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 1.10 2003/09/02 10:04:47 hsu 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 int useTS; /* use timestamps in Eifel detection */ 367 struct ip6_hdr *ip6 = NULL; 368 #ifdef INET6 369 int isipv6; 370 #else 371 const int isipv6 = 0; 372 #endif 373 #ifdef TCPDEBUG 374 short ostate = 0; 375 #endif 376 377 /* Grab info from MT_TAG mbufs prepended to the chain. */ 378 for (;m && m->m_type == MT_TAG; m = m->m_next) { 379 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 380 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 381 } 382 #ifdef INET6 383 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 384 #endif 385 bzero((char *)&to, sizeof(to)); 386 387 tcpstat.tcps_rcvtotal++; 388 389 if (isipv6) { 390 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 391 ip6 = mtod(m, struct ip6_hdr *); 392 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 393 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 394 tcpstat.tcps_rcvbadsum++; 395 goto drop; 396 } 397 th = (struct tcphdr *)((caddr_t)ip6 + off0); 398 399 /* 400 * Be proactive about unspecified IPv6 address in source. 401 * As we use all-zero to indicate unbounded/unconnected pcb, 402 * unspecified IPv6 address can be used to confuse us. 403 * 404 * Note that packets with unspecified IPv6 destination is 405 * already dropped in ip6_input. 406 */ 407 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 408 /* XXX stat */ 409 goto drop; 410 } 411 } else { 412 /* 413 * Get IP and TCP header together in first mbuf. 414 * Note: IP leaves IP header in first mbuf. 415 */ 416 if (off0 > sizeof(struct ip)) { 417 ip_stripoptions(m, (struct mbuf *)0); 418 off0 = sizeof(struct ip); 419 } 420 if (m->m_len < sizeof(struct tcpiphdr)) { 421 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) { 422 tcpstat.tcps_rcvshort++; 423 return; 424 } 425 } 426 ip = mtod(m, struct ip *); 427 ipov = (struct ipovly *)ip; 428 th = (struct tcphdr *)((caddr_t)ip + off0); 429 tlen = ip->ip_len; 430 431 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 432 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 433 th->th_sum = m->m_pkthdr.csum_data; 434 else 435 th->th_sum = in_pseudo(ip->ip_src.s_addr, 436 ip->ip_dst.s_addr, 437 htonl(m->m_pkthdr.csum_data + 438 ip->ip_len + 439 IPPROTO_TCP)); 440 th->th_sum ^= 0xffff; 441 } else { 442 /* 443 * Checksum extended TCP header and data. 444 */ 445 len = sizeof(struct ip) + tlen; 446 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 447 ipov->ih_len = (u_short)tlen; 448 ipov->ih_len = htons(ipov->ih_len); 449 th->th_sum = in_cksum(m, len); 450 } 451 if (th->th_sum) { 452 tcpstat.tcps_rcvbadsum++; 453 goto drop; 454 } 455 #ifdef INET6 456 /* Re-initialization for later version check */ 457 ip->ip_v = IPVERSION; 458 #endif 459 } 460 461 /* 462 * Check that TCP offset makes sense, 463 * pull out TCP options and adjust length. XXX 464 */ 465 off = th->th_off << 2; 466 if (off < sizeof(struct tcphdr) || off > tlen) { 467 tcpstat.tcps_rcvbadoff++; 468 goto drop; 469 } 470 tlen -= off; /* tlen is used instead of ti->ti_len */ 471 if (off > sizeof(struct tcphdr)) { 472 if (isipv6) { 473 IP6_EXTHDR_CHECK(m, off0, off, ); 474 ip6 = mtod(m, struct ip6_hdr *); 475 th = (struct tcphdr *)((caddr_t)ip6 + off0); 476 } else { 477 if (m->m_len < sizeof(struct ip) + off) { 478 if ((m = m_pullup(m, sizeof(struct ip) + off)) 479 == 0) { 480 tcpstat.tcps_rcvshort++; 481 return; 482 } 483 ip = mtod(m, struct ip *); 484 ipov = (struct ipovly *)ip; 485 th = (struct tcphdr *)((caddr_t)ip + off0); 486 } 487 } 488 optlen = off - sizeof(struct tcphdr); 489 optp = (u_char *)(th + 1); 490 } 491 thflags = th->th_flags; 492 493 #ifdef TCP_DROP_SYNFIN 494 /* 495 * If the drop_synfin option is enabled, drop all packets with 496 * both the SYN and FIN bits set. This prevents e.g. nmap from 497 * identifying the TCP/IP stack. 498 * 499 * This is a violation of the TCP specification. 500 */ 501 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 502 goto drop; 503 #endif 504 505 /* 506 * Convert TCP protocol specific fields to host format. 507 */ 508 th->th_seq = ntohl(th->th_seq); 509 th->th_ack = ntohl(th->th_ack); 510 th->th_win = ntohs(th->th_win); 511 th->th_urp = ntohs(th->th_urp); 512 513 /* 514 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 515 * until after ip6_savecontrol() is called and before other functions 516 * which don't want those proto headers. 517 * Because ip6_savecontrol() is going to parse the mbuf to 518 * search for data to be passed up to user-land, it wants mbuf 519 * parameters to be unchanged. 520 * XXX: the call of ip6_savecontrol() has been obsoleted based on 521 * latest version of the advanced API (20020110). 522 */ 523 drop_hdrlen = off0 + off; 524 525 /* 526 * Locate pcb for segment. 527 */ 528 findpcb: 529 /* IPFIREWALL_FORWARD section */ 530 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 531 /* 532 * Transparently forwarded. Pretend to be the destination. 533 * already got one like this? 534 */ 535 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 536 ip->ip_dst, th->th_dport, 537 0, m->m_pkthdr.rcvif); 538 if (!inp) { 539 /* It's new. Try find the ambushing socket. */ 540 inp = in_pcblookup_hash(&tcbinfo, 541 ip->ip_src, th->th_sport, 542 next_hop->sin_addr, 543 next_hop->sin_port ? 544 ntohs(next_hop->sin_port) : 545 th->th_dport, 546 1, m->m_pkthdr.rcvif); 547 } 548 } else { 549 if (isipv6) 550 inp = in6_pcblookup_hash(&tcbinfo, 551 &ip6->ip6_src, th->th_sport, 552 &ip6->ip6_dst, th->th_dport, 553 1, m->m_pkthdr.rcvif); 554 else 555 inp = in_pcblookup_hash(&tcbinfo, 556 ip->ip_src, th->th_sport, 557 ip->ip_dst, th->th_dport, 558 1, m->m_pkthdr.rcvif); 559 } 560 561 #ifdef IPSEC 562 if (isipv6) { 563 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 564 ipsec6stat.in_polvio++; 565 goto drop; 566 } 567 } else { 568 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 569 ipsecstat.in_polvio++; 570 goto drop; 571 } 572 } 573 #endif 574 #ifdef FAST_IPSEC 575 if (isipv6) { 576 if (inp != NULL && ipsec6_in_reject(m, inp)) { 577 goto drop; 578 } 579 } else { 580 if (inp != NULL && ipsec4_in_reject(m, inp)) { 581 goto drop; 582 } 583 } 584 #endif 585 586 /* 587 * If the state is CLOSED (i.e., TCB does not exist) then 588 * all data in the incoming segment is discarded. 589 * If the TCB exists but is in CLOSED state, it is embryonic, 590 * but should either do a listen or a connect soon. 591 */ 592 if (inp == NULL) { 593 if (log_in_vain) { 594 #ifdef INET6 595 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 596 #else 597 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 598 #endif 599 if (isipv6) { 600 strcpy(dbuf, "["); 601 strcpy(sbuf, "["); 602 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 603 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 604 strcat(dbuf, "]"); 605 strcat(sbuf, "]"); 606 } else { 607 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 608 strcpy(sbuf, inet_ntoa(ip->ip_src)); 609 } 610 switch (log_in_vain) { 611 case 1: 612 if ((thflags & TH_SYN) == 0) 613 break; 614 case 2: 615 log(LOG_INFO, 616 "Connection attempt to TCP %s:%d " 617 "from %s:%d flags:0x%02x\n", 618 dbuf, ntohs(th->th_dport), sbuf, 619 ntohs(th->th_sport), thflags); 620 break; 621 default: 622 break; 623 } 624 } 625 if (blackhole) { 626 switch (blackhole) { 627 case 1: 628 if (thflags & TH_SYN) 629 goto drop; 630 break; 631 case 2: 632 goto drop; 633 default: 634 goto drop; 635 } 636 } 637 rstreason = BANDLIM_RST_CLOSEDPORT; 638 goto dropwithreset; 639 } 640 tp = intotcpcb(inp); 641 if (tp == NULL) { 642 rstreason = BANDLIM_RST_CLOSEDPORT; 643 goto dropwithreset; 644 } 645 if (tp->t_state == TCPS_CLOSED) 646 goto drop; 647 648 /* Unscale the window into a 32-bit value. */ 649 if ((thflags & TH_SYN) == 0) 650 tiwin = th->th_win << tp->snd_scale; 651 else 652 tiwin = th->th_win; 653 654 so = inp->inp_socket; 655 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 656 struct in_conninfo inc; 657 #ifdef TCPDEBUG 658 if (so->so_options & SO_DEBUG) { 659 ostate = tp->t_state; 660 if (isipv6) 661 bcopy((char *)ip6, (char *)tcp_saveipgen, 662 sizeof(*ip6)); 663 else 664 bcopy((char *)ip, (char *)tcp_saveipgen, 665 sizeof(*ip)); 666 tcp_savetcp = *th; 667 } 668 #endif 669 /* skip if this isn't a listen socket */ 670 if ((so->so_options & SO_ACCEPTCONN) == 0) 671 goto after_listen; 672 #ifdef INET6 673 inc.inc_isipv6 = isipv6; 674 #endif 675 if (isipv6) { 676 inc.inc6_faddr = ip6->ip6_src; 677 inc.inc6_laddr = ip6->ip6_dst; 678 inc.inc6_route.ro_rt = NULL; /* XXX */ 679 } else { 680 inc.inc_faddr = ip->ip_src; 681 inc.inc_laddr = ip->ip_dst; 682 inc.inc_route.ro_rt = NULL; /* XXX */ 683 } 684 inc.inc_fport = th->th_sport; 685 inc.inc_lport = th->th_dport; 686 687 /* 688 * If the state is LISTEN then ignore segment if it contains 689 * a RST. If the segment contains an ACK then it is bad and 690 * send a RST. If it does not contain a SYN then it is not 691 * interesting; drop it. 692 * 693 * If the state is SYN_RECEIVED (syncache) and seg contains 694 * an ACK, but not for our SYN/ACK, send a RST. If the seg 695 * contains a RST, check the sequence number to see if it 696 * is a valid reset segment. 697 */ 698 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 699 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 700 if (!syncache_expand(&inc, th, &so, m)) { 701 /* 702 * No syncache entry, or ACK was not 703 * for our SYN/ACK. Send a RST. 704 */ 705 tcpstat.tcps_badsyn++; 706 rstreason = BANDLIM_RST_OPENPORT; 707 goto dropwithreset; 708 } 709 if (so == NULL) 710 /* 711 * Could not complete 3-way handshake, 712 * connection is being closed down, and 713 * syncache will free mbuf. 714 */ 715 return; 716 /* 717 * Socket is created in state SYN_RECEIVED. 718 * Continue processing segment. 719 */ 720 inp = sotoinpcb(so); 721 tp = intotcpcb(inp); 722 /* 723 * This is what would have happened in 724 * tcp_output() when the SYN,ACK was sent. 725 */ 726 tp->snd_up = tp->snd_una; 727 tp->snd_max = tp->snd_nxt = tp->iss + 1; 728 tp->last_ack_sent = tp->rcv_nxt; 729 /* 730 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 731 * until the _second_ ACK is received: 732 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 733 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 734 * move to ESTAB, set snd_wnd to tiwin. 735 */ 736 tp->snd_wnd = tiwin; /* unscaled */ 737 goto after_listen; 738 } 739 if (thflags & TH_RST) { 740 syncache_chkrst(&inc, th); 741 goto drop; 742 } 743 if (thflags & TH_ACK) { 744 syncache_badack(&inc); 745 tcpstat.tcps_badsyn++; 746 rstreason = BANDLIM_RST_OPENPORT; 747 goto dropwithreset; 748 } 749 goto drop; 750 } 751 752 /* 753 * Segment's flags are (SYN) or (SYN|FIN). 754 */ 755 #ifdef INET6 756 /* 757 * If deprecated address is forbidden, 758 * we do not accept SYN to deprecated interface 759 * address to prevent any new inbound connection from 760 * getting established. 761 * When we do not accept SYN, we send a TCP RST, 762 * with deprecated source address (instead of dropping 763 * it). We compromise it as it is much better for peer 764 * to send a RST, and RST will be the final packet 765 * for the exchange. 766 * 767 * If we do not forbid deprecated addresses, we accept 768 * the SYN packet. RFC2462 does not suggest dropping 769 * SYN in this case. 770 * If we decipher RFC2462 5.5.4, it says like this: 771 * 1. use of deprecated addr with existing 772 * communication is okay - "SHOULD continue to be 773 * used" 774 * 2. use of it with new communication: 775 * (2a) "SHOULD NOT be used if alternate address 776 * with sufficient scope is available" 777 * (2b) nothing mentioned otherwise. 778 * Here we fall into (2b) case as we have no choice in 779 * our source address selection - we must obey the peer. 780 * 781 * The wording in RFC2462 is confusing, and there are 782 * multiple description text for deprecated address 783 * handling - worse, they are not exactly the same. 784 * I believe 5.5.4 is the best one, so we follow 5.5.4. 785 */ 786 if (isipv6 && !ip6_use_deprecated) { 787 struct in6_ifaddr *ia6; 788 789 if ((ia6 = ip6_getdstifaddr(m)) && 790 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 791 tp = NULL; 792 rstreason = BANDLIM_RST_OPENPORT; 793 goto dropwithreset; 794 } 795 } 796 #endif 797 /* 798 * If it is from this socket, drop it, it must be forged. 799 * Don't bother responding if the destination was a broadcast. 800 */ 801 if (th->th_dport == th->th_sport) { 802 if (isipv6) { 803 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 804 &ip6->ip6_src)) 805 goto drop; 806 } else { 807 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 808 goto drop; 809 } 810 } 811 /* 812 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 813 * 814 * Note that it is quite possible to receive unicast 815 * link-layer packets with a broadcast IP address. Use 816 * in_broadcast() to find them. 817 */ 818 if (m->m_flags & (M_BCAST|M_MCAST)) 819 goto drop; 820 if (isipv6) { 821 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 822 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 823 goto drop; 824 } else { 825 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 826 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 827 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 828 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 829 goto drop; 830 } 831 /* 832 * SYN appears to be valid; create compressed TCP state 833 * for syncache, or perform t/tcp connection. 834 */ 835 if (so->so_qlen <= so->so_qlimit) { 836 tcp_dooptions(&to, optp, optlen, 1); 837 if (!syncache_add(&inc, &to, th, &so, m)) 838 goto drop; 839 if (so == NULL) 840 /* 841 * Entry added to syncache, mbuf used to 842 * send SYN,ACK packet. 843 */ 844 return; 845 /* 846 * Segment passed TAO tests. 847 */ 848 inp = sotoinpcb(so); 849 tp = intotcpcb(inp); 850 tp->snd_wnd = tiwin; 851 tp->t_starttime = ticks; 852 tp->t_state = TCPS_ESTABLISHED; 853 854 /* 855 * If there is a FIN, or if there is data and the 856 * connection is local, then delay SYN,ACK(SYN) in 857 * the hope of piggy-backing it on a response 858 * segment. Otherwise must send ACK now in case 859 * the other side is slow starting. 860 */ 861 if (DELAY_ACK(tp) && 862 ((thflags & TH_FIN) || 863 (tlen != 0 && 864 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 865 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 866 callout_reset(tp->tt_delack, tcp_delacktime, 867 tcp_timer_delack, tp); 868 tp->t_flags |= TF_NEEDSYN; 869 } else 870 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 871 872 tcpstat.tcps_connects++; 873 soisconnected(so); 874 goto trimthenstep6; 875 } 876 goto drop; 877 } 878 after_listen: 879 880 /* XXX temp debugging */ 881 /* should not happen - syncache should pick up these connections */ 882 if (tp->t_state == TCPS_LISTEN) 883 panic("tcp_input: TCPS_LISTEN"); 884 885 /* 886 * Segment received on connection. 887 * Reset idle time and keep-alive timer. 888 */ 889 tp->t_rcvtime = ticks; 890 if (TCPS_HAVEESTABLISHED(tp->t_state)) 891 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 892 893 /* 894 * Process options. 895 * XXX this is tradtitional behavior, may need to be cleaned up. 896 */ 897 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 898 if (thflags & TH_SYN) { 899 if (to.to_flags & TOF_SCALE) { 900 tp->t_flags |= TF_RCVD_SCALE; 901 tp->requested_s_scale = to.to_requested_s_scale; 902 } 903 if (to.to_flags & TOF_TS) { 904 tp->t_flags |= TF_RCVD_TSTMP; 905 tp->ts_recent = to.to_tsval; 906 tp->ts_recent_age = ticks; 907 } 908 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 909 tp->t_flags |= TF_RCVD_CC; 910 if (to.to_flags & TOF_MSS) 911 tcp_mss(tp, to.to_mss); 912 } 913 914 /* 915 * Header prediction: check for the two common cases 916 * of a uni-directional data xfer. If the packet has 917 * no control flags, is in-sequence, the window didn't 918 * change and we're not retransmitting, it's a 919 * candidate. If the length is zero and the ack moved 920 * forward, we're the sender side of the xfer. Just 921 * free the data acked & wake any higher level process 922 * that was blocked waiting for space. If the length 923 * is non-zero and the ack didn't move, we're the 924 * receiver side. If we're getting packets in-order 925 * (the reassembly queue is empty), add the data to 926 * the socket buffer and note that we need a delayed ack. 927 * Make sure that the hidden state-flags are also off. 928 * Since we check for TCPS_ESTABLISHED above, it can only 929 * be TH_NEEDSYN. 930 */ 931 if (tp->t_state == TCPS_ESTABLISHED && 932 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 933 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 934 ((to.to_flags & TOF_TS) == 0 || 935 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 936 /* 937 * Using the CC option is compulsory if once started: 938 * the segment is OK if no T/TCP was negotiated or 939 * if the segment has a CC option equal to CCrecv 940 */ 941 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 942 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 943 th->th_seq == tp->rcv_nxt && 944 tiwin && tiwin == tp->snd_wnd && 945 tp->snd_nxt == tp->snd_max) { 946 947 /* 948 * If last ACK falls within this segment's sequence numbers, 949 * record the timestamp. 950 * NOTE that the test is modified according to the latest 951 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 952 */ 953 if ((to.to_flags & TOF_TS) != 0 && 954 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 955 tp->ts_recent_age = ticks; 956 tp->ts_recent = to.to_tsval; 957 } 958 959 if (tlen == 0) { 960 if (SEQ_GT(th->th_ack, tp->snd_una) && 961 SEQ_LEQ(th->th_ack, tp->snd_max) && 962 tp->snd_cwnd >= tp->snd_wnd && 963 ((!tcp_do_newreno && 964 tp->t_dupacks < tcprexmtthresh) || 965 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 966 /* 967 * this is a pure ack for outstanding data. 968 */ 969 ++tcpstat.tcps_predack; 970 /* 971 * "bad retransmit" recovery 972 * 973 * If Eifel detection applies, then 974 * it is deterministic, so use it 975 * unconditionally over the old heuristic. 976 * Otherwise, fall back to the old heuristic. 977 */ 978 if (tcp_do_eifel_detect && 979 (to.to_flags & TOF_TS) && to.to_tsecr && 980 (tp->t_flags & TF_FIRSTACCACK)) { 981 /* Eifel detection applicable. */ 982 if (to.to_tsecr < tp->t_rexmtTS) { 983 tcp_revert_congestion_state(tp); 984 ++tcpstat.tcps_eifeldetected; 985 } 986 } else if (tp->t_rxtshift == 1 && 987 ticks < tp->t_badrxtwin) { 988 tcp_revert_congestion_state(tp); 989 ++tcpstat.tcps_rttdetected; 990 } 991 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT); 992 /* 993 * Recalculate the retransmit timer / rtt. 994 * 995 * Some machines (certain windows boxes) 996 * send broken timestamp replies during the 997 * SYN+ACK phase, ignore timestamps of 0. 998 */ 999 if ((to.to_flags & TOF_TS) != 0 && 1000 to.to_tsecr) { 1001 tcp_xmit_timer(tp, 1002 ticks - to.to_tsecr + 1); 1003 } else if (tp->t_rtttime && 1004 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1005 tcp_xmit_timer(tp, 1006 ticks - tp->t_rtttime); 1007 } 1008 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1009 acked = th->th_ack - tp->snd_una; 1010 tcpstat.tcps_rcvackpack++; 1011 tcpstat.tcps_rcvackbyte += acked; 1012 sbdrop(&so->so_snd, acked); 1013 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1014 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1015 tp->snd_recover = th->th_ack - 1; 1016 tp->snd_una = th->th_ack; 1017 tp->t_dupacks = 0; 1018 m_freem(m); 1019 ND6_HINT(tp); /* some progress has been done */ 1020 1021 /* 1022 * If all outstanding data are acked, stop 1023 * retransmit timer, otherwise restart timer 1024 * using current (possibly backed-off) value. 1025 * If process is waiting for space, 1026 * wakeup/selwakeup/signal. If data 1027 * are ready to send, let tcp_output 1028 * decide between more output or persist. 1029 */ 1030 if (tp->snd_una == tp->snd_max) 1031 callout_stop(tp->tt_rexmt); 1032 else if (!callout_active(tp->tt_persist)) 1033 callout_reset(tp->tt_rexmt, 1034 tp->t_rxtcur, 1035 tcp_timer_rexmt, tp); 1036 1037 sowwakeup(so); 1038 if (so->so_snd.sb_cc) 1039 (void) tcp_output(tp); 1040 return; 1041 } 1042 } else if (th->th_ack == tp->snd_una && 1043 LIST_EMPTY(&tp->t_segq) && 1044 tlen <= sbspace(&so->so_rcv)) { 1045 /* 1046 * this is a pure, in-sequence data packet 1047 * with nothing on the reassembly queue and 1048 * we have enough buffer space to take it. 1049 */ 1050 ++tcpstat.tcps_preddat; 1051 tp->rcv_nxt += tlen; 1052 tcpstat.tcps_rcvpack++; 1053 tcpstat.tcps_rcvbyte += tlen; 1054 ND6_HINT(tp); /* some progress has been done */ 1055 /* 1056 * Add data to socket buffer. 1057 */ 1058 if (so->so_state & SS_CANTRCVMORE) { 1059 m_freem(m); 1060 } else { 1061 m_adj(m, drop_hdrlen); /* delayed header drop */ 1062 sbappend(&so->so_rcv, m); 1063 } 1064 sorwakeup(so); 1065 if (DELAY_ACK(tp)) { 1066 callout_reset(tp->tt_delack, tcp_delacktime, 1067 tcp_timer_delack, tp); 1068 } else { 1069 tp->t_flags |= TF_ACKNOW; 1070 tcp_output(tp); 1071 } 1072 return; 1073 } 1074 } 1075 1076 /* 1077 * Calculate amount of space in receive window, 1078 * and then do TCP input processing. 1079 * Receive window is amount of space in rcv queue, 1080 * but not less than advertised window. 1081 */ 1082 { int win; 1083 1084 win = sbspace(&so->so_rcv); 1085 if (win < 0) 1086 win = 0; 1087 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1088 } 1089 1090 switch (tp->t_state) { 1091 1092 /* 1093 * If the state is SYN_RECEIVED: 1094 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1095 */ 1096 case TCPS_SYN_RECEIVED: 1097 if ((thflags & TH_ACK) && 1098 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1099 SEQ_GT(th->th_ack, tp->snd_max))) { 1100 rstreason = BANDLIM_RST_OPENPORT; 1101 goto dropwithreset; 1102 } 1103 break; 1104 1105 /* 1106 * If the state is SYN_SENT: 1107 * if seg contains an ACK, but not for our SYN, drop the input. 1108 * if seg contains a RST, then drop the connection. 1109 * if seg does not contain SYN, then drop it. 1110 * Otherwise this is an acceptable SYN segment 1111 * initialize tp->rcv_nxt and tp->irs 1112 * if seg contains ack then advance tp->snd_una 1113 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1114 * arrange for segment to be acked (eventually) 1115 * continue processing rest of data/controls, beginning with URG 1116 */ 1117 case TCPS_SYN_SENT: 1118 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1119 taop = &tao_noncached; 1120 bzero(taop, sizeof(*taop)); 1121 } 1122 1123 if ((thflags & TH_ACK) && 1124 (SEQ_LEQ(th->th_ack, tp->iss) || 1125 SEQ_GT(th->th_ack, tp->snd_max))) { 1126 /* 1127 * If we have a cached CCsent for the remote host, 1128 * hence we haven't just crashed and restarted, 1129 * do not send a RST. This may be a retransmission 1130 * from the other side after our earlier ACK was lost. 1131 * Our new SYN, when it arrives, will serve as the 1132 * needed ACK. 1133 */ 1134 if (taop->tao_ccsent != 0) 1135 goto drop; 1136 else { 1137 rstreason = BANDLIM_UNLIMITED; 1138 goto dropwithreset; 1139 } 1140 } 1141 if (thflags & TH_RST) { 1142 if (thflags & TH_ACK) 1143 tp = tcp_drop(tp, ECONNREFUSED); 1144 goto drop; 1145 } 1146 if ((thflags & TH_SYN) == 0) 1147 goto drop; 1148 tp->snd_wnd = th->th_win; /* initial send window */ 1149 tp->cc_recv = to.to_cc; /* foreign CC */ 1150 1151 tp->irs = th->th_seq; 1152 tcp_rcvseqinit(tp); 1153 if (thflags & TH_ACK) { 1154 /* 1155 * Our SYN was acked. If segment contains CC.ECHO 1156 * option, check it to make sure this segment really 1157 * matches our SYN. If not, just drop it as old 1158 * duplicate, but send an RST if we're still playing 1159 * by the old rules. If no CC.ECHO option, make sure 1160 * we don't get fooled into using T/TCP. 1161 */ 1162 if (to.to_flags & TOF_CCECHO) { 1163 if (tp->cc_send != to.to_ccecho) { 1164 if (taop->tao_ccsent != 0) 1165 goto drop; 1166 else { 1167 rstreason = BANDLIM_UNLIMITED; 1168 goto dropwithreset; 1169 } 1170 } 1171 } else 1172 tp->t_flags &= ~TF_RCVD_CC; 1173 tcpstat.tcps_connects++; 1174 soisconnected(so); 1175 /* Do window scaling on this connection? */ 1176 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1177 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1178 tp->snd_scale = tp->requested_s_scale; 1179 tp->rcv_scale = tp->request_r_scale; 1180 } 1181 /* Segment is acceptable, update cache if undefined. */ 1182 if (taop->tao_ccsent == 0) 1183 taop->tao_ccsent = to.to_ccecho; 1184 1185 tp->rcv_adv += tp->rcv_wnd; 1186 tp->snd_una++; /* SYN is acked */ 1187 /* 1188 * If there's data, delay ACK; if there's also a FIN 1189 * ACKNOW will be turned on later. 1190 */ 1191 if (DELAY_ACK(tp) && tlen != 0) 1192 callout_reset(tp->tt_delack, tcp_delacktime, 1193 tcp_timer_delack, tp); 1194 else 1195 tp->t_flags |= TF_ACKNOW; 1196 /* 1197 * Received <SYN,ACK> in SYN_SENT[*] state. 1198 * Transitions: 1199 * SYN_SENT --> ESTABLISHED 1200 * SYN_SENT* --> FIN_WAIT_1 1201 */ 1202 tp->t_starttime = ticks; 1203 if (tp->t_flags & TF_NEEDFIN) { 1204 tp->t_state = TCPS_FIN_WAIT_1; 1205 tp->t_flags &= ~TF_NEEDFIN; 1206 thflags &= ~TH_SYN; 1207 } else { 1208 tp->t_state = TCPS_ESTABLISHED; 1209 callout_reset(tp->tt_keep, tcp_keepidle, 1210 tcp_timer_keep, tp); 1211 } 1212 } else { 1213 /* 1214 * Received initial SYN in SYN-SENT[*] state => 1215 * simultaneous open. If segment contains CC option 1216 * and there is a cached CC, apply TAO test. 1217 * If it succeeds, connection is * half-synchronized. 1218 * Otherwise, do 3-way handshake: 1219 * SYN-SENT -> SYN-RECEIVED 1220 * SYN-SENT* -> SYN-RECEIVED* 1221 * If there was no CC option, clear cached CC value. 1222 */ 1223 tp->t_flags |= TF_ACKNOW; 1224 callout_stop(tp->tt_rexmt); 1225 if (to.to_flags & TOF_CC) { 1226 if (taop->tao_cc != 0 && 1227 CC_GT(to.to_cc, taop->tao_cc)) { 1228 /* 1229 * update cache and make transition: 1230 * SYN-SENT -> ESTABLISHED* 1231 * SYN-SENT* -> FIN-WAIT-1* 1232 */ 1233 taop->tao_cc = to.to_cc; 1234 tp->t_starttime = ticks; 1235 if (tp->t_flags & TF_NEEDFIN) { 1236 tp->t_state = TCPS_FIN_WAIT_1; 1237 tp->t_flags &= ~TF_NEEDFIN; 1238 } else { 1239 tp->t_state = TCPS_ESTABLISHED; 1240 callout_reset(tp->tt_keep, 1241 tcp_keepidle, 1242 tcp_timer_keep, 1243 tp); 1244 } 1245 tp->t_flags |= TF_NEEDSYN; 1246 } else 1247 tp->t_state = TCPS_SYN_RECEIVED; 1248 } else { 1249 /* CC.NEW or no option => invalidate cache */ 1250 taop->tao_cc = 0; 1251 tp->t_state = TCPS_SYN_RECEIVED; 1252 } 1253 } 1254 1255 trimthenstep6: 1256 /* 1257 * Advance th->th_seq to correspond to first data byte. 1258 * If data, trim to stay within window, 1259 * dropping FIN if necessary. 1260 */ 1261 th->th_seq++; 1262 if (tlen > tp->rcv_wnd) { 1263 todrop = tlen - tp->rcv_wnd; 1264 m_adj(m, -todrop); 1265 tlen = tp->rcv_wnd; 1266 thflags &= ~TH_FIN; 1267 tcpstat.tcps_rcvpackafterwin++; 1268 tcpstat.tcps_rcvbyteafterwin += todrop; 1269 } 1270 tp->snd_wl1 = th->th_seq - 1; 1271 tp->rcv_up = th->th_seq; 1272 /* 1273 * Client side of transaction: already sent SYN and data. 1274 * If the remote host used T/TCP to validate the SYN, 1275 * our data will be ACK'd; if so, enter normal data segment 1276 * processing in the middle of step 5, ack processing. 1277 * Otherwise, goto step 6. 1278 */ 1279 if (thflags & TH_ACK) 1280 goto process_ACK; 1281 1282 goto step6; 1283 1284 /* 1285 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1286 * if segment contains a SYN and CC [not CC.NEW] option: 1287 * if state == TIME_WAIT and connection duration > MSL, 1288 * drop packet and send RST; 1289 * 1290 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1291 * ack the FIN (and data) in retransmission queue. 1292 * Complete close and delete TCPCB. Then reprocess 1293 * segment, hoping to find new TCPCB in LISTEN state; 1294 * 1295 * else must be old SYN; drop it. 1296 * else do normal processing. 1297 */ 1298 case TCPS_LAST_ACK: 1299 case TCPS_CLOSING: 1300 case TCPS_TIME_WAIT: 1301 if ((thflags & TH_SYN) && 1302 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1303 if (tp->t_state == TCPS_TIME_WAIT && 1304 (ticks - tp->t_starttime) > tcp_msl) { 1305 rstreason = BANDLIM_UNLIMITED; 1306 goto dropwithreset; 1307 } 1308 if (CC_GT(to.to_cc, tp->cc_recv)) { 1309 tp = tcp_close(tp); 1310 goto findpcb; 1311 } 1312 else 1313 goto drop; 1314 } 1315 break; /* continue normal processing */ 1316 } 1317 1318 /* 1319 * States other than LISTEN or SYN_SENT. 1320 * First check the RST flag and sequence number since reset segments 1321 * are exempt from the timestamp and connection count tests. This 1322 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1323 * below which allowed reset segments in half the sequence space 1324 * to fall though and be processed (which gives forged reset 1325 * segments with a random sequence number a 50 percent chance of 1326 * killing a connection). 1327 * Then check timestamp, if present. 1328 * Then check the connection count, if present. 1329 * Then check that at least some bytes of segment are within 1330 * receive window. If segment begins before rcv_nxt, 1331 * drop leading data (and SYN); if nothing left, just ack. 1332 * 1333 * 1334 * If the RST bit is set, check the sequence number to see 1335 * if this is a valid reset segment. 1336 * RFC 793 page 37: 1337 * In all states except SYN-SENT, all reset (RST) segments 1338 * are validated by checking their SEQ-fields. A reset is 1339 * valid if its sequence number is in the window. 1340 * Note: this does not take into account delayed ACKs, so 1341 * we should test against last_ack_sent instead of rcv_nxt. 1342 * The sequence number in the reset segment is normally an 1343 * echo of our outgoing acknowlegement numbers, but some hosts 1344 * send a reset with the sequence number at the rightmost edge 1345 * of our receive window, and we have to handle this case. 1346 * If we have multiple segments in flight, the intial reset 1347 * segment sequence numbers will be to the left of last_ack_sent, 1348 * but they will eventually catch up. 1349 * In any case, it never made sense to trim reset segments to 1350 * fit the receive window since RFC 1122 says: 1351 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1352 * 1353 * A TCP SHOULD allow a received RST segment to include data. 1354 * 1355 * DISCUSSION 1356 * It has been suggested that a RST segment could contain 1357 * ASCII text that encoded and explained the cause of the 1358 * RST. No standard has yet been established for such 1359 * data. 1360 * 1361 * If the reset segment passes the sequence number test examine 1362 * the state: 1363 * SYN_RECEIVED STATE: 1364 * If passive open, return to LISTEN state. 1365 * If active open, inform user that connection was refused. 1366 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1367 * Inform user that connection was reset, and close tcb. 1368 * CLOSING, LAST_ACK STATES: 1369 * Close the tcb. 1370 * TIME_WAIT STATE: 1371 * Drop the segment - see Stevens, vol. 2, p. 964 and 1372 * RFC 1337. 1373 */ 1374 if (thflags & TH_RST) { 1375 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1376 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1377 switch (tp->t_state) { 1378 1379 case TCPS_SYN_RECEIVED: 1380 so->so_error = ECONNREFUSED; 1381 goto close; 1382 1383 case TCPS_ESTABLISHED: 1384 case TCPS_FIN_WAIT_1: 1385 case TCPS_FIN_WAIT_2: 1386 case TCPS_CLOSE_WAIT: 1387 so->so_error = ECONNRESET; 1388 close: 1389 tp->t_state = TCPS_CLOSED; 1390 tcpstat.tcps_drops++; 1391 tp = tcp_close(tp); 1392 break; 1393 1394 case TCPS_CLOSING: 1395 case TCPS_LAST_ACK: 1396 tp = tcp_close(tp); 1397 break; 1398 1399 case TCPS_TIME_WAIT: 1400 break; 1401 } 1402 } 1403 goto drop; 1404 } 1405 1406 /* 1407 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1408 * and it's less than ts_recent, drop it. 1409 */ 1410 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1411 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1412 1413 /* Check to see if ts_recent is over 24 days old. */ 1414 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1415 /* 1416 * Invalidate ts_recent. If this segment updates 1417 * ts_recent, the age will be reset later and ts_recent 1418 * will get a valid value. If it does not, setting 1419 * ts_recent to zero will at least satisfy the 1420 * requirement that zero be placed in the timestamp 1421 * echo reply when ts_recent isn't valid. The 1422 * age isn't reset until we get a valid ts_recent 1423 * because we don't want out-of-order segments to be 1424 * dropped when ts_recent is old. 1425 */ 1426 tp->ts_recent = 0; 1427 } else { 1428 tcpstat.tcps_rcvduppack++; 1429 tcpstat.tcps_rcvdupbyte += tlen; 1430 tcpstat.tcps_pawsdrop++; 1431 if (tlen) 1432 goto dropafterack; 1433 goto drop; 1434 } 1435 } 1436 1437 /* 1438 * T/TCP mechanism 1439 * If T/TCP was negotiated and the segment doesn't have CC, 1440 * or if its CC is wrong then drop the segment. 1441 * RST segments do not have to comply with this. 1442 */ 1443 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1444 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1445 goto dropafterack; 1446 1447 /* 1448 * In the SYN-RECEIVED state, validate that the packet belongs to 1449 * this connection before trimming the data to fit the receive 1450 * window. Check the sequence number versus IRS since we know 1451 * the sequence numbers haven't wrapped. This is a partial fix 1452 * for the "LAND" DoS attack. 1453 */ 1454 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1455 rstreason = BANDLIM_RST_OPENPORT; 1456 goto dropwithreset; 1457 } 1458 1459 todrop = tp->rcv_nxt - th->th_seq; 1460 if (todrop > 0) { 1461 if (thflags & TH_SYN) { 1462 thflags &= ~TH_SYN; 1463 th->th_seq++; 1464 if (th->th_urp > 1) 1465 th->th_urp--; 1466 else 1467 thflags &= ~TH_URG; 1468 todrop--; 1469 } 1470 /* 1471 * Following if statement from Stevens, vol. 2, p. 960. 1472 */ 1473 if (todrop > tlen 1474 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1475 /* 1476 * Any valid FIN must be to the left of the window. 1477 * At this point the FIN must be a duplicate or out 1478 * of sequence; drop it. 1479 */ 1480 thflags &= ~TH_FIN; 1481 1482 /* 1483 * Send an ACK to resynchronize and drop any data. 1484 * But keep on processing for RST or ACK. 1485 */ 1486 tp->t_flags |= TF_ACKNOW; 1487 todrop = tlen; 1488 tcpstat.tcps_rcvduppack++; 1489 tcpstat.tcps_rcvdupbyte += todrop; 1490 } else { 1491 tcpstat.tcps_rcvpartduppack++; 1492 tcpstat.tcps_rcvpartdupbyte += todrop; 1493 } 1494 drop_hdrlen += todrop; /* drop from the top afterwards */ 1495 th->th_seq += todrop; 1496 tlen -= todrop; 1497 if (th->th_urp > todrop) 1498 th->th_urp -= todrop; 1499 else { 1500 thflags &= ~TH_URG; 1501 th->th_urp = 0; 1502 } 1503 } 1504 1505 /* 1506 * If new data are received on a connection after the 1507 * user processes are gone, then RST the other end. 1508 */ 1509 if ((so->so_state & SS_NOFDREF) && 1510 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1511 tp = tcp_close(tp); 1512 tcpstat.tcps_rcvafterclose++; 1513 rstreason = BANDLIM_UNLIMITED; 1514 goto dropwithreset; 1515 } 1516 1517 /* 1518 * If segment ends after window, drop trailing data 1519 * (and PUSH and FIN); if nothing left, just ACK. 1520 */ 1521 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1522 if (todrop > 0) { 1523 tcpstat.tcps_rcvpackafterwin++; 1524 if (todrop >= tlen) { 1525 tcpstat.tcps_rcvbyteafterwin += tlen; 1526 /* 1527 * If a new connection request is received 1528 * while in TIME_WAIT, drop the old connection 1529 * and start over if the sequence numbers 1530 * are above the previous ones. 1531 */ 1532 if (thflags & TH_SYN && 1533 tp->t_state == TCPS_TIME_WAIT && 1534 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1535 tp = tcp_close(tp); 1536 goto findpcb; 1537 } 1538 /* 1539 * If window is closed can only take segments at 1540 * window edge, and have to drop data and PUSH from 1541 * incoming segments. Continue processing, but 1542 * remember to ack. Otherwise, drop segment 1543 * and ack. 1544 */ 1545 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1546 tp->t_flags |= TF_ACKNOW; 1547 tcpstat.tcps_rcvwinprobe++; 1548 } else 1549 goto dropafterack; 1550 } else 1551 tcpstat.tcps_rcvbyteafterwin += todrop; 1552 m_adj(m, -todrop); 1553 tlen -= todrop; 1554 thflags &= ~(TH_PUSH|TH_FIN); 1555 } 1556 1557 /* 1558 * If last ACK falls within this segment's sequence numbers, 1559 * record its timestamp. 1560 * NOTE that the test is modified according to the latest 1561 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1562 */ 1563 if ((to.to_flags & TOF_TS) != 0 && 1564 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1565 tp->ts_recent_age = ticks; 1566 tp->ts_recent = to.to_tsval; 1567 } 1568 1569 /* 1570 * If a SYN is in the window, then this is an 1571 * error and we send an RST and drop the connection. 1572 */ 1573 if (thflags & TH_SYN) { 1574 tp = tcp_drop(tp, ECONNRESET); 1575 rstreason = BANDLIM_UNLIMITED; 1576 goto dropwithreset; 1577 } 1578 1579 /* 1580 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1581 * flag is on (half-synchronized state), then queue data for 1582 * later processing; else drop segment and return. 1583 */ 1584 if ((thflags & TH_ACK) == 0) { 1585 if (tp->t_state == TCPS_SYN_RECEIVED || 1586 (tp->t_flags & TF_NEEDSYN)) 1587 goto step6; 1588 else 1589 goto drop; 1590 } 1591 1592 /* 1593 * Ack processing. 1594 */ 1595 switch (tp->t_state) { 1596 1597 /* 1598 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1599 * ESTABLISHED state and continue processing. 1600 * The ACK was checked above. 1601 */ 1602 case TCPS_SYN_RECEIVED: 1603 1604 tcpstat.tcps_connects++; 1605 soisconnected(so); 1606 /* Do window scaling? */ 1607 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1608 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1609 tp->snd_scale = tp->requested_s_scale; 1610 tp->rcv_scale = tp->request_r_scale; 1611 } 1612 /* 1613 * Upon successful completion of 3-way handshake, 1614 * update cache.CC if it was undefined, pass any queued 1615 * data to the user, and advance state appropriately. 1616 */ 1617 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1618 taop->tao_cc == 0) 1619 taop->tao_cc = tp->cc_recv; 1620 1621 /* 1622 * Make transitions: 1623 * SYN-RECEIVED -> ESTABLISHED 1624 * SYN-RECEIVED* -> FIN-WAIT-1 1625 */ 1626 tp->t_starttime = ticks; 1627 if (tp->t_flags & TF_NEEDFIN) { 1628 tp->t_state = TCPS_FIN_WAIT_1; 1629 tp->t_flags &= ~TF_NEEDFIN; 1630 } else { 1631 tp->t_state = TCPS_ESTABLISHED; 1632 callout_reset(tp->tt_keep, tcp_keepidle, 1633 tcp_timer_keep, tp); 1634 } 1635 /* 1636 * If segment contains data or ACK, will call tcp_reass() 1637 * later; if not, do so now to pass queued data to user. 1638 */ 1639 if (tlen == 0 && (thflags & TH_FIN) == 0) 1640 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1641 (struct mbuf *)0); 1642 tp->snd_wl1 = th->th_seq - 1; 1643 /* fall into ... */ 1644 1645 /* 1646 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1647 * ACKs. If the ack is in the range 1648 * tp->snd_una < th->th_ack <= tp->snd_max 1649 * then advance tp->snd_una to th->th_ack and drop 1650 * data from the retransmission queue. If this ACK reflects 1651 * more up to date window information we update our window information. 1652 */ 1653 case TCPS_ESTABLISHED: 1654 case TCPS_FIN_WAIT_1: 1655 case TCPS_FIN_WAIT_2: 1656 case TCPS_CLOSE_WAIT: 1657 case TCPS_CLOSING: 1658 case TCPS_LAST_ACK: 1659 case TCPS_TIME_WAIT: 1660 1661 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1662 if (tlen == 0 && tiwin == tp->snd_wnd) { 1663 tcpstat.tcps_rcvdupack++; 1664 /* 1665 * If we have outstanding data (other than 1666 * a window probe), this is a completely 1667 * duplicate ack (ie, window info didn't 1668 * change), the ack is the biggest we've 1669 * seen and we've seen exactly our rexmt 1670 * threshhold of them, assume a packet 1671 * has been dropped and retransmit it. 1672 * Kludge snd_nxt & the congestion 1673 * window so we send only this one 1674 * packet. 1675 * 1676 * We know we're losing at the current 1677 * window size so do congestion avoidance 1678 * (set ssthresh to half the current window 1679 * and pull our congestion window back to 1680 * the new ssthresh). 1681 * 1682 * Dup acks mean that packets have left the 1683 * network (they're now cached at the receiver) 1684 * so bump cwnd by the amount in the receiver 1685 * to keep a constant cwnd packets in the 1686 * network. 1687 */ 1688 if (!callout_active(tp->tt_rexmt) || 1689 th->th_ack != tp->snd_una) 1690 tp->t_dupacks = 0; 1691 else if (++tp->t_dupacks > tcprexmtthresh || 1692 (tcp_do_newreno && 1693 IN_FASTRECOVERY(tp))) { 1694 tp->snd_cwnd += tp->t_maxseg; 1695 (void) tcp_output(tp); 1696 goto drop; 1697 } else if (tp->t_dupacks == tcprexmtthresh) { 1698 tcp_seq onxt = tp->snd_nxt; 1699 u_int win; 1700 if (tcp_do_newreno && 1701 SEQ_LEQ(th->th_ack, 1702 tp->snd_recover)) { 1703 tp->t_dupacks = 0; 1704 break; 1705 } 1706 if (tcp_do_eifel_detect && 1707 (tp->t_flags & TF_RCVD_TSTMP)) { 1708 tcp_save_congestion_state(tp); 1709 tp->t_flags |= TF_FASTREXMT; 1710 } 1711 win = min(tp->snd_wnd, tp->snd_cwnd) / 1712 2 / tp->t_maxseg; 1713 if (win < 2) 1714 win = 2; 1715 tp->snd_ssthresh = win * tp->t_maxseg; 1716 ENTER_FASTRECOVERY(tp); 1717 tp->snd_recover = tp->snd_max; 1718 callout_stop(tp->tt_rexmt); 1719 tp->t_rtttime = 0; 1720 tp->snd_nxt = th->th_ack; 1721 tp->snd_cwnd = tp->t_maxseg; 1722 (void) tcp_output(tp); 1723 KASSERT(tp->snd_limited <= 2, 1724 ("tp->snd_limited too big")); 1725 tp->snd_cwnd = tp->snd_ssthresh + 1726 (tp->t_maxseg * 1727 (tp->t_dupacks - tp->snd_limited)); 1728 if (SEQ_GT(onxt, tp->snd_nxt)) 1729 tp->snd_nxt = onxt; 1730 goto drop; 1731 } else if (tcp_do_limitedtransmit) { 1732 u_long oldcwnd = tp->snd_cwnd; 1733 tcp_seq oldsndmax = tp->snd_max; 1734 u_int sent; 1735 KASSERT(tp->t_dupacks == 1 || 1736 tp->t_dupacks == 2, 1737 ("dupacks not 1 or 2")); 1738 if (tp->t_dupacks == 1) 1739 tp->snd_limited = 0; 1740 tp->snd_cwnd = 1741 (tp->snd_nxt - tp->snd_una) + 1742 (tp->t_dupacks - tp->snd_limited) * 1743 tp->t_maxseg; 1744 (void) tcp_output(tp); 1745 sent = tp->snd_max - oldsndmax; 1746 if (sent > tp->t_maxseg) { 1747 KASSERT(tp->snd_limited == 0 && 1748 tp->t_dupacks == 2, 1749 ("sent too much")); 1750 tp->snd_limited = 2; 1751 } else if (sent > 0) 1752 ++tp->snd_limited; 1753 tp->snd_cwnd = oldcwnd; 1754 goto drop; 1755 } 1756 } else 1757 tp->t_dupacks = 0; 1758 break; 1759 } 1760 1761 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1762 1763 /* 1764 * If the congestion window was inflated to account 1765 * for the other side's cached packets, retract it. 1766 */ 1767 if (tcp_do_newreno) { 1768 if (IN_FASTRECOVERY(tp)) { 1769 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1770 tcp_newreno_partial_ack(tp, th); 1771 } else { 1772 /* 1773 * Window inflation should have left us 1774 * with approximately snd_ssthresh 1775 * outstanding data. 1776 * But in case we would be inclined to 1777 * send a burst, better to do it via 1778 * the slow start mechanism. 1779 */ 1780 if (SEQ_GT(th->th_ack + 1781 tp->snd_ssthresh, 1782 tp->snd_max)) 1783 tp->snd_cwnd = tp->snd_max - 1784 th->th_ack + 1785 tp->t_maxseg; 1786 else 1787 tp->snd_cwnd = tp->snd_ssthresh; 1788 } 1789 } 1790 } else { 1791 if (tp->t_dupacks >= tcprexmtthresh && 1792 tp->snd_cwnd > tp->snd_ssthresh) 1793 tp->snd_cwnd = tp->snd_ssthresh; 1794 } 1795 tp->t_dupacks = 0; 1796 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1797 tcpstat.tcps_rcvacktoomuch++; 1798 goto dropafterack; 1799 } 1800 /* 1801 * If we reach this point, ACK is not a duplicate, 1802 * i.e., it ACKs something we sent. 1803 */ 1804 if (tp->t_flags & TF_NEEDSYN) { 1805 /* 1806 * T/TCP: Connection was half-synchronized, and our 1807 * SYN has been ACK'd (so connection is now fully 1808 * synchronized). Go to non-starred state, 1809 * increment snd_una for ACK of SYN, and check if 1810 * we can do window scaling. 1811 */ 1812 tp->t_flags &= ~TF_NEEDSYN; 1813 tp->snd_una++; 1814 /* Do window scaling? */ 1815 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1816 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1817 tp->snd_scale = tp->requested_s_scale; 1818 tp->rcv_scale = tp->request_r_scale; 1819 } 1820 } 1821 1822 process_ACK: 1823 acked = th->th_ack - tp->snd_una; 1824 tcpstat.tcps_rcvackpack++; 1825 tcpstat.tcps_rcvackbyte += acked; 1826 1827 /* 1828 * If we just performed our first retransmit, and the ACK 1829 * arrives within our recovery window, then it was a mistake 1830 * to do the retransmit in the first place. Recover our 1831 * original cwnd and ssthresh, and proceed to transmit where 1832 * we left off. 1833 */ 1834 if (tcp_do_eifel_detect && acked && 1835 (to.to_flags & TOF_TS) && to.to_tsecr && 1836 (tp->t_flags & TF_FIRSTACCACK)) { 1837 /* Eifel detection applicable. */ 1838 if (to.to_tsecr < tp->t_rexmtTS) { 1839 tcp_revert_congestion_state(tp); 1840 ++tcpstat.tcps_eifeldetected; 1841 } 1842 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1843 tcp_revert_congestion_state(tp); 1844 ++tcpstat.tcps_rttdetected; 1845 } 1846 1847 /* 1848 * If we have a timestamp reply, update smoothed 1849 * round trip time. If no timestamp is present but 1850 * transmit timer is running and timed sequence 1851 * number was acked, update smoothed round trip time. 1852 * Since we now have an rtt measurement, cancel the 1853 * timer backoff (cf., Phil Karn's retransmit alg.). 1854 * Recompute the initial retransmit timer. 1855 * 1856 * Some machines (certain windows boxes) send broken 1857 * timestamp replies during the SYN+ACK phase, ignore 1858 * timestamps of 0. 1859 */ 1860 if ((to.to_flags & TOF_TS) != 0 && 1861 to.to_tsecr) { 1862 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1863 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1864 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1865 } 1866 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1867 1868 /* 1869 * If all outstanding data is acked, stop retransmit 1870 * timer and remember to restart (more output or persist). 1871 * If there is more data to be acked, restart retransmit 1872 * timer, using current (possibly backed-off) value. 1873 */ 1874 if (th->th_ack == tp->snd_max) { 1875 callout_stop(tp->tt_rexmt); 1876 needoutput = 1; 1877 } else if (!callout_active(tp->tt_persist)) 1878 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1879 tcp_timer_rexmt, tp); 1880 1881 /* 1882 * If no data (only SYN) was ACK'd, 1883 * skip rest of ACK processing. 1884 */ 1885 if (acked == 0) 1886 goto step6; 1887 1888 /* Stop looking for an acceptable ACK since one was received. */ 1889 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT); 1890 1891 /* 1892 * When new data is acked, open the congestion window. 1893 * If the window gives us less than ssthresh packets 1894 * in flight, open exponentially (maxseg per packet). 1895 * Otherwise open linearly: maxseg per window 1896 * (maxseg^2 / cwnd per packet). 1897 */ 1898 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 1899 u_int cw = tp->snd_cwnd; 1900 u_int incr = tp->t_maxseg; 1901 if (cw > tp->snd_ssthresh) 1902 incr = incr * incr / cw; 1903 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1904 } 1905 if (acked > so->so_snd.sb_cc) { 1906 tp->snd_wnd -= so->so_snd.sb_cc; 1907 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1908 ourfinisacked = 1; 1909 } else { 1910 sbdrop(&so->so_snd, acked); 1911 tp->snd_wnd -= acked; 1912 ourfinisacked = 0; 1913 } 1914 sowwakeup(so); 1915 /* detect una wraparound */ 1916 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 1917 SEQ_GT(tp->snd_una, tp->snd_recover) && 1918 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1919 tp->snd_recover = th->th_ack - 1; 1920 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 1921 SEQ_GEQ(th->th_ack, tp->snd_recover)) 1922 EXIT_FASTRECOVERY(tp); 1923 tp->snd_una = th->th_ack; 1924 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1925 tp->snd_nxt = tp->snd_una; 1926 1927 switch (tp->t_state) { 1928 1929 /* 1930 * In FIN_WAIT_1 STATE in addition to the processing 1931 * for the ESTABLISHED state if our FIN is now acknowledged 1932 * then enter FIN_WAIT_2. 1933 */ 1934 case TCPS_FIN_WAIT_1: 1935 if (ourfinisacked) { 1936 /* 1937 * If we can't receive any more 1938 * data, then closing user can proceed. 1939 * Starting the timer is contrary to the 1940 * specification, but if we don't get a FIN 1941 * we'll hang forever. 1942 */ 1943 if (so->so_state & SS_CANTRCVMORE) { 1944 soisdisconnected(so); 1945 callout_reset(tp->tt_2msl, tcp_maxidle, 1946 tcp_timer_2msl, tp); 1947 } 1948 tp->t_state = TCPS_FIN_WAIT_2; 1949 } 1950 break; 1951 1952 /* 1953 * In CLOSING STATE in addition to the processing for 1954 * the ESTABLISHED state if the ACK acknowledges our FIN 1955 * then enter the TIME-WAIT state, otherwise ignore 1956 * the segment. 1957 */ 1958 case TCPS_CLOSING: 1959 if (ourfinisacked) { 1960 tp->t_state = TCPS_TIME_WAIT; 1961 tcp_canceltimers(tp); 1962 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1963 if (tp->cc_recv != 0 && 1964 (ticks - tp->t_starttime) < tcp_msl) 1965 callout_reset(tp->tt_2msl, 1966 tp->t_rxtcur * 1967 TCPTV_TWTRUNC, 1968 tcp_timer_2msl, tp); 1969 else 1970 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1971 tcp_timer_2msl, tp); 1972 soisdisconnected(so); 1973 } 1974 break; 1975 1976 /* 1977 * In LAST_ACK, we may still be waiting for data to drain 1978 * and/or to be acked, as well as for the ack of our FIN. 1979 * If our FIN is now acknowledged, delete the TCB, 1980 * enter the closed state and return. 1981 */ 1982 case TCPS_LAST_ACK: 1983 if (ourfinisacked) { 1984 tp = tcp_close(tp); 1985 goto drop; 1986 } 1987 break; 1988 1989 /* 1990 * In TIME_WAIT state the only thing that should arrive 1991 * is a retransmission of the remote FIN. Acknowledge 1992 * it and restart the finack timer. 1993 */ 1994 case TCPS_TIME_WAIT: 1995 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1996 tcp_timer_2msl, tp); 1997 goto dropafterack; 1998 } 1999 } 2000 2001 step6: 2002 /* 2003 * Update window information. 2004 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2005 */ 2006 if ((thflags & TH_ACK) && 2007 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2008 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2009 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2010 /* keep track of pure window updates */ 2011 if (tlen == 0 && 2012 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2013 tcpstat.tcps_rcvwinupd++; 2014 tp->snd_wnd = tiwin; 2015 tp->snd_wl1 = th->th_seq; 2016 tp->snd_wl2 = th->th_ack; 2017 if (tp->snd_wnd > tp->max_sndwnd) 2018 tp->max_sndwnd = tp->snd_wnd; 2019 needoutput = 1; 2020 } 2021 2022 /* 2023 * Process segments with URG. 2024 */ 2025 if ((thflags & TH_URG) && th->th_urp && 2026 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2027 /* 2028 * This is a kludge, but if we receive and accept 2029 * random urgent pointers, we'll crash in 2030 * soreceive. It's hard to imagine someone 2031 * actually wanting to send this much urgent data. 2032 */ 2033 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2034 th->th_urp = 0; /* XXX */ 2035 thflags &= ~TH_URG; /* XXX */ 2036 goto dodata; /* XXX */ 2037 } 2038 /* 2039 * If this segment advances the known urgent pointer, 2040 * then mark the data stream. This should not happen 2041 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2042 * a FIN has been received from the remote side. 2043 * In these states we ignore the URG. 2044 * 2045 * According to RFC961 (Assigned Protocols), 2046 * the urgent pointer points to the last octet 2047 * of urgent data. We continue, however, 2048 * to consider it to indicate the first octet 2049 * of data past the urgent section as the original 2050 * spec states (in one of two places). 2051 */ 2052 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2053 tp->rcv_up = th->th_seq + th->th_urp; 2054 so->so_oobmark = so->so_rcv.sb_cc + 2055 (tp->rcv_up - tp->rcv_nxt) - 1; 2056 if (so->so_oobmark == 0) 2057 so->so_state |= SS_RCVATMARK; 2058 sohasoutofband(so); 2059 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2060 } 2061 /* 2062 * Remove out of band data so doesn't get presented to user. 2063 * This can happen independent of advancing the URG pointer, 2064 * but if two URG's are pending at once, some out-of-band 2065 * data may creep in... ick. 2066 */ 2067 if (th->th_urp <= (u_long)tlen 2068 #ifdef SO_OOBINLINE 2069 && (so->so_options & SO_OOBINLINE) == 0 2070 #endif 2071 ) 2072 tcp_pulloutofband(so, th, m, 2073 drop_hdrlen); /* hdr drop is delayed */ 2074 } else { 2075 /* 2076 * If no out of band data is expected, 2077 * pull receive urgent pointer along 2078 * with the receive window. 2079 */ 2080 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2081 tp->rcv_up = tp->rcv_nxt; 2082 } 2083 dodata: /* XXX */ 2084 2085 /* 2086 * Process the segment text, merging it into the TCP sequencing queue, 2087 * and arranging for acknowledgment of receipt if necessary. 2088 * This process logically involves adjusting tp->rcv_wnd as data 2089 * is presented to the user (this happens in tcp_usrreq.c, 2090 * case PRU_RCVD). If a FIN has already been received on this 2091 * connection then we just ignore the text. 2092 */ 2093 if ((tlen || (thflags & TH_FIN)) && 2094 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2095 m_adj(m, drop_hdrlen); /* delayed header drop */ 2096 /* 2097 * Insert segment which includes th into TCP reassembly queue 2098 * with control block tp. Set thflags to whether reassembly now 2099 * includes a segment with FIN. This handles the common case 2100 * inline (segment is the next to be received on an established 2101 * connection, and the queue is empty), avoiding linkage into 2102 * and removal from the queue and repetition of various 2103 * conversions. 2104 * Set DELACK for segments received in order, but ack 2105 * immediately when segments are out of order (so 2106 * fast retransmit can work). 2107 */ 2108 if (th->th_seq == tp->rcv_nxt && 2109 LIST_EMPTY(&tp->t_segq) && 2110 TCPS_HAVEESTABLISHED(tp->t_state)) { 2111 if (DELAY_ACK(tp)) 2112 callout_reset(tp->tt_delack, tcp_delacktime, 2113 tcp_timer_delack, tp); 2114 else 2115 tp->t_flags |= TF_ACKNOW; 2116 tp->rcv_nxt += tlen; 2117 thflags = th->th_flags & TH_FIN; 2118 tcpstat.tcps_rcvpack++; 2119 tcpstat.tcps_rcvbyte += tlen; 2120 ND6_HINT(tp); 2121 if (so->so_state & SS_CANTRCVMORE) 2122 m_freem(m); 2123 else 2124 sbappend(&so->so_rcv, m); 2125 sorwakeup(so); 2126 } else { 2127 thflags = tcp_reass(tp, th, &tlen, m); 2128 tp->t_flags |= TF_ACKNOW; 2129 } 2130 2131 /* 2132 * Note the amount of data that peer has sent into 2133 * our window, in order to estimate the sender's 2134 * buffer size. 2135 */ 2136 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2137 } else { 2138 m_freem(m); 2139 thflags &= ~TH_FIN; 2140 } 2141 2142 /* 2143 * If FIN is received ACK the FIN and let the user know 2144 * that the connection is closing. 2145 */ 2146 if (thflags & TH_FIN) { 2147 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2148 socantrcvmore(so); 2149 /* 2150 * If connection is half-synchronized 2151 * (ie NEEDSYN flag on) then delay ACK, 2152 * so it may be piggybacked when SYN is sent. 2153 * Otherwise, since we received a FIN then no 2154 * more input can be expected, send ACK now. 2155 */ 2156 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2157 callout_reset(tp->tt_delack, tcp_delacktime, 2158 tcp_timer_delack, tp); 2159 else 2160 tp->t_flags |= TF_ACKNOW; 2161 tp->rcv_nxt++; 2162 } 2163 switch (tp->t_state) { 2164 2165 /* 2166 * In SYN_RECEIVED and ESTABLISHED STATES 2167 * enter the CLOSE_WAIT state. 2168 */ 2169 case TCPS_SYN_RECEIVED: 2170 tp->t_starttime = ticks; 2171 /*FALLTHROUGH*/ 2172 case TCPS_ESTABLISHED: 2173 tp->t_state = TCPS_CLOSE_WAIT; 2174 break; 2175 2176 /* 2177 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2178 * enter the CLOSING state. 2179 */ 2180 case TCPS_FIN_WAIT_1: 2181 tp->t_state = TCPS_CLOSING; 2182 break; 2183 2184 /* 2185 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2186 * starting the time-wait timer, turning off the other 2187 * standard timers. 2188 */ 2189 case TCPS_FIN_WAIT_2: 2190 tp->t_state = TCPS_TIME_WAIT; 2191 tcp_canceltimers(tp); 2192 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2193 if (tp->cc_recv != 0 && 2194 (ticks - tp->t_starttime) < tcp_msl) { 2195 callout_reset(tp->tt_2msl, 2196 tp->t_rxtcur * TCPTV_TWTRUNC, 2197 tcp_timer_2msl, tp); 2198 /* For transaction client, force ACK now. */ 2199 tp->t_flags |= TF_ACKNOW; 2200 } 2201 else 2202 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2203 tcp_timer_2msl, tp); 2204 soisdisconnected(so); 2205 break; 2206 2207 /* 2208 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2209 */ 2210 case TCPS_TIME_WAIT: 2211 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2212 tcp_timer_2msl, tp); 2213 break; 2214 } 2215 } 2216 #ifdef TCPDEBUG 2217 if (so->so_options & SO_DEBUG) 2218 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2219 &tcp_savetcp, 0); 2220 #endif 2221 2222 /* 2223 * Return any desired output. 2224 */ 2225 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2226 (void) tcp_output(tp); 2227 return; 2228 2229 dropafterack: 2230 /* 2231 * Generate an ACK dropping incoming segment if it occupies 2232 * sequence space, where the ACK reflects our state. 2233 * 2234 * We can now skip the test for the RST flag since all 2235 * paths to this code happen after packets containing 2236 * RST have been dropped. 2237 * 2238 * In the SYN-RECEIVED state, don't send an ACK unless the 2239 * segment we received passes the SYN-RECEIVED ACK test. 2240 * If it fails send a RST. This breaks the loop in the 2241 * "LAND" DoS attack, and also prevents an ACK storm 2242 * between two listening ports that have been sent forged 2243 * SYN segments, each with the source address of the other. 2244 */ 2245 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2246 (SEQ_GT(tp->snd_una, th->th_ack) || 2247 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2248 rstreason = BANDLIM_RST_OPENPORT; 2249 goto dropwithreset; 2250 } 2251 #ifdef TCPDEBUG 2252 if (so->so_options & SO_DEBUG) 2253 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2254 &tcp_savetcp, 0); 2255 #endif 2256 m_freem(m); 2257 tp->t_flags |= TF_ACKNOW; 2258 (void) tcp_output(tp); 2259 return; 2260 2261 dropwithreset: 2262 /* 2263 * Generate a RST, dropping incoming segment. 2264 * Make ACK acceptable to originator of segment. 2265 * Don't bother to respond if destination was broadcast/multicast. 2266 */ 2267 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2268 goto drop; 2269 if (isipv6) { 2270 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2271 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2272 goto drop; 2273 } else { 2274 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2275 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2276 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2277 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2278 goto drop; 2279 } 2280 /* IPv6 anycast check is done at tcp6_input() */ 2281 2282 /* 2283 * Perform bandwidth limiting. 2284 */ 2285 #ifdef ICMP_BANDLIM 2286 if (badport_bandlim(rstreason) < 0) 2287 goto drop; 2288 #endif 2289 2290 #ifdef TCPDEBUG 2291 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2292 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2293 &tcp_savetcp, 0); 2294 #endif 2295 if (thflags & TH_ACK) 2296 /* mtod() below is safe as long as hdr dropping is delayed */ 2297 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2298 TH_RST); 2299 else { 2300 if (thflags & TH_SYN) 2301 tlen++; 2302 /* mtod() below is safe as long as hdr dropping is delayed */ 2303 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2304 (tcp_seq)0, TH_RST|TH_ACK); 2305 } 2306 return; 2307 2308 drop: 2309 /* 2310 * Drop space held by incoming segment and return. 2311 */ 2312 #ifdef TCPDEBUG 2313 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2314 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2315 &tcp_savetcp, 0); 2316 #endif 2317 m_freem(m); 2318 return; 2319 } 2320 2321 /* 2322 * Parse TCP options and place in tcpopt. 2323 */ 2324 static void 2325 tcp_dooptions(to, cp, cnt, is_syn) 2326 struct tcpopt *to; 2327 u_char *cp; 2328 int cnt; 2329 { 2330 int opt, optlen; 2331 2332 to->to_flags = 0; 2333 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2334 opt = cp[0]; 2335 if (opt == TCPOPT_EOL) 2336 break; 2337 if (opt == TCPOPT_NOP) 2338 optlen = 1; 2339 else { 2340 if (cnt < 2) 2341 break; 2342 optlen = cp[1]; 2343 if (optlen < 2 || optlen > cnt) 2344 break; 2345 } 2346 switch (opt) { 2347 case TCPOPT_MAXSEG: 2348 if (optlen != TCPOLEN_MAXSEG) 2349 continue; 2350 if (!is_syn) 2351 continue; 2352 to->to_flags |= TOF_MSS; 2353 bcopy((char *)cp + 2, 2354 (char *)&to->to_mss, sizeof(to->to_mss)); 2355 to->to_mss = ntohs(to->to_mss); 2356 break; 2357 case TCPOPT_WINDOW: 2358 if (optlen != TCPOLEN_WINDOW) 2359 continue; 2360 if (! is_syn) 2361 continue; 2362 to->to_flags |= TOF_SCALE; 2363 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2364 break; 2365 case TCPOPT_TIMESTAMP: 2366 if (optlen != TCPOLEN_TIMESTAMP) 2367 continue; 2368 to->to_flags |= TOF_TS; 2369 bcopy((char *)cp + 2, 2370 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2371 to->to_tsval = ntohl(to->to_tsval); 2372 bcopy((char *)cp + 6, 2373 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2374 to->to_tsecr = ntohl(to->to_tsecr); 2375 break; 2376 case TCPOPT_CC: 2377 if (optlen != TCPOLEN_CC) 2378 continue; 2379 to->to_flags |= TOF_CC; 2380 bcopy((char *)cp + 2, 2381 (char *)&to->to_cc, sizeof(to->to_cc)); 2382 to->to_cc = ntohl(to->to_cc); 2383 break; 2384 case TCPOPT_CCNEW: 2385 if (optlen != TCPOLEN_CC) 2386 continue; 2387 if (!is_syn) 2388 continue; 2389 to->to_flags |= TOF_CCNEW; 2390 bcopy((char *)cp + 2, 2391 (char *)&to->to_cc, sizeof(to->to_cc)); 2392 to->to_cc = ntohl(to->to_cc); 2393 break; 2394 case TCPOPT_CCECHO: 2395 if (optlen != TCPOLEN_CC) 2396 continue; 2397 if (!is_syn) 2398 continue; 2399 to->to_flags |= TOF_CCECHO; 2400 bcopy((char *)cp + 2, 2401 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2402 to->to_ccecho = ntohl(to->to_ccecho); 2403 break; 2404 default: 2405 continue; 2406 } 2407 } 2408 } 2409 2410 /* 2411 * Pull out of band byte out of a segment so 2412 * it doesn't appear in the user's data queue. 2413 * It is still reflected in the segment length for 2414 * sequencing purposes. 2415 */ 2416 static void 2417 tcp_pulloutofband(so, th, m, off) 2418 struct socket *so; 2419 struct tcphdr *th; 2420 struct mbuf *m; 2421 int off; /* delayed to be droped hdrlen */ 2422 { 2423 int cnt = off + th->th_urp - 1; 2424 2425 while (cnt >= 0) { 2426 if (m->m_len > cnt) { 2427 char *cp = mtod(m, caddr_t) + cnt; 2428 struct tcpcb *tp = sototcpcb(so); 2429 2430 tp->t_iobc = *cp; 2431 tp->t_oobflags |= TCPOOB_HAVEDATA; 2432 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2433 m->m_len--; 2434 if (m->m_flags & M_PKTHDR) 2435 m->m_pkthdr.len--; 2436 return; 2437 } 2438 cnt -= m->m_len; 2439 m = m->m_next; 2440 if (m == 0) 2441 break; 2442 } 2443 panic("tcp_pulloutofband"); 2444 } 2445 2446 /* 2447 * Collect new round-trip time estimate 2448 * and update averages and current timeout. 2449 */ 2450 static void 2451 tcp_xmit_timer(tp, rtt) 2452 struct tcpcb *tp; 2453 int rtt; 2454 { 2455 int delta; 2456 2457 tcpstat.tcps_rttupdated++; 2458 tp->t_rttupdated++; 2459 if (tp->t_srtt != 0) { 2460 /* 2461 * srtt is stored as fixed point with 5 bits after the 2462 * binary point (i.e., scaled by 8). The following magic 2463 * is equivalent to the smoothing algorithm in rfc793 with 2464 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2465 * point). Adjust rtt to origin 0. 2466 */ 2467 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2468 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2469 2470 if ((tp->t_srtt += delta) <= 0) 2471 tp->t_srtt = 1; 2472 2473 /* 2474 * We accumulate a smoothed rtt variance (actually, a 2475 * smoothed mean difference), then set the retransmit 2476 * timer to smoothed rtt + 4 times the smoothed variance. 2477 * rttvar is stored as fixed point with 4 bits after the 2478 * binary point (scaled by 16). The following is 2479 * equivalent to rfc793 smoothing with an alpha of .75 2480 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2481 * rfc793's wired-in beta. 2482 */ 2483 if (delta < 0) 2484 delta = -delta; 2485 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2486 if ((tp->t_rttvar += delta) <= 0) 2487 tp->t_rttvar = 1; 2488 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2489 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2490 } else { 2491 /* 2492 * No rtt measurement yet - use the unsmoothed rtt. 2493 * Set the variance to half the rtt (so our first 2494 * retransmit happens at 3*rtt). 2495 */ 2496 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2497 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2498 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2499 } 2500 tp->t_rtttime = 0; 2501 tp->t_rxtshift = 0; 2502 2503 /* 2504 * the retransmit should happen at rtt + 4 * rttvar. 2505 * Because of the way we do the smoothing, srtt and rttvar 2506 * will each average +1/2 tick of bias. When we compute 2507 * the retransmit timer, we want 1/2 tick of rounding and 2508 * 1 extra tick because of +-1/2 tick uncertainty in the 2509 * firing of the timer. The bias will give us exactly the 2510 * 1.5 tick we need. But, because the bias is 2511 * statistical, we have to test that we don't drop below 2512 * the minimum feasible timer (which is 2 ticks). 2513 */ 2514 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2515 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2516 2517 /* 2518 * We received an ack for a packet that wasn't retransmitted; 2519 * it is probably safe to discard any error indications we've 2520 * received recently. This isn't quite right, but close enough 2521 * for now (a route might have failed after we sent a segment, 2522 * and the return path might not be symmetrical). 2523 */ 2524 tp->t_softerror = 0; 2525 } 2526 2527 /* 2528 * Determine a reasonable value for maxseg size. 2529 * If the route is known, check route for mtu. 2530 * If none, use an mss that can be handled on the outgoing 2531 * interface without forcing IP to fragment; if bigger than 2532 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2533 * to utilize large mbufs. If no route is found, route has no mtu, 2534 * or the destination isn't local, use a default, hopefully conservative 2535 * size (usually 512 or the default IP max size, but no more than the mtu 2536 * of the interface), as we can't discover anything about intervening 2537 * gateways or networks. We also initialize the congestion/slow start 2538 * window to be a single segment if the destination isn't local. 2539 * While looking at the routing entry, we also initialize other path-dependent 2540 * parameters from pre-set or cached values in the routing entry. 2541 * 2542 * Also take into account the space needed for options that we 2543 * send regularly. Make maxseg shorter by that amount to assure 2544 * that we can send maxseg amount of data even when the options 2545 * are present. Store the upper limit of the length of options plus 2546 * data in maxopd. 2547 * 2548 * NOTE that this routine is only called when we process an incoming 2549 * segment, for outgoing segments only tcp_mssopt is called. 2550 * 2551 * In case of T/TCP, we call this routine during implicit connection 2552 * setup as well (offer = -1), to initialize maxseg from the cached 2553 * MSS of our peer. 2554 */ 2555 void 2556 tcp_mss(tp, offer) 2557 struct tcpcb *tp; 2558 int offer; 2559 { 2560 struct rtentry *rt; 2561 struct ifnet *ifp; 2562 int rtt, mss; 2563 u_long bufsize; 2564 struct inpcb *inp = tp->t_inpcb; 2565 struct socket *so; 2566 struct rmxp_tao *taop; 2567 int origoffer = offer; 2568 #ifdef INET6 2569 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2570 size_t min_protoh = isipv6 ? 2571 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2572 sizeof(struct tcpiphdr); 2573 #else 2574 const int isipv6 = 0; 2575 const size_t min_protoh = sizeof(struct tcpiphdr); 2576 #endif 2577 2578 if (isipv6) 2579 rt = tcp_rtlookup6(&inp->inp_inc); 2580 else 2581 rt = tcp_rtlookup(&inp->inp_inc); 2582 if (rt == NULL) { 2583 tp->t_maxopd = tp->t_maxseg = 2584 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2585 return; 2586 } 2587 ifp = rt->rt_ifp; 2588 so = inp->inp_socket; 2589 2590 taop = rmx_taop(rt->rt_rmx); 2591 /* 2592 * Offer == -1 means that we didn't receive SYN yet, 2593 * use cached value in that case; 2594 */ 2595 if (offer == -1) 2596 offer = taop->tao_mssopt; 2597 /* 2598 * Offer == 0 means that there was no MSS on the SYN segment, 2599 * in this case we use tcp_mssdflt. 2600 */ 2601 if (offer == 0) 2602 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2603 else 2604 /* 2605 * Sanity check: make sure that maxopd will be large 2606 * enough to allow some data on segments even is the 2607 * all the option space is used (40bytes). Otherwise 2608 * funny things may happen in tcp_output. 2609 */ 2610 offer = max(offer, 64); 2611 taop->tao_mssopt = offer; 2612 2613 /* 2614 * While we're here, check if there's an initial rtt 2615 * or rttvar. Convert from the route-table units 2616 * to scaled multiples of the slow timeout timer. 2617 */ 2618 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2619 /* 2620 * XXX the lock bit for RTT indicates that the value 2621 * is also a minimum value; this is subject to time. 2622 */ 2623 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2624 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2625 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2626 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2627 tcpstat.tcps_usedrtt++; 2628 if (rt->rt_rmx.rmx_rttvar) { 2629 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2630 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2631 tcpstat.tcps_usedrttvar++; 2632 } else { 2633 /* default variation is +- 1 rtt */ 2634 tp->t_rttvar = 2635 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2636 } 2637 TCPT_RANGESET(tp->t_rxtcur, 2638 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2639 tp->t_rttmin, TCPTV_REXMTMAX); 2640 } 2641 /* 2642 * if there's an mtu associated with the route, use it 2643 * else, use the link mtu. 2644 */ 2645 if (rt->rt_rmx.rmx_mtu) 2646 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2647 else { 2648 if (isipv6) { 2649 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2650 min_protoh; 2651 if (!in6_localaddr(&inp->in6p_faddr)) 2652 mss = min(mss, tcp_v6mssdflt); 2653 } else { 2654 mss = ifp->if_mtu - min_protoh; 2655 if (!in_localaddr(inp->inp_faddr)) 2656 mss = min(mss, tcp_mssdflt); 2657 } 2658 } 2659 mss = min(mss, offer); 2660 /* 2661 * maxopd stores the maximum length of data AND options 2662 * in a segment; maxseg is the amount of data in a normal 2663 * segment. We need to store this value (maxopd) apart 2664 * from maxseg, because now every segment carries options 2665 * and thus we normally have somewhat less data in segments. 2666 */ 2667 tp->t_maxopd = mss; 2668 2669 /* 2670 * In case of T/TCP, origoffer==-1 indicates, that no segments 2671 * were received yet. In this case we just guess, otherwise 2672 * we do the same as before T/TCP. 2673 */ 2674 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2675 (origoffer == -1 || 2676 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2677 mss -= TCPOLEN_TSTAMP_APPA; 2678 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2679 (origoffer == -1 || 2680 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2681 mss -= TCPOLEN_CC_APPA; 2682 2683 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2684 if (mss > MCLBYTES) 2685 mss &= ~(MCLBYTES-1); 2686 #else 2687 if (mss > MCLBYTES) 2688 mss = mss / MCLBYTES * MCLBYTES; 2689 #endif 2690 /* 2691 * If there's a pipesize, change the socket buffer 2692 * to that size. Make the socket buffers an integral 2693 * number of mss units; if the mss is larger than 2694 * the socket buffer, decrease the mss. 2695 */ 2696 #ifdef RTV_SPIPE 2697 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2698 #endif 2699 bufsize = so->so_snd.sb_hiwat; 2700 if (bufsize < mss) 2701 mss = bufsize; 2702 else { 2703 bufsize = roundup(bufsize, mss); 2704 if (bufsize > sb_max) 2705 bufsize = sb_max; 2706 if (bufsize > so->so_snd.sb_hiwat) 2707 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2708 } 2709 tp->t_maxseg = mss; 2710 2711 #ifdef RTV_RPIPE 2712 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2713 #endif 2714 bufsize = so->so_rcv.sb_hiwat; 2715 if (bufsize > mss) { 2716 bufsize = roundup(bufsize, mss); 2717 if (bufsize > sb_max) 2718 bufsize = sb_max; 2719 if (bufsize > so->so_rcv.sb_hiwat) 2720 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2721 } 2722 2723 /* 2724 * Set the slow-start flight size depending on whether this 2725 * is a local network or not. 2726 */ 2727 if (tcp_do_rfc3390) 2728 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2729 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2730 (!isipv6 && in_localaddr(inp->inp_faddr))) 2731 tp->snd_cwnd = mss * ss_fltsz_local; 2732 else 2733 tp->snd_cwnd = mss * ss_fltsz; 2734 2735 if (rt->rt_rmx.rmx_ssthresh) { 2736 /* 2737 * There's some sort of gateway or interface 2738 * buffer limit on the path. Use this to set 2739 * the slow start threshhold, but set the 2740 * threshold to no less than 2*mss. 2741 */ 2742 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2743 tcpstat.tcps_usedssthresh++; 2744 } 2745 } 2746 2747 /* 2748 * Determine the MSS option to send on an outgoing SYN. 2749 */ 2750 int 2751 tcp_mssopt(tp) 2752 struct tcpcb *tp; 2753 { 2754 struct rtentry *rt; 2755 #ifdef INET6 2756 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2757 int min_protoh = isipv6 ? 2758 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2759 sizeof(struct tcpiphdr); 2760 #else 2761 const int isipv6 = 0; 2762 const size_t min_protoh = sizeof(struct tcpiphdr); 2763 #endif 2764 2765 if (isipv6) 2766 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2767 else 2768 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2769 if (rt == NULL) 2770 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2771 2772 return (rt->rt_ifp->if_mtu - min_protoh); 2773 } 2774 2775 2776 /* 2777 * When a partial ack arrives, force the retransmission of the 2778 * next unacknowledged segment. Do not clear tp->t_dupacks. 2779 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2780 * be started again. 2781 */ 2782 static void 2783 tcp_newreno_partial_ack(tp, th) 2784 struct tcpcb *tp; 2785 struct tcphdr *th; 2786 { 2787 tcp_seq onxt = tp->snd_nxt; 2788 u_long ocwnd = tp->snd_cwnd; 2789 2790 callout_stop(tp->tt_rexmt); 2791 tp->t_rtttime = 0; 2792 tp->snd_nxt = th->th_ack; 2793 /* 2794 * Set snd_cwnd to one segment beyond acknowledged offset 2795 * (tp->snd_una has not yet been updated when this function is called.) 2796 */ 2797 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2798 tp->t_flags |= TF_ACKNOW; 2799 (void) tcp_output(tp); 2800 tp->snd_cwnd = ocwnd; 2801 if (SEQ_GT(onxt, tp->snd_nxt)) 2802 tp->snd_nxt = onxt; 2803 /* 2804 * Partial window deflation. Relies on fact that tp->snd_una 2805 * not updated yet. 2806 */ 2807 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2808 } 2809