_family = AF_INET; 1116 sin->sin_len = sizeof(*sin); 1117 sin->sin_addr = ip->ip_src; 1118 sin->sin_port = th->th_sport; 1119 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 1120 laddr = inp->inp_laddr; 1121 if (inp->inp_laddr.s_addr == INADDR_ANY) 1122 inp->inp_laddr = ip->ip_dst; 1123 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 1124 inp->inp_laddr = laddr; 1125 FREE(sin, M_SONAME); 1126 goto drop; 1127 } 1128 FREE(sin, M_SONAME); 1129 } 1130 tp->t_template = tcp_template(tp); 1131 if (tp->t_template == 0) { 1132 tp = tcp_drop(tp, ENOBUFS); 1133 dropsocket = 0; /* socket is already gone */ 1134 goto drop; 1135 } 1136 if ((taop = tcp_gettaocache(inp)) == NULL) { 1137 taop = &tao_noncached; 1138 bzero(taop, sizeof(*taop)); 1139 } 1140 tcp_dooptions(tp, optp, optlen, th, &to); 1141 if (iss) 1142 tp->iss = iss; 1143 else { 1144 tp->iss = tcp_rndiss_next(); 1145 } 1146 tp->irs = th->th_seq; 1147 tcp_sendseqinit(tp); 1148 tcp_rcvseqinit(tp); 1149 tp->snd_recover = tp->snd_una; 1150 /* 1151 * Initialization of the tcpcb for transaction; 1152 * set SND.WND = SEG.WND, 1153 * initialize CCsend and CCrecv. 1154 */ 1155 tp->snd_wnd = tiwin; /* initial send-window */ 1156 tp->cc_send = CC_INC(tcp_ccgen); 1157 tp->cc_recv = to.to_cc; 1158 /* 1159 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1160 * - compare SEG.CC against cached CC from the same host, 1161 * if any. 1162 * - if SEG.CC > chached value, SYN must be new and is accepted 1163 * immediately: save new CC in the cache, mark the socket 1164 * connected, enter ESTABLISHED state, turn on flag to 1165 * send a SYN in the next segment. 1166 * A virtual advertised window is set in rcv_adv to 1167 * initialize SWS prevention. Then enter normal segment 1168 * processing: drop SYN, process data and FIN. 1169 * - otherwise do a normal 3-way handshake. 1170 */ 1171 if ((to.to_flag & TOF_CC) != 0) { 1172 if (((tp->t_flags & TF_NOPUSH) != 0) && 1173 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 1174 1175 taop->tao_cc = to.to_cc; 1176 tp->t_starttime = ticks; 1177 tp->t_state = TCPS_ESTABLISHED; 1178 1179 /* 1180 * If there is a FIN, or if there is data and the 1181 * connection is local, then delay SYN,ACK(SYN) in 1182 * the hope of piggy-backing it on a response 1183 * segment. Otherwise must send ACK now in case 1184 * the other side is slow starting. 1185 */ 1186 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 1187 (tlen != 0 && 1188#ifdef INET6 1189 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 1190 || 1191 (!isipv6 && 1192#endif 1193 in_localaddr(inp->inp_faddr) 1194#ifdef INET6 1195 )) 1196#endif 1197 ))) { 1198 callout_reset(tp->tt_delack, tcp_delacktime, 1199 tcp_timer_delack, tp); 1200 tp->t_flags |= TF_NEEDSYN; 1201 } else 1202 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1203 1204 /* 1205 * Limit the `virtual advertised window' to TCP_MAXWIN 1206 * here. Even if we requested window scaling, it will 1207 * become effective only later when our SYN is acked. 1208 */ 1209 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 1210 tcpstat.tcps_connects++; 1211 soisconnected(so); 1212 callout_reset(tp->tt_keep, tcp_keepinit, 1213 tcp_timer_keep, tp); 1214 dropsocket = 0; /* committed to socket */ 1215 tcpstat.tcps_accepts++; 1216 goto trimthenstep6; 1217 } 1218 /* else do standard 3-way handshake */ 1219 } else { 1220 /* 1221 * No CC option, but maybe CC.NEW: 1222 * invalidate cached value. 1223 */ 1224 taop->tao_cc = 0; 1225 } 1226 /* 1227 * TAO test failed or there was no CC option, 1228 * do a standard 3-way handshake. 1229 */ 1230 tp->t_flags |= TF_ACKNOW; 1231 tp->t_state = TCPS_SYN_RECEIVED; 1232 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 1233 dropsocket = 0; /* committed to socket */ 1234 tcpstat.tcps_accepts++; 1235 goto trimthenstep6; 1236 } 1237 1238 /* 1239 * If the state is SYN_RECEIVED: 1240 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1241 */ 1242 case TCPS_SYN_RECEIVED: 1243 if ((thflags & TH_ACK) && 1244 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1245 SEQ_GT(th->th_ack, tp->snd_max))) { 1246 rstreason = BANDLIM_RST_OPENPORT; 1247 goto dropwithreset; 1248 } 1249 break; 1250 1251 /* 1252 * If the state is SYN_SENT: 1253 * if seg contains an ACK, but not for our SYN, drop the input. 1254 * if seg contains a RST, then drop the connection. 1255 * if seg does not contain SYN, then drop it. 1256 * Otherwise this is an acceptable SYN segment 1257 * initialize tp->rcv_nxt and tp->irs 1258 * if seg contains ack then advance tp->snd_una 1259 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1260 * arrange for segment to be acked (eventually) 1261 * continue processing rest of data/controls, beginning with URG 1262 */ 1263 case TCPS_SYN_SENT: 1264 if ((taop = tcp_gettaocache(inp)) == NULL) { 1265 taop = &tao_noncached; 1266 bzero(taop, sizeof(*taop)); 1267 } 1268 1269 if ((thflags & TH_ACK) && 1270 (SEQ_LEQ(th->th_ack, tp->iss) || 1271 SEQ_GT(th->th_ack, tp->snd_max))) { 1272 /* 1273 * If we have a cached CCsent for the remote host, 1274 * hence we haven't just crashed and restarted, 1275 * do not send a RST. This may be a retransmission 1276 * from the other side after our earlier ACK was lost. 1277 * Our new SYN, when it arrives, will serve as the 1278 * needed ACK. 1279 */ 1280 if (taop->tao_ccsent != 0) 1281 goto drop; 1282 else { 1283 rstreason = BANDLIM_UNLIMITED; 1284 goto dropwithreset; 1285 } 1286 } 1287 if (thflags & TH_RST) { 1288 if (thflags & TH_ACK) 1289 tp = tcp_drop(tp, ECONNREFUSED); 1290 goto drop; 1291 } 1292 if ((thflags & TH_SYN) == 0) 1293 goto drop; 1294 tp->snd_wnd = th->th_win; /* initial send window */ 1295 tp->cc_recv = to.to_cc; /* foreign CC */ 1296 1297 tp->irs = th->th_seq; 1298 tcp_rcvseqinit(tp); 1299 if (thflags & TH_ACK) { 1300 /* 1301 * Our SYN was acked. If segment contains CC.ECHO 1302 * option, check it to make sure this segment really 1303 * matches our SYN. If not, just drop it as old 1304 * duplicate, but send an RST if we're still playing 1305 * by the old rules. If no CC.ECHO option, make sure 1306 * we don't get fooled into using T/TCP. 1307 */ 1308 if (to.to_flag & TOF_CCECHO) { 1309 if (tp->cc_send != to.to_ccecho) { 1310 if (taop->tao_ccsent != 0) 1311 goto drop; 1312 else { 1313 rstreason = BANDLIM_UNLIMITED; 1314 goto dropwithreset; 1315 } 1316 } 1317 } else 1318 tp->t_flags &= ~TF_RCVD_CC; 1319 tcpstat.tcps_connects++; 1320 soisconnected(so); 1321 /* Do window scaling on this connection? */ 1322 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1323 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1324 tp->snd_scale = tp->requested_s_scale; 1325 tp->rcv_scale = tp->request_r_scale; 1326 } 1327 /* Segment is acceptable, update cache if undefined. */ 1328 if (taop->tao_ccsent == 0) 1329 taop->tao_ccsent = to.to_ccecho; 1330 1331 tp->rcv_adv += tp->rcv_wnd; 1332 tp->snd_una++; /* SYN is acked */ 1333 /* 1334 * If there's data, delay ACK; if there's also a FIN 1335 * ACKNOW will be turned on later. 1336 */ 1337 if (DELAY_ACK(tp) && tlen != 0) 1338 callout_reset(tp->tt_delack, tcp_delacktime, 1339 tcp_timer_delack, tp); 1340 else 1341 tp->t_flags |= TF_ACKNOW; 1342 /* 1343 * Received <SYN,ACK> in SYN_SENT[*] state. 1344 * Transitions: 1345 * SYN_SENT --> ESTABLISHED 1346 * SYN_SENT* --> FIN_WAIT_1 1347 */ 1348 tp->t_starttime = ticks; 1349 if (tp->t_flags & TF_NEEDFIN) { 1350 tp->t_state = TCPS_FIN_WAIT_1; 1351 tp->t_flags &= ~TF_NEEDFIN; 1352 thflags &= ~TH_SYN; 1353 } else { 1354 tp->t_state = TCPS_ESTABLISHED; 1355 callout_reset(tp->tt_keep, tcp_keepidle, 1356 tcp_timer_keep, tp); 1357 } 1358 } else { 1359 /* 1360 * Received initial SYN in SYN-SENT[*] state => simul- 1361 * taneous open. If segment contains CC option and there is 1362 * a cached CC, apply TAO test; if it succeeds, connection is 1363 * half-synchronized. Otherwise, do 3-way handshake: 1364 * SYN-SENT -> SYN-RECEIVED 1365 * SYN-SENT* -> SYN-RECEIVED* 1366 * If there was no CC option, clear cached CC value. 1367 */ 1368 tp->t_flags |= TF_ACKNOW; 1369 callout_stop(tp->tt_rexmt); 1370 if (to.to_flag & TOF_CC) { 1371 if (taop->tao_cc != 0 && 1372 CC_GT(to.to_cc, taop->tao_cc)) { 1373 /* 1374 * update cache and make transition: 1375 * SYN-SENT -> ESTABLISHED* 1376 * SYN-SENT* -> FIN-WAIT-1* 1377 */ 1378 taop->tao_cc = to.to_cc; 1379 tp->t_starttime = ticks; 1380 if (tp->t_flags & TF_NEEDFIN) { 1381 tp->t_state = TCPS_FIN_WAIT_1; 1382 tp->t_flags &= ~TF_NEEDFIN; 1383 } else { 1384 tp->t_state = TCPS_ESTABLISHED; 1385 callout_reset(tp->tt_keep, 1386 tcp_keepidle, 1387 tcp_timer_keep, 1388 tp); 1389 } 1390 tp->t_flags |= TF_NEEDSYN; 1391 } else 1392 tp->t_state = TCPS_SYN_RECEIVED; 1393 } else { 1394 /* CC.NEW or no option => invalidate cache */ 1395 taop->tao_cc = 0; 1396 tp->t_state = TCPS_SYN_RECEIVED; 1397 } 1398 } 1399 1400trimthenstep6: 1401 /* 1402 * Advance th->th_seq to correspond to first data byte. 1403 * If data, trim to stay within window, 1404 * dropping FIN if necessary. 1405 */ 1406 th->th_seq++; 1407 if (tlen > tp->rcv_wnd) { 1408 todrop = tlen - tp->rcv_wnd; 1409 m_adj(m, -todrop); 1410 tlen = tp->rcv_wnd; 1411 thflags &= ~TH_FIN; 1412 tcpstat.tcps_rcvpackafterwin++; 1413 tcpstat.tcps_rcvbyteafterwin += todrop; 1414 } 1415 tp->snd_wl1 = th->th_seq - 1; 1416 tp->rcv_up = th->th_seq; 1417 /* 1418 * Client side of transaction: already sent SYN and data. 1419 * If the remote host used T/TCP to validate the SYN, 1420 * our data will be ACK'd; if so, enter normal data segment 1421 * processing in the middle of step 5, ack processing. 1422 * Otherwise, goto step 6. 1423 */ 1424 if (thflags & TH_ACK) 1425 goto process_ACK; 1426 goto step6; 1427 /* 1428 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1429 * if segment contains a SYN and CC [not CC.NEW] option: 1430 * if state == TIME_WAIT and connection duration > MSL, 1431 * drop packet and send RST; 1432 * 1433 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1434 * ack the FIN (and data) in retransmission queue. 1435 * Complete close and delete TCPCB. Then reprocess 1436 * segment, hoping to find new TCPCB in LISTEN state; 1437 * 1438 * else must be old SYN; drop it. 1439 * else do normal processing. 1440 */ 1441 case TCPS_LAST_ACK: 1442 case TCPS_CLOSING: 1443 case TCPS_TIME_WAIT: 1444 if ((thflags & TH_SYN) && 1445 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 1446 if (tp->t_state == TCPS_TIME_WAIT && 1447 (ticks - tp->t_starttime) > tcp_msl) { 1448 rstreason = BANDLIM_UNLIMITED; 1449 goto dropwithreset; 1450 } 1451 if (CC_GT(to.to_cc, tp->cc_recv)) { 1452 tp = tcp_close(tp); 1453 goto findpcb; 1454 } 1455 else 1456 goto drop; 1457 } 1458 break; /* continue normal processing */ 1459 } 1460 1461 /* 1462 * States other than LISTEN or SYN_SENT. 1463 * First check the RST flag and sequence number since reset segments 1464 * are exempt from the timestamp and connection count tests. This 1465 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1466 * below which allowed reset segments in half the sequence space 1467 * to fall though and be processed (which gives forged reset 1468 * segments with a random sequence number a 50 percent chance of 1469 * killing a connection). 1470 * Then check timestamp, if present. 1471 * Then check the connection count, if present. 1472 * Then check that at least some bytes of segment are within 1473 * receive window. If segment begins before rcv_nxt, 1474 * drop leading data (and SYN); if nothing left, just ack. 1475 * 1476 * 1477 * If the RST bit is set, check the sequence number to see 1478 * if this is a valid reset segment. 1479 * RFC 793 page 37: 1480 * In all states except SYN-SENT, all reset (RST) segments 1481 * are validated by checking their SEQ-fields. A reset is 1482 * valid if its sequence number is in the window. 1483 * Note: this does not take into account delayed ACKs, so 1484 * we should test against last_ack_sent instead of rcv_nxt. 1485 * The sequence number in the reset segment is normally an 1486 * echo of our outgoing acknowlegement numbers, but some hosts 1487 * send a reset with the sequence number at the rightmost edge 1488 * of our receive window, and we have to handle this case. 1489 * If we have multiple segments in flight, the intial reset 1490 * segment sequence numbers will be to the left of last_ack_sent, 1491 * but they will eventually catch up. 1492 * In any case, it never made sense to trim reset segments to 1493 * fit the receive window since RFC 1122 says: 1494 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1495 * 1496 * A TCP SHOULD allow a received RST segment to include data. 1497 * 1498 * DISCUSSION 1499 * It has been suggested that a RST segment could contain 1500 * ASCII text that encoded and explained the cause of the 1501 * RST. No standard has yet been established for such 1502 * data. 1503 * 1504 * If the reset segment passes the sequence number test examine 1505 * the state: 1506 * SYN_RECEIVED STATE: 1507 * If passive open, return to LISTEN state. 1508 * If active open, inform user that connection was refused. 1509 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1510 * Inform user that connection was reset, and close tcb. 1511 * CLOSING, LAST_ACK STATES: 1512 * Close the tcb. 1513 * TIME_WAIT STATE: 1514 * Drop the segment - see Stevens, vol. 2, p. 964 and 1515 * RFC 1337. 1516 */ 1517 if (thflags & TH_RST) { 1518 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1519 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1520 switch (tp->t_state) { 1521 1522 case TCPS_SYN_RECEIVED: 1523 so->so_error = ECONNREFUSED; 1524 goto close; 1525 1526 case TCPS_ESTABLISHED: 1527 case TCPS_FIN_WAIT_1: 1528 case TCPS_FIN_WAIT_2: 1529 case TCPS_CLOSE_WAIT: 1530 so->so_error = ECONNRESET; 1531 close: 1532 tp->t_state = TCPS_CLOSED; 1533 tcpstat.tcps_drops++; 1534 tp = tcp_close(tp); 1535 break; 1536 1537 case TCPS_CLOSING: 1538 case TCPS_LAST_ACK: 1539 tp = tcp_close(tp); 1540 break; 1541 1542 case TCPS_TIME_WAIT: 1543 break; 1544 } 1545 } 1546 goto drop; 1547 } 1548 1549 /* 1550 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1551 * and it's less than ts_recent, drop it. 1552 */ 1553 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1554 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1555 1556 /* Check to see if ts_recent is over 24 days old. */ 1557 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1558 /* 1559 * Invalidate ts_recent. If this segment updates 1560 * ts_recent, the age will be reset later and ts_recent 1561 * will get a valid value. If it does not, setting 1562 * ts_recent to zero will at least satisfy the 1563 * requirement that zero be placed in the timestamp 1564 * echo reply when ts_recent isn't valid. The 1565 * age isn't reset until we get a valid ts_recent 1566 * because we don't want out-of-order segments to be 1567 * dropped when ts_recent is old. 1568 */ 1569 tp->ts_recent = 0; 1570 } else { 1571 tcpstat.tcps_rcvduppack++; 1572 tcpstat.tcps_rcvdupbyte += tlen; 1573 tcpstat.tcps_pawsdrop++; 1574 goto dropafterack; 1575 } 1576 } 1577 1578 /* 1579 * T/TCP mechanism 1580 * If T/TCP was negotiated and the segment doesn't have CC, 1581 * or if its CC is wrong then drop the segment. 1582 * RST segments do not have to comply with this. 1583 */ 1584 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1585 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1586 goto dropafterack; 1587 1588 /* 1589 * In the SYN-RECEIVED state, validate that the packet belongs to 1590 * this connection before trimming the data to fit the receive 1591 * window. Check the sequence number versus IRS since we know 1592 * the sequence numbers haven't wrapped. This is a partial fix 1593 * for the "LAND" DoS attack. 1594 */ 1595 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1596 rstreason = BANDLIM_RST_OPENPORT; 1597 goto dropwithreset; 1598 } 1599 1600 todrop = tp->rcv_nxt - th->th_seq; 1601 if (todrop > 0) { 1602 if (thflags & TH_SYN) { 1603 thflags &= ~TH_SYN; 1604 th->th_seq++; 1605 if (th->th_urp > 1) 1606 th->th_urp--; 1607 else 1608 thflags &= ~TH_URG; 1609 todrop--; 1610 } 1611 /* 1612 * Following if statement from Stevens, vol. 2, p. 960. 1613 */ 1614 if (todrop > tlen 1615 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1616 /* 1617 * Any valid FIN must be to the left of the window. 1618 * At this point the FIN must be a duplicate or out 1619 * of sequence; drop it. 1620 */ 1621 thflags &= ~TH_FIN; 1622 1623 /* 1624 * Send an ACK to resynchronize and drop any data. 1625 * But keep on processing for RST or ACK. 1626 */ 1627 tp->t_flags |= TF_ACKNOW; 1628 todrop = tlen; 1629 tcpstat.tcps_rcvduppack++; 1630 tcpstat.tcps_rcvdupbyte += todrop; 1631 } else { 1632 tcpstat.tcps_rcvpartduppack++; 1633 tcpstat.tcps_rcvpartdupbyte += todrop; 1634 } 1635 drop_hdrlen += todrop; /* drop from the top afterwards */ 1636 th->th_seq += todrop; 1637 tlen -= todrop; 1638 if (th->th_urp > todrop) 1639 th->th_urp -= todrop; 1640 else { 1641 thflags &= ~TH_URG; 1642 th->th_urp = 0; 1643 } 1644 } 1645 1646 /* 1647 * If new data are received on a connection after the 1648 * user processes are gone, then RST the other end. 1649 */ 1650 if ((so->so_state & SS_NOFDREF) && 1651 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1652 tp = tcp_close(tp); 1653 tcpstat.tcps_rcvafterclose++; 1654 rstreason = BANDLIM_UNLIMITED; 1655 goto dropwithreset; 1656 } 1657 1658 /* 1659 * If segment ends after window, drop trailing data 1660 * (and PUSH and FIN); if nothing left, just ACK. 1661 */ 1662 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1663 if (todrop > 0) { 1664 tcpstat.tcps_rcvpackafterwin++; 1665 if (todrop >= tlen) { 1666 tcpstat.tcps_rcvbyteafterwin += tlen; 1667 /* 1668 * If a new connection request is received 1669 * while in TIME_WAIT, drop the old connection 1670 * and start over if the sequence numbers 1671 * are above the previous ones. 1672 */ 1673 if (thflags & TH_SYN && 1674 tp->t_state == TCPS_TIME_WAIT && 1675 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1676 iss = tcp_rndiss_next(); 1677 tp = tcp_close(tp); 1678 goto findpcb; 1679 } 1680 /* 1681 * If window is closed can only take segments at 1682 * window edge, and have to drop data and PUSH from 1683 * incoming segments. Continue processing, but 1684 * remember to ack. Otherwise, drop segment 1685 * and ack. 1686 */ 1687 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1688 tp->t_flags |= TF_ACKNOW; 1689 tcpstat.tcps_rcvwinprobe++; 1690 } else 1691 goto dropafterack; 1692 } else 1693 tcpstat.tcps_rcvbyteafterwin += todrop; 1694 m_adj(m, -todrop); 1695 tlen -= todrop; 1696 thflags &= ~(TH_PUSH|TH_FIN); 1697 } 1698 1699 /* 1700 * If last ACK falls within this segment's sequence numbers, 1701 * record its timestamp. 1702 * NOTE that the test is modified according to the latest 1703 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1704 */ 1705 if ((to.to_flag & TOF_TS) != 0 && 1706 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1707 tp->ts_recent_age = ticks; 1708 tp->ts_recent = to.to_tsval; 1709 } 1710 1711 /* 1712 * If a SYN is in the window, then this is an 1713 * error and we send an RST and drop the connection. 1714 */ 1715 if (thflags & TH_SYN) { 1716 tp = tcp_drop(tp, ECONNRESET); 1717 rstreason = BANDLIM_UNLIMITED; 1718 goto dropwithreset; 1719 } 1720 1721 /* 1722 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1723 * flag is on (half-synchronized state), then queue data for 1724 * later processing; else drop segment and return. 1725 */ 1726 if ((thflags & TH_ACK) == 0) { 1727 if (tp->t_state == TCPS_SYN_RECEIVED || 1728 (tp->t_flags & TF_NEEDSYN)) 1729 goto step6; 1730 else 1731 goto drop; 1732 } 1733 1734 /* 1735 * Ack processing. 1736 */ 1737 switch (tp->t_state) { 1738 1739 /* 1740 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1741 * ESTABLISHED state and continue processing. 1742 * The ACK was checked above. 1743 */ 1744 case TCPS_SYN_RECEIVED: 1745 1746 tcpstat.tcps_connects++; 1747 soisconnected(so); 1748 /* Do window scaling? */ 1749 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1750 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1751 tp->snd_scale = tp->requested_s_scale; 1752 tp->rcv_scale = tp->request_r_scale; 1753 } 1754 /* 1755 * Upon successful completion of 3-way handshake, 1756 * update cache.CC if it was undefined, pass any queued 1757 * data to the user, and advance state appropriately. 1758 */ 1759 if ((taop = tcp_gettaocache(inp)) != NULL && 1760 taop->tao_cc == 0) 1761 taop->tao_cc = tp->cc_recv; 1762 1763 /* 1764 * Make transitions: 1765 * SYN-RECEIVED -> ESTABLISHED 1766 * SYN-RECEIVED* -> FIN-WAIT-1 1767 */ 1768 tp->t_starttime = ticks; 1769 if (tp->t_flags & TF_NEEDFIN) { 1770 tp->t_state = TCPS_FIN_WAIT_1; 1771 tp->t_flags &= ~TF_NEEDFIN; 1772 } else { 1773 tp->t_state = TCPS_ESTABLISHED; 1774 callout_reset(tp->tt_keep, tcp_keepidle, 1775 tcp_timer_keep, tp); 1776 } 1777 /* 1778 * If segment contains data or ACK, will call tcp_reass() 1779 * later; if not, do so now to pass queued data to user. 1780 */ 1781 if (tlen == 0 && (thflags & TH_FIN) == 0) 1782 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1783 (struct mbuf *)0); 1784 tp->snd_wl1 = th->th_seq - 1; 1785 /* fall into ... */ 1786 1787 /* 1788 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1789 * ACKs. If the ack is in the range 1790 * tp->snd_una < th->th_ack <= tp->snd_max 1791 * then advance tp->snd_una to th->th_ack and drop 1792 * data from the retransmission queue. If this ACK reflects 1793 * more up to date window information we update our window information. 1794 */ 1795 case TCPS_ESTABLISHED: 1796 case TCPS_FIN_WAIT_1: 1797 case TCPS_FIN_WAIT_2: 1798 case TCPS_CLOSE_WAIT: 1799 case TCPS_CLOSING: 1800 case TCPS_LAST_ACK: 1801 case TCPS_TIME_WAIT: 1802 1803 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1804 if (tlen == 0 && tiwin == tp->snd_wnd) { 1805 tcpstat.tcps_rcvdupack++; 1806 /* 1807 * If we have outstanding data (other than 1808 * a window probe), this is a completely 1809 * duplicate ack (ie, window info didn't 1810 * change), the ack is the biggest we've 1811 * seen and we've seen exactly our rexmt 1812 * threshhold of them, assume a packet 1813 * has been dropped and retransmit it. 1814 * Kludge snd_nxt & the congestion 1815 * window so we send only this one 1816 * packet. 1817 * 1818 * We know we're losing at the current 1819 * window size so do congestion avoidance 1820 * (set ssthresh to half the current window 1821 * and pull our congestion window back to 1822 * the new ssthresh). 1823 * 1824 * Dup acks mean that packets have left the 1825 * network (they're now cached at the receiver) 1826 * so bump cwnd by the amount in the receiver 1827 * to keep a constant cwnd packets in the 1828 * network. 1829 */ 1830 if (!callout_active(tp->tt_rexmt) || 1831 th->th_ack != tp->snd_una) 1832 tp->t_dupacks = 0; 1833 else if (++tp->t_dupacks == tcprexmtthresh) { 1834 tcp_seq onxt = tp->snd_nxt; 1835 u_int win = 1836 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1837 tp->t_maxseg; 1838 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1839 tp->snd_recover)) { 1840 /* False retransmit, should not 1841 * cut window 1842 */ 1843 tp->snd_cwnd += tp->t_maxseg; 1844 tp->t_dupacks = 0; 1845 (void) tcp_output(tp); 1846 goto drop; 1847 } 1848 if (win < 2) 1849 win = 2; 1850 tp->snd_ssthresh = win * tp->t_maxseg; 1851 tp->snd_recover = tp->snd_max; 1852 callout_stop(tp->tt_rexmt); 1853 tp->t_rtttime = 0; 1854 tp->snd_nxt = th->th_ack; 1855 tp->snd_cwnd = tp->t_maxseg; 1856 (void) tcp_output(tp); 1857 tp->snd_cwnd = tp->snd_ssthresh + 1858 tp->t_maxseg * tp->t_dupacks; 1859 if (SEQ_GT(onxt, tp->snd_nxt)) 1860 tp->snd_nxt = onxt; 1861 goto drop; 1862 } else if (tp->t_dupacks > tcprexmtthresh) { 1863 tp->snd_cwnd += tp->t_maxseg; 1864 (void) tcp_output(tp); 1865 goto drop; 1866 } 1867 } else 1868 tp->t_dupacks = 0; 1869 break; 1870 } 1871 /* 1872 * If the congestion window was inflated to account 1873 * for the other side's cached packets, retract it. 1874 */ 1875 if (tcp_do_newreno == 0) { 1876 if (tp->t_dupacks >= tcprexmtthresh && 1877 tp->snd_cwnd > tp->snd_ssthresh) 1878 tp->snd_cwnd = tp->snd_ssthresh; 1879 tp->t_dupacks = 0; 1880 } else if (tp->t_dupacks >= tcprexmtthresh && 1881 !tcp_newreno(tp, th)) { 1882 /* 1883 * Window inflation should have left us with approx. 1884 * snd_ssthresh outstanding data. But in case we 1885 * would be inclined to send a burst, better to do 1886 * it via the slow start mechanism. 1887 */ 1888 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1889 tp->snd_cwnd = 1890 tp->snd_max - th->th_ack + tp->t_maxseg; 1891 else 1892 tp->snd_cwnd = tp->snd_ssthresh; 1893 tp->t_dupacks = 0; 1894 } 1895 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1896 tcpstat.tcps_rcvacktoomuch++; 1897 goto dropafterack; 1898 } 1899 /* 1900 * If we reach this point, ACK is not a duplicate, 1901 * i.e., it ACKs something we sent. 1902 */ 1903 if (tp->t_flags & TF_NEEDSYN) { 1904 /* 1905 * T/TCP: Connection was half-synchronized, and our 1906 * SYN has been ACK'd (so connection is now fully 1907 * synchronized). Go to non-starred state, 1908 * increment snd_una for ACK of SYN, and check if 1909 * we can do window scaling. 1910 */ 1911 tp->t_flags &= ~TF_NEEDSYN; 1912 tp->snd_una++; 1913 /* Do window scaling? */ 1914 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1915 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1916 tp->snd_scale = tp->requested_s_scale; 1917 tp->rcv_scale = tp->request_r_scale; 1918 } 1919 } 1920 1921process_ACK: 1922 acked = th->th_ack - tp->snd_una; 1923 tcpstat.tcps_rcvackpack++; 1924 tcpstat.tcps_rcvackbyte += acked; 1925 1926 /* 1927 * If we just performed our first retransmit, and the ACK 1928 * arrives within our recovery window, then it was a mistake 1929 * to do the retransmit in the first place. Recover our 1930 * original cwnd and ssthresh, and proceed to transmit where 1931 * we left off. 1932 */ 1933 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1934 tp->snd_cwnd = tp->snd_cwnd_prev; 1935 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1936 tp->snd_nxt = tp->snd_max; 1937 tp->t_badrxtwin = 0; /* XXX probably not required */ 1938 } 1939 1940 /* 1941 * If we have a timestamp reply, update smoothed 1942 * round trip time. If no timestamp is present but 1943 * transmit timer is running and timed sequence 1944 * number was acked, update smoothed round trip time. 1945 * Since we now have an rtt measurement, cancel the 1946 * timer backoff (cf., Phil Karn's retransmit alg.). 1947 * Recompute the initial retransmit timer. 1948 */ 1949 if (to.to_flag & TOF_TS) 1950 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1951 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1952 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1953 1954 /* 1955 * If all outstanding data is acked, stop retransmit 1956 * timer and remember to restart (more output or persist). 1957 * If there is more data to be acked, restart retransmit 1958 * timer, using current (possibly backed-off) value. 1959 */ 1960 if (th->th_ack == tp->snd_max) { 1961 callout_stop(tp->tt_rexmt); 1962 needoutput = 1; 1963 } else if (!callout_active(tp->tt_persist)) 1964 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1965 tcp_timer_rexmt, tp); 1966 1967 /* 1968 * If no data (only SYN) was ACK'd, 1969 * skip rest of ACK processing. 1970 */ 1971 if (acked == 0) 1972 goto step6; 1973 1974 /* 1975 * When new data is acked, open the congestion window. 1976 * If the window gives us less than ssthresh packets 1977 * in flight, open exponentially (maxseg per packet). 1978 * Otherwise open linearly: maxseg per window 1979 * (maxseg^2 / cwnd per packet). 1980 */ 1981 { 1982 register u_int cw = tp->snd_cwnd; 1983 register u_int incr = tp->t_maxseg; 1984 1985 if (cw > tp->snd_ssthresh) 1986 incr = incr * incr / cw; 1987 /* 1988 * If t_dupacks != 0 here, it indicates that we are still 1989 * in NewReno fast recovery mode, so we leave the congestion 1990 * window alone. 1991 */ 1992 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1993 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1994 } 1995 if (acked > so->so_snd.sb_cc) { 1996 tp->snd_wnd -= so->so_snd.sb_cc; 1997 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1998 ourfinisacked = 1; 1999 } else { 2000 sbdrop(&so->so_snd, acked); 2001 tp->snd_wnd -= acked; 2002 ourfinisacked = 0; 2003 } 2004 sowwakeup(so); 2005 tp->snd_una = th->th_ack; 2006 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2007 tp->snd_nxt = tp->snd_una; 2008 2009 switch (tp->t_state) { 2010 2011 /* 2012 * In FIN_WAIT_1 STATE in addition to the processing 2013 * for the ESTABLISHED state if our FIN is now acknowledged 2014 * then enter FIN_WAIT_2. 2015 */ 2016 case TCPS_FIN_WAIT_1: 2017 if (ourfinisacked) { 2018 /* 2019 * If we can't receive any more 2020 * data, then closing user can proceed. 2021 * Starting the timer is contrary to the 2022 * specification, but if we don't get a FIN 2023 * we'll hang forever. 2024 */ 2025 if (so->so_state & SS_CANTRCVMORE) { 2026 soisdisconnected(so); 2027 callout_reset(tp->tt_2msl, tcp_maxidle, 2028 tcp_timer_2msl, tp); 2029 } 2030 tp->t_state = TCPS_FIN_WAIT_2; 2031 } 2032 break; 2033 2034 /* 2035 * In CLOSING STATE in addition to the processing for 2036 * the ESTABLISHED state if the ACK acknowledges our FIN 2037 * then enter the TIME-WAIT state, otherwise ignore 2038 * the segment. 2039 */ 2040 case TCPS_CLOSING: 2041 if (ourfinisacked) { 2042 tp->t_state = TCPS_TIME_WAIT; 2043 tcp_canceltimers(tp); 2044 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2045 if (tp->cc_recv != 0 && 2046 (ticks - tp->t_starttime) < tcp_msl) 2047 callout_reset(tp->tt_2msl, 2048 tp->t_rxtcur * 2049 TCPTV_TWTRUNC, 2050 tcp_timer_2msl, tp); 2051 else 2052 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2053 tcp_timer_2msl, tp); 2054 soisdisconnected(so); 2055 } 2056 break; 2057 2058 /* 2059 * In LAST_ACK, we may still be waiting for data to drain 2060 * and/or to be acked, as well as for the ack of our FIN. 2061 * If our FIN is now acknowledged, delete the TCB, 2062 * enter the closed state and return. 2063 */ 2064 case TCPS_LAST_ACK: 2065 if (ourfinisacked) { 2066 tp = tcp_close(tp); 2067 goto drop; 2068 } 2069 break; 2070 2071 /* 2072 * In TIME_WAIT state the only thing that should arrive 2073 * is a retransmission of the remote FIN. Acknowledge 2074 * it and restart the finack timer. 2075 */ 2076 case TCPS_TIME_WAIT: 2077 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2078 tcp_timer_2msl, tp); 2079 goto dropafterack; 2080 } 2081 } 2082 2083step6: 2084 /* 2085 * Update window information. 2086 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2087 */ 2088 if ((thflags & TH_ACK) && 2089 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2090 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2091 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2092 /* keep track of pure window updates */ 2093 if (tlen == 0 && 2094 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2095 tcpstat.tcps_rcvwinupd++; 2096 tp->snd_wnd = tiwin; 2097 tp->snd_wl1 = th->th_seq; 2098 tp->snd_wl2 = th->th_ack; 2099 if (tp->snd_wnd > tp->max_sndwnd) 2100 tp->max_sndwnd = tp->snd_wnd; 2101 needoutput = 1; 2102 } 2103 2104 /* 2105 * Process segments with URG. 2106 */ 2107 if ((thflags & TH_URG) && th->th_urp && 2108 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2109 /* 2110 * This is a kludge, but if we receive and accept 2111 * random urgent pointers, we'll crash in 2112 * soreceive. It's hard to imagine someone 2113 * actually wanting to send this much urgent data. 2114 */ 2115 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2116 th->th_urp = 0; /* XXX */ 2117 thflags &= ~TH_URG; /* XXX */ 2118 goto dodata; /* XXX */ 2119 } 2120 /* 2121 * If this segment advances the known urgent pointer, 2122 * then mark the data stream. This should not happen 2123 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2124 * a FIN has been received from the remote side. 2125 * In these states we ignore the URG. 2126 * 2127 * According to RFC961 (Assigned Protocols), 2128 * the urgent pointer points to the last octet 2129 * of urgent data. We continue, however, 2130 * to consider it to indicate the first octet 2131 * of data past the urgent section as the original 2132 * spec states (in one of two places). 2133 */ 2134 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2135 tp->rcv_up = th->th_seq + th->th_urp; 2136 so->so_oobmark = so->so_rcv.sb_cc + 2137 (tp->rcv_up - tp->rcv_nxt) - 1; 2138 if (so->so_oobmark == 0) 2139 so->so_state |= SS_RCVATMARK; 2140 sohasoutofband(so); 2141 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2142 } 2143 /* 2144 * Remove out of band data so doesn't get presented to user. 2145 * This can happen independent of advancing the URG pointer, 2146 * but if two URG's are pending at once, some out-of-band 2147 * data may creep in... ick. 2148 */ 2149 if (th->th_urp <= (u_long)tlen 2150#ifdef SO_OOBINLINE 2151 && (so->so_options & SO_OOBINLINE) == 0 2152#endif 2153 ) 2154 tcp_pulloutofband(so, th, m, 2155 drop_hdrlen); /* hdr drop is delayed */ 2156 } else 2157 /* 2158 * If no out of band data is expected, 2159 * pull receive urgent pointer along 2160 * with the receive window. 2161 */ 2162 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2163 tp->rcv_up = tp->rcv_nxt; 2164dodata: /* XXX */ 2165 2166 /* 2167 * Process the segment text, merging it into the TCP sequencing queue, 2168 * and arranging for acknowledgment of receipt if necessary. 2169 * This process logically involves adjusting tp->rcv_wnd as data 2170 * is presented to the user (this happens in tcp_usrreq.c, 2171 * case PRU_RCVD). If a FIN has already been received on this 2172 * connection then we just ignore the text. 2173 */ 2174 if ((tlen || (thflags&TH_FIN)) && 2175 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2176 m_adj(m, drop_hdrlen); /* delayed header drop */ 2177 /* 2178 * Insert segment which inludes th into reassembly queue of tcp with 2179 * control block tp. Return TH_FIN if reassembly now includes 2180 * a segment with FIN. This handle the common case inline (segment 2181 * is the next to be received on an established connection, and the 2182 * queue is empty), avoiding linkage into and removal from the queue 2183 * and repetition of various conversions. 2184 * Set DELACK for segments received in order, but ack immediately 2185 * when segments are out of order (so fast retransmit can work). 2186 */ 2187 if (th->th_seq == tp->rcv_nxt && 2188 LIST_EMPTY(&tp->t_segq) && 2189 TCPS_HAVEESTABLISHED(tp->t_state)) { 2190 if (DELAY_ACK(tp)) 2191 callout_reset(tp->tt_delack, tcp_delacktime, 2192 tcp_timer_delack, tp); 2193 else 2194 tp->t_flags |= TF_ACKNOW; 2195 tp->rcv_nxt += tlen; 2196 thflags = th->th_flags & TH_FIN; 2197 tcpstat.tcps_rcvpack++; 2198 tcpstat.tcps_rcvbyte += tlen; 2199 ND6_HINT(tp); 2200 sbappend(&so->so_rcv, m); 2201 sorwakeup(so); 2202 } else { 2203 thflags = tcp_reass(tp, th, &tlen, m); 2204 tp->t_flags |= TF_ACKNOW; 2205 } 2206 2207 /* 2208 * Note the amount of data that peer has sent into 2209 * our window, in order to estimate the sender's 2210 * buffer size. 2211 */ 2212 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2213 } else { 2214 m_freem(m); 2215 thflags &= ~TH_FIN; 2216 } 2217 2218 /* 2219 * If FIN is received ACK the FIN and let the user know 2220 * that the connection is closing. 2221 */ 2222 if (thflags & TH_FIN) { 2223 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2224 socantrcvmore(so); 2225 /* 2226 * If connection is half-synchronized 2227 * (ie NEEDSYN flag on) then delay ACK, 2228 * so it may be piggybacked when SYN is sent. 2229 * Otherwise, since we received a FIN then no 2230 * more input can be expected, send ACK now. 2231 */ 2232 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2233 callout_reset(tp->tt_delack, tcp_delacktime, 2234 tcp_timer_delack, tp); 2235 else 2236 tp->t_flags |= TF_ACKNOW; 2237 tp->rcv_nxt++; 2238 } 2239 switch (tp->t_state) { 2240 2241 /* 2242 * In SYN_RECEIVED and ESTABLISHED STATES 2243 * enter the CLOSE_WAIT state. 2244 */ 2245 case TCPS_SYN_RECEIVED: 2246 tp->t_starttime = ticks; 2247 /*FALLTHROUGH*/ 2248 case TCPS_ESTABLISHED: 2249 tp->t_state = TCPS_CLOSE_WAIT; 2250 break; 2251 2252 /* 2253 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2254 * enter the CLOSING state. 2255 */ 2256 case TCPS_FIN_WAIT_1: 2257 tp->t_state = TCPS_CLOSING; 2258 break; 2259 2260 /* 2261 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2262 * starting the time-wait timer, turning off the other 2263 * standard timers. 2264 */ 2265 case TCPS_FIN_WAIT_2: 2266 tp->t_state = TCPS_TIME_WAIT; 2267 tcp_canceltimers(tp); 2268 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2269 if (tp->cc_recv != 0 && 2270 (ticks - tp->t_starttime) < tcp_msl) { 2271 callout_reset(tp->tt_2msl, 2272 tp->t_rxtcur * TCPTV_TWTRUNC, 2273 tcp_timer_2msl, tp); 2274 /* For transaction client, force ACK now. */ 2275 tp->t_flags |= TF_ACKNOW; 2276 } 2277 else 2278 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2279 tcp_timer_2msl, tp); 2280 soisdisconnected(so); 2281 break; 2282 2283 /* 2284 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2285 */ 2286 case TCPS_TIME_WAIT: 2287 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2288 tcp_timer_2msl, tp); 2289 break; 2290 } 2291 } 2292#ifdef TCPDEBUG 2293 if (so->so_options & SO_DEBUG) 2294 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2295 &tcp_savetcp, 0); 2296#endif 2297 2298 /* 2299 * Return any desired output. 2300 */ 2301 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2302 (void) tcp_output(tp); 2303 return; 2304 2305dropafterack: 2306 /* 2307 * Generate an ACK dropping incoming segment if it occupies 2308 * sequence space, where the ACK reflects our state. 2309 * 2310 * We can now skip the test for the RST flag since all 2311 * paths to this code happen after packets containing 2312 * RST have been dropped. 2313 * 2314 * In the SYN-RECEIVED state, don't send an ACK unless the 2315 * segment we received passes the SYN-RECEIVED ACK test. 2316 * If it fails send a RST. This breaks the loop in the 2317 * "LAND" DoS attack, and also prevents an ACK storm 2318 * between two listening ports that have been sent forged 2319 * SYN segments, each with the source address of the other. 2320 */ 2321 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2322 (SEQ_GT(tp->snd_una, th->th_ack) || 2323 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2324 rstreason = BANDLIM_RST_OPENPORT; 2325 goto dropwithreset; 2326 } 2327#ifdef TCPDEBUG 2328 if (so->so_options & SO_DEBUG) 2329 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2330 &tcp_savetcp, 0); 2331#endif 2332 m_freem(m); 2333 tp->t_flags |= TF_ACKNOW; 2334 (void) tcp_output(tp); 2335 return; 2336 2337dropwithreset: 2338 /* 2339 * Generate a RST, dropping incoming segment. 2340 * Make ACK acceptable to originator of segment. 2341 * Don't bother to respond if destination was broadcast/multicast. 2342 */ 2343 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2344 goto drop; 2345#ifdef INET6 2346 if (isipv6) { 2347 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2348 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2349 goto drop; 2350 } else 2351#endif /* INET6 */ 2352 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2353 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2354 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 2355 goto drop; 2356 /* IPv6 anycast check is done at tcp6_input() */ 2357 2358 /* 2359 * Perform bandwidth limiting. 2360 */ 2361 if (badport_bandlim(rstreason) < 0) 2362 goto drop; 2363 2364#ifdef TCPDEBUG 2365 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2366 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2367 &tcp_savetcp, 0); 2368#endif 2369 if (thflags & TH_ACK) 2370 /* mtod() below is safe as long as hdr dropping is delayed */ 2371 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2372 TH_RST); 2373 else { 2374 if (thflags & TH_SYN) 2375 tlen++; 2376 /* mtod() below is safe as long as hdr dropping is delayed */ 2377 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2378 (tcp_seq)0, TH_RST|TH_ACK); 2379 } 2380 /* destroy temporarily created socket */ 2381 if (dropsocket) 2382 (void) soabort(so); 2383 return; 2384 2385drop: 2386 /* 2387 * Drop space held by incoming segment and return. 2388 */ 2389#ifdef TCPDEBUG 2390 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2391 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2392 &tcp_savetcp, 0); 2393#endif 2394 m_freem(m); 2395 /* destroy temporarily created socket */ 2396 if (dropsocket) 2397 (void) soabort(so); 2398 return; 2399} 2400 2401static void 2402tcp_dooptions(tp, cp, cnt, th, to) 2403 struct tcpcb *tp; 2404 u_char *cp; 2405 int cnt; 2406 struct tcphdr *th; 2407 struct tcpopt *to; 2408{ 2409 u_short mss = 0; 2410 int opt, optlen; 2411 2412 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2413 opt = cp[0]; 2414 if (opt == TCPOPT_EOL) 2415 break; 2416 if (opt == TCPOPT_NOP) 2417 optlen = 1; 2418 else { 2419 if (cnt < 2) 2420 break; 2421 optlen = cp[1]; 2422 if (optlen < 2 || optlen > cnt) 2423 break; 2424 } 2425 switch (opt) { 2426 2427 default: 2428 continue; 2429 2430 case TCPOPT_MAXSEG: 2431 if (optlen != TCPOLEN_MAXSEG) 2432 continue; 2433 if (!(th->th_flags & TH_SYN)) 2434 continue; 2435 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2436 NTOHS(mss); 2437 break; 2438 2439 case TCPOPT_WINDOW: 2440 if (optlen != TCPOLEN_WINDOW) 2441 continue; 2442 if (!(th->th_flags & TH_SYN)) 2443 continue; 2444 tp->t_flags |= TF_RCVD_SCALE; 2445 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2446 break; 2447 2448 case TCPOPT_TIMESTAMP: 2449 if (optlen != TCPOLEN_TIMESTAMP) 2450 continue; 2451 to->to_flag |= TOF_TS; 2452 bcopy((char *)cp + 2, 2453 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2454 NTOHL(to->to_tsval); 2455 bcopy((char *)cp + 6, 2456 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2457 NTOHL(to->to_tsecr); 2458 2459 /* 2460 * A timestamp received in a SYN makes 2461 * it ok to send timestamp requests and replies. 2462 */ 2463 if (th->th_flags & TH_SYN) { 2464 tp->t_flags |= TF_RCVD_TSTMP; 2465 tp->ts_recent = to->to_tsval; 2466 tp->ts_recent_age = ticks; 2467 } 2468 break; 2469 case TCPOPT_CC: 2470 if (optlen != TCPOLEN_CC) 2471 continue; 2472 to->to_flag |= TOF_CC; 2473 bcopy((char *)cp + 2, 2474 (char *)&to->to_cc, sizeof(to->to_cc)); 2475 NTOHL(to->to_cc); 2476 /* 2477 * A CC or CC.new option received in a SYN makes 2478 * it ok to send CC in subsequent segments. 2479 */ 2480 if (th->th_flags & TH_SYN) 2481 tp->t_flags |= TF_RCVD_CC; 2482 break; 2483 case TCPOPT_CCNEW: 2484 if (optlen != TCPOLEN_CC) 2485 continue; 2486 if (!(th->th_flags & TH_SYN)) 2487 continue; 2488 to->to_flag |= TOF_CCNEW; 2489 bcopy((char *)cp + 2, 2490 (char *)&to->to_cc, sizeof(to->to_cc)); 2491 NTOHL(to->to_cc); 2492 /* 2493 * A CC or CC.new option received in a SYN makes 2494 * it ok to send CC in subsequent segments. 2495 */ 2496 tp->t_flags |= TF_RCVD_CC; 2497 break; 2498 case TCPOPT_CCECHO: 2499 if (optlen != TCPOLEN_CC) 2500 continue; 2501 if (!(th->th_flags & TH_SYN)) 2502 continue; 2503 to->to_flag |= TOF_CCECHO; 2504 bcopy((char *)cp + 2, 2505 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2506 NTOHL(to->to_ccecho); 2507 break; 2508 } 2509 } 2510 if (th->th_flags & TH_SYN) 2511 tcp_mss(tp, mss); /* sets t_maxseg */ 2512} 2513 2514/* 2515 * Pull out of band byte out of a segment so 2516 * it doesn't appear in the user's data queue. 2517 * It is still reflected in the segment length for 2518 * sequencing purposes. 2519 */ 2520static void 2521tcp_pulloutofband(so, th, m, off) 2522 struct socket *so; 2523 struct tcphdr *th; 2524 register struct mbuf *m; 2525 int off; /* delayed to be droped hdrlen */ 2526{ 2527 int cnt = off + th->th_urp - 1; 2528 2529 while (cnt >= 0) { 2530 if (m->m_len > cnt) { 2531 char *cp = mtod(m, caddr_t) + cnt; 2532 struct tcpcb *tp = sototcpcb(so); 2533 2534 tp->t_iobc = *cp; 2535 tp->t_oobflags |= TCPOOB_HAVEDATA; 2536 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2537 m->m_len--; 2538 if (m->m_flags & M_PKTHDR) 2539 m->m_pkthdr.len--; 2540 return; 2541 } 2542 cnt -= m->m_len; 2543 m = m->m_next; 2544 if (m == 0) 2545 break; 2546 } 2547 panic("tcp_pulloutofband"); 2548} 2549 2550/* 2551 * Collect new round-trip time estimate 2552 * and update averages and current timeout. 2553 */ 2554static void 2555tcp_xmit_timer(tp, rtt) 2556 register struct tcpcb *tp; 2557 int rtt; 2558{ 2559 register int delta; 2560 2561 tcpstat.tcps_rttupdated++; 2562 tp->t_rttupdated++; 2563 if (tp->t_srtt != 0) { 2564 /* 2565 * srtt is stored as fixed point with 5 bits after the 2566 * binary point (i.e., scaled by 8). The following magic 2567 * is equivalent to the smoothing algorithm in rfc793 with 2568 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2569 * point). Adjust rtt to origin 0. 2570 */ 2571 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2572 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2573 2574 if ((tp->t_srtt += delta) <= 0) 2575 tp->t_srtt = 1; 2576 2577 /* 2578 * We accumulate a smoothed rtt variance (actually, a 2579 * smoothed mean difference), then set the retransmit 2580 * timer to smoothed rtt + 4 times the smoothed variance. 2581 * rttvar is stored as fixed point with 4 bits after the 2582 * binary point (scaled by 16). The following is 2583 * equivalent to rfc793 smoothing with an alpha of .75 2584 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2585 * rfc793's wired-in beta. 2586 */ 2587 if (delta < 0) 2588 delta = -delta; 2589 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2590 if ((tp->t_rttvar += delta) <= 0) 2591 tp->t_rttvar = 1; 2592 } else { 2593 /* 2594 * No rtt measurement yet - use the unsmoothed rtt. 2595 * Set the variance to half the rtt (so our first 2596 * retransmit happens at 3*rtt). 2597 */ 2598 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2599 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2600 } 2601 tp->t_rtttime = 0; 2602 tp->t_rxtshift = 0; 2603 2604 /* 2605 * the retransmit should happen at rtt + 4 * rttvar. 2606 * Because of the way we do the smoothing, srtt and rttvar 2607 * will each average +1/2 tick of bias. When we compute 2608 * the retransmit timer, we want 1/2 tick of rounding and 2609 * 1 extra tick because of +-1/2 tick uncertainty in the 2610 * firing of the timer. The bias will give us exactly the 2611 * 1.5 tick we need. But, because the bias is 2612 * statistical, we have to test that we don't drop below 2613 * the minimum feasible timer (which is 2 ticks). 2614 */ 2615 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2616 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2617 2618 /* 2619 * We received an ack for a packet that wasn't retransmitted; 2620 * it is probably safe to discard any error indications we've 2621 * received recently. This isn't quite right, but close enough 2622 * for now (a route might have failed after we sent a segment, 2623 * and the return path might not be symmetrical). 2624 */ 2625 tp->t_softerror = 0; 2626} 2627 2628/* 2629 * Determine a reasonable value for maxseg size. 2630 * If the route is known, check route for mtu. 2631 * If none, use an mss that can be handled on the outgoing 2632 * interface without forcing IP to fragment; if bigger than 2633 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2634 * to utilize large mbufs. If no route is found, route has no mtu, 2635 * or the destination isn't local, use a default, hopefully conservative 2636 * size (usually 512 or the default IP max size, but no more than the mtu 2637 * of the interface), as we can't discover anything about intervening 2638 * gateways or networks. We also initialize the congestion/slow start 2639 * window to be a single segment if the destination isn't local. 2640 * While looking at the routing entry, we also initialize other path-dependent 2641 * parameters from pre-set or cached values in the routing entry. 2642 * 2643 * Also take into account the space needed for options that we 2644 * send regularly. Make maxseg shorter by that amount to assure 2645 * that we can send maxseg amount of data even when the options 2646 * are present. Store the upper limit of the length of options plus 2647 * data in maxopd. 2648 * 2649 * NOTE that this routine is only called when we process an incoming 2650 * segment, for outgoing segments only tcp_mssopt is called. 2651 * 2652 * In case of T/TCP, we call this routine during implicit connection 2653 * setup as well (offer = -1), to initialize maxseg from the cached 2654 * MSS of our peer. 2655 */ 2656void 2657tcp_mss(tp, offer) 2658 struct tcpcb *tp; 2659 int offer; 2660{ 2661 register struct rtentry *rt; 2662 struct ifnet *ifp; 2663 register int rtt, mss; 2664 u_long bufsize; 2665 struct inpcb *inp; 2666 struct socket *so; 2667 struct rmxp_tao *taop; 2668 int origoffer = offer; 2669#ifdef INET6 2670 int isipv6; 2671 int min_protoh; 2672#endif 2673 2674 inp = tp->t_inpcb; 2675#ifdef INET6 2676 isipv6 = ((inp->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(inp); 2685 else 2686#endif 2687 rt = tcp_rtlookup(inp); 2688 if (rt == NULL) { 2689 tp->t_maxopd = tp->t_maxseg = 2690#ifdef INET6 2691 isipv6 ? tcp_v6mssdflt : 2692#endif /* INET6 */ 2693 tcp_mssdflt; 2694 return; 2695 } 2696 ifp = rt->rt_ifp; 2697 so = inp->inp_socket; 2698 2699 taop = rmx_taop(rt->rt_rmx); 2700 /* 2701 * Offer == -1 means that we didn't receive SYN yet, 2702 * use cached value in that case; 2703 */ 2704 if (offer == -1) 2705 offer = taop->tao_mssopt; 2706 /* 2707 * Offer == 0 means that there was no MSS on the SYN segment, 2708 * in this case we use tcp_mssdflt. 2709 */ 2710 if (offer == 0) 2711 offer = 2712#ifdef INET6 2713 isipv6 ? tcp_v6mssdflt : 2714#endif /* INET6 */ 2715 tcp_mssdflt; 2716 else 2717 /* 2718 * Sanity check: make sure that maxopd will be large 2719 * enough to allow some data on segments even is the 2720 * all the option space is used (40bytes). Otherwise 2721 * funny things may happen in tcp_output. 2722 */ 2723 offer = max(offer, 64); 2724 taop->tao_mssopt = offer; 2725 2726 /* 2727 * While we're here, check if there's an initial rtt 2728 * or rttvar. Convert from the route-table units 2729 * to scaled multiples of the slow timeout timer. 2730 */ 2731 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2732 /* 2733 * XXX the lock bit for RTT indicates that the value 2734 * is also a minimum value; this is subject to time. 2735 */ 2736 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2737 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2738 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2739 tcpstat.tcps_usedrtt++; 2740 if (rt->rt_rmx.rmx_rttvar) { 2741 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2742 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2743 tcpstat.tcps_usedrttvar++; 2744 } else { 2745 /* default variation is +- 1 rtt */ 2746 tp->t_rttvar = 2747 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2748 } 2749 TCPT_RANGESET(tp->t_rxtcur, 2750 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2751 tp->t_rttmin, TCPTV_REXMTMAX); 2752 } 2753 /* 2754 * if there's an mtu associated with the route, use it 2755 * else, use the link mtu. 2756 */ 2757 if (rt->rt_rmx.rmx_mtu) 2758 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2759 else 2760 { 2761 mss = 2762#ifdef INET6 2763 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2764#endif 2765 ifp->if_mtu 2766#ifdef INET6 2767 ) 2768#endif 2769 - min_protoh; 2770#ifdef INET6 2771 if (isipv6) { 2772 if (!in6_localaddr(&inp->in6p_faddr)) 2773 mss = min(mss, tcp_v6mssdflt); 2774 } else 2775#endif 2776 if (!in_localaddr(inp->inp_faddr)) 2777 mss = min(mss, tcp_mssdflt); 2778 } 2779 mss = min(mss, offer); 2780 /* 2781 * maxopd stores the maximum length of data AND options 2782 * in a segment; maxseg is the amount of data in a normal 2783 * segment. We need to store this value (maxopd) apart 2784 * from maxseg, because now every segment carries options 2785 * and thus we normally have somewhat less data in segments. 2786 */ 2787 tp->t_maxopd = mss; 2788 2789 /* 2790 * In case of T/TCP, origoffer==-1 indicates, that no segments 2791 * were received yet. In this case we just guess, otherwise 2792 * we do the same as before T/TCP. 2793 */ 2794 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2795 (origoffer == -1 || 2796 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2797 mss -= TCPOLEN_TSTAMP_APPA; 2798 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2799 (origoffer == -1 || 2800 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2801 mss -= TCPOLEN_CC_APPA; 2802 2803#if (MCLBYTES & (MCLBYTES - 1)) == 0 2804 if (mss > MCLBYTES) 2805 mss &= ~(MCLBYTES-1); 2806#else 2807 if (mss > MCLBYTES) 2808 mss = mss / MCLBYTES * MCLBYTES; 2809#endif 2810 /* 2811 * If there's a pipesize, change the socket buffer 2812 * to that size. Make the socket buffers an integral 2813 * number of mss units; if the mss is larger than 2814 * the socket buffer, decrease the mss. 2815 */ 2816#ifdef RTV_SPIPE 2817 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2818#endif 2819 bufsize = so->so_snd.sb_hiwat; 2820 if (bufsize < mss) 2821 mss = bufsize; 2822 else { 2823 bufsize = roundup(bufsize, mss); 2824 if (bufsize > sb_max) 2825 bufsize = sb_max; 2826 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2827 } 2828 tp->t_maxseg = mss; 2829 2830#ifdef RTV_RPIPE 2831 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2832#endif 2833 bufsize = so->so_rcv.sb_hiwat; 2834 if (bufsize > mss) { 2835 bufsize = roundup(bufsize, mss); 2836 if (bufsize > sb_max) 2837 bufsize = sb_max; 2838 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2839 } 2840 2841 /* 2842 * Set the slow-start flight size depending on whether this 2843 * is a local network or not. 2844 */ 2845 if ( 2846#ifdef INET6 2847 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2848 (!isipv6 && 2849#endif 2850 in_localaddr(inp->inp_faddr) 2851#ifdef INET6 2852 ) 2853#endif 2854 ) 2855 tp->snd_cwnd = mss * ss_fltsz_local; 2856 else 2857 tp->snd_cwnd = mss * ss_fltsz; 2858 2859 if (rt->rt_rmx.rmx_ssthresh) { 2860 /* 2861 * There's some sort of gateway or interface 2862 * buffer limit on the path. Use this to set 2863 * the slow start threshhold, but set the 2864 * threshold to no less than 2*mss. 2865 */ 2866 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2867 tcpstat.tcps_usedssthresh++; 2868 } 2869} 2870 2871/* 2872 * Determine the MSS option to send on an outgoing SYN. 2873 */ 2874int 2875tcp_mssopt(tp) 2876 struct tcpcb *tp; 2877{ 2878 struct rtentry *rt; 2879#ifdef INET6 2880 int isipv6; 2881 int min_protoh; 2882#endif 2883 2884#ifdef INET6 2885 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2886 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2887 : sizeof (struct tcpiphdr); 2888#else 2889#define min_protoh (sizeof (struct tcpiphdr)) 2890#endif 2891#ifdef INET6 2892 if (isipv6) 2893 rt = tcp_rtlookup6(tp->t_inpcb); 2894 else 2895#endif /* INET6 */ 2896 rt = tcp_rtlookup(tp->t_inpcb); 2897 if (rt == NULL) 2898 return 2899#ifdef INET6 2900 isipv6 ? tcp_v6mssdflt : 2901#endif /* INET6 */ 2902 tcp_mssdflt; 2903 2904 return rt->rt_ifp->if_mtu - min_protoh; 2905} 2906 2907 2908/* 2909 * Checks for partial ack. If partial ack arrives, force the retransmission 2910 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2911 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2912 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2913 */ 2914static int 2915tcp_newreno(tp, th) 2916 struct tcpcb *tp; 2917 struct tcphdr *th; 2918{ 2919 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2920 tcp_seq onxt = tp->snd_nxt; 2921 u_long ocwnd = tp->snd_cwnd; 2922 2923 callout_stop(tp->tt_rexmt); 2924 tp->t_rtttime = 0; 2925 tp->snd_nxt = th->th_ack; 2926 /* 2927 * Set snd_cwnd to one segment beyond acknowledged offset 2928 * (tp->snd_una has not yet been updated when this function 2929 * is called) 2930 */ 2931 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2932 (void) tcp_output(tp); 2933 tp->snd_cwnd = ocwnd; 2934 if (SEQ_GT(onxt, tp->snd_nxt)) 2935 tp->snd_nxt = onxt; 2936 /* 2937 * Partial window deflation. Relies on fact that tp->snd_una 2938 * not updated yet. 2939 */ 2940 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2941 return (1); 2942 } 2943 return (0); 2944} 2945