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 #include <netipsec/ipsec6.h> 90 #endif /*FAST_IPSEC*/ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #include <netinet6/ipsec6.h> 95 #include <netkey/key.h> 96 #endif /*IPSEC*/ 97 98 #include <machine/in_cksum.h> 99 100 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 101 102 static const int tcprexmtthresh = 3; 103 tcp_cc tcp_ccgen; 104 105 struct tcpstat tcpstat; 106 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 107 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 108 109 static int log_in_vain = 0; 110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 111 &log_in_vain, 0, "Log all incoming TCP connections"); 112 113 static int blackhole = 0; 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 115 &blackhole, 0, "Do not send RST when dropping refused connections"); 116 117 int tcp_delack_enabled = 1; 118 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 119 &tcp_delack_enabled, 0, 120 "Delay ACK to try and piggyback it onto a data packet"); 121 122 #ifdef TCP_DROP_SYNFIN 123 static int drop_synfin = 0; 124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 125 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 126 #endif 127 128 static int tcp_do_rfc3042 = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 130 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 131 132 static int tcp_do_rfc3390 = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 134 &tcp_do_rfc3390, 0, 135 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 136 137 struct inpcbhead tcb; 138 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 139 struct inpcbinfo tcbinfo; 140 struct mtx *tcbinfo_mtx; 141 142 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 143 static void tcp_pulloutofband(struct socket *, 144 struct tcphdr *, struct mbuf *, int); 145 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 146 struct mbuf *); 147 static void tcp_xmit_timer(struct tcpcb *, int); 148 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 149 static int tcp_timewait(struct tcptw *, struct tcpopt *, 150 struct tcphdr *, struct mbuf *, int); 151 152 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 153 #ifdef INET6 154 #define ND6_HINT(tp) \ 155 do { \ 156 if ((tp) && (tp)->t_inpcb && \ 157 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 158 (tp)->t_inpcb->in6p_route.ro_rt) \ 159 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 160 } while (0) 161 #else 162 #define ND6_HINT(tp) 163 #endif 164 165 /* 166 * Indicate whether this ack should be delayed. We can delay the ack if 167 * - there is no delayed ack timer in progress and 168 * - our last ack wasn't a 0-sized window. We never want to delay 169 * the ack that opens up a 0-sized window and 170 * - delayed acks are enabled or 171 * - this is a half-synchronized T/TCP connection. 172 */ 173 #define DELAY_ACK(tp) \ 174 ((!callout_active(tp->tt_delack) && \ 175 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 176 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 177 178 static int 179 tcp_reass(tp, th, tlenp, m) 180 register struct tcpcb *tp; 181 register struct tcphdr *th; 182 int *tlenp; 183 struct mbuf *m; 184 { 185 struct tseg_qent *q; 186 struct tseg_qent *p = NULL; 187 struct tseg_qent *nq; 188 struct tseg_qent *te; 189 struct socket *so = tp->t_inpcb->inp_socket; 190 int flags; 191 192 /* 193 * Call with th==0 after become established to 194 * force pre-ESTABLISHED data up to user socket. 195 */ 196 if (th == 0) 197 goto present; 198 199 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 200 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 201 M_NOWAIT); 202 if (te == NULL) { 203 tcpstat.tcps_rcvmemdrop++; 204 m_freem(m); 205 return (0); 206 } 207 208 /* 209 * Find a segment which begins after this one does. 210 */ 211 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 212 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 213 break; 214 p = q; 215 } 216 217 /* 218 * If there is a preceding segment, it may provide some of 219 * our data already. If so, drop the data from the incoming 220 * segment. If it provides all of our data, drop us. 221 */ 222 if (p != NULL) { 223 register int i; 224 /* conversion to int (in i) handles seq wraparound */ 225 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 226 if (i > 0) { 227 if (i >= *tlenp) { 228 tcpstat.tcps_rcvduppack++; 229 tcpstat.tcps_rcvdupbyte += *tlenp; 230 m_freem(m); 231 FREE(te, M_TSEGQ); 232 /* 233 * Try to present any queued data 234 * at the left window edge to the user. 235 * This is needed after the 3-WHS 236 * completes. 237 */ 238 goto present; /* ??? */ 239 } 240 m_adj(m, i); 241 *tlenp -= i; 242 th->th_seq += i; 243 } 244 } 245 tcpstat.tcps_rcvoopack++; 246 tcpstat.tcps_rcvoobyte += *tlenp; 247 248 /* 249 * While we overlap succeeding segments trim them or, 250 * if they are completely covered, dequeue them. 251 */ 252 while (q) { 253 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 254 if (i <= 0) 255 break; 256 if (i < q->tqe_len) { 257 q->tqe_th->th_seq += i; 258 q->tqe_len -= i; 259 m_adj(q->tqe_m, i); 260 break; 261 } 262 263 nq = LIST_NEXT(q, tqe_q); 264 LIST_REMOVE(q, tqe_q); 265 m_freem(q->tqe_m); 266 FREE(q, M_TSEGQ); 267 q = nq; 268 } 269 270 /* Insert the new segment queue entry into place. */ 271 te->tqe_m = m; 272 te->tqe_th = th; 273 te->tqe_len = *tlenp; 274 275 if (p == NULL) { 276 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 277 } else { 278 LIST_INSERT_AFTER(p, te, tqe_q); 279 } 280 281 present: 282 /* 283 * Present data to user, advancing rcv_nxt through 284 * completed sequence space. 285 */ 286 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 287 return (0); 288 q = LIST_FIRST(&tp->t_segq); 289 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 290 return (0); 291 do { 292 tp->rcv_nxt += q->tqe_len; 293 flags = q->tqe_th->th_flags & TH_FIN; 294 nq = LIST_NEXT(q, tqe_q); 295 LIST_REMOVE(q, tqe_q); 296 if (so->so_state & SS_CANTRCVMORE) 297 m_freem(q->tqe_m); 298 else 299 sbappend(&so->so_rcv, q->tqe_m); 300 FREE(q, M_TSEGQ); 301 q = nq; 302 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 303 ND6_HINT(tp); 304 sorwakeup(so); 305 return (flags); 306 } 307 308 /* 309 * TCP input routine, follows pages 65-76 of the 310 * protocol specification dated September, 1981 very closely. 311 */ 312 #ifdef INET6 313 int 314 tcp6_input(mp, offp, proto) 315 struct mbuf **mp; 316 int *offp, proto; 317 { 318 register struct mbuf *m = *mp; 319 struct in6_ifaddr *ia6; 320 321 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 322 323 /* 324 * draft-itojun-ipv6-tcp-to-anycast 325 * better place to put this in? 326 */ 327 ia6 = ip6_getdstifaddr(m); 328 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 329 struct ip6_hdr *ip6; 330 331 ip6 = mtod(m, struct ip6_hdr *); 332 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 333 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 334 return IPPROTO_DONE; 335 } 336 337 tcp_input(m, *offp); 338 return IPPROTO_DONE; 339 } 340 #endif 341 342 void 343 tcp_input(m, off0) 344 register struct mbuf *m; 345 int off0; 346 { 347 register struct tcphdr *th; 348 register struct ip *ip = NULL; 349 register struct ipovly *ipov; 350 register struct inpcb *inp = NULL; 351 u_char *optp = NULL; 352 int optlen = 0; 353 int len, tlen, off; 354 int drop_hdrlen; 355 register struct tcpcb *tp = 0; 356 register int thflags; 357 struct socket *so = 0; 358 int todrop, acked, ourfinisacked, needoutput = 0; 359 u_long tiwin; 360 struct tcpopt to; /* options in this segment */ 361 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 362 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 363 int headlocked = 0; 364 struct sockaddr_in *next_hop = NULL; 365 int rstreason; /* For badport_bandlim accounting purposes */ 366 367 struct ip6_hdr *ip6 = NULL; 368 #ifdef INET6 369 int isipv6; 370 #else 371 const int isipv6 = 0; 372 #endif 373 374 #ifdef TCPDEBUG 375 /* 376 * The size of tcp_saveipgen must be the size of the max ip header, 377 * now IPv6. 378 */ 379 u_char tcp_saveipgen[40]; 380 struct tcphdr tcp_savetcp; 381 short ostate = 0; 382 #endif 383 384 /* Grab info from MT_TAG mbufs prepended to the chain. */ 385 for (;m && m->m_type == MT_TAG; m = m->m_next) { 386 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 387 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 388 } 389 #ifdef INET6 390 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 391 #endif 392 bzero((char *)&to, sizeof(to)); 393 394 tcpstat.tcps_rcvtotal++; 395 396 if (isipv6) { 397 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 398 ip6 = mtod(m, struct ip6_hdr *); 399 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 400 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 401 tcpstat.tcps_rcvbadsum++; 402 goto drop; 403 } 404 th = (struct tcphdr *)((caddr_t)ip6 + off0); 405 406 /* 407 * Be proactive about unspecified IPv6 address in source. 408 * As we use all-zero to indicate unbounded/unconnected pcb, 409 * unspecified IPv6 address can be used to confuse us. 410 * 411 * Note that packets with unspecified IPv6 destination is 412 * already dropped in ip6_input. 413 */ 414 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 415 /* XXX stat */ 416 goto drop; 417 } 418 } else { 419 /* 420 * Get IP and TCP header together in first mbuf. 421 * Note: IP leaves IP header in first mbuf. 422 */ 423 if (off0 > sizeof (struct ip)) { 424 ip_stripoptions(m, (struct mbuf *)0); 425 off0 = sizeof(struct ip); 426 } 427 if (m->m_len < sizeof (struct tcpiphdr)) { 428 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 429 tcpstat.tcps_rcvshort++; 430 return; 431 } 432 } 433 ip = mtod(m, struct ip *); 434 ipov = (struct ipovly *)ip; 435 th = (struct tcphdr *)((caddr_t)ip + off0); 436 tlen = ip->ip_len; 437 438 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 439 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 440 th->th_sum = m->m_pkthdr.csum_data; 441 else 442 th->th_sum = in_pseudo(ip->ip_src.s_addr, 443 ip->ip_dst.s_addr, 444 htonl(m->m_pkthdr.csum_data + 445 ip->ip_len + 446 IPPROTO_TCP)); 447 th->th_sum ^= 0xffff; 448 } else { 449 /* 450 * Checksum extended TCP header and data. 451 */ 452 len = sizeof (struct ip) + tlen; 453 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 454 ipov->ih_len = (u_short)tlen; 455 ipov->ih_len = htons(ipov->ih_len); 456 th->th_sum = in_cksum(m, len); 457 } 458 if (th->th_sum) { 459 tcpstat.tcps_rcvbadsum++; 460 goto drop; 461 } 462 #ifdef INET6 463 /* Re-initialization for later version check */ 464 ip->ip_v = IPVERSION; 465 #endif 466 } 467 468 /* 469 * Check that TCP offset makes sense, 470 * pull out TCP options and adjust length. XXX 471 */ 472 off = th->th_off << 2; 473 if (off < sizeof (struct tcphdr) || off > tlen) { 474 tcpstat.tcps_rcvbadoff++; 475 goto drop; 476 } 477 tlen -= off; /* tlen is used instead of ti->ti_len */ 478 if (off > sizeof (struct tcphdr)) { 479 if (isipv6) { 480 IP6_EXTHDR_CHECK(m, off0, off, ); 481 ip6 = mtod(m, struct ip6_hdr *); 482 th = (struct tcphdr *)((caddr_t)ip6 + off0); 483 } else { 484 if (m->m_len < sizeof(struct ip) + off) { 485 if ((m = m_pullup(m, sizeof (struct ip) + off)) 486 == 0) { 487 tcpstat.tcps_rcvshort++; 488 return; 489 } 490 ip = mtod(m, struct ip *); 491 ipov = (struct ipovly *)ip; 492 th = (struct tcphdr *)((caddr_t)ip + off0); 493 } 494 } 495 optlen = off - sizeof (struct tcphdr); 496 optp = (u_char *)(th + 1); 497 } 498 thflags = th->th_flags; 499 500 #ifdef TCP_DROP_SYNFIN 501 /* 502 * If the drop_synfin option is enabled, drop all packets with 503 * both the SYN and FIN bits set. This prevents e.g. nmap from 504 * identifying the TCP/IP stack. 505 * 506 * This is a violation of the TCP specification. 507 */ 508 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 509 goto drop; 510 #endif 511 512 /* 513 * Convert TCP protocol specific fields to host format. 514 */ 515 th->th_seq = ntohl(th->th_seq); 516 th->th_ack = ntohl(th->th_ack); 517 th->th_win = ntohs(th->th_win); 518 th->th_urp = ntohs(th->th_urp); 519 520 /* 521 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 522 * until after ip6_savecontrol() is called and before other functions 523 * which don't want those proto headers. 524 * Because ip6_savecontrol() is going to parse the mbuf to 525 * search for data to be passed up to user-land, it wants mbuf 526 * parameters to be unchanged. 527 * XXX: the call of ip6_savecontrol() has been obsoleted based on 528 * latest version of the advanced API (20020110). 529 */ 530 drop_hdrlen = off0 + off; 531 532 /* 533 * Locate pcb for segment. 534 */ 535 INP_INFO_WLOCK(&tcbinfo); 536 headlocked = 1; 537 findpcb: 538 /* IPFIREWALL_FORWARD section */ 539 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 540 /* 541 * Transparently forwarded. Pretend to be the destination. 542 * already got one like this? 543 */ 544 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 545 ip->ip_dst, th->th_dport, 546 0, m->m_pkthdr.rcvif); 547 if (!inp) { 548 /* It's new. Try find the ambushing socket. */ 549 inp = in_pcblookup_hash(&tcbinfo, 550 ip->ip_src, th->th_sport, 551 next_hop->sin_addr, 552 next_hop->sin_port ? 553 ntohs(next_hop->sin_port) : 554 th->th_dport, 555 1, m->m_pkthdr.rcvif); 556 } 557 } else { 558 if (isipv6) 559 inp = in6_pcblookup_hash(&tcbinfo, 560 &ip6->ip6_src, th->th_sport, 561 &ip6->ip6_dst, th->th_dport, 562 1, m->m_pkthdr.rcvif); 563 else 564 inp = in_pcblookup_hash(&tcbinfo, 565 ip->ip_src, th->th_sport, 566 ip->ip_dst, th->th_dport, 567 1, m->m_pkthdr.rcvif); 568 } 569 570 #ifdef IPSEC 571 if (isipv6) { 572 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 573 ipsec6stat.in_polvio++; 574 goto drop; 575 } 576 } else { 577 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 578 ipsecstat.in_polvio++; 579 goto drop; 580 } 581 } 582 #endif 583 #ifdef FAST_IPSEC 584 if (isipv6) { 585 if (inp != NULL && ipsec6_in_reject(m, inp)) { 586 goto drop; 587 } 588 } else 589 if (inp != NULL && ipsec4_in_reject(m, inp)) { 590 goto drop; 591 } 592 #endif /*FAST_IPSEC*/ 593 594 /* 595 * If the state is CLOSED (i.e., TCB does not exist) then 596 * all data in the incoming segment is discarded. 597 * If the TCB exists but is in CLOSED state, it is embryonic, 598 * but should either do a listen or a connect soon. 599 */ 600 if (inp == NULL) { 601 if (log_in_vain) { 602 #ifdef INET6 603 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 604 #else 605 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 606 #endif 607 608 if (isipv6) { 609 strcpy(dbuf, "["); 610 strcpy(sbuf, "["); 611 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 612 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 613 strcat(dbuf, "]"); 614 strcat(sbuf, "]"); 615 } else { 616 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 617 strcpy(sbuf, inet_ntoa(ip->ip_src)); 618 } 619 switch (log_in_vain) { 620 case 1: 621 if ((thflags & TH_SYN) == 0) 622 break; 623 case 2: 624 log(LOG_INFO, 625 "Connection attempt to TCP %s:%d " 626 "from %s:%d flags:0x%02x\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 if (inp->inp_vflag & INP_TIMEWAIT) { 651 /* 652 * The only option of relevance is TOF_CC, and only if 653 * present in a SYN segment. See tcp_timewait(). 654 */ 655 if (thflags & TH_SYN) 656 tcp_dooptions(&to, optp, optlen, 1); 657 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 658 &to, th, m, tlen)) 659 goto findpcb; 660 /* 661 * tcp_timewait unlocks inp. 662 */ 663 INP_INFO_WUNLOCK(&tcbinfo); 664 return; 665 } 666 tp = intotcpcb(inp); 667 if (tp == 0) { 668 INP_UNLOCK(inp); 669 rstreason = BANDLIM_RST_CLOSEDPORT; 670 goto dropwithreset; 671 } 672 if (tp->t_state == TCPS_CLOSED) 673 goto drop; 674 675 /* Unscale the window into a 32-bit value. */ 676 if ((thflags & TH_SYN) == 0) 677 tiwin = th->th_win << tp->snd_scale; 678 else 679 tiwin = th->th_win; 680 681 so = inp->inp_socket; 682 #ifdef MAC 683 if (mac_check_socket_deliver(so, m)) 684 goto drop; 685 #endif 686 #ifdef TCPDEBUG 687 if (so->so_options & SO_DEBUG) { 688 ostate = tp->t_state; 689 if (isipv6) 690 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 691 else 692 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 693 tcp_savetcp = *th; 694 } 695 #endif 696 if (so->so_options & SO_ACCEPTCONN) { 697 struct in_conninfo inc; 698 699 #ifdef INET6 700 inc.inc_isipv6 = isipv6; 701 #endif 702 if (isipv6) { 703 inc.inc6_faddr = ip6->ip6_src; 704 inc.inc6_laddr = ip6->ip6_dst; 705 inc.inc6_route.ro_rt = NULL; /* XXX */ 706 } else { 707 inc.inc_faddr = ip->ip_src; 708 inc.inc_laddr = ip->ip_dst; 709 inc.inc_route.ro_rt = NULL; /* XXX */ 710 } 711 inc.inc_fport = th->th_sport; 712 inc.inc_lport = th->th_dport; 713 714 /* 715 * If the state is LISTEN then ignore segment if it contains 716 * a RST. If the segment contains an ACK then it is bad and 717 * send a RST. If it does not contain a SYN then it is not 718 * interesting; drop it. 719 * 720 * If the state is SYN_RECEIVED (syncache) and seg contains 721 * an ACK, but not for our SYN/ACK, send a RST. If the seg 722 * contains a RST, check the sequence number to see if it 723 * is a valid reset segment. 724 */ 725 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 726 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 727 if (!syncache_expand(&inc, th, &so, m)) { 728 /* 729 * No syncache entry, or ACK was not 730 * for our SYN/ACK. Send a RST. 731 */ 732 tcpstat.tcps_badsyn++; 733 rstreason = BANDLIM_RST_OPENPORT; 734 goto dropwithreset; 735 } 736 if (so == NULL) { 737 /* 738 * Could not complete 3-way handshake, 739 * connection is being closed down, and 740 * syncache will free mbuf. 741 */ 742 INP_UNLOCK(inp); 743 INP_INFO_WUNLOCK(&tcbinfo); 744 return; 745 } 746 /* 747 * Socket is created in state SYN_RECEIVED. 748 * Continue processing segment. 749 */ 750 INP_UNLOCK(inp); 751 inp = sotoinpcb(so); 752 INP_LOCK(inp); 753 tp = intotcpcb(inp); 754 /* 755 * This is what would have happened in 756 * tcp_output() when the SYN,ACK was sent. 757 */ 758 tp->snd_up = tp->snd_una; 759 tp->snd_max = tp->snd_nxt = tp->iss + 1; 760 tp->last_ack_sent = tp->rcv_nxt; 761 /* 762 * RFC1323: The window in SYN & SYN/ACK 763 * segments is never scaled. 764 */ 765 tp->snd_wnd = tiwin; /* unscaled */ 766 goto after_listen; 767 } 768 if (thflags & TH_RST) { 769 syncache_chkrst(&inc, th); 770 goto drop; 771 } 772 if (thflags & TH_ACK) { 773 syncache_badack(&inc); 774 tcpstat.tcps_badsyn++; 775 rstreason = BANDLIM_RST_OPENPORT; 776 goto dropwithreset; 777 } 778 goto drop; 779 } 780 781 /* 782 * Segment's flags are (SYN) or (SYN|FIN). 783 */ 784 #ifdef INET6 785 /* 786 * If deprecated address is forbidden, 787 * we do not accept SYN to deprecated interface 788 * address to prevent any new inbound connection from 789 * getting established. 790 * When we do not accept SYN, we send a TCP RST, 791 * with deprecated source address (instead of dropping 792 * it). We compromise it as it is much better for peer 793 * to send a RST, and RST will be the final packet 794 * for the exchange. 795 * 796 * If we do not forbid deprecated addresses, we accept 797 * the SYN packet. RFC2462 does not suggest dropping 798 * SYN in this case. 799 * If we decipher RFC2462 5.5.4, it says like this: 800 * 1. use of deprecated addr with existing 801 * communication is okay - "SHOULD continue to be 802 * used" 803 * 2. use of it with new communication: 804 * (2a) "SHOULD NOT be used if alternate address 805 * with sufficient scope is available" 806 * (2b) nothing mentioned otherwise. 807 * Here we fall into (2b) case as we have no choice in 808 * our source address selection - we must obey the peer. 809 * 810 * The wording in RFC2462 is confusing, and there are 811 * multiple description text for deprecated address 812 * handling - worse, they are not exactly the same. 813 * I believe 5.5.4 is the best one, so we follow 5.5.4. 814 */ 815 if (isipv6 && !ip6_use_deprecated) { 816 struct in6_ifaddr *ia6; 817 818 if ((ia6 = ip6_getdstifaddr(m)) && 819 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 820 INP_UNLOCK(inp); 821 tp = NULL; 822 rstreason = BANDLIM_RST_OPENPORT; 823 goto dropwithreset; 824 } 825 } 826 #endif 827 /* 828 * If it is from this socket, drop it, it must be forged. 829 * Don't bother responding if the destination was a broadcast. 830 */ 831 if (th->th_dport == th->th_sport) { 832 if (isipv6) { 833 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 834 &ip6->ip6_src)) 835 goto drop; 836 } else { 837 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 838 goto drop; 839 } 840 } 841 /* 842 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 843 * 844 * Note that it is quite possible to receive unicast 845 * link-layer packets with a broadcast IP address. Use 846 * in_broadcast() to find them. 847 */ 848 if (m->m_flags & (M_BCAST|M_MCAST)) 849 goto drop; 850 if (isipv6) { 851 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 852 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 853 goto drop; 854 } else { 855 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 856 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 857 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 858 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 859 goto drop; 860 } 861 /* 862 * SYN appears to be valid; create compressed TCP state 863 * for syncache, or perform t/tcp connection. 864 */ 865 if (so->so_qlen <= so->so_qlimit) { 866 tcp_dooptions(&to, optp, optlen, 1); 867 if (!syncache_add(&inc, &to, th, &so, m)) 868 goto drop; 869 if (so == NULL) { 870 /* 871 * Entry added to syncache, mbuf used to 872 * send SYN,ACK packet. 873 */ 874 KASSERT(headlocked, ("headlocked")); 875 INP_UNLOCK(inp); 876 INP_INFO_WUNLOCK(&tcbinfo); 877 return; 878 } 879 /* 880 * Segment passed TAO tests. 881 */ 882 INP_UNLOCK(inp); 883 inp = sotoinpcb(so); 884 INP_LOCK(inp); 885 tp = intotcpcb(inp); 886 tp->snd_wnd = tiwin; 887 tp->t_starttime = ticks; 888 tp->t_state = TCPS_ESTABLISHED; 889 890 /* 891 * T/TCP logic: 892 * If there is a FIN or if there is data, then 893 * delay SYN,ACK(SYN) in the hope of piggy-backing 894 * it on a response segment. Otherwise must send 895 * ACK now in case the other side is slow starting. 896 */ 897 if (thflags & TH_FIN || tlen != 0) 898 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 899 else 900 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 901 tcpstat.tcps_connects++; 902 soisconnected(so); 903 goto trimthenstep6; 904 } 905 goto drop; 906 } 907 after_listen: 908 909 /* XXX temp debugging */ 910 /* should not happen - syncache should pick up these connections */ 911 if (tp->t_state == TCPS_LISTEN) 912 panic("tcp_input: TCPS_LISTEN"); 913 914 /* 915 * Segment received on connection. 916 * Reset idle time and keep-alive timer. 917 */ 918 tp->t_rcvtime = ticks; 919 if (TCPS_HAVEESTABLISHED(tp->t_state)) 920 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 921 922 /* 923 * Process options. 924 * XXX this is tradtitional behavior, may need to be cleaned up. 925 */ 926 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 927 if (thflags & TH_SYN) { 928 if (to.to_flags & TOF_SCALE) { 929 tp->t_flags |= TF_RCVD_SCALE; 930 tp->requested_s_scale = to.to_requested_s_scale; 931 } 932 if (to.to_flags & TOF_TS) { 933 tp->t_flags |= TF_RCVD_TSTMP; 934 tp->ts_recent = to.to_tsval; 935 tp->ts_recent_age = ticks; 936 } 937 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 938 tp->t_flags |= TF_RCVD_CC; 939 if (to.to_flags & TOF_MSS) 940 tcp_mss(tp, to.to_mss); 941 } 942 943 /* 944 * Header prediction: check for the two common cases 945 * of a uni-directional data xfer. If the packet has 946 * no control flags, is in-sequence, the window didn't 947 * change and we're not retransmitting, it's a 948 * candidate. If the length is zero and the ack moved 949 * forward, we're the sender side of the xfer. Just 950 * free the data acked & wake any higher level process 951 * that was blocked waiting for space. If the length 952 * is non-zero and the ack didn't move, we're the 953 * receiver side. If we're getting packets in-order 954 * (the reassembly queue is empty), add the data to 955 * the socket buffer and note that we need a delayed ack. 956 * Make sure that the hidden state-flags are also off. 957 * Since we check for TCPS_ESTABLISHED above, it can only 958 * be TH_NEEDSYN. 959 */ 960 if (tp->t_state == TCPS_ESTABLISHED && 961 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 962 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 963 ((to.to_flags & TOF_TS) == 0 || 964 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 965 /* 966 * Using the CC option is compulsory if once started: 967 * the segment is OK if no T/TCP was negotiated or 968 * if the segment has a CC option equal to CCrecv 969 */ 970 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 971 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 972 th->th_seq == tp->rcv_nxt && 973 tiwin && tiwin == tp->snd_wnd && 974 tp->snd_nxt == tp->snd_max) { 975 976 /* 977 * If last ACK falls within this segment's sequence numbers, 978 * record the timestamp. 979 * NOTE that the test is modified according to the latest 980 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 981 */ 982 if ((to.to_flags & TOF_TS) != 0 && 983 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 984 tp->ts_recent_age = ticks; 985 tp->ts_recent = to.to_tsval; 986 } 987 988 if (tlen == 0) { 989 if (SEQ_GT(th->th_ack, tp->snd_una) && 990 SEQ_LEQ(th->th_ack, tp->snd_max) && 991 tp->snd_cwnd >= tp->snd_wnd && 992 ((!tcp_do_newreno && 993 tp->t_dupacks < tcprexmtthresh) || 994 (tcp_do_newreno && 995 !SEQ_LT(tp->snd_una, tp->snd_recover)))) { 996 KASSERT(headlocked, ("headlocked")); 997 INP_INFO_WUNLOCK(&tcbinfo); 998 /* 999 * this is a pure ack for outstanding data. 1000 */ 1001 ++tcpstat.tcps_predack; 1002 /* 1003 * "bad retransmit" recovery 1004 */ 1005 if (tp->t_rxtshift == 1 && 1006 ticks < tp->t_badrxtwin) { 1007 ++tcpstat.tcps_sndrexmitbad; 1008 tp->snd_cwnd = tp->snd_cwnd_prev; 1009 tp->snd_ssthresh = 1010 tp->snd_ssthresh_prev; 1011 tp->snd_high = tp->snd_high_prev; 1012 tp->snd_nxt = tp->snd_max; 1013 tp->t_badrxtwin = 0; 1014 } 1015 1016 /* 1017 * Recalculate the transmit timer / rtt. 1018 * 1019 * Some boxes send broken timestamp replies 1020 * during the SYN+ACK phase, ignore 1021 * timestamps of 0 or we could calculate a 1022 * huge RTT and blow up the retransmit timer. 1023 */ 1024 if ((to.to_flags & TOF_TS) != 0 && 1025 to.to_tsecr) { 1026 tcp_xmit_timer(tp, 1027 ticks - to.to_tsecr + 1); 1028 } else if (tp->t_rtttime && 1029 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1030 tcp_xmit_timer(tp, 1031 ticks - tp->t_rtttime); 1032 } 1033 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1034 acked = th->th_ack - tp->snd_una; 1035 tcpstat.tcps_rcvackpack++; 1036 tcpstat.tcps_rcvackbyte += acked; 1037 sbdrop(&so->so_snd, acked); 1038 if (SEQ_GT(tp->snd_una, tp->snd_high) && 1039 SEQ_LEQ(th->th_ack, tp->snd_high)) 1040 tp->snd_high = th->th_ack - 1; 1041 tp->snd_una = tp->snd_recover = th->th_ack; 1042 /* 1043 * pull snd_wl2 up to prevent seq wrap relative 1044 * to th_ack. 1045 */ 1046 tp->snd_wl2 = th->th_ack; 1047 tp->t_dupacks = 0; 1048 m_freem(m); 1049 ND6_HINT(tp); /* some progress has been done */ 1050 1051 /* 1052 * If all outstanding data are acked, stop 1053 * retransmit timer, otherwise restart timer 1054 * using current (possibly backed-off) value. 1055 * If process is waiting for space, 1056 * wakeup/selwakeup/signal. If data 1057 * are ready to send, let tcp_output 1058 * decide between more output or persist. 1059 */ 1060 if (tp->snd_una == tp->snd_max) 1061 callout_stop(tp->tt_rexmt); 1062 else if (!callout_active(tp->tt_persist)) 1063 callout_reset(tp->tt_rexmt, 1064 tp->t_rxtcur, 1065 tcp_timer_rexmt, tp); 1066 1067 sowwakeup(so); 1068 if (so->so_snd.sb_cc) 1069 (void) tcp_output(tp); 1070 goto check_delack; 1071 } 1072 } else if (th->th_ack == tp->snd_una && 1073 LIST_EMPTY(&tp->t_segq) && 1074 tlen <= sbspace(&so->so_rcv)) { 1075 KASSERT(headlocked, ("headlocked")); 1076 INP_INFO_WUNLOCK(&tcbinfo); 1077 /* 1078 * this is a pure, in-sequence data packet 1079 * with nothing on the reassembly queue and 1080 * we have enough buffer space to take it. 1081 */ 1082 ++tcpstat.tcps_preddat; 1083 tp->rcv_nxt += tlen; 1084 /* 1085 * Pull snd_wl1 up to prevent seq wrap relative to 1086 * th_seq. 1087 */ 1088 tp->snd_wl1 = th->th_seq; 1089 /* 1090 * Pull rcv_up up to prevent seq wrap relative to 1091 * rcv_nxt. 1092 */ 1093 tp->rcv_up = tp->rcv_nxt; 1094 tcpstat.tcps_rcvpack++; 1095 tcpstat.tcps_rcvbyte += tlen; 1096 ND6_HINT(tp); /* some progress has been done */ 1097 /* 1098 * Add data to socket buffer. 1099 */ 1100 if (so->so_state & SS_CANTRCVMORE) { 1101 m_freem(m); 1102 } else { 1103 m_adj(m, drop_hdrlen); /* delayed header drop */ 1104 sbappend(&so->so_rcv, m); 1105 } 1106 sorwakeup(so); 1107 if (DELAY_ACK(tp)) { 1108 tp->t_flags |= TF_DELACK; 1109 } else { 1110 tp->t_flags |= TF_ACKNOW; 1111 tcp_output(tp); 1112 } 1113 goto check_delack; 1114 } 1115 } 1116 1117 /* 1118 * Calculate amount of space in receive window, 1119 * and then do TCP input processing. 1120 * Receive window is amount of space in rcv queue, 1121 * but not less than advertised window. 1122 */ 1123 { int win; 1124 1125 win = sbspace(&so->so_rcv); 1126 if (win < 0) 1127 win = 0; 1128 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1129 } 1130 1131 switch (tp->t_state) { 1132 1133 /* 1134 * If the state is SYN_RECEIVED: 1135 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1136 */ 1137 case TCPS_SYN_RECEIVED: 1138 if ((thflags & TH_ACK) && 1139 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1140 SEQ_GT(th->th_ack, tp->snd_max))) { 1141 rstreason = BANDLIM_RST_OPENPORT; 1142 goto dropwithreset; 1143 } 1144 break; 1145 1146 /* 1147 * If the state is SYN_SENT: 1148 * if seg contains an ACK, but not for our SYN, drop the input. 1149 * if seg contains a RST, then drop the connection. 1150 * if seg does not contain SYN, then drop it. 1151 * Otherwise this is an acceptable SYN segment 1152 * initialize tp->rcv_nxt and tp->irs 1153 * if seg contains ack then advance tp->snd_una 1154 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1155 * arrange for segment to be acked (eventually) 1156 * continue processing rest of data/controls, beginning with URG 1157 */ 1158 case TCPS_SYN_SENT: 1159 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1160 taop = &tao_noncached; 1161 bzero(taop, sizeof(*taop)); 1162 } 1163 1164 if ((thflags & TH_ACK) && 1165 (SEQ_LEQ(th->th_ack, tp->iss) || 1166 SEQ_GT(th->th_ack, tp->snd_max))) { 1167 /* 1168 * If we have a cached CCsent for the remote host, 1169 * hence we haven't just crashed and restarted, 1170 * do not send a RST. This may be a retransmission 1171 * from the other side after our earlier ACK was lost. 1172 * Our new SYN, when it arrives, will serve as the 1173 * needed ACK. 1174 */ 1175 if (taop->tao_ccsent != 0) 1176 goto drop; 1177 else { 1178 rstreason = BANDLIM_UNLIMITED; 1179 goto dropwithreset; 1180 } 1181 } 1182 if (thflags & TH_RST) { 1183 if (thflags & TH_ACK) 1184 tp = tcp_drop(tp, ECONNREFUSED); 1185 goto drop; 1186 } 1187 if ((thflags & TH_SYN) == 0) 1188 goto drop; 1189 tp->snd_wnd = th->th_win; /* initial send window */ 1190 tp->cc_recv = to.to_cc; /* foreign CC */ 1191 1192 tp->irs = th->th_seq; 1193 tcp_rcvseqinit(tp); 1194 if (thflags & TH_ACK) { 1195 /* 1196 * Our SYN was acked. If segment contains CC.ECHO 1197 * option, check it to make sure this segment really 1198 * matches our SYN. If not, just drop it as old 1199 * duplicate, but send an RST if we're still playing 1200 * by the old rules. If no CC.ECHO option, make sure 1201 * we don't get fooled into using T/TCP. 1202 */ 1203 if (to.to_flags & TOF_CCECHO) { 1204 if (tp->cc_send != to.to_ccecho) { 1205 if (taop->tao_ccsent != 0) 1206 goto drop; 1207 else { 1208 rstreason = BANDLIM_UNLIMITED; 1209 goto dropwithreset; 1210 } 1211 } 1212 } else 1213 tp->t_flags &= ~TF_RCVD_CC; 1214 tcpstat.tcps_connects++; 1215 soisconnected(so); 1216 #ifdef MAC 1217 mac_set_socket_peer_from_mbuf(m, so); 1218 #endif 1219 /* Do window scaling on this connection? */ 1220 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1221 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1222 tp->snd_scale = tp->requested_s_scale; 1223 tp->rcv_scale = tp->request_r_scale; 1224 } 1225 /* Segment is acceptable, update cache if undefined. */ 1226 if (taop->tao_ccsent == 0) 1227 taop->tao_ccsent = to.to_ccecho; 1228 1229 tp->rcv_adv += tp->rcv_wnd; 1230 tp->snd_una++; /* SYN is acked */ 1231 /* 1232 * If there's data, delay ACK; if there's also a FIN 1233 * ACKNOW will be turned on later. 1234 */ 1235 if (DELAY_ACK(tp) && tlen != 0) 1236 callout_reset(tp->tt_delack, tcp_delacktime, 1237 tcp_timer_delack, tp); 1238 else 1239 tp->t_flags |= TF_ACKNOW; 1240 /* 1241 * Received <SYN,ACK> in SYN_SENT[*] state. 1242 * Transitions: 1243 * SYN_SENT --> ESTABLISHED 1244 * SYN_SENT* --> FIN_WAIT_1 1245 */ 1246 tp->t_starttime = ticks; 1247 if (tp->t_flags & TF_NEEDFIN) { 1248 tp->t_state = TCPS_FIN_WAIT_1; 1249 tp->t_flags &= ~TF_NEEDFIN; 1250 thflags &= ~TH_SYN; 1251 } else { 1252 tp->t_state = TCPS_ESTABLISHED; 1253 callout_reset(tp->tt_keep, tcp_keepidle, 1254 tcp_timer_keep, tp); 1255 } 1256 } else { 1257 /* 1258 * Received initial SYN in SYN-SENT[*] state => 1259 * simultaneous open. If segment contains CC option 1260 * and there is a cached CC, apply TAO test. 1261 * If it succeeds, connection is * half-synchronized. 1262 * Otherwise, do 3-way handshake: 1263 * SYN-SENT -> SYN-RECEIVED 1264 * SYN-SENT* -> SYN-RECEIVED* 1265 * If there was no CC option, clear cached CC value. 1266 */ 1267 tp->t_flags |= TF_ACKNOW; 1268 callout_stop(tp->tt_rexmt); 1269 if (to.to_flags & TOF_CC) { 1270 if (taop->tao_cc != 0 && 1271 CC_GT(to.to_cc, taop->tao_cc)) { 1272 /* 1273 * update cache and make transition: 1274 * SYN-SENT -> ESTABLISHED* 1275 * SYN-SENT* -> FIN-WAIT-1* 1276 */ 1277 taop->tao_cc = to.to_cc; 1278 tp->t_starttime = ticks; 1279 if (tp->t_flags & TF_NEEDFIN) { 1280 tp->t_state = TCPS_FIN_WAIT_1; 1281 tp->t_flags &= ~TF_NEEDFIN; 1282 } else { 1283 tp->t_state = TCPS_ESTABLISHED; 1284 callout_reset(tp->tt_keep, 1285 tcp_keepidle, 1286 tcp_timer_keep, 1287 tp); 1288 } 1289 tp->t_flags |= TF_NEEDSYN; 1290 } else 1291 tp->t_state = TCPS_SYN_RECEIVED; 1292 } else { 1293 /* CC.NEW or no option => invalidate cache */ 1294 taop->tao_cc = 0; 1295 tp->t_state = TCPS_SYN_RECEIVED; 1296 } 1297 } 1298 1299 trimthenstep6: 1300 /* 1301 * Advance th->th_seq to correspond to first data byte. 1302 * If data, trim to stay within window, 1303 * dropping FIN if necessary. 1304 */ 1305 th->th_seq++; 1306 if (tlen > tp->rcv_wnd) { 1307 todrop = tlen - tp->rcv_wnd; 1308 m_adj(m, -todrop); 1309 tlen = tp->rcv_wnd; 1310 thflags &= ~TH_FIN; 1311 tcpstat.tcps_rcvpackafterwin++; 1312 tcpstat.tcps_rcvbyteafterwin += todrop; 1313 } 1314 tp->snd_wl1 = th->th_seq - 1; 1315 tp->rcv_up = th->th_seq; 1316 /* 1317 * Client side of transaction: already sent SYN and data. 1318 * If the remote host used T/TCP to validate the SYN, 1319 * our data will be ACK'd; if so, enter normal data segment 1320 * processing in the middle of step 5, ack processing. 1321 * Otherwise, goto step 6. 1322 */ 1323 if (thflags & TH_ACK) 1324 goto process_ACK; 1325 1326 goto step6; 1327 1328 /* 1329 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1330 * if segment contains a SYN and CC [not CC.NEW] option: 1331 * if state == TIME_WAIT and connection duration > MSL, 1332 * drop packet and send RST; 1333 * 1334 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1335 * ack the FIN (and data) in retransmission queue. 1336 * Complete close and delete TCPCB. Then reprocess 1337 * segment, hoping to find new TCPCB in LISTEN state; 1338 * 1339 * else must be old SYN; drop it. 1340 * else do normal processing. 1341 */ 1342 case TCPS_LAST_ACK: 1343 case TCPS_CLOSING: 1344 case TCPS_TIME_WAIT: 1345 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1346 if ((thflags & TH_SYN) && 1347 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1348 if (tp->t_state == TCPS_TIME_WAIT && 1349 (ticks - tp->t_starttime) > tcp_msl) { 1350 rstreason = BANDLIM_UNLIMITED; 1351 goto dropwithreset; 1352 } 1353 if (CC_GT(to.to_cc, tp->cc_recv)) { 1354 tp = tcp_close(tp); 1355 goto findpcb; 1356 } 1357 else 1358 goto drop; 1359 } 1360 break; /* continue normal processing */ 1361 } 1362 1363 /* 1364 * States other than LISTEN or SYN_SENT. 1365 * First check the RST flag and sequence number since reset segments 1366 * are exempt from the timestamp and connection count tests. This 1367 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1368 * below which allowed reset segments in half the sequence space 1369 * to fall though and be processed (which gives forged reset 1370 * segments with a random sequence number a 50 percent chance of 1371 * killing a connection). 1372 * Then check timestamp, if present. 1373 * Then check the connection count, if present. 1374 * Then check that at least some bytes of segment are within 1375 * receive window. If segment begins before rcv_nxt, 1376 * drop leading data (and SYN); if nothing left, just ack. 1377 * 1378 * 1379 * If the RST bit is set, check the sequence number to see 1380 * if this is a valid reset segment. 1381 * RFC 793 page 37: 1382 * In all states except SYN-SENT, all reset (RST) segments 1383 * are validated by checking their SEQ-fields. A reset is 1384 * valid if its sequence number is in the window. 1385 * Note: this does not take into account delayed ACKs, so 1386 * we should test against last_ack_sent instead of rcv_nxt. 1387 * The sequence number in the reset segment is normally an 1388 * echo of our outgoing acknowlegement numbers, but some hosts 1389 * send a reset with the sequence number at the rightmost edge 1390 * of our receive window, and we have to handle this case. 1391 * If we have multiple segments in flight, the intial reset 1392 * segment sequence numbers will be to the left of last_ack_sent, 1393 * but they will eventually catch up. 1394 * In any case, it never made sense to trim reset segments to 1395 * fit the receive window since RFC 1122 says: 1396 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1397 * 1398 * A TCP SHOULD allow a received RST segment to include data. 1399 * 1400 * DISCUSSION 1401 * It has been suggested that a RST segment could contain 1402 * ASCII text that encoded and explained the cause of the 1403 * RST. No standard has yet been established for such 1404 * data. 1405 * 1406 * If the reset segment passes the sequence number test examine 1407 * the state: 1408 * SYN_RECEIVED STATE: 1409 * If passive open, return to LISTEN state. 1410 * If active open, inform user that connection was refused. 1411 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1412 * Inform user that connection was reset, and close tcb. 1413 * CLOSING, LAST_ACK STATES: 1414 * Close the tcb. 1415 * TIME_WAIT STATE: 1416 * Drop the segment - see Stevens, vol. 2, p. 964 and 1417 * RFC 1337. 1418 */ 1419 if (thflags & TH_RST) { 1420 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1421 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1422 switch (tp->t_state) { 1423 1424 case TCPS_SYN_RECEIVED: 1425 so->so_error = ECONNREFUSED; 1426 goto close; 1427 1428 case TCPS_ESTABLISHED: 1429 case TCPS_FIN_WAIT_1: 1430 case TCPS_FIN_WAIT_2: 1431 case TCPS_CLOSE_WAIT: 1432 so->so_error = ECONNRESET; 1433 close: 1434 tp->t_state = TCPS_CLOSED; 1435 tcpstat.tcps_drops++; 1436 tp = tcp_close(tp); 1437 break; 1438 1439 case TCPS_CLOSING: 1440 case TCPS_LAST_ACK: 1441 tp = tcp_close(tp); 1442 break; 1443 1444 case TCPS_TIME_WAIT: 1445 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1446 ("timewait")); 1447 break; 1448 } 1449 } 1450 goto drop; 1451 } 1452 1453 /* 1454 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1455 * and it's less than ts_recent, drop it. 1456 */ 1457 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1458 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1459 1460 /* Check to see if ts_recent is over 24 days old. */ 1461 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1462 /* 1463 * Invalidate ts_recent. If this segment updates 1464 * ts_recent, the age will be reset later and ts_recent 1465 * will get a valid value. If it does not, setting 1466 * ts_recent to zero will at least satisfy the 1467 * requirement that zero be placed in the timestamp 1468 * echo reply when ts_recent isn't valid. The 1469 * age isn't reset until we get a valid ts_recent 1470 * because we don't want out-of-order segments to be 1471 * dropped when ts_recent is old. 1472 */ 1473 tp->ts_recent = 0; 1474 } else { 1475 tcpstat.tcps_rcvduppack++; 1476 tcpstat.tcps_rcvdupbyte += tlen; 1477 tcpstat.tcps_pawsdrop++; 1478 if (tlen) 1479 goto dropafterack; 1480 goto drop; 1481 } 1482 } 1483 1484 /* 1485 * T/TCP mechanism 1486 * If T/TCP was negotiated and the segment doesn't have CC, 1487 * or if its CC is wrong then drop the segment. 1488 * RST segments do not have to comply with this. 1489 */ 1490 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1491 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1492 goto dropafterack; 1493 1494 /* 1495 * In the SYN-RECEIVED state, validate that the packet belongs to 1496 * this connection before trimming the data to fit the receive 1497 * window. Check the sequence number versus IRS since we know 1498 * the sequence numbers haven't wrapped. This is a partial fix 1499 * for the "LAND" DoS attack. 1500 */ 1501 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1502 rstreason = BANDLIM_RST_OPENPORT; 1503 goto dropwithreset; 1504 } 1505 1506 todrop = tp->rcv_nxt - th->th_seq; 1507 if (todrop > 0) { 1508 if (thflags & TH_SYN) { 1509 thflags &= ~TH_SYN; 1510 th->th_seq++; 1511 if (th->th_urp > 1) 1512 th->th_urp--; 1513 else 1514 thflags &= ~TH_URG; 1515 todrop--; 1516 } 1517 /* 1518 * Following if statement from Stevens, vol. 2, p. 960. 1519 */ 1520 if (todrop > tlen 1521 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1522 /* 1523 * Any valid FIN must be to the left of the window. 1524 * At this point the FIN must be a duplicate or out 1525 * of sequence; drop it. 1526 */ 1527 thflags &= ~TH_FIN; 1528 1529 /* 1530 * Send an ACK to resynchronize and drop any data. 1531 * But keep on processing for RST or ACK. 1532 */ 1533 tp->t_flags |= TF_ACKNOW; 1534 todrop = tlen; 1535 tcpstat.tcps_rcvduppack++; 1536 tcpstat.tcps_rcvdupbyte += todrop; 1537 } else { 1538 tcpstat.tcps_rcvpartduppack++; 1539 tcpstat.tcps_rcvpartdupbyte += todrop; 1540 } 1541 drop_hdrlen += todrop; /* drop from the top afterwards */ 1542 th->th_seq += todrop; 1543 tlen -= todrop; 1544 if (th->th_urp > todrop) 1545 th->th_urp -= todrop; 1546 else { 1547 thflags &= ~TH_URG; 1548 th->th_urp = 0; 1549 } 1550 } 1551 1552 /* 1553 * If new data are received on a connection after the 1554 * user processes are gone, then RST the other end. 1555 */ 1556 if ((so->so_state & SS_NOFDREF) && 1557 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1558 tp = tcp_close(tp); 1559 tcpstat.tcps_rcvafterclose++; 1560 rstreason = BANDLIM_UNLIMITED; 1561 goto dropwithreset; 1562 } 1563 1564 /* 1565 * If segment ends after window, drop trailing data 1566 * (and PUSH and FIN); if nothing left, just ACK. 1567 */ 1568 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1569 if (todrop > 0) { 1570 tcpstat.tcps_rcvpackafterwin++; 1571 if (todrop >= tlen) { 1572 tcpstat.tcps_rcvbyteafterwin += tlen; 1573 /* 1574 * If a new connection request is received 1575 * while in TIME_WAIT, drop the old connection 1576 * and start over if the sequence numbers 1577 * are above the previous ones. 1578 */ 1579 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1580 if (thflags & TH_SYN && 1581 tp->t_state == TCPS_TIME_WAIT && 1582 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1583 tp = tcp_close(tp); 1584 goto findpcb; 1585 } 1586 /* 1587 * If window is closed can only take segments at 1588 * window edge, and have to drop data and PUSH from 1589 * incoming segments. Continue processing, but 1590 * remember to ack. Otherwise, drop segment 1591 * and ack. 1592 */ 1593 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1594 tp->t_flags |= TF_ACKNOW; 1595 tcpstat.tcps_rcvwinprobe++; 1596 } else 1597 goto dropafterack; 1598 } else 1599 tcpstat.tcps_rcvbyteafterwin += todrop; 1600 m_adj(m, -todrop); 1601 tlen -= todrop; 1602 thflags &= ~(TH_PUSH|TH_FIN); 1603 } 1604 1605 /* 1606 * If last ACK falls within this segment's sequence numbers, 1607 * record its timestamp. 1608 * NOTE that the test is modified according to the latest 1609 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1610 */ 1611 if ((to.to_flags & TOF_TS) != 0 && 1612 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1613 tp->ts_recent_age = ticks; 1614 tp->ts_recent = to.to_tsval; 1615 } 1616 1617 /* 1618 * If a SYN is in the window, then this is an 1619 * error and we send an RST and drop the connection. 1620 */ 1621 if (thflags & TH_SYN) { 1622 tp = tcp_drop(tp, ECONNRESET); 1623 rstreason = BANDLIM_UNLIMITED; 1624 goto dropwithreset; 1625 } 1626 1627 /* 1628 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1629 * flag is on (half-synchronized state), then queue data for 1630 * later processing; else drop segment and return. 1631 */ 1632 if ((thflags & TH_ACK) == 0) { 1633 if (tp->t_state == TCPS_SYN_RECEIVED || 1634 (tp->t_flags & TF_NEEDSYN)) 1635 goto step6; 1636 else 1637 goto drop; 1638 } 1639 1640 /* 1641 * Ack processing. 1642 */ 1643 switch (tp->t_state) { 1644 1645 /* 1646 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1647 * ESTABLISHED state and continue processing. 1648 * The ACK was checked above. 1649 */ 1650 case TCPS_SYN_RECEIVED: 1651 1652 tcpstat.tcps_connects++; 1653 soisconnected(so); 1654 /* Do window scaling? */ 1655 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1656 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1657 tp->snd_scale = tp->requested_s_scale; 1658 tp->rcv_scale = tp->request_r_scale; 1659 } 1660 /* 1661 * Upon successful completion of 3-way handshake, 1662 * update cache.CC if it was undefined, pass any queued 1663 * data to the user, and advance state appropriately. 1664 */ 1665 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1666 taop->tao_cc == 0) 1667 taop->tao_cc = tp->cc_recv; 1668 1669 /* 1670 * Make transitions: 1671 * SYN-RECEIVED -> ESTABLISHED 1672 * SYN-RECEIVED* -> FIN-WAIT-1 1673 */ 1674 tp->t_starttime = ticks; 1675 if (tp->t_flags & TF_NEEDFIN) { 1676 tp->t_state = TCPS_FIN_WAIT_1; 1677 tp->t_flags &= ~TF_NEEDFIN; 1678 } else { 1679 tp->t_state = TCPS_ESTABLISHED; 1680 callout_reset(tp->tt_keep, tcp_keepidle, 1681 tcp_timer_keep, tp); 1682 } 1683 /* 1684 * If segment contains data or ACK, will call tcp_reass() 1685 * later; if not, do so now to pass queued data to user. 1686 */ 1687 if (tlen == 0 && (thflags & TH_FIN) == 0) 1688 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1689 (struct mbuf *)0); 1690 tp->snd_wl1 = th->th_seq - 1; 1691 /* FALLTHROUGH */ 1692 1693 /* 1694 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1695 * ACKs. If the ack is in the range 1696 * tp->snd_una < th->th_ack <= tp->snd_max 1697 * then advance tp->snd_una to th->th_ack and drop 1698 * data from the retransmission queue. If this ACK reflects 1699 * more up to date window information we update our window information. 1700 */ 1701 case TCPS_ESTABLISHED: 1702 case TCPS_FIN_WAIT_1: 1703 case TCPS_FIN_WAIT_2: 1704 case TCPS_CLOSE_WAIT: 1705 case TCPS_CLOSING: 1706 case TCPS_LAST_ACK: 1707 case TCPS_TIME_WAIT: 1708 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1709 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1710 if (tlen == 0 && tiwin == tp->snd_wnd) { 1711 tcpstat.tcps_rcvdupack++; 1712 /* 1713 * If we have outstanding data (other than 1714 * a window probe), this is a completely 1715 * duplicate ack (ie, window info didn't 1716 * change), the ack is the biggest we've 1717 * seen and we've seen exactly our rexmt 1718 * threshhold of them, assume a packet 1719 * has been dropped and retransmit it. 1720 * Kludge snd_nxt & the congestion 1721 * window so we send only this one 1722 * packet. 1723 * 1724 * We know we're losing at the current 1725 * window size so do congestion avoidance 1726 * (set ssthresh to half the current window 1727 * and pull our congestion window back to 1728 * the new ssthresh). 1729 * 1730 * Dup acks mean that packets have left the 1731 * network (they're now cached at the receiver) 1732 * so bump cwnd by the amount in the receiver 1733 * to keep a constant cwnd packets in the 1734 * network. 1735 */ 1736 if (!callout_active(tp->tt_rexmt) || 1737 th->th_ack != tp->snd_una) 1738 tp->t_dupacks = 0; 1739 else if (++tp->t_dupacks > tcprexmtthresh || 1740 (tcp_do_newreno && 1741 SEQ_LT(tp->snd_una, 1742 tp->snd_recover))) { 1743 tp->snd_cwnd += tp->t_maxseg; 1744 (void) tcp_output(tp); 1745 goto drop; 1746 } else if (tp->t_dupacks == tcprexmtthresh) { 1747 tcp_seq onxt = tp->snd_nxt; 1748 u_int win; 1749 if (tcp_do_newreno && 1750 SEQ_LEQ(th->th_ack, tp->snd_high)) { 1751 tp->t_dupacks = 0; 1752 break; 1753 } 1754 win = min(tp->snd_wnd, tp->snd_cwnd) / 1755 2 / tp->t_maxseg; 1756 if (win < 2) 1757 win = 2; 1758 tp->snd_ssthresh = win * tp->t_maxseg; 1759 tp->snd_recover = tp->snd_max; 1760 callout_stop(tp->tt_rexmt); 1761 tp->t_rtttime = 0; 1762 tp->snd_nxt = th->th_ack; 1763 tp->snd_cwnd = tp->t_maxseg; 1764 (void) tcp_output(tp); 1765 tp->snd_cwnd = tp->snd_ssthresh + 1766 tp->t_maxseg * tp->t_dupacks; 1767 if (SEQ_GT(onxt, tp->snd_nxt)) 1768 tp->snd_nxt = onxt; 1769 goto drop; 1770 } else if (tcp_do_rfc3042) { 1771 u_long oldcwnd = tp->snd_cwnd; 1772 KASSERT(tp->t_dupacks == 1 || 1773 tp->t_dupacks == 2, 1774 ("dupacks not 1 or 2")); 1775 tp->snd_cwnd += tp->t_dupacks * 1776 tp->t_maxseg; 1777 (void) tcp_output(tp); 1778 tp->snd_cwnd = oldcwnd; 1779 goto drop; 1780 } 1781 } else 1782 tp->t_dupacks = 0; 1783 break; 1784 } 1785 1786 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1787 1788 /* 1789 * If the congestion window was inflated to account 1790 * for the other side's cached packets, retract it. 1791 */ 1792 if (tcp_do_newreno) { 1793 if (SEQ_LT(tp->snd_una, tp->snd_recover)) { 1794 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1795 tcp_newreno_partial_ack(tp, th); 1796 } else { 1797 /* 1798 * Window inflation should have left us 1799 * with approximately snd_ssthresh 1800 * outstanding data. 1801 * But in case we would be inclined to 1802 * send a burst, better to do it via 1803 * the slow start mechanism. 1804 */ 1805 if (SEQ_GT(th->th_ack + 1806 tp->snd_ssthresh, 1807 tp->snd_max)) 1808 tp->snd_cwnd = tp->snd_max - 1809 th->th_ack + 1810 tp->t_maxseg; 1811 else 1812 tp->snd_cwnd = tp->snd_ssthresh; 1813 } 1814 } 1815 } else { 1816 if (tp->t_dupacks >= tcprexmtthresh && 1817 tp->snd_cwnd > tp->snd_ssthresh) 1818 tp->snd_cwnd = tp->snd_ssthresh; 1819 } 1820 tp->t_dupacks = 0; 1821 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1822 tcpstat.tcps_rcvacktoomuch++; 1823 goto dropafterack; 1824 } 1825 /* 1826 * If we reach this point, ACK is not a duplicate, 1827 * i.e., it ACKs something we sent. 1828 */ 1829 if (tp->t_flags & TF_NEEDSYN) { 1830 /* 1831 * T/TCP: Connection was half-synchronized, and our 1832 * SYN has been ACK'd (so connection is now fully 1833 * synchronized). Go to non-starred state, 1834 * increment snd_una for ACK of SYN, and check if 1835 * we can do window scaling. 1836 */ 1837 tp->t_flags &= ~TF_NEEDSYN; 1838 tp->snd_una++; 1839 /* Do window scaling? */ 1840 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1841 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1842 tp->snd_scale = tp->requested_s_scale; 1843 tp->rcv_scale = tp->request_r_scale; 1844 } 1845 } 1846 1847 process_ACK: 1848 acked = th->th_ack - tp->snd_una; 1849 tcpstat.tcps_rcvackpack++; 1850 tcpstat.tcps_rcvackbyte += acked; 1851 1852 /* 1853 * If we just performed our first retransmit, and the ACK 1854 * arrives within our recovery window, then it was a mistake 1855 * to do the retransmit in the first place. Recover our 1856 * original cwnd and ssthresh, and proceed to transmit where 1857 * we left off. 1858 */ 1859 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1860 ++tcpstat.tcps_sndrexmitbad; 1861 tp->snd_cwnd = tp->snd_cwnd_prev; 1862 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1863 tp->snd_high = tp->snd_high_prev; 1864 tp->snd_nxt = tp->snd_max; 1865 tp->t_badrxtwin = 0; /* XXX probably not required */ 1866 } 1867 1868 /* 1869 * If we have a timestamp reply, update smoothed 1870 * round trip time. If no timestamp is present but 1871 * transmit timer is running and timed sequence 1872 * number was acked, update smoothed round trip time. 1873 * Since we now have an rtt measurement, cancel the 1874 * timer backoff (cf., Phil Karn's retransmit alg.). 1875 * Recompute the initial retransmit timer. 1876 * 1877 * Some boxes send broken timestamp replies 1878 * during the SYN+ACK phase, ignore 1879 * timestamps of 0 or we could calculate a 1880 * huge RTT and blow up the retransmit timer. 1881 */ 1882 if ((to.to_flags & TOF_TS) != 0 && 1883 to.to_tsecr) { 1884 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1885 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1886 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1887 } 1888 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1889 1890 /* 1891 * If all outstanding data is acked, stop retransmit 1892 * timer and remember to restart (more output or persist). 1893 * If there is more data to be acked, restart retransmit 1894 * timer, using current (possibly backed-off) value. 1895 */ 1896 if (th->th_ack == tp->snd_max) { 1897 callout_stop(tp->tt_rexmt); 1898 needoutput = 1; 1899 } else if (!callout_active(tp->tt_persist)) 1900 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1901 tcp_timer_rexmt, tp); 1902 1903 /* 1904 * If no data (only SYN) was ACK'd, 1905 * skip rest of ACK processing. 1906 */ 1907 if (acked == 0) 1908 goto step6; 1909 1910 /* 1911 * When new data is acked, open the congestion window. 1912 * If the window gives us less than ssthresh packets 1913 * in flight, open exponentially (maxseg per packet). 1914 * Otherwise open linearly: maxseg per window 1915 * (maxseg^2 / cwnd per packet). 1916 */ 1917 if (!tcp_do_newreno || SEQ_GEQ(tp->snd_una, tp->snd_recover)) { 1918 register u_int cw = tp->snd_cwnd; 1919 register u_int incr = tp->t_maxseg; 1920 if (cw > tp->snd_ssthresh) 1921 incr = incr * incr / cw; 1922 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1923 } 1924 if (acked > so->so_snd.sb_cc) { 1925 tp->snd_wnd -= so->so_snd.sb_cc; 1926 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1927 ourfinisacked = 1; 1928 } else { 1929 sbdrop(&so->so_snd, acked); 1930 tp->snd_wnd -= acked; 1931 ourfinisacked = 0; 1932 } 1933 sowwakeup(so); 1934 /* detect una wraparound */ 1935 if (SEQ_GEQ(tp->snd_una, tp->snd_recover) && 1936 SEQ_LT(th->th_ack, tp->snd_recover)) 1937 tp->snd_recover = th->th_ack; 1938 if (SEQ_GT(tp->snd_una, tp->snd_high) && 1939 SEQ_LEQ(th->th_ack, tp->snd_high)) 1940 tp->snd_high = th->th_ack - 1; 1941 tp->snd_una = th->th_ack; 1942 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1943 tp->snd_nxt = tp->snd_una; 1944 1945 switch (tp->t_state) { 1946 1947 /* 1948 * In FIN_WAIT_1 STATE in addition to the processing 1949 * for the ESTABLISHED state if our FIN is now acknowledged 1950 * then enter FIN_WAIT_2. 1951 */ 1952 case TCPS_FIN_WAIT_1: 1953 if (ourfinisacked) { 1954 /* 1955 * If we can't receive any more 1956 * data, then closing user can proceed. 1957 * Starting the timer is contrary to the 1958 * specification, but if we don't get a FIN 1959 * we'll hang forever. 1960 */ 1961 /* XXXjl 1962 * we should release the tp also, and use a 1963 * compressed state. 1964 */ 1965 if (so->so_state & SS_CANTRCVMORE) { 1966 soisdisconnected(so); 1967 callout_reset(tp->tt_2msl, tcp_maxidle, 1968 tcp_timer_2msl, tp); 1969 } 1970 tp->t_state = TCPS_FIN_WAIT_2; 1971 } 1972 break; 1973 1974 /* 1975 * In CLOSING STATE in addition to the processing for 1976 * the ESTABLISHED state if the ACK acknowledges our FIN 1977 * then enter the TIME-WAIT state, otherwise ignore 1978 * the segment. 1979 */ 1980 case TCPS_CLOSING: 1981 if (ourfinisacked) { 1982 KASSERT(headlocked, ("headlocked")); 1983 tcp_twstart(tp); 1984 INP_INFO_WUNLOCK(&tcbinfo); 1985 m_freem(m); 1986 return; 1987 } 1988 break; 1989 1990 /* 1991 * In LAST_ACK, we may still be waiting for data to drain 1992 * and/or to be acked, as well as for the ack of our FIN. 1993 * If our FIN is now acknowledged, delete the TCB, 1994 * enter the closed state and return. 1995 */ 1996 case TCPS_LAST_ACK: 1997 if (ourfinisacked) { 1998 tp = tcp_close(tp); 1999 goto drop; 2000 } 2001 break; 2002 2003 /* 2004 * In TIME_WAIT state the only thing that should arrive 2005 * is a retransmission of the remote FIN. Acknowledge 2006 * it and restart the finack timer. 2007 */ 2008 case TCPS_TIME_WAIT: 2009 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2010 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2011 tcp_timer_2msl, tp); 2012 goto dropafterack; 2013 } 2014 } 2015 2016 step6: 2017 /* 2018 * Update window information. 2019 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2020 */ 2021 if ((thflags & TH_ACK) && 2022 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2023 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2024 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2025 /* keep track of pure window updates */ 2026 if (tlen == 0 && 2027 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2028 tcpstat.tcps_rcvwinupd++; 2029 tp->snd_wnd = tiwin; 2030 tp->snd_wl1 = th->th_seq; 2031 tp->snd_wl2 = th->th_ack; 2032 if (tp->snd_wnd > tp->max_sndwnd) 2033 tp->max_sndwnd = tp->snd_wnd; 2034 needoutput = 1; 2035 } 2036 2037 /* 2038 * Process segments with URG. 2039 */ 2040 if ((thflags & TH_URG) && th->th_urp && 2041 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2042 /* 2043 * This is a kludge, but if we receive and accept 2044 * random urgent pointers, we'll crash in 2045 * soreceive. It's hard to imagine someone 2046 * actually wanting to send this much urgent data. 2047 */ 2048 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2049 th->th_urp = 0; /* XXX */ 2050 thflags &= ~TH_URG; /* XXX */ 2051 goto dodata; /* XXX */ 2052 } 2053 /* 2054 * If this segment advances the known urgent pointer, 2055 * then mark the data stream. This should not happen 2056 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2057 * a FIN has been received from the remote side. 2058 * In these states we ignore the URG. 2059 * 2060 * According to RFC961 (Assigned Protocols), 2061 * the urgent pointer points to the last octet 2062 * of urgent data. We continue, however, 2063 * to consider it to indicate the first octet 2064 * of data past the urgent section as the original 2065 * spec states (in one of two places). 2066 */ 2067 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2068 tp->rcv_up = th->th_seq + th->th_urp; 2069 so->so_oobmark = so->so_rcv.sb_cc + 2070 (tp->rcv_up - tp->rcv_nxt) - 1; 2071 if (so->so_oobmark == 0) 2072 so->so_state |= SS_RCVATMARK; 2073 sohasoutofband(so); 2074 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2075 } 2076 /* 2077 * Remove out of band data so doesn't get presented to user. 2078 * This can happen independent of advancing the URG pointer, 2079 * but if two URG's are pending at once, some out-of-band 2080 * data may creep in... ick. 2081 */ 2082 if (th->th_urp <= (u_long)tlen && 2083 !(so->so_options & SO_OOBINLINE)) { 2084 /* hdr drop is delayed */ 2085 tcp_pulloutofband(so, th, m, drop_hdrlen); 2086 } 2087 } else { 2088 /* 2089 * If no out of band data is expected, 2090 * pull receive urgent pointer along 2091 * with the receive window. 2092 */ 2093 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2094 tp->rcv_up = tp->rcv_nxt; 2095 } 2096 dodata: /* XXX */ 2097 KASSERT(headlocked, ("headlocked")); 2098 /* 2099 * Process the segment text, merging it into the TCP sequencing queue, 2100 * and arranging for acknowledgment of receipt if necessary. 2101 * This process logically involves adjusting tp->rcv_wnd as data 2102 * is presented to the user (this happens in tcp_usrreq.c, 2103 * case PRU_RCVD). If a FIN has already been received on this 2104 * connection then we just ignore the text. 2105 */ 2106 if ((tlen || (thflags & TH_FIN)) && 2107 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2108 m_adj(m, drop_hdrlen); /* delayed header drop */ 2109 /* 2110 * Insert segment which includes th into TCP reassembly queue 2111 * with control block tp. Set thflags to whether reassembly now 2112 * includes a segment with FIN. This handles the common case 2113 * inline (segment is the next to be received on an established 2114 * connection, and the queue is empty), avoiding linkage into 2115 * and removal from the queue and repetition of various 2116 * conversions. 2117 * Set DELACK for segments received in order, but ack 2118 * immediately when segments are out of order (so 2119 * fast retransmit can work). 2120 */ 2121 if (th->th_seq == tp->rcv_nxt && 2122 LIST_EMPTY(&tp->t_segq) && 2123 TCPS_HAVEESTABLISHED(tp->t_state)) { 2124 if (DELAY_ACK(tp)) 2125 tp->t_flags |= TF_DELACK; 2126 else 2127 tp->t_flags |= TF_ACKNOW; 2128 tp->rcv_nxt += tlen; 2129 thflags = th->th_flags & TH_FIN; 2130 tcpstat.tcps_rcvpack++; 2131 tcpstat.tcps_rcvbyte += tlen; 2132 ND6_HINT(tp); 2133 if (so->so_state & SS_CANTRCVMORE) 2134 m_freem(m); 2135 else 2136 sbappend(&so->so_rcv, m); 2137 sorwakeup(so); 2138 } else { 2139 thflags = tcp_reass(tp, th, &tlen, m); 2140 tp->t_flags |= TF_ACKNOW; 2141 } 2142 2143 /* 2144 * Note the amount of data that peer has sent into 2145 * our window, in order to estimate the sender's 2146 * buffer size. 2147 */ 2148 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2149 } else { 2150 m_freem(m); 2151 thflags &= ~TH_FIN; 2152 } 2153 2154 /* 2155 * If FIN is received ACK the FIN and let the user know 2156 * that the connection is closing. 2157 */ 2158 if (thflags & TH_FIN) { 2159 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2160 socantrcvmore(so); 2161 /* 2162 * If connection is half-synchronized 2163 * (ie NEEDSYN flag on) then delay ACK, 2164 * so it may be piggybacked when SYN is sent. 2165 * Otherwise, since we received a FIN then no 2166 * more input can be expected, send ACK now. 2167 */ 2168 if (tp->t_flags & TF_NEEDSYN) 2169 tp->t_flags |= TF_DELACK; 2170 else 2171 tp->t_flags |= TF_ACKNOW; 2172 tp->rcv_nxt++; 2173 } 2174 switch (tp->t_state) { 2175 2176 /* 2177 * In SYN_RECEIVED and ESTABLISHED STATES 2178 * enter the CLOSE_WAIT state. 2179 */ 2180 case TCPS_SYN_RECEIVED: 2181 tp->t_starttime = ticks; 2182 /*FALLTHROUGH*/ 2183 case TCPS_ESTABLISHED: 2184 tp->t_state = TCPS_CLOSE_WAIT; 2185 break; 2186 2187 /* 2188 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2189 * enter the CLOSING state. 2190 */ 2191 case TCPS_FIN_WAIT_1: 2192 tp->t_state = TCPS_CLOSING; 2193 break; 2194 2195 /* 2196 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2197 * starting the time-wait timer, turning off the other 2198 * standard timers. 2199 */ 2200 case TCPS_FIN_WAIT_2: 2201 KASSERT(headlocked == 1, ("headlocked should be 1")); 2202 tcp_twstart(tp); 2203 INP_INFO_WUNLOCK(&tcbinfo); 2204 return; 2205 2206 /* 2207 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2208 */ 2209 case TCPS_TIME_WAIT: 2210 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2211 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2212 tcp_timer_2msl, tp); 2213 break; 2214 } 2215 } 2216 INP_INFO_WUNLOCK(&tcbinfo); 2217 #ifdef TCPDEBUG 2218 if (so->so_options & SO_DEBUG) 2219 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2220 &tcp_savetcp, 0); 2221 #endif 2222 2223 /* 2224 * Return any desired output. 2225 */ 2226 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2227 (void) tcp_output(tp); 2228 2229 check_delack: 2230 if (tp->t_flags & TF_DELACK) { 2231 tp->t_flags &= ~TF_DELACK; 2232 callout_reset(tp->tt_delack, tcp_delacktime, 2233 tcp_timer_delack, tp); 2234 } 2235 INP_UNLOCK(inp); 2236 return; 2237 2238 dropafterack: 2239 /* 2240 * Generate an ACK dropping incoming segment if it occupies 2241 * sequence space, where the ACK reflects our state. 2242 * 2243 * We can now skip the test for the RST flag since all 2244 * paths to this code happen after packets containing 2245 * RST have been dropped. 2246 * 2247 * In the SYN-RECEIVED state, don't send an ACK unless the 2248 * segment we received passes the SYN-RECEIVED ACK test. 2249 * If it fails send a RST. This breaks the loop in the 2250 * "LAND" DoS attack, and also prevents an ACK storm 2251 * between two listening ports that have been sent forged 2252 * SYN segments, each with the source address of the other. 2253 */ 2254 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2255 (SEQ_GT(tp->snd_una, th->th_ack) || 2256 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2257 rstreason = BANDLIM_RST_OPENPORT; 2258 goto dropwithreset; 2259 } 2260 #ifdef TCPDEBUG 2261 if (so->so_options & SO_DEBUG) 2262 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2263 &tcp_savetcp, 0); 2264 #endif 2265 KASSERT(headlocked, ("headlocked should be 1")); 2266 INP_INFO_WUNLOCK(&tcbinfo); 2267 m_freem(m); 2268 tp->t_flags |= TF_ACKNOW; 2269 (void) tcp_output(tp); 2270 INP_UNLOCK(inp); 2271 return; 2272 2273 dropwithreset: 2274 /* 2275 * Generate a RST, dropping incoming segment. 2276 * Make ACK acceptable to originator of segment. 2277 * Don't bother to respond if destination was broadcast/multicast. 2278 */ 2279 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2280 goto drop; 2281 if (isipv6) { 2282 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2283 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2284 goto drop; 2285 } else { 2286 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2287 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2288 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2289 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2290 goto drop; 2291 } 2292 /* IPv6 anycast check is done at tcp6_input() */ 2293 2294 /* 2295 * Perform bandwidth limiting. 2296 */ 2297 if (badport_bandlim(rstreason) < 0) 2298 goto drop; 2299 2300 #ifdef TCPDEBUG 2301 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2302 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2303 &tcp_savetcp, 0); 2304 #endif 2305 2306 if (tp) 2307 INP_UNLOCK(inp); 2308 2309 if (thflags & TH_ACK) 2310 /* mtod() below is safe as long as hdr dropping is delayed */ 2311 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2312 TH_RST); 2313 else { 2314 if (thflags & TH_SYN) 2315 tlen++; 2316 /* mtod() below is safe as long as hdr dropping is delayed */ 2317 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2318 (tcp_seq)0, TH_RST|TH_ACK); 2319 } 2320 if (headlocked) 2321 INP_INFO_WUNLOCK(&tcbinfo); 2322 return; 2323 2324 drop: 2325 /* 2326 * Drop space held by incoming segment and return. 2327 */ 2328 #ifdef TCPDEBUG 2329 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2330 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2331 &tcp_savetcp, 0); 2332 #endif 2333 if (tp) 2334 INP_UNLOCK(inp); 2335 m_freem(m); 2336 if (headlocked) 2337 INP_INFO_WUNLOCK(&tcbinfo); 2338 return; 2339 } 2340 2341 /* 2342 * Parse TCP options and place in tcpopt. 2343 */ 2344 static void 2345 tcp_dooptions(to, cp, cnt, is_syn) 2346 struct tcpopt *to; 2347 u_char *cp; 2348 int cnt; 2349 { 2350 int opt, optlen; 2351 2352 to->to_flags = 0; 2353 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2354 opt = cp[0]; 2355 if (opt == TCPOPT_EOL) 2356 break; 2357 if (opt == TCPOPT_NOP) 2358 optlen = 1; 2359 else { 2360 if (cnt < 2) 2361 break; 2362 optlen = cp[1]; 2363 if (optlen < 2 || optlen > cnt) 2364 break; 2365 } 2366 switch (opt) { 2367 case TCPOPT_MAXSEG: 2368 if (optlen != TCPOLEN_MAXSEG) 2369 continue; 2370 if (!is_syn) 2371 continue; 2372 to->to_flags |= TOF_MSS; 2373 bcopy((char *)cp + 2, 2374 (char *)&to->to_mss, sizeof(to->to_mss)); 2375 to->to_mss = ntohs(to->to_mss); 2376 break; 2377 case TCPOPT_WINDOW: 2378 if (optlen != TCPOLEN_WINDOW) 2379 continue; 2380 if (! is_syn) 2381 continue; 2382 to->to_flags |= TOF_SCALE; 2383 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2384 break; 2385 case TCPOPT_TIMESTAMP: 2386 if (optlen != TCPOLEN_TIMESTAMP) 2387 continue; 2388 to->to_flags |= TOF_TS; 2389 bcopy((char *)cp + 2, 2390 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2391 to->to_tsval = ntohl(to->to_tsval); 2392 bcopy((char *)cp + 6, 2393 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2394 to->to_tsecr = ntohl(to->to_tsecr); 2395 break; 2396 case TCPOPT_CC: 2397 if (optlen != TCPOLEN_CC) 2398 continue; 2399 to->to_flags |= TOF_CC; 2400 bcopy((char *)cp + 2, 2401 (char *)&to->to_cc, sizeof(to->to_cc)); 2402 to->to_cc = ntohl(to->to_cc); 2403 break; 2404 case TCPOPT_CCNEW: 2405 if (optlen != TCPOLEN_CC) 2406 continue; 2407 if (!is_syn) 2408 continue; 2409 to->to_flags |= TOF_CCNEW; 2410 bcopy((char *)cp + 2, 2411 (char *)&to->to_cc, sizeof(to->to_cc)); 2412 to->to_cc = ntohl(to->to_cc); 2413 break; 2414 case TCPOPT_CCECHO: 2415 if (optlen != TCPOLEN_CC) 2416 continue; 2417 if (!is_syn) 2418 continue; 2419 to->to_flags |= TOF_CCECHO; 2420 bcopy((char *)cp + 2, 2421 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2422 to->to_ccecho = ntohl(to->to_ccecho); 2423 break; 2424 default: 2425 continue; 2426 } 2427 } 2428 } 2429 2430 /* 2431 * Pull out of band byte out of a segment so 2432 * it doesn't appear in the user's data queue. 2433 * It is still reflected in the segment length for 2434 * sequencing purposes. 2435 */ 2436 static void 2437 tcp_pulloutofband(so, th, m, off) 2438 struct socket *so; 2439 struct tcphdr *th; 2440 register struct mbuf *m; 2441 int off; /* delayed to be droped hdrlen */ 2442 { 2443 int cnt = off + th->th_urp - 1; 2444 2445 while (cnt >= 0) { 2446 if (m->m_len > cnt) { 2447 char *cp = mtod(m, caddr_t) + cnt; 2448 struct tcpcb *tp = sototcpcb(so); 2449 2450 tp->t_iobc = *cp; 2451 tp->t_oobflags |= TCPOOB_HAVEDATA; 2452 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2453 m->m_len--; 2454 if (m->m_flags & M_PKTHDR) 2455 m->m_pkthdr.len--; 2456 return; 2457 } 2458 cnt -= m->m_len; 2459 m = m->m_next; 2460 if (m == 0) 2461 break; 2462 } 2463 panic("tcp_pulloutofband"); 2464 } 2465 2466 /* 2467 * Collect new round-trip time estimate 2468 * and update averages and current timeout. 2469 */ 2470 static void 2471 tcp_xmit_timer(tp, rtt) 2472 register struct tcpcb *tp; 2473 int rtt; 2474 { 2475 register int delta; 2476 2477 tcpstat.tcps_rttupdated++; 2478 tp->t_rttupdated++; 2479 if (tp->t_srtt != 0) { 2480 /* 2481 * srtt is stored as fixed point with 5 bits after the 2482 * binary point (i.e., scaled by 8). The following magic 2483 * is equivalent to the smoothing algorithm in rfc793 with 2484 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2485 * point). Adjust rtt to origin 0. 2486 */ 2487 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2488 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2489 2490 if ((tp->t_srtt += delta) <= 0) 2491 tp->t_srtt = 1; 2492 2493 /* 2494 * We accumulate a smoothed rtt variance (actually, a 2495 * smoothed mean difference), then set the retransmit 2496 * timer to smoothed rtt + 4 times the smoothed variance. 2497 * rttvar is stored as fixed point with 4 bits after the 2498 * binary point (scaled by 16). The following is 2499 * equivalent to rfc793 smoothing with an alpha of .75 2500 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2501 * rfc793's wired-in beta. 2502 */ 2503 if (delta < 0) 2504 delta = -delta; 2505 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2506 if ((tp->t_rttvar += delta) <= 0) 2507 tp->t_rttvar = 1; 2508 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2509 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2510 } else { 2511 /* 2512 * No rtt measurement yet - use the unsmoothed rtt. 2513 * Set the variance to half the rtt (so our first 2514 * retransmit happens at 3*rtt). 2515 */ 2516 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2517 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2518 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2519 } 2520 tp->t_rtttime = 0; 2521 tp->t_rxtshift = 0; 2522 2523 /* 2524 * the retransmit should happen at rtt + 4 * rttvar. 2525 * Because of the way we do the smoothing, srtt and rttvar 2526 * will each average +1/2 tick of bias. When we compute 2527 * the retransmit timer, we want 1/2 tick of rounding and 2528 * 1 extra tick because of +-1/2 tick uncertainty in the 2529 * firing of the timer. The bias will give us exactly the 2530 * 1.5 tick we need. But, because the bias is 2531 * statistical, we have to test that we don't drop below 2532 * the minimum feasible timer (which is 2 ticks). 2533 */ 2534 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2535 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2536 2537 /* 2538 * We received an ack for a packet that wasn't retransmitted; 2539 * it is probably safe to discard any error indications we've 2540 * received recently. This isn't quite right, but close enough 2541 * for now (a route might have failed after we sent a segment, 2542 * and the return path might not be symmetrical). 2543 */ 2544 tp->t_softerror = 0; 2545 } 2546 2547 /* 2548 * Determine a reasonable value for maxseg size. 2549 * If the route is known, check route for mtu. 2550 * If none, use an mss that can be handled on the outgoing 2551 * interface without forcing IP to fragment; if bigger than 2552 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2553 * to utilize large mbufs. If no route is found, route has no mtu, 2554 * or the destination isn't local, use a default, hopefully conservative 2555 * size (usually 512 or the default IP max size, but no more than the mtu 2556 * of the interface), as we can't discover anything about intervening 2557 * gateways or networks. We also initialize the congestion/slow start 2558 * window to be a single segment if the destination isn't local. 2559 * While looking at the routing entry, we also initialize other path-dependent 2560 * parameters from pre-set or cached values in the routing entry. 2561 * 2562 * Also take into account the space needed for options that we 2563 * send regularly. Make maxseg shorter by that amount to assure 2564 * that we can send maxseg amount of data even when the options 2565 * are present. Store the upper limit of the length of options plus 2566 * data in maxopd. 2567 * 2568 * NOTE that this routine is only called when we process an incoming 2569 * segment, for outgoing segments only tcp_mssopt is called. 2570 * 2571 * In case of T/TCP, we call this routine during implicit connection 2572 * setup as well (offer = -1), to initialize maxseg from the cached 2573 * MSS of our peer. 2574 */ 2575 void 2576 tcp_mss(tp, offer) 2577 struct tcpcb *tp; 2578 int offer; 2579 { 2580 register struct rtentry *rt; 2581 struct ifnet *ifp; 2582 register int rtt, mss; 2583 u_long bufsize; 2584 struct inpcb *inp = tp->t_inpcb; 2585 struct socket *so; 2586 struct rmxp_tao *taop; 2587 int origoffer = offer; 2588 #ifdef INET6 2589 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2590 size_t min_protoh = isipv6 ? 2591 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2592 sizeof (struct tcpiphdr); 2593 #else 2594 const int isipv6 = 0; 2595 const size_t min_protoh = sizeof (struct tcpiphdr); 2596 #endif 2597 2598 if (isipv6) 2599 rt = tcp_rtlookup6(&inp->inp_inc); 2600 else 2601 rt = tcp_rtlookup(&inp->inp_inc); 2602 if (rt == NULL) { 2603 tp->t_maxopd = tp->t_maxseg = 2604 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2605 return; 2606 } 2607 ifp = rt->rt_ifp; 2608 so = inp->inp_socket; 2609 2610 taop = rmx_taop(rt->rt_rmx); 2611 /* 2612 * Offer == -1 means that we didn't receive SYN yet, 2613 * use cached value in that case; 2614 */ 2615 if (offer == -1) 2616 offer = taop->tao_mssopt; 2617 /* 2618 * Offer == 0 means that there was no MSS on the SYN segment, 2619 * in this case we use tcp_mssdflt. 2620 */ 2621 if (offer == 0) 2622 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2623 else 2624 /* 2625 * Sanity check: make sure that maxopd will be large 2626 * enough to allow some data on segments even is the 2627 * all the option space is used (40bytes). Otherwise 2628 * funny things may happen in tcp_output. 2629 */ 2630 offer = max(offer, 64); 2631 taop->tao_mssopt = offer; 2632 2633 /* 2634 * While we're here, check if there's an initial rtt 2635 * or rttvar. Convert from the route-table units 2636 * to scaled multiples of the slow timeout timer. 2637 */ 2638 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2639 /* 2640 * XXX the lock bit for RTT indicates that the value 2641 * is also a minimum value; this is subject to time. 2642 */ 2643 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2644 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2645 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2646 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2647 tcpstat.tcps_usedrtt++; 2648 if (rt->rt_rmx.rmx_rttvar) { 2649 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2650 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2651 tcpstat.tcps_usedrttvar++; 2652 } else { 2653 /* default variation is +- 1 rtt */ 2654 tp->t_rttvar = 2655 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2656 } 2657 TCPT_RANGESET(tp->t_rxtcur, 2658 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2659 tp->t_rttmin, TCPTV_REXMTMAX); 2660 } 2661 /* 2662 * if there's an mtu associated with the route, use it 2663 * else, use the link mtu. 2664 */ 2665 if (rt->rt_rmx.rmx_mtu) 2666 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2667 else { 2668 if (isipv6) { 2669 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2670 min_protoh; 2671 if (!in6_localaddr(&inp->in6p_faddr)) 2672 mss = min(mss, tcp_v6mssdflt); 2673 } else { 2674 mss = ifp->if_mtu - min_protoh; 2675 if (!in_localaddr(inp->inp_faddr)) 2676 mss = min(mss, tcp_mssdflt); 2677 } 2678 } 2679 mss = min(mss, offer); 2680 /* 2681 * maxopd stores the maximum length of data AND options 2682 * in a segment; maxseg is the amount of data in a normal 2683 * segment. We need to store this value (maxopd) apart 2684 * from maxseg, because now every segment carries options 2685 * and thus we normally have somewhat less data in segments. 2686 */ 2687 tp->t_maxopd = mss; 2688 2689 /* 2690 * In case of T/TCP, origoffer==-1 indicates, that no segments 2691 * were received yet. In this case we just guess, otherwise 2692 * we do the same as before T/TCP. 2693 */ 2694 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2695 (origoffer == -1 || 2696 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2697 mss -= TCPOLEN_TSTAMP_APPA; 2698 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2699 (origoffer == -1 || 2700 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2701 mss -= TCPOLEN_CC_APPA; 2702 2703 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2704 if (mss > MCLBYTES) 2705 mss &= ~(MCLBYTES-1); 2706 #else 2707 if (mss > MCLBYTES) 2708 mss = mss / MCLBYTES * MCLBYTES; 2709 #endif 2710 /* 2711 * If there's a pipesize, change the socket buffer 2712 * to that size. Make the socket buffers an integral 2713 * number of mss units; if the mss is larger than 2714 * the socket buffer, decrease the mss. 2715 */ 2716 #ifdef RTV_SPIPE 2717 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2718 #endif 2719 bufsize = so->so_snd.sb_hiwat; 2720 if (bufsize < mss) 2721 mss = bufsize; 2722 else { 2723 bufsize = roundup(bufsize, mss); 2724 if (bufsize > sb_max) 2725 bufsize = sb_max; 2726 if (bufsize > so->so_snd.sb_hiwat) 2727 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2728 } 2729 tp->t_maxseg = mss; 2730 2731 #ifdef RTV_RPIPE 2732 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2733 #endif 2734 bufsize = so->so_rcv.sb_hiwat; 2735 if (bufsize > mss) { 2736 bufsize = roundup(bufsize, mss); 2737 if (bufsize > sb_max) 2738 bufsize = sb_max; 2739 if (bufsize > so->so_rcv.sb_hiwat) 2740 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2741 } 2742 2743 /* 2744 * Set the slow-start flight size depending on whether this 2745 * is a local network or not. 2746 */ 2747 if (tcp_do_rfc3390) 2748 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2749 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2750 (!isipv6 && in_localaddr(inp->inp_faddr))) 2751 tp->snd_cwnd = mss * ss_fltsz_local; 2752 else 2753 tp->snd_cwnd = mss * ss_fltsz; 2754 2755 if (rt->rt_rmx.rmx_ssthresh) { 2756 /* 2757 * There's some sort of gateway or interface 2758 * buffer limit on the path. Use this to set 2759 * the slow start threshhold, but set the 2760 * threshold to no less than 2*mss. 2761 */ 2762 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2763 tcpstat.tcps_usedssthresh++; 2764 } 2765 } 2766 2767 /* 2768 * Determine the MSS option to send on an outgoing SYN. 2769 */ 2770 int 2771 tcp_mssopt(tp) 2772 struct tcpcb *tp; 2773 { 2774 struct rtentry *rt; 2775 #ifdef INET6 2776 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2777 size_t min_protoh = isipv6 ? 2778 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2779 sizeof (struct tcpiphdr); 2780 #else 2781 const int isipv6 = 0; 2782 const size_t min_protoh = sizeof (struct tcpiphdr); 2783 #endif 2784 2785 if (isipv6) 2786 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2787 else 2788 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2789 if (rt == NULL) 2790 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2791 2792 return (rt->rt_ifp->if_mtu - min_protoh); 2793 } 2794 2795 2796 /* 2797 * On a partial ack arrives, force the retransmission of the 2798 * next unacknowledged segment. Do not clear tp->t_dupacks. 2799 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2800 * be started again. 2801 */ 2802 static void 2803 tcp_newreno_partial_ack(tp, th) 2804 struct tcpcb *tp; 2805 struct tcphdr *th; 2806 { 2807 tcp_seq onxt = tp->snd_nxt; 2808 u_long ocwnd = tp->snd_cwnd; 2809 2810 callout_stop(tp->tt_rexmt); 2811 tp->t_rtttime = 0; 2812 tp->snd_nxt = th->th_ack; 2813 /* 2814 * Set snd_cwnd to one segment beyond acknowledged offset. 2815 * (tp->snd_una has not yet been updated when this function is called.) 2816 */ 2817 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2818 tp->t_flags |= TF_ACKNOW; 2819 (void) tcp_output(tp); 2820 tp->snd_cwnd = ocwnd; 2821 if (SEQ_GT(onxt, tp->snd_nxt)) 2822 tp->snd_nxt = onxt; 2823 /* 2824 * Partial window deflation. Relies on fact that tp->snd_una 2825 * not updated yet. 2826 */ 2827 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2828 } 2829 2830 /* 2831 * Returns 1 if the TIME_WAIT state was killed and we should start over, 2832 * looking for a pcb in the listen state. Returns 0 otherwise. 2833 */ 2834 static int 2835 tcp_timewait(tw, to, th, m, tlen) 2836 struct tcptw *tw; 2837 struct tcpopt *to; 2838 struct tcphdr *th; 2839 struct mbuf *m; 2840 int tlen; 2841 { 2842 int thflags; 2843 tcp_seq seq; 2844 #ifdef INET6 2845 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 2846 #else 2847 const int isipv6 = 0; 2848 #endif 2849 2850 thflags = th->th_flags; 2851 2852 /* 2853 * NOTE: for FIN_WAIT_2 (to be added later), 2854 * must validate sequence number before accepting RST 2855 */ 2856 2857 /* 2858 * If the segment contains RST: 2859 * Drop the segment - see Stevens, vol. 2, p. 964 and 2860 * RFC 1337. 2861 */ 2862 if (thflags & TH_RST) 2863 goto drop; 2864 2865 /* 2866 * If segment contains a SYN and CC [not CC.NEW] option: 2867 * if connection duration > MSL, drop packet and send RST; 2868 * 2869 * if SEG.CC > CCrecv then is new SYN. 2870 * Complete close and delete TCPCB. Then reprocess 2871 * segment, hoping to find new TCPCB in LISTEN state; 2872 * 2873 * else must be old SYN; drop it. 2874 * else do normal processing. 2875 */ 2876 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 2877 if ((ticks - tw->t_starttime) > tcp_msl) 2878 goto reset; 2879 if (CC_GT(to->to_cc, tw->cc_recv)) { 2880 (void) tcp_twclose(tw, 0); 2881 return (1); 2882 } 2883 goto drop; 2884 } 2885 2886 #if 0 2887 /* PAWS not needed at the moment */ 2888 /* 2889 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2890 * and it's less than ts_recent, drop it. 2891 */ 2892 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2893 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2894 if ((thflags & TH_ACK) == 0) 2895 goto drop; 2896 goto ack; 2897 } 2898 /* 2899 * ts_recent is never updated because we never accept new segments. 2900 */ 2901 #endif 2902 2903 /* 2904 * If a new connection request is received 2905 * while in TIME_WAIT, drop the old connection 2906 * and start over if the sequence numbers 2907 * are above the previous ones. 2908 */ 2909 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 2910 (void) tcp_twclose(tw, 0); 2911 return (1); 2912 } 2913 2914 /* 2915 * Drop the the segment if it does not contain an ACK. 2916 */ 2917 if ((thflags & TH_ACK) == 0) 2918 goto drop; 2919 2920 /* 2921 * Reset the 2MSL timer if this is a duplicate FIN. 2922 */ 2923 if (thflags & TH_FIN) { 2924 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 2925 if (seq + 1 == tw->rcv_nxt) 2926 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 2927 } 2928 2929 /* 2930 * Acknowledge the segment if it has data or is not a duplicate ACK. 2931 */ 2932 if (thflags != TH_ACK || tlen != 0 || 2933 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 2934 tcp_twrespond(tw, TH_ACK); 2935 goto drop; 2936 2937 reset: 2938 /* 2939 * Generate a RST, dropping incoming segment. 2940 * Make ACK acceptable to originator of segment. 2941 * Don't bother to respond if destination was broadcast/multicast. 2942 */ 2943 if (m->m_flags & (M_BCAST|M_MCAST)) 2944 goto drop; 2945 if (isipv6) { 2946 struct ip6_hdr *ip6; 2947 2948 /* IPv6 anycast check is done at tcp6_input() */ 2949 ip6 = mtod(m, struct ip6_hdr *); 2950 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2951 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2952 goto drop; 2953 } else { 2954 struct ip *ip; 2955 2956 ip = mtod(m, struct ip *); 2957 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2958 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2959 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2960 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2961 goto drop; 2962 } 2963 if (thflags & TH_ACK) { 2964 tcp_respond(NULL, 2965 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 2966 } else { 2967 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 2968 tcp_respond(NULL, 2969 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 2970 } 2971 INP_UNLOCK(tw->tw_inpcb); 2972 return (0); 2973 2974 drop: 2975 INP_UNLOCK(tw->tw_inpcb); 2976 m_freem(m); 2977 return (0); 2978 } 2979