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 * $FreeBSD$ 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/kernel.h> 45 #include <sys/malloc.h> 46 #include <sys/mbuf.h> 47 #include <sys/proc.h> /* for proc0 declaration */ 48 #include <sys/protosw.h> 49 #include <sys/signalvar.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/sysctl.h> 53 #include <sys/syslog.h> 54 #include <sys/systm.h> 55 56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58 #include <net/if.h> 59 #include <net/route.h> 60 61 #include <netinet/in.h> 62 #include <netinet/in_pcb.h> 63 #include <netinet/in_systm.h> 64 #include <netinet/in_var.h> 65 #include <netinet/ip.h> 66 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 67 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 68 #include <netinet/ip_var.h> 69 #ifdef INET6 70 #include <netinet/ip6.h> 71 #include <netinet/icmp6.h> 72 #include <netinet6/in6_pcb.h> 73 #include <netinet6/ip6_var.h> 74 #include <netinet6/nd6.h> 75 #endif 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #ifdef INET6 82 #include <netinet6/tcp6_var.h> 83 #endif 84 #include <netinet/tcpip.h> 85 #ifdef TCPDEBUG 86 #include <netinet/tcp_debug.h> 87 88 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 89 struct tcphdr tcp_savetcp; 90 #endif /* TCPDEBUG */ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #ifdef INET6 95 #include <netinet6/ipsec6.h> 96 #endif 97 #include <netkey/key.h> 98 #endif /*IPSEC*/ 99 100 #include <machine/in_cksum.h> 101 102 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 103 104 static int tcprexmtthresh = 3; 105 tcp_cc tcp_ccgen; 106 107 struct tcpstat tcpstat; 108 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 109 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 110 111 static int log_in_vain = 0; 112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 113 &log_in_vain, 0, "Log all incoming TCP connections"); 114 115 static int blackhole = 0; 116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 117 &blackhole, 0, "Do not send RST when dropping refused connections"); 118 119 int tcp_delack_enabled = 1; 120 SYSCTL_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"); 123 124 #ifdef TCP_DROP_SYNFIN 125 static int drop_synfin = 0; 126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 127 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 128 #endif 129 130 struct inpcbhead tcb; 131 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 132 struct inpcbinfo tcbinfo; 133 134 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 135 static void tcp_pulloutofband(struct socket *, 136 struct tcphdr *, struct mbuf *, int); 137 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 138 struct mbuf *); 139 static void tcp_xmit_timer(struct tcpcb *, int); 140 static int tcp_newreno(struct tcpcb *, struct tcphdr *); 141 142 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 143 #ifdef INET6 144 #define ND6_HINT(tp) \ 145 do { \ 146 if ((tp) && (tp)->t_inpcb && \ 147 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 148 (tp)->t_inpcb->in6p_route.ro_rt) \ 149 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 150 } while (0) 151 #else 152 #define ND6_HINT(tp) 153 #endif 154 155 /* 156 * Indicate whether this ack should be delayed. We can delay the ack if 157 * - delayed acks are enabled and 158 * - there is no delayed ack timer in progress and 159 * - our last ack wasn't a 0-sized window. We never want to delay 160 * the ack that opens up a 0-sized window. 161 */ 162 #define DELAY_ACK(tp) \ 163 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 164 (tp->t_flags & TF_RXWIN0SENT) == 0) 165 166 static int 167 tcp_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 } 268 269 present: 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 301 int 302 tcp6_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); 326 return IPPROTO_DONE; 327 } 328 #endif 329 330 void 331 tcp_input(m, off0) 332 register struct mbuf *m; 333 int off0; 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 u_long tiwin; 348 struct tcpopt to; /* options in this segment */ 349 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 350 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 351 #ifdef TCPDEBUG 352 short ostate = 0; 353 #endif 354 #ifdef INET6 355 struct ip6_hdr *ip6 = NULL; 356 int isipv6; 357 #endif /* INET6 */ 358 int rstreason; /* For badport_bandlim accounting purposes */ 359 360 #ifdef INET6 361 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 362 #endif 363 bzero((char *)&to, sizeof(to)); 364 365 tcpstat.tcps_rcvtotal++; 366 367 #ifdef INET6 368 if (isipv6) { 369 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 370 ip6 = mtod(m, struct ip6_hdr *); 371 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 372 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 373 tcpstat.tcps_rcvbadsum++; 374 goto drop; 375 } 376 th = (struct tcphdr *)((caddr_t)ip6 + off0); 377 378 /* 379 * Be proactive about unspecified IPv6 address in source. 380 * As we use all-zero to indicate unbounded/unconnected pcb, 381 * unspecified IPv6 address can be used to confuse us. 382 * 383 * Note that packets with unspecified IPv6 destination is 384 * already dropped in ip6_input. 385 */ 386 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 387 /* XXX stat */ 388 goto drop; 389 } 390 } else 391 #endif /* INET6 */ 392 { 393 /* 394 * Get IP and TCP header together in first mbuf. 395 * Note: IP leaves IP header in first mbuf. 396 */ 397 if (off0 > sizeof (struct ip)) { 398 ip_stripoptions(m, (struct mbuf *)0); 399 off0 = sizeof(struct ip); 400 } 401 if (m->m_len < sizeof (struct tcpiphdr)) { 402 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 403 tcpstat.tcps_rcvshort++; 404 return; 405 } 406 } 407 ip = mtod(m, struct ip *); 408 ipov = (struct ipovly *)ip; 409 th = (struct tcphdr *)((caddr_t)ip + off0); 410 tlen = ip->ip_len; 411 412 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 413 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 414 th->th_sum = m->m_pkthdr.csum_data; 415 else 416 th->th_sum = in_pseudo(ip->ip_src.s_addr, 417 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 418 ip->ip_len + IPPROTO_TCP)); 419 th->th_sum ^= 0xffff; 420 } else { 421 /* 422 * Checksum extended TCP header and data. 423 */ 424 len = sizeof (struct ip) + tlen; 425 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 426 ipov->ih_len = (u_short)tlen; 427 ipov->ih_len = htons(ipov->ih_len); 428 th->th_sum = in_cksum(m, len); 429 } 430 if (th->th_sum) { 431 tcpstat.tcps_rcvbadsum++; 432 goto drop; 433 } 434 #ifdef INET6 435 /* Re-initialization for later version check */ 436 ip->ip_v = IPVERSION; 437 #endif 438 } 439 440 /* 441 * Check that TCP offset makes sense, 442 * pull out TCP options and adjust length. XXX 443 */ 444 off = th->th_off << 2; 445 if (off < sizeof (struct tcphdr) || off > tlen) { 446 tcpstat.tcps_rcvbadoff++; 447 goto drop; 448 } 449 tlen -= off; /* tlen is used instead of ti->ti_len */ 450 if (off > sizeof (struct tcphdr)) { 451 #ifdef INET6 452 if (isipv6) { 453 IP6_EXTHDR_CHECK(m, off0, off, ); 454 ip6 = mtod(m, struct ip6_hdr *); 455 th = (struct tcphdr *)((caddr_t)ip6 + off0); 456 } else 457 #endif /* INET6 */ 458 { 459 if (m->m_len < sizeof(struct ip) + off) { 460 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 461 tcpstat.tcps_rcvshort++; 462 return; 463 } 464 ip = mtod(m, struct ip *); 465 ipov = (struct ipovly *)ip; 466 th = (struct tcphdr *)((caddr_t)ip + off0); 467 } 468 } 469 optlen = off - sizeof (struct tcphdr); 470 optp = (u_char *)(th + 1); 471 } 472 thflags = th->th_flags; 473 474 #ifdef TCP_DROP_SYNFIN 475 /* 476 * If the drop_synfin option is enabled, drop all packets with 477 * both the SYN and FIN bits set. This prevents e.g. nmap from 478 * identifying the TCP/IP stack. 479 * 480 * This is a violation of the TCP specification. 481 */ 482 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 483 goto drop; 484 #endif 485 486 /* 487 * Convert TCP protocol specific fields to host format. 488 */ 489 th->th_seq = ntohl(th->th_seq); 490 th->th_ack = ntohl(th->th_ack); 491 th->th_win = ntohs(th->th_win); 492 th->th_urp = ntohs(th->th_urp); 493 494 /* 495 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 496 * until after ip6_savecontrol() is called and before other functions 497 * which don't want those proto headers. 498 * Because ip6_savecontrol() is going to parse the mbuf to 499 * search for data to be passed up to user-land, it wants mbuf 500 * parameters to be unchanged. 501 * XXX: the call of ip6_savecontrol() has been obsoleted based on 502 * latest version of the advanced API (20020110). 503 */ 504 drop_hdrlen = off0 + off; 505 506 /* 507 * Locate pcb for segment. 508 */ 509 findpcb: 510 #ifdef IPFIREWALL_FORWARD 511 if (ip_fw_fwd_addr != NULL 512 #ifdef INET6 513 && isipv6 == NULL /* IPv6 support is not yet */ 514 #endif /* INET6 */ 515 ) { 516 /* 517 * Diverted. Pretend to be the destination. 518 * already got one like this? 519 */ 520 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 521 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 522 if (!inp) { 523 /* 524 * No, then it's new. Try find the ambushing socket 525 */ 526 if (!ip_fw_fwd_addr->sin_port) { 527 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 528 th->th_sport, ip_fw_fwd_addr->sin_addr, 529 th->th_dport, 1, m->m_pkthdr.rcvif); 530 } else { 531 inp = in_pcblookup_hash(&tcbinfo, 532 ip->ip_src, th->th_sport, 533 ip_fw_fwd_addr->sin_addr, 534 ntohs(ip_fw_fwd_addr->sin_port), 1, 535 m->m_pkthdr.rcvif); 536 } 537 } 538 ip_fw_fwd_addr = NULL; 539 } else 540 #endif /* IPFIREWALL_FORWARD */ 541 { 542 #ifdef INET6 543 if (isipv6) 544 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 545 &ip6->ip6_dst, th->th_dport, 1, 546 m->m_pkthdr.rcvif); 547 else 548 #endif /* INET6 */ 549 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 550 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 551 } 552 553 #ifdef IPSEC 554 #ifdef INET6 555 if (isipv6) { 556 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 557 ipsec6stat.in_polvio++; 558 goto drop; 559 } 560 } else 561 #endif /* INET6 */ 562 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 563 ipsecstat.in_polvio++; 564 goto drop; 565 } 566 #endif /*IPSEC*/ 567 568 /* 569 * If the state is CLOSED (i.e., TCB does not exist) then 570 * all data in the incoming segment is discarded. 571 * If the TCB exists but is in CLOSED state, it is embryonic, 572 * but should either do a listen or a connect soon. 573 */ 574 if (inp == NULL) { 575 if (log_in_vain) { 576 #ifdef INET6 577 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 578 #else /* INET6 */ 579 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 580 #endif /* INET6 */ 581 582 #ifdef INET6 583 if (isipv6) { 584 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 585 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 586 } else 587 #endif 588 { 589 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 590 strcpy(sbuf, inet_ntoa(ip->ip_src)); 591 } 592 switch (log_in_vain) { 593 case 1: 594 if(thflags & TH_SYN) 595 log(LOG_INFO, 596 "Connection attempt to TCP %s:%d from %s:%d\n", 597 dbuf, ntohs(th->th_dport), 598 sbuf, 599 ntohs(th->th_sport)); 600 break; 601 case 2: 602 log(LOG_INFO, 603 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 604 dbuf, ntohs(th->th_dport), sbuf, 605 ntohs(th->th_sport), thflags); 606 break; 607 default: 608 break; 609 } 610 } 611 if (blackhole) { 612 switch (blackhole) { 613 case 1: 614 if (thflags & TH_SYN) 615 goto drop; 616 break; 617 case 2: 618 goto drop; 619 default: 620 goto drop; 621 } 622 } 623 rstreason = BANDLIM_RST_CLOSEDPORT; 624 goto dropwithreset; 625 } 626 tp = intotcpcb(inp); 627 if (tp == 0) { 628 rstreason = BANDLIM_RST_CLOSEDPORT; 629 goto dropwithreset; 630 } 631 if (tp->t_state == TCPS_CLOSED) 632 goto drop; 633 634 /* Unscale the window into a 32-bit value. */ 635 if ((thflags & TH_SYN) == 0) 636 tiwin = th->th_win << tp->snd_scale; 637 else 638 tiwin = th->th_win; 639 640 so = inp->inp_socket; 641 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 642 struct in_conninfo inc; 643 #ifdef TCPDEBUG 644 if (so->so_options & SO_DEBUG) { 645 ostate = tp->t_state; 646 #ifdef INET6 647 if (isipv6) 648 bcopy((char *)ip6, (char *)tcp_saveipgen, 649 sizeof(*ip6)); 650 else 651 #endif /* INET6 */ 652 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 653 tcp_savetcp = *th; 654 } 655 #endif 656 /* skip if this isn't a listen socket */ 657 if ((so->so_options & SO_ACCEPTCONN) == 0) 658 goto after_listen; 659 #ifdef INET6 660 inc.inc_isipv6 = isipv6; 661 if (isipv6) { 662 inc.inc6_faddr = ip6->ip6_src; 663 inc.inc6_laddr = ip6->ip6_dst; 664 inc.inc6_route.ro_rt = NULL; /* XXX */ 665 666 } else 667 #endif /* INET6 */ 668 { 669 inc.inc_faddr = ip->ip_src; 670 inc.inc_laddr = ip->ip_dst; 671 inc.inc_route.ro_rt = NULL; /* XXX */ 672 } 673 inc.inc_fport = th->th_sport; 674 inc.inc_lport = th->th_dport; 675 676 /* 677 * If the state is LISTEN then ignore segment if it contains 678 * a RST. If the segment contains an ACK then it is bad and 679 * send a RST. If it does not contain a SYN then it is not 680 * interesting; drop it. 681 * 682 * If the state is SYN_RECEIVED (syncache) and seg contains 683 * an ACK, but not for our SYN/ACK, send a RST. If the seg 684 * contains a RST, check the sequence number to see if it 685 * is a valid reset segment. 686 */ 687 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 688 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 689 if (!syncache_expand(&inc, th, &so, m)) { 690 /* 691 * No syncache entry, or ACK was not 692 * for our SYN/ACK. Send a RST. 693 */ 694 tcpstat.tcps_badsyn++; 695 rstreason = BANDLIM_RST_OPENPORT; 696 goto dropwithreset; 697 } 698 if (so == NULL) 699 /* 700 * Could not complete 3-way handshake, 701 * connection is being closed down, and 702 * syncache will free mbuf. 703 */ 704 return; 705 /* 706 * Socket is created in state SYN_RECEIVED. 707 * Continue processing segment. 708 */ 709 inp = sotoinpcb(so); 710 tp = intotcpcb(inp); 711 /* 712 * This is what would have happened in 713 * tcp_output() when the SYN,ACK was sent. 714 */ 715 tp->snd_up = tp->snd_una; 716 tp->snd_max = tp->snd_nxt = tp->iss + 1; 717 tp->last_ack_sent = tp->rcv_nxt; 718 /* 719 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 720 * until the _second_ ACK is received: 721 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 722 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 723 * move to ESTAB, set snd_wnd to tiwin. 724 */ 725 tp->snd_wnd = tiwin; /* unscaled */ 726 goto after_listen; 727 } 728 if (thflags & TH_RST) { 729 syncache_chkrst(&inc, th); 730 goto drop; 731 } 732 if (thflags & TH_ACK) { 733 syncache_badack(&inc); 734 tcpstat.tcps_badsyn++; 735 rstreason = BANDLIM_RST_OPENPORT; 736 goto dropwithreset; 737 } 738 goto drop; 739 } 740 741 /* 742 * Segment's flags are (SYN) or (SYN|FIN). 743 */ 744 #ifdef INET6 745 /* 746 * If deprecated address is forbidden, 747 * we do not accept SYN to deprecated interface 748 * address to prevent any new inbound connection from 749 * getting established. 750 * When we do not accept SYN, we send a TCP RST, 751 * with deprecated source address (instead of dropping 752 * it). We compromise it as it is much better for peer 753 * to send a RST, and RST will be the final packet 754 * for the exchange. 755 * 756 * If we do not forbid deprecated addresses, we accept 757 * the SYN packet. RFC2462 does not suggest dropping 758 * SYN in this case. 759 * If we decipher RFC2462 5.5.4, it says like this: 760 * 1. use of deprecated addr with existing 761 * communication is okay - "SHOULD continue to be 762 * used" 763 * 2. use of it with new communication: 764 * (2a) "SHOULD NOT be used if alternate address 765 * with sufficient scope is available" 766 * (2b) nothing mentioned otherwise. 767 * Here we fall into (2b) case as we have no choice in 768 * our source address selection - we must obey the peer. 769 * 770 * The wording in RFC2462 is confusing, and there are 771 * multiple description text for deprecated address 772 * handling - worse, they are not exactly the same. 773 * I believe 5.5.4 is the best one, so we follow 5.5.4. 774 */ 775 if (isipv6 && !ip6_use_deprecated) { 776 struct in6_ifaddr *ia6; 777 778 if ((ia6 = ip6_getdstifaddr(m)) && 779 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 780 tp = NULL; 781 rstreason = BANDLIM_RST_OPENPORT; 782 goto dropwithreset; 783 } 784 } 785 #endif 786 /* 787 * If it is from this socket, drop it, it must be forged. 788 * Don't bother responding if the destination was a broadcast. 789 */ 790 if (th->th_dport == th->th_sport) { 791 #ifdef INET6 792 if (isipv6) { 793 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 794 &ip6->ip6_src)) 795 goto drop; 796 } else 797 #endif /* INET6 */ 798 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 799 goto drop; 800 } 801 /* 802 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 803 * 804 * Note that it is quite possible to receive unicast 805 * link-layer packets with a broadcast IP address. Use 806 * in_broadcast() to find them. 807 */ 808 if (m->m_flags & (M_BCAST|M_MCAST)) 809 goto drop; 810 #ifdef INET6 811 if (isipv6) { 812 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 813 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 814 goto drop; 815 } else 816 #endif 817 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 818 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 819 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 820 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 821 goto drop; 822 /* 823 * SYN appears to be valid; create compressed TCP state 824 * for syncache, or perform t/tcp connection. 825 */ 826 if (so->so_qlen <= so->so_qlimit) { 827 tcp_dooptions(&to, optp, optlen, 1); 828 if (!syncache_add(&inc, &to, th, &so, m)) 829 goto drop; 830 if (so == NULL) 831 /* 832 * Entry added to syncache, mbuf used to 833 * send SYN,ACK packet. 834 */ 835 return; 836 /* 837 * Segment passed TAO tests. 838 */ 839 inp = sotoinpcb(so); 840 tp = intotcpcb(inp); 841 tp->snd_wnd = tiwin; 842 tp->t_starttime = ticks; 843 tp->t_state = TCPS_ESTABLISHED; 844 845 /* 846 * If there is a FIN, or if there is data and the 847 * connection is local, then delay SYN,ACK(SYN) in 848 * the hope of piggy-backing it on a response 849 * segment. Otherwise must send ACK now in case 850 * the other side is slow starting. 851 */ 852 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 853 (tlen != 0 && 854 #ifdef INET6 855 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 856 || 857 (!isipv6 && 858 #endif 859 in_localaddr(inp->inp_faddr) 860 #ifdef INET6 861 )) 862 #endif 863 ))) { 864 callout_reset(tp->tt_delack, tcp_delacktime, 865 tcp_timer_delack, tp); 866 tp->t_flags |= TF_NEEDSYN; 867 } else 868 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 869 870 tcpstat.tcps_connects++; 871 soisconnected(so); 872 goto trimthenstep6; 873 } 874 goto drop; 875 } 876 after_listen: 877 878 /* XXX temp debugging */ 879 /* should not happen - syncache should pick up these connections */ 880 if (tp->t_state == TCPS_LISTEN) 881 panic("tcp_input: TCPS_LISTEN"); 882 883 /* 884 * Segment received on connection. 885 * Reset idle time and keep-alive timer. 886 */ 887 tp->t_rcvtime = ticks; 888 if (TCPS_HAVEESTABLISHED(tp->t_state)) 889 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 890 891 /* 892 * Process options. 893 * XXX this is tradtitional behavior, may need to be cleaned up. 894 */ 895 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 896 if (thflags & TH_SYN) { 897 if (to.to_flags & TOF_SCALE) { 898 tp->t_flags |= TF_RCVD_SCALE; 899 tp->requested_s_scale = to.to_requested_s_scale; 900 } 901 if (to.to_flags & TOF_TS) { 902 tp->t_flags |= TF_RCVD_TSTMP; 903 tp->ts_recent = to.to_tsval; 904 tp->ts_recent_age = ticks; 905 } 906 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 907 tp->t_flags |= TF_RCVD_CC; 908 if (to.to_flags & TOF_MSS) 909 tcp_mss(tp, to.to_mss); 910 } 911 912 /* 913 * Header prediction: check for the two common cases 914 * of a uni-directional data xfer. If the packet has 915 * no control flags, is in-sequence, the window didn't 916 * change and we're not retransmitting, it's a 917 * candidate. If the length is zero and the ack moved 918 * forward, we're the sender side of the xfer. Just 919 * free the data acked & wake any higher level process 920 * that was blocked waiting for space. If the length 921 * is non-zero and the ack didn't move, we're the 922 * receiver side. If we're getting packets in-order 923 * (the reassembly queue is empty), add the data to 924 * the socket buffer and note that we need a delayed ack. 925 * Make sure that the hidden state-flags are also off. 926 * Since we check for TCPS_ESTABLISHED above, it can only 927 * be TH_NEEDSYN. 928 */ 929 if (tp->t_state == TCPS_ESTABLISHED && 930 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 931 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 932 ((to.to_flags & TOF_TS) == 0 || 933 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 934 /* 935 * Using the CC option is compulsory if once started: 936 * the segment is OK if no T/TCP was negotiated or 937 * if the segment has a CC option equal to CCrecv 938 */ 939 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 940 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 941 th->th_seq == tp->rcv_nxt && 942 tiwin && tiwin == tp->snd_wnd && 943 tp->snd_nxt == tp->snd_max) { 944 945 /* 946 * If last ACK falls within this segment's sequence numbers, 947 * record the timestamp. 948 * NOTE that the test is modified according to the latest 949 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 950 */ 951 if ((to.to_flags & TOF_TS) != 0 && 952 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 953 tp->ts_recent_age = ticks; 954 tp->ts_recent = to.to_tsval; 955 } 956 957 if (tlen == 0) { 958 if (SEQ_GT(th->th_ack, tp->snd_una) && 959 SEQ_LEQ(th->th_ack, tp->snd_max) && 960 tp->snd_cwnd >= tp->snd_wnd && 961 tp->t_dupacks < tcprexmtthresh) { 962 /* 963 * this is a pure ack for outstanding data. 964 */ 965 ++tcpstat.tcps_predack; 966 /* 967 * "bad retransmit" recovery 968 */ 969 if (tp->t_rxtshift == 1 && 970 ticks < tp->t_badrxtwin) { 971 tp->snd_cwnd = tp->snd_cwnd_prev; 972 tp->snd_ssthresh = 973 tp->snd_ssthresh_prev; 974 tp->snd_nxt = tp->snd_max; 975 tp->t_badrxtwin = 0; 976 } 977 if ((to.to_flags & TOF_TS) != 0) 978 tcp_xmit_timer(tp, 979 ticks - to.to_tsecr + 1); 980 else if (tp->t_rtttime && 981 SEQ_GT(th->th_ack, tp->t_rtseq)) 982 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 983 acked = th->th_ack - tp->snd_una; 984 tcpstat.tcps_rcvackpack++; 985 tcpstat.tcps_rcvackbyte += acked; 986 sbdrop(&so->so_snd, acked); 987 tp->snd_una = th->th_ack; 988 m_freem(m); 989 ND6_HINT(tp); /* some progress has been done */ 990 991 /* 992 * If all outstanding data are acked, stop 993 * retransmit timer, otherwise restart timer 994 * using current (possibly backed-off) value. 995 * If process is waiting for space, 996 * wakeup/selwakeup/signal. If data 997 * are ready to send, let tcp_output 998 * decide between more output or persist. 999 */ 1000 if (tp->snd_una == tp->snd_max) 1001 callout_stop(tp->tt_rexmt); 1002 else if (!callout_active(tp->tt_persist)) 1003 callout_reset(tp->tt_rexmt, 1004 tp->t_rxtcur, 1005 tcp_timer_rexmt, tp); 1006 1007 sowwakeup(so); 1008 if (so->so_snd.sb_cc) 1009 (void) tcp_output(tp); 1010 return; 1011 } 1012 } else if (th->th_ack == tp->snd_una && 1013 LIST_EMPTY(&tp->t_segq) && 1014 tlen <= sbspace(&so->so_rcv)) { 1015 /* 1016 * this is a pure, in-sequence data packet 1017 * with nothing on the reassembly queue and 1018 * we have enough buffer space to take it. 1019 */ 1020 ++tcpstat.tcps_preddat; 1021 tp->rcv_nxt += tlen; 1022 tcpstat.tcps_rcvpack++; 1023 tcpstat.tcps_rcvbyte += tlen; 1024 ND6_HINT(tp); /* some progress has been done */ 1025 /* 1026 * Add data to socket buffer. 1027 */ 1028 m_adj(m, drop_hdrlen); /* delayed header drop */ 1029 sbappend(&so->so_rcv, m); 1030 sorwakeup(so); 1031 if (DELAY_ACK(tp)) { 1032 callout_reset(tp->tt_delack, tcp_delacktime, 1033 tcp_timer_delack, tp); 1034 } else { 1035 tp->t_flags |= TF_ACKNOW; 1036 tcp_output(tp); 1037 } 1038 return; 1039 } 1040 } 1041 1042 /* 1043 * Calculate amount of space in receive window, 1044 * and then do TCP input processing. 1045 * Receive window is amount of space in rcv queue, 1046 * but not less than advertised window. 1047 */ 1048 { int win; 1049 1050 win = sbspace(&so->so_rcv); 1051 if (win < 0) 1052 win = 0; 1053 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1054 } 1055 1056 switch (tp->t_state) { 1057 1058 /* 1059 * If the state is SYN_RECEIVED: 1060 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1061 */ 1062 case TCPS_SYN_RECEIVED: 1063 if ((thflags & TH_ACK) && 1064 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1065 SEQ_GT(th->th_ack, tp->snd_max))) { 1066 rstreason = BANDLIM_RST_OPENPORT; 1067 goto dropwithreset; 1068 } 1069 break; 1070 1071 /* 1072 * If the state is SYN_SENT: 1073 * if seg contains an ACK, but not for our SYN, drop the input. 1074 * if seg contains a RST, then drop the connection. 1075 * if seg does not contain SYN, then drop it. 1076 * Otherwise this is an acceptable SYN segment 1077 * initialize tp->rcv_nxt and tp->irs 1078 * if seg contains ack then advance tp->snd_una 1079 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1080 * arrange for segment to be acked (eventually) 1081 * continue processing rest of data/controls, beginning with URG 1082 */ 1083 case TCPS_SYN_SENT: 1084 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1085 taop = &tao_noncached; 1086 bzero(taop, sizeof(*taop)); 1087 } 1088 1089 if ((thflags & TH_ACK) && 1090 (SEQ_LEQ(th->th_ack, tp->iss) || 1091 SEQ_GT(th->th_ack, tp->snd_max))) { 1092 /* 1093 * If we have a cached CCsent for the remote host, 1094 * hence we haven't just crashed and restarted, 1095 * do not send a RST. This may be a retransmission 1096 * from the other side after our earlier ACK was lost. 1097 * Our new SYN, when it arrives, will serve as the 1098 * needed ACK. 1099 */ 1100 if (taop->tao_ccsent != 0) 1101 goto drop; 1102 else { 1103 rstreason = BANDLIM_UNLIMITED; 1104 goto dropwithreset; 1105 } 1106 } 1107 if (thflags & TH_RST) { 1108 if (thflags & TH_ACK) 1109 tp = tcp_drop(tp, ECONNREFUSED); 1110 goto drop; 1111 } 1112 if ((thflags & TH_SYN) == 0) 1113 goto drop; 1114 tp->snd_wnd = th->th_win; /* initial send window */ 1115 tp->cc_recv = to.to_cc; /* foreign CC */ 1116 1117 tp->irs = th->th_seq; 1118 tcp_rcvseqinit(tp); 1119 if (thflags & TH_ACK) { 1120 /* 1121 * Our SYN was acked. If segment contains CC.ECHO 1122 * option, check it to make sure this segment really 1123 * matches our SYN. If not, just drop it as old 1124 * duplicate, but send an RST if we're still playing 1125 * by the old rules. If no CC.ECHO option, make sure 1126 * we don't get fooled into using T/TCP. 1127 */ 1128 if (to.to_flags & TOF_CCECHO) { 1129 if (tp->cc_send != to.to_ccecho) { 1130 if (taop->tao_ccsent != 0) 1131 goto drop; 1132 else { 1133 rstreason = BANDLIM_UNLIMITED; 1134 goto dropwithreset; 1135 } 1136 } 1137 } else 1138 tp->t_flags &= ~TF_RCVD_CC; 1139 tcpstat.tcps_connects++; 1140 soisconnected(so); 1141 /* Do window scaling on this connection? */ 1142 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1143 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1144 tp->snd_scale = tp->requested_s_scale; 1145 tp->rcv_scale = tp->request_r_scale; 1146 } 1147 /* Segment is acceptable, update cache if undefined. */ 1148 if (taop->tao_ccsent == 0) 1149 taop->tao_ccsent = to.to_ccecho; 1150 1151 tp->rcv_adv += tp->rcv_wnd; 1152 tp->snd_una++; /* SYN is acked */ 1153 /* 1154 * If there's data, delay ACK; if there's also a FIN 1155 * ACKNOW will be turned on later. 1156 */ 1157 if (DELAY_ACK(tp) && tlen != 0) 1158 callout_reset(tp->tt_delack, tcp_delacktime, 1159 tcp_timer_delack, tp); 1160 else 1161 tp->t_flags |= TF_ACKNOW; 1162 /* 1163 * Received <SYN,ACK> in SYN_SENT[*] state. 1164 * Transitions: 1165 * SYN_SENT --> ESTABLISHED 1166 * SYN_SENT* --> FIN_WAIT_1 1167 */ 1168 tp->t_starttime = ticks; 1169 if (tp->t_flags & TF_NEEDFIN) { 1170 tp->t_state = TCPS_FIN_WAIT_1; 1171 tp->t_flags &= ~TF_NEEDFIN; 1172 thflags &= ~TH_SYN; 1173 } else { 1174 tp->t_state = TCPS_ESTABLISHED; 1175 callout_reset(tp->tt_keep, tcp_keepidle, 1176 tcp_timer_keep, tp); 1177 } 1178 } else { 1179 /* 1180 * Received initial SYN in SYN-SENT[*] state => simul- 1181 * taneous open. If segment contains CC option and there is 1182 * a cached CC, apply TAO test; if it succeeds, connection is 1183 * half-synchronized. Otherwise, do 3-way handshake: 1184 * SYN-SENT -> SYN-RECEIVED 1185 * SYN-SENT* -> SYN-RECEIVED* 1186 * If there was no CC option, clear cached CC value. 1187 */ 1188 tp->t_flags |= TF_ACKNOW; 1189 callout_stop(tp->tt_rexmt); 1190 if (to.to_flags & TOF_CC) { 1191 if (taop->tao_cc != 0 && 1192 CC_GT(to.to_cc, taop->tao_cc)) { 1193 /* 1194 * update cache and make transition: 1195 * SYN-SENT -> ESTABLISHED* 1196 * SYN-SENT* -> FIN-WAIT-1* 1197 */ 1198 taop->tao_cc = to.to_cc; 1199 tp->t_starttime = ticks; 1200 if (tp->t_flags & TF_NEEDFIN) { 1201 tp->t_state = TCPS_FIN_WAIT_1; 1202 tp->t_flags &= ~TF_NEEDFIN; 1203 } else { 1204 tp->t_state = TCPS_ESTABLISHED; 1205 callout_reset(tp->tt_keep, 1206 tcp_keepidle, 1207 tcp_timer_keep, 1208 tp); 1209 } 1210 tp->t_flags |= TF_NEEDSYN; 1211 } else 1212 tp->t_state = TCPS_SYN_RECEIVED; 1213 } else { 1214 /* CC.NEW or no option => invalidate cache */ 1215 taop->tao_cc = 0; 1216 tp->t_state = TCPS_SYN_RECEIVED; 1217 } 1218 } 1219 1220 trimthenstep6: 1221 /* 1222 * Advance th->th_seq to correspond to first data byte. 1223 * If data, trim to stay within window, 1224 * dropping FIN if necessary. 1225 */ 1226 th->th_seq++; 1227 if (tlen > tp->rcv_wnd) { 1228 todrop = tlen - tp->rcv_wnd; 1229 m_adj(m, -todrop); 1230 tlen = tp->rcv_wnd; 1231 thflags &= ~TH_FIN; 1232 tcpstat.tcps_rcvpackafterwin++; 1233 tcpstat.tcps_rcvbyteafterwin += todrop; 1234 } 1235 tp->snd_wl1 = th->th_seq - 1; 1236 tp->rcv_up = th->th_seq; 1237 /* 1238 * Client side of transaction: already sent SYN and data. 1239 * If the remote host used T/TCP to validate the SYN, 1240 * our data will be ACK'd; if so, enter normal data segment 1241 * processing in the middle of step 5, ack processing. 1242 * Otherwise, goto step 6. 1243 */ 1244 if (thflags & TH_ACK) 1245 goto process_ACK; 1246 goto step6; 1247 /* 1248 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1249 * if segment contains a SYN and CC [not CC.NEW] option: 1250 * if state == TIME_WAIT and connection duration > MSL, 1251 * drop packet and send RST; 1252 * 1253 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1254 * ack the FIN (and data) in retransmission queue. 1255 * Complete close and delete TCPCB. Then reprocess 1256 * segment, hoping to find new TCPCB in LISTEN state; 1257 * 1258 * else must be old SYN; drop it. 1259 * else do normal processing. 1260 */ 1261 case TCPS_LAST_ACK: 1262 case TCPS_CLOSING: 1263 case TCPS_TIME_WAIT: 1264 if ((thflags & TH_SYN) && 1265 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1266 if (tp->t_state == TCPS_TIME_WAIT && 1267 (ticks - tp->t_starttime) > tcp_msl) { 1268 rstreason = BANDLIM_UNLIMITED; 1269 goto dropwithreset; 1270 } 1271 if (CC_GT(to.to_cc, tp->cc_recv)) { 1272 tp = tcp_close(tp); 1273 goto findpcb; 1274 } 1275 else 1276 goto drop; 1277 } 1278 break; /* continue normal processing */ 1279 } 1280 1281 /* 1282 * States other than LISTEN or SYN_SENT. 1283 * First check the RST flag and sequence number since reset segments 1284 * are exempt from the timestamp and connection count tests. This 1285 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1286 * below which allowed reset segments in half the sequence space 1287 * to fall though and be processed (which gives forged reset 1288 * segments with a random sequence number a 50 percent chance of 1289 * killing a connection). 1290 * Then check timestamp, if present. 1291 * Then check the connection count, if present. 1292 * Then check that at least some bytes of segment are within 1293 * receive window. If segment begins before rcv_nxt, 1294 * drop leading data (and SYN); if nothing left, just ack. 1295 * 1296 * 1297 * If the RST bit is set, check the sequence number to see 1298 * if this is a valid reset segment. 1299 * RFC 793 page 37: 1300 * In all states except SYN-SENT, all reset (RST) segments 1301 * are validated by checking their SEQ-fields. A reset is 1302 * valid if its sequence number is in the window. 1303 * Note: this does not take into account delayed ACKs, so 1304 * we should test against last_ack_sent instead of rcv_nxt. 1305 * The sequence number in the reset segment is normally an 1306 * echo of our outgoing acknowlegement numbers, but some hosts 1307 * send a reset with the sequence number at the rightmost edge 1308 * of our receive window, and we have to handle this case. 1309 * If we have multiple segments in flight, the intial reset 1310 * segment sequence numbers will be to the left of last_ack_sent, 1311 * but they will eventually catch up. 1312 * In any case, it never made sense to trim reset segments to 1313 * fit the receive window since RFC 1122 says: 1314 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1315 * 1316 * A TCP SHOULD allow a received RST segment to include data. 1317 * 1318 * DISCUSSION 1319 * It has been suggested that a RST segment could contain 1320 * ASCII text that encoded and explained the cause of the 1321 * RST. No standard has yet been established for such 1322 * data. 1323 * 1324 * If the reset segment passes the sequence number test examine 1325 * the state: 1326 * SYN_RECEIVED STATE: 1327 * If passive open, return to LISTEN state. 1328 * If active open, inform user that connection was refused. 1329 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1330 * Inform user that connection was reset, and close tcb. 1331 * CLOSING, LAST_ACK STATES: 1332 * Close the tcb. 1333 * TIME_WAIT STATE: 1334 * Drop the segment - see Stevens, vol. 2, p. 964 and 1335 * RFC 1337. 1336 */ 1337 if (thflags & TH_RST) { 1338 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1339 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1340 switch (tp->t_state) { 1341 1342 case TCPS_SYN_RECEIVED: 1343 so->so_error = ECONNREFUSED; 1344 goto close; 1345 1346 case TCPS_ESTABLISHED: 1347 case TCPS_FIN_WAIT_1: 1348 case TCPS_FIN_WAIT_2: 1349 case TCPS_CLOSE_WAIT: 1350 so->so_error = ECONNRESET; 1351 close: 1352 tp->t_state = TCPS_CLOSED; 1353 tcpstat.tcps_drops++; 1354 tp = tcp_close(tp); 1355 break; 1356 1357 case TCPS_CLOSING: 1358 case TCPS_LAST_ACK: 1359 tp = tcp_close(tp); 1360 break; 1361 1362 case TCPS_TIME_WAIT: 1363 break; 1364 } 1365 } 1366 goto drop; 1367 } 1368 1369 /* 1370 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1371 * and it's less than ts_recent, drop it. 1372 */ 1373 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1374 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1375 1376 /* Check to see if ts_recent is over 24 days old. */ 1377 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1378 /* 1379 * Invalidate ts_recent. If this segment updates 1380 * ts_recent, the age will be reset later and ts_recent 1381 * will get a valid value. If it does not, setting 1382 * ts_recent to zero will at least satisfy the 1383 * requirement that zero be placed in the timestamp 1384 * echo reply when ts_recent isn't valid. The 1385 * age isn't reset until we get a valid ts_recent 1386 * because we don't want out-of-order segments to be 1387 * dropped when ts_recent is old. 1388 */ 1389 tp->ts_recent = 0; 1390 } else { 1391 tcpstat.tcps_rcvduppack++; 1392 tcpstat.tcps_rcvdupbyte += tlen; 1393 tcpstat.tcps_pawsdrop++; 1394 goto dropafterack; 1395 } 1396 } 1397 1398 /* 1399 * T/TCP mechanism 1400 * If T/TCP was negotiated and the segment doesn't have CC, 1401 * or if its CC is wrong then drop the segment. 1402 * RST segments do not have to comply with this. 1403 */ 1404 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1405 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1406 goto dropafterack; 1407 1408 /* 1409 * In the SYN-RECEIVED state, validate that the packet belongs to 1410 * this connection before trimming the data to fit the receive 1411 * window. Check the sequence number versus IRS since we know 1412 * the sequence numbers haven't wrapped. This is a partial fix 1413 * for the "LAND" DoS attack. 1414 */ 1415 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1416 rstreason = BANDLIM_RST_OPENPORT; 1417 goto dropwithreset; 1418 } 1419 1420 todrop = tp->rcv_nxt - th->th_seq; 1421 if (todrop > 0) { 1422 if (thflags & TH_SYN) { 1423 thflags &= ~TH_SYN; 1424 th->th_seq++; 1425 if (th->th_urp > 1) 1426 th->th_urp--; 1427 else 1428 thflags &= ~TH_URG; 1429 todrop--; 1430 } 1431 /* 1432 * Following if statement from Stevens, vol. 2, p. 960. 1433 */ 1434 if (todrop > tlen 1435 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1436 /* 1437 * Any valid FIN must be to the left of the window. 1438 * At this point the FIN must be a duplicate or out 1439 * of sequence; drop it. 1440 */ 1441 thflags &= ~TH_FIN; 1442 1443 /* 1444 * Send an ACK to resynchronize and drop any data. 1445 * But keep on processing for RST or ACK. 1446 */ 1447 tp->t_flags |= TF_ACKNOW; 1448 todrop = tlen; 1449 tcpstat.tcps_rcvduppack++; 1450 tcpstat.tcps_rcvdupbyte += todrop; 1451 } else { 1452 tcpstat.tcps_rcvpartduppack++; 1453 tcpstat.tcps_rcvpartdupbyte += todrop; 1454 } 1455 drop_hdrlen += todrop; /* drop from the top afterwards */ 1456 th->th_seq += todrop; 1457 tlen -= todrop; 1458 if (th->th_urp > todrop) 1459 th->th_urp -= todrop; 1460 else { 1461 thflags &= ~TH_URG; 1462 th->th_urp = 0; 1463 } 1464 } 1465 1466 /* 1467 * If new data are received on a connection after the 1468 * user processes are gone, then RST the other end. 1469 */ 1470 if ((so->so_state & SS_NOFDREF) && 1471 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1472 tp = tcp_close(tp); 1473 tcpstat.tcps_rcvafterclose++; 1474 rstreason = BANDLIM_UNLIMITED; 1475 goto dropwithreset; 1476 } 1477 1478 /* 1479 * If segment ends after window, drop trailing data 1480 * (and PUSH and FIN); if nothing left, just ACK. 1481 */ 1482 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1483 if (todrop > 0) { 1484 tcpstat.tcps_rcvpackafterwin++; 1485 if (todrop >= tlen) { 1486 tcpstat.tcps_rcvbyteafterwin += tlen; 1487 /* 1488 * If a new connection request is received 1489 * while in TIME_WAIT, drop the old connection 1490 * and start over if the sequence numbers 1491 * are above the previous ones. 1492 */ 1493 if (thflags & TH_SYN && 1494 tp->t_state == TCPS_TIME_WAIT && 1495 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1496 tp = tcp_close(tp); 1497 goto findpcb; 1498 } 1499 /* 1500 * If window is closed can only take segments at 1501 * window edge, and have to drop data and PUSH from 1502 * incoming segments. Continue processing, but 1503 * remember to ack. Otherwise, drop segment 1504 * and ack. 1505 */ 1506 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1507 tp->t_flags |= TF_ACKNOW; 1508 tcpstat.tcps_rcvwinprobe++; 1509 } else 1510 goto dropafterack; 1511 } else 1512 tcpstat.tcps_rcvbyteafterwin += todrop; 1513 m_adj(m, -todrop); 1514 tlen -= todrop; 1515 thflags &= ~(TH_PUSH|TH_FIN); 1516 } 1517 1518 /* 1519 * If last ACK falls within this segment's sequence numbers, 1520 * record its timestamp. 1521 * NOTE that the test is modified according to the latest 1522 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1523 */ 1524 if ((to.to_flags & TOF_TS) != 0 && 1525 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1526 tp->ts_recent_age = ticks; 1527 tp->ts_recent = to.to_tsval; 1528 } 1529 1530 /* 1531 * If a SYN is in the window, then this is an 1532 * error and we send an RST and drop the connection. 1533 */ 1534 if (thflags & TH_SYN) { 1535 tp = tcp_drop(tp, ECONNRESET); 1536 rstreason = BANDLIM_UNLIMITED; 1537 goto dropwithreset; 1538 } 1539 1540 /* 1541 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1542 * flag is on (half-synchronized state), then queue data for 1543 * later processing; else drop segment and return. 1544 */ 1545 if ((thflags & TH_ACK) == 0) { 1546 if (tp->t_state == TCPS_SYN_RECEIVED || 1547 (tp->t_flags & TF_NEEDSYN)) 1548 goto step6; 1549 else 1550 goto drop; 1551 } 1552 1553 /* 1554 * Ack processing. 1555 */ 1556 switch (tp->t_state) { 1557 1558 /* 1559 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1560 * ESTABLISHED state and continue processing. 1561 * The ACK was checked above. 1562 */ 1563 case TCPS_SYN_RECEIVED: 1564 1565 tcpstat.tcps_connects++; 1566 soisconnected(so); 1567 /* Do window scaling? */ 1568 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1569 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1570 tp->snd_scale = tp->requested_s_scale; 1571 tp->rcv_scale = tp->request_r_scale; 1572 } 1573 /* 1574 * Upon successful completion of 3-way handshake, 1575 * update cache.CC if it was undefined, pass any queued 1576 * data to the user, and advance state appropriately. 1577 */ 1578 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1579 taop->tao_cc == 0) 1580 taop->tao_cc = tp->cc_recv; 1581 1582 /* 1583 * Make transitions: 1584 * SYN-RECEIVED -> ESTABLISHED 1585 * SYN-RECEIVED* -> FIN-WAIT-1 1586 */ 1587 tp->t_starttime = ticks; 1588 if (tp->t_flags & TF_NEEDFIN) { 1589 tp->t_state = TCPS_FIN_WAIT_1; 1590 tp->t_flags &= ~TF_NEEDFIN; 1591 } else { 1592 tp->t_state = TCPS_ESTABLISHED; 1593 callout_reset(tp->tt_keep, tcp_keepidle, 1594 tcp_timer_keep, tp); 1595 } 1596 /* 1597 * If segment contains data or ACK, will call tcp_reass() 1598 * later; if not, do so now to pass queued data to user. 1599 */ 1600 if (tlen == 0 && (thflags & TH_FIN) == 0) 1601 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1602 (struct mbuf *)0); 1603 tp->snd_wl1 = th->th_seq - 1; 1604 /* fall into ... */ 1605 1606 /* 1607 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1608 * ACKs. If the ack is in the range 1609 * tp->snd_una < th->th_ack <= tp->snd_max 1610 * then advance tp->snd_una to th->th_ack and drop 1611 * data from the retransmission queue. If this ACK reflects 1612 * more up to date window information we update our window information. 1613 */ 1614 case TCPS_ESTABLISHED: 1615 case TCPS_FIN_WAIT_1: 1616 case TCPS_FIN_WAIT_2: 1617 case TCPS_CLOSE_WAIT: 1618 case TCPS_CLOSING: 1619 case TCPS_LAST_ACK: 1620 case TCPS_TIME_WAIT: 1621 1622 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1623 if (tlen == 0 && tiwin == tp->snd_wnd) { 1624 tcpstat.tcps_rcvdupack++; 1625 /* 1626 * If we have outstanding data (other than 1627 * a window probe), this is a completely 1628 * duplicate ack (ie, window info didn't 1629 * change), the ack is the biggest we've 1630 * seen and we've seen exactly our rexmt 1631 * threshhold of them, assume a packet 1632 * has been dropped and retransmit it. 1633 * Kludge snd_nxt & the congestion 1634 * window so we send only this one 1635 * packet. 1636 * 1637 * We know we're losing at the current 1638 * window size so do congestion avoidance 1639 * (set ssthresh to half the current window 1640 * and pull our congestion window back to 1641 * the new ssthresh). 1642 * 1643 * Dup acks mean that packets have left the 1644 * network (they're now cached at the receiver) 1645 * so bump cwnd by the amount in the receiver 1646 * to keep a constant cwnd packets in the 1647 * network. 1648 */ 1649 if (!callout_active(tp->tt_rexmt) || 1650 th->th_ack != tp->snd_una) 1651 tp->t_dupacks = 0; 1652 else if (++tp->t_dupacks == tcprexmtthresh) { 1653 tcp_seq onxt = tp->snd_nxt; 1654 u_int win = 1655 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1656 tp->t_maxseg; 1657 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1658 tp->snd_recover)) { 1659 /* False retransmit, should not 1660 * cut window 1661 */ 1662 tp->snd_cwnd += tp->t_maxseg; 1663 tp->t_dupacks = 0; 1664 (void) tcp_output(tp); 1665 goto drop; 1666 } 1667 if (win < 2) 1668 win = 2; 1669 tp->snd_ssthresh = win * tp->t_maxseg; 1670 tp->snd_recover = tp->snd_max; 1671 callout_stop(tp->tt_rexmt); 1672 tp->t_rtttime = 0; 1673 tp->snd_nxt = th->th_ack; 1674 tp->snd_cwnd = tp->t_maxseg; 1675 (void) tcp_output(tp); 1676 tp->snd_cwnd = tp->snd_ssthresh + 1677 tp->t_maxseg * tp->t_dupacks; 1678 if (SEQ_GT(onxt, tp->snd_nxt)) 1679 tp->snd_nxt = onxt; 1680 goto drop; 1681 } else if (tp->t_dupacks > tcprexmtthresh) { 1682 tp->snd_cwnd += tp->t_maxseg; 1683 (void) tcp_output(tp); 1684 goto drop; 1685 } 1686 } else 1687 tp->t_dupacks = 0; 1688 break; 1689 } 1690 /* 1691 * If the congestion window was inflated to account 1692 * for the other side's cached packets, retract it. 1693 */ 1694 if (tcp_do_newreno == 0) { 1695 if (tp->t_dupacks >= tcprexmtthresh && 1696 tp->snd_cwnd > tp->snd_ssthresh) 1697 tp->snd_cwnd = tp->snd_ssthresh; 1698 tp->t_dupacks = 0; 1699 } else if (tp->t_dupacks >= tcprexmtthresh && 1700 !tcp_newreno(tp, th)) { 1701 /* 1702 * Window inflation should have left us with approx. 1703 * snd_ssthresh outstanding data. But in case we 1704 * would be inclined to send a burst, better to do 1705 * it via the slow start mechanism. 1706 */ 1707 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1708 tp->snd_cwnd = 1709 tp->snd_max - th->th_ack + tp->t_maxseg; 1710 else 1711 tp->snd_cwnd = tp->snd_ssthresh; 1712 tp->t_dupacks = 0; 1713 } 1714 if (tp->t_dupacks < tcprexmtthresh) 1715 tp->t_dupacks = 0; 1716 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1717 tcpstat.tcps_rcvacktoomuch++; 1718 goto dropafterack; 1719 } 1720 /* 1721 * If we reach this point, ACK is not a duplicate, 1722 * i.e., it ACKs something we sent. 1723 */ 1724 if (tp->t_flags & TF_NEEDSYN) { 1725 /* 1726 * T/TCP: Connection was half-synchronized, and our 1727 * SYN has been ACK'd (so connection is now fully 1728 * synchronized). Go to non-starred state, 1729 * increment snd_una for ACK of SYN, and check if 1730 * we can do window scaling. 1731 */ 1732 tp->t_flags &= ~TF_NEEDSYN; 1733 tp->snd_una++; 1734 /* Do window scaling? */ 1735 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1736 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1737 tp->snd_scale = tp->requested_s_scale; 1738 tp->rcv_scale = tp->request_r_scale; 1739 } 1740 } 1741 1742 process_ACK: 1743 acked = th->th_ack - tp->snd_una; 1744 tcpstat.tcps_rcvackpack++; 1745 tcpstat.tcps_rcvackbyte += acked; 1746 1747 /* 1748 * If we just performed our first retransmit, and the ACK 1749 * arrives within our recovery window, then it was a mistake 1750 * to do the retransmit in the first place. Recover our 1751 * original cwnd and ssthresh, and proceed to transmit where 1752 * we left off. 1753 */ 1754 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1755 tp->snd_cwnd = tp->snd_cwnd_prev; 1756 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1757 tp->snd_nxt = tp->snd_max; 1758 tp->t_badrxtwin = 0; /* XXX probably not required */ 1759 } 1760 1761 /* 1762 * If we have a timestamp reply, update smoothed 1763 * round trip time. If no timestamp is present but 1764 * transmit timer is running and timed sequence 1765 * number was acked, update smoothed round trip time. 1766 * Since we now have an rtt measurement, cancel the 1767 * timer backoff (cf., Phil Karn's retransmit alg.). 1768 * Recompute the initial retransmit timer. 1769 */ 1770 if (to.to_flags & TOF_TS) 1771 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1772 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1773 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1774 1775 /* 1776 * If all outstanding data is acked, stop retransmit 1777 * timer and remember to restart (more output or persist). 1778 * If there is more data to be acked, restart retransmit 1779 * timer, using current (possibly backed-off) value. 1780 */ 1781 if (th->th_ack == tp->snd_max) { 1782 callout_stop(tp->tt_rexmt); 1783 needoutput = 1; 1784 } else if (!callout_active(tp->tt_persist)) 1785 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1786 tcp_timer_rexmt, tp); 1787 1788 /* 1789 * If no data (only SYN) was ACK'd, 1790 * skip rest of ACK processing. 1791 */ 1792 if (acked == 0) 1793 goto step6; 1794 1795 /* 1796 * When new data is acked, open the congestion window. 1797 * If the window gives us less than ssthresh packets 1798 * in flight, open exponentially (maxseg per packet). 1799 * Otherwise open linearly: maxseg per window 1800 * (maxseg^2 / cwnd per packet). 1801 */ 1802 { 1803 register u_int cw = tp->snd_cwnd; 1804 register u_int incr = tp->t_maxseg; 1805 1806 if (cw > tp->snd_ssthresh) 1807 incr = incr * incr / cw; 1808 /* 1809 * If t_dupacks != 0 here, it indicates that we are still 1810 * in NewReno fast recovery mode, so we leave the congestion 1811 * window alone. 1812 */ 1813 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1814 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1815 } 1816 if (acked > so->so_snd.sb_cc) { 1817 tp->snd_wnd -= so->so_snd.sb_cc; 1818 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1819 ourfinisacked = 1; 1820 } else { 1821 sbdrop(&so->so_snd, acked); 1822 tp->snd_wnd -= acked; 1823 ourfinisacked = 0; 1824 } 1825 sowwakeup(so); 1826 tp->snd_una = th->th_ack; 1827 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1828 tp->snd_nxt = tp->snd_una; 1829 1830 switch (tp->t_state) { 1831 1832 /* 1833 * In FIN_WAIT_1 STATE in addition to the processing 1834 * for the ESTABLISHED state if our FIN is now acknowledged 1835 * then enter FIN_WAIT_2. 1836 */ 1837 case TCPS_FIN_WAIT_1: 1838 if (ourfinisacked) { 1839 /* 1840 * If we can't receive any more 1841 * data, then closing user can proceed. 1842 * Starting the timer is contrary to the 1843 * specification, but if we don't get a FIN 1844 * we'll hang forever. 1845 */ 1846 if (so->so_state & SS_CANTRCVMORE) { 1847 soisdisconnected_locked(so); 1848 callout_reset(tp->tt_2msl, tcp_maxidle, 1849 tcp_timer_2msl, tp); 1850 } 1851 tp->t_state = TCPS_FIN_WAIT_2; 1852 } 1853 break; 1854 1855 /* 1856 * In CLOSING STATE in addition to the processing for 1857 * the ESTABLISHED state if the ACK acknowledges our FIN 1858 * then enter the TIME-WAIT state, otherwise ignore 1859 * the segment. 1860 */ 1861 case TCPS_CLOSING: 1862 if (ourfinisacked) { 1863 tp->t_state = TCPS_TIME_WAIT; 1864 tcp_canceltimers(tp); 1865 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1866 if (tp->cc_recv != 0 && 1867 (ticks - tp->t_starttime) < tcp_msl) 1868 callout_reset(tp->tt_2msl, 1869 tp->t_rxtcur * 1870 TCPTV_TWTRUNC, 1871 tcp_timer_2msl, tp); 1872 else 1873 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1874 tcp_timer_2msl, tp); 1875 soisdisconnected(so); 1876 } 1877 break; 1878 1879 /* 1880 * In LAST_ACK, we may still be waiting for data to drain 1881 * and/or to be acked, as well as for the ack of our FIN. 1882 * If our FIN is now acknowledged, delete the TCB, 1883 * enter the closed state and return. 1884 */ 1885 case TCPS_LAST_ACK: 1886 if (ourfinisacked) { 1887 tp = tcp_close(tp); 1888 goto drop; 1889 } 1890 break; 1891 1892 /* 1893 * In TIME_WAIT state the only thing that should arrive 1894 * is a retransmission of the remote FIN. Acknowledge 1895 * it and restart the finack timer. 1896 */ 1897 case TCPS_TIME_WAIT: 1898 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1899 tcp_timer_2msl, tp); 1900 goto dropafterack; 1901 } 1902 } 1903 1904 step6: 1905 /* 1906 * Update window information. 1907 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1908 */ 1909 if ((thflags & TH_ACK) && 1910 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1911 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1912 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1913 /* keep track of pure window updates */ 1914 if (tlen == 0 && 1915 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1916 tcpstat.tcps_rcvwinupd++; 1917 tp->snd_wnd = tiwin; 1918 tp->snd_wl1 = th->th_seq; 1919 tp->snd_wl2 = th->th_ack; 1920 if (tp->snd_wnd > tp->max_sndwnd) 1921 tp->max_sndwnd = tp->snd_wnd; 1922 needoutput = 1; 1923 } 1924 1925 /* 1926 * Process segments with URG. 1927 */ 1928 if ((thflags & TH_URG) && th->th_urp && 1929 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1930 /* 1931 * This is a kludge, but if we receive and accept 1932 * random urgent pointers, we'll crash in 1933 * soreceive. It's hard to imagine someone 1934 * actually wanting to send this much urgent data. 1935 */ 1936 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1937 th->th_urp = 0; /* XXX */ 1938 thflags &= ~TH_URG; /* XXX */ 1939 goto dodata; /* XXX */ 1940 } 1941 /* 1942 * If this segment advances the known urgent pointer, 1943 * then mark the data stream. This should not happen 1944 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1945 * a FIN has been received from the remote side. 1946 * In these states we ignore the URG. 1947 * 1948 * According to RFC961 (Assigned Protocols), 1949 * the urgent pointer points to the last octet 1950 * of urgent data. We continue, however, 1951 * to consider it to indicate the first octet 1952 * of data past the urgent section as the original 1953 * spec states (in one of two places). 1954 */ 1955 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1956 tp->rcv_up = th->th_seq + th->th_urp; 1957 so->so_oobmark = so->so_rcv.sb_cc + 1958 (tp->rcv_up - tp->rcv_nxt) - 1; 1959 if (so->so_oobmark == 0) 1960 so->so_state |= SS_RCVATMARK; 1961 sohasoutofband(so); 1962 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1963 } 1964 /* 1965 * Remove out of band data so doesn't get presented to user. 1966 * This can happen independent of advancing the URG pointer, 1967 * but if two URG's are pending at once, some out-of-band 1968 * data may creep in... ick. 1969 */ 1970 if (th->th_urp <= (u_long)tlen 1971 #ifdef SO_OOBINLINE 1972 && (so->so_options & SO_OOBINLINE) == 0 1973 #endif 1974 ) 1975 tcp_pulloutofband(so, th, m, 1976 drop_hdrlen); /* hdr drop is delayed */ 1977 } else 1978 /* 1979 * If no out of band data is expected, 1980 * pull receive urgent pointer along 1981 * with the receive window. 1982 */ 1983 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1984 tp->rcv_up = tp->rcv_nxt; 1985 dodata: /* XXX */ 1986 1987 /* 1988 * Process the segment text, merging it into the TCP sequencing queue, 1989 * and arranging for acknowledgment of receipt if necessary. 1990 * This process logically involves adjusting tp->rcv_wnd as data 1991 * is presented to the user (this happens in tcp_usrreq.c, 1992 * case PRU_RCVD). If a FIN has already been received on this 1993 * connection then we just ignore the text. 1994 */ 1995 if ((tlen || (thflags&TH_FIN)) && 1996 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1997 m_adj(m, drop_hdrlen); /* delayed header drop */ 1998 /* 1999 * Insert segment which inludes th into reassembly queue of tcp with 2000 * control block tp. Return TH_FIN if reassembly now includes 2001 * a segment with FIN. This handle the common case inline (segment 2002 * is the next to be received on an established connection, and the 2003 * queue is empty), avoiding linkage into and removal from the queue 2004 * and repetition of various conversions. 2005 * Set DELACK for segments received in order, but ack immediately 2006 * when segments are out of order (so fast retransmit can work). 2007 */ 2008 if (th->th_seq == tp->rcv_nxt && 2009 LIST_EMPTY(&tp->t_segq) && 2010 TCPS_HAVEESTABLISHED(tp->t_state)) { 2011 if (DELAY_ACK(tp)) 2012 callout_reset(tp->tt_delack, tcp_delacktime, 2013 tcp_timer_delack, tp); 2014 else 2015 tp->t_flags |= TF_ACKNOW; 2016 tp->rcv_nxt += tlen; 2017 thflags = th->th_flags & TH_FIN; 2018 tcpstat.tcps_rcvpack++; 2019 tcpstat.tcps_rcvbyte += tlen; 2020 ND6_HINT(tp); 2021 sbappend(&so->so_rcv, m); 2022 sorwakeup(so); 2023 } else { 2024 thflags = tcp_reass(tp, th, &tlen, m); 2025 tp->t_flags |= TF_ACKNOW; 2026 } 2027 2028 /* 2029 * Note the amount of data that peer has sent into 2030 * our window, in order to estimate the sender's 2031 * buffer size. 2032 */ 2033 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2034 } else { 2035 m_freem(m); 2036 thflags &= ~TH_FIN; 2037 } 2038 2039 /* 2040 * If FIN is received ACK the FIN and let the user know 2041 * that the connection is closing. 2042 */ 2043 if (thflags & TH_FIN) { 2044 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2045 socantrcvmore(so); 2046 /* 2047 * If connection is half-synchronized 2048 * (ie NEEDSYN flag on) then delay ACK, 2049 * so it may be piggybacked when SYN is sent. 2050 * Otherwise, since we received a FIN then no 2051 * more input can be expected, send ACK now. 2052 */ 2053 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2054 callout_reset(tp->tt_delack, tcp_delacktime, 2055 tcp_timer_delack, tp); 2056 else 2057 tp->t_flags |= TF_ACKNOW; 2058 tp->rcv_nxt++; 2059 } 2060 switch (tp->t_state) { 2061 2062 /* 2063 * In SYN_RECEIVED and ESTABLISHED STATES 2064 * enter the CLOSE_WAIT state. 2065 */ 2066 case TCPS_SYN_RECEIVED: 2067 tp->t_starttime = ticks; 2068 /*FALLTHROUGH*/ 2069 case TCPS_ESTABLISHED: 2070 tp->t_state = TCPS_CLOSE_WAIT; 2071 break; 2072 2073 /* 2074 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2075 * enter the CLOSING state. 2076 */ 2077 case TCPS_FIN_WAIT_1: 2078 tp->t_state = TCPS_CLOSING; 2079 break; 2080 2081 /* 2082 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2083 * starting the time-wait timer, turning off the other 2084 * standard timers. 2085 */ 2086 case TCPS_FIN_WAIT_2: 2087 tp->t_state = TCPS_TIME_WAIT; 2088 tcp_canceltimers(tp); 2089 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2090 if (tp->cc_recv != 0 && 2091 (ticks - tp->t_starttime) < tcp_msl) { 2092 callout_reset(tp->tt_2msl, 2093 tp->t_rxtcur * TCPTV_TWTRUNC, 2094 tcp_timer_2msl, tp); 2095 /* For transaction client, force ACK now. */ 2096 tp->t_flags |= TF_ACKNOW; 2097 } 2098 else 2099 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2100 tcp_timer_2msl, tp); 2101 soisdisconnected(so); 2102 break; 2103 2104 /* 2105 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2106 */ 2107 case TCPS_TIME_WAIT: 2108 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2109 tcp_timer_2msl, tp); 2110 break; 2111 } 2112 } 2113 #ifdef TCPDEBUG 2114 if (so->so_options & SO_DEBUG) 2115 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2116 &tcp_savetcp, 0); 2117 #endif 2118 2119 /* 2120 * Return any desired output. 2121 */ 2122 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2123 (void) tcp_output(tp); 2124 return; 2125 2126 dropafterack: 2127 /* 2128 * Generate an ACK dropping incoming segment if it occupies 2129 * sequence space, where the ACK reflects our state. 2130 * 2131 * We can now skip the test for the RST flag since all 2132 * paths to this code happen after packets containing 2133 * RST have been dropped. 2134 * 2135 * In the SYN-RECEIVED state, don't send an ACK unless the 2136 * segment we received passes the SYN-RECEIVED ACK test. 2137 * If it fails send a RST. This breaks the loop in the 2138 * "LAND" DoS attack, and also prevents an ACK storm 2139 * between two listening ports that have been sent forged 2140 * SYN segments, each with the source address of the other. 2141 */ 2142 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2143 (SEQ_GT(tp->snd_una, th->th_ack) || 2144 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2145 rstreason = BANDLIM_RST_OPENPORT; 2146 goto dropwithreset; 2147 } 2148 #ifdef TCPDEBUG 2149 if (so->so_options & SO_DEBUG) 2150 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2151 &tcp_savetcp, 0); 2152 #endif 2153 m_freem(m); 2154 tp->t_flags |= TF_ACKNOW; 2155 (void) tcp_output(tp); 2156 return; 2157 2158 dropwithreset: 2159 /* 2160 * Generate a RST, dropping incoming segment. 2161 * Make ACK acceptable to originator of segment. 2162 * Don't bother to respond if destination was broadcast/multicast. 2163 */ 2164 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2165 goto drop; 2166 #ifdef INET6 2167 if (isipv6) { 2168 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2169 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2170 goto drop; 2171 } else 2172 #endif /* INET6 */ 2173 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2174 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2175 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2176 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2177 goto drop; 2178 /* IPv6 anycast check is done at tcp6_input() */ 2179 2180 /* 2181 * Perform bandwidth limiting. 2182 */ 2183 if (badport_bandlim(rstreason) < 0) 2184 goto drop; 2185 2186 #ifdef TCPDEBUG 2187 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2188 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2189 &tcp_savetcp, 0); 2190 #endif 2191 if (thflags & TH_ACK) 2192 /* mtod() below is safe as long as hdr dropping is delayed */ 2193 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2194 TH_RST); 2195 else { 2196 if (thflags & TH_SYN) 2197 tlen++; 2198 /* mtod() below is safe as long as hdr dropping is delayed */ 2199 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2200 (tcp_seq)0, TH_RST|TH_ACK); 2201 } 2202 return; 2203 2204 drop: 2205 /* 2206 * Drop space held by incoming segment and return. 2207 */ 2208 #ifdef TCPDEBUG 2209 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2210 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2211 &tcp_savetcp, 0); 2212 #endif 2213 m_freem(m); 2214 return; 2215 } 2216 2217 /* 2218 * Parse TCP options and place in tcpopt. 2219 */ 2220 static void 2221 tcp_dooptions(to, cp, cnt, is_syn) 2222 struct tcpopt *to; 2223 u_char *cp; 2224 int cnt; 2225 { 2226 int opt, optlen; 2227 2228 to->to_flags = 0; 2229 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2230 opt = cp[0]; 2231 if (opt == TCPOPT_EOL) 2232 break; 2233 if (opt == TCPOPT_NOP) 2234 optlen = 1; 2235 else { 2236 if (cnt < 2) 2237 break; 2238 optlen = cp[1]; 2239 if (optlen < 2 || optlen > cnt) 2240 break; 2241 } 2242 switch (opt) { 2243 case TCPOPT_MAXSEG: 2244 if (optlen != TCPOLEN_MAXSEG) 2245 continue; 2246 if (!is_syn) 2247 continue; 2248 to->to_flags |= TOF_MSS; 2249 bcopy((char *)cp + 2, 2250 (char *)&to->to_mss, sizeof(to->to_mss)); 2251 to->to_mss = ntohs(to->to_mss); 2252 break; 2253 case TCPOPT_WINDOW: 2254 if (optlen != TCPOLEN_WINDOW) 2255 continue; 2256 if (! is_syn) 2257 continue; 2258 to->to_flags |= TOF_SCALE; 2259 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2260 break; 2261 case TCPOPT_TIMESTAMP: 2262 if (optlen != TCPOLEN_TIMESTAMP) 2263 continue; 2264 to->to_flags |= TOF_TS; 2265 bcopy((char *)cp + 2, 2266 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2267 to->to_tsval = ntohl(to->to_tsval); 2268 bcopy((char *)cp + 6, 2269 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2270 to->to_tsecr = ntohl(to->to_tsecr); 2271 break; 2272 case TCPOPT_CC: 2273 if (optlen != TCPOLEN_CC) 2274 continue; 2275 to->to_flags |= TOF_CC; 2276 bcopy((char *)cp + 2, 2277 (char *)&to->to_cc, sizeof(to->to_cc)); 2278 to->to_cc = ntohl(to->to_cc); 2279 break; 2280 case TCPOPT_CCNEW: 2281 if (optlen != TCPOLEN_CC) 2282 continue; 2283 if (!is_syn) 2284 continue; 2285 to->to_flags |= TOF_CCNEW; 2286 bcopy((char *)cp + 2, 2287 (char *)&to->to_cc, sizeof(to->to_cc)); 2288 to->to_cc = ntohl(to->to_cc); 2289 break; 2290 case TCPOPT_CCECHO: 2291 if (optlen != TCPOLEN_CC) 2292 continue; 2293 if (!is_syn) 2294 continue; 2295 to->to_flags |= TOF_CCECHO; 2296 bcopy((char *)cp + 2, 2297 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2298 to->to_ccecho = ntohl(to->to_ccecho); 2299 break; 2300 default: 2301 continue; 2302 } 2303 } 2304 } 2305 2306 /* 2307 * Pull out of band byte out of a segment so 2308 * it doesn't appear in the user's data queue. 2309 * It is still reflected in the segment length for 2310 * sequencing purposes. 2311 */ 2312 static void 2313 tcp_pulloutofband(so, th, m, off) 2314 struct socket *so; 2315 struct tcphdr *th; 2316 register struct mbuf *m; 2317 int off; /* delayed to be droped hdrlen */ 2318 { 2319 int cnt = off + th->th_urp - 1; 2320 2321 while (cnt >= 0) { 2322 if (m->m_len > cnt) { 2323 char *cp = mtod(m, caddr_t) + cnt; 2324 struct tcpcb *tp = sototcpcb(so); 2325 2326 tp->t_iobc = *cp; 2327 tp->t_oobflags |= TCPOOB_HAVEDATA; 2328 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2329 m->m_len--; 2330 if (m->m_flags & M_PKTHDR) 2331 m->m_pkthdr.len--; 2332 return; 2333 } 2334 cnt -= m->m_len; 2335 m = m->m_next; 2336 if (m == 0) 2337 break; 2338 } 2339 panic("tcp_pulloutofband"); 2340 } 2341 2342 /* 2343 * Collect new round-trip time estimate 2344 * and update averages and current timeout. 2345 */ 2346 static void 2347 tcp_xmit_timer(tp, rtt) 2348 register struct tcpcb *tp; 2349 int rtt; 2350 { 2351 register int delta; 2352 2353 tcpstat.tcps_rttupdated++; 2354 tp->t_rttupdated++; 2355 if (tp->t_srtt != 0) { 2356 /* 2357 * srtt is stored as fixed point with 5 bits after the 2358 * binary point (i.e., scaled by 8). The following magic 2359 * is equivalent to the smoothing algorithm in rfc793 with 2360 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2361 * point). Adjust rtt to origin 0. 2362 */ 2363 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2364 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2365 2366 if ((tp->t_srtt += delta) <= 0) 2367 tp->t_srtt = 1; 2368 2369 /* 2370 * We accumulate a smoothed rtt variance (actually, a 2371 * smoothed mean difference), then set the retransmit 2372 * timer to smoothed rtt + 4 times the smoothed variance. 2373 * rttvar is stored as fixed point with 4 bits after the 2374 * binary point (scaled by 16). The following is 2375 * equivalent to rfc793 smoothing with an alpha of .75 2376 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2377 * rfc793's wired-in beta. 2378 */ 2379 if (delta < 0) 2380 delta = -delta; 2381 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2382 if ((tp->t_rttvar += delta) <= 0) 2383 tp->t_rttvar = 1; 2384 } else { 2385 /* 2386 * No rtt measurement yet - use the unsmoothed rtt. 2387 * Set the variance to half the rtt (so our first 2388 * retransmit happens at 3*rtt). 2389 */ 2390 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2391 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2392 } 2393 tp->t_rtttime = 0; 2394 tp->t_rxtshift = 0; 2395 2396 /* 2397 * the retransmit should happen at rtt + 4 * rttvar. 2398 * Because of the way we do the smoothing, srtt and rttvar 2399 * will each average +1/2 tick of bias. When we compute 2400 * the retransmit timer, we want 1/2 tick of rounding and 2401 * 1 extra tick because of +-1/2 tick uncertainty in the 2402 * firing of the timer. The bias will give us exactly the 2403 * 1.5 tick we need. But, because the bias is 2404 * statistical, we have to test that we don't drop below 2405 * the minimum feasible timer (which is 2 ticks). 2406 */ 2407 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2408 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2409 2410 /* 2411 * We received an ack for a packet that wasn't retransmitted; 2412 * it is probably safe to discard any error indications we've 2413 * received recently. This isn't quite right, but close enough 2414 * for now (a route might have failed after we sent a segment, 2415 * and the return path might not be symmetrical). 2416 */ 2417 tp->t_softerror = 0; 2418 } 2419 2420 /* 2421 * Determine a reasonable value for maxseg size. 2422 * If the route is known, check route for mtu. 2423 * If none, use an mss that can be handled on the outgoing 2424 * interface without forcing IP to fragment; if bigger than 2425 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2426 * to utilize large mbufs. If no route is found, route has no mtu, 2427 * or the destination isn't local, use a default, hopefully conservative 2428 * size (usually 512 or the default IP max size, but no more than the mtu 2429 * of the interface), as we can't discover anything about intervening 2430 * gateways or networks. We also initialize the congestion/slow start 2431 * window to be a single segment if the destination isn't local. 2432 * While looking at the routing entry, we also initialize other path-dependent 2433 * parameters from pre-set or cached values in the routing entry. 2434 * 2435 * Also take into account the space needed for options that we 2436 * send regularly. Make maxseg shorter by that amount to assure 2437 * that we can send maxseg amount of data even when the options 2438 * are present. Store the upper limit of the length of options plus 2439 * data in maxopd. 2440 * 2441 * NOTE that this routine is only called when we process an incoming 2442 * segment, for outgoing segments only tcp_mssopt is called. 2443 * 2444 * In case of T/TCP, we call this routine during implicit connection 2445 * setup as well (offer = -1), to initialize maxseg from the cached 2446 * MSS of our peer. 2447 */ 2448 void 2449 tcp_mss(tp, offer) 2450 struct tcpcb *tp; 2451 int offer; 2452 { 2453 register struct rtentry *rt; 2454 struct ifnet *ifp; 2455 register int rtt, mss; 2456 u_long bufsize; 2457 struct inpcb *inp; 2458 struct socket *so; 2459 struct rmxp_tao *taop; 2460 int origoffer = offer; 2461 #ifdef INET6 2462 int isipv6; 2463 int min_protoh; 2464 #endif 2465 2466 inp = tp->t_inpcb; 2467 #ifdef INET6 2468 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2469 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2470 : sizeof (struct tcpiphdr); 2471 #else 2472 #define min_protoh (sizeof (struct tcpiphdr)) 2473 #endif 2474 #ifdef INET6 2475 if (isipv6) 2476 rt = tcp_rtlookup6(&inp->inp_inc); 2477 else 2478 #endif 2479 rt = tcp_rtlookup(&inp->inp_inc); 2480 if (rt == NULL) { 2481 tp->t_maxopd = tp->t_maxseg = 2482 #ifdef INET6 2483 isipv6 ? tcp_v6mssdflt : 2484 #endif /* INET6 */ 2485 tcp_mssdflt; 2486 return; 2487 } 2488 ifp = rt->rt_ifp; 2489 so = inp->inp_socket; 2490 2491 taop = rmx_taop(rt->rt_rmx); 2492 /* 2493 * Offer == -1 means that we didn't receive SYN yet, 2494 * use cached value in that case; 2495 */ 2496 if (offer == -1) 2497 offer = taop->tao_mssopt; 2498 /* 2499 * Offer == 0 means that there was no MSS on the SYN segment, 2500 * in this case we use tcp_mssdflt. 2501 */ 2502 if (offer == 0) 2503 offer = 2504 #ifdef INET6 2505 isipv6 ? tcp_v6mssdflt : 2506 #endif /* INET6 */ 2507 tcp_mssdflt; 2508 else 2509 /* 2510 * Sanity check: make sure that maxopd will be large 2511 * enough to allow some data on segments even is the 2512 * all the option space is used (40bytes). Otherwise 2513 * funny things may happen in tcp_output. 2514 */ 2515 offer = max(offer, 64); 2516 taop->tao_mssopt = offer; 2517 2518 /* 2519 * While we're here, check if there's an initial rtt 2520 * or rttvar. Convert from the route-table units 2521 * to scaled multiples of the slow timeout timer. 2522 */ 2523 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2524 /* 2525 * XXX the lock bit for RTT indicates that the value 2526 * is also a minimum value; this is subject to time. 2527 */ 2528 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2529 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2530 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2531 tcpstat.tcps_usedrtt++; 2532 if (rt->rt_rmx.rmx_rttvar) { 2533 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2534 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2535 tcpstat.tcps_usedrttvar++; 2536 } else { 2537 /* default variation is +- 1 rtt */ 2538 tp->t_rttvar = 2539 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2540 } 2541 TCPT_RANGESET(tp->t_rxtcur, 2542 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2543 tp->t_rttmin, TCPTV_REXMTMAX); 2544 } 2545 /* 2546 * if there's an mtu associated with the route, use it 2547 * else, use the link mtu. 2548 */ 2549 if (rt->rt_rmx.rmx_mtu) 2550 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2551 else 2552 { 2553 mss = 2554 #ifdef INET6 2555 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2556 #endif 2557 ifp->if_mtu 2558 #ifdef INET6 2559 ) 2560 #endif 2561 - min_protoh; 2562 #ifdef INET6 2563 if (isipv6) { 2564 if (!in6_localaddr(&inp->in6p_faddr)) 2565 mss = min(mss, tcp_v6mssdflt); 2566 } else 2567 #endif 2568 if (!in_localaddr(inp->inp_faddr)) 2569 mss = min(mss, tcp_mssdflt); 2570 } 2571 mss = min(mss, offer); 2572 /* 2573 * maxopd stores the maximum length of data AND options 2574 * in a segment; maxseg is the amount of data in a normal 2575 * segment. We need to store this value (maxopd) apart 2576 * from maxseg, because now every segment carries options 2577 * and thus we normally have somewhat less data in segments. 2578 */ 2579 tp->t_maxopd = mss; 2580 2581 /* 2582 * In case of T/TCP, origoffer==-1 indicates, that no segments 2583 * were received yet. In this case we just guess, otherwise 2584 * we do the same as before T/TCP. 2585 */ 2586 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2587 (origoffer == -1 || 2588 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2589 mss -= TCPOLEN_TSTAMP_APPA; 2590 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2591 (origoffer == -1 || 2592 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2593 mss -= TCPOLEN_CC_APPA; 2594 2595 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2596 if (mss > MCLBYTES) 2597 mss &= ~(MCLBYTES-1); 2598 #else 2599 if (mss > MCLBYTES) 2600 mss = mss / MCLBYTES * MCLBYTES; 2601 #endif 2602 /* 2603 * If there's a pipesize, change the socket buffer 2604 * to that size. Make the socket buffers an integral 2605 * number of mss units; if the mss is larger than 2606 * the socket buffer, decrease the mss. 2607 */ 2608 #ifdef RTV_SPIPE 2609 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2610 #endif 2611 bufsize = so->so_snd.sb_hiwat; 2612 if (bufsize < mss) 2613 mss = bufsize; 2614 else { 2615 bufsize = roundup(bufsize, mss); 2616 if (bufsize > sb_max) 2617 bufsize = sb_max; 2618 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2619 } 2620 tp->t_maxseg = mss; 2621 2622 #ifdef RTV_RPIPE 2623 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2624 #endif 2625 bufsize = so->so_rcv.sb_hiwat; 2626 if (bufsize > mss) { 2627 bufsize = roundup(bufsize, mss); 2628 if (bufsize > sb_max) 2629 bufsize = sb_max; 2630 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2631 } 2632 2633 /* 2634 * Set the slow-start flight size depending on whether this 2635 * is a local network or not. 2636 */ 2637 if ( 2638 #ifdef INET6 2639 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2640 (!isipv6 && 2641 #endif 2642 in_localaddr(inp->inp_faddr) 2643 #ifdef INET6 2644 ) 2645 #endif 2646 ) 2647 tp->snd_cwnd = mss * ss_fltsz_local; 2648 else 2649 tp->snd_cwnd = mss * ss_fltsz; 2650 2651 if (rt->rt_rmx.rmx_ssthresh) { 2652 /* 2653 * There's some sort of gateway or interface 2654 * buffer limit on the path. Use this to set 2655 * the slow start threshhold, but set the 2656 * threshold to no less than 2*mss. 2657 */ 2658 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2659 tcpstat.tcps_usedssthresh++; 2660 } 2661 } 2662 2663 /* 2664 * Determine the MSS option to send on an outgoing SYN. 2665 */ 2666 int 2667 tcp_mssopt(tp) 2668 struct tcpcb *tp; 2669 { 2670 struct rtentry *rt; 2671 #ifdef INET6 2672 int isipv6; 2673 int min_protoh; 2674 #endif 2675 2676 #ifdef INET6 2677 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2678 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2679 : sizeof (struct tcpiphdr); 2680 #else 2681 #define min_protoh (sizeof (struct tcpiphdr)) 2682 #endif 2683 #ifdef INET6 2684 if (isipv6) 2685 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2686 else 2687 #endif /* INET6 */ 2688 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2689 if (rt == NULL) 2690 return 2691 #ifdef INET6 2692 isipv6 ? tcp_v6mssdflt : 2693 #endif /* INET6 */ 2694 tcp_mssdflt; 2695 2696 return rt->rt_ifp->if_mtu - min_protoh; 2697 } 2698 2699 2700 /* 2701 * Checks for partial ack. If partial ack arrives, force the retransmission 2702 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2703 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2704 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2705 */ 2706 static int 2707 tcp_newreno(tp, th) 2708 struct tcpcb *tp; 2709 struct tcphdr *th; 2710 { 2711 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2712 tcp_seq onxt = tp->snd_nxt; 2713 u_long ocwnd = tp->snd_cwnd; 2714 2715 callout_stop(tp->tt_rexmt); 2716 tp->t_rtttime = 0; 2717 tp->snd_nxt = th->th_ack; 2718 /* 2719 * Set snd_cwnd to one segment beyond acknowledged offset 2720 * (tp->snd_una has not yet been updated when this function 2721 * is called) 2722 */ 2723 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2724 (void) tcp_output(tp); 2725 tp->snd_cwnd = ocwnd; 2726 if (SEQ_GT(onxt, tp->snd_nxt)) 2727 tp->snd_nxt = onxt; 2728 /* 2729 * Partial window deflation. Relies on fact that tp->snd_una 2730 * not updated yet. 2731 */ 2732 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2733 return (1); 2734 } 2735 return (0); 2736 } 2737