1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_ipfw.h" /* for ipfw_fwd */ 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_mac.h" 41 #include "opt_tcpdebug.h" 42 #include "opt_tcp_input.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/mac.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/proc.h> /* for proc0 declaration */ 50 #include <sys/protosw.h> 51 #include <sys/signalvar.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/sysctl.h> 55 #include <sys/syslog.h> 56 #include <sys/systm.h> 57 58 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 59 60 #include <net/if.h> 61 #include <net/route.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip.h> 68 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70 #include <netinet/ip_var.h> 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/in6_pcb.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet6/tcp6_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif /* TCPDEBUG */ 86 87 #ifdef FAST_IPSEC 88 #include <netipsec/ipsec.h> 89 #ifdef INET6 90 #include <netipsec/ipsec6.h> 91 #endif 92 #endif /*FAST_IPSEC*/ 93 94 #ifdef IPSEC 95 #include <netinet6/ipsec.h> 96 #include <netinet6/ipsec6.h> 97 #include <netkey/key.h> 98 #endif /*IPSEC*/ 99 100 #include <machine/in_cksum.h> 101 102 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 103 104 static 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 struct inpcbhead tcb; 131 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 132 struct inpcbinfo tcbinfo; 133 struct mtx *tcbinfo_mtx; 134 135 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 136 static void tcp_pulloutofband(struct socket *, 137 struct tcphdr *, struct mbuf *, int); 138 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 139 struct mbuf *); 140 static void tcp_xmit_timer(struct tcpcb *, int); 141 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 142 143 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 144 #ifdef INET6 145 #define ND6_HINT(tp) \ 146 do { \ 147 if ((tp) && (tp)->t_inpcb && \ 148 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 149 (tp)->t_inpcb->in6p_route.ro_rt) \ 150 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 151 } while (0) 152 #else 153 #define ND6_HINT(tp) 154 #endif 155 156 /* 157 * Indicate whether this ack should be delayed. We can delay the ack if 158 * - delayed acks are enabled and 159 * - there is no delayed ack timer in progress and 160 * - our last ack wasn't a 0-sized window. We never want to delay 161 * the ack that opens up a 0-sized window. 162 */ 163 #define DELAY_ACK(tp) \ 164 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 165 (tp->t_flags & TF_RXWIN0SENT) == 0) 166 167 static int 168 tcp_reass(tp, th, tlenp, m) 169 register struct tcpcb *tp; 170 register struct tcphdr *th; 171 int *tlenp; 172 struct mbuf *m; 173 { 174 struct tseg_qent *q; 175 struct tseg_qent *p = NULL; 176 struct tseg_qent *nq; 177 struct tseg_qent *te; 178 struct socket *so = tp->t_inpcb->inp_socket; 179 int flags; 180 181 /* 182 * Call with th==0 after become established to 183 * force pre-ESTABLISHED data up to user socket. 184 */ 185 if (th == 0) 186 goto present; 187 188 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 189 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 190 M_NOWAIT); 191 if (te == NULL) { 192 tcpstat.tcps_rcvmemdrop++; 193 m_freem(m); 194 return (0); 195 } 196 197 /* 198 * Find a segment which begins after this one does. 199 */ 200 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 201 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 202 break; 203 p = q; 204 } 205 206 /* 207 * If there is a preceding segment, it may provide some of 208 * our data already. If so, drop the data from the incoming 209 * segment. If it provides all of our data, drop us. 210 */ 211 if (p != NULL) { 212 register int i; 213 /* conversion to int (in i) handles seq wraparound */ 214 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 215 if (i > 0) { 216 if (i >= *tlenp) { 217 tcpstat.tcps_rcvduppack++; 218 tcpstat.tcps_rcvdupbyte += *tlenp; 219 m_freem(m); 220 FREE(te, M_TSEGQ); 221 /* 222 * Try to present any queued data 223 * at the left window edge to the user. 224 * This is needed after the 3-WHS 225 * completes. 226 */ 227 goto present; /* ??? */ 228 } 229 m_adj(m, i); 230 *tlenp -= i; 231 th->th_seq += i; 232 } 233 } 234 tcpstat.tcps_rcvoopack++; 235 tcpstat.tcps_rcvoobyte += *tlenp; 236 237 /* 238 * While we overlap succeeding segments trim them or, 239 * if they are completely covered, dequeue them. 240 */ 241 while (q) { 242 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 243 if (i <= 0) 244 break; 245 if (i < q->tqe_len) { 246 q->tqe_th->th_seq += i; 247 q->tqe_len -= i; 248 m_adj(q->tqe_m, i); 249 break; 250 } 251 252 nq = LIST_NEXT(q, tqe_q); 253 LIST_REMOVE(q, tqe_q); 254 m_freem(q->tqe_m); 255 FREE(q, M_TSEGQ); 256 q = nq; 257 } 258 259 /* Insert the new segment queue entry into place. */ 260 te->tqe_m = m; 261 te->tqe_th = th; 262 te->tqe_len = *tlenp; 263 264 if (p == NULL) { 265 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 266 } else { 267 LIST_INSERT_AFTER(p, te, tqe_q); 268 } 269 270 present: 271 /* 272 * Present data to user, advancing rcv_nxt through 273 * completed sequence space. 274 */ 275 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 276 return (0); 277 q = LIST_FIRST(&tp->t_segq); 278 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 279 return (0); 280 do { 281 tp->rcv_nxt += q->tqe_len; 282 flags = q->tqe_th->th_flags & TH_FIN; 283 nq = LIST_NEXT(q, tqe_q); 284 LIST_REMOVE(q, tqe_q); 285 if (so->so_state & SS_CANTRCVMORE) 286 m_freem(q->tqe_m); 287 else 288 sbappend(&so->so_rcv, q->tqe_m); 289 FREE(q, M_TSEGQ); 290 q = nq; 291 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 292 ND6_HINT(tp); 293 sorwakeup(so); 294 return (flags); 295 } 296 297 /* 298 * TCP input routine, follows pages 65-76 of the 299 * protocol specification dated September, 1981 very closely. 300 */ 301 #ifdef INET6 302 int 303 tcp6_input(mp, offp, proto) 304 struct mbuf **mp; 305 int *offp, proto; 306 { 307 register struct mbuf *m = *mp; 308 struct in6_ifaddr *ia6; 309 310 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 311 312 /* 313 * draft-itojun-ipv6-tcp-to-anycast 314 * better place to put this in? 315 */ 316 ia6 = ip6_getdstifaddr(m); 317 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 318 struct ip6_hdr *ip6; 319 320 ip6 = mtod(m, struct ip6_hdr *); 321 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 322 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 323 return IPPROTO_DONE; 324 } 325 326 tcp_input(m, *offp); 327 return IPPROTO_DONE; 328 } 329 #endif 330 331 void 332 tcp_input(m, off0) 333 register struct mbuf *m; 334 int off0; 335 { 336 register struct tcphdr *th; 337 register struct ip *ip = NULL; 338 register struct ipovly *ipov; 339 register struct inpcb *inp = NULL; 340 u_char *optp = NULL; 341 int optlen = 0; 342 int len, tlen, off; 343 int drop_hdrlen; 344 register struct tcpcb *tp = 0; 345 register int thflags; 346 struct socket *so = 0; 347 int todrop, acked, ourfinisacked, needoutput = 0; 348 u_long tiwin; 349 struct tcpopt to; /* options in this segment */ 350 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 351 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 352 int headlocked = 0; 353 struct sockaddr_in *next_hop = NULL; 354 int rstreason; /* For badport_bandlim accounting purposes */ 355 356 struct ip6_hdr *ip6 = NULL; 357 #ifdef INET6 358 int isipv6; 359 #else 360 const int isipv6 = 0; 361 #endif 362 363 #ifdef TCPDEBUG 364 /* 365 * The size of tcp_saveipgen must be the size of the max ip header, 366 * now IPv6. 367 */ 368 u_char tcp_saveipgen[40]; 369 struct tcphdr tcp_savetcp; 370 short ostate = 0; 371 #endif 372 373 #ifdef MAC 374 int error; 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 INP_INFO_WLOCK(&tcbinfo); 529 headlocked = 1; 530 findpcb: 531 /* IPFIREWALL_FORWARD section */ 532 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 533 /* 534 * Transparently forwarded. Pretend to be the destination. 535 * already got one like this? 536 */ 537 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 538 ip->ip_dst, th->th_dport, 539 0, m->m_pkthdr.rcvif); 540 if (!inp) { 541 /* It's new. Try find the ambushing socket. */ 542 inp = in_pcblookup_hash(&tcbinfo, 543 ip->ip_src, th->th_sport, 544 next_hop->sin_addr, 545 next_hop->sin_port ? 546 ntohs(next_hop->sin_port) : 547 th->th_dport, 548 1, m->m_pkthdr.rcvif); 549 } 550 } else { 551 if (isipv6) 552 inp = in6_pcblookup_hash(&tcbinfo, 553 &ip6->ip6_src, th->th_sport, 554 &ip6->ip6_dst, th->th_dport, 555 1, m->m_pkthdr.rcvif); 556 else 557 inp = in_pcblookup_hash(&tcbinfo, 558 ip->ip_src, th->th_sport, 559 ip->ip_dst, th->th_dport, 560 1, m->m_pkthdr.rcvif); 561 } 562 563 #ifdef IPSEC 564 if (isipv6) { 565 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 566 ipsec6stat.in_polvio++; 567 goto drop; 568 } 569 } else { 570 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 571 ipsecstat.in_polvio++; 572 goto drop; 573 } 574 } 575 #endif 576 #ifdef FAST_IPSEC 577 #ifdef INET6 578 if (isipv6) { 579 if (inp != NULL && ipsec6_in_reject(m, inp)) { 580 goto drop; 581 } 582 } else 583 #endif /* INET6 */ 584 if (inp != NULL && ipsec4_in_reject(m, inp)) { 585 goto drop; 586 } 587 #endif /*FAST_IPSEC*/ 588 589 /* 590 * If the state is CLOSED (i.e., TCB does not exist) then 591 * all data in the incoming segment is discarded. 592 * If the TCB exists but is in CLOSED state, it is embryonic, 593 * but should either do a listen or a connect soon. 594 */ 595 if (inp == NULL) { 596 if (log_in_vain) { 597 #ifdef INET6 598 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 599 #else 600 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 601 #endif 602 603 if (isipv6) { 604 strcpy(dbuf, "["); 605 strcpy(sbuf, "["); 606 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 607 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 608 strcat(dbuf, "]"); 609 strcat(sbuf, "]"); 610 } else { 611 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 612 strcpy(sbuf, inet_ntoa(ip->ip_src)); 613 } 614 switch (log_in_vain) { 615 case 1: 616 if (thflags & TH_SYN) 617 log(LOG_INFO, 618 "Connection attempt to TCP %s:%d " 619 "from %s:%d\n", 620 dbuf, ntohs(th->th_dport), sbuf, 621 ntohs(th->th_sport)); 622 break; 623 case 2: 624 log(LOG_INFO, 625 "Connection attempt to TCP %s:%d " 626 "from %s:%d flags:0x%x\n", 627 dbuf, ntohs(th->th_dport), sbuf, 628 ntohs(th->th_sport), thflags); 629 break; 630 default: 631 break; 632 } 633 } 634 if (blackhole) { 635 switch (blackhole) { 636 case 1: 637 if (thflags & TH_SYN) 638 goto drop; 639 break; 640 case 2: 641 goto drop; 642 default: 643 goto drop; 644 } 645 } 646 rstreason = BANDLIM_RST_CLOSEDPORT; 647 goto dropwithreset; 648 } 649 INP_LOCK(inp); 650 tp = intotcpcb(inp); 651 if (tp == 0) { 652 INP_UNLOCK(inp); 653 rstreason = BANDLIM_RST_CLOSEDPORT; 654 goto dropwithreset; 655 } 656 if (tp->t_state == TCPS_CLOSED) 657 goto drop; 658 659 /* Unscale the window into a 32-bit value. */ 660 if ((thflags & TH_SYN) == 0) 661 tiwin = th->th_win << tp->snd_scale; 662 else 663 tiwin = th->th_win; 664 665 so = inp->inp_socket; 666 #ifdef MAC 667 error = mac_check_socket_deliver(so, m); 668 if (error) 669 goto drop; 670 #endif 671 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 672 struct in_conninfo inc; 673 #ifdef TCPDEBUG 674 if (so->so_options & SO_DEBUG) { 675 ostate = tp->t_state; 676 if (isipv6) 677 bcopy((char *)ip6, (char *)tcp_saveipgen, 678 sizeof(*ip6)); 679 else 680 bcopy((char *)ip, (char *)tcp_saveipgen, 681 sizeof(*ip)); 682 tcp_savetcp = *th; 683 } 684 #endif 685 /* skip if this isn't a listen socket */ 686 if ((so->so_options & SO_ACCEPTCONN) == 0) 687 goto after_listen; 688 #ifdef INET6 689 inc.inc_isipv6 = isipv6; 690 #endif 691 if (isipv6) { 692 inc.inc6_faddr = ip6->ip6_src; 693 inc.inc6_laddr = ip6->ip6_dst; 694 inc.inc6_route.ro_rt = NULL; /* XXX */ 695 } else { 696 inc.inc_faddr = ip->ip_src; 697 inc.inc_laddr = ip->ip_dst; 698 inc.inc_route.ro_rt = NULL; /* XXX */ 699 } 700 inc.inc_fport = th->th_sport; 701 inc.inc_lport = th->th_dport; 702 703 /* 704 * If the state is LISTEN then ignore segment if it contains 705 * a RST. If the segment contains an ACK then it is bad and 706 * send a RST. If it does not contain a SYN then it is not 707 * interesting; drop it. 708 * 709 * If the state is SYN_RECEIVED (syncache) and seg contains 710 * an ACK, but not for our SYN/ACK, send a RST. If the seg 711 * contains a RST, check the sequence number to see if it 712 * is a valid reset segment. 713 */ 714 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 715 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 716 if (!syncache_expand(&inc, th, &so, m)) { 717 /* 718 * No syncache entry, or ACK was not 719 * for our SYN/ACK. Send a RST. 720 */ 721 tcpstat.tcps_badsyn++; 722 rstreason = BANDLIM_RST_OPENPORT; 723 goto dropwithreset; 724 } 725 if (so == NULL) { 726 /* 727 * Could not complete 3-way handshake, 728 * connection is being closed down, and 729 * syncache will free mbuf. 730 */ 731 INP_UNLOCK(inp); 732 INP_INFO_WUNLOCK(&tcbinfo); 733 return; 734 } 735 /* 736 * Socket is created in state SYN_RECEIVED. 737 * Continue processing segment. 738 */ 739 INP_UNLOCK(inp); 740 inp = sotoinpcb(so); 741 INP_LOCK(inp); 742 tp = intotcpcb(inp); 743 /* 744 * This is what would have happened in 745 * tcp_output() when the SYN,ACK was sent. 746 */ 747 tp->snd_up = tp->snd_una; 748 tp->snd_max = tp->snd_nxt = tp->iss + 1; 749 tp->last_ack_sent = tp->rcv_nxt; 750 /* 751 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 752 * until the _second_ ACK is received: 753 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 754 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 755 * move to ESTAB, set snd_wnd to tiwin. 756 */ 757 tp->snd_wnd = tiwin; /* unscaled */ 758 goto after_listen; 759 } 760 if (thflags & TH_RST) { 761 syncache_chkrst(&inc, th); 762 goto drop; 763 } 764 if (thflags & TH_ACK) { 765 syncache_badack(&inc); 766 tcpstat.tcps_badsyn++; 767 rstreason = BANDLIM_RST_OPENPORT; 768 goto dropwithreset; 769 } 770 goto drop; 771 } 772 773 /* 774 * Segment's flags are (SYN) or (SYN|FIN). 775 */ 776 #ifdef INET6 777 /* 778 * If deprecated address is forbidden, 779 * we do not accept SYN to deprecated interface 780 * address to prevent any new inbound connection from 781 * getting established. 782 * When we do not accept SYN, we send a TCP RST, 783 * with deprecated source address (instead of dropping 784 * it). We compromise it as it is much better for peer 785 * to send a RST, and RST will be the final packet 786 * for the exchange. 787 * 788 * If we do not forbid deprecated addresses, we accept 789 * the SYN packet. RFC2462 does not suggest dropping 790 * SYN in this case. 791 * If we decipher RFC2462 5.5.4, it says like this: 792 * 1. use of deprecated addr with existing 793 * communication is okay - "SHOULD continue to be 794 * used" 795 * 2. use of it with new communication: 796 * (2a) "SHOULD NOT be used if alternate address 797 * with sufficient scope is available" 798 * (2b) nothing mentioned otherwise. 799 * Here we fall into (2b) case as we have no choice in 800 * our source address selection - we must obey the peer. 801 * 802 * The wording in RFC2462 is confusing, and there are 803 * multiple description text for deprecated address 804 * handling - worse, they are not exactly the same. 805 * I believe 5.5.4 is the best one, so we follow 5.5.4. 806 */ 807 if (isipv6 && !ip6_use_deprecated) { 808 struct in6_ifaddr *ia6; 809 810 if ((ia6 = ip6_getdstifaddr(m)) && 811 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 812 INP_UNLOCK(inp); 813 tp = NULL; 814 rstreason = BANDLIM_RST_OPENPORT; 815 goto dropwithreset; 816 } 817 } 818 #endif 819 /* 820 * If it is from this socket, drop it, it must be forged. 821 * Don't bother responding if the destination was a broadcast. 822 */ 823 if (th->th_dport == th->th_sport) { 824 if (isipv6) { 825 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 826 &ip6->ip6_src)) 827 goto drop; 828 } else { 829 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 830 goto drop; 831 } 832 } 833 /* 834 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 835 * 836 * Note that it is quite possible to receive unicast 837 * link-layer packets with a broadcast IP address. Use 838 * in_broadcast() to find them. 839 */ 840 if (m->m_flags & (M_BCAST|M_MCAST)) 841 goto drop; 842 if (isipv6) { 843 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 844 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 845 goto drop; 846 } else { 847 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 848 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 849 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 850 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 851 goto drop; 852 } 853 /* 854 * SYN appears to be valid; create compressed TCP state 855 * for syncache, or perform t/tcp connection. 856 */ 857 if (so->so_qlen <= so->so_qlimit) { 858 tcp_dooptions(&to, optp, optlen, 1); 859 if (!syncache_add(&inc, &to, th, &so, m)) 860 goto drop; 861 if (so == NULL) { 862 /* 863 * Entry added to syncache, mbuf used to 864 * send SYN,ACK packet. 865 */ 866 KASSERT(headlocked, ("headlocked")); 867 INP_UNLOCK(inp); 868 INP_INFO_WUNLOCK(&tcbinfo); 869 return; 870 } 871 /* 872 * Segment passed TAO tests. 873 */ 874 INP_UNLOCK(inp); 875 inp = sotoinpcb(so); 876 INP_LOCK(inp); 877 tp = intotcpcb(inp); 878 tp->snd_wnd = tiwin; 879 tp->t_starttime = ticks; 880 tp->t_state = TCPS_ESTABLISHED; 881 882 /* 883 * If there is a FIN, or if there is data and the 884 * connection is local, then delay SYN,ACK(SYN) in 885 * the hope of piggy-backing it on a response 886 * segment. Otherwise must send ACK now in case 887 * the other side is slow starting. 888 */ 889 if (DELAY_ACK(tp) && 890 ((thflags & TH_FIN) || 891 (tlen != 0 && 892 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 893 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 894 callout_reset(tp->tt_delack, tcp_delacktime, 895 tcp_timer_delack, tp); 896 tp->t_flags |= TF_NEEDSYN; 897 } else 898 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 899 900 tcpstat.tcps_connects++; 901 soisconnected(so); 902 goto trimthenstep6; 903 } 904 goto drop; 905 } 906 after_listen: 907 908 /* XXX temp debugging */ 909 /* should not happen - syncache should pick up these connections */ 910 if (tp->t_state == TCPS_LISTEN) 911 panic("tcp_input: TCPS_LISTEN"); 912 913 /* 914 * Segment received on connection. 915 * Reset idle time and keep-alive timer. 916 */ 917 tp->t_rcvtime = ticks; 918 if (TCPS_HAVEESTABLISHED(tp->t_state)) 919 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 920 921 /* 922 * Process options. 923 * XXX this is tradtitional behavior, may need to be cleaned up. 924 */ 925 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 926 if (thflags & TH_SYN) { 927 if (to.to_flags & TOF_SCALE) { 928 tp->t_flags |= TF_RCVD_SCALE; 929 tp->requested_s_scale = to.to_requested_s_scale; 930 } 931 if (to.to_flags & TOF_TS) { 932 tp->t_flags |= TF_RCVD_TSTMP; 933 tp->ts_recent = to.to_tsval; 934 tp->ts_recent_age = ticks; 935 } 936 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 937 tp->t_flags |= TF_RCVD_CC; 938 if (to.to_flags & TOF_MSS) 939 tcp_mss(tp, to.to_mss); 940 } 941 942 /* 943 * Header prediction: check for the two common cases 944 * of a uni-directional data xfer. If the packet has 945 * no control flags, is in-sequence, the window didn't 946 * change and we're not retransmitting, it's a 947 * candidate. If the length is zero and the ack moved 948 * forward, we're the sender side of the xfer. Just 949 * free the data acked & wake any higher level process 950 * that was blocked waiting for space. If the length 951 * is non-zero and the ack didn't move, we're the 952 * receiver side. If we're getting packets in-order 953 * (the reassembly queue is empty), add the data to 954 * the socket buffer and note that we need a delayed ack. 955 * Make sure that the hidden state-flags are also off. 956 * Since we check for TCPS_ESTABLISHED above, it can only 957 * be TH_NEEDSYN. 958 */ 959 if (tp->t_state == TCPS_ESTABLISHED && 960 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 961 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 962 ((to.to_flags & TOF_TS) == 0 || 963 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 964 /* 965 * Using the CC option is compulsory if once started: 966 * the segment is OK if no T/TCP was negotiated or 967 * if the segment has a CC option equal to CCrecv 968 */ 969 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 970 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 971 th->th_seq == tp->rcv_nxt && 972 tiwin && tiwin == tp->snd_wnd && 973 tp->snd_nxt == tp->snd_max) { 974 975 /* 976 * If last ACK falls within this segment's sequence numbers, 977 * record the timestamp. 978 * NOTE that the test is modified according to the latest 979 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 980 */ 981 if ((to.to_flags & TOF_TS) != 0 && 982 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 983 tp->ts_recent_age = ticks; 984 tp->ts_recent = to.to_tsval; 985 } 986 987 if (tlen == 0) { 988 if (SEQ_GT(th->th_ack, tp->snd_una) && 989 SEQ_LEQ(th->th_ack, tp->snd_max) && 990 tp->snd_cwnd >= tp->snd_wnd && 991 tp->t_dupacks < tcprexmtthresh) { 992 KASSERT(headlocked, ("headlocked")); 993 INP_INFO_WUNLOCK(&tcbinfo); 994 headlocked = 0; 995 /* 996 * this is a pure ack for outstanding data. 997 */ 998 ++tcpstat.tcps_predack; 999 /* 1000 * "bad retransmit" recovery 1001 */ 1002 if (tp->t_rxtshift == 1 && 1003 ticks < tp->t_badrxtwin) { 1004 tp->snd_cwnd = tp->snd_cwnd_prev; 1005 tp->snd_ssthresh = 1006 tp->snd_ssthresh_prev; 1007 tp->snd_nxt = tp->snd_max; 1008 tp->t_badrxtwin = 0; 1009 } 1010 1011 /* 1012 * Recalculate the transmit timer / rtt. 1013 * 1014 * Some boxes send broken timestamp replies 1015 * during the SYN+ACK phase, ignore 1016 * timestamps of 0 or we could calculate a 1017 * huge RTT and blow up the retransmit timer. 1018 */ 1019 if ((to.to_flags & TOF_TS) != 0 && 1020 to.to_tsecr) { 1021 tcp_xmit_timer(tp, 1022 ticks - to.to_tsecr + 1); 1023 } else if (tp->t_rtttime && 1024 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1025 tcp_xmit_timer(tp, 1026 ticks - tp->t_rtttime); 1027 } 1028 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1029 acked = th->th_ack - tp->snd_una; 1030 tcpstat.tcps_rcvackpack++; 1031 tcpstat.tcps_rcvackbyte += acked; 1032 sbdrop(&so->so_snd, acked); 1033 tp->snd_una = th->th_ack; 1034 tp->t_dupacks = 0; 1035 m_freem(m); 1036 ND6_HINT(tp); /* some progress has been done */ 1037 1038 /* 1039 * If all outstanding data are acked, stop 1040 * retransmit timer, otherwise restart timer 1041 * using current (possibly backed-off) value. 1042 * If process is waiting for space, 1043 * wakeup/selwakeup/signal. If data 1044 * are ready to send, let tcp_output 1045 * decide between more output or persist. 1046 */ 1047 if (tp->snd_una == tp->snd_max) 1048 callout_stop(tp->tt_rexmt); 1049 else if (!callout_active(tp->tt_persist)) 1050 callout_reset(tp->tt_rexmt, 1051 tp->t_rxtcur, 1052 tcp_timer_rexmt, tp); 1053 1054 sowwakeup(so); 1055 if (so->so_snd.sb_cc) 1056 (void) tcp_output(tp); 1057 INP_UNLOCK(inp); 1058 return; 1059 } 1060 } else if (th->th_ack == tp->snd_una && 1061 LIST_EMPTY(&tp->t_segq) && 1062 tlen <= sbspace(&so->so_rcv)) { 1063 KASSERT(headlocked, ("headlocked")); 1064 INP_INFO_WUNLOCK(&tcbinfo); 1065 headlocked = 0; 1066 /* 1067 * this is a pure, in-sequence data packet 1068 * with nothing on the reassembly queue and 1069 * we have enough buffer space to take it. 1070 */ 1071 ++tcpstat.tcps_preddat; 1072 tp->rcv_nxt += tlen; 1073 tcpstat.tcps_rcvpack++; 1074 tcpstat.tcps_rcvbyte += tlen; 1075 ND6_HINT(tp); /* some progress has been done */ 1076 /* 1077 * Add data to socket buffer. 1078 */ 1079 if (so->so_state & SS_CANTRCVMORE) { 1080 m_freem(m); 1081 } else { 1082 m_adj(m, drop_hdrlen); /* delayed header drop */ 1083 sbappend(&so->so_rcv, m); 1084 } 1085 sorwakeup(so); 1086 if (DELAY_ACK(tp)) { 1087 callout_reset(tp->tt_delack, tcp_delacktime, 1088 tcp_timer_delack, tp); 1089 } else { 1090 tp->t_flags |= TF_ACKNOW; 1091 tcp_output(tp); 1092 } 1093 INP_UNLOCK(inp); 1094 return; 1095 } 1096 } 1097 1098 /* 1099 * Calculate amount of space in receive window, 1100 * and then do TCP input processing. 1101 * Receive window is amount of space in rcv queue, 1102 * but not less than advertised window. 1103 */ 1104 { int win; 1105 1106 win = sbspace(&so->so_rcv); 1107 if (win < 0) 1108 win = 0; 1109 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1110 } 1111 1112 switch (tp->t_state) { 1113 1114 /* 1115 * If the state is SYN_RECEIVED: 1116 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1117 */ 1118 case TCPS_SYN_RECEIVED: 1119 if ((thflags & TH_ACK) && 1120 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1121 SEQ_GT(th->th_ack, tp->snd_max))) { 1122 rstreason = BANDLIM_RST_OPENPORT; 1123 goto dropwithreset; 1124 } 1125 break; 1126 1127 /* 1128 * If the state is SYN_SENT: 1129 * if seg contains an ACK, but not for our SYN, drop the input. 1130 * if seg contains a RST, then drop the connection. 1131 * if seg does not contain SYN, then drop it. 1132 * Otherwise this is an acceptable SYN segment 1133 * initialize tp->rcv_nxt and tp->irs 1134 * if seg contains ack then advance tp->snd_una 1135 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1136 * arrange for segment to be acked (eventually) 1137 * continue processing rest of data/controls, beginning with URG 1138 */ 1139 case TCPS_SYN_SENT: 1140 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1141 taop = &tao_noncached; 1142 bzero(taop, sizeof(*taop)); 1143 } 1144 1145 if ((thflags & TH_ACK) && 1146 (SEQ_LEQ(th->th_ack, tp->iss) || 1147 SEQ_GT(th->th_ack, tp->snd_max))) { 1148 /* 1149 * If we have a cached CCsent for the remote host, 1150 * hence we haven't just crashed and restarted, 1151 * do not send a RST. This may be a retransmission 1152 * from the other side after our earlier ACK was lost. 1153 * Our new SYN, when it arrives, will serve as the 1154 * needed ACK. 1155 */ 1156 if (taop->tao_ccsent != 0) 1157 goto drop; 1158 else { 1159 rstreason = BANDLIM_UNLIMITED; 1160 goto dropwithreset; 1161 } 1162 } 1163 if (thflags & TH_RST) { 1164 if (thflags & TH_ACK) 1165 tp = tcp_drop(tp, ECONNREFUSED); 1166 goto drop; 1167 } 1168 if ((thflags & TH_SYN) == 0) 1169 goto drop; 1170 tp->snd_wnd = th->th_win; /* initial send window */ 1171 tp->cc_recv = to.to_cc; /* foreign CC */ 1172 1173 tp->irs = th->th_seq; 1174 tcp_rcvseqinit(tp); 1175 if (thflags & TH_ACK) { 1176 /* 1177 * Our SYN was acked. If segment contains CC.ECHO 1178 * option, check it to make sure this segment really 1179 * matches our SYN. If not, just drop it as old 1180 * duplicate, but send an RST if we're still playing 1181 * by the old rules. If no CC.ECHO option, make sure 1182 * we don't get fooled into using T/TCP. 1183 */ 1184 if (to.to_flags & TOF_CCECHO) { 1185 if (tp->cc_send != to.to_ccecho) { 1186 if (taop->tao_ccsent != 0) 1187 goto drop; 1188 else { 1189 rstreason = BANDLIM_UNLIMITED; 1190 goto dropwithreset; 1191 } 1192 } 1193 } else 1194 tp->t_flags &= ~TF_RCVD_CC; 1195 tcpstat.tcps_connects++; 1196 soisconnected(so); 1197 #ifdef MAC 1198 mac_set_socket_peer_from_mbuf(m, so); 1199 #endif 1200 /* Do window scaling on this connection? */ 1201 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1202 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1203 tp->snd_scale = tp->requested_s_scale; 1204 tp->rcv_scale = tp->request_r_scale; 1205 } 1206 /* Segment is acceptable, update cache if undefined. */ 1207 if (taop->tao_ccsent == 0) 1208 taop->tao_ccsent = to.to_ccecho; 1209 1210 tp->rcv_adv += tp->rcv_wnd; 1211 tp->snd_una++; /* SYN is acked */ 1212 /* 1213 * If there's data, delay ACK; if there's also a FIN 1214 * ACKNOW will be turned on later. 1215 */ 1216 if (DELAY_ACK(tp) && tlen != 0) 1217 callout_reset(tp->tt_delack, tcp_delacktime, 1218 tcp_timer_delack, tp); 1219 else 1220 tp->t_flags |= TF_ACKNOW; 1221 /* 1222 * Received <SYN,ACK> in SYN_SENT[*] state. 1223 * Transitions: 1224 * SYN_SENT --> ESTABLISHED 1225 * SYN_SENT* --> FIN_WAIT_1 1226 */ 1227 tp->t_starttime = ticks; 1228 if (tp->t_flags & TF_NEEDFIN) { 1229 tp->t_state = TCPS_FIN_WAIT_1; 1230 tp->t_flags &= ~TF_NEEDFIN; 1231 thflags &= ~TH_SYN; 1232 } else { 1233 tp->t_state = TCPS_ESTABLISHED; 1234 callout_reset(tp->tt_keep, tcp_keepidle, 1235 tcp_timer_keep, tp); 1236 } 1237 } else { 1238 /* 1239 * Received initial SYN in SYN-SENT[*] state => 1240 * simultaneous open. If segment contains CC option 1241 * and there is a cached CC, apply TAO test. 1242 * If it succeeds, connection is * half-synchronized. 1243 * Otherwise, do 3-way handshake: 1244 * SYN-SENT -> SYN-RECEIVED 1245 * SYN-SENT* -> SYN-RECEIVED* 1246 * If there was no CC option, clear cached CC value. 1247 */ 1248 tp->t_flags |= TF_ACKNOW; 1249 callout_stop(tp->tt_rexmt); 1250 if (to.to_flags & TOF_CC) { 1251 if (taop->tao_cc != 0 && 1252 CC_GT(to.to_cc, taop->tao_cc)) { 1253 /* 1254 * update cache and make transition: 1255 * SYN-SENT -> ESTABLISHED* 1256 * SYN-SENT* -> FIN-WAIT-1* 1257 */ 1258 taop->tao_cc = to.to_cc; 1259 tp->t_starttime = ticks; 1260 if (tp->t_flags & TF_NEEDFIN) { 1261 tp->t_state = TCPS_FIN_WAIT_1; 1262 tp->t_flags &= ~TF_NEEDFIN; 1263 } else { 1264 tp->t_state = TCPS_ESTABLISHED; 1265 callout_reset(tp->tt_keep, 1266 tcp_keepidle, 1267 tcp_timer_keep, 1268 tp); 1269 } 1270 tp->t_flags |= TF_NEEDSYN; 1271 } else 1272 tp->t_state = TCPS_SYN_RECEIVED; 1273 } else { 1274 /* CC.NEW or no option => invalidate cache */ 1275 taop->tao_cc = 0; 1276 tp->t_state = TCPS_SYN_RECEIVED; 1277 } 1278 } 1279 1280 trimthenstep6: 1281 /* 1282 * Advance th->th_seq to correspond to first data byte. 1283 * If data, trim to stay within window, 1284 * dropping FIN if necessary. 1285 */ 1286 th->th_seq++; 1287 if (tlen > tp->rcv_wnd) { 1288 todrop = tlen - tp->rcv_wnd; 1289 m_adj(m, -todrop); 1290 tlen = tp->rcv_wnd; 1291 thflags &= ~TH_FIN; 1292 tcpstat.tcps_rcvpackafterwin++; 1293 tcpstat.tcps_rcvbyteafterwin += todrop; 1294 } 1295 tp->snd_wl1 = th->th_seq - 1; 1296 tp->rcv_up = th->th_seq; 1297 /* 1298 * Client side of transaction: already sent SYN and data. 1299 * If the remote host used T/TCP to validate the SYN, 1300 * our data will be ACK'd; if so, enter normal data segment 1301 * processing in the middle of step 5, ack processing. 1302 * Otherwise, goto step 6. 1303 */ 1304 if (thflags & TH_ACK) 1305 goto process_ACK; 1306 1307 goto step6; 1308 1309 /* 1310 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1311 * if segment contains a SYN and CC [not CC.NEW] option: 1312 * if state == TIME_WAIT and connection duration > MSL, 1313 * drop packet and send RST; 1314 * 1315 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1316 * ack the FIN (and data) in retransmission queue. 1317 * Complete close and delete TCPCB. Then reprocess 1318 * segment, hoping to find new TCPCB in LISTEN state; 1319 * 1320 * else must be old SYN; drop it. 1321 * else do normal processing. 1322 */ 1323 case TCPS_LAST_ACK: 1324 case TCPS_CLOSING: 1325 case TCPS_TIME_WAIT: 1326 if ((thflags & TH_SYN) && 1327 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1328 if (tp->t_state == TCPS_TIME_WAIT && 1329 (ticks - tp->t_starttime) > tcp_msl) { 1330 rstreason = BANDLIM_UNLIMITED; 1331 goto dropwithreset; 1332 } 1333 if (CC_GT(to.to_cc, tp->cc_recv)) { 1334 tp = tcp_close(tp); 1335 goto findpcb; 1336 } 1337 else 1338 goto drop; 1339 } 1340 break; /* continue normal processing */ 1341 } 1342 1343 /* 1344 * States other than LISTEN or SYN_SENT. 1345 * First check the RST flag and sequence number since reset segments 1346 * are exempt from the timestamp and connection count tests. This 1347 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1348 * below which allowed reset segments in half the sequence space 1349 * to fall though and be processed (which gives forged reset 1350 * segments with a random sequence number a 50 percent chance of 1351 * killing a connection). 1352 * Then check timestamp, if present. 1353 * Then check the connection count, if present. 1354 * Then check that at least some bytes of segment are within 1355 * receive window. If segment begins before rcv_nxt, 1356 * drop leading data (and SYN); if nothing left, just ack. 1357 * 1358 * 1359 * If the RST bit is set, check the sequence number to see 1360 * if this is a valid reset segment. 1361 * RFC 793 page 37: 1362 * In all states except SYN-SENT, all reset (RST) segments 1363 * are validated by checking their SEQ-fields. A reset is 1364 * valid if its sequence number is in the window. 1365 * Note: this does not take into account delayed ACKs, so 1366 * we should test against last_ack_sent instead of rcv_nxt. 1367 * The sequence number in the reset segment is normally an 1368 * echo of our outgoing acknowlegement numbers, but some hosts 1369 * send a reset with the sequence number at the rightmost edge 1370 * of our receive window, and we have to handle this case. 1371 * If we have multiple segments in flight, the intial reset 1372 * segment sequence numbers will be to the left of last_ack_sent, 1373 * but they will eventually catch up. 1374 * In any case, it never made sense to trim reset segments to 1375 * fit the receive window since RFC 1122 says: 1376 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1377 * 1378 * A TCP SHOULD allow a received RST segment to include data. 1379 * 1380 * DISCUSSION 1381 * It has been suggested that a RST segment could contain 1382 * ASCII text that encoded and explained the cause of the 1383 * RST. No standard has yet been established for such 1384 * data. 1385 * 1386 * If the reset segment passes the sequence number test examine 1387 * the state: 1388 * SYN_RECEIVED STATE: 1389 * If passive open, return to LISTEN state. 1390 * If active open, inform user that connection was refused. 1391 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1392 * Inform user that connection was reset, and close tcb. 1393 * CLOSING, LAST_ACK STATES: 1394 * Close the tcb. 1395 * TIME_WAIT STATE: 1396 * Drop the segment - see Stevens, vol. 2, p. 964 and 1397 * RFC 1337. 1398 */ 1399 if (thflags & TH_RST) { 1400 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1401 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1402 switch (tp->t_state) { 1403 1404 case TCPS_SYN_RECEIVED: 1405 so->so_error = ECONNREFUSED; 1406 goto close; 1407 1408 case TCPS_ESTABLISHED: 1409 case TCPS_FIN_WAIT_1: 1410 case TCPS_FIN_WAIT_2: 1411 case TCPS_CLOSE_WAIT: 1412 so->so_error = ECONNRESET; 1413 close: 1414 tp->t_state = TCPS_CLOSED; 1415 tcpstat.tcps_drops++; 1416 tp = tcp_close(tp); 1417 break; 1418 1419 case TCPS_CLOSING: 1420 case TCPS_LAST_ACK: 1421 tp = tcp_close(tp); 1422 break; 1423 1424 case TCPS_TIME_WAIT: 1425 break; 1426 } 1427 } 1428 goto drop; 1429 } 1430 1431 /* 1432 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1433 * and it's less than ts_recent, drop it. 1434 */ 1435 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1436 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1437 1438 /* Check to see if ts_recent is over 24 days old. */ 1439 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1440 /* 1441 * Invalidate ts_recent. If this segment updates 1442 * ts_recent, the age will be reset later and ts_recent 1443 * will get a valid value. If it does not, setting 1444 * ts_recent to zero will at least satisfy the 1445 * requirement that zero be placed in the timestamp 1446 * echo reply when ts_recent isn't valid. The 1447 * age isn't reset until we get a valid ts_recent 1448 * because we don't want out-of-order segments to be 1449 * dropped when ts_recent is old. 1450 */ 1451 tp->ts_recent = 0; 1452 } else { 1453 tcpstat.tcps_rcvduppack++; 1454 tcpstat.tcps_rcvdupbyte += tlen; 1455 tcpstat.tcps_pawsdrop++; 1456 goto dropafterack; 1457 } 1458 } 1459 1460 /* 1461 * T/TCP mechanism 1462 * If T/TCP was negotiated and the segment doesn't have CC, 1463 * or if its CC is wrong then drop the segment. 1464 * RST segments do not have to comply with this. 1465 */ 1466 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1467 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1468 goto dropafterack; 1469 1470 /* 1471 * In the SYN-RECEIVED state, validate that the packet belongs to 1472 * this connection before trimming the data to fit the receive 1473 * window. Check the sequence number versus IRS since we know 1474 * the sequence numbers haven't wrapped. This is a partial fix 1475 * for the "LAND" DoS attack. 1476 */ 1477 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1478 rstreason = BANDLIM_RST_OPENPORT; 1479 goto dropwithreset; 1480 } 1481 1482 todrop = tp->rcv_nxt - th->th_seq; 1483 if (todrop > 0) { 1484 if (thflags & TH_SYN) { 1485 thflags &= ~TH_SYN; 1486 th->th_seq++; 1487 if (th->th_urp > 1) 1488 th->th_urp--; 1489 else 1490 thflags &= ~TH_URG; 1491 todrop--; 1492 } 1493 /* 1494 * Following if statement from Stevens, vol. 2, p. 960. 1495 */ 1496 if (todrop > tlen 1497 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1498 /* 1499 * Any valid FIN must be to the left of the window. 1500 * At this point the FIN must be a duplicate or out 1501 * of sequence; drop it. 1502 */ 1503 thflags &= ~TH_FIN; 1504 1505 /* 1506 * Send an ACK to resynchronize and drop any data. 1507 * But keep on processing for RST or ACK. 1508 */ 1509 tp->t_flags |= TF_ACKNOW; 1510 todrop = tlen; 1511 tcpstat.tcps_rcvduppack++; 1512 tcpstat.tcps_rcvdupbyte += todrop; 1513 } else { 1514 tcpstat.tcps_rcvpartduppack++; 1515 tcpstat.tcps_rcvpartdupbyte += todrop; 1516 } 1517 drop_hdrlen += todrop; /* drop from the top afterwards */ 1518 th->th_seq += todrop; 1519 tlen -= todrop; 1520 if (th->th_urp > todrop) 1521 th->th_urp -= todrop; 1522 else { 1523 thflags &= ~TH_URG; 1524 th->th_urp = 0; 1525 } 1526 } 1527 1528 /* 1529 * If new data are received on a connection after the 1530 * user processes are gone, then RST the other end. 1531 */ 1532 if ((so->so_state & SS_NOFDREF) && 1533 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1534 tp = tcp_close(tp); 1535 tcpstat.tcps_rcvafterclose++; 1536 rstreason = BANDLIM_UNLIMITED; 1537 goto dropwithreset; 1538 } 1539 1540 /* 1541 * If segment ends after window, drop trailing data 1542 * (and PUSH and FIN); if nothing left, just ACK. 1543 */ 1544 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1545 if (todrop > 0) { 1546 tcpstat.tcps_rcvpackafterwin++; 1547 if (todrop >= tlen) { 1548 tcpstat.tcps_rcvbyteafterwin += tlen; 1549 /* 1550 * If a new connection request is received 1551 * while in TIME_WAIT, drop the old connection 1552 * and start over if the sequence numbers 1553 * are above the previous ones. 1554 */ 1555 if (thflags & TH_SYN && 1556 tp->t_state == TCPS_TIME_WAIT && 1557 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1558 tp = tcp_close(tp); 1559 goto findpcb; 1560 } 1561 /* 1562 * If window is closed can only take segments at 1563 * window edge, and have to drop data and PUSH from 1564 * incoming segments. Continue processing, but 1565 * remember to ack. Otherwise, drop segment 1566 * and ack. 1567 */ 1568 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1569 tp->t_flags |= TF_ACKNOW; 1570 tcpstat.tcps_rcvwinprobe++; 1571 } else 1572 goto dropafterack; 1573 } else 1574 tcpstat.tcps_rcvbyteafterwin += todrop; 1575 m_adj(m, -todrop); 1576 tlen -= todrop; 1577 thflags &= ~(TH_PUSH|TH_FIN); 1578 } 1579 1580 /* 1581 * If last ACK falls within this segment's sequence numbers, 1582 * record its timestamp. 1583 * NOTE that the test is modified according to the latest 1584 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1585 */ 1586 if ((to.to_flags & TOF_TS) != 0 && 1587 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1588 tp->ts_recent_age = ticks; 1589 tp->ts_recent = to.to_tsval; 1590 } 1591 1592 /* 1593 * If a SYN is in the window, then this is an 1594 * error and we send an RST and drop the connection. 1595 */ 1596 if (thflags & TH_SYN) { 1597 tp = tcp_drop(tp, ECONNRESET); 1598 rstreason = BANDLIM_UNLIMITED; 1599 goto dropwithreset; 1600 } 1601 1602 /* 1603 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1604 * flag is on (half-synchronized state), then queue data for 1605 * later processing; else drop segment and return. 1606 */ 1607 if ((thflags & TH_ACK) == 0) { 1608 if (tp->t_state == TCPS_SYN_RECEIVED || 1609 (tp->t_flags & TF_NEEDSYN)) 1610 goto step6; 1611 else 1612 goto drop; 1613 } 1614 1615 /* 1616 * Ack processing. 1617 */ 1618 switch (tp->t_state) { 1619 1620 /* 1621 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1622 * ESTABLISHED state and continue processing. 1623 * The ACK was checked above. 1624 */ 1625 case TCPS_SYN_RECEIVED: 1626 1627 tcpstat.tcps_connects++; 1628 soisconnected(so); 1629 /* Do window scaling? */ 1630 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1631 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1632 tp->snd_scale = tp->requested_s_scale; 1633 tp->rcv_scale = tp->request_r_scale; 1634 } 1635 /* 1636 * Upon successful completion of 3-way handshake, 1637 * update cache.CC if it was undefined, pass any queued 1638 * data to the user, and advance state appropriately. 1639 */ 1640 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1641 taop->tao_cc == 0) 1642 taop->tao_cc = tp->cc_recv; 1643 1644 /* 1645 * Make transitions: 1646 * SYN-RECEIVED -> ESTABLISHED 1647 * SYN-RECEIVED* -> FIN-WAIT-1 1648 */ 1649 tp->t_starttime = ticks; 1650 if (tp->t_flags & TF_NEEDFIN) { 1651 tp->t_state = TCPS_FIN_WAIT_1; 1652 tp->t_flags &= ~TF_NEEDFIN; 1653 } else { 1654 tp->t_state = TCPS_ESTABLISHED; 1655 callout_reset(tp->tt_keep, tcp_keepidle, 1656 tcp_timer_keep, tp); 1657 } 1658 /* 1659 * If segment contains data or ACK, will call tcp_reass() 1660 * later; if not, do so now to pass queued data to user. 1661 */ 1662 if (tlen == 0 && (thflags & TH_FIN) == 0) 1663 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1664 (struct mbuf *)0); 1665 tp->snd_wl1 = th->th_seq - 1; 1666 /* FALLTHROUGH */ 1667 1668 /* 1669 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1670 * ACKs. If the ack is in the range 1671 * tp->snd_una < th->th_ack <= tp->snd_max 1672 * then advance tp->snd_una to th->th_ack and drop 1673 * data from the retransmission queue. If this ACK reflects 1674 * more up to date window information we update our window information. 1675 */ 1676 case TCPS_ESTABLISHED: 1677 case TCPS_FIN_WAIT_1: 1678 case TCPS_FIN_WAIT_2: 1679 case TCPS_CLOSE_WAIT: 1680 case TCPS_CLOSING: 1681 case TCPS_LAST_ACK: 1682 case TCPS_TIME_WAIT: 1683 1684 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1685 if (tlen == 0 && tiwin == tp->snd_wnd) { 1686 tcpstat.tcps_rcvdupack++; 1687 /* 1688 * If we have outstanding data (other than 1689 * a window probe), this is a completely 1690 * duplicate ack (ie, window info didn't 1691 * change), the ack is the biggest we've 1692 * seen and we've seen exactly our rexmt 1693 * threshhold of them, assume a packet 1694 * has been dropped and retransmit it. 1695 * Kludge snd_nxt & the congestion 1696 * window so we send only this one 1697 * packet. 1698 * 1699 * We know we're losing at the current 1700 * window size so do congestion avoidance 1701 * (set ssthresh to half the current window 1702 * and pull our congestion window back to 1703 * the new ssthresh). 1704 * 1705 * Dup acks mean that packets have left the 1706 * network (they're now cached at the receiver) 1707 * so bump cwnd by the amount in the receiver 1708 * to keep a constant cwnd packets in the 1709 * network. 1710 */ 1711 if (!callout_active(tp->tt_rexmt) || 1712 th->th_ack != tp->snd_una) 1713 tp->t_dupacks = 0; 1714 else if (++tp->t_dupacks == tcprexmtthresh) { 1715 tcp_seq onxt = tp->snd_nxt; 1716 u_int win = 1717 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1718 tp->t_maxseg; 1719 if (tcp_do_newreno && 1720 SEQ_LT(th->th_ack, 1721 tp->snd_recover)) { 1722 /* False retransmit, should not 1723 * cut window 1724 */ 1725 tp->snd_cwnd += tp->t_maxseg; 1726 tp->t_dupacks = 0; 1727 (void) tcp_output(tp); 1728 goto drop; 1729 } 1730 if (win < 2) 1731 win = 2; 1732 tp->snd_ssthresh = win * tp->t_maxseg; 1733 tp->snd_recover = tp->snd_max; 1734 callout_stop(tp->tt_rexmt); 1735 tp->t_rtttime = 0; 1736 tp->snd_nxt = th->th_ack; 1737 tp->snd_cwnd = tp->t_maxseg; 1738 (void) tcp_output(tp); 1739 tp->snd_cwnd = tp->snd_ssthresh + 1740 tp->t_maxseg * tp->t_dupacks; 1741 if (SEQ_GT(onxt, tp->snd_nxt)) 1742 tp->snd_nxt = onxt; 1743 goto drop; 1744 } else if (tp->t_dupacks > tcprexmtthresh) { 1745 tp->snd_cwnd += tp->t_maxseg; 1746 (void) tcp_output(tp); 1747 goto drop; 1748 } 1749 } else 1750 tp->t_dupacks = 0; 1751 break; 1752 } 1753 /* 1754 * If the congestion window was inflated to account 1755 * for the other side's cached packets, retract it. 1756 */ 1757 if (tcp_do_newreno) { 1758 int is_partialack = SEQ_LT(th->th_ack, tp->snd_recover); 1759 if (tp->t_dupacks >= tcprexmtthresh) { 1760 if (is_partialack) { 1761 tcp_newreno_partial_ack(tp, th); 1762 } else { 1763 /* 1764 * Window inflation should have left us 1765 * with approximately snd_ssthresh 1766 * outstanding data. 1767 * But in case we would be inclined to 1768 * send a burst, better to do it via 1769 * the slow start mechanism. 1770 */ 1771 if (SEQ_GT(th->th_ack + 1772 tp->snd_ssthresh, 1773 tp->snd_max)) 1774 tp->snd_cwnd = tp->snd_max - 1775 th->th_ack + 1776 tp->t_maxseg; 1777 else 1778 tp->snd_cwnd = tp->snd_ssthresh; 1779 } 1780 } 1781 /* 1782 * Reset dupacks, except on partial acks in 1783 * fast recovery. 1784 */ 1785 if (!(tp->t_dupacks >= tcprexmtthresh && is_partialack)) 1786 tp->t_dupacks = 0; 1787 } else { 1788 if (tp->t_dupacks >= tcprexmtthresh && 1789 tp->snd_cwnd > tp->snd_ssthresh) 1790 tp->snd_cwnd = tp->snd_ssthresh; 1791 tp->t_dupacks = 0; 1792 } 1793 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1794 tcpstat.tcps_rcvacktoomuch++; 1795 goto dropafterack; 1796 } 1797 /* 1798 * If we reach this point, ACK is not a duplicate, 1799 * i.e., it ACKs something we sent. 1800 */ 1801 if (tp->t_flags & TF_NEEDSYN) { 1802 /* 1803 * T/TCP: Connection was half-synchronized, and our 1804 * SYN has been ACK'd (so connection is now fully 1805 * synchronized). Go to non-starred state, 1806 * increment snd_una for ACK of SYN, and check if 1807 * we can do window scaling. 1808 */ 1809 tp->t_flags &= ~TF_NEEDSYN; 1810 tp->snd_una++; 1811 /* Do window scaling? */ 1812 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1813 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1814 tp->snd_scale = tp->requested_s_scale; 1815 tp->rcv_scale = tp->request_r_scale; 1816 } 1817 } 1818 1819 process_ACK: 1820 acked = th->th_ack - tp->snd_una; 1821 tcpstat.tcps_rcvackpack++; 1822 tcpstat.tcps_rcvackbyte += acked; 1823 1824 /* 1825 * If we just performed our first retransmit, and the ACK 1826 * arrives within our recovery window, then it was a mistake 1827 * to do the retransmit in the first place. Recover our 1828 * original cwnd and ssthresh, and proceed to transmit where 1829 * we left off. 1830 */ 1831 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1832 ++tcpstat.tcps_sndrexmitbad; 1833 tp->snd_cwnd = tp->snd_cwnd_prev; 1834 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1835 tp->snd_nxt = tp->snd_max; 1836 tp->t_badrxtwin = 0; /* XXX probably not required */ 1837 } 1838 1839 /* 1840 * If we have a timestamp reply, update smoothed 1841 * round trip time. If no timestamp is present but 1842 * transmit timer is running and timed sequence 1843 * number was acked, update smoothed round trip time. 1844 * Since we now have an rtt measurement, cancel the 1845 * timer backoff (cf., Phil Karn's retransmit alg.). 1846 * Recompute the initial retransmit timer. 1847 * 1848 * Some boxes send broken timestamp replies 1849 * during the SYN+ACK phase, ignore 1850 * timestamps of 0 or we could calculate a 1851 * huge RTT and blow up the retransmit timer. 1852 */ 1853 if ((to.to_flags & TOF_TS) != 0 && 1854 to.to_tsecr) { 1855 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1856 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1857 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1858 } 1859 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1860 1861 /* 1862 * If all outstanding data is acked, stop retransmit 1863 * timer and remember to restart (more output or persist). 1864 * If there is more data to be acked, restart retransmit 1865 * timer, using current (possibly backed-off) value. 1866 */ 1867 if (th->th_ack == tp->snd_max) { 1868 callout_stop(tp->tt_rexmt); 1869 needoutput = 1; 1870 } else if (!callout_active(tp->tt_persist)) 1871 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1872 tcp_timer_rexmt, tp); 1873 1874 /* 1875 * If no data (only SYN) was ACK'd, 1876 * skip rest of ACK processing. 1877 */ 1878 if (acked == 0) 1879 goto step6; 1880 1881 /* 1882 * When new data is acked, open the congestion window. 1883 * If the window gives us less than ssthresh packets 1884 * in flight, open exponentially (maxseg per packet). 1885 * Otherwise open linearly: maxseg per window 1886 * (maxseg^2 / cwnd per packet). 1887 */ 1888 { 1889 register u_int cw = tp->snd_cwnd; 1890 register u_int incr = tp->t_maxseg; 1891 1892 if (cw > tp->snd_ssthresh) 1893 incr = incr * incr / cw; 1894 /* 1895 * If t_dupacks != 0 here, it indicates that we are still 1896 * in NewReno fast recovery mode, so we leave the congestion 1897 * window alone. 1898 */ 1899 if (!tcp_do_newreno || tp->t_dupacks == 0) 1900 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1901 } 1902 if (acked > so->so_snd.sb_cc) { 1903 tp->snd_wnd -= so->so_snd.sb_cc; 1904 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1905 ourfinisacked = 1; 1906 } else { 1907 sbdrop(&so->so_snd, acked); 1908 tp->snd_wnd -= acked; 1909 ourfinisacked = 0; 1910 } 1911 sowwakeup(so); 1912 tp->snd_una = th->th_ack; 1913 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1914 tp->snd_nxt = tp->snd_una; 1915 1916 switch (tp->t_state) { 1917 1918 /* 1919 * In FIN_WAIT_1 STATE in addition to the processing 1920 * for the ESTABLISHED state if our FIN is now acknowledged 1921 * then enter FIN_WAIT_2. 1922 */ 1923 case TCPS_FIN_WAIT_1: 1924 if (ourfinisacked) { 1925 /* 1926 * If we can't receive any more 1927 * data, then closing user can proceed. 1928 * Starting the timer is contrary to the 1929 * specification, but if we don't get a FIN 1930 * we'll hang forever. 1931 */ 1932 if (so->so_state & SS_CANTRCVMORE) { 1933 soisdisconnected(so); 1934 callout_reset(tp->tt_2msl, tcp_maxidle, 1935 tcp_timer_2msl, tp); 1936 } 1937 tp->t_state = TCPS_FIN_WAIT_2; 1938 } 1939 break; 1940 1941 /* 1942 * In CLOSING STATE in addition to the processing for 1943 * the ESTABLISHED state if the ACK acknowledges our FIN 1944 * then enter the TIME-WAIT state, otherwise ignore 1945 * the segment. 1946 */ 1947 case TCPS_CLOSING: 1948 if (ourfinisacked) { 1949 tp->t_state = TCPS_TIME_WAIT; 1950 tcp_canceltimers(tp); 1951 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1952 if (tp->cc_recv != 0 && 1953 (ticks - tp->t_starttime) < tcp_msl) 1954 callout_reset(tp->tt_2msl, 1955 tp->t_rxtcur * 1956 TCPTV_TWTRUNC, 1957 tcp_timer_2msl, tp); 1958 else 1959 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1960 tcp_timer_2msl, tp); 1961 soisdisconnected(so); 1962 } 1963 break; 1964 1965 /* 1966 * In LAST_ACK, we may still be waiting for data to drain 1967 * and/or to be acked, as well as for the ack of our FIN. 1968 * If our FIN is now acknowledged, delete the TCB, 1969 * enter the closed state and return. 1970 */ 1971 case TCPS_LAST_ACK: 1972 if (ourfinisacked) { 1973 tp = tcp_close(tp); 1974 goto drop; 1975 } 1976 break; 1977 1978 /* 1979 * In TIME_WAIT state the only thing that should arrive 1980 * is a retransmission of the remote FIN. Acknowledge 1981 * it and restart the finack timer. 1982 */ 1983 case TCPS_TIME_WAIT: 1984 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1985 tcp_timer_2msl, tp); 1986 goto dropafterack; 1987 } 1988 } 1989 1990 step6: 1991 /* 1992 * Update window information. 1993 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1994 */ 1995 if ((thflags & TH_ACK) && 1996 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1997 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1998 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1999 /* keep track of pure window updates */ 2000 if (tlen == 0 && 2001 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2002 tcpstat.tcps_rcvwinupd++; 2003 tp->snd_wnd = tiwin; 2004 tp->snd_wl1 = th->th_seq; 2005 tp->snd_wl2 = th->th_ack; 2006 if (tp->snd_wnd > tp->max_sndwnd) 2007 tp->max_sndwnd = tp->snd_wnd; 2008 needoutput = 1; 2009 } 2010 2011 /* 2012 * Process segments with URG. 2013 */ 2014 if ((thflags & TH_URG) && th->th_urp && 2015 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2016 /* 2017 * This is a kludge, but if we receive and accept 2018 * random urgent pointers, we'll crash in 2019 * soreceive. It's hard to imagine someone 2020 * actually wanting to send this much urgent data. 2021 */ 2022 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2023 th->th_urp = 0; /* XXX */ 2024 thflags &= ~TH_URG; /* XXX */ 2025 goto dodata; /* XXX */ 2026 } 2027 /* 2028 * If this segment advances the known urgent pointer, 2029 * then mark the data stream. This should not happen 2030 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2031 * a FIN has been received from the remote side. 2032 * In these states we ignore the URG. 2033 * 2034 * According to RFC961 (Assigned Protocols), 2035 * the urgent pointer points to the last octet 2036 * of urgent data. We continue, however, 2037 * to consider it to indicate the first octet 2038 * of data past the urgent section as the original 2039 * spec states (in one of two places). 2040 */ 2041 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2042 tp->rcv_up = th->th_seq + th->th_urp; 2043 so->so_oobmark = so->so_rcv.sb_cc + 2044 (tp->rcv_up - tp->rcv_nxt) - 1; 2045 if (so->so_oobmark == 0) 2046 so->so_state |= SS_RCVATMARK; 2047 sohasoutofband(so); 2048 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2049 } 2050 /* 2051 * Remove out of band data so doesn't get presented to user. 2052 * This can happen independent of advancing the URG pointer, 2053 * but if two URG's are pending at once, some out-of-band 2054 * data may creep in... ick. 2055 */ 2056 if (th->th_urp <= (u_long)tlen 2057 #ifdef SO_OOBINLINE 2058 && (so->so_options & SO_OOBINLINE) == 0 2059 #endif 2060 ) 2061 tcp_pulloutofband(so, th, m, 2062 drop_hdrlen); /* hdr drop is delayed */ 2063 } else { 2064 /* 2065 * If no out of band data is expected, 2066 * pull receive urgent pointer along 2067 * with the receive window. 2068 */ 2069 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2070 tp->rcv_up = tp->rcv_nxt; 2071 } 2072 dodata: /* XXX */ 2073 KASSERT(headlocked, ("headlocked")); 2074 INP_INFO_WUNLOCK(&tcbinfo); 2075 headlocked = 0; 2076 /* 2077 * Process the segment text, merging it into the TCP sequencing queue, 2078 * and arranging for acknowledgment of receipt if necessary. 2079 * This process logically involves adjusting tp->rcv_wnd as data 2080 * is presented to the user (this happens in tcp_usrreq.c, 2081 * case PRU_RCVD). If a FIN has already been received on this 2082 * connection then we just ignore the text. 2083 */ 2084 if ((tlen || (thflags & TH_FIN)) && 2085 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2086 m_adj(m, drop_hdrlen); /* delayed header drop */ 2087 /* 2088 * Insert segment which includes th into TCP reassembly queue 2089 * with control block tp. Set thflags to whether reassembly now 2090 * includes a segment with FIN. This handles the common case 2091 * inline (segment is the next to be received on an established 2092 * connection, and the queue is empty), avoiding linkage into 2093 * and removal from the queue and repetition of various 2094 * conversions. 2095 * Set DELACK for segments received in order, but ack 2096 * immediately when segments are out of order (so 2097 * fast retransmit can work). 2098 */ 2099 if (th->th_seq == tp->rcv_nxt && 2100 LIST_EMPTY(&tp->t_segq) && 2101 TCPS_HAVEESTABLISHED(tp->t_state)) { 2102 if (DELAY_ACK(tp)) 2103 callout_reset(tp->tt_delack, tcp_delacktime, 2104 tcp_timer_delack, tp); 2105 else 2106 tp->t_flags |= TF_ACKNOW; 2107 tp->rcv_nxt += tlen; 2108 thflags = th->th_flags & TH_FIN; 2109 tcpstat.tcps_rcvpack++; 2110 tcpstat.tcps_rcvbyte += tlen; 2111 ND6_HINT(tp); 2112 if (so->so_state & SS_CANTRCVMORE) 2113 m_freem(m); 2114 else 2115 sbappend(&so->so_rcv, m); 2116 sorwakeup(so); 2117 } else { 2118 thflags = tcp_reass(tp, th, &tlen, m); 2119 tp->t_flags |= TF_ACKNOW; 2120 } 2121 2122 /* 2123 * Note the amount of data that peer has sent into 2124 * our window, in order to estimate the sender's 2125 * buffer size. 2126 */ 2127 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2128 } else { 2129 m_freem(m); 2130 thflags &= ~TH_FIN; 2131 } 2132 2133 /* 2134 * If FIN is received ACK the FIN and let the user know 2135 * that the connection is closing. 2136 */ 2137 if (thflags & TH_FIN) { 2138 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2139 socantrcvmore(so); 2140 /* 2141 * If connection is half-synchronized 2142 * (ie NEEDSYN flag on) then delay ACK, 2143 * so it may be piggybacked when SYN is sent. 2144 * Otherwise, since we received a FIN then no 2145 * more input can be expected, send ACK now. 2146 */ 2147 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2148 callout_reset(tp->tt_delack, tcp_delacktime, 2149 tcp_timer_delack, tp); 2150 else 2151 tp->t_flags |= TF_ACKNOW; 2152 tp->rcv_nxt++; 2153 } 2154 switch (tp->t_state) { 2155 2156 /* 2157 * In SYN_RECEIVED and ESTABLISHED STATES 2158 * enter the CLOSE_WAIT state. 2159 */ 2160 case TCPS_SYN_RECEIVED: 2161 tp->t_starttime = ticks; 2162 /*FALLTHROUGH*/ 2163 case TCPS_ESTABLISHED: 2164 tp->t_state = TCPS_CLOSE_WAIT; 2165 break; 2166 2167 /* 2168 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2169 * enter the CLOSING state. 2170 */ 2171 case TCPS_FIN_WAIT_1: 2172 tp->t_state = TCPS_CLOSING; 2173 break; 2174 2175 /* 2176 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2177 * starting the time-wait timer, turning off the other 2178 * standard timers. 2179 */ 2180 case TCPS_FIN_WAIT_2: 2181 tp->t_state = TCPS_TIME_WAIT; 2182 tcp_canceltimers(tp); 2183 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2184 if (tp->cc_recv != 0 && 2185 (ticks - tp->t_starttime) < tcp_msl) { 2186 callout_reset(tp->tt_2msl, 2187 tp->t_rxtcur * TCPTV_TWTRUNC, 2188 tcp_timer_2msl, tp); 2189 /* For transaction client, force ACK now. */ 2190 tp->t_flags |= TF_ACKNOW; 2191 } 2192 else 2193 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2194 tcp_timer_2msl, tp); 2195 soisdisconnected(so); 2196 break; 2197 2198 /* 2199 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2200 */ 2201 case TCPS_TIME_WAIT: 2202 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2203 tcp_timer_2msl, tp); 2204 break; 2205 } 2206 } 2207 #ifdef TCPDEBUG 2208 if (so->so_options & SO_DEBUG) 2209 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2210 &tcp_savetcp, 0); 2211 #endif 2212 2213 /* 2214 * Return any desired output. 2215 */ 2216 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2217 (void) tcp_output(tp); 2218 INP_UNLOCK(inp); 2219 return; 2220 2221 dropafterack: 2222 /* 2223 * Generate an ACK dropping incoming segment if it occupies 2224 * sequence space, where the ACK reflects our state. 2225 * 2226 * We can now skip the test for the RST flag since all 2227 * paths to this code happen after packets containing 2228 * RST have been dropped. 2229 * 2230 * In the SYN-RECEIVED state, don't send an ACK unless the 2231 * segment we received passes the SYN-RECEIVED ACK test. 2232 * If it fails send a RST. This breaks the loop in the 2233 * "LAND" DoS attack, and also prevents an ACK storm 2234 * between two listening ports that have been sent forged 2235 * SYN segments, each with the source address of the other. 2236 */ 2237 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2238 (SEQ_GT(tp->snd_una, th->th_ack) || 2239 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2240 rstreason = BANDLIM_RST_OPENPORT; 2241 goto dropwithreset; 2242 } 2243 #ifdef TCPDEBUG 2244 if (so->so_options & SO_DEBUG) 2245 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2246 &tcp_savetcp, 0); 2247 #endif 2248 if (headlocked) 2249 INP_INFO_WUNLOCK(&tcbinfo); 2250 m_freem(m); 2251 tp->t_flags |= TF_ACKNOW; 2252 (void) tcp_output(tp); 2253 INP_UNLOCK(inp); 2254 return; 2255 2256 dropwithreset: 2257 /* 2258 * Generate a RST, dropping incoming segment. 2259 * Make ACK acceptable to originator of segment. 2260 * Don't bother to respond if destination was broadcast/multicast. 2261 */ 2262 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2263 goto drop; 2264 if (isipv6) { 2265 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2266 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2267 goto drop; 2268 } else { 2269 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2270 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2271 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2272 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2273 goto drop; 2274 } 2275 /* IPv6 anycast check is done at tcp6_input() */ 2276 2277 /* 2278 * Perform bandwidth limiting. 2279 */ 2280 if (badport_bandlim(rstreason) < 0) 2281 goto drop; 2282 2283 #ifdef TCPDEBUG 2284 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2285 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2286 &tcp_savetcp, 0); 2287 #endif 2288 2289 if (tp) 2290 INP_UNLOCK(inp); 2291 2292 if (thflags & TH_ACK) 2293 /* mtod() below is safe as long as hdr dropping is delayed */ 2294 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2295 TH_RST); 2296 else { 2297 if (thflags & TH_SYN) 2298 tlen++; 2299 /* mtod() below is safe as long as hdr dropping is delayed */ 2300 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2301 (tcp_seq)0, TH_RST|TH_ACK); 2302 } 2303 if (headlocked) 2304 INP_INFO_WUNLOCK(&tcbinfo); 2305 return; 2306 2307 drop: 2308 /* 2309 * Drop space held by incoming segment and return. 2310 */ 2311 #ifdef TCPDEBUG 2312 if (tp == 0 || (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 if (tp) 2317 INP_UNLOCK(inp); 2318 m_freem(m); 2319 if (headlocked) 2320 INP_INFO_WUNLOCK(&tcbinfo); 2321 return; 2322 } 2323 2324 /* 2325 * Parse TCP options and place in tcpopt. 2326 */ 2327 static void 2328 tcp_dooptions(to, cp, cnt, is_syn) 2329 struct tcpopt *to; 2330 u_char *cp; 2331 int cnt; 2332 { 2333 int opt, optlen; 2334 2335 to->to_flags = 0; 2336 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2337 opt = cp[0]; 2338 if (opt == TCPOPT_EOL) 2339 break; 2340 if (opt == TCPOPT_NOP) 2341 optlen = 1; 2342 else { 2343 if (cnt < 2) 2344 break; 2345 optlen = cp[1]; 2346 if (optlen < 2 || optlen > cnt) 2347 break; 2348 } 2349 switch (opt) { 2350 case TCPOPT_MAXSEG: 2351 if (optlen != TCPOLEN_MAXSEG) 2352 continue; 2353 if (!is_syn) 2354 continue; 2355 to->to_flags |= TOF_MSS; 2356 bcopy((char *)cp + 2, 2357 (char *)&to->to_mss, sizeof(to->to_mss)); 2358 to->to_mss = ntohs(to->to_mss); 2359 break; 2360 case TCPOPT_WINDOW: 2361 if (optlen != TCPOLEN_WINDOW) 2362 continue; 2363 if (! is_syn) 2364 continue; 2365 to->to_flags |= TOF_SCALE; 2366 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2367 break; 2368 case TCPOPT_TIMESTAMP: 2369 if (optlen != TCPOLEN_TIMESTAMP) 2370 continue; 2371 to->to_flags |= TOF_TS; 2372 bcopy((char *)cp + 2, 2373 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2374 to->to_tsval = ntohl(to->to_tsval); 2375 bcopy((char *)cp + 6, 2376 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2377 to->to_tsecr = ntohl(to->to_tsecr); 2378 break; 2379 case TCPOPT_CC: 2380 if (optlen != TCPOLEN_CC) 2381 continue; 2382 to->to_flags |= TOF_CC; 2383 bcopy((char *)cp + 2, 2384 (char *)&to->to_cc, sizeof(to->to_cc)); 2385 to->to_cc = ntohl(to->to_cc); 2386 break; 2387 case TCPOPT_CCNEW: 2388 if (optlen != TCPOLEN_CC) 2389 continue; 2390 if (!is_syn) 2391 continue; 2392 to->to_flags |= TOF_CCNEW; 2393 bcopy((char *)cp + 2, 2394 (char *)&to->to_cc, sizeof(to->to_cc)); 2395 to->to_cc = ntohl(to->to_cc); 2396 break; 2397 case TCPOPT_CCECHO: 2398 if (optlen != TCPOLEN_CC) 2399 continue; 2400 if (!is_syn) 2401 continue; 2402 to->to_flags |= TOF_CCECHO; 2403 bcopy((char *)cp + 2, 2404 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2405 to->to_ccecho = ntohl(to->to_ccecho); 2406 break; 2407 default: 2408 continue; 2409 } 2410 } 2411 } 2412 2413 /* 2414 * Pull out of band byte out of a segment so 2415 * it doesn't appear in the user's data queue. 2416 * It is still reflected in the segment length for 2417 * sequencing purposes. 2418 */ 2419 static void 2420 tcp_pulloutofband(so, th, m, off) 2421 struct socket *so; 2422 struct tcphdr *th; 2423 register struct mbuf *m; 2424 int off; /* delayed to be droped hdrlen */ 2425 { 2426 int cnt = off + th->th_urp - 1; 2427 2428 while (cnt >= 0) { 2429 if (m->m_len > cnt) { 2430 char *cp = mtod(m, caddr_t) + cnt; 2431 struct tcpcb *tp = sototcpcb(so); 2432 2433 tp->t_iobc = *cp; 2434 tp->t_oobflags |= TCPOOB_HAVEDATA; 2435 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2436 m->m_len--; 2437 if (m->m_flags & M_PKTHDR) 2438 m->m_pkthdr.len--; 2439 return; 2440 } 2441 cnt -= m->m_len; 2442 m = m->m_next; 2443 if (m == 0) 2444 break; 2445 } 2446 panic("tcp_pulloutofband"); 2447 } 2448 2449 /* 2450 * Collect new round-trip time estimate 2451 * and update averages and current timeout. 2452 */ 2453 static void 2454 tcp_xmit_timer(tp, rtt) 2455 register struct tcpcb *tp; 2456 int rtt; 2457 { 2458 register int delta; 2459 2460 tcpstat.tcps_rttupdated++; 2461 tp->t_rttupdated++; 2462 if (tp->t_srtt != 0) { 2463 /* 2464 * srtt is stored as fixed point with 5 bits after the 2465 * binary point (i.e., scaled by 8). The following magic 2466 * is equivalent to the smoothing algorithm in rfc793 with 2467 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2468 * point). Adjust rtt to origin 0. 2469 */ 2470 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2471 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2472 2473 if ((tp->t_srtt += delta) <= 0) 2474 tp->t_srtt = 1; 2475 2476 /* 2477 * We accumulate a smoothed rtt variance (actually, a 2478 * smoothed mean difference), then set the retransmit 2479 * timer to smoothed rtt + 4 times the smoothed variance. 2480 * rttvar is stored as fixed point with 4 bits after the 2481 * binary point (scaled by 16). The following is 2482 * equivalent to rfc793 smoothing with an alpha of .75 2483 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2484 * rfc793's wired-in beta. 2485 */ 2486 if (delta < 0) 2487 delta = -delta; 2488 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2489 if ((tp->t_rttvar += delta) <= 0) 2490 tp->t_rttvar = 1; 2491 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2492 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2493 } else { 2494 /* 2495 * No rtt measurement yet - use the unsmoothed rtt. 2496 * Set the variance to half the rtt (so our first 2497 * retransmit happens at 3*rtt). 2498 */ 2499 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2500 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2501 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2502 } 2503 tp->t_rtttime = 0; 2504 tp->t_rxtshift = 0; 2505 2506 /* 2507 * the retransmit should happen at rtt + 4 * rttvar. 2508 * Because of the way we do the smoothing, srtt and rttvar 2509 * will each average +1/2 tick of bias. When we compute 2510 * the retransmit timer, we want 1/2 tick of rounding and 2511 * 1 extra tick because of +-1/2 tick uncertainty in the 2512 * firing of the timer. The bias will give us exactly the 2513 * 1.5 tick we need. But, because the bias is 2514 * statistical, we have to test that we don't drop below 2515 * the minimum feasible timer (which is 2 ticks). 2516 */ 2517 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2518 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2519 2520 /* 2521 * We received an ack for a packet that wasn't retransmitted; 2522 * it is probably safe to discard any error indications we've 2523 * received recently. This isn't quite right, but close enough 2524 * for now (a route might have failed after we sent a segment, 2525 * and the return path might not be symmetrical). 2526 */ 2527 tp->t_softerror = 0; 2528 } 2529 2530 /* 2531 * Determine a reasonable value for maxseg size. 2532 * If the route is known, check route for mtu. 2533 * If none, use an mss that can be handled on the outgoing 2534 * interface without forcing IP to fragment; if bigger than 2535 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2536 * to utilize large mbufs. If no route is found, route has no mtu, 2537 * or the destination isn't local, use a default, hopefully conservative 2538 * size (usually 512 or the default IP max size, but no more than the mtu 2539 * of the interface), as we can't discover anything about intervening 2540 * gateways or networks. We also initialize the congestion/slow start 2541 * window to be a single segment if the destination isn't local. 2542 * While looking at the routing entry, we also initialize other path-dependent 2543 * parameters from pre-set or cached values in the routing entry. 2544 * 2545 * Also take into account the space needed for options that we 2546 * send regularly. Make maxseg shorter by that amount to assure 2547 * that we can send maxseg amount of data even when the options 2548 * are present. Store the upper limit of the length of options plus 2549 * data in maxopd. 2550 * 2551 * NOTE that this routine is only called when we process an incoming 2552 * segment, for outgoing segments only tcp_mssopt is called. 2553 * 2554 * In case of T/TCP, we call this routine during implicit connection 2555 * setup as well (offer = -1), to initialize maxseg from the cached 2556 * MSS of our peer. 2557 */ 2558 void 2559 tcp_mss(tp, offer) 2560 struct tcpcb *tp; 2561 int offer; 2562 { 2563 register struct rtentry *rt; 2564 struct ifnet *ifp; 2565 register int rtt, mss; 2566 u_long bufsize; 2567 struct inpcb *inp = tp->t_inpcb; 2568 struct socket *so; 2569 struct rmxp_tao *taop; 2570 int origoffer = offer; 2571 #ifdef INET6 2572 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2573 size_t min_protoh = isipv6 ? 2574 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2575 sizeof (struct tcpiphdr); 2576 #else 2577 const int isipv6 = 0; 2578 const size_t min_protoh = sizeof (struct tcpiphdr); 2579 #endif 2580 2581 if (isipv6) 2582 rt = tcp_rtlookup6(&inp->inp_inc); 2583 else 2584 rt = tcp_rtlookup(&inp->inp_inc); 2585 if (rt == NULL) { 2586 tp->t_maxopd = tp->t_maxseg = 2587 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2588 return; 2589 } 2590 ifp = rt->rt_ifp; 2591 so = inp->inp_socket; 2592 2593 taop = rmx_taop(rt->rt_rmx); 2594 /* 2595 * Offer == -1 means that we didn't receive SYN yet, 2596 * use cached value in that case; 2597 */ 2598 if (offer == -1) 2599 offer = taop->tao_mssopt; 2600 /* 2601 * Offer == 0 means that there was no MSS on the SYN segment, 2602 * in this case we use tcp_mssdflt. 2603 */ 2604 if (offer == 0) 2605 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2606 else 2607 /* 2608 * Sanity check: make sure that maxopd will be large 2609 * enough to allow some data on segments even is the 2610 * all the option space is used (40bytes). Otherwise 2611 * funny things may happen in tcp_output. 2612 */ 2613 offer = max(offer, 64); 2614 taop->tao_mssopt = offer; 2615 2616 /* 2617 * While we're here, check if there's an initial rtt 2618 * or rttvar. Convert from the route-table units 2619 * to scaled multiples of the slow timeout timer. 2620 */ 2621 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2622 /* 2623 * XXX the lock bit for RTT indicates that the value 2624 * is also a minimum value; this is subject to time. 2625 */ 2626 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2627 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2628 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2629 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2630 tcpstat.tcps_usedrtt++; 2631 if (rt->rt_rmx.rmx_rttvar) { 2632 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2633 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2634 tcpstat.tcps_usedrttvar++; 2635 } else { 2636 /* default variation is +- 1 rtt */ 2637 tp->t_rttvar = 2638 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2639 } 2640 TCPT_RANGESET(tp->t_rxtcur, 2641 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2642 tp->t_rttmin, TCPTV_REXMTMAX); 2643 } 2644 /* 2645 * if there's an mtu associated with the route, use it 2646 * else, use the link mtu. 2647 */ 2648 if (rt->rt_rmx.rmx_mtu) 2649 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2650 else { 2651 if (isipv6) { 2652 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2653 min_protoh; 2654 if (!in6_localaddr(&inp->in6p_faddr)) 2655 mss = min(mss, tcp_v6mssdflt); 2656 } else { 2657 mss = ifp->if_mtu - min_protoh; 2658 if (!in_localaddr(inp->inp_faddr)) 2659 mss = min(mss, tcp_mssdflt); 2660 } 2661 } 2662 mss = min(mss, offer); 2663 /* 2664 * maxopd stores the maximum length of data AND options 2665 * in a segment; maxseg is the amount of data in a normal 2666 * segment. We need to store this value (maxopd) apart 2667 * from maxseg, because now every segment carries options 2668 * and thus we normally have somewhat less data in segments. 2669 */ 2670 tp->t_maxopd = mss; 2671 2672 /* 2673 * In case of T/TCP, origoffer==-1 indicates, that no segments 2674 * were received yet. In this case we just guess, otherwise 2675 * we do the same as before T/TCP. 2676 */ 2677 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2678 (origoffer == -1 || 2679 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2680 mss -= TCPOLEN_TSTAMP_APPA; 2681 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2682 (origoffer == -1 || 2683 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2684 mss -= TCPOLEN_CC_APPA; 2685 2686 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2687 if (mss > MCLBYTES) 2688 mss &= ~(MCLBYTES-1); 2689 #else 2690 if (mss > MCLBYTES) 2691 mss = mss / MCLBYTES * MCLBYTES; 2692 #endif 2693 /* 2694 * If there's a pipesize, change the socket buffer 2695 * to that size. Make the socket buffers an integral 2696 * number of mss units; if the mss is larger than 2697 * the socket buffer, decrease the mss. 2698 */ 2699 #ifdef RTV_SPIPE 2700 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2701 #endif 2702 bufsize = so->so_snd.sb_hiwat; 2703 if (bufsize < mss) 2704 mss = bufsize; 2705 else { 2706 bufsize = roundup(bufsize, mss); 2707 if (bufsize > sb_max) 2708 bufsize = sb_max; 2709 if (bufsize > so->so_snd.sb_hiwat) 2710 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2711 } 2712 tp->t_maxseg = mss; 2713 2714 #ifdef RTV_RPIPE 2715 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2716 #endif 2717 bufsize = so->so_rcv.sb_hiwat; 2718 if (bufsize > mss) { 2719 bufsize = roundup(bufsize, mss); 2720 if (bufsize > sb_max) 2721 bufsize = sb_max; 2722 if (bufsize > so->so_rcv.sb_hiwat) 2723 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2724 } 2725 2726 /* 2727 * Set the slow-start flight size depending on whether this 2728 * is a local network or not. 2729 */ 2730 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2731 (!isipv6 && in_localaddr(inp->inp_faddr))) 2732 tp->snd_cwnd = mss * ss_fltsz_local; 2733 else 2734 tp->snd_cwnd = mss * ss_fltsz; 2735 2736 if (rt->rt_rmx.rmx_ssthresh) { 2737 /* 2738 * There's some sort of gateway or interface 2739 * buffer limit on the path. Use this to set 2740 * the slow start threshhold, but set the 2741 * threshold to no less than 2*mss. 2742 */ 2743 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2744 tcpstat.tcps_usedssthresh++; 2745 } 2746 } 2747 2748 /* 2749 * Determine the MSS option to send on an outgoing SYN. 2750 */ 2751 int 2752 tcp_mssopt(tp) 2753 struct tcpcb *tp; 2754 { 2755 struct rtentry *rt; 2756 #ifdef INET6 2757 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2758 size_t min_protoh = isipv6 ? 2759 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2760 sizeof (struct tcpiphdr); 2761 #else 2762 const int isipv6 = 0; 2763 const size_t min_protoh = sizeof (struct tcpiphdr); 2764 #endif 2765 2766 if (isipv6) 2767 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2768 else 2769 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2770 if (rt == NULL) 2771 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2772 2773 return (rt->rt_ifp->if_mtu - min_protoh); 2774 } 2775 2776 2777 /* 2778 * On a partial ack arrives, force the retransmission of the 2779 * next unacknowledged segment. Do not clear tp->t_dupacks. 2780 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2781 * be started again. 2782 */ 2783 static void 2784 tcp_newreno_partial_ack(tp, th) 2785 struct tcpcb *tp; 2786 struct tcphdr *th; 2787 { 2788 tcp_seq onxt = tp->snd_nxt; 2789 u_long ocwnd = tp->snd_cwnd; 2790 2791 callout_stop(tp->tt_rexmt); 2792 tp->t_rtttime = 0; 2793 tp->snd_nxt = th->th_ack; 2794 /* 2795 * Set snd_cwnd to one segment beyond acknowledged offset. 2796 * (tp->snd_una has not yet been updated when this function is called.) 2797 */ 2798 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2799 tp->t_flags |= TF_ACKNOW; 2800 (void) tcp_output(tp); 2801 tp->snd_cwnd = ocwnd; 2802 if (SEQ_GT(onxt, tp->snd_nxt)) 2803 tp->snd_nxt = onxt; 2804 /* 2805 * Partial window deflation. Relies on fact that tp->snd_una 2806 * not updated yet. 2807 */ 2808 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2809 } 2810