1 /* 2 * Copyright (c) 2002-2003 Jeffrey Hsu 3 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 4 * The Regents of the University of California. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. All advertising materials mentioning features or use of this software 15 * must display the following acknowledgement: 16 * This product includes software developed by the University of 17 * California, Berkeley and its contributors. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 35 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $ 36 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.11 2003/09/13 18:35:20 dillon Exp $ 37 */ 38 39 #include "opt_ipfw.h" /* for ipfw_fwd */ 40 #include "opt_inet6.h" 41 #include "opt_ipsec.h" 42 #include "opt_tcpdebug.h" 43 #include "opt_tcp_input.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/kernel.h> 48 #include <sys/sysctl.h> 49 #include <sys/malloc.h> 50 #include <sys/mbuf.h> 51 #include <sys/proc.h> /* for proc0 declaration */ 52 #include <sys/protosw.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/syslog.h> 56 57 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 58 59 #include <net/if.h> 60 #include <net/route.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_systm.h> 64 #include <netinet/ip.h> 65 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 66 #include <netinet/in_var.h> 67 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 68 #include <netinet/in_pcb.h> 69 #include <netinet/ip_var.h> 70 #include <netinet/ip6.h> 71 #include <netinet/icmp6.h> 72 #include <netinet6/nd6.h> 73 #include <netinet6/ip6_var.h> 74 #include <netinet6/in6_pcb.h> 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 #include <netinet6/tcp6_var.h> 81 #include <netinet/tcpip.h> 82 #ifdef TCPDEBUG 83 #include <netinet/tcp_debug.h> 84 85 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 86 struct tcphdr tcp_savetcp; 87 #endif /* TCPDEBUG */ 88 89 #ifdef FAST_IPSEC 90 #include <netipsec/ipsec.h> 91 #include <netipsec/ipsec6.h> 92 #endif 93 94 #ifdef IPSEC 95 #include <netinet6/ipsec.h> 96 #include <netinet6/ipsec6.h> 97 #include <netproto/key/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 const 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 static int tcp_do_limitedtransmit = 1; 131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW, 132 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)"); 133 134 static int tcp_do_rfc3390 = 1; 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 136 &tcp_do_rfc3390, 0, 137 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 138 139 static int tcp_do_eifel_detect = 1; 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 141 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 142 143 struct inpcbhead tcb; 144 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 145 struct inpcbinfo tcbinfo; 146 147 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 148 static void tcp_pulloutofband(struct socket *, 149 struct tcphdr *, struct mbuf *, int); 150 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 151 struct mbuf *); 152 static void tcp_xmit_timer(struct tcpcb *, int); 153 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 154 155 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 156 #ifdef INET6 157 #define ND6_HINT(tp) \ 158 do { \ 159 if ((tp) && (tp)->t_inpcb && \ 160 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 161 (tp)->t_inpcb->in6p_route.ro_rt) \ 162 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 163 } while (0) 164 #else 165 #define ND6_HINT(tp) 166 #endif 167 168 /* 169 * Indicate whether this ack should be delayed. We can delay the ack if 170 * - delayed acks are enabled and 171 * - there is no delayed ack timer in progress and 172 * - our last ack wasn't a 0-sized window. We never want to delay 173 * the ack that opens up a 0-sized window. 174 */ 175 #define DELAY_ACK(tp) \ 176 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 177 (tp->t_flags & TF_RXWIN0SENT) == 0) 178 179 static int 180 tcp_reass(tp, th, tlenp, m) 181 struct tcpcb *tp; 182 struct tcphdr *th; 183 int *tlenp; 184 struct mbuf *m; 185 { 186 struct tseg_qent *q; 187 struct tseg_qent *p = NULL; 188 struct tseg_qent *nq; 189 struct tseg_qent *te; 190 struct socket *so = tp->t_inpcb->inp_socket; 191 int flags; 192 193 /* 194 * Call with th==0 after become established to 195 * force pre-ESTABLISHED data up to user socket. 196 */ 197 if (th == 0) 198 goto present; 199 200 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 201 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ, 202 M_NOWAIT); 203 if (te == NULL) { 204 tcpstat.tcps_rcvmemdrop++; 205 m_freem(m); 206 return (0); 207 } 208 209 /* 210 * Find a segment which begins after this one does. 211 */ 212 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 213 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 214 break; 215 p = q; 216 } 217 218 /* 219 * If there is a preceding segment, it may provide some of 220 * our data already. If so, drop the data from the incoming 221 * segment. If it provides all of our data, drop us. 222 */ 223 if (p != NULL) { 224 int i; 225 /* conversion to int (in i) handles seq wraparound */ 226 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 227 if (i > 0) { 228 if (i >= *tlenp) { 229 tcpstat.tcps_rcvduppack++; 230 tcpstat.tcps_rcvdupbyte += *tlenp; 231 m_freem(m); 232 free(te, M_TSEGQ); 233 /* 234 * Try to present any queued data 235 * at the left window edge to the user. 236 * This is needed after the 3-WHS 237 * completes. 238 */ 239 goto present; /* ??? */ 240 } 241 m_adj(m, i); 242 *tlenp -= i; 243 th->th_seq += i; 244 } 245 } 246 tcpstat.tcps_rcvoopack++; 247 tcpstat.tcps_rcvoobyte += *tlenp; 248 249 /* 250 * While we overlap succeeding segments trim them or, 251 * if they are completely covered, dequeue them. 252 */ 253 while (q) { 254 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 255 if (i <= 0) 256 break; 257 if (i < q->tqe_len) { 258 q->tqe_th->th_seq += i; 259 q->tqe_len -= i; 260 m_adj(q->tqe_m, i); 261 break; 262 } 263 264 nq = LIST_NEXT(q, tqe_q); 265 LIST_REMOVE(q, tqe_q); 266 m_freem(q->tqe_m); 267 free(q, M_TSEGQ); 268 q = nq; 269 } 270 271 /* Insert the new segment queue entry into place. */ 272 te->tqe_m = m; 273 te->tqe_th = th; 274 te->tqe_len = *tlenp; 275 276 if (p == NULL) { 277 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 278 } else { 279 LIST_INSERT_AFTER(p, te, tqe_q); 280 } 281 282 present: 283 /* 284 * Present data to user, advancing rcv_nxt through 285 * completed sequence space. 286 */ 287 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 288 return (0); 289 q = LIST_FIRST(&tp->t_segq); 290 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 291 return (0); 292 do { 293 tp->rcv_nxt += q->tqe_len; 294 flags = q->tqe_th->th_flags & TH_FIN; 295 nq = LIST_NEXT(q, tqe_q); 296 LIST_REMOVE(q, tqe_q); 297 if (so->so_state & SS_CANTRCVMORE) 298 m_freem(q->tqe_m); 299 else 300 sbappend(&so->so_rcv, q->tqe_m); 301 free(q, M_TSEGQ); 302 q = nq; 303 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 304 ND6_HINT(tp); 305 sorwakeup(so); 306 return (flags); 307 } 308 309 /* 310 * TCP input routine, follows pages 65-76 of the 311 * protocol specification dated September, 1981 very closely. 312 */ 313 #ifdef INET6 314 int 315 tcp6_input(mp, offp, proto) 316 struct mbuf **mp; 317 int *offp, proto; 318 { 319 struct mbuf *m = *mp; 320 struct in6_ifaddr *ia6; 321 322 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 323 324 /* 325 * draft-itojun-ipv6-tcp-to-anycast 326 * better place to put this in? 327 */ 328 ia6 = ip6_getdstifaddr(m); 329 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 330 struct ip6_hdr *ip6; 331 332 ip6 = mtod(m, struct ip6_hdr *); 333 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 334 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 335 return IPPROTO_DONE; 336 } 337 338 tcp_input(m, *offp, proto); 339 return IPPROTO_DONE; 340 } 341 #endif 342 343 void 344 tcp_input(m, off0, proto) 345 struct mbuf *m; 346 int off0, proto; 347 { 348 struct tcphdr *th; 349 struct ip *ip = NULL; 350 struct ipovly *ipov; 351 struct inpcb *inp = NULL; 352 u_char *optp = NULL; 353 int optlen = 0; 354 int len, tlen, off; 355 int drop_hdrlen; 356 struct tcpcb *tp = NULL; 357 int thflags; 358 struct socket *so = 0; 359 int todrop, acked, ourfinisacked, needoutput = 0; 360 u_long tiwin; 361 struct tcpopt to; /* options in this segment */ 362 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 363 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 364 struct sockaddr_in *next_hop = NULL; 365 int rstreason; /* For badport_bandlim accounting purposes */ 366 struct ip6_hdr *ip6 = NULL; 367 #ifdef INET6 368 int isipv6; 369 #else 370 const int isipv6 = 0; 371 #endif 372 #ifdef TCPDEBUG 373 short ostate = 0; 374 #endif 375 376 /* Grab info from MT_TAG mbufs prepended to the chain. */ 377 for (;m && m->m_type == MT_TAG; m = m->m_next) { 378 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 379 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 380 } 381 #ifdef INET6 382 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 383 #endif 384 bzero((char *)&to, sizeof(to)); 385 386 tcpstat.tcps_rcvtotal++; 387 388 if (isipv6) { 389 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 390 ip6 = mtod(m, struct ip6_hdr *); 391 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 392 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 393 tcpstat.tcps_rcvbadsum++; 394 goto drop; 395 } 396 th = (struct tcphdr *)((caddr_t)ip6 + off0); 397 398 /* 399 * Be proactive about unspecified IPv6 address in source. 400 * As we use all-zero to indicate unbounded/unconnected pcb, 401 * unspecified IPv6 address can be used to confuse us. 402 * 403 * Note that packets with unspecified IPv6 destination is 404 * already dropped in ip6_input. 405 */ 406 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 407 /* XXX stat */ 408 goto drop; 409 } 410 } else { 411 /* 412 * Get IP and TCP header together in first mbuf. 413 * Note: IP leaves IP header in first mbuf. 414 */ 415 if (off0 > sizeof(struct ip)) { 416 ip_stripoptions(m, (struct mbuf *)0); 417 off0 = sizeof(struct ip); 418 } 419 if (m->m_len < sizeof(struct tcpiphdr)) { 420 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) { 421 tcpstat.tcps_rcvshort++; 422 return; 423 } 424 } 425 ip = mtod(m, struct ip *); 426 ipov = (struct ipovly *)ip; 427 th = (struct tcphdr *)((caddr_t)ip + off0); 428 tlen = ip->ip_len; 429 430 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 431 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 432 th->th_sum = m->m_pkthdr.csum_data; 433 else 434 th->th_sum = in_pseudo(ip->ip_src.s_addr, 435 ip->ip_dst.s_addr, 436 htonl(m->m_pkthdr.csum_data + 437 ip->ip_len + 438 IPPROTO_TCP)); 439 th->th_sum ^= 0xffff; 440 } else { 441 /* 442 * Checksum extended TCP header and data. 443 */ 444 len = sizeof(struct ip) + tlen; 445 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 446 ipov->ih_len = (u_short)tlen; 447 ipov->ih_len = htons(ipov->ih_len); 448 th->th_sum = in_cksum(m, len); 449 } 450 if (th->th_sum) { 451 tcpstat.tcps_rcvbadsum++; 452 goto drop; 453 } 454 #ifdef INET6 455 /* Re-initialization for later version check */ 456 ip->ip_v = IPVERSION; 457 #endif 458 } 459 460 /* 461 * Check that TCP offset makes sense, 462 * pull out TCP options and adjust length. XXX 463 */ 464 off = th->th_off << 2; 465 if (off < sizeof(struct tcphdr) || off > tlen) { 466 tcpstat.tcps_rcvbadoff++; 467 goto drop; 468 } 469 tlen -= off; /* tlen is used instead of ti->ti_len */ 470 if (off > sizeof(struct tcphdr)) { 471 if (isipv6) { 472 IP6_EXTHDR_CHECK(m, off0, off, ); 473 ip6 = mtod(m, struct ip6_hdr *); 474 th = (struct tcphdr *)((caddr_t)ip6 + off0); 475 } else { 476 if (m->m_len < sizeof(struct ip) + off) { 477 if ((m = m_pullup(m, sizeof(struct ip) + off)) 478 == 0) { 479 tcpstat.tcps_rcvshort++; 480 return; 481 } 482 ip = mtod(m, struct ip *); 483 ipov = (struct ipovly *)ip; 484 th = (struct tcphdr *)((caddr_t)ip + off0); 485 } 486 } 487 optlen = off - sizeof(struct tcphdr); 488 optp = (u_char *)(th + 1); 489 } 490 thflags = th->th_flags; 491 492 #ifdef TCP_DROP_SYNFIN 493 /* 494 * If the drop_synfin option is enabled, drop all packets with 495 * both the SYN and FIN bits set. This prevents e.g. nmap from 496 * identifying the TCP/IP stack. 497 * 498 * This is a violation of the TCP specification. 499 */ 500 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 501 goto drop; 502 #endif 503 504 /* 505 * Convert TCP protocol specific fields to host format. 506 */ 507 th->th_seq = ntohl(th->th_seq); 508 th->th_ack = ntohl(th->th_ack); 509 th->th_win = ntohs(th->th_win); 510 th->th_urp = ntohs(th->th_urp); 511 512 /* 513 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 514 * until after ip6_savecontrol() is called and before other functions 515 * which don't want those proto headers. 516 * Because ip6_savecontrol() is going to parse the mbuf to 517 * search for data to be passed up to user-land, it wants mbuf 518 * parameters to be unchanged. 519 * XXX: the call of ip6_savecontrol() has been obsoleted based on 520 * latest version of the advanced API (20020110). 521 */ 522 drop_hdrlen = off0 + off; 523 524 /* 525 * Locate pcb for segment. 526 */ 527 findpcb: 528 /* IPFIREWALL_FORWARD section */ 529 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 530 /* 531 * Transparently forwarded. Pretend to be the destination. 532 * already got one like this? 533 */ 534 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 535 ip->ip_dst, th->th_dport, 536 0, m->m_pkthdr.rcvif); 537 if (!inp) { 538 /* It's new. Try find the ambushing socket. */ 539 inp = in_pcblookup_hash(&tcbinfo, 540 ip->ip_src, th->th_sport, 541 next_hop->sin_addr, 542 next_hop->sin_port ? 543 ntohs(next_hop->sin_port) : 544 th->th_dport, 545 1, m->m_pkthdr.rcvif); 546 } 547 } else { 548 if (isipv6) 549 inp = in6_pcblookup_hash(&tcbinfo, 550 &ip6->ip6_src, th->th_sport, 551 &ip6->ip6_dst, th->th_dport, 552 1, m->m_pkthdr.rcvif); 553 else 554 inp = in_pcblookup_hash(&tcbinfo, 555 ip->ip_src, th->th_sport, 556 ip->ip_dst, th->th_dport, 557 1, m->m_pkthdr.rcvif); 558 } 559 560 #ifdef IPSEC 561 if (isipv6) { 562 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 563 ipsec6stat.in_polvio++; 564 goto drop; 565 } 566 } else { 567 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 568 ipsecstat.in_polvio++; 569 goto drop; 570 } 571 } 572 #endif 573 #ifdef FAST_IPSEC 574 if (isipv6) { 575 if (inp != NULL && ipsec6_in_reject(m, inp)) { 576 goto drop; 577 } 578 } else { 579 if (inp != NULL && ipsec4_in_reject(m, inp)) { 580 goto drop; 581 } 582 } 583 #endif 584 585 /* 586 * If the state is CLOSED (i.e., TCB does not exist) then 587 * all data in the incoming segment is discarded. 588 * If the TCB exists but is in CLOSED state, it is embryonic, 589 * but should either do a listen or a connect soon. 590 */ 591 if (inp == NULL) { 592 if (log_in_vain) { 593 #ifdef INET6 594 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 595 #else 596 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 597 #endif 598 if (isipv6) { 599 strcpy(dbuf, "["); 600 strcpy(sbuf, "["); 601 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 602 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 603 strcat(dbuf, "]"); 604 strcat(sbuf, "]"); 605 } else { 606 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 607 strcpy(sbuf, inet_ntoa(ip->ip_src)); 608 } 609 switch (log_in_vain) { 610 case 1: 611 if ((thflags & TH_SYN) == 0) 612 break; 613 case 2: 614 log(LOG_INFO, 615 "Connection attempt to TCP %s:%d " 616 "from %s:%d flags:0x%02x\n", 617 dbuf, ntohs(th->th_dport), sbuf, 618 ntohs(th->th_sport), thflags); 619 break; 620 default: 621 break; 622 } 623 } 624 if (blackhole) { 625 switch (blackhole) { 626 case 1: 627 if (thflags & TH_SYN) 628 goto drop; 629 break; 630 case 2: 631 goto drop; 632 default: 633 goto drop; 634 } 635 } 636 rstreason = BANDLIM_RST_CLOSEDPORT; 637 goto dropwithreset; 638 } 639 tp = intotcpcb(inp); 640 if (tp == NULL) { 641 rstreason = BANDLIM_RST_CLOSEDPORT; 642 goto dropwithreset; 643 } 644 if (tp->t_state == TCPS_CLOSED) 645 goto drop; 646 647 /* Unscale the window into a 32-bit value. */ 648 if ((thflags & TH_SYN) == 0) 649 tiwin = th->th_win << tp->snd_scale; 650 else 651 tiwin = th->th_win; 652 653 so = inp->inp_socket; 654 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 655 struct in_conninfo inc; 656 #ifdef TCPDEBUG 657 if (so->so_options & SO_DEBUG) { 658 ostate = tp->t_state; 659 if (isipv6) 660 bcopy((char *)ip6, (char *)tcp_saveipgen, 661 sizeof(*ip6)); 662 else 663 bcopy((char *)ip, (char *)tcp_saveipgen, 664 sizeof(*ip)); 665 tcp_savetcp = *th; 666 } 667 #endif 668 /* skip if this isn't a listen socket */ 669 if ((so->so_options & SO_ACCEPTCONN) == 0) 670 goto after_listen; 671 #ifdef INET6 672 inc.inc_isipv6 = isipv6; 673 #endif 674 if (isipv6) { 675 inc.inc6_faddr = ip6->ip6_src; 676 inc.inc6_laddr = ip6->ip6_dst; 677 inc.inc6_route.ro_rt = NULL; /* XXX */ 678 } else { 679 inc.inc_faddr = ip->ip_src; 680 inc.inc_laddr = ip->ip_dst; 681 inc.inc_route.ro_rt = NULL; /* XXX */ 682 } 683 inc.inc_fport = th->th_sport; 684 inc.inc_lport = th->th_dport; 685 686 /* 687 * If the state is LISTEN then ignore segment if it contains 688 * a RST. If the segment contains an ACK then it is bad and 689 * send a RST. If it does not contain a SYN then it is not 690 * interesting; drop it. 691 * 692 * If the state is SYN_RECEIVED (syncache) and seg contains 693 * an ACK, but not for our SYN/ACK, send a RST. If the seg 694 * contains a RST, check the sequence number to see if it 695 * is a valid reset segment. 696 */ 697 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 698 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 699 if (!syncache_expand(&inc, th, &so, m)) { 700 /* 701 * No syncache entry, or ACK was not 702 * for our SYN/ACK. Send a RST. 703 */ 704 tcpstat.tcps_badsyn++; 705 rstreason = BANDLIM_RST_OPENPORT; 706 goto dropwithreset; 707 } 708 if (so == NULL) 709 /* 710 * Could not complete 3-way handshake, 711 * connection is being closed down, and 712 * syncache will free mbuf. 713 */ 714 return; 715 /* 716 * Socket is created in state SYN_RECEIVED. 717 * Continue processing segment. 718 */ 719 inp = sotoinpcb(so); 720 tp = intotcpcb(inp); 721 /* 722 * This is what would have happened in 723 * tcp_output() when the SYN,ACK was sent. 724 */ 725 tp->snd_up = tp->snd_una; 726 tp->snd_max = tp->snd_nxt = tp->iss + 1; 727 tp->last_ack_sent = tp->rcv_nxt; 728 /* 729 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 730 * until the _second_ ACK is received: 731 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 732 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 733 * move to ESTAB, set snd_wnd to tiwin. 734 */ 735 tp->snd_wnd = tiwin; /* unscaled */ 736 goto after_listen; 737 } 738 if (thflags & TH_RST) { 739 syncache_chkrst(&inc, th); 740 goto drop; 741 } 742 if (thflags & TH_ACK) { 743 syncache_badack(&inc); 744 tcpstat.tcps_badsyn++; 745 rstreason = BANDLIM_RST_OPENPORT; 746 goto dropwithreset; 747 } 748 goto drop; 749 } 750 751 /* 752 * Segment's flags are (SYN) or (SYN|FIN). 753 */ 754 #ifdef INET6 755 /* 756 * If deprecated address is forbidden, 757 * we do not accept SYN to deprecated interface 758 * address to prevent any new inbound connection from 759 * getting established. 760 * When we do not accept SYN, we send a TCP RST, 761 * with deprecated source address (instead of dropping 762 * it). We compromise it as it is much better for peer 763 * to send a RST, and RST will be the final packet 764 * for the exchange. 765 * 766 * If we do not forbid deprecated addresses, we accept 767 * the SYN packet. RFC2462 does not suggest dropping 768 * SYN in this case. 769 * If we decipher RFC2462 5.5.4, it says like this: 770 * 1. use of deprecated addr with existing 771 * communication is okay - "SHOULD continue to be 772 * used" 773 * 2. use of it with new communication: 774 * (2a) "SHOULD NOT be used if alternate address 775 * with sufficient scope is available" 776 * (2b) nothing mentioned otherwise. 777 * Here we fall into (2b) case as we have no choice in 778 * our source address selection - we must obey the peer. 779 * 780 * The wording in RFC2462 is confusing, and there are 781 * multiple description text for deprecated address 782 * handling - worse, they are not exactly the same. 783 * I believe 5.5.4 is the best one, so we follow 5.5.4. 784 */ 785 if (isipv6 && !ip6_use_deprecated) { 786 struct in6_ifaddr *ia6; 787 788 if ((ia6 = ip6_getdstifaddr(m)) && 789 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 790 tp = NULL; 791 rstreason = BANDLIM_RST_OPENPORT; 792 goto dropwithreset; 793 } 794 } 795 #endif 796 /* 797 * If it is from this socket, drop it, it must be forged. 798 * Don't bother responding if the destination was a broadcast. 799 */ 800 if (th->th_dport == th->th_sport) { 801 if (isipv6) { 802 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 803 &ip6->ip6_src)) 804 goto drop; 805 } else { 806 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 807 goto drop; 808 } 809 } 810 /* 811 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 812 * 813 * Note that it is quite possible to receive unicast 814 * link-layer packets with a broadcast IP address. Use 815 * in_broadcast() to find them. 816 */ 817 if (m->m_flags & (M_BCAST|M_MCAST)) 818 goto drop; 819 if (isipv6) { 820 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 821 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 822 goto drop; 823 } else { 824 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 825 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 826 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 827 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 828 goto drop; 829 } 830 /* 831 * SYN appears to be valid; create compressed TCP state 832 * for syncache, or perform t/tcp connection. 833 */ 834 if (so->so_qlen <= so->so_qlimit) { 835 tcp_dooptions(&to, optp, optlen, 1); 836 if (!syncache_add(&inc, &to, th, &so, m)) 837 goto drop; 838 if (so == NULL) 839 /* 840 * Entry added to syncache, mbuf used to 841 * send SYN,ACK packet. 842 */ 843 return; 844 /* 845 * Segment passed TAO tests. 846 */ 847 inp = sotoinpcb(so); 848 tp = intotcpcb(inp); 849 tp->snd_wnd = tiwin; 850 tp->t_starttime = ticks; 851 tp->t_state = TCPS_ESTABLISHED; 852 853 /* 854 * If there is a FIN, or if there is data and the 855 * connection is local, then delay SYN,ACK(SYN) in 856 * the hope of piggy-backing it on a response 857 * segment. Otherwise must send ACK now in case 858 * the other side is slow starting. 859 */ 860 if (DELAY_ACK(tp) && 861 ((thflags & TH_FIN) || 862 (tlen != 0 && 863 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 864 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 865 callout_reset(tp->tt_delack, tcp_delacktime, 866 tcp_timer_delack, tp); 867 tp->t_flags |= TF_NEEDSYN; 868 } else 869 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 870 871 tcpstat.tcps_connects++; 872 soisconnected(so); 873 goto trimthenstep6; 874 } 875 goto drop; 876 } 877 after_listen: 878 879 /* XXX temp debugging */ 880 /* should not happen - syncache should pick up these connections */ 881 if (tp->t_state == TCPS_LISTEN) 882 panic("tcp_input: TCPS_LISTEN"); 883 884 /* 885 * Segment received on connection. 886 * Reset idle time and keep-alive timer. 887 */ 888 tp->t_rcvtime = ticks; 889 if (TCPS_HAVEESTABLISHED(tp->t_state)) 890 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 891 892 /* 893 * Process options. 894 * XXX this is tradtitional behavior, may need to be cleaned up. 895 */ 896 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 897 if (thflags & TH_SYN) { 898 if (to.to_flags & TOF_SCALE) { 899 tp->t_flags |= TF_RCVD_SCALE; 900 tp->requested_s_scale = to.to_requested_s_scale; 901 } 902 if (to.to_flags & TOF_TS) { 903 tp->t_flags |= TF_RCVD_TSTMP; 904 tp->ts_recent = to.to_tsval; 905 tp->ts_recent_age = ticks; 906 } 907 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 908 tp->t_flags |= TF_RCVD_CC; 909 if (to.to_flags & TOF_MSS) 910 tcp_mss(tp, to.to_mss); 911 } 912 913 /* 914 * Header prediction: check for the two common cases 915 * of a uni-directional data xfer. If the packet has 916 * no control flags, is in-sequence, the window didn't 917 * change and we're not retransmitting, it's a 918 * candidate. If the length is zero and the ack moved 919 * forward, we're the sender side of the xfer. Just 920 * free the data acked & wake any higher level process 921 * that was blocked waiting for space. If the length 922 * is non-zero and the ack didn't move, we're the 923 * receiver side. If we're getting packets in-order 924 * (the reassembly queue is empty), add the data to 925 * the socket buffer and note that we need a delayed ack. 926 * Make sure that the hidden state-flags are also off. 927 * Since we check for TCPS_ESTABLISHED above, it can only 928 * be TH_NEEDSYN. 929 */ 930 if (tp->t_state == TCPS_ESTABLISHED && 931 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 932 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 933 ((to.to_flags & TOF_TS) == 0 || 934 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 935 /* 936 * Using the CC option is compulsory if once started: 937 * the segment is OK if no T/TCP was negotiated or 938 * if the segment has a CC option equal to CCrecv 939 */ 940 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 941 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 942 th->th_seq == tp->rcv_nxt && 943 tiwin && tiwin == tp->snd_wnd && 944 tp->snd_nxt == tp->snd_max) { 945 946 /* 947 * If last ACK falls within this segment's sequence numbers, 948 * record the timestamp. 949 * NOTE that the test is modified according to the latest 950 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 951 */ 952 if ((to.to_flags & TOF_TS) != 0 && 953 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 954 tp->ts_recent_age = ticks; 955 tp->ts_recent = to.to_tsval; 956 } 957 958 if (tlen == 0) { 959 if (SEQ_GT(th->th_ack, tp->snd_una) && 960 SEQ_LEQ(th->th_ack, tp->snd_max) && 961 tp->snd_cwnd >= tp->snd_wnd && 962 ((!tcp_do_newreno && 963 tp->t_dupacks < tcprexmtthresh) || 964 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 965 /* 966 * this is a pure ack for outstanding data. 967 */ 968 ++tcpstat.tcps_predack; 969 /* 970 * "bad retransmit" recovery 971 * 972 * If Eifel detection applies, then 973 * it is deterministic, so use it 974 * unconditionally over the old heuristic. 975 * Otherwise, fall back to the old heuristic. 976 */ 977 if (tcp_do_eifel_detect && 978 (to.to_flags & TOF_TS) && to.to_tsecr && 979 (tp->t_flags & TF_FIRSTACCACK)) { 980 /* Eifel detection applicable. */ 981 if (to.to_tsecr < tp->t_rexmtTS) { 982 tcp_revert_congestion_state(tp); 983 ++tcpstat.tcps_eifeldetected; 984 } 985 } else if (tp->t_rxtshift == 1 && 986 ticks < tp->t_badrxtwin) { 987 tcp_revert_congestion_state(tp); 988 ++tcpstat.tcps_rttdetected; 989 } 990 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT); 991 /* 992 * Recalculate the retransmit timer / rtt. 993 * 994 * Some machines (certain windows boxes) 995 * send broken timestamp replies during the 996 * SYN+ACK phase, ignore timestamps of 0. 997 */ 998 if ((to.to_flags & TOF_TS) != 0 && 999 to.to_tsecr) { 1000 tcp_xmit_timer(tp, 1001 ticks - to.to_tsecr + 1); 1002 } else if (tp->t_rtttime && 1003 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1004 tcp_xmit_timer(tp, 1005 ticks - tp->t_rtttime); 1006 } 1007 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1008 acked = th->th_ack - tp->snd_una; 1009 tcpstat.tcps_rcvackpack++; 1010 tcpstat.tcps_rcvackbyte += acked; 1011 sbdrop(&so->so_snd, acked); 1012 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1013 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1014 tp->snd_recover = th->th_ack - 1; 1015 tp->snd_una = th->th_ack; 1016 tp->t_dupacks = 0; 1017 m_freem(m); 1018 ND6_HINT(tp); /* some progress has been done */ 1019 1020 /* 1021 * If all outstanding data are acked, stop 1022 * retransmit timer, otherwise restart timer 1023 * using current (possibly backed-off) value. 1024 * If process is waiting for space, 1025 * wakeup/selwakeup/signal. If data 1026 * are ready to send, let tcp_output 1027 * decide between more output or persist. 1028 */ 1029 if (tp->snd_una == tp->snd_max) 1030 callout_stop(tp->tt_rexmt); 1031 else if (!callout_active(tp->tt_persist)) 1032 callout_reset(tp->tt_rexmt, 1033 tp->t_rxtcur, 1034 tcp_timer_rexmt, tp); 1035 1036 sowwakeup(so); 1037 if (so->so_snd.sb_cc) 1038 (void) tcp_output(tp); 1039 return; 1040 } 1041 } else if (th->th_ack == tp->snd_una && 1042 LIST_EMPTY(&tp->t_segq) && 1043 tlen <= sbspace(&so->so_rcv)) { 1044 /* 1045 * this is a pure, in-sequence data packet 1046 * with nothing on the reassembly queue and 1047 * we have enough buffer space to take it. 1048 */ 1049 ++tcpstat.tcps_preddat; 1050 tp->rcv_nxt += tlen; 1051 tcpstat.tcps_rcvpack++; 1052 tcpstat.tcps_rcvbyte += tlen; 1053 ND6_HINT(tp); /* some progress has been done */ 1054 /* 1055 * Add data to socket buffer. 1056 */ 1057 if (so->so_state & SS_CANTRCVMORE) { 1058 m_freem(m); 1059 } else { 1060 m_adj(m, drop_hdrlen); /* delayed header drop */ 1061 sbappend(&so->so_rcv, m); 1062 } 1063 sorwakeup(so); 1064 if (DELAY_ACK(tp)) { 1065 callout_reset(tp->tt_delack, tcp_delacktime, 1066 tcp_timer_delack, tp); 1067 } else { 1068 tp->t_flags |= TF_ACKNOW; 1069 tcp_output(tp); 1070 } 1071 return; 1072 } 1073 } 1074 1075 /* 1076 * Calculate amount of space in receive window, 1077 * and then do TCP input processing. 1078 * Receive window is amount of space in rcv queue, 1079 * but not less than advertised window. 1080 */ 1081 { int win; 1082 1083 win = sbspace(&so->so_rcv); 1084 if (win < 0) 1085 win = 0; 1086 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1087 } 1088 1089 switch (tp->t_state) { 1090 1091 /* 1092 * If the state is SYN_RECEIVED: 1093 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1094 */ 1095 case TCPS_SYN_RECEIVED: 1096 if ((thflags & TH_ACK) && 1097 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1098 SEQ_GT(th->th_ack, tp->snd_max))) { 1099 rstreason = BANDLIM_RST_OPENPORT; 1100 goto dropwithreset; 1101 } 1102 break; 1103 1104 /* 1105 * If the state is SYN_SENT: 1106 * if seg contains an ACK, but not for our SYN, drop the input. 1107 * if seg contains a RST, then drop the connection. 1108 * if seg does not contain SYN, then drop it. 1109 * Otherwise this is an acceptable SYN segment 1110 * initialize tp->rcv_nxt and tp->irs 1111 * if seg contains ack then advance tp->snd_una 1112 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1113 * arrange for segment to be acked (eventually) 1114 * continue processing rest of data/controls, beginning with URG 1115 */ 1116 case TCPS_SYN_SENT: 1117 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1118 taop = &tao_noncached; 1119 bzero(taop, sizeof(*taop)); 1120 } 1121 1122 if ((thflags & TH_ACK) && 1123 (SEQ_LEQ(th->th_ack, tp->iss) || 1124 SEQ_GT(th->th_ack, tp->snd_max))) { 1125 /* 1126 * If we have a cached CCsent for the remote host, 1127 * hence we haven't just crashed and restarted, 1128 * do not send a RST. This may be a retransmission 1129 * from the other side after our earlier ACK was lost. 1130 * Our new SYN, when it arrives, will serve as the 1131 * needed ACK. 1132 */ 1133 if (taop->tao_ccsent != 0) 1134 goto drop; 1135 else { 1136 rstreason = BANDLIM_UNLIMITED; 1137 goto dropwithreset; 1138 } 1139 } 1140 if (thflags & TH_RST) { 1141 if (thflags & TH_ACK) 1142 tp = tcp_drop(tp, ECONNREFUSED); 1143 goto drop; 1144 } 1145 if ((thflags & TH_SYN) == 0) 1146 goto drop; 1147 tp->snd_wnd = th->th_win; /* initial send window */ 1148 tp->cc_recv = to.to_cc; /* foreign CC */ 1149 1150 tp->irs = th->th_seq; 1151 tcp_rcvseqinit(tp); 1152 if (thflags & TH_ACK) { 1153 /* 1154 * Our SYN was acked. If segment contains CC.ECHO 1155 * option, check it to make sure this segment really 1156 * matches our SYN. If not, just drop it as old 1157 * duplicate, but send an RST if we're still playing 1158 * by the old rules. If no CC.ECHO option, make sure 1159 * we don't get fooled into using T/TCP. 1160 */ 1161 if (to.to_flags & TOF_CCECHO) { 1162 if (tp->cc_send != to.to_ccecho) { 1163 if (taop->tao_ccsent != 0) 1164 goto drop; 1165 else { 1166 rstreason = BANDLIM_UNLIMITED; 1167 goto dropwithreset; 1168 } 1169 } 1170 } else 1171 tp->t_flags &= ~TF_RCVD_CC; 1172 tcpstat.tcps_connects++; 1173 soisconnected(so); 1174 /* Do window scaling on this connection? */ 1175 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1176 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1177 tp->snd_scale = tp->requested_s_scale; 1178 tp->rcv_scale = tp->request_r_scale; 1179 } 1180 /* Segment is acceptable, update cache if undefined. */ 1181 if (taop->tao_ccsent == 0) 1182 taop->tao_ccsent = to.to_ccecho; 1183 1184 tp->rcv_adv += tp->rcv_wnd; 1185 tp->snd_una++; /* SYN is acked */ 1186 /* 1187 * If there's data, delay ACK; if there's also a FIN 1188 * ACKNOW will be turned on later. 1189 */ 1190 if (DELAY_ACK(tp) && tlen != 0) 1191 callout_reset(tp->tt_delack, tcp_delacktime, 1192 tcp_timer_delack, tp); 1193 else 1194 tp->t_flags |= TF_ACKNOW; 1195 /* 1196 * Received <SYN,ACK> in SYN_SENT[*] state. 1197 * Transitions: 1198 * SYN_SENT --> ESTABLISHED 1199 * SYN_SENT* --> FIN_WAIT_1 1200 */ 1201 tp->t_starttime = ticks; 1202 if (tp->t_flags & TF_NEEDFIN) { 1203 tp->t_state = TCPS_FIN_WAIT_1; 1204 tp->t_flags &= ~TF_NEEDFIN; 1205 thflags &= ~TH_SYN; 1206 } else { 1207 tp->t_state = TCPS_ESTABLISHED; 1208 callout_reset(tp->tt_keep, tcp_keepidle, 1209 tcp_timer_keep, tp); 1210 } 1211 } else { 1212 /* 1213 * Received initial SYN in SYN-SENT[*] state => 1214 * simultaneous open. If segment contains CC option 1215 * and there is a cached CC, apply TAO test. 1216 * If it succeeds, connection is * half-synchronized. 1217 * Otherwise, do 3-way handshake: 1218 * SYN-SENT -> SYN-RECEIVED 1219 * SYN-SENT* -> SYN-RECEIVED* 1220 * If there was no CC option, clear cached CC value. 1221 */ 1222 tp->t_flags |= TF_ACKNOW; 1223 callout_stop(tp->tt_rexmt); 1224 if (to.to_flags & TOF_CC) { 1225 if (taop->tao_cc != 0 && 1226 CC_GT(to.to_cc, taop->tao_cc)) { 1227 /* 1228 * update cache and make transition: 1229 * SYN-SENT -> ESTABLISHED* 1230 * SYN-SENT* -> FIN-WAIT-1* 1231 */ 1232 taop->tao_cc = to.to_cc; 1233 tp->t_starttime = ticks; 1234 if (tp->t_flags & TF_NEEDFIN) { 1235 tp->t_state = TCPS_FIN_WAIT_1; 1236 tp->t_flags &= ~TF_NEEDFIN; 1237 } else { 1238 tp->t_state = TCPS_ESTABLISHED; 1239 callout_reset(tp->tt_keep, 1240 tcp_keepidle, 1241 tcp_timer_keep, 1242 tp); 1243 } 1244 tp->t_flags |= TF_NEEDSYN; 1245 } else 1246 tp->t_state = TCPS_SYN_RECEIVED; 1247 } else { 1248 /* CC.NEW or no option => invalidate cache */ 1249 taop->tao_cc = 0; 1250 tp->t_state = TCPS_SYN_RECEIVED; 1251 } 1252 } 1253 1254 trimthenstep6: 1255 /* 1256 * Advance th->th_seq to correspond to first data byte. 1257 * If data, trim to stay within window, 1258 * dropping FIN if necessary. 1259 */ 1260 th->th_seq++; 1261 if (tlen > tp->rcv_wnd) { 1262 todrop = tlen - tp->rcv_wnd; 1263 m_adj(m, -todrop); 1264 tlen = tp->rcv_wnd; 1265 thflags &= ~TH_FIN; 1266 tcpstat.tcps_rcvpackafterwin++; 1267 tcpstat.tcps_rcvbyteafterwin += todrop; 1268 } 1269 tp->snd_wl1 = th->th_seq - 1; 1270 tp->rcv_up = th->th_seq; 1271 /* 1272 * Client side of transaction: already sent SYN and data. 1273 * If the remote host used T/TCP to validate the SYN, 1274 * our data will be ACK'd; if so, enter normal data segment 1275 * processing in the middle of step 5, ack processing. 1276 * Otherwise, goto step 6. 1277 */ 1278 if (thflags & TH_ACK) 1279 goto process_ACK; 1280 1281 goto step6; 1282 1283 /* 1284 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1285 * if segment contains a SYN and CC [not CC.NEW] option: 1286 * if state == TIME_WAIT and connection duration > MSL, 1287 * drop packet and send RST; 1288 * 1289 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1290 * ack the FIN (and data) in retransmission queue. 1291 * Complete close and delete TCPCB. Then reprocess 1292 * segment, hoping to find new TCPCB in LISTEN state; 1293 * 1294 * else must be old SYN; drop it. 1295 * else do normal processing. 1296 */ 1297 case TCPS_LAST_ACK: 1298 case TCPS_CLOSING: 1299 case TCPS_TIME_WAIT: 1300 if ((thflags & TH_SYN) && 1301 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1302 if (tp->t_state == TCPS_TIME_WAIT && 1303 (ticks - tp->t_starttime) > tcp_msl) { 1304 rstreason = BANDLIM_UNLIMITED; 1305 goto dropwithreset; 1306 } 1307 if (CC_GT(to.to_cc, tp->cc_recv)) { 1308 tp = tcp_close(tp); 1309 goto findpcb; 1310 } 1311 else 1312 goto drop; 1313 } 1314 break; /* continue normal processing */ 1315 } 1316 1317 /* 1318 * States other than LISTEN or SYN_SENT. 1319 * First check the RST flag and sequence number since reset segments 1320 * are exempt from the timestamp and connection count tests. This 1321 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1322 * below which allowed reset segments in half the sequence space 1323 * to fall though and be processed (which gives forged reset 1324 * segments with a random sequence number a 50 percent chance of 1325 * killing a connection). 1326 * Then check timestamp, if present. 1327 * Then check the connection count, if present. 1328 * Then check that at least some bytes of segment are within 1329 * receive window. If segment begins before rcv_nxt, 1330 * drop leading data (and SYN); if nothing left, just ack. 1331 * 1332 * 1333 * If the RST bit is set, check the sequence number to see 1334 * if this is a valid reset segment. 1335 * RFC 793 page 37: 1336 * In all states except SYN-SENT, all reset (RST) segments 1337 * are validated by checking their SEQ-fields. A reset is 1338 * valid if its sequence number is in the window. 1339 * Note: this does not take into account delayed ACKs, so 1340 * we should test against last_ack_sent instead of rcv_nxt. 1341 * The sequence number in the reset segment is normally an 1342 * echo of our outgoing acknowlegement numbers, but some hosts 1343 * send a reset with the sequence number at the rightmost edge 1344 * of our receive window, and we have to handle this case. 1345 * If we have multiple segments in flight, the intial reset 1346 * segment sequence numbers will be to the left of last_ack_sent, 1347 * but they will eventually catch up. 1348 * In any case, it never made sense to trim reset segments to 1349 * fit the receive window since RFC 1122 says: 1350 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1351 * 1352 * A TCP SHOULD allow a received RST segment to include data. 1353 * 1354 * DISCUSSION 1355 * It has been suggested that a RST segment could contain 1356 * ASCII text that encoded and explained the cause of the 1357 * RST. No standard has yet been established for such 1358 * data. 1359 * 1360 * If the reset segment passes the sequence number test examine 1361 * the state: 1362 * SYN_RECEIVED STATE: 1363 * If passive open, return to LISTEN state. 1364 * If active open, inform user that connection was refused. 1365 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1366 * Inform user that connection was reset, and close tcb. 1367 * CLOSING, LAST_ACK STATES: 1368 * Close the tcb. 1369 * TIME_WAIT STATE: 1370 * Drop the segment - see Stevens, vol. 2, p. 964 and 1371 * RFC 1337. 1372 */ 1373 if (thflags & TH_RST) { 1374 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1375 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1376 switch (tp->t_state) { 1377 1378 case TCPS_SYN_RECEIVED: 1379 so->so_error = ECONNREFUSED; 1380 goto close; 1381 1382 case TCPS_ESTABLISHED: 1383 case TCPS_FIN_WAIT_1: 1384 case TCPS_FIN_WAIT_2: 1385 case TCPS_CLOSE_WAIT: 1386 so->so_error = ECONNRESET; 1387 close: 1388 tp->t_state = TCPS_CLOSED; 1389 tcpstat.tcps_drops++; 1390 tp = tcp_close(tp); 1391 break; 1392 1393 case TCPS_CLOSING: 1394 case TCPS_LAST_ACK: 1395 tp = tcp_close(tp); 1396 break; 1397 1398 case TCPS_TIME_WAIT: 1399 break; 1400 } 1401 } 1402 goto drop; 1403 } 1404 1405 /* 1406 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1407 * and it's less than ts_recent, drop it. 1408 */ 1409 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1410 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1411 1412 /* Check to see if ts_recent is over 24 days old. */ 1413 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1414 /* 1415 * Invalidate ts_recent. If this segment updates 1416 * ts_recent, the age will be reset later and ts_recent 1417 * will get a valid value. If it does not, setting 1418 * ts_recent to zero will at least satisfy the 1419 * requirement that zero be placed in the timestamp 1420 * echo reply when ts_recent isn't valid. The 1421 * age isn't reset until we get a valid ts_recent 1422 * because we don't want out-of-order segments to be 1423 * dropped when ts_recent is old. 1424 */ 1425 tp->ts_recent = 0; 1426 } else { 1427 tcpstat.tcps_rcvduppack++; 1428 tcpstat.tcps_rcvdupbyte += tlen; 1429 tcpstat.tcps_pawsdrop++; 1430 if (tlen) 1431 goto dropafterack; 1432 goto drop; 1433 } 1434 } 1435 1436 /* 1437 * T/TCP mechanism 1438 * If T/TCP was negotiated and the segment doesn't have CC, 1439 * or if its CC is wrong then drop the segment. 1440 * RST segments do not have to comply with this. 1441 */ 1442 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1443 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1444 goto dropafterack; 1445 1446 /* 1447 * In the SYN-RECEIVED state, validate that the packet belongs to 1448 * this connection before trimming the data to fit the receive 1449 * window. Check the sequence number versus IRS since we know 1450 * the sequence numbers haven't wrapped. This is a partial fix 1451 * for the "LAND" DoS attack. 1452 */ 1453 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1454 rstreason = BANDLIM_RST_OPENPORT; 1455 goto dropwithreset; 1456 } 1457 1458 todrop = tp->rcv_nxt - th->th_seq; 1459 if (todrop > 0) { 1460 if (thflags & TH_SYN) { 1461 thflags &= ~TH_SYN; 1462 th->th_seq++; 1463 if (th->th_urp > 1) 1464 th->th_urp--; 1465 else 1466 thflags &= ~TH_URG; 1467 todrop--; 1468 } 1469 /* 1470 * Following if statement from Stevens, vol. 2, p. 960. 1471 */ 1472 if (todrop > tlen 1473 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1474 /* 1475 * Any valid FIN must be to the left of the window. 1476 * At this point the FIN must be a duplicate or out 1477 * of sequence; drop it. 1478 */ 1479 thflags &= ~TH_FIN; 1480 1481 /* 1482 * Send an ACK to resynchronize and drop any data. 1483 * But keep on processing for RST or ACK. 1484 */ 1485 tp->t_flags |= TF_ACKNOW; 1486 todrop = tlen; 1487 tcpstat.tcps_rcvduppack++; 1488 tcpstat.tcps_rcvdupbyte += todrop; 1489 } else { 1490 tcpstat.tcps_rcvpartduppack++; 1491 tcpstat.tcps_rcvpartdupbyte += todrop; 1492 } 1493 drop_hdrlen += todrop; /* drop from the top afterwards */ 1494 th->th_seq += todrop; 1495 tlen -= todrop; 1496 if (th->th_urp > todrop) 1497 th->th_urp -= todrop; 1498 else { 1499 thflags &= ~TH_URG; 1500 th->th_urp = 0; 1501 } 1502 } 1503 1504 /* 1505 * If new data are received on a connection after the 1506 * user processes are gone, then RST the other end. 1507 */ 1508 if ((so->so_state & SS_NOFDREF) && 1509 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1510 tp = tcp_close(tp); 1511 tcpstat.tcps_rcvafterclose++; 1512 rstreason = BANDLIM_UNLIMITED; 1513 goto dropwithreset; 1514 } 1515 1516 /* 1517 * If segment ends after window, drop trailing data 1518 * (and PUSH and FIN); if nothing left, just ACK. 1519 */ 1520 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1521 if (todrop > 0) { 1522 tcpstat.tcps_rcvpackafterwin++; 1523 if (todrop >= tlen) { 1524 tcpstat.tcps_rcvbyteafterwin += tlen; 1525 /* 1526 * If a new connection request is received 1527 * while in TIME_WAIT, drop the old connection 1528 * and start over if the sequence numbers 1529 * are above the previous ones. 1530 */ 1531 if (thflags & TH_SYN && 1532 tp->t_state == TCPS_TIME_WAIT && 1533 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1534 tp = tcp_close(tp); 1535 goto findpcb; 1536 } 1537 /* 1538 * If window is closed can only take segments at 1539 * window edge, and have to drop data and PUSH from 1540 * incoming segments. Continue processing, but 1541 * remember to ack. Otherwise, drop segment 1542 * and ack. 1543 */ 1544 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1545 tp->t_flags |= TF_ACKNOW; 1546 tcpstat.tcps_rcvwinprobe++; 1547 } else 1548 goto dropafterack; 1549 } else 1550 tcpstat.tcps_rcvbyteafterwin += todrop; 1551 m_adj(m, -todrop); 1552 tlen -= todrop; 1553 thflags &= ~(TH_PUSH|TH_FIN); 1554 } 1555 1556 /* 1557 * If last ACK falls within this segment's sequence numbers, 1558 * record its timestamp. 1559 * NOTE that the test is modified according to the latest 1560 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1561 */ 1562 if ((to.to_flags & TOF_TS) != 0 && 1563 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1564 tp->ts_recent_age = ticks; 1565 tp->ts_recent = to.to_tsval; 1566 } 1567 1568 /* 1569 * If a SYN is in the window, then this is an 1570 * error and we send an RST and drop the connection. 1571 */ 1572 if (thflags & TH_SYN) { 1573 tp = tcp_drop(tp, ECONNRESET); 1574 rstreason = BANDLIM_UNLIMITED; 1575 goto dropwithreset; 1576 } 1577 1578 /* 1579 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1580 * flag is on (half-synchronized state), then queue data for 1581 * later processing; else drop segment and return. 1582 */ 1583 if ((thflags & TH_ACK) == 0) { 1584 if (tp->t_state == TCPS_SYN_RECEIVED || 1585 (tp->t_flags & TF_NEEDSYN)) 1586 goto step6; 1587 else 1588 goto drop; 1589 } 1590 1591 /* 1592 * Ack processing. 1593 */ 1594 switch (tp->t_state) { 1595 1596 /* 1597 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1598 * ESTABLISHED state and continue processing. 1599 * The ACK was checked above. 1600 */ 1601 case TCPS_SYN_RECEIVED: 1602 1603 tcpstat.tcps_connects++; 1604 soisconnected(so); 1605 /* Do window scaling? */ 1606 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1607 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1608 tp->snd_scale = tp->requested_s_scale; 1609 tp->rcv_scale = tp->request_r_scale; 1610 } 1611 /* 1612 * Upon successful completion of 3-way handshake, 1613 * update cache.CC if it was undefined, pass any queued 1614 * data to the user, and advance state appropriately. 1615 */ 1616 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1617 taop->tao_cc == 0) 1618 taop->tao_cc = tp->cc_recv; 1619 1620 /* 1621 * Make transitions: 1622 * SYN-RECEIVED -> ESTABLISHED 1623 * SYN-RECEIVED* -> FIN-WAIT-1 1624 */ 1625 tp->t_starttime = ticks; 1626 if (tp->t_flags & TF_NEEDFIN) { 1627 tp->t_state = TCPS_FIN_WAIT_1; 1628 tp->t_flags &= ~TF_NEEDFIN; 1629 } else { 1630 tp->t_state = TCPS_ESTABLISHED; 1631 callout_reset(tp->tt_keep, tcp_keepidle, 1632 tcp_timer_keep, tp); 1633 } 1634 /* 1635 * If segment contains data or ACK, will call tcp_reass() 1636 * later; if not, do so now to pass queued data to user. 1637 */ 1638 if (tlen == 0 && (thflags & TH_FIN) == 0) 1639 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1640 (struct mbuf *)0); 1641 tp->snd_wl1 = th->th_seq - 1; 1642 /* fall into ... */ 1643 1644 /* 1645 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1646 * ACKs. If the ack is in the range 1647 * tp->snd_una < th->th_ack <= tp->snd_max 1648 * then advance tp->snd_una to th->th_ack and drop 1649 * data from the retransmission queue. If this ACK reflects 1650 * more up to date window information we update our window information. 1651 */ 1652 case TCPS_ESTABLISHED: 1653 case TCPS_FIN_WAIT_1: 1654 case TCPS_FIN_WAIT_2: 1655 case TCPS_CLOSE_WAIT: 1656 case TCPS_CLOSING: 1657 case TCPS_LAST_ACK: 1658 case TCPS_TIME_WAIT: 1659 1660 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1661 if (tlen == 0 && tiwin == tp->snd_wnd) { 1662 tcpstat.tcps_rcvdupack++; 1663 /* 1664 * If we have outstanding data (other than 1665 * a window probe), this is a completely 1666 * duplicate ack (ie, window info didn't 1667 * change), the ack is the biggest we've 1668 * seen and we've seen exactly our rexmt 1669 * threshhold of them, assume a packet 1670 * has been dropped and retransmit it. 1671 * Kludge snd_nxt & the congestion 1672 * window so we send only this one 1673 * packet. 1674 * 1675 * We know we're losing at the current 1676 * window size so do congestion avoidance 1677 * (set ssthresh to half the current window 1678 * and pull our congestion window back to 1679 * the new ssthresh). 1680 * 1681 * Dup acks mean that packets have left the 1682 * network (they're now cached at the receiver) 1683 * so bump cwnd by the amount in the receiver 1684 * to keep a constant cwnd packets in the 1685 * network. 1686 */ 1687 if (!callout_active(tp->tt_rexmt) || 1688 th->th_ack != tp->snd_una) 1689 tp->t_dupacks = 0; 1690 else if (++tp->t_dupacks > tcprexmtthresh || 1691 (tcp_do_newreno && 1692 IN_FASTRECOVERY(tp))) { 1693 tp->snd_cwnd += tp->t_maxseg; 1694 (void) tcp_output(tp); 1695 goto drop; 1696 } else if (tp->t_dupacks == tcprexmtthresh) { 1697 tcp_seq onxt = tp->snd_nxt; 1698 u_int win; 1699 if (tcp_do_newreno && 1700 SEQ_LEQ(th->th_ack, 1701 tp->snd_recover)) { 1702 tp->t_dupacks = 0; 1703 break; 1704 } 1705 if (tcp_do_eifel_detect && 1706 (tp->t_flags & TF_RCVD_TSTMP)) { 1707 tcp_save_congestion_state(tp); 1708 tp->t_flags |= TF_FASTREXMT; 1709 } 1710 win = min(tp->snd_wnd, tp->snd_cwnd) / 1711 2 / tp->t_maxseg; 1712 if (win < 2) 1713 win = 2; 1714 tp->snd_ssthresh = win * tp->t_maxseg; 1715 ENTER_FASTRECOVERY(tp); 1716 tp->snd_recover = tp->snd_max; 1717 callout_stop(tp->tt_rexmt); 1718 tp->t_rtttime = 0; 1719 tp->snd_nxt = th->th_ack; 1720 tp->snd_cwnd = tp->t_maxseg; 1721 (void) tcp_output(tp); 1722 KASSERT(tp->snd_limited <= 2, 1723 ("tp->snd_limited too big")); 1724 tp->snd_cwnd = tp->snd_ssthresh + 1725 (tp->t_maxseg * 1726 (tp->t_dupacks - tp->snd_limited)); 1727 if (SEQ_GT(onxt, tp->snd_nxt)) 1728 tp->snd_nxt = onxt; 1729 goto drop; 1730 } else if (tcp_do_limitedtransmit) { 1731 u_long oldcwnd = tp->snd_cwnd; 1732 tcp_seq oldsndmax = tp->snd_max; 1733 u_int sent; 1734 KASSERT(tp->t_dupacks == 1 || 1735 tp->t_dupacks == 2, 1736 ("dupacks not 1 or 2")); 1737 if (tp->t_dupacks == 1) 1738 tp->snd_limited = 0; 1739 tp->snd_cwnd = 1740 (tp->snd_nxt - tp->snd_una) + 1741 (tp->t_dupacks - tp->snd_limited) * 1742 tp->t_maxseg; 1743 (void) tcp_output(tp); 1744 sent = tp->snd_max - oldsndmax; 1745 if (sent > tp->t_maxseg) { 1746 KASSERT(tp->snd_limited == 0 && 1747 tp->t_dupacks == 2, 1748 ("sent too much")); 1749 tp->snd_limited = 2; 1750 } else if (sent > 0) 1751 ++tp->snd_limited; 1752 tp->snd_cwnd = oldcwnd; 1753 goto drop; 1754 } 1755 } else 1756 tp->t_dupacks = 0; 1757 break; 1758 } 1759 1760 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1761 1762 /* 1763 * If the congestion window was inflated to account 1764 * for the other side's cached packets, retract it. 1765 */ 1766 if (tcp_do_newreno) { 1767 if (IN_FASTRECOVERY(tp)) { 1768 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1769 tcp_newreno_partial_ack(tp, th); 1770 } else { 1771 /* 1772 * Window inflation should have left us 1773 * with approximately snd_ssthresh 1774 * outstanding data. 1775 * But in case we would be inclined to 1776 * send a burst, better to do it via 1777 * the slow start mechanism. 1778 */ 1779 if (SEQ_GT(th->th_ack + 1780 tp->snd_ssthresh, 1781 tp->snd_max)) 1782 tp->snd_cwnd = tp->snd_max - 1783 th->th_ack + 1784 tp->t_maxseg; 1785 else 1786 tp->snd_cwnd = tp->snd_ssthresh; 1787 } 1788 } 1789 } else { 1790 if (tp->t_dupacks >= tcprexmtthresh && 1791 tp->snd_cwnd > tp->snd_ssthresh) 1792 tp->snd_cwnd = tp->snd_ssthresh; 1793 } 1794 tp->t_dupacks = 0; 1795 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1796 tcpstat.tcps_rcvacktoomuch++; 1797 goto dropafterack; 1798 } 1799 /* 1800 * If we reach this point, ACK is not a duplicate, 1801 * i.e., it ACKs something we sent. 1802 */ 1803 if (tp->t_flags & TF_NEEDSYN) { 1804 /* 1805 * T/TCP: Connection was half-synchronized, and our 1806 * SYN has been ACK'd (so connection is now fully 1807 * synchronized). Go to non-starred state, 1808 * increment snd_una for ACK of SYN, and check if 1809 * we can do window scaling. 1810 */ 1811 tp->t_flags &= ~TF_NEEDSYN; 1812 tp->snd_una++; 1813 /* Do window scaling? */ 1814 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1815 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1816 tp->snd_scale = tp->requested_s_scale; 1817 tp->rcv_scale = tp->request_r_scale; 1818 } 1819 } 1820 1821 process_ACK: 1822 acked = th->th_ack - tp->snd_una; 1823 tcpstat.tcps_rcvackpack++; 1824 tcpstat.tcps_rcvackbyte += acked; 1825 1826 /* 1827 * If we just performed our first retransmit, and the ACK 1828 * arrives within our recovery window, then it was a mistake 1829 * to do the retransmit in the first place. Recover our 1830 * original cwnd and ssthresh, and proceed to transmit where 1831 * we left off. 1832 */ 1833 if (tcp_do_eifel_detect && acked && 1834 (to.to_flags & TOF_TS) && to.to_tsecr && 1835 (tp->t_flags & TF_FIRSTACCACK)) { 1836 /* Eifel detection applicable. */ 1837 if (to.to_tsecr < tp->t_rexmtTS) { 1838 tcp_revert_congestion_state(tp); 1839 ++tcpstat.tcps_eifeldetected; 1840 } 1841 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1842 tcp_revert_congestion_state(tp); 1843 ++tcpstat.tcps_rttdetected; 1844 } 1845 1846 /* 1847 * If we have a timestamp reply, update smoothed 1848 * round trip time. If no timestamp is present but 1849 * transmit timer is running and timed sequence 1850 * number was acked, update smoothed round trip time. 1851 * Since we now have an rtt measurement, cancel the 1852 * timer backoff (cf., Phil Karn's retransmit alg.). 1853 * Recompute the initial retransmit timer. 1854 * 1855 * Some machines (certain windows boxes) send broken 1856 * timestamp replies during the SYN+ACK phase, ignore 1857 * timestamps of 0. 1858 */ 1859 if ((to.to_flags & TOF_TS) != 0 && 1860 to.to_tsecr) { 1861 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1862 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1863 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1864 } 1865 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1866 1867 /* 1868 * If all outstanding data is acked, stop retransmit 1869 * timer and remember to restart (more output or persist). 1870 * If there is more data to be acked, restart retransmit 1871 * timer, using current (possibly backed-off) value. 1872 */ 1873 if (th->th_ack == tp->snd_max) { 1874 callout_stop(tp->tt_rexmt); 1875 needoutput = 1; 1876 } else if (!callout_active(tp->tt_persist)) 1877 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1878 tcp_timer_rexmt, tp); 1879 1880 /* 1881 * If no data (only SYN) was ACK'd, 1882 * skip rest of ACK processing. 1883 */ 1884 if (acked == 0) 1885 goto step6; 1886 1887 /* Stop looking for an acceptable ACK since one was received. */ 1888 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT); 1889 1890 /* 1891 * When new data is acked, open the congestion window. 1892 * If the window gives us less than ssthresh packets 1893 * in flight, open exponentially (maxseg per packet). 1894 * Otherwise open linearly: maxseg per window 1895 * (maxseg^2 / cwnd per packet). 1896 */ 1897 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 1898 u_int cw = tp->snd_cwnd; 1899 u_int incr = tp->t_maxseg; 1900 if (cw > tp->snd_ssthresh) 1901 incr = incr * incr / cw; 1902 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1903 } 1904 if (acked > so->so_snd.sb_cc) { 1905 tp->snd_wnd -= so->so_snd.sb_cc; 1906 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1907 ourfinisacked = 1; 1908 } else { 1909 sbdrop(&so->so_snd, acked); 1910 tp->snd_wnd -= acked; 1911 ourfinisacked = 0; 1912 } 1913 sowwakeup(so); 1914 /* detect una wraparound */ 1915 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 1916 SEQ_GT(tp->snd_una, tp->snd_recover) && 1917 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1918 tp->snd_recover = th->th_ack - 1; 1919 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 1920 SEQ_GEQ(th->th_ack, tp->snd_recover)) 1921 EXIT_FASTRECOVERY(tp); 1922 tp->snd_una = th->th_ack; 1923 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1924 tp->snd_nxt = tp->snd_una; 1925 1926 switch (tp->t_state) { 1927 1928 /* 1929 * In FIN_WAIT_1 STATE in addition to the processing 1930 * for the ESTABLISHED state if our FIN is now acknowledged 1931 * then enter FIN_WAIT_2. 1932 */ 1933 case TCPS_FIN_WAIT_1: 1934 if (ourfinisacked) { 1935 /* 1936 * If we can't receive any more 1937 * data, then closing user can proceed. 1938 * Starting the timer is contrary to the 1939 * specification, but if we don't get a FIN 1940 * we'll hang forever. 1941 */ 1942 if (so->so_state & SS_CANTRCVMORE) { 1943 soisdisconnected(so); 1944 callout_reset(tp->tt_2msl, tcp_maxidle, 1945 tcp_timer_2msl, tp); 1946 } 1947 tp->t_state = TCPS_FIN_WAIT_2; 1948 } 1949 break; 1950 1951 /* 1952 * In CLOSING STATE in addition to the processing for 1953 * the ESTABLISHED state if the ACK acknowledges our FIN 1954 * then enter the TIME-WAIT state, otherwise ignore 1955 * the segment. 1956 */ 1957 case TCPS_CLOSING: 1958 if (ourfinisacked) { 1959 tp->t_state = TCPS_TIME_WAIT; 1960 tcp_canceltimers(tp); 1961 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1962 if (tp->cc_recv != 0 && 1963 (ticks - tp->t_starttime) < tcp_msl) 1964 callout_reset(tp->tt_2msl, 1965 tp->t_rxtcur * 1966 TCPTV_TWTRUNC, 1967 tcp_timer_2msl, tp); 1968 else 1969 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1970 tcp_timer_2msl, tp); 1971 soisdisconnected(so); 1972 } 1973 break; 1974 1975 /* 1976 * In LAST_ACK, we may still be waiting for data to drain 1977 * and/or to be acked, as well as for the ack of our FIN. 1978 * If our FIN is now acknowledged, delete the TCB, 1979 * enter the closed state and return. 1980 */ 1981 case TCPS_LAST_ACK: 1982 if (ourfinisacked) { 1983 tp = tcp_close(tp); 1984 goto drop; 1985 } 1986 break; 1987 1988 /* 1989 * In TIME_WAIT state the only thing that should arrive 1990 * is a retransmission of the remote FIN. Acknowledge 1991 * it and restart the finack timer. 1992 */ 1993 case TCPS_TIME_WAIT: 1994 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1995 tcp_timer_2msl, tp); 1996 goto dropafterack; 1997 } 1998 } 1999 2000 step6: 2001 /* 2002 * Update window information. 2003 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2004 */ 2005 if ((thflags & TH_ACK) && 2006 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2007 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2008 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2009 /* keep track of pure window updates */ 2010 if (tlen == 0 && 2011 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2012 tcpstat.tcps_rcvwinupd++; 2013 tp->snd_wnd = tiwin; 2014 tp->snd_wl1 = th->th_seq; 2015 tp->snd_wl2 = th->th_ack; 2016 if (tp->snd_wnd > tp->max_sndwnd) 2017 tp->max_sndwnd = tp->snd_wnd; 2018 needoutput = 1; 2019 } 2020 2021 /* 2022 * Process segments with URG. 2023 */ 2024 if ((thflags & TH_URG) && th->th_urp && 2025 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2026 /* 2027 * This is a kludge, but if we receive and accept 2028 * random urgent pointers, we'll crash in 2029 * soreceive. It's hard to imagine someone 2030 * actually wanting to send this much urgent data. 2031 */ 2032 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2033 th->th_urp = 0; /* XXX */ 2034 thflags &= ~TH_URG; /* XXX */ 2035 goto dodata; /* XXX */ 2036 } 2037 /* 2038 * If this segment advances the known urgent pointer, 2039 * then mark the data stream. This should not happen 2040 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2041 * a FIN has been received from the remote side. 2042 * In these states we ignore the URG. 2043 * 2044 * According to RFC961 (Assigned Protocols), 2045 * the urgent pointer points to the last octet 2046 * of urgent data. We continue, however, 2047 * to consider it to indicate the first octet 2048 * of data past the urgent section as the original 2049 * spec states (in one of two places). 2050 */ 2051 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2052 tp->rcv_up = th->th_seq + th->th_urp; 2053 so->so_oobmark = so->so_rcv.sb_cc + 2054 (tp->rcv_up - tp->rcv_nxt) - 1; 2055 if (so->so_oobmark == 0) 2056 so->so_state |= SS_RCVATMARK; 2057 sohasoutofband(so); 2058 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2059 } 2060 /* 2061 * Remove out of band data so doesn't get presented to user. 2062 * This can happen independent of advancing the URG pointer, 2063 * but if two URG's are pending at once, some out-of-band 2064 * data may creep in... ick. 2065 */ 2066 if (th->th_urp <= (u_long)tlen 2067 #ifdef SO_OOBINLINE 2068 && (so->so_options & SO_OOBINLINE) == 0 2069 #endif 2070 ) 2071 tcp_pulloutofband(so, th, m, 2072 drop_hdrlen); /* hdr drop is delayed */ 2073 } else { 2074 /* 2075 * If no out of band data is expected, 2076 * pull receive urgent pointer along 2077 * with the receive window. 2078 */ 2079 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2080 tp->rcv_up = tp->rcv_nxt; 2081 } 2082 dodata: /* XXX */ 2083 2084 /* 2085 * Process the segment text, merging it into the TCP sequencing queue, 2086 * and arranging for acknowledgment of receipt if necessary. 2087 * This process logically involves adjusting tp->rcv_wnd as data 2088 * is presented to the user (this happens in tcp_usrreq.c, 2089 * case PRU_RCVD). If a FIN has already been received on this 2090 * connection then we just ignore the text. 2091 */ 2092 if ((tlen || (thflags & TH_FIN)) && 2093 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2094 m_adj(m, drop_hdrlen); /* delayed header drop */ 2095 /* 2096 * Insert segment which includes th into TCP reassembly queue 2097 * with control block tp. Set thflags to whether reassembly now 2098 * includes a segment with FIN. This handles the common case 2099 * inline (segment is the next to be received on an established 2100 * connection, and the queue is empty), avoiding linkage into 2101 * and removal from the queue and repetition of various 2102 * conversions. 2103 * Set DELACK for segments received in order, but ack 2104 * immediately when segments are out of order (so 2105 * fast retransmit can work). 2106 */ 2107 if (th->th_seq == tp->rcv_nxt && 2108 LIST_EMPTY(&tp->t_segq) && 2109 TCPS_HAVEESTABLISHED(tp->t_state)) { 2110 if (DELAY_ACK(tp)) 2111 callout_reset(tp->tt_delack, tcp_delacktime, 2112 tcp_timer_delack, tp); 2113 else 2114 tp->t_flags |= TF_ACKNOW; 2115 tp->rcv_nxt += tlen; 2116 thflags = th->th_flags & TH_FIN; 2117 tcpstat.tcps_rcvpack++; 2118 tcpstat.tcps_rcvbyte += tlen; 2119 ND6_HINT(tp); 2120 if (so->so_state & SS_CANTRCVMORE) 2121 m_freem(m); 2122 else 2123 sbappend(&so->so_rcv, m); 2124 sorwakeup(so); 2125 } else { 2126 thflags = tcp_reass(tp, th, &tlen, m); 2127 tp->t_flags |= TF_ACKNOW; 2128 } 2129 2130 /* 2131 * Note the amount of data that peer has sent into 2132 * our window, in order to estimate the sender's 2133 * buffer size. 2134 */ 2135 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2136 } else { 2137 m_freem(m); 2138 thflags &= ~TH_FIN; 2139 } 2140 2141 /* 2142 * If FIN is received ACK the FIN and let the user know 2143 * that the connection is closing. 2144 */ 2145 if (thflags & TH_FIN) { 2146 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2147 socantrcvmore(so); 2148 /* 2149 * If connection is half-synchronized 2150 * (ie NEEDSYN flag on) then delay ACK, 2151 * so it may be piggybacked when SYN is sent. 2152 * Otherwise, since we received a FIN then no 2153 * more input can be expected, send ACK now. 2154 */ 2155 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2156 callout_reset(tp->tt_delack, tcp_delacktime, 2157 tcp_timer_delack, tp); 2158 else 2159 tp->t_flags |= TF_ACKNOW; 2160 tp->rcv_nxt++; 2161 } 2162 switch (tp->t_state) { 2163 2164 /* 2165 * In SYN_RECEIVED and ESTABLISHED STATES 2166 * enter the CLOSE_WAIT state. 2167 */ 2168 case TCPS_SYN_RECEIVED: 2169 tp->t_starttime = ticks; 2170 /*FALLTHROUGH*/ 2171 case TCPS_ESTABLISHED: 2172 tp->t_state = TCPS_CLOSE_WAIT; 2173 break; 2174 2175 /* 2176 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2177 * enter the CLOSING state. 2178 */ 2179 case TCPS_FIN_WAIT_1: 2180 tp->t_state = TCPS_CLOSING; 2181 break; 2182 2183 /* 2184 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2185 * starting the time-wait timer, turning off the other 2186 * standard timers. 2187 */ 2188 case TCPS_FIN_WAIT_2: 2189 tp->t_state = TCPS_TIME_WAIT; 2190 tcp_canceltimers(tp); 2191 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2192 if (tp->cc_recv != 0 && 2193 (ticks - tp->t_starttime) < tcp_msl) { 2194 callout_reset(tp->tt_2msl, 2195 tp->t_rxtcur * TCPTV_TWTRUNC, 2196 tcp_timer_2msl, tp); 2197 /* For transaction client, force ACK now. */ 2198 tp->t_flags |= TF_ACKNOW; 2199 } 2200 else 2201 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2202 tcp_timer_2msl, tp); 2203 soisdisconnected(so); 2204 break; 2205 2206 /* 2207 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2208 */ 2209 case TCPS_TIME_WAIT: 2210 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2211 tcp_timer_2msl, tp); 2212 break; 2213 } 2214 } 2215 #ifdef TCPDEBUG 2216 if (so->so_options & SO_DEBUG) 2217 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2218 &tcp_savetcp, 0); 2219 #endif 2220 2221 /* 2222 * Return any desired output. 2223 */ 2224 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2225 (void) tcp_output(tp); 2226 return; 2227 2228 dropafterack: 2229 /* 2230 * Generate an ACK dropping incoming segment if it occupies 2231 * sequence space, where the ACK reflects our state. 2232 * 2233 * We can now skip the test for the RST flag since all 2234 * paths to this code happen after packets containing 2235 * RST have been dropped. 2236 * 2237 * In the SYN-RECEIVED state, don't send an ACK unless the 2238 * segment we received passes the SYN-RECEIVED ACK test. 2239 * If it fails send a RST. This breaks the loop in the 2240 * "LAND" DoS attack, and also prevents an ACK storm 2241 * between two listening ports that have been sent forged 2242 * SYN segments, each with the source address of the other. 2243 */ 2244 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2245 (SEQ_GT(tp->snd_una, th->th_ack) || 2246 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2247 rstreason = BANDLIM_RST_OPENPORT; 2248 goto dropwithreset; 2249 } 2250 #ifdef TCPDEBUG 2251 if (so->so_options & SO_DEBUG) 2252 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2253 &tcp_savetcp, 0); 2254 #endif 2255 m_freem(m); 2256 tp->t_flags |= TF_ACKNOW; 2257 (void) tcp_output(tp); 2258 return; 2259 2260 dropwithreset: 2261 /* 2262 * Generate a RST, dropping incoming segment. 2263 * Make ACK acceptable to originator of segment. 2264 * Don't bother to respond if destination was broadcast/multicast. 2265 */ 2266 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2267 goto drop; 2268 if (isipv6) { 2269 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2270 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2271 goto drop; 2272 } else { 2273 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2274 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2275 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2276 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2277 goto drop; 2278 } 2279 /* IPv6 anycast check is done at tcp6_input() */ 2280 2281 /* 2282 * Perform bandwidth limiting. 2283 */ 2284 #ifdef ICMP_BANDLIM 2285 if (badport_bandlim(rstreason) < 0) 2286 goto drop; 2287 #endif 2288 2289 #ifdef TCPDEBUG 2290 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2291 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2292 &tcp_savetcp, 0); 2293 #endif 2294 if (thflags & TH_ACK) 2295 /* mtod() below is safe as long as hdr dropping is delayed */ 2296 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2297 TH_RST); 2298 else { 2299 if (thflags & TH_SYN) 2300 tlen++; 2301 /* mtod() below is safe as long as hdr dropping is delayed */ 2302 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2303 (tcp_seq)0, TH_RST|TH_ACK); 2304 } 2305 return; 2306 2307 drop: 2308 /* 2309 * Drop space held by incoming segment and return. 2310 */ 2311 #ifdef TCPDEBUG 2312 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2313 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2314 &tcp_savetcp, 0); 2315 #endif 2316 m_freem(m); 2317 return; 2318 } 2319 2320 /* 2321 * Parse TCP options and place in tcpopt. 2322 */ 2323 static void 2324 tcp_dooptions(to, cp, cnt, is_syn) 2325 struct tcpopt *to; 2326 u_char *cp; 2327 int cnt; 2328 { 2329 int opt, optlen; 2330 2331 to->to_flags = 0; 2332 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2333 opt = cp[0]; 2334 if (opt == TCPOPT_EOL) 2335 break; 2336 if (opt == TCPOPT_NOP) 2337 optlen = 1; 2338 else { 2339 if (cnt < 2) 2340 break; 2341 optlen = cp[1]; 2342 if (optlen < 2 || optlen > cnt) 2343 break; 2344 } 2345 switch (opt) { 2346 case TCPOPT_MAXSEG: 2347 if (optlen != TCPOLEN_MAXSEG) 2348 continue; 2349 if (!is_syn) 2350 continue; 2351 to->to_flags |= TOF_MSS; 2352 bcopy((char *)cp + 2, 2353 (char *)&to->to_mss, sizeof(to->to_mss)); 2354 to->to_mss = ntohs(to->to_mss); 2355 break; 2356 case TCPOPT_WINDOW: 2357 if (optlen != TCPOLEN_WINDOW) 2358 continue; 2359 if (! is_syn) 2360 continue; 2361 to->to_flags |= TOF_SCALE; 2362 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2363 break; 2364 case TCPOPT_TIMESTAMP: 2365 if (optlen != TCPOLEN_TIMESTAMP) 2366 continue; 2367 to->to_flags |= TOF_TS; 2368 bcopy((char *)cp + 2, 2369 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2370 to->to_tsval = ntohl(to->to_tsval); 2371 bcopy((char *)cp + 6, 2372 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2373 to->to_tsecr = ntohl(to->to_tsecr); 2374 break; 2375 case TCPOPT_CC: 2376 if (optlen != TCPOLEN_CC) 2377 continue; 2378 to->to_flags |= TOF_CC; 2379 bcopy((char *)cp + 2, 2380 (char *)&to->to_cc, sizeof(to->to_cc)); 2381 to->to_cc = ntohl(to->to_cc); 2382 break; 2383 case TCPOPT_CCNEW: 2384 if (optlen != TCPOLEN_CC) 2385 continue; 2386 if (!is_syn) 2387 continue; 2388 to->to_flags |= TOF_CCNEW; 2389 bcopy((char *)cp + 2, 2390 (char *)&to->to_cc, sizeof(to->to_cc)); 2391 to->to_cc = ntohl(to->to_cc); 2392 break; 2393 case TCPOPT_CCECHO: 2394 if (optlen != TCPOLEN_CC) 2395 continue; 2396 if (!is_syn) 2397 continue; 2398 to->to_flags |= TOF_CCECHO; 2399 bcopy((char *)cp + 2, 2400 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2401 to->to_ccecho = ntohl(to->to_ccecho); 2402 break; 2403 default: 2404 continue; 2405 } 2406 } 2407 } 2408 2409 /* 2410 * Pull out of band byte out of a segment so 2411 * it doesn't appear in the user's data queue. 2412 * It is still reflected in the segment length for 2413 * sequencing purposes. 2414 */ 2415 static void 2416 tcp_pulloutofband(so, th, m, off) 2417 struct socket *so; 2418 struct tcphdr *th; 2419 struct mbuf *m; 2420 int off; /* delayed to be droped hdrlen */ 2421 { 2422 int cnt = off + th->th_urp - 1; 2423 2424 while (cnt >= 0) { 2425 if (m->m_len > cnt) { 2426 char *cp = mtod(m, caddr_t) + cnt; 2427 struct tcpcb *tp = sototcpcb(so); 2428 2429 tp->t_iobc = *cp; 2430 tp->t_oobflags |= TCPOOB_HAVEDATA; 2431 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2432 m->m_len--; 2433 if (m->m_flags & M_PKTHDR) 2434 m->m_pkthdr.len--; 2435 return; 2436 } 2437 cnt -= m->m_len; 2438 m = m->m_next; 2439 if (m == 0) 2440 break; 2441 } 2442 panic("tcp_pulloutofband"); 2443 } 2444 2445 /* 2446 * Collect new round-trip time estimate 2447 * and update averages and current timeout. 2448 */ 2449 static void 2450 tcp_xmit_timer(tp, rtt) 2451 struct tcpcb *tp; 2452 int rtt; 2453 { 2454 int delta; 2455 2456 tcpstat.tcps_rttupdated++; 2457 tp->t_rttupdated++; 2458 if (tp->t_srtt != 0) { 2459 /* 2460 * srtt is stored as fixed point with 5 bits after the 2461 * binary point (i.e., scaled by 8). The following magic 2462 * is equivalent to the smoothing algorithm in rfc793 with 2463 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2464 * point). Adjust rtt to origin 0. 2465 */ 2466 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2467 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2468 2469 if ((tp->t_srtt += delta) <= 0) 2470 tp->t_srtt = 1; 2471 2472 /* 2473 * We accumulate a smoothed rtt variance (actually, a 2474 * smoothed mean difference), then set the retransmit 2475 * timer to smoothed rtt + 4 times the smoothed variance. 2476 * rttvar is stored as fixed point with 4 bits after the 2477 * binary point (scaled by 16). The following is 2478 * equivalent to rfc793 smoothing with an alpha of .75 2479 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2480 * rfc793's wired-in beta. 2481 */ 2482 if (delta < 0) 2483 delta = -delta; 2484 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2485 if ((tp->t_rttvar += delta) <= 0) 2486 tp->t_rttvar = 1; 2487 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2488 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2489 } else { 2490 /* 2491 * No rtt measurement yet - use the unsmoothed rtt. 2492 * Set the variance to half the rtt (so our first 2493 * retransmit happens at 3*rtt). 2494 */ 2495 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2496 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2497 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2498 } 2499 tp->t_rtttime = 0; 2500 tp->t_rxtshift = 0; 2501 2502 /* 2503 * the retransmit should happen at rtt + 4 * rttvar. 2504 * Because of the way we do the smoothing, srtt and rttvar 2505 * will each average +1/2 tick of bias. When we compute 2506 * the retransmit timer, we want 1/2 tick of rounding and 2507 * 1 extra tick because of +-1/2 tick uncertainty in the 2508 * firing of the timer. The bias will give us exactly the 2509 * 1.5 tick we need. But, because the bias is 2510 * statistical, we have to test that we don't drop below 2511 * the minimum feasible timer (which is 2 ticks). 2512 */ 2513 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2514 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2515 2516 /* 2517 * We received an ack for a packet that wasn't retransmitted; 2518 * it is probably safe to discard any error indications we've 2519 * received recently. This isn't quite right, but close enough 2520 * for now (a route might have failed after we sent a segment, 2521 * and the return path might not be symmetrical). 2522 */ 2523 tp->t_softerror = 0; 2524 } 2525 2526 /* 2527 * Determine a reasonable value for maxseg size. 2528 * If the route is known, check route for mtu. 2529 * If none, use an mss that can be handled on the outgoing 2530 * interface without forcing IP to fragment; if bigger than 2531 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2532 * to utilize large mbufs. If no route is found, route has no mtu, 2533 * or the destination isn't local, use a default, hopefully conservative 2534 * size (usually 512 or the default IP max size, but no more than the mtu 2535 * of the interface), as we can't discover anything about intervening 2536 * gateways or networks. We also initialize the congestion/slow start 2537 * window to be a single segment if the destination isn't local. 2538 * While looking at the routing entry, we also initialize other path-dependent 2539 * parameters from pre-set or cached values in the routing entry. 2540 * 2541 * Also take into account the space needed for options that we 2542 * send regularly. Make maxseg shorter by that amount to assure 2543 * that we can send maxseg amount of data even when the options 2544 * are present. Store the upper limit of the length of options plus 2545 * data in maxopd. 2546 * 2547 * NOTE that this routine is only called when we process an incoming 2548 * segment, for outgoing segments only tcp_mssopt is called. 2549 * 2550 * In case of T/TCP, we call this routine during implicit connection 2551 * setup as well (offer = -1), to initialize maxseg from the cached 2552 * MSS of our peer. 2553 */ 2554 void 2555 tcp_mss(tp, offer) 2556 struct tcpcb *tp; 2557 int offer; 2558 { 2559 struct rtentry *rt; 2560 struct ifnet *ifp; 2561 int rtt, mss; 2562 u_long bufsize; 2563 struct inpcb *inp = tp->t_inpcb; 2564 struct socket *so; 2565 struct rmxp_tao *taop; 2566 int origoffer = offer; 2567 #ifdef INET6 2568 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2569 size_t min_protoh = isipv6 ? 2570 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2571 sizeof(struct tcpiphdr); 2572 #else 2573 const int isipv6 = 0; 2574 const size_t min_protoh = sizeof(struct tcpiphdr); 2575 #endif 2576 2577 if (isipv6) 2578 rt = tcp_rtlookup6(&inp->inp_inc); 2579 else 2580 rt = tcp_rtlookup(&inp->inp_inc); 2581 if (rt == NULL) { 2582 tp->t_maxopd = tp->t_maxseg = 2583 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2584 return; 2585 } 2586 ifp = rt->rt_ifp; 2587 so = inp->inp_socket; 2588 2589 taop = rmx_taop(rt->rt_rmx); 2590 /* 2591 * Offer == -1 means that we didn't receive SYN yet, 2592 * use cached value in that case; 2593 */ 2594 if (offer == -1) 2595 offer = taop->tao_mssopt; 2596 /* 2597 * Offer == 0 means that there was no MSS on the SYN segment, 2598 * in this case we use tcp_mssdflt. 2599 */ 2600 if (offer == 0) 2601 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2602 else 2603 /* 2604 * Sanity check: make sure that maxopd will be large 2605 * enough to allow some data on segments even is the 2606 * all the option space is used (40bytes). Otherwise 2607 * funny things may happen in tcp_output. 2608 */ 2609 offer = max(offer, 64); 2610 taop->tao_mssopt = offer; 2611 2612 /* 2613 * While we're here, check if there's an initial rtt 2614 * or rttvar. Convert from the route-table units 2615 * to scaled multiples of the slow timeout timer. 2616 */ 2617 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2618 /* 2619 * XXX the lock bit for RTT indicates that the value 2620 * is also a minimum value; this is subject to time. 2621 */ 2622 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2623 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2624 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2625 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2626 tcpstat.tcps_usedrtt++; 2627 if (rt->rt_rmx.rmx_rttvar) { 2628 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2629 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2630 tcpstat.tcps_usedrttvar++; 2631 } else { 2632 /* default variation is +- 1 rtt */ 2633 tp->t_rttvar = 2634 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2635 } 2636 TCPT_RANGESET(tp->t_rxtcur, 2637 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2638 tp->t_rttmin, TCPTV_REXMTMAX); 2639 } 2640 /* 2641 * if there's an mtu associated with the route, use it 2642 * else, use the link mtu. 2643 */ 2644 if (rt->rt_rmx.rmx_mtu) 2645 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2646 else { 2647 if (isipv6) { 2648 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2649 min_protoh; 2650 if (!in6_localaddr(&inp->in6p_faddr)) 2651 mss = min(mss, tcp_v6mssdflt); 2652 } else { 2653 mss = ifp->if_mtu - min_protoh; 2654 if (!in_localaddr(inp->inp_faddr)) 2655 mss = min(mss, tcp_mssdflt); 2656 } 2657 } 2658 mss = min(mss, offer); 2659 /* 2660 * maxopd stores the maximum length of data AND options 2661 * in a segment; maxseg is the amount of data in a normal 2662 * segment. We need to store this value (maxopd) apart 2663 * from maxseg, because now every segment carries options 2664 * and thus we normally have somewhat less data in segments. 2665 */ 2666 tp->t_maxopd = mss; 2667 2668 /* 2669 * In case of T/TCP, origoffer==-1 indicates, that no segments 2670 * were received yet. In this case we just guess, otherwise 2671 * we do the same as before T/TCP. 2672 */ 2673 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2674 (origoffer == -1 || 2675 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2676 mss -= TCPOLEN_TSTAMP_APPA; 2677 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2678 (origoffer == -1 || 2679 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2680 mss -= TCPOLEN_CC_APPA; 2681 2682 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2683 if (mss > MCLBYTES) 2684 mss &= ~(MCLBYTES-1); 2685 #else 2686 if (mss > MCLBYTES) 2687 mss = mss / MCLBYTES * MCLBYTES; 2688 #endif 2689 /* 2690 * If there's a pipesize, change the socket buffer 2691 * to that size. Make the socket buffers an integral 2692 * number of mss units; if the mss is larger than 2693 * the socket buffer, decrease the mss. 2694 */ 2695 #ifdef RTV_SPIPE 2696 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2697 #endif 2698 bufsize = so->so_snd.sb_hiwat; 2699 if (bufsize < mss) 2700 mss = bufsize; 2701 else { 2702 bufsize = roundup(bufsize, mss); 2703 if (bufsize > sb_max) 2704 bufsize = sb_max; 2705 if (bufsize > so->so_snd.sb_hiwat) 2706 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2707 } 2708 tp->t_maxseg = mss; 2709 2710 #ifdef RTV_RPIPE 2711 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2712 #endif 2713 bufsize = so->so_rcv.sb_hiwat; 2714 if (bufsize > mss) { 2715 bufsize = roundup(bufsize, mss); 2716 if (bufsize > sb_max) 2717 bufsize = sb_max; 2718 if (bufsize > so->so_rcv.sb_hiwat) 2719 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2720 } 2721 2722 /* 2723 * Set the slow-start flight size depending on whether this 2724 * is a local network or not. 2725 */ 2726 if (tcp_do_rfc3390) 2727 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2728 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2729 (!isipv6 && in_localaddr(inp->inp_faddr))) 2730 tp->snd_cwnd = mss * ss_fltsz_local; 2731 else 2732 tp->snd_cwnd = mss * ss_fltsz; 2733 2734 if (rt->rt_rmx.rmx_ssthresh) { 2735 /* 2736 * There's some sort of gateway or interface 2737 * buffer limit on the path. Use this to set 2738 * the slow start threshhold, but set the 2739 * threshold to no less than 2*mss. 2740 */ 2741 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2742 tcpstat.tcps_usedssthresh++; 2743 } 2744 } 2745 2746 /* 2747 * Determine the MSS option to send on an outgoing SYN. 2748 */ 2749 int 2750 tcp_mssopt(tp) 2751 struct tcpcb *tp; 2752 { 2753 struct rtentry *rt; 2754 #ifdef INET6 2755 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2756 int min_protoh = isipv6 ? 2757 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2758 sizeof(struct tcpiphdr); 2759 #else 2760 const int isipv6 = 0; 2761 const size_t min_protoh = sizeof(struct tcpiphdr); 2762 #endif 2763 2764 if (isipv6) 2765 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2766 else 2767 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2768 if (rt == NULL) 2769 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2770 2771 return (rt->rt_ifp->if_mtu - min_protoh); 2772 } 2773 2774 2775 /* 2776 * When a partial ack arrives, force the retransmission of the 2777 * next unacknowledged segment. Do not clear tp->t_dupacks. 2778 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2779 * be started again. 2780 */ 2781 static void 2782 tcp_newreno_partial_ack(tp, th) 2783 struct tcpcb *tp; 2784 struct tcphdr *th; 2785 { 2786 tcp_seq onxt = tp->snd_nxt; 2787 u_long ocwnd = tp->snd_cwnd; 2788 2789 callout_stop(tp->tt_rexmt); 2790 tp->t_rtttime = 0; 2791 tp->snd_nxt = th->th_ack; 2792 /* 2793 * Set snd_cwnd to one segment beyond acknowledged offset 2794 * (tp->snd_una has not yet been updated when this function is called.) 2795 */ 2796 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2797 tp->t_flags |= TF_ACKNOW; 2798 (void) tcp_output(tp); 2799 tp->snd_cwnd = ocwnd; 2800 if (SEQ_GT(onxt, tp->snd_nxt)) 2801 tp->snd_nxt = onxt; 2802 /* 2803 * Partial window deflation. Relies on fact that tp->snd_una 2804 * not updated yet. 2805 */ 2806 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2807 } 2808