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/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 87 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 88 struct 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> 100 101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 102 103 static int tcprexmtthresh = 3; 104 tcp_cc tcp_ccgen; 105 106 struct tcpstat tcpstat; 107 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 109 110 static int log_in_vain = 0; 111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 112 &log_in_vain, 0, "Log all incoming TCP connections"); 113 114 static int blackhole = 0; 115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 116 &blackhole, 0, "Do not send RST when dropping refused connections"); 117 118 int tcp_delack_enabled = 1; 119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 120 &tcp_delack_enabled, 0, 121 "Delay ACK to try and piggyback it onto a data packet"); 122 123 #ifdef TCP_DROP_SYNFIN 124 static int drop_synfin = 0; 125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 127 #endif 128 129 struct inpcbhead tcb; 130 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 131 struct inpcbinfo tcbinfo; 132 133 static void tcp_dooptions __P((struct tcpopt *, u_char *, int, int)); 134 static void tcp_pulloutofband __P((struct socket *, 135 struct tcphdr *, struct mbuf *, int)); 136 static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *, 137 struct mbuf *)); 138 static void tcp_xmit_timer __P((struct tcpcb *, int)); 139 static int tcp_newreno __P((struct tcpcb *, struct tcphdr *)); 140 141 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 142 #ifdef INET6 143 #define ND6_HINT(tp) \ 144 do { \ 145 if ((tp) && (tp)->t_inpcb && \ 146 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 147 (tp)->t_inpcb->in6p_route.ro_rt) \ 148 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 149 } while (0) 150 #else 151 #define ND6_HINT(tp) 152 #endif 153 154 /* 155 * Indicate whether this ack should be delayed. We can delay the ack if 156 * - delayed acks are enabled and 157 * - there is no delayed ack timer in progress and 158 * - our last ack wasn't a 0-sized window. We never want to delay 159 * the ack that opens up a 0-sized window. 160 */ 161 #define DELAY_ACK(tp) \ 162 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 163 (tp->t_flags & TF_RXWIN0SENT) == 0) 164 165 static int 166 tcp_reass(tp, th, tlenp, m) 167 register struct tcpcb *tp; 168 register struct tcphdr *th; 169 int *tlenp; 170 struct mbuf *m; 171 { 172 struct tseg_qent *q; 173 struct tseg_qent *p = NULL; 174 struct tseg_qent *nq; 175 struct tseg_qent *te; 176 struct socket *so = tp->t_inpcb->inp_socket; 177 int flags; 178 179 /* 180 * Call with th==0 after become established to 181 * force pre-ESTABLISHED data up to user socket. 182 */ 183 if (th == 0) 184 goto present; 185 186 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 187 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 188 M_NOWAIT); 189 if (te == NULL) { 190 tcpstat.tcps_rcvmemdrop++; 191 m_freem(m); 192 return (0); 193 } 194 195 /* 196 * Find a segment which begins after this one does. 197 */ 198 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 199 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 200 break; 201 p = q; 202 } 203 204 /* 205 * If there is a preceding segment, it may provide some of 206 * our data already. If so, drop the data from the incoming 207 * segment. If it provides all of our data, drop us. 208 */ 209 if (p != NULL) { 210 register int i; 211 /* conversion to int (in i) handles seq wraparound */ 212 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 213 if (i > 0) { 214 if (i >= *tlenp) { 215 tcpstat.tcps_rcvduppack++; 216 tcpstat.tcps_rcvdupbyte += *tlenp; 217 m_freem(m); 218 FREE(te, M_TSEGQ); 219 /* 220 * Try to present any queued data 221 * at the left window edge to the user. 222 * This is needed after the 3-WHS 223 * completes. 224 */ 225 goto present; /* ??? */ 226 } 227 m_adj(m, i); 228 *tlenp -= i; 229 th->th_seq += i; 230 } 231 } 232 tcpstat.tcps_rcvoopack++; 233 tcpstat.tcps_rcvoobyte += *tlenp; 234 235 /* 236 * While we overlap succeeding segments trim them or, 237 * if they are completely covered, dequeue them. 238 */ 239 while (q) { 240 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 241 if (i <= 0) 242 break; 243 if (i < q->tqe_len) { 244 q->tqe_th->th_seq += i; 245 q->tqe_len -= i; 246 m_adj(q->tqe_m, i); 247 break; 248 } 249 250 nq = LIST_NEXT(q, tqe_q); 251 LIST_REMOVE(q, tqe_q); 252 m_freem(q->tqe_m); 253 FREE(q, M_TSEGQ); 254 q = nq; 255 } 256 257 /* Insert the new segment queue entry into place. */ 258 te->tqe_m = m; 259 te->tqe_th = th; 260 te->tqe_len = *tlenp; 261 262 if (p == NULL) { 263 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 264 } else { 265 LIST_INSERT_AFTER(p, te, tqe_q); 266 } 267 268 present: 269 /* 270 * Present data to user, advancing rcv_nxt through 271 * completed sequence space. 272 */ 273 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 274 return (0); 275 q = LIST_FIRST(&tp->t_segq); 276 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 277 return (0); 278 do { 279 tp->rcv_nxt += q->tqe_len; 280 flags = q->tqe_th->th_flags & TH_FIN; 281 nq = LIST_NEXT(q, tqe_q); 282 LIST_REMOVE(q, tqe_q); 283 if (so->so_state & SS_CANTRCVMORE) 284 m_freem(q->tqe_m); 285 else 286 sbappend(&so->so_rcv, q->tqe_m); 287 FREE(q, M_TSEGQ); 288 q = nq; 289 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 290 ND6_HINT(tp); 291 sorwakeup(so); 292 return (flags); 293 } 294 295 /* 296 * TCP input routine, follows pages 65-76 of the 297 * protocol specification dated September, 1981 very closely. 298 */ 299 #ifdef INET6 300 int 301 tcp6_input(mp, offp, proto) 302 struct mbuf **mp; 303 int *offp, proto; 304 { 305 register struct mbuf *m = *mp; 306 struct in6_ifaddr *ia6; 307 308 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 309 310 /* 311 * draft-itojun-ipv6-tcp-to-anycast 312 * better place to put this in? 313 */ 314 ia6 = ip6_getdstifaddr(m); 315 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 316 struct ip6_hdr *ip6; 317 318 ip6 = mtod(m, struct ip6_hdr *); 319 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 320 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 321 return IPPROTO_DONE; 322 } 323 324 tcp_input(m, *offp); 325 return IPPROTO_DONE; 326 } 327 #endif 328 329 void 330 tcp_input(m, off0) 331 register struct mbuf *m; 332 int off0; 333 { 334 register struct tcphdr *th; 335 register struct ip *ip = NULL; 336 register struct ipovly *ipov; 337 register struct inpcb *inp; 338 u_char *optp = NULL; 339 int optlen = 0; 340 int len, tlen, off; 341 int drop_hdrlen; 342 register struct tcpcb *tp = 0; 343 register int thflags; 344 struct socket *so = 0; 345 int todrop, acked, ourfinisacked, needoutput = 0; 346 int iss = 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 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 NTOHL(th->th_seq); 490 NTOHL(th->th_ack); 491 NTOHS(th->th_win); 492 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 */ 502 drop_hdrlen = off0 + off; 503 504 /* 505 * Locate pcb for segment. 506 */ 507 findpcb: 508 #ifdef IPFIREWALL_FORWARD 509 if (ip_fw_fwd_addr != NULL 510 #ifdef INET6 511 && isipv6 == NULL /* IPv6 support is not yet */ 512 #endif /* INET6 */ 513 ) { 514 /* 515 * Diverted. Pretend to be the destination. 516 * already got one like this? 517 */ 518 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 519 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 520 if (!inp) { 521 /* 522 * No, then it's new. Try find the ambushing socket 523 */ 524 if (!ip_fw_fwd_addr->sin_port) { 525 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 526 th->th_sport, ip_fw_fwd_addr->sin_addr, 527 th->th_dport, 1, m->m_pkthdr.rcvif); 528 } else { 529 inp = in_pcblookup_hash(&tcbinfo, 530 ip->ip_src, th->th_sport, 531 ip_fw_fwd_addr->sin_addr, 532 ntohs(ip_fw_fwd_addr->sin_port), 1, 533 m->m_pkthdr.rcvif); 534 } 535 } 536 ip_fw_fwd_addr = NULL; 537 } else 538 #endif /* IPFIREWALL_FORWARD */ 539 { 540 #ifdef INET6 541 if (isipv6) 542 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 543 &ip6->ip6_dst, th->th_dport, 1, 544 m->m_pkthdr.rcvif); 545 else 546 #endif /* INET6 */ 547 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 548 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 549 } 550 551 #ifdef IPSEC 552 #ifdef INET6 553 if (isipv6) { 554 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 555 ipsec6stat.in_polvio++; 556 goto drop; 557 } 558 } else 559 #endif /* INET6 */ 560 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 561 ipsecstat.in_polvio++; 562 goto drop; 563 } 564 #endif /*IPSEC*/ 565 566 /* 567 * If the state is CLOSED (i.e., TCB does not exist) then 568 * all data in the incoming segment is discarded. 569 * If the TCB exists but is in CLOSED state, it is embryonic, 570 * but should either do a listen or a connect soon. 571 */ 572 if (inp == NULL) { 573 if (log_in_vain) { 574 #ifdef INET6 575 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 576 #else /* INET6 */ 577 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 578 #endif /* INET6 */ 579 580 #ifdef INET6 581 if (isipv6) { 582 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 583 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 584 } else 585 #endif 586 { 587 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 588 strcpy(sbuf, inet_ntoa(ip->ip_src)); 589 } 590 switch (log_in_vain) { 591 case 1: 592 if(thflags & TH_SYN) 593 log(LOG_INFO, 594 "Connection attempt to TCP %s:%d from %s:%d\n", 595 dbuf, ntohs(th->th_dport), 596 sbuf, 597 ntohs(th->th_sport)); 598 break; 599 case 2: 600 log(LOG_INFO, 601 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 602 dbuf, ntohs(th->th_dport), sbuf, 603 ntohs(th->th_sport), thflags); 604 break; 605 default: 606 break; 607 } 608 } 609 if (blackhole) { 610 switch (blackhole) { 611 case 1: 612 if (thflags & TH_SYN) 613 goto drop; 614 break; 615 case 2: 616 goto drop; 617 default: 618 goto drop; 619 } 620 } 621 rstreason = BANDLIM_RST_CLOSEDPORT; 622 goto dropwithreset; 623 } 624 tp = intotcpcb(inp); 625 if (tp == 0) { 626 rstreason = BANDLIM_RST_CLOSEDPORT; 627 goto dropwithreset; 628 } 629 if (tp->t_state == TCPS_CLOSED) 630 goto drop; 631 632 /* Unscale the window into a 32-bit value. */ 633 if ((thflags & TH_SYN) == 0) 634 tiwin = th->th_win << tp->snd_scale; 635 else 636 tiwin = th->th_win; 637 638 so = inp->inp_socket; 639 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 640 struct in_conninfo inc; 641 #ifdef TCPDEBUG 642 if (so->so_options & SO_DEBUG) { 643 ostate = tp->t_state; 644 #ifdef INET6 645 if (isipv6) 646 bcopy((char *)ip6, (char *)tcp_saveipgen, 647 sizeof(*ip6)); 648 else 649 #endif /* INET6 */ 650 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 651 tcp_savetcp = *th; 652 } 653 #endif 654 /* skip if this isn't a listen socket */ 655 if ((so->so_options & SO_ACCEPTCONN) == 0) 656 goto after_listen; 657 #ifdef INET6 658 inc.inc_isipv6 = isipv6; 659 if (isipv6) { 660 inc.inc6_faddr = ip6->ip6_src; 661 inc.inc6_laddr = ip6->ip6_dst; 662 inc.inc6_route.ro_rt = NULL; /* XXX */ 663 664 } else 665 #endif /* INET6 */ 666 { 667 inc.inc_faddr = ip->ip_src; 668 inc.inc_laddr = ip->ip_dst; 669 inc.inc_route.ro_rt = NULL; /* XXX */ 670 } 671 inc.inc_fport = th->th_sport; 672 inc.inc_lport = th->th_dport; 673 674 /* 675 * If the state is LISTEN then ignore segment if it contains 676 * a RST. If the segment contains an ACK then it is bad and 677 * send a RST. If it does not contain a SYN then it is not 678 * interesting; drop it. 679 * 680 * If the state is SYN_RECEIVED (syncache) and seg contains 681 * an ACK, but not for our SYN/ACK, send a RST. If the seg 682 * contains a RST, check the sequence number to see if it 683 * is a valid reset segment. 684 */ 685 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 686 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 687 if (!syncache_expand(&inc, th, &so, m)) { 688 /* 689 * No syncache entry, or ACK was not 690 * for our SYN/ACK. Send a RST. 691 */ 692 tcpstat.tcps_badsyn++; 693 rstreason = BANDLIM_RST_OPENPORT; 694 goto dropwithreset; 695 } 696 if (so == NULL) 697 /* 698 * Could not complete 3-way handshake, 699 * connection is being closed down, and 700 * syncache will free mbuf. 701 */ 702 return; 703 /* 704 * Socket is created in state SYN_RECEIVED. 705 * Continue processing segment. 706 */ 707 inp = sotoinpcb(so); 708 tp = intotcpcb(inp); 709 /* 710 * This is what would have happened in 711 * tcp_ouput() when the SYN,ACK was sent. 712 */ 713 tp->snd_up = tp->snd_una; 714 tp->snd_max = tp->snd_nxt = tp->iss + 1; 715 tp->last_ack_sent = tp->rcv_nxt; 716 /* 717 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 718 * until the _second_ ACK is received: 719 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 720 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 721 * move to ESTAB, set snd_wnd to tiwin. 722 */ 723 tp->snd_wnd = tiwin; /* unscaled */ 724 goto after_listen; 725 } 726 if (thflags & TH_RST) { 727 syncache_chkrst(&inc, th); 728 goto drop; 729 } 730 if (thflags & TH_ACK) { 731 syncache_badack(&inc); 732 tcpstat.tcps_badsyn++; 733 rstreason = BANDLIM_RST_OPENPORT; 734 goto dropwithreset; 735 } 736 goto drop; 737 } 738 739 /* 740 * Segment's flags are (SYN) or (SYN|FIN). 741 */ 742 #ifdef INET6 743 /* 744 * If deprecated address is forbidden, 745 * we do not accept SYN to deprecated interface 746 * address to prevent any new inbound connection from 747 * getting established. 748 * When we do not accept SYN, we send a TCP RST, 749 * with deprecated source address (instead of dropping 750 * it). We compromise it as it is much better for peer 751 * to send a RST, and RST will be the final packet 752 * for the exchange. 753 * 754 * If we do not forbid deprecated addresses, we accept 755 * the SYN packet. RFC2462 does not suggest dropping 756 * SYN in this case. 757 * If we decipher RFC2462 5.5.4, it says like this: 758 * 1. use of deprecated addr with existing 759 * communication is okay - "SHOULD continue to be 760 * used" 761 * 2. use of it with new communication: 762 * (2a) "SHOULD NOT be used if alternate address 763 * with sufficient scope is available" 764 * (2b) nothing mentioned otherwise. 765 * Here we fall into (2b) case as we have no choice in 766 * our source address selection - we must obey the peer. 767 * 768 * The wording in RFC2462 is confusing, and there are 769 * multiple description text for deprecated address 770 * handling - worse, they are not exactly the same. 771 * I believe 5.5.4 is the best one, so we follow 5.5.4. 772 */ 773 if (isipv6 && !ip6_use_deprecated) { 774 struct in6_ifaddr *ia6; 775 776 if ((ia6 = ip6_getdstifaddr(m)) && 777 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 778 tp = NULL; 779 rstreason = BANDLIM_RST_OPENPORT; 780 goto dropwithreset; 781 } 782 } 783 #endif 784 /* 785 * If it is from this socket, drop it, it must be forged. 786 * Don't bother responding if the destination was a broadcast. 787 */ 788 if (th->th_dport == th->th_sport) { 789 #ifdef INET6 790 if (isipv6) { 791 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 792 &ip6->ip6_src)) 793 goto drop; 794 } else 795 #endif /* INET6 */ 796 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 797 goto drop; 798 } 799 /* 800 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 801 * in_broadcast() should never return true on a received 802 * packet with M_BCAST not set. 803 * 804 * Packets with a multicast source address should also 805 * be discarded. 806 */ 807 if (m->m_flags & (M_BCAST|M_MCAST)) 808 goto drop; 809 #ifdef INET6 810 if (isipv6) { 811 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 812 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 813 goto drop; 814 } else 815 #endif 816 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 817 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 818 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 819 goto drop; 820 /* 821 * SYN appears to be valid; create compressed TCP state 822 * for syncache, or perform t/tcp connection. 823 */ 824 if (so->so_qlen <= so->so_qlimit) { 825 tcp_dooptions(&to, optp, optlen, 1); 826 if (!syncache_add(&inc, &to, th, &so, m)) 827 goto drop; 828 if (so == NULL) 829 /* 830 * Entry added to syncache, mbuf used to 831 * send SYN,ACK packet. 832 */ 833 return; 834 /* 835 * Segment passed TAO tests. 836 */ 837 inp = sotoinpcb(so); 838 tp = intotcpcb(inp); 839 tp->snd_wnd = tiwin; 840 tp->t_starttime = ticks; 841 tp->t_state = TCPS_ESTABLISHED; 842 843 /* 844 * If there is a FIN, or if there is data and the 845 * connection is local, then delay SYN,ACK(SYN) in 846 * the hope of piggy-backing it on a response 847 * segment. Otherwise must send ACK now in case 848 * the other side is slow starting. 849 */ 850 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 851 (tlen != 0 && 852 #ifdef INET6 853 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 854 || 855 (!isipv6 && 856 #endif 857 in_localaddr(inp->inp_faddr) 858 #ifdef INET6 859 )) 860 #endif 861 ))) { 862 callout_reset(tp->tt_delack, tcp_delacktime, 863 tcp_timer_delack, tp); 864 tp->t_flags |= TF_NEEDSYN; 865 } else 866 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 867 868 tcpstat.tcps_connects++; 869 soisconnected(so); 870 goto trimthenstep6; 871 } 872 goto drop; 873 } 874 after_listen: 875 876 /* XXX temp debugging */ 877 /* should not happen - syncache should pick up these connections */ 878 if (tp->t_state == TCPS_LISTEN) 879 panic("tcp_input: TCPS_LISTEN"); 880 881 /* 882 * Segment received on connection. 883 * Reset idle time and keep-alive timer. 884 */ 885 tp->t_rcvtime = ticks; 886 if (TCPS_HAVEESTABLISHED(tp->t_state)) 887 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 888 889 /* 890 * Process options. 891 * XXX this is tradtitional behavior, may need to be cleaned up. 892 */ 893 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 894 if (thflags & TH_SYN) { 895 if (to.to_flags & TOF_SCALE) { 896 tp->t_flags |= TF_RCVD_SCALE; 897 tp->requested_s_scale = to.to_requested_s_scale; 898 } 899 if (to.to_flags & TOF_TS) { 900 tp->t_flags |= TF_RCVD_TSTMP; 901 tp->ts_recent = to.to_tsval; 902 tp->ts_recent_age = ticks; 903 } 904 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 905 tp->t_flags |= TF_RCVD_CC; 906 if (to.to_flags & TOF_MSS) 907 tcp_mss(tp, to.to_mss); 908 } 909 910 /* 911 * Header prediction: check for the two common cases 912 * of a uni-directional data xfer. If the packet has 913 * no control flags, is in-sequence, the window didn't 914 * change and we're not retransmitting, it's a 915 * candidate. If the length is zero and the ack moved 916 * forward, we're the sender side of the xfer. Just 917 * free the data acked & wake any higher level process 918 * that was blocked waiting for space. If the length 919 * is non-zero and the ack didn't move, we're the 920 * receiver side. If we're getting packets in-order 921 * (the reassembly queue is empty), add the data to 922 * the socket buffer and note that we need a delayed ack. 923 * Make sure that the hidden state-flags are also off. 924 * Since we check for TCPS_ESTABLISHED above, it can only 925 * be TH_NEEDSYN. 926 */ 927 if (tp->t_state == TCPS_ESTABLISHED && 928 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 929 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 930 ((to.to_flags & TOF_TS) == 0 || 931 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 932 /* 933 * Using the CC option is compulsory if once started: 934 * the segment is OK if no T/TCP was negotiated or 935 * if the segment has a CC option equal to CCrecv 936 */ 937 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 938 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 939 th->th_seq == tp->rcv_nxt && 940 tiwin && tiwin == tp->snd_wnd && 941 tp->snd_nxt == tp->snd_max) { 942 943 /* 944 * If last ACK falls within this segment's sequence numbers, 945 * record the timestamp. 946 * NOTE that the test is modified according to the latest 947 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 948 */ 949 if ((to.to_flags & TOF_TS) != 0 && 950 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 951 tp->ts_recent_age = ticks; 952 tp->ts_recent = to.to_tsval; 953 } 954 955 if (tlen == 0) { 956 if (SEQ_GT(th->th_ack, tp->snd_una) && 957 SEQ_LEQ(th->th_ack, tp->snd_max) && 958 tp->snd_cwnd >= tp->snd_wnd && 959 tp->t_dupacks < tcprexmtthresh) { 960 /* 961 * this is a pure ack for outstanding data. 962 */ 963 ++tcpstat.tcps_predack; 964 /* 965 * "bad retransmit" recovery 966 */ 967 if (tp->t_rxtshift == 1 && 968 ticks < tp->t_badrxtwin) { 969 tp->snd_cwnd = tp->snd_cwnd_prev; 970 tp->snd_ssthresh = 971 tp->snd_ssthresh_prev; 972 tp->snd_nxt = tp->snd_max; 973 tp->t_badrxtwin = 0; 974 } 975 if ((to.to_flags & TOF_TS) != 0) 976 tcp_xmit_timer(tp, 977 ticks - to.to_tsecr + 1); 978 else if (tp->t_rtttime && 979 SEQ_GT(th->th_ack, tp->t_rtseq)) 980 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 981 acked = th->th_ack - tp->snd_una; 982 tcpstat.tcps_rcvackpack++; 983 tcpstat.tcps_rcvackbyte += acked; 984 sbdrop(&so->so_snd, acked); 985 tp->snd_una = th->th_ack; 986 m_freem(m); 987 ND6_HINT(tp); /* some progress has been done */ 988 989 /* 990 * If all outstanding data are acked, stop 991 * retransmit timer, otherwise restart timer 992 * using current (possibly backed-off) value. 993 * If process is waiting for space, 994 * wakeup/selwakeup/signal. If data 995 * are ready to send, let tcp_output 996 * decide between more output or persist. 997 */ 998 if (tp->snd_una == tp->snd_max) 999 callout_stop(tp->tt_rexmt); 1000 else if (!callout_active(tp->tt_persist)) 1001 callout_reset(tp->tt_rexmt, 1002 tp->t_rxtcur, 1003 tcp_timer_rexmt, tp); 1004 1005 sowwakeup(so); 1006 if (so->so_snd.sb_cc) 1007 (void) tcp_output(tp); 1008 return; 1009 } 1010 } else if (th->th_ack == tp->snd_una && 1011 LIST_EMPTY(&tp->t_segq) && 1012 tlen <= sbspace(&so->so_rcv)) { 1013 /* 1014 * this is a pure, in-sequence data packet 1015 * with nothing on the reassembly queue and 1016 * we have enough buffer space to take it. 1017 */ 1018 ++tcpstat.tcps_preddat; 1019 tp->rcv_nxt += tlen; 1020 tcpstat.tcps_rcvpack++; 1021 tcpstat.tcps_rcvbyte += tlen; 1022 ND6_HINT(tp); /* some progress has been done */ 1023 /* 1024 * Add data to socket buffer. 1025 */ 1026 m_adj(m, drop_hdrlen); /* delayed header drop */ 1027 sbappend(&so->so_rcv, m); 1028 sorwakeup(so); 1029 if (DELAY_ACK(tp)) { 1030 callout_reset(tp->tt_delack, tcp_delacktime, 1031 tcp_timer_delack, tp); 1032 } else { 1033 tp->t_flags |= TF_ACKNOW; 1034 tcp_output(tp); 1035 } 1036 return; 1037 } 1038 } 1039 1040 /* 1041 * Calculate amount of space in receive window, 1042 * and then do TCP input processing. 1043 * Receive window is amount of space in rcv queue, 1044 * but not less than advertised window. 1045 */ 1046 { int win; 1047 1048 win = sbspace(&so->so_rcv); 1049 if (win < 0) 1050 win = 0; 1051 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1052 } 1053 1054 switch (tp->t_state) { 1055 1056 /* 1057 * If the state is SYN_RECEIVED: 1058 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1059 */ 1060 case TCPS_SYN_RECEIVED: 1061 if ((thflags & TH_ACK) && 1062 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1063 SEQ_GT(th->th_ack, tp->snd_max))) { 1064 rstreason = BANDLIM_RST_OPENPORT; 1065 goto dropwithreset; 1066 } 1067 break; 1068 1069 /* 1070 * If the state is SYN_SENT: 1071 * if seg contains an ACK, but not for our SYN, drop the input. 1072 * if seg contains a RST, then drop the connection. 1073 * if seg does not contain SYN, then drop it. 1074 * Otherwise this is an acceptable SYN segment 1075 * initialize tp->rcv_nxt and tp->irs 1076 * if seg contains ack then advance tp->snd_una 1077 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1078 * arrange for segment to be acked (eventually) 1079 * continue processing rest of data/controls, beginning with URG 1080 */ 1081 case TCPS_SYN_SENT: 1082 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1083 taop = &tao_noncached; 1084 bzero(taop, sizeof(*taop)); 1085 } 1086 1087 if ((thflags & TH_ACK) && 1088 (SEQ_LEQ(th->th_ack, tp->iss) || 1089 SEQ_GT(th->th_ack, tp->snd_max))) { 1090 /* 1091 * If we have a cached CCsent for the remote host, 1092 * hence we haven't just crashed and restarted, 1093 * do not send a RST. This may be a retransmission 1094 * from the other side after our earlier ACK was lost. 1095 * Our new SYN, when it arrives, will serve as the 1096 * needed ACK. 1097 */ 1098 if (taop->tao_ccsent != 0) 1099 goto drop; 1100 else { 1101 rstreason = BANDLIM_UNLIMITED; 1102 goto dropwithreset; 1103 } 1104 } 1105 if (thflags & TH_RST) { 1106 if (thflags & TH_ACK) 1107 tp = tcp_drop(tp, ECONNREFUSED); 1108 goto drop; 1109 } 1110 if ((thflags & TH_SYN) == 0) 1111 goto drop; 1112 tp->snd_wnd = th->th_win; /* initial send window */ 1113 tp->cc_recv = to.to_cc; /* foreign CC */ 1114 1115 tp->irs = th->th_seq; 1116 tcp_rcvseqinit(tp); 1117 if (thflags & TH_ACK) { 1118 /* 1119 * Our SYN was acked. If segment contains CC.ECHO 1120 * option, check it to make sure this segment really 1121 * matches our SYN. If not, just drop it as old 1122 * duplicate, but send an RST if we're still playing 1123 * by the old rules. If no CC.ECHO option, make sure 1124 * we don't get fooled into using T/TCP. 1125 */ 1126 if (to.to_flags & TOF_CCECHO) { 1127 if (tp->cc_send != to.to_ccecho) { 1128 if (taop->tao_ccsent != 0) 1129 goto drop; 1130 else { 1131 rstreason = BANDLIM_UNLIMITED; 1132 goto dropwithreset; 1133 } 1134 } 1135 } else 1136 tp->t_flags &= ~TF_RCVD_CC; 1137 tcpstat.tcps_connects++; 1138 soisconnected(so); 1139 /* Do window scaling on this connection? */ 1140 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1141 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1142 tp->snd_scale = tp->requested_s_scale; 1143 tp->rcv_scale = tp->request_r_scale; 1144 } 1145 /* Segment is acceptable, update cache if undefined. */ 1146 if (taop->tao_ccsent == 0) 1147 taop->tao_ccsent = to.to_ccecho; 1148 1149 tp->rcv_adv += tp->rcv_wnd; 1150 tp->snd_una++; /* SYN is acked */ 1151 /* 1152 * If there's data, delay ACK; if there's also a FIN 1153 * ACKNOW will be turned on later. 1154 */ 1155 if (DELAY_ACK(tp) && tlen != 0) 1156 callout_reset(tp->tt_delack, tcp_delacktime, 1157 tcp_timer_delack, tp); 1158 else 1159 tp->t_flags |= TF_ACKNOW; 1160 /* 1161 * Received <SYN,ACK> in SYN_SENT[*] state. 1162 * Transitions: 1163 * SYN_SENT --> ESTABLISHED 1164 * SYN_SENT* --> FIN_WAIT_1 1165 */ 1166 tp->t_starttime = ticks; 1167 if (tp->t_flags & TF_NEEDFIN) { 1168 tp->t_state = TCPS_FIN_WAIT_1; 1169 tp->t_flags &= ~TF_NEEDFIN; 1170 thflags &= ~TH_SYN; 1171 } else { 1172 tp->t_state = TCPS_ESTABLISHED; 1173 callout_reset(tp->tt_keep, tcp_keepidle, 1174 tcp_timer_keep, tp); 1175 } 1176 } else { 1177 /* 1178 * Received initial SYN in SYN-SENT[*] state => simul- 1179 * taneous open. If segment contains CC option and there is 1180 * a cached CC, apply TAO test; if it succeeds, connection is 1181 * half-synchronized. Otherwise, do 3-way handshake: 1182 * SYN-SENT -> SYN-RECEIVED 1183 * SYN-SENT* -> SYN-RECEIVED* 1184 * If there was no CC option, clear cached CC value. 1185 */ 1186 tp->t_flags |= TF_ACKNOW; 1187 callout_stop(tp->tt_rexmt); 1188 if (to.to_flags & TOF_CC) { 1189 if (taop->tao_cc != 0 && 1190 CC_GT(to.to_cc, taop->tao_cc)) { 1191 /* 1192 * update cache and make transition: 1193 * SYN-SENT -> ESTABLISHED* 1194 * SYN-SENT* -> FIN-WAIT-1* 1195 */ 1196 taop->tao_cc = to.to_cc; 1197 tp->t_starttime = ticks; 1198 if (tp->t_flags & TF_NEEDFIN) { 1199 tp->t_state = TCPS_FIN_WAIT_1; 1200 tp->t_flags &= ~TF_NEEDFIN; 1201 } else { 1202 tp->t_state = TCPS_ESTABLISHED; 1203 callout_reset(tp->tt_keep, 1204 tcp_keepidle, 1205 tcp_timer_keep, 1206 tp); 1207 } 1208 tp->t_flags |= TF_NEEDSYN; 1209 } else 1210 tp->t_state = TCPS_SYN_RECEIVED; 1211 } else { 1212 /* CC.NEW or no option => invalidate cache */ 1213 taop->tao_cc = 0; 1214 tp->t_state = TCPS_SYN_RECEIVED; 1215 } 1216 } 1217 1218 trimthenstep6: 1219 /* 1220 * Advance th->th_seq to correspond to first data byte. 1221 * If data, trim to stay within window, 1222 * dropping FIN if necessary. 1223 */ 1224 th->th_seq++; 1225 if (tlen > tp->rcv_wnd) { 1226 todrop = tlen - tp->rcv_wnd; 1227 m_adj(m, -todrop); 1228 tlen = tp->rcv_wnd; 1229 thflags &= ~TH_FIN; 1230 tcpstat.tcps_rcvpackafterwin++; 1231 tcpstat.tcps_rcvbyteafterwin += todrop; 1232 } 1233 tp->snd_wl1 = th->th_seq - 1; 1234 tp->rcv_up = th->th_seq; 1235 /* 1236 * Client side of transaction: already sent SYN and data. 1237 * If the remote host used T/TCP to validate the SYN, 1238 * our data will be ACK'd; if so, enter normal data segment 1239 * processing in the middle of step 5, ack processing. 1240 * Otherwise, goto step 6. 1241 */ 1242 if (thflags & TH_ACK) 1243 goto process_ACK; 1244 goto step6; 1245 /* 1246 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1247 * if segment contains a SYN and CC [not CC.NEW] option: 1248 * if state == TIME_WAIT and connection duration > MSL, 1249 * drop packet and send RST; 1250 * 1251 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1252 * ack the FIN (and data) in retransmission queue. 1253 * Complete close and delete TCPCB. Then reprocess 1254 * segment, hoping to find new TCPCB in LISTEN state; 1255 * 1256 * else must be old SYN; drop it. 1257 * else do normal processing. 1258 */ 1259 case TCPS_LAST_ACK: 1260 case TCPS_CLOSING: 1261 case TCPS_TIME_WAIT: 1262 if ((thflags & TH_SYN) && 1263 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1264 if (tp->t_state == TCPS_TIME_WAIT && 1265 (ticks - tp->t_starttime) > tcp_msl) { 1266 rstreason = BANDLIM_UNLIMITED; 1267 goto dropwithreset; 1268 } 1269 if (CC_GT(to.to_cc, tp->cc_recv)) { 1270 tp = tcp_close(tp); 1271 goto findpcb; 1272 } 1273 else 1274 goto drop; 1275 } 1276 break; /* continue normal processing */ 1277 } 1278 1279 /* 1280 * States other than LISTEN or SYN_SENT. 1281 * First check the RST flag and sequence number since reset segments 1282 * are exempt from the timestamp and connection count tests. This 1283 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1284 * below which allowed reset segments in half the sequence space 1285 * to fall though and be processed (which gives forged reset 1286 * segments with a random sequence number a 50 percent chance of 1287 * killing a connection). 1288 * Then check timestamp, if present. 1289 * Then check the connection count, if present. 1290 * Then check that at least some bytes of segment are within 1291 * receive window. If segment begins before rcv_nxt, 1292 * drop leading data (and SYN); if nothing left, just ack. 1293 * 1294 * 1295 * If the RST bit is set, check the sequence number to see 1296 * if this is a valid reset segment. 1297 * RFC 793 page 37: 1298 * In all states except SYN-SENT, all reset (RST) segments 1299 * are validated by checking their SEQ-fields. A reset is 1300 * valid if its sequence number is in the window. 1301 * Note: this does not take into account delayed ACKs, so 1302 * we should test against last_ack_sent instead of rcv_nxt. 1303 * The sequence number in the reset segment is normally an 1304 * echo of our outgoing acknowlegement numbers, but some hosts 1305 * send a reset with the sequence number at the rightmost edge 1306 * of our receive window, and we have to handle this case. 1307 * If we have multiple segments in flight, the intial reset 1308 * segment sequence numbers will be to the left of last_ack_sent, 1309 * but they will eventually catch up. 1310 * In any case, it never made sense to trim reset segments to 1311 * fit the receive window since RFC 1122 says: 1312 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1313 * 1314 * A TCP SHOULD allow a received RST segment to include data. 1315 * 1316 * DISCUSSION 1317 * It has been suggested that a RST segment could contain 1318 * ASCII text that encoded and explained the cause of the 1319 * RST. No standard has yet been established for such 1320 * data. 1321 * 1322 * If the reset segment passes the sequence number test examine 1323 * the state: 1324 * SYN_RECEIVED STATE: 1325 * If passive open, return to LISTEN state. 1326 * If active open, inform user that connection was refused. 1327 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1328 * Inform user that connection was reset, and close tcb. 1329 * CLOSING, LAST_ACK STATES: 1330 * Close the tcb. 1331 * TIME_WAIT STATE: 1332 * Drop the segment - see Stevens, vol. 2, p. 964 and 1333 * RFC 1337. 1334 */ 1335 if (thflags & TH_RST) { 1336 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1337 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1338 switch (tp->t_state) { 1339 1340 case TCPS_SYN_RECEIVED: 1341 so->so_error = ECONNREFUSED; 1342 goto close; 1343 1344 case TCPS_ESTABLISHED: 1345 case TCPS_FIN_WAIT_1: 1346 case TCPS_FIN_WAIT_2: 1347 case TCPS_CLOSE_WAIT: 1348 so->so_error = ECONNRESET; 1349 close: 1350 tp->t_state = TCPS_CLOSED; 1351 tcpstat.tcps_drops++; 1352 tp = tcp_close(tp); 1353 break; 1354 1355 case TCPS_CLOSING: 1356 case TCPS_LAST_ACK: 1357 tp = tcp_close(tp); 1358 break; 1359 1360 case TCPS_TIME_WAIT: 1361 break; 1362 } 1363 } 1364 goto drop; 1365 } 1366 1367 /* 1368 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1369 * and it's less than ts_recent, drop it. 1370 */ 1371 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1372 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1373 1374 /* Check to see if ts_recent is over 24 days old. */ 1375 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1376 /* 1377 * Invalidate ts_recent. If this segment updates 1378 * ts_recent, the age will be reset later and ts_recent 1379 * will get a valid value. If it does not, setting 1380 * ts_recent to zero will at least satisfy the 1381 * requirement that zero be placed in the timestamp 1382 * echo reply when ts_recent isn't valid. The 1383 * age isn't reset until we get a valid ts_recent 1384 * because we don't want out-of-order segments to be 1385 * dropped when ts_recent is old. 1386 */ 1387 tp->ts_recent = 0; 1388 } else { 1389 tcpstat.tcps_rcvduppack++; 1390 tcpstat.tcps_rcvdupbyte += tlen; 1391 tcpstat.tcps_pawsdrop++; 1392 goto dropafterack; 1393 } 1394 } 1395 1396 /* 1397 * T/TCP mechanism 1398 * If T/TCP was negotiated and the segment doesn't have CC, 1399 * or if its CC is wrong then drop the segment. 1400 * RST segments do not have to comply with this. 1401 */ 1402 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1403 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1404 goto dropafterack; 1405 1406 /* 1407 * In the SYN-RECEIVED state, validate that the packet belongs to 1408 * this connection before trimming the data to fit the receive 1409 * window. Check the sequence number versus IRS since we know 1410 * the sequence numbers haven't wrapped. This is a partial fix 1411 * for the "LAND" DoS attack. 1412 */ 1413 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1414 rstreason = BANDLIM_RST_OPENPORT; 1415 goto dropwithreset; 1416 } 1417 1418 todrop = tp->rcv_nxt - th->th_seq; 1419 if (todrop > 0) { 1420 if (thflags & TH_SYN) { 1421 thflags &= ~TH_SYN; 1422 th->th_seq++; 1423 if (th->th_urp > 1) 1424 th->th_urp--; 1425 else 1426 thflags &= ~TH_URG; 1427 todrop--; 1428 } 1429 /* 1430 * Following if statement from Stevens, vol. 2, p. 960. 1431 */ 1432 if (todrop > tlen 1433 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1434 /* 1435 * Any valid FIN must be to the left of the window. 1436 * At this point the FIN must be a duplicate or out 1437 * of sequence; drop it. 1438 */ 1439 thflags &= ~TH_FIN; 1440 1441 /* 1442 * Send an ACK to resynchronize and drop any data. 1443 * But keep on processing for RST or ACK. 1444 */ 1445 tp->t_flags |= TF_ACKNOW; 1446 todrop = tlen; 1447 tcpstat.tcps_rcvduppack++; 1448 tcpstat.tcps_rcvdupbyte += todrop; 1449 } else { 1450 tcpstat.tcps_rcvpartduppack++; 1451 tcpstat.tcps_rcvpartdupbyte += todrop; 1452 } 1453 drop_hdrlen += todrop; /* drop from the top afterwards */ 1454 th->th_seq += todrop; 1455 tlen -= todrop; 1456 if (th->th_urp > todrop) 1457 th->th_urp -= todrop; 1458 else { 1459 thflags &= ~TH_URG; 1460 th->th_urp = 0; 1461 } 1462 } 1463 1464 /* 1465 * If new data are received on a connection after the 1466 * user processes are gone, then RST the other end. 1467 */ 1468 if ((so->so_state & SS_NOFDREF) && 1469 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1470 tp = tcp_close(tp); 1471 tcpstat.tcps_rcvafterclose++; 1472 rstreason = BANDLIM_UNLIMITED; 1473 goto dropwithreset; 1474 } 1475 1476 /* 1477 * If segment ends after window, drop trailing data 1478 * (and PUSH and FIN); if nothing left, just ACK. 1479 */ 1480 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1481 if (todrop > 0) { 1482 tcpstat.tcps_rcvpackafterwin++; 1483 if (todrop >= tlen) { 1484 tcpstat.tcps_rcvbyteafterwin += tlen; 1485 /* 1486 * If a new connection request is received 1487 * while in TIME_WAIT, drop the old connection 1488 * and start over if the sequence numbers 1489 * are above the previous ones. 1490 */ 1491 if (thflags & TH_SYN && 1492 tp->t_state == TCPS_TIME_WAIT && 1493 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1494 iss = tcp_new_isn(tp); 1495 tp = tcp_close(tp); 1496 goto findpcb; 1497 } 1498 /* 1499 * If window is closed can only take segments at 1500 * window edge, and have to drop data and PUSH from 1501 * incoming segments. Continue processing, but 1502 * remember to ack. Otherwise, drop segment 1503 * and ack. 1504 */ 1505 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1506 tp->t_flags |= TF_ACKNOW; 1507 tcpstat.tcps_rcvwinprobe++; 1508 } else 1509 goto dropafterack; 1510 } else 1511 tcpstat.tcps_rcvbyteafterwin += todrop; 1512 m_adj(m, -todrop); 1513 tlen -= todrop; 1514 thflags &= ~(TH_PUSH|TH_FIN); 1515 } 1516 1517 /* 1518 * If last ACK falls within this segment's sequence numbers, 1519 * record its timestamp. 1520 * NOTE that the test is modified according to the latest 1521 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1522 */ 1523 if ((to.to_flags & TOF_TS) != 0 && 1524 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1525 tp->ts_recent_age = ticks; 1526 tp->ts_recent = to.to_tsval; 1527 } 1528 1529 /* 1530 * If a SYN is in the window, then this is an 1531 * error and we send an RST and drop the connection. 1532 */ 1533 if (thflags & TH_SYN) { 1534 tp = tcp_drop(tp, ECONNRESET); 1535 rstreason = BANDLIM_UNLIMITED; 1536 goto dropwithreset; 1537 } 1538 1539 /* 1540 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1541 * flag is on (half-synchronized state), then queue data for 1542 * later processing; else drop segment and return. 1543 */ 1544 if ((thflags & TH_ACK) == 0) { 1545 if (tp->t_state == TCPS_SYN_RECEIVED || 1546 (tp->t_flags & TF_NEEDSYN)) 1547 goto step6; 1548 else 1549 goto drop; 1550 } 1551 1552 /* 1553 * Ack processing. 1554 */ 1555 switch (tp->t_state) { 1556 1557 /* 1558 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1559 * ESTABLISHED state and continue processing. 1560 * The ACK was checked above. 1561 */ 1562 case TCPS_SYN_RECEIVED: 1563 1564 tcpstat.tcps_connects++; 1565 soisconnected(so); 1566 /* Do window scaling? */ 1567 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1568 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1569 tp->snd_scale = tp->requested_s_scale; 1570 tp->rcv_scale = tp->request_r_scale; 1571 } 1572 /* 1573 * Upon successful completion of 3-way handshake, 1574 * update cache.CC if it was undefined, pass any queued 1575 * data to the user, and advance state appropriately. 1576 */ 1577 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1578 taop->tao_cc == 0) 1579 taop->tao_cc = tp->cc_recv; 1580 1581 /* 1582 * Make transitions: 1583 * SYN-RECEIVED -> ESTABLISHED 1584 * SYN-RECEIVED* -> FIN-WAIT-1 1585 */ 1586 tp->t_starttime = ticks; 1587 if (tp->t_flags & TF_NEEDFIN) { 1588 tp->t_state = TCPS_FIN_WAIT_1; 1589 tp->t_flags &= ~TF_NEEDFIN; 1590 } else { 1591 tp->t_state = TCPS_ESTABLISHED; 1592 callout_reset(tp->tt_keep, tcp_keepidle, 1593 tcp_timer_keep, tp); 1594 } 1595 /* 1596 * If segment contains data or ACK, will call tcp_reass() 1597 * later; if not, do so now to pass queued data to user. 1598 */ 1599 if (tlen == 0 && (thflags & TH_FIN) == 0) 1600 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1601 (struct mbuf *)0); 1602 tp->snd_wl1 = th->th_seq - 1; 1603 /* fall into ... */ 1604 1605 /* 1606 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1607 * ACKs. If the ack is in the range 1608 * tp->snd_una < th->th_ack <= tp->snd_max 1609 * then advance tp->snd_una to th->th_ack and drop 1610 * data from the retransmission queue. If this ACK reflects 1611 * more up to date window information we update our window information. 1612 */ 1613 case TCPS_ESTABLISHED: 1614 case TCPS_FIN_WAIT_1: 1615 case TCPS_FIN_WAIT_2: 1616 case TCPS_CLOSE_WAIT: 1617 case TCPS_CLOSING: 1618 case TCPS_LAST_ACK: 1619 case TCPS_TIME_WAIT: 1620 1621 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1622 if (tlen == 0 && tiwin == tp->snd_wnd) { 1623 tcpstat.tcps_rcvdupack++; 1624 /* 1625 * If we have outstanding data (other than 1626 * a window probe), this is a completely 1627 * duplicate ack (ie, window info didn't 1628 * change), the ack is the biggest we've 1629 * seen and we've seen exactly our rexmt 1630 * threshhold of them, assume a packet 1631 * has been dropped and retransmit it. 1632 * Kludge snd_nxt & the congestion 1633 * window so we send only this one 1634 * packet. 1635 * 1636 * We know we're losing at the current 1637 * window size so do congestion avoidance 1638 * (set ssthresh to half the current window 1639 * and pull our congestion window back to 1640 * the new ssthresh). 1641 * 1642 * Dup acks mean that packets have left the 1643 * network (they're now cached at the receiver) 1644 * so bump cwnd by the amount in the receiver 1645 * to keep a constant cwnd packets in the 1646 * network. 1647 */ 1648 if (!callout_active(tp->tt_rexmt) || 1649 th->th_ack != tp->snd_una) 1650 tp->t_dupacks = 0; 1651 else if (++tp->t_dupacks == tcprexmtthresh) { 1652 tcp_seq onxt = tp->snd_nxt; 1653 u_int win = 1654 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1655 tp->t_maxseg; 1656 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1657 tp->snd_recover)) { 1658 /* False retransmit, should not 1659 * cut window 1660 */ 1661 tp->snd_cwnd += tp->t_maxseg; 1662 tp->t_dupacks = 0; 1663 (void) tcp_output(tp); 1664 goto drop; 1665 } 1666 if (win < 2) 1667 win = 2; 1668 tp->snd_ssthresh = win * tp->t_maxseg; 1669 tp->snd_recover = tp->snd_max; 1670 callout_stop(tp->tt_rexmt); 1671 tp->t_rtttime = 0; 1672 tp->snd_nxt = th->th_ack; 1673 tp->snd_cwnd = tp->t_maxseg; 1674 (void) tcp_output(tp); 1675 tp->snd_cwnd = tp->snd_ssthresh + 1676 tp->t_maxseg * tp->t_dupacks; 1677 if (SEQ_GT(onxt, tp->snd_nxt)) 1678 tp->snd_nxt = onxt; 1679 goto drop; 1680 } else if (tp->t_dupacks > tcprexmtthresh) { 1681 tp->snd_cwnd += tp->t_maxseg; 1682 (void) tcp_output(tp); 1683 goto drop; 1684 } 1685 } else 1686 tp->t_dupacks = 0; 1687 break; 1688 } 1689 /* 1690 * If the congestion window was inflated to account 1691 * for the other side's cached packets, retract it. 1692 */ 1693 if (tcp_do_newreno == 0) { 1694 if (tp->t_dupacks >= tcprexmtthresh && 1695 tp->snd_cwnd > tp->snd_ssthresh) 1696 tp->snd_cwnd = tp->snd_ssthresh; 1697 tp->t_dupacks = 0; 1698 } else if (tp->t_dupacks >= tcprexmtthresh && 1699 !tcp_newreno(tp, th)) { 1700 /* 1701 * Window inflation should have left us with approx. 1702 * snd_ssthresh outstanding data. But in case we 1703 * would be inclined to send a burst, better to do 1704 * it via the slow start mechanism. 1705 */ 1706 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1707 tp->snd_cwnd = 1708 tp->snd_max - th->th_ack + tp->t_maxseg; 1709 else 1710 tp->snd_cwnd = tp->snd_ssthresh; 1711 tp->t_dupacks = 0; 1712 } 1713 if (tp->t_dupacks < tcprexmtthresh) 1714 tp->t_dupacks = 0; 1715 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1716 tcpstat.tcps_rcvacktoomuch++; 1717 goto dropafterack; 1718 } 1719 /* 1720 * If we reach this point, ACK is not a duplicate, 1721 * i.e., it ACKs something we sent. 1722 */ 1723 if (tp->t_flags & TF_NEEDSYN) { 1724 /* 1725 * T/TCP: Connection was half-synchronized, and our 1726 * SYN has been ACK'd (so connection is now fully 1727 * synchronized). Go to non-starred state, 1728 * increment snd_una for ACK of SYN, and check if 1729 * we can do window scaling. 1730 */ 1731 tp->t_flags &= ~TF_NEEDSYN; 1732 tp->snd_una++; 1733 /* Do window scaling? */ 1734 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1735 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1736 tp->snd_scale = tp->requested_s_scale; 1737 tp->rcv_scale = tp->request_r_scale; 1738 } 1739 } 1740 1741 process_ACK: 1742 acked = th->th_ack - tp->snd_una; 1743 tcpstat.tcps_rcvackpack++; 1744 tcpstat.tcps_rcvackbyte += acked; 1745 1746 /* 1747 * If we just performed our first retransmit, and the ACK 1748 * arrives within our recovery window, then it was a mistake 1749 * to do the retransmit in the first place. Recover our 1750 * original cwnd and ssthresh, and proceed to transmit where 1751 * we left off. 1752 */ 1753 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1754 tp->snd_cwnd = tp->snd_cwnd_prev; 1755 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1756 tp->snd_nxt = tp->snd_max; 1757 tp->t_badrxtwin = 0; /* XXX probably not required */ 1758 } 1759 1760 /* 1761 * If we have a timestamp reply, update smoothed 1762 * round trip time. If no timestamp is present but 1763 * transmit timer is running and timed sequence 1764 * number was acked, update smoothed round trip time. 1765 * Since we now have an rtt measurement, cancel the 1766 * timer backoff (cf., Phil Karn's retransmit alg.). 1767 * Recompute the initial retransmit timer. 1768 */ 1769 if (to.to_flags & TOF_TS) 1770 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1771 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1772 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1773 1774 /* 1775 * If all outstanding data is acked, stop retransmit 1776 * timer and remember to restart (more output or persist). 1777 * If there is more data to be acked, restart retransmit 1778 * timer, using current (possibly backed-off) value. 1779 */ 1780 if (th->th_ack == tp->snd_max) { 1781 callout_stop(tp->tt_rexmt); 1782 needoutput = 1; 1783 } else if (!callout_active(tp->tt_persist)) 1784 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1785 tcp_timer_rexmt, tp); 1786 1787 /* 1788 * If no data (only SYN) was ACK'd, 1789 * skip rest of ACK processing. 1790 */ 1791 if (acked == 0) 1792 goto step6; 1793 1794 /* 1795 * When new data is acked, open the congestion window. 1796 * If the window gives us less than ssthresh packets 1797 * in flight, open exponentially (maxseg per packet). 1798 * Otherwise open linearly: maxseg per window 1799 * (maxseg^2 / cwnd per packet). 1800 */ 1801 { 1802 register u_int cw = tp->snd_cwnd; 1803 register u_int incr = tp->t_maxseg; 1804 1805 if (cw > tp->snd_ssthresh) 1806 incr = incr * incr / cw; 1807 /* 1808 * If t_dupacks != 0 here, it indicates that we are still 1809 * in NewReno fast recovery mode, so we leave the congestion 1810 * window alone. 1811 */ 1812 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1813 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1814 } 1815 if (acked > so->so_snd.sb_cc) { 1816 tp->snd_wnd -= so->so_snd.sb_cc; 1817 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1818 ourfinisacked = 1; 1819 } else { 1820 sbdrop(&so->so_snd, acked); 1821 tp->snd_wnd -= acked; 1822 ourfinisacked = 0; 1823 } 1824 sowwakeup(so); 1825 tp->snd_una = th->th_ack; 1826 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1827 tp->snd_nxt = tp->snd_una; 1828 1829 switch (tp->t_state) { 1830 1831 /* 1832 * In FIN_WAIT_1 STATE in addition to the processing 1833 * for the ESTABLISHED state if our FIN is now acknowledged 1834 * then enter FIN_WAIT_2. 1835 */ 1836 case TCPS_FIN_WAIT_1: 1837 if (ourfinisacked) { 1838 /* 1839 * If we can't receive any more 1840 * data, then closing user can proceed. 1841 * Starting the timer is contrary to the 1842 * specification, but if we don't get a FIN 1843 * we'll hang forever. 1844 */ 1845 if (so->so_state & SS_CANTRCVMORE) { 1846 soisdisconnected(so); 1847 callout_reset(tp->tt_2msl, tcp_maxidle, 1848 tcp_timer_2msl, tp); 1849 } 1850 tp->t_state = TCPS_FIN_WAIT_2; 1851 } 1852 break; 1853 1854 /* 1855 * In CLOSING STATE in addition to the processing for 1856 * the ESTABLISHED state if the ACK acknowledges our FIN 1857 * then enter the TIME-WAIT state, otherwise ignore 1858 * the segment. 1859 */ 1860 case TCPS_CLOSING: 1861 if (ourfinisacked) { 1862 tp->t_state = TCPS_TIME_WAIT; 1863 tcp_canceltimers(tp); 1864 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1865 if (tp->cc_recv != 0 && 1866 (ticks - tp->t_starttime) < tcp_msl) 1867 callout_reset(tp->tt_2msl, 1868 tp->t_rxtcur * 1869 TCPTV_TWTRUNC, 1870 tcp_timer_2msl, tp); 1871 else 1872 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1873 tcp_timer_2msl, tp); 1874 soisdisconnected(so); 1875 } 1876 break; 1877 1878 /* 1879 * In LAST_ACK, we may still be waiting for data to drain 1880 * and/or to be acked, as well as for the ack of our FIN. 1881 * If our FIN is now acknowledged, delete the TCB, 1882 * enter the closed state and return. 1883 */ 1884 case TCPS_LAST_ACK: 1885 if (ourfinisacked) { 1886 tp = tcp_close(tp); 1887 goto drop; 1888 } 1889 break; 1890 1891 /* 1892 * In TIME_WAIT state the only thing that should arrive 1893 * is a retransmission of the remote FIN. Acknowledge 1894 * it and restart the finack timer. 1895 */ 1896 case TCPS_TIME_WAIT: 1897 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1898 tcp_timer_2msl, tp); 1899 goto dropafterack; 1900 } 1901 } 1902 1903 step6: 1904 /* 1905 * Update window information. 1906 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1907 */ 1908 if ((thflags & TH_ACK) && 1909 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1910 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1911 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1912 /* keep track of pure window updates */ 1913 if (tlen == 0 && 1914 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1915 tcpstat.tcps_rcvwinupd++; 1916 tp->snd_wnd = tiwin; 1917 tp->snd_wl1 = th->th_seq; 1918 tp->snd_wl2 = th->th_ack; 1919 if (tp->snd_wnd > tp->max_sndwnd) 1920 tp->max_sndwnd = tp->snd_wnd; 1921 needoutput = 1; 1922 } 1923 1924 /* 1925 * Process segments with URG. 1926 */ 1927 if ((thflags & TH_URG) && th->th_urp && 1928 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1929 /* 1930 * This is a kludge, but if we receive and accept 1931 * random urgent pointers, we'll crash in 1932 * soreceive. It's hard to imagine someone 1933 * actually wanting to send this much urgent data. 1934 */ 1935 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1936 th->th_urp = 0; /* XXX */ 1937 thflags &= ~TH_URG; /* XXX */ 1938 goto dodata; /* XXX */ 1939 } 1940 /* 1941 * If this segment advances the known urgent pointer, 1942 * then mark the data stream. This should not happen 1943 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1944 * a FIN has been received from the remote side. 1945 * In these states we ignore the URG. 1946 * 1947 * According to RFC961 (Assigned Protocols), 1948 * the urgent pointer points to the last octet 1949 * of urgent data. We continue, however, 1950 * to consider it to indicate the first octet 1951 * of data past the urgent section as the original 1952 * spec states (in one of two places). 1953 */ 1954 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1955 tp->rcv_up = th->th_seq + th->th_urp; 1956 so->so_oobmark = so->so_rcv.sb_cc + 1957 (tp->rcv_up - tp->rcv_nxt) - 1; 1958 if (so->so_oobmark == 0) 1959 so->so_state |= SS_RCVATMARK; 1960 sohasoutofband(so); 1961 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1962 } 1963 /* 1964 * Remove out of band data so doesn't get presented to user. 1965 * This can happen independent of advancing the URG pointer, 1966 * but if two URG's are pending at once, some out-of-band 1967 * data may creep in... ick. 1968 */ 1969 if (th->th_urp <= (u_long)tlen 1970 #ifdef SO_OOBINLINE 1971 && (so->so_options & SO_OOBINLINE) == 0 1972 #endif 1973 ) 1974 tcp_pulloutofband(so, th, m, 1975 drop_hdrlen); /* hdr drop is delayed */ 1976 } else 1977 /* 1978 * If no out of band data is expected, 1979 * pull receive urgent pointer along 1980 * with the receive window. 1981 */ 1982 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1983 tp->rcv_up = tp->rcv_nxt; 1984 dodata: /* XXX */ 1985 1986 /* 1987 * Process the segment text, merging it into the TCP sequencing queue, 1988 * and arranging for acknowledgment of receipt if necessary. 1989 * This process logically involves adjusting tp->rcv_wnd as data 1990 * is presented to the user (this happens in tcp_usrreq.c, 1991 * case PRU_RCVD). If a FIN has already been received on this 1992 * connection then we just ignore the text. 1993 */ 1994 if ((tlen || (thflags&TH_FIN)) && 1995 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1996 m_adj(m, drop_hdrlen); /* delayed header drop */ 1997 /* 1998 * Insert segment which inludes th into reassembly queue of tcp with 1999 * control block tp. Return TH_FIN if reassembly now includes 2000 * a segment with FIN. This handle the common case inline (segment 2001 * is the next to be received on an established connection, and the 2002 * queue is empty), avoiding linkage into and removal from the queue 2003 * and repetition of various conversions. 2004 * Set DELACK for segments received in order, but ack immediately 2005 * when segments are out of order (so fast retransmit can work). 2006 */ 2007 if (th->th_seq == tp->rcv_nxt && 2008 LIST_EMPTY(&tp->t_segq) && 2009 TCPS_HAVEESTABLISHED(tp->t_state)) { 2010 if (DELAY_ACK(tp)) 2011 callout_reset(tp->tt_delack, tcp_delacktime, 2012 tcp_timer_delack, tp); 2013 else 2014 tp->t_flags |= TF_ACKNOW; 2015 tp->rcv_nxt += tlen; 2016 thflags = th->th_flags & TH_FIN; 2017 tcpstat.tcps_rcvpack++; 2018 tcpstat.tcps_rcvbyte += tlen; 2019 ND6_HINT(tp); 2020 sbappend(&so->so_rcv, m); 2021 sorwakeup(so); 2022 } else { 2023 thflags = tcp_reass(tp, th, &tlen, m); 2024 tp->t_flags |= TF_ACKNOW; 2025 } 2026 2027 /* 2028 * Note the amount of data that peer has sent into 2029 * our window, in order to estimate the sender's 2030 * buffer size. 2031 */ 2032 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2033 } else { 2034 m_freem(m); 2035 thflags &= ~TH_FIN; 2036 } 2037 2038 /* 2039 * If FIN is received ACK the FIN and let the user know 2040 * that the connection is closing. 2041 */ 2042 if (thflags & TH_FIN) { 2043 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2044 socantrcvmore(so); 2045 /* 2046 * If connection is half-synchronized 2047 * (ie NEEDSYN flag on) then delay ACK, 2048 * so it may be piggybacked when SYN is sent. 2049 * Otherwise, since we received a FIN then no 2050 * more input can be expected, send ACK now. 2051 */ 2052 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2053 callout_reset(tp->tt_delack, tcp_delacktime, 2054 tcp_timer_delack, tp); 2055 else 2056 tp->t_flags |= TF_ACKNOW; 2057 tp->rcv_nxt++; 2058 } 2059 switch (tp->t_state) { 2060 2061 /* 2062 * In SYN_RECEIVED and ESTABLISHED STATES 2063 * enter the CLOSE_WAIT state. 2064 */ 2065 case TCPS_SYN_RECEIVED: 2066 tp->t_starttime = ticks; 2067 /*FALLTHROUGH*/ 2068 case TCPS_ESTABLISHED: 2069 tp->t_state = TCPS_CLOSE_WAIT; 2070 break; 2071 2072 /* 2073 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2074 * enter the CLOSING state. 2075 */ 2076 case TCPS_FIN_WAIT_1: 2077 tp->t_state = TCPS_CLOSING; 2078 break; 2079 2080 /* 2081 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2082 * starting the time-wait timer, turning off the other 2083 * standard timers. 2084 */ 2085 case TCPS_FIN_WAIT_2: 2086 tp->t_state = TCPS_TIME_WAIT; 2087 tcp_canceltimers(tp); 2088 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2089 if (tp->cc_recv != 0 && 2090 (ticks - tp->t_starttime) < tcp_msl) { 2091 callout_reset(tp->tt_2msl, 2092 tp->t_rxtcur * TCPTV_TWTRUNC, 2093 tcp_timer_2msl, tp); 2094 /* For transaction client, force ACK now. */ 2095 tp->t_flags |= TF_ACKNOW; 2096 } 2097 else 2098 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2099 tcp_timer_2msl, tp); 2100 soisdisconnected(so); 2101 break; 2102 2103 /* 2104 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2105 */ 2106 case TCPS_TIME_WAIT: 2107 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2108 tcp_timer_2msl, tp); 2109 break; 2110 } 2111 } 2112 #ifdef TCPDEBUG 2113 if (so->so_options & SO_DEBUG) 2114 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2115 &tcp_savetcp, 0); 2116 #endif 2117 2118 /* 2119 * Return any desired output. 2120 */ 2121 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2122 (void) tcp_output(tp); 2123 return; 2124 2125 dropafterack: 2126 /* 2127 * Generate an ACK dropping incoming segment if it occupies 2128 * sequence space, where the ACK reflects our state. 2129 * 2130 * We can now skip the test for the RST flag since all 2131 * paths to this code happen after packets containing 2132 * RST have been dropped. 2133 * 2134 * In the SYN-RECEIVED state, don't send an ACK unless the 2135 * segment we received passes the SYN-RECEIVED ACK test. 2136 * If it fails send a RST. This breaks the loop in the 2137 * "LAND" DoS attack, and also prevents an ACK storm 2138 * between two listening ports that have been sent forged 2139 * SYN segments, each with the source address of the other. 2140 */ 2141 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2142 (SEQ_GT(tp->snd_una, th->th_ack) || 2143 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2144 rstreason = BANDLIM_RST_OPENPORT; 2145 goto dropwithreset; 2146 } 2147 #ifdef TCPDEBUG 2148 if (so->so_options & SO_DEBUG) 2149 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2150 &tcp_savetcp, 0); 2151 #endif 2152 m_freem(m); 2153 tp->t_flags |= TF_ACKNOW; 2154 (void) tcp_output(tp); 2155 return; 2156 2157 dropwithreset: 2158 /* 2159 * Generate a RST, dropping incoming segment. 2160 * Make ACK acceptable to originator of segment. 2161 * Don't bother to respond if destination was broadcast/multicast. 2162 */ 2163 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2164 goto drop; 2165 #ifdef INET6 2166 if (isipv6) { 2167 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2168 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2169 goto drop; 2170 } else 2171 #endif /* INET6 */ 2172 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2173 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2174 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 2175 goto drop; 2176 /* IPv6 anycast check is done at tcp6_input() */ 2177 2178 /* 2179 * Perform bandwidth limiting. 2180 */ 2181 if (badport_bandlim(rstreason) < 0) 2182 goto drop; 2183 2184 #ifdef TCPDEBUG 2185 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2186 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2187 &tcp_savetcp, 0); 2188 #endif 2189 if (thflags & TH_ACK) 2190 /* mtod() below is safe as long as hdr dropping is delayed */ 2191 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2192 TH_RST); 2193 else { 2194 if (thflags & TH_SYN) 2195 tlen++; 2196 /* mtod() below is safe as long as hdr dropping is delayed */ 2197 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2198 (tcp_seq)0, TH_RST|TH_ACK); 2199 } 2200 return; 2201 2202 drop: 2203 /* 2204 * Drop space held by incoming segment and return. 2205 */ 2206 #ifdef TCPDEBUG 2207 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2208 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2209 &tcp_savetcp, 0); 2210 #endif 2211 m_freem(m); 2212 return; 2213 } 2214 2215 /* 2216 * Parse TCP options and place in tcpopt. 2217 */ 2218 static void 2219 tcp_dooptions(to, cp, cnt, is_syn) 2220 struct tcpopt *to; 2221 u_char *cp; 2222 int cnt; 2223 { 2224 int opt, optlen; 2225 2226 to->to_flags = 0; 2227 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2228 opt = cp[0]; 2229 if (opt == TCPOPT_EOL) 2230 break; 2231 if (opt == TCPOPT_NOP) 2232 optlen = 1; 2233 else { 2234 if (cnt < 2) 2235 break; 2236 optlen = cp[1]; 2237 if (optlen < 2 || optlen > cnt) 2238 break; 2239 } 2240 switch (opt) { 2241 case TCPOPT_MAXSEG: 2242 if (optlen != TCPOLEN_MAXSEG) 2243 continue; 2244 if (!is_syn) 2245 continue; 2246 to->to_flags |= TOF_MSS; 2247 bcopy((char *)cp + 2, 2248 (char *)&to->to_mss, sizeof(to->to_mss)); 2249 NTOHS(to->to_mss); 2250 break; 2251 case TCPOPT_WINDOW: 2252 if (optlen != TCPOLEN_WINDOW) 2253 continue; 2254 if (! is_syn) 2255 continue; 2256 to->to_flags |= TOF_SCALE; 2257 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2258 break; 2259 case TCPOPT_TIMESTAMP: 2260 if (optlen != TCPOLEN_TIMESTAMP) 2261 continue; 2262 to->to_flags |= TOF_TS; 2263 bcopy((char *)cp + 2, 2264 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2265 NTOHL(to->to_tsval); 2266 bcopy((char *)cp + 6, 2267 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2268 NTOHL(to->to_tsecr); 2269 break; 2270 case TCPOPT_CC: 2271 if (optlen != TCPOLEN_CC) 2272 continue; 2273 to->to_flags |= TOF_CC; 2274 bcopy((char *)cp + 2, 2275 (char *)&to->to_cc, sizeof(to->to_cc)); 2276 NTOHL(to->to_cc); 2277 break; 2278 case TCPOPT_CCNEW: 2279 if (optlen != TCPOLEN_CC) 2280 continue; 2281 if (!is_syn) 2282 continue; 2283 to->to_flags |= TOF_CCNEW; 2284 bcopy((char *)cp + 2, 2285 (char *)&to->to_cc, sizeof(to->to_cc)); 2286 NTOHL(to->to_cc); 2287 break; 2288 case TCPOPT_CCECHO: 2289 if (optlen != TCPOLEN_CC) 2290 continue; 2291 if (!is_syn) 2292 continue; 2293 to->to_flags |= TOF_CCECHO; 2294 bcopy((char *)cp + 2, 2295 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2296 NTOHL(to->to_ccecho); 2297 break; 2298 default: 2299 continue; 2300 } 2301 } 2302 } 2303 2304 /* 2305 * Pull out of band byte out of a segment so 2306 * it doesn't appear in the user's data queue. 2307 * It is still reflected in the segment length for 2308 * sequencing purposes. 2309 */ 2310 static void 2311 tcp_pulloutofband(so, th, m, off) 2312 struct socket *so; 2313 struct tcphdr *th; 2314 register struct mbuf *m; 2315 int off; /* delayed to be droped hdrlen */ 2316 { 2317 int cnt = off + th->th_urp - 1; 2318 2319 while (cnt >= 0) { 2320 if (m->m_len > cnt) { 2321 char *cp = mtod(m, caddr_t) + cnt; 2322 struct tcpcb *tp = sototcpcb(so); 2323 2324 tp->t_iobc = *cp; 2325 tp->t_oobflags |= TCPOOB_HAVEDATA; 2326 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2327 m->m_len--; 2328 if (m->m_flags & M_PKTHDR) 2329 m->m_pkthdr.len--; 2330 return; 2331 } 2332 cnt -= m->m_len; 2333 m = m->m_next; 2334 if (m == 0) 2335 break; 2336 } 2337 panic("tcp_pulloutofband"); 2338 } 2339 2340 /* 2341 * Collect new round-trip time estimate 2342 * and update averages and current timeout. 2343 */ 2344 static void 2345 tcp_xmit_timer(tp, rtt) 2346 register struct tcpcb *tp; 2347 int rtt; 2348 { 2349 register int delta; 2350 2351 tcpstat.tcps_rttupdated++; 2352 tp->t_rttupdated++; 2353 if (tp->t_srtt != 0) { 2354 /* 2355 * srtt is stored as fixed point with 5 bits after the 2356 * binary point (i.e., scaled by 8). The following magic 2357 * is equivalent to the smoothing algorithm in rfc793 with 2358 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2359 * point). Adjust rtt to origin 0. 2360 */ 2361 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2362 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2363 2364 if ((tp->t_srtt += delta) <= 0) 2365 tp->t_srtt = 1; 2366 2367 /* 2368 * We accumulate a smoothed rtt variance (actually, a 2369 * smoothed mean difference), then set the retransmit 2370 * timer to smoothed rtt + 4 times the smoothed variance. 2371 * rttvar is stored as fixed point with 4 bits after the 2372 * binary point (scaled by 16). The following is 2373 * equivalent to rfc793 smoothing with an alpha of .75 2374 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2375 * rfc793's wired-in beta. 2376 */ 2377 if (delta < 0) 2378 delta = -delta; 2379 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2380 if ((tp->t_rttvar += delta) <= 0) 2381 tp->t_rttvar = 1; 2382 } else { 2383 /* 2384 * No rtt measurement yet - use the unsmoothed rtt. 2385 * Set the variance to half the rtt (so our first 2386 * retransmit happens at 3*rtt). 2387 */ 2388 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2389 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2390 } 2391 tp->t_rtttime = 0; 2392 tp->t_rxtshift = 0; 2393 2394 /* 2395 * the retransmit should happen at rtt + 4 * rttvar. 2396 * Because of the way we do the smoothing, srtt and rttvar 2397 * will each average +1/2 tick of bias. When we compute 2398 * the retransmit timer, we want 1/2 tick of rounding and 2399 * 1 extra tick because of +-1/2 tick uncertainty in the 2400 * firing of the timer. The bias will give us exactly the 2401 * 1.5 tick we need. But, because the bias is 2402 * statistical, we have to test that we don't drop below 2403 * the minimum feasible timer (which is 2 ticks). 2404 */ 2405 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2406 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2407 2408 /* 2409 * We received an ack for a packet that wasn't retransmitted; 2410 * it is probably safe to discard any error indications we've 2411 * received recently. This isn't quite right, but close enough 2412 * for now (a route might have failed after we sent a segment, 2413 * and the return path might not be symmetrical). 2414 */ 2415 tp->t_softerror = 0; 2416 } 2417 2418 /* 2419 * Determine a reasonable value for maxseg size. 2420 * If the route is known, check route for mtu. 2421 * If none, use an mss that can be handled on the outgoing 2422 * interface without forcing IP to fragment; if bigger than 2423 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2424 * to utilize large mbufs. If no route is found, route has no mtu, 2425 * or the destination isn't local, use a default, hopefully conservative 2426 * size (usually 512 or the default IP max size, but no more than the mtu 2427 * of the interface), as we can't discover anything about intervening 2428 * gateways or networks. We also initialize the congestion/slow start 2429 * window to be a single segment if the destination isn't local. 2430 * While looking at the routing entry, we also initialize other path-dependent 2431 * parameters from pre-set or cached values in the routing entry. 2432 * 2433 * Also take into account the space needed for options that we 2434 * send regularly. Make maxseg shorter by that amount to assure 2435 * that we can send maxseg amount of data even when the options 2436 * are present. Store the upper limit of the length of options plus 2437 * data in maxopd. 2438 * 2439 * NOTE that this routine is only called when we process an incoming 2440 * segment, for outgoing segments only tcp_mssopt is called. 2441 * 2442 * In case of T/TCP, we call this routine during implicit connection 2443 * setup as well (offer = -1), to initialize maxseg from the cached 2444 * MSS of our peer. 2445 */ 2446 void 2447 tcp_mss(tp, offer) 2448 struct tcpcb *tp; 2449 int offer; 2450 { 2451 register struct rtentry *rt; 2452 struct ifnet *ifp; 2453 register int rtt, mss; 2454 u_long bufsize; 2455 struct inpcb *inp; 2456 struct socket *so; 2457 struct rmxp_tao *taop; 2458 int origoffer = offer; 2459 #ifdef INET6 2460 int isipv6; 2461 int min_protoh; 2462 #endif 2463 2464 inp = tp->t_inpcb; 2465 #ifdef INET6 2466 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2467 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2468 : sizeof (struct tcpiphdr); 2469 #else 2470 #define min_protoh (sizeof (struct tcpiphdr)) 2471 #endif 2472 #ifdef INET6 2473 if (isipv6) 2474 rt = tcp_rtlookup6(&inp->inp_inc); 2475 else 2476 #endif 2477 rt = tcp_rtlookup(&inp->inp_inc); 2478 if (rt == NULL) { 2479 tp->t_maxopd = tp->t_maxseg = 2480 #ifdef INET6 2481 isipv6 ? tcp_v6mssdflt : 2482 #endif /* INET6 */ 2483 tcp_mssdflt; 2484 return; 2485 } 2486 ifp = rt->rt_ifp; 2487 so = inp->inp_socket; 2488 2489 taop = rmx_taop(rt->rt_rmx); 2490 /* 2491 * Offer == -1 means that we didn't receive SYN yet, 2492 * use cached value in that case; 2493 */ 2494 if (offer == -1) 2495 offer = taop->tao_mssopt; 2496 /* 2497 * Offer == 0 means that there was no MSS on the SYN segment, 2498 * in this case we use tcp_mssdflt. 2499 */ 2500 if (offer == 0) 2501 offer = 2502 #ifdef INET6 2503 isipv6 ? tcp_v6mssdflt : 2504 #endif /* INET6 */ 2505 tcp_mssdflt; 2506 else 2507 /* 2508 * Sanity check: make sure that maxopd will be large 2509 * enough to allow some data on segments even is the 2510 * all the option space is used (40bytes). Otherwise 2511 * funny things may happen in tcp_output. 2512 */ 2513 offer = max(offer, 64); 2514 taop->tao_mssopt = offer; 2515 2516 /* 2517 * While we're here, check if there's an initial rtt 2518 * or rttvar. Convert from the route-table units 2519 * to scaled multiples of the slow timeout timer. 2520 */ 2521 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2522 /* 2523 * XXX the lock bit for RTT indicates that the value 2524 * is also a minimum value; this is subject to time. 2525 */ 2526 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2527 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2528 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2529 tcpstat.tcps_usedrtt++; 2530 if (rt->rt_rmx.rmx_rttvar) { 2531 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2532 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2533 tcpstat.tcps_usedrttvar++; 2534 } else { 2535 /* default variation is +- 1 rtt */ 2536 tp->t_rttvar = 2537 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2538 } 2539 TCPT_RANGESET(tp->t_rxtcur, 2540 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2541 tp->t_rttmin, TCPTV_REXMTMAX); 2542 } 2543 /* 2544 * if there's an mtu associated with the route, use it 2545 * else, use the link mtu. 2546 */ 2547 if (rt->rt_rmx.rmx_mtu) 2548 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2549 else 2550 { 2551 mss = 2552 #ifdef INET6 2553 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2554 #endif 2555 ifp->if_mtu 2556 #ifdef INET6 2557 ) 2558 #endif 2559 - min_protoh; 2560 #ifdef INET6 2561 if (isipv6) { 2562 if (!in6_localaddr(&inp->in6p_faddr)) 2563 mss = min(mss, tcp_v6mssdflt); 2564 } else 2565 #endif 2566 if (!in_localaddr(inp->inp_faddr)) 2567 mss = min(mss, tcp_mssdflt); 2568 } 2569 mss = min(mss, offer); 2570 /* 2571 * maxopd stores the maximum length of data AND options 2572 * in a segment; maxseg is the amount of data in a normal 2573 * segment. We need to store this value (maxopd) apart 2574 * from maxseg, because now every segment carries options 2575 * and thus we normally have somewhat less data in segments. 2576 */ 2577 tp->t_maxopd = mss; 2578 2579 /* 2580 * In case of T/TCP, origoffer==-1 indicates, that no segments 2581 * were received yet. In this case we just guess, otherwise 2582 * we do the same as before T/TCP. 2583 */ 2584 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2585 (origoffer == -1 || 2586 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2587 mss -= TCPOLEN_TSTAMP_APPA; 2588 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2589 (origoffer == -1 || 2590 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2591 mss -= TCPOLEN_CC_APPA; 2592 2593 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2594 if (mss > MCLBYTES) 2595 mss &= ~(MCLBYTES-1); 2596 #else 2597 if (mss > MCLBYTES) 2598 mss = mss / MCLBYTES * MCLBYTES; 2599 #endif 2600 /* 2601 * If there's a pipesize, change the socket buffer 2602 * to that size. Make the socket buffers an integral 2603 * number of mss units; if the mss is larger than 2604 * the socket buffer, decrease the mss. 2605 */ 2606 #ifdef RTV_SPIPE 2607 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2608 #endif 2609 bufsize = so->so_snd.sb_hiwat; 2610 if (bufsize < mss) 2611 mss = bufsize; 2612 else { 2613 bufsize = roundup(bufsize, mss); 2614 if (bufsize > sb_max) 2615 bufsize = sb_max; 2616 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2617 } 2618 tp->t_maxseg = mss; 2619 2620 #ifdef RTV_RPIPE 2621 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2622 #endif 2623 bufsize = so->so_rcv.sb_hiwat; 2624 if (bufsize > mss) { 2625 bufsize = roundup(bufsize, mss); 2626 if (bufsize > sb_max) 2627 bufsize = sb_max; 2628 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2629 } 2630 2631 /* 2632 * Set the slow-start flight size depending on whether this 2633 * is a local network or not. 2634 */ 2635 if ( 2636 #ifdef INET6 2637 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2638 (!isipv6 && 2639 #endif 2640 in_localaddr(inp->inp_faddr) 2641 #ifdef INET6 2642 ) 2643 #endif 2644 ) 2645 tp->snd_cwnd = mss * ss_fltsz_local; 2646 else 2647 tp->snd_cwnd = mss * ss_fltsz; 2648 2649 if (rt->rt_rmx.rmx_ssthresh) { 2650 /* 2651 * There's some sort of gateway or interface 2652 * buffer limit on the path. Use this to set 2653 * the slow start threshhold, but set the 2654 * threshold to no less than 2*mss. 2655 */ 2656 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2657 tcpstat.tcps_usedssthresh++; 2658 } 2659 } 2660 2661 /* 2662 * Determine the MSS option to send on an outgoing SYN. 2663 */ 2664 int 2665 tcp_mssopt(tp) 2666 struct tcpcb *tp; 2667 { 2668 struct rtentry *rt; 2669 #ifdef INET6 2670 int isipv6; 2671 int min_protoh; 2672 #endif 2673 2674 #ifdef INET6 2675 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2676 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2677 : sizeof (struct tcpiphdr); 2678 #else 2679 #define min_protoh (sizeof (struct tcpiphdr)) 2680 #endif 2681 #ifdef INET6 2682 if (isipv6) 2683 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2684 else 2685 #endif /* INET6 */ 2686 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2687 if (rt == NULL) 2688 return 2689 #ifdef INET6 2690 isipv6 ? tcp_v6mssdflt : 2691 #endif /* INET6 */ 2692 tcp_mssdflt; 2693 2694 return rt->rt_ifp->if_mtu - min_protoh; 2695 } 2696 2697 2698 /* 2699 * Checks for partial ack. If partial ack arrives, force the retransmission 2700 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2701 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2702 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2703 */ 2704 static int 2705 tcp_newreno(tp, th) 2706 struct tcpcb *tp; 2707 struct tcphdr *th; 2708 { 2709 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2710 tcp_seq onxt = tp->snd_nxt; 2711 u_long ocwnd = tp->snd_cwnd; 2712 2713 callout_stop(tp->tt_rexmt); 2714 tp->t_rtttime = 0; 2715 tp->snd_nxt = th->th_ack; 2716 /* 2717 * Set snd_cwnd to one segment beyond acknowledged offset 2718 * (tp->snd_una has not yet been updated when this function 2719 * is called) 2720 */ 2721 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2722 (void) tcp_output(tp); 2723 tp->snd_cwnd = ocwnd; 2724 if (SEQ_GT(onxt, tp->snd_nxt)) 2725 tp->snd_nxt = onxt; 2726 /* 2727 * Partial window deflation. Relies on fact that tp->snd_una 2728 * not updated yet. 2729 */ 2730 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2731 return (1); 2732 } 2733 return (0); 2734 } 2735